xref: /freebsd/sys/net/if_tuntap.c (revision 96c1d8db39dfeea78ea3f27d67649252a39bbf2e)
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 #include <sys/osd.h>
78 
79 #include <net/ethernet.h>
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/if_clone.h>
83 #include <net/if_dl.h>
84 #include <net/if_media.h>
85 #include <net/if_private.h>
86 #include <net/if_types.h>
87 #include <net/if_vlan_var.h>
88 #include <net/netisr.h>
89 #include <net/route.h>
90 #include <net/vnet.h>
91 #include <netinet/in.h>
92 #ifdef INET
93 #include <netinet/ip.h>
94 #endif
95 #ifdef INET6
96 #include <netinet/ip6.h>
97 #include <netinet6/ip6_var.h>
98 #endif
99 #include <netinet/udp.h>
100 #include <netinet/tcp.h>
101 #include <netinet/tcp_lro.h>
102 #include <net/bpf.h>
103 #include <net/if_tap.h>
104 #include <net/if_tun.h>
105 
106 #include <dev/virtio/network/virtio_net.h>
107 
108 #include <sys/queue.h>
109 #include <sys/condvar.h>
110 #include <security/mac/mac_framework.h>
111 
112 struct tuntap_driver;
113 
114 /*
115  * tun_list is protected by global tunmtx.  Other mutable fields are
116  * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
117  * static for the duration of a tunnel interface.
118  */
119 struct tuntap_softc {
120 	TAILQ_ENTRY(tuntap_softc)	 tun_list;
121 	struct cdev			*tun_alias;
122 	struct cdev			*tun_dev;
123 	u_short				 tun_flags;	/* misc flags */
124 #define	TUN_OPEN	0x0001
125 #define	TUN_INITED	0x0002
126 #define	TUN_UNUSED1	0x0008
127 #define	TUN_UNUSED2	0x0010
128 #define	TUN_LMODE	0x0020
129 #define	TUN_RWAIT	0x0040
130 #define	TUN_ASYNC	0x0080
131 #define	TUN_IFHEAD	0x0100
132 #define	TUN_DYING	0x0200
133 #define	TUN_L2		0x0400
134 #define	TUN_VMNET	0x0800
135 #define	TUN_TRANSIENT	0x1000
136 
137 #define	TUN_DRIVER_IDENT_MASK	(TUN_L2 | TUN_VMNET)
138 #define	TUN_READY		(TUN_OPEN | TUN_INITED)
139 
140 	pid_t			 tun_pid;	/* owning pid */
141 	struct epoch_context	 tun_epoch_ctx;
142 	struct ifnet		*tun_ifp;	/* the interface */
143 	struct sigio		*tun_sigio;	/* async I/O info */
144 	struct tuntap_driver	*tun_drv;	/* appropriate driver */
145 	struct selinfo		 tun_rsel;	/* read select */
146 	struct mtx		 tun_mtx;	/* softc field mutex */
147 	struct cv		 tun_cv;	/* for ref'd dev destroy */
148 	struct ether_addr	 tun_ether;	/* remote address */
149 	int			 tun_busy;	/* busy count */
150 	int			 tun_vhdrlen;	/* virtio-net header length */
151 	struct lro_ctrl		 tun_lro;	/* for TCP LRO */
152 	bool			 tun_lro_ready;	/* TCP LRO initialized */
153 };
154 #define	TUN2IFP(sc)	((sc)->tun_ifp)
155 
156 #define	TUNDEBUG	if (tundebug) if_printf
157 
158 #define	TUN_LOCK(tp)		mtx_lock(&(tp)->tun_mtx)
159 #define	TUN_UNLOCK(tp)		mtx_unlock(&(tp)->tun_mtx)
160 #define	TUN_LOCK_ASSERT(tp)	mtx_assert(&(tp)->tun_mtx, MA_OWNED);
161 
162 #define	TUN_VMIO_FLAG_MASK	0x0fff
163 
164 /*
165  * Interface capabilities of a tap device that supports the virtio-net
166  * header.
167  */
168 #define TAP_VNET_HDR_CAPS	(IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6	\
169 				| IFCAP_VLAN_HWCSUM			\
170 				| IFCAP_TSO | IFCAP_LRO			\
171 				| IFCAP_VLAN_HWTSO)
172 
173 #define TAP_ALL_OFFLOAD		(CSUM_TSO | CSUM_TCP | CSUM_UDP |\
174 				    CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
175 
176 /*
177  * All mutable global variables in if_tun are locked using tunmtx, with
178  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
179  * which are static after setup.
180  */
181 static struct mtx tunmtx;
182 static eventhandler_tag arrival_tag;
183 static eventhandler_tag clone_tag;
184 static int tuntap_osd_jail_slot;
185 static const char tunname[] = "tun";
186 static const char tapname[] = "tap";
187 static const char vmnetname[] = "vmnet";
188 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
189 static int tundebug = 0;
190 static int tundclone = 1;
191 static int tap_allow_uopen = 0;	/* allow user devfs cloning */
192 static int tapuponopen = 0;	/* IFF_UP on open() */
193 static int tapdclone = 1;	/* enable devfs cloning */
194 
195 static TAILQ_HEAD(,tuntap_softc)	tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
196 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
197 
198 static struct sx tun_ioctl_sx;
199 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
200 
201 SYSCTL_DECL(_net_link);
202 /* tun */
203 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
204     "IP tunnel software network interface");
205 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
206     "Enable legacy devfs interface creation");
207 
208 /* tap */
209 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
210     "Ethernet tunnel software network interface");
211 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
212     "Enable legacy devfs interface creation for all users");
213 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
214     "Bring interface up when /dev/tap is opened");
215 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
216     "Enable legacy devfs interface creation");
217 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
218 
219 static int	tun_create_device(struct tuntap_driver *drv, int unit,
220     struct ucred *cr, struct cdev **dev, const char *name);
221 static int	tun_busy_locked(struct tuntap_softc *tp);
222 static void	tun_unbusy_locked(struct tuntap_softc *tp);
223 static int	tun_busy(struct tuntap_softc *tp);
224 static void	tun_unbusy(struct tuntap_softc *tp);
225 
226 static int	tuntap_name2info(const char *name, int *unit, int *flags);
227 static void	tunclone(void *arg, struct ucred *cred, char *name,
228 		    int namelen, struct cdev **dev);
229 static void	tuncreate(struct cdev *dev);
230 static void	tundtor(void *data);
231 static void	tunrename(void *arg, struct ifnet *ifp);
232 static int	tunifioctl(struct ifnet *, u_long, caddr_t);
233 static void	tuninit(struct ifnet *);
234 static void	tunifinit(void *xtp);
235 static int	tuntapmodevent(module_t, int, void *);
236 static int	tunoutput(struct ifnet *, struct mbuf *,
237 		    const struct sockaddr *, struct route *ro);
238 static void	tunstart(struct ifnet *);
239 static void	tunstart_l2(struct ifnet *);
240 
241 static int	tun_clone_match(struct if_clone *ifc, const char *name);
242 static int	tap_clone_match(struct if_clone *ifc, const char *name);
243 static int	vmnet_clone_match(struct if_clone *ifc, const char *name);
244 static int	tun_clone_create(struct if_clone *, char *, size_t,
245 		    struct ifc_data *, struct ifnet **);
246 static int	tun_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
247 static void	tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
248 
249 static d_open_t		tunopen;
250 static d_read_t		tunread;
251 static d_write_t	tunwrite;
252 static d_ioctl_t	tunioctl;
253 static d_poll_t		tunpoll;
254 static d_kqfilter_t	tunkqfilter;
255 
256 static int		tunkqread(struct knote *, long);
257 static int		tunkqwrite(struct knote *, long);
258 static void		tunkqdetach(struct knote *);
259 
260 static const struct filterops tun_read_filterops = {
261 	.f_isfd =	1,
262 	.f_attach =	NULL,
263 	.f_detach =	tunkqdetach,
264 	.f_event =	tunkqread,
265 	.f_copy =	knote_triv_copy,
266 };
267 
268 static const struct filterops tun_write_filterops = {
269 	.f_isfd =	1,
270 	.f_attach =	NULL,
271 	.f_detach =	tunkqdetach,
272 	.f_event =	tunkqwrite,
273 	.f_copy =	knote_triv_copy,
274 };
275 
276 static struct tuntap_driver {
277 	struct cdevsw		 cdevsw;
278 	int			 ident_flags;
279 	struct unrhdr		*unrhdr;
280 	struct clonedevs	*clones;
281 	ifc_match_f		*clone_match_fn;
282 	ifc_create_f		*clone_create_fn;
283 	ifc_destroy_f		*clone_destroy_fn;
284 } tuntap_drivers[] = {
285 	{
286 		.ident_flags =	0,
287 		.cdevsw =	{
288 		    .d_version =	D_VERSION,
289 		    .d_flags =		D_NEEDMINOR,
290 		    .d_open =		tunopen,
291 		    .d_read =		tunread,
292 		    .d_write =		tunwrite,
293 		    .d_ioctl =		tunioctl,
294 		    .d_poll =		tunpoll,
295 		    .d_kqfilter =	tunkqfilter,
296 		    .d_name =		tunname,
297 		},
298 		.clone_match_fn =	tun_clone_match,
299 		.clone_create_fn =	tun_clone_create,
300 		.clone_destroy_fn =	tun_clone_destroy,
301 	},
302 	{
303 		.ident_flags =	TUN_L2,
304 		.cdevsw =	{
305 		    .d_version =	D_VERSION,
306 		    .d_flags =		D_NEEDMINOR,
307 		    .d_open =		tunopen,
308 		    .d_read =		tunread,
309 		    .d_write =		tunwrite,
310 		    .d_ioctl =		tunioctl,
311 		    .d_poll =		tunpoll,
312 		    .d_kqfilter =	tunkqfilter,
313 		    .d_name =		tapname,
314 		},
315 		.clone_match_fn =	tap_clone_match,
316 		.clone_create_fn =	tun_clone_create,
317 		.clone_destroy_fn =	tun_clone_destroy,
318 	},
319 	{
320 		.ident_flags =	TUN_L2 | TUN_VMNET,
321 		.cdevsw =	{
322 		    .d_version =	D_VERSION,
323 		    .d_flags =		D_NEEDMINOR,
324 		    .d_open =		tunopen,
325 		    .d_read =		tunread,
326 		    .d_write =		tunwrite,
327 		    .d_ioctl =		tunioctl,
328 		    .d_poll =		tunpoll,
329 		    .d_kqfilter =	tunkqfilter,
330 		    .d_name =		vmnetname,
331 		},
332 		.clone_match_fn =	vmnet_clone_match,
333 		.clone_create_fn =	tun_clone_create,
334 		.clone_destroy_fn =	tun_clone_destroy,
335 	},
336 };
337 #define	NDRV	nitems(tuntap_drivers)
338 
339 VNET_DEFINE_STATIC(struct if_clone *, tuntap_driver_cloners[NDRV]);
340 #define	V_tuntap_driver_cloners	VNET(tuntap_driver_cloners)
341 
342 /*
343  * Mechanism for marking a tunnel device as busy so that we can safely do some
344  * orthogonal operations (such as operations on devices) without racing against
345  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
346  * open, to be woken up when the condition is alleviated.
347  */
348 static int
tun_busy_locked(struct tuntap_softc * tp)349 tun_busy_locked(struct tuntap_softc *tp)
350 {
351 
352 	TUN_LOCK_ASSERT(tp);
353 	if ((tp->tun_flags & TUN_DYING) != 0) {
354 		/*
355 		 * Perhaps unintuitive, but the device is busy going away.
356 		 * Other interpretations of EBUSY from tun_busy make little
357 		 * sense, since making a busy device even more busy doesn't
358 		 * sound like a problem.
359 		 */
360 		return (EBUSY);
361 	}
362 
363 	++tp->tun_busy;
364 	return (0);
365 }
366 
367 static void
tun_unbusy_locked(struct tuntap_softc * tp)368 tun_unbusy_locked(struct tuntap_softc *tp)
369 {
370 
371 	TUN_LOCK_ASSERT(tp);
372 	KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
373 
374 	--tp->tun_busy;
375 	/* Wake up anything that may be waiting on our busy tunnel. */
376 	if (tp->tun_busy == 0)
377 		cv_broadcast(&tp->tun_cv);
378 }
379 
380 static int
tun_busy(struct tuntap_softc * tp)381 tun_busy(struct tuntap_softc *tp)
382 {
383 	int ret;
384 
385 	TUN_LOCK(tp);
386 	ret = tun_busy_locked(tp);
387 	TUN_UNLOCK(tp);
388 	return (ret);
389 }
390 
391 static void
tun_unbusy(struct tuntap_softc * tp)392 tun_unbusy(struct tuntap_softc *tp)
393 {
394 
395 	TUN_LOCK(tp);
396 	tun_unbusy_locked(tp);
397 	TUN_UNLOCK(tp);
398 }
399 
400 /*
401  * Sets unit and/or flags given the device name.  Must be called with correct
402  * vnet context.
403  */
404 static int
tuntap_name2info(const char * name,int * outunit,int * outflags)405 tuntap_name2info(const char *name, int *outunit, int *outflags)
406 {
407 	struct tuntap_driver *drv;
408 	char *dname;
409 	int flags, unit;
410 	bool found;
411 
412 	if (name == NULL)
413 		return (EINVAL);
414 
415 	/*
416 	 * Needed for dev_stdclone, but dev_stdclone will not modify, it just
417 	 * wants to be able to pass back a char * through the second param. We
418 	 * will always set that as NULL here, so we'll fake it.
419 	 */
420 	dname = __DECONST(char *, name);
421 	found = false;
422 
423 	for (u_int i = 0; i < NDRV; i++) {
424 		drv = &tuntap_drivers[i];
425 
426 		if (strcmp(name, drv->cdevsw.d_name) == 0) {
427 			found = true;
428 			unit = -1;
429 			flags = drv->ident_flags;
430 			break;
431 		}
432 
433 		if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
434 			found = true;
435 			flags = drv->ident_flags;
436 			break;
437 		}
438 	}
439 
440 	if (!found)
441 		return (ENXIO);
442 
443 	if (outunit != NULL)
444 		*outunit = unit;
445 	if (outflags != NULL)
446 		*outflags = flags;
447 	return (0);
448 }
449 
450 static struct if_clone *
tuntap_cloner_from_flags(int tun_flags)451 tuntap_cloner_from_flags(int tun_flags)
452 {
453 
454 	for (u_int i = 0; i < NDRV; i++)
455 		if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
456 		    tuntap_drivers[i].ident_flags)
457 			return (V_tuntap_driver_cloners[i]);
458 
459 	return (NULL);
460 }
461 
462 /*
463  * Get driver information from a set of flags specified.  Masks the identifying
464  * part of the flags and compares it against all of the available
465  * tuntap_drivers.
466  */
467 static struct tuntap_driver *
tuntap_driver_from_flags(int tun_flags)468 tuntap_driver_from_flags(int tun_flags)
469 {
470 
471 	for (u_int i = 0; i < NDRV; i++)
472 		if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
473 		    tuntap_drivers[i].ident_flags)
474 			return (&tuntap_drivers[i]);
475 
476 	return (NULL);
477 }
478 
479 static int
tun_clone_match(struct if_clone * ifc,const char * name)480 tun_clone_match(struct if_clone *ifc, const char *name)
481 {
482 	int tunflags;
483 
484 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
485 		if ((tunflags & TUN_L2) == 0)
486 			return (1);
487 	}
488 
489 	return (0);
490 }
491 
492 static int
tap_clone_match(struct if_clone * ifc,const char * name)493 tap_clone_match(struct if_clone *ifc, const char *name)
494 {
495 	int tunflags;
496 
497 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
498 		if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
499 			return (1);
500 	}
501 
502 	return (0);
503 }
504 
505 static int
vmnet_clone_match(struct if_clone * ifc,const char * name)506 vmnet_clone_match(struct if_clone *ifc, const char *name)
507 {
508 	int tunflags;
509 
510 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
511 		if ((tunflags & TUN_VMNET) != 0)
512 			return (1);
513 	}
514 
515 	return (0);
516 }
517 
518 /*
519  * Create a clone via the ifnet cloning mechanism.  Note that this is invoked
520  * indirectly by tunclone() below.
521  */
522 static int
tun_clone_create(struct if_clone * ifc,char * name,size_t len,struct ifc_data * ifd,struct ifnet ** ifpp)523 tun_clone_create(struct if_clone *ifc, char *name, size_t len,
524     struct ifc_data *ifd, struct ifnet **ifpp)
525 {
526 	struct tuntap_driver *drv;
527 	struct cdev *dev;
528 	int err, i, tunflags, unit;
529 
530 	tunflags = 0;
531 	/* The name here tells us exactly what we're creating */
532 	err = tuntap_name2info(name, &unit, &tunflags);
533 	if (err != 0)
534 		return (err);
535 
536 	drv = tuntap_driver_from_flags(tunflags);
537 	if (drv == NULL)
538 		return (ENXIO);
539 
540 	if (unit != -1) {
541 		/* If this unit number is still available that's okay. */
542 		if (alloc_unr_specific(drv->unrhdr, unit) == -1)
543 			return (EEXIST);
544 	} else {
545 		unit = alloc_unr(drv->unrhdr);
546 	}
547 
548 	snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
549 
550 	/* find any existing device, or allocate new unit number */
551 	dev = NULL;
552 	i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
553 	/* No preexisting struct cdev *, create one */
554 	if (i != 0)
555 		i = tun_create_device(drv, unit, NULL, &dev, name);
556 	if (i == 0) {
557 		struct tuntap_softc *tp;
558 
559 		tuncreate(dev);
560 		tp = dev->si_drv1;
561 		*ifpp = tp->tun_ifp;
562 	}
563 
564 	return (i);
565 }
566 
567 /*
568  * Create a clone via devfs access.
569  */
570 static void
tunclone(void * arg,struct ucred * cred,char * name,int namelen,struct cdev ** dev)571 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
572     struct cdev **dev)
573 {
574 	char devname[SPECNAMELEN + 1];
575 	struct tuntap_driver *drv;
576 	int append_unit, i, u, tunflags;
577 	bool mayclone;
578 
579 	if (*dev != NULL)
580 		return;
581 
582 	tunflags = 0;
583 	CURVNET_SET(CRED_TO_VNET(cred));
584 	if (tuntap_name2info(name, &u, &tunflags) != 0)
585 		goto out;	/* Not recognized */
586 
587 	if (u != -1 && u > IF_MAXUNIT)
588 		goto out;	/* Unit number too high */
589 
590 	mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
591 	if ((tunflags & TUN_L2) != 0) {
592 		/* tap/vmnet allow user open with a sysctl */
593 		mayclone = (mayclone || tap_allow_uopen) && tapdclone;
594 	} else {
595 		mayclone = mayclone && tundclone;
596 	}
597 
598 	/*
599 	 * If tun cloning is enabled, only the superuser can create an
600 	 * interface.
601 	 */
602 	if (!mayclone)
603 		goto out;
604 
605 	if (u == -1)
606 		append_unit = 1;
607 	else
608 		append_unit = 0;
609 
610 	drv = tuntap_driver_from_flags(tunflags);
611 	if (drv == NULL)
612 		goto out;
613 
614 	/* find any existing device, or allocate new unit number */
615 	i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
616 	if (i) {
617 		if (append_unit) {
618 			namelen = snprintf(devname, sizeof(devname), "%s%d",
619 			    name, u);
620 			name = devname;
621 		}
622 
623 		i = tun_create_device(drv, u, cred, dev, name);
624 	} else {
625 		/* Consumed by the dev_clone invoker. */
626 		dev_ref(*dev);
627 	}
628 	if (i == 0)
629 		if_clone_create(name, namelen, NULL);
630 out:
631 	CURVNET_RESTORE();
632 }
633 
634 static void
tunfree(struct epoch_context * ctx)635 tunfree(struct epoch_context *ctx)
636 {
637 	struct tuntap_softc *tp;
638 
639 	tp = __containerof(ctx, struct tuntap_softc, tun_epoch_ctx);
640 
641 	/* Any remaining resources that would be needed by a concurrent open. */
642 	mtx_destroy(&tp->tun_mtx);
643 	free(tp, M_TUN);
644 }
645 
646 static int
tun_destroy(struct tuntap_softc * tp,bool may_intr)647 tun_destroy(struct tuntap_softc *tp, bool may_intr)
648 {
649 	int error;
650 
651 	TUN_LOCK(tp);
652 
653 	/*
654 	 * Transient tunnels may have set TUN_DYING if we're being destroyed as
655 	 * a result of the last close, which we'll allow.
656 	 */
657 	MPASS((tp->tun_flags & (TUN_DYING | TUN_TRANSIENT)) != TUN_DYING);
658 	tp->tun_flags |= TUN_DYING;
659 	error = 0;
660 	while (tp->tun_busy != 0) {
661 		if (may_intr)
662 			error = cv_wait_sig(&tp->tun_cv, &tp->tun_mtx);
663 		else
664 			cv_wait(&tp->tun_cv, &tp->tun_mtx);
665 		if (error != 0 && tp->tun_busy != 0) {
666 			tp->tun_flags &= ~TUN_DYING;
667 			TUN_UNLOCK(tp);
668 			return (error);
669 		}
670 	}
671 	TUN_UNLOCK(tp);
672 
673 	CURVNET_SET(TUN2IFP(tp)->if_vnet);
674 
675 	mtx_lock(&tunmtx);
676 	TAILQ_REMOVE(&tunhead, tp, tun_list);
677 	mtx_unlock(&tunmtx);
678 
679 	/*
680 	 * destroy_dev will take care of any alias.  For transient tunnels,
681 	 * we're being called from close(2) so we can't destroy it ourselves
682 	 * without deadlocking, but we already know that we can cleanup
683 	 * everything else and just continue to prevent it from being reopened.
684 	 */
685 	if ((tp->tun_flags & TUN_TRANSIENT) != 0) {
686 		atomic_store_ptr(&tp->tun_dev->si_drv1, tp->tun_dev);
687 		destroy_dev_sched(tp->tun_dev);
688 	} else {
689 		destroy_dev(tp->tun_dev);
690 	}
691 	seldrain(&tp->tun_rsel);
692 	knlist_clear(&tp->tun_rsel.si_note, 0);
693 	knlist_destroy(&tp->tun_rsel.si_note);
694 	if ((tp->tun_flags & TUN_L2) != 0) {
695 		ether_ifdetach(TUN2IFP(tp));
696 	} else {
697 		bpfdetach(TUN2IFP(tp));
698 		if_detach(TUN2IFP(tp));
699 	}
700 	sx_xlock(&tun_ioctl_sx);
701 	TUN2IFP(tp)->if_softc = NULL;
702 	sx_xunlock(&tun_ioctl_sx);
703 	free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
704 	if_free(TUN2IFP(tp));
705 	cv_destroy(&tp->tun_cv);
706 	NET_EPOCH_CALL(tunfree, &tp->tun_epoch_ctx);
707 	CURVNET_RESTORE();
708 
709 	return (0);
710 }
711 
712 static int
tun_clone_destroy(struct if_clone * ifc __unused,struct ifnet * ifp,uint32_t flags)713 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags)
714 {
715 	struct tuntap_softc *tp = ifp->if_softc;
716 
717 	return (tun_destroy(tp, true));
718 }
719 
720 static void
vnet_tun_init(const void * unused __unused)721 vnet_tun_init(const void *unused __unused)
722 {
723 
724 	for (u_int i = 0; i < NDRV; ++i) {
725 		struct if_clone_addreq req = {
726 			.match_f = tuntap_drivers[i].clone_match_fn,
727 			.create_f = tuntap_drivers[i].clone_create_fn,
728 			.destroy_f = tuntap_drivers[i].clone_destroy_fn,
729 		};
730 		V_tuntap_driver_cloners[i] =
731 		    ifc_attach_cloner(tuntap_drivers[i].cdevsw.d_name, &req);
732 	};
733 }
734 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
735 		vnet_tun_init, NULL);
736 
737 static void
tun_uninit(const void * unused __unused)738 tun_uninit(const void *unused __unused)
739 {
740 	struct tuntap_driver *drv;
741 	struct tuntap_softc *tp;
742 	int i;
743 
744 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
745 	EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
746 
747 	CURVNET_SET(vnet0);
748 	for (u_int i = 0; i < NDRV; i++) {
749 		if_clone_detach(V_tuntap_driver_cloners[i]);
750 		V_tuntap_driver_cloners[i] = NULL;
751 	}
752 	CURVNET_RESTORE();
753 
754 	if (tuntap_osd_jail_slot != 0)
755 		osd_jail_deregister(tuntap_osd_jail_slot);
756 
757 	mtx_lock(&tunmtx);
758 	while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
759 		mtx_unlock(&tunmtx);
760 		/* tun_destroy() will remove it from the tailq. */
761 		tun_destroy(tp, false);
762 		mtx_lock(&tunmtx);
763 	}
764 	mtx_unlock(&tunmtx);
765 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
766 		drv = &tuntap_drivers[i];
767 		destroy_dev_drain(&drv->cdevsw);
768 		delete_unrhdr(drv->unrhdr);
769 		clone_cleanup(&drv->clones);
770 	}
771 	NET_EPOCH_DRAIN_CALLBACKS();
772 	mtx_destroy(&tunmtx);
773 }
774 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
775 
776 static struct tuntap_driver *
tuntap_driver_from_ifnet(const struct ifnet * ifp)777 tuntap_driver_from_ifnet(const struct ifnet *ifp)
778 {
779 	struct tuntap_driver *drv;
780 	int i;
781 
782 	if (ifp == NULL)
783 		return (NULL);
784 
785 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
786 		drv = &tuntap_drivers[i];
787 		if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
788 			return (drv);
789 	}
790 
791 	return (NULL);
792 }
793 
794 /*
795  * Remove devices that were created by devfs cloning, as they hold references
796  * which prevent the prison from collapsing, in which state VNET sysuninits will
797  * not be invoked.
798  */
799 static int
tuntap_prison_remove(void * obj,void * data __unused)800 tuntap_prison_remove(void *obj, void *data __unused)
801 {
802 #ifdef VIMAGE
803 	struct prison *pr;
804 
805 	pr = obj;
806 	if (prison_owns_vnet(pr)) {
807 		CURVNET_SET(pr->pr_vnet);
808 		for (u_int i = 0; i < NDRV; i++) {
809 			if_clone_detach(V_tuntap_driver_cloners[i]);
810 			V_tuntap_driver_cloners[i] = NULL;
811 		}
812 		CURVNET_RESTORE();
813 	}
814 #endif
815 	return (0);
816 }
817 
818 static int
tuntapmodevent(module_t mod,int type,void * data)819 tuntapmodevent(module_t mod, int type, void *data)
820 {
821 	struct tuntap_driver *drv;
822 	int i;
823 
824 	switch (type) {
825 	case MOD_LOAD:
826 		mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
827 		for (i = 0; i < nitems(tuntap_drivers); ++i) {
828 			drv = &tuntap_drivers[i];
829 			clone_setup(&drv->clones);
830 			drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
831 		}
832 		osd_method_t methods[PR_MAXMETHOD] = {
833 			[PR_METHOD_REMOVE] = tuntap_prison_remove,
834 		};
835 		tuntap_osd_jail_slot = osd_jail_register(NULL, methods);
836 		arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
837 		    tunrename, 0, 1000);
838 		if (arrival_tag == NULL)
839 			return (ENOMEM);
840 		clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
841 		if (clone_tag == NULL)
842 			return (ENOMEM);
843 		break;
844 	case MOD_UNLOAD:
845 		/* See tun_uninit(). */
846 		break;
847 	default:
848 		return EOPNOTSUPP;
849 	}
850 	return 0;
851 }
852 
853 static moduledata_t tuntap_mod = {
854 	"if_tuntap",
855 	tuntapmodevent,
856 	0
857 };
858 
859 /* We'll only ever have these two, so no need for a macro. */
860 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
861 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
862 
863 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
864 MODULE_VERSION(if_tuntap, 1);
865 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
866 MODULE_VERSION(if_tun, 1);
867 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
868 MODULE_VERSION(if_tap, 1);
869 
870 static int
tun_create_device(struct tuntap_driver * drv,int unit,struct ucred * cr,struct cdev ** dev,const char * name)871 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
872     struct cdev **dev, const char *name)
873 {
874 	struct make_dev_args args;
875 	struct tuntap_softc *tp;
876 	int error;
877 
878 	tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
879 	mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
880 	cv_init(&tp->tun_cv, "tun_condvar");
881 	tp->tun_flags = drv->ident_flags;
882 	tp->tun_drv = drv;
883 
884 	make_dev_args_init(&args);
885 	if (cr != NULL)
886 		args.mda_flags = MAKEDEV_REF | MAKEDEV_CHECKNAME;
887 	args.mda_devsw = &drv->cdevsw;
888 	args.mda_cr = cr;
889 	args.mda_uid = UID_UUCP;
890 	args.mda_gid = GID_DIALER;
891 	args.mda_mode = 0600;
892 	args.mda_unit = unit;
893 	args.mda_si_drv1 = tp;
894 	error = make_dev_s(&args, dev, "%s", name);
895 	if (error != 0) {
896 		mtx_destroy(&tp->tun_mtx);
897 		cv_destroy(&tp->tun_cv);
898 		free(tp, M_TUN);
899 		return (error);
900 	}
901 
902 	KASSERT((*dev)->si_drv1 != NULL,
903 	    ("Failed to set si_drv1 at %s creation", name));
904 	tp->tun_dev = *dev;
905 	knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
906 	mtx_lock(&tunmtx);
907 	TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
908 	mtx_unlock(&tunmtx);
909 	return (0);
910 }
911 
912 static void
tunstart(struct ifnet * ifp)913 tunstart(struct ifnet *ifp)
914 {
915 	struct tuntap_softc *tp = ifp->if_softc;
916 	struct mbuf *m;
917 
918 	TUNDEBUG(ifp, "starting\n");
919 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
920 		IFQ_LOCK(&ifp->if_snd);
921 		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
922 		if (m == NULL) {
923 			IFQ_UNLOCK(&ifp->if_snd);
924 			return;
925 		}
926 		IFQ_UNLOCK(&ifp->if_snd);
927 	}
928 
929 	TUN_LOCK(tp);
930 	if (tp->tun_flags & TUN_RWAIT) {
931 		tp->tun_flags &= ~TUN_RWAIT;
932 		wakeup(tp);
933 	}
934 	selwakeuppri(&tp->tun_rsel, PZERO);
935 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
936 	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
937 		TUN_UNLOCK(tp);
938 		pgsigio(&tp->tun_sigio, SIGIO, 0);
939 	} else
940 		TUN_UNLOCK(tp);
941 }
942 
943 /*
944  * tunstart_l2
945  *
946  * queue packets from higher level ready to put out
947  */
948 static void
tunstart_l2(struct ifnet * ifp)949 tunstart_l2(struct ifnet *ifp)
950 {
951 	struct tuntap_softc	*tp = ifp->if_softc;
952 
953 	TUNDEBUG(ifp, "starting\n");
954 
955 	/*
956 	 * do not junk pending output if we are in VMnet mode.
957 	 * XXX: can this do any harm because of queue overflow?
958 	 */
959 
960 	TUN_LOCK(tp);
961 	if (((tp->tun_flags & TUN_VMNET) == 0) &&
962 	    ((tp->tun_flags & TUN_READY) != TUN_READY)) {
963 		struct mbuf *m;
964 
965 		/* Unlocked read. */
966 		TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
967 
968 		for (;;) {
969 			IF_DEQUEUE(&ifp->if_snd, m);
970 			if (m != NULL) {
971 				m_freem(m);
972 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
973 			} else
974 				break;
975 		}
976 		TUN_UNLOCK(tp);
977 
978 		return;
979 	}
980 
981 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
982 
983 	if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
984 		if (tp->tun_flags & TUN_RWAIT) {
985 			tp->tun_flags &= ~TUN_RWAIT;
986 			wakeup(tp);
987 		}
988 
989 		if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
990 			TUN_UNLOCK(tp);
991 			pgsigio(&tp->tun_sigio, SIGIO, 0);
992 			TUN_LOCK(tp);
993 		}
994 
995 		selwakeuppri(&tp->tun_rsel, PZERO);
996 		KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
997 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
998 	}
999 
1000 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1001 	TUN_UNLOCK(tp);
1002 } /* tunstart_l2 */
1003 
1004 static int
tap_transmit(struct ifnet * ifp,struct mbuf * m)1005 tap_transmit(struct ifnet *ifp, struct mbuf *m)
1006 {
1007 	int error;
1008 
1009 	BPF_MTAP(ifp, m);
1010 	IFQ_HANDOFF(ifp, m, error);
1011 	return (error);
1012 }
1013 
1014 static void
tuncreate(struct cdev * dev)1015 tuncreate(struct cdev *dev)
1016 {
1017 	struct tuntap_driver *drv;
1018 	struct tuntap_softc *tp;
1019 	struct ifnet *ifp;
1020 	struct ether_addr eaddr;
1021 	int iflags;
1022 	u_char type;
1023 
1024 	tp = dev->si_drv1;
1025 	KASSERT(tp != NULL,
1026 	    ("si_drv1 should have been initialized at creation"));
1027 
1028 	drv = tp->tun_drv;
1029 	iflags = IFF_MULTICAST;
1030 	if ((tp->tun_flags & TUN_L2) != 0) {
1031 		type = IFT_ETHER;
1032 		iflags |= IFF_BROADCAST | IFF_SIMPLEX;
1033 	} else {
1034 		type = IFT_PPP;
1035 		iflags |= IFF_POINTOPOINT;
1036 	}
1037 	ifp = tp->tun_ifp = if_alloc(type);
1038 	ifp->if_softc = tp;
1039 	if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
1040 	ifp->if_ioctl = tunifioctl;
1041 	ifp->if_flags = iflags;
1042 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
1043 	ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_MEXTPG;
1044 	if ((tp->tun_flags & TUN_L2) != 0)
1045 		ifp->if_capabilities |=
1046 		    IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO;
1047 	ifp->if_capenable |= IFCAP_LINKSTATE | IFCAP_MEXTPG;
1048 
1049 	if ((tp->tun_flags & TUN_L2) != 0) {
1050 		ifp->if_init = tunifinit;
1051 		ifp->if_start = tunstart_l2;
1052 		ifp->if_transmit = tap_transmit;
1053 		ifp->if_qflush = if_qflush;
1054 
1055 		ether_gen_addr(ifp, &eaddr);
1056 		ether_ifattach(ifp, eaddr.octet);
1057 	} else {
1058 		ifp->if_mtu = TUNMTU;
1059 		ifp->if_start = tunstart;
1060 		ifp->if_output = tunoutput;
1061 
1062 		ifp->if_snd.ifq_drv_maxlen = 0;
1063 		IFQ_SET_READY(&ifp->if_snd);
1064 
1065 		if_attach(ifp);
1066 		bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
1067 	}
1068 
1069 	TUN_LOCK(tp);
1070 	tp->tun_flags |= TUN_INITED;
1071 	TUN_UNLOCK(tp);
1072 
1073 	TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
1074 	    ifp->if_xname, dev2unit(dev));
1075 }
1076 
1077 static void
tunrename(void * arg __unused,struct ifnet * ifp)1078 tunrename(void *arg __unused, struct ifnet *ifp)
1079 {
1080 	struct tuntap_softc *tp;
1081 	int error;
1082 
1083 	if ((ifp->if_flags & IFF_RENAMING) == 0)
1084 		return;
1085 
1086 	if (tuntap_driver_from_ifnet(ifp) == NULL)
1087 		return;
1088 
1089 	/*
1090 	 * We need to grab the ioctl sx long enough to make sure the softc is
1091 	 * still there.  If it is, we can safely try to busy the tun device.
1092 	 * The busy may fail if the device is currently dying, in which case
1093 	 * we do nothing.  If it doesn't fail, the busy count stops the device
1094 	 * from dying until we've created the alias (that will then be
1095 	 * subsequently destroyed).
1096 	 */
1097 	sx_xlock(&tun_ioctl_sx);
1098 	tp = ifp->if_softc;
1099 	if (tp == NULL) {
1100 		sx_xunlock(&tun_ioctl_sx);
1101 		return;
1102 	}
1103 	error = tun_busy(tp);
1104 	sx_xunlock(&tun_ioctl_sx);
1105 	if (error != 0)
1106 		return;
1107 	if (tp->tun_alias != NULL) {
1108 		destroy_dev(tp->tun_alias);
1109 		tp->tun_alias = NULL;
1110 	}
1111 
1112 	if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1113 		goto out;
1114 
1115 	/*
1116 	 * Failure's ok, aliases are created on a best effort basis.  If a
1117 	 * tun user/consumer decides to rename the interface to conflict with
1118 	 * another device (non-ifnet) on the system, we will assume they know
1119 	 * what they are doing.  make_dev_alias_p won't touch tun_alias on
1120 	 * failure, so we use it but ignore the return value.
1121 	 */
1122 	make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1123 	    ifp->if_xname);
1124 out:
1125 	tun_unbusy(tp);
1126 }
1127 
1128 static int
tunopen(struct cdev * dev,int flag,int mode,struct thread * td)1129 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1130 {
1131 	struct epoch_tracker et;
1132 	struct ifnet	*ifp;
1133 	struct tuntap_softc *tp;
1134 	void		*p;
1135 	int error __diagused, tunflags;
1136 
1137 	/*
1138 	 * Transient tunnels do deferred destroy of the tun device but want
1139 	 * to immediately cleanup state, so they clobber si_drv1 to avoid a
1140 	 * use-after-free in case someone does happen to open it in the interim.
1141 	 * We avoid using NULL to be able to distinguish from an uninitialized
1142 	 * cdev.
1143 	 *
1144 	 * We use the net epoch here to let a concurrent tun_destroy() schedule
1145 	 * freeing our tuntap_softc, in case we entered here and loaded si_drv1
1146 	 * before it was swapped out.  If we managed to load this while it was
1147 	 * still a softc, then the concurrent tun_destroy() hasn't yet scheduled
1148 	 * it to be free- that will take place sometime after the epoch we just
1149 	 * entered, so we can safely use it.
1150 	 */
1151 	NET_EPOCH_ENTER(et);
1152 	p = atomic_load_ptr(&dev->si_drv1);
1153 	if (p == dev) {
1154 		NET_EPOCH_EXIT(et);
1155 		return (ENXIO);
1156 	}
1157 
1158 	tunflags = 0;
1159 	CURVNET_SET(TD_TO_VNET(td));
1160 	error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1161 	if (error != 0) {
1162 		CURVNET_RESTORE();
1163 		NET_EPOCH_EXIT(et);
1164 		return (error);	/* Shouldn't happen */
1165 	}
1166 
1167 	tp = p;
1168 	KASSERT(tp != NULL,
1169 	    ("si_drv1 should have been initialized at creation"));
1170 
1171 	TUN_LOCK(tp);
1172 	if ((tp->tun_flags & TUN_INITED) == 0) {
1173 		TUN_UNLOCK(tp);
1174 		CURVNET_RESTORE();
1175 		NET_EPOCH_EXIT(et);
1176 		return (ENXIO);
1177 	}
1178 	if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1179 		TUN_UNLOCK(tp);
1180 		CURVNET_RESTORE();
1181 		NET_EPOCH_EXIT(et);
1182 		return (EBUSY);
1183 	}
1184 
1185 	NET_EPOCH_EXIT(et);
1186 	error = tun_busy_locked(tp);
1187 	KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1188 	ifp = TUN2IFP(tp);
1189 
1190 	if ((tp->tun_flags & TUN_L2) != 0) {
1191 		bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1192 		    sizeof(tp->tun_ether.octet));
1193 
1194 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
1195 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1196 
1197 		if (tapuponopen)
1198 			ifp->if_flags |= IFF_UP;
1199 	}
1200 
1201 	tp->tun_pid = td->td_proc->p_pid;
1202 	tp->tun_flags |= TUN_OPEN;
1203 
1204 	if_link_state_change(ifp, LINK_STATE_UP);
1205 	TUNDEBUG(ifp, "open\n");
1206 	TUN_UNLOCK(tp);
1207 
1208 	/*
1209 	 * This can fail with either ENOENT or EBUSY.  This is in the middle of
1210 	 * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1211 	 * the only cdevpriv dtor being set will be tundtor and the softc being
1212 	 * passed is constant for a given cdev.  We ignore the possible error
1213 	 * because of this as either "unlikely" or "not actually a problem."
1214 	 */
1215 	(void)devfs_set_cdevpriv(tp, tundtor);
1216 	CURVNET_RESTORE();
1217 	return (0);
1218 }
1219 
1220 /*
1221  * tundtor - tear down the device - mark i/f down & delete
1222  * routing info
1223  */
1224 static void
tundtor(void * data)1225 tundtor(void *data)
1226 {
1227 	struct proc *p;
1228 	struct tuntap_softc *tp;
1229 	struct ifnet *ifp;
1230 	bool l2tun;
1231 
1232 	tp = data;
1233 	p = curproc;
1234 	ifp = TUN2IFP(tp);
1235 
1236 	TUN_LOCK(tp);
1237 
1238 	/*
1239 	 * Realistically, we can't be obstinate here.  This only means that the
1240 	 * tuntap device was closed out of order, and the last closer wasn't the
1241 	 * controller.  These are still good to know about, though, as software
1242 	 * should avoid multiple processes with a tuntap device open and
1243 	 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1244 	 * parent).
1245 	 */
1246 	if (p->p_pid != tp->tun_pid) {
1247 		log(LOG_INFO,
1248 		    "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1249 		    p->p_pid, p->p_comm, tp->tun_dev->si_name);
1250 	}
1251 
1252 	/*
1253 	 * junk all pending output
1254 	 */
1255 	CURVNET_SET(ifp->if_vnet);
1256 
1257 	l2tun = false;
1258 	if ((tp->tun_flags & TUN_L2) != 0) {
1259 		l2tun = true;
1260 		IF_DRAIN(&ifp->if_snd);
1261 	} else {
1262 		IFQ_PURGE(&ifp->if_snd);
1263 	}
1264 
1265 	/* For vmnet, we won't do most of the address/route bits */
1266 	if ((tp->tun_flags & TUN_VMNET) != 0 ||
1267 	    (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1268 		goto out;
1269 #if defined(INET) || defined(INET6)
1270 	if (l2tun && tp->tun_lro_ready) {
1271 		TUNDEBUG (ifp, "LRO disabled\n");
1272 		tcp_lro_free(&tp->tun_lro);
1273 		tp->tun_lro_ready = false;
1274 	}
1275 #endif
1276 	if (ifp->if_flags & IFF_UP) {
1277 		TUN_UNLOCK(tp);
1278 		if_down(ifp);
1279 		TUN_LOCK(tp);
1280 	}
1281 
1282 	/* Delete all addresses and routes which reference this interface. */
1283 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1284 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1285 		TUN_UNLOCK(tp);
1286 		if_purgeaddrs(ifp);
1287 		TUN_LOCK(tp);
1288 	}
1289 
1290 out:
1291 	if_link_state_change(ifp, LINK_STATE_DOWN);
1292 	CURVNET_RESTORE();
1293 
1294 	funsetown(&tp->tun_sigio);
1295 	selwakeuppri(&tp->tun_rsel, PZERO);
1296 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1297 	TUNDEBUG (ifp, "closed\n");
1298 	tp->tun_flags &= ~TUN_OPEN;
1299 	tp->tun_pid = 0;
1300 	tun_vnethdr_set(ifp, 0);
1301 
1302 	tun_unbusy_locked(tp);
1303 	if ((tp->tun_flags & TUN_TRANSIENT) != 0) {
1304 		struct if_clone *cloner;
1305 		int error __diagused;
1306 
1307 		/* Mark it busy so that nothing can re-open it. */
1308 		tp->tun_flags |= TUN_DYING;
1309 		TUN_UNLOCK(tp);
1310 
1311 		CURVNET_SET_QUIET(ifp->if_home_vnet);
1312 		cloner = tuntap_cloner_from_flags(tp->tun_flags);
1313 		CURVNET_RESTORE();
1314 
1315 		error = if_clone_destroyif(cloner, ifp);
1316 		MPASS(error == 0 || error == EINTR || error == ERESTART);
1317 		return;
1318 	}
1319 
1320 	TUN_UNLOCK(tp);
1321 }
1322 
1323 static void
tuninit(struct ifnet * ifp)1324 tuninit(struct ifnet *ifp)
1325 {
1326 	struct tuntap_softc *tp = ifp->if_softc;
1327 
1328 	TUNDEBUG(ifp, "tuninit\n");
1329 
1330 	TUN_LOCK(tp);
1331 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1332 	if ((tp->tun_flags & TUN_L2) == 0) {
1333 		ifp->if_flags |= IFF_UP;
1334 		getmicrotime(&ifp->if_lastchange);
1335 		TUN_UNLOCK(tp);
1336 	} else {
1337 #if defined(INET) || defined(INET6)
1338 		if (tcp_lro_init(&tp->tun_lro) == 0) {
1339 			TUNDEBUG(ifp, "LRO enabled\n");
1340 			tp->tun_lro.ifp = ifp;
1341 			tp->tun_lro_ready = true;
1342 		} else {
1343 			TUNDEBUG(ifp, "Could not enable LRO\n");
1344 			tp->tun_lro_ready = false;
1345 		}
1346 #endif
1347 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1348 		TUN_UNLOCK(tp);
1349 		/* attempt to start output */
1350 		tunstart_l2(ifp);
1351 	}
1352 
1353 }
1354 
1355 /*
1356  * Used only for l2 tunnel.
1357  */
1358 static void
tunifinit(void * xtp)1359 tunifinit(void *xtp)
1360 {
1361 	struct tuntap_softc *tp;
1362 
1363 	tp = (struct tuntap_softc *)xtp;
1364 	tuninit(tp->tun_ifp);
1365 }
1366 
1367 /*
1368  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1369  * current value of ifp->if_capenable.
1370  */
1371 static void
tun_caps_changed(struct ifnet * ifp)1372 tun_caps_changed(struct ifnet *ifp)
1373 {
1374 	uint64_t hwassist = 0;
1375 
1376 	TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1377 	if (ifp->if_capenable & IFCAP_TXCSUM)
1378 		hwassist |= CSUM_TCP | CSUM_UDP;
1379 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1380 		hwassist |= CSUM_TCP_IPV6
1381 		    | CSUM_UDP_IPV6;
1382 	if (ifp->if_capenable & IFCAP_TSO4)
1383 		hwassist |= CSUM_IP_TSO;
1384 	if (ifp->if_capenable & IFCAP_TSO6)
1385 		hwassist |= CSUM_IP6_TSO;
1386 	ifp->if_hwassist = hwassist;
1387 }
1388 
1389 /*
1390  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1391  * if_capabilities and if_capenable as needed.
1392  */
1393 static void
tun_vnethdr_set(struct ifnet * ifp,int vhdrlen)1394 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1395 {
1396 	struct tuntap_softc *tp = ifp->if_softc;
1397 
1398 	TUN_LOCK_ASSERT(tp);
1399 
1400 	if (tp->tun_vhdrlen == vhdrlen)
1401 		return;
1402 
1403 	/*
1404 	 * Update if_capabilities to reflect the
1405 	 * functionalities offered by the virtio-net
1406 	 * header.
1407 	 */
1408 	if (vhdrlen != 0)
1409 		ifp->if_capabilities |=
1410 			TAP_VNET_HDR_CAPS;
1411 	else
1412 		ifp->if_capabilities &=
1413 			~TAP_VNET_HDR_CAPS;
1414 	/*
1415 	 * Disable any capabilities that we don't
1416 	 * support anymore.
1417 	 */
1418 	ifp->if_capenable &= ifp->if_capabilities;
1419 	tun_caps_changed(ifp);
1420 	tp->tun_vhdrlen = vhdrlen;
1421 
1422 	TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1423 	    vhdrlen, ifp->if_capabilities);
1424 }
1425 
1426 /*
1427  * Process an ioctl request.
1428  */
1429 static int
tunifioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1430 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1431 {
1432 	struct ifreq *ifr = (struct ifreq *)data;
1433 	struct tuntap_softc *tp;
1434 	struct ifstat *ifs;
1435 	struct ifmediareq	*ifmr;
1436 	int		dummy, error = 0;
1437 	bool		l2tun;
1438 
1439 	ifmr = NULL;
1440 	sx_xlock(&tun_ioctl_sx);
1441 	tp = ifp->if_softc;
1442 	if (tp == NULL) {
1443 		error = ENXIO;
1444 		goto bad;
1445 	}
1446 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1447 	switch(cmd) {
1448 	case SIOCGIFSTATUS:
1449 		ifs = (struct ifstat *)data;
1450 		TUN_LOCK(tp);
1451 		if (tp->tun_pid)
1452 			snprintf(ifs->ascii, sizeof(ifs->ascii),
1453 			    "\tOpened by PID %d\n", tp->tun_pid);
1454 		else
1455 			ifs->ascii[0] = '\0';
1456 		TUN_UNLOCK(tp);
1457 		break;
1458 	case SIOCSIFADDR:
1459 		if (l2tun)
1460 			error = ether_ioctl(ifp, cmd, data);
1461 		else
1462 			tuninit(ifp);
1463 		if (error == 0)
1464 		    TUNDEBUG(ifp, "address set\n");
1465 		break;
1466 	case SIOCSIFMTU:
1467 		ifp->if_mtu = ifr->ifr_mtu;
1468 		TUNDEBUG(ifp, "mtu set\n");
1469 		break;
1470 	case SIOCSIFFLAGS:
1471 	case SIOCADDMULTI:
1472 	case SIOCDELMULTI:
1473 		break;
1474 	case SIOCGIFMEDIA:
1475 		if (!l2tun) {
1476 			error = EINVAL;
1477 			break;
1478 		}
1479 
1480 		ifmr = (struct ifmediareq *)data;
1481 		dummy = ifmr->ifm_count;
1482 		ifmr->ifm_count = 1;
1483 		ifmr->ifm_status = IFM_AVALID;
1484 		ifmr->ifm_active = IFM_ETHER | IFM_FDX | IFM_1000_T;
1485 		if (tp->tun_flags & TUN_OPEN)
1486 			ifmr->ifm_status |= IFM_ACTIVE;
1487 		ifmr->ifm_current = ifmr->ifm_active;
1488 		if (dummy >= 1) {
1489 			int media = IFM_ETHER;
1490 			error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1491 		}
1492 		break;
1493 	case SIOCSIFCAP:
1494 		TUN_LOCK(tp);
1495 		ifp->if_capenable = ifr->ifr_reqcap;
1496 		tun_caps_changed(ifp);
1497 		TUN_UNLOCK(tp);
1498 		VLAN_CAPABILITIES(ifp);
1499 		break;
1500 	default:
1501 		if (l2tun) {
1502 			error = ether_ioctl(ifp, cmd, data);
1503 		} else {
1504 			error = EINVAL;
1505 		}
1506 	}
1507 bad:
1508 	sx_xunlock(&tun_ioctl_sx);
1509 	return (error);
1510 }
1511 
1512 /*
1513  * tunoutput - queue packets from higher level ready to put out.
1514  */
1515 static int
tunoutput(struct ifnet * ifp,struct mbuf * m0,const struct sockaddr * dst,struct route * ro)1516 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1517     struct route *ro)
1518 {
1519 	struct tuntap_softc *tp = ifp->if_softc;
1520 	u_short cached_tun_flags;
1521 	int error;
1522 	u_int32_t af;
1523 
1524 	TUNDEBUG (ifp, "tunoutput\n");
1525 
1526 #ifdef MAC
1527 	error = mac_ifnet_check_transmit(ifp, m0);
1528 	if (error) {
1529 		m_freem(m0);
1530 		return (error);
1531 	}
1532 #endif
1533 
1534 	/* Could be unlocked read? */
1535 	TUN_LOCK(tp);
1536 	cached_tun_flags = tp->tun_flags;
1537 	TUN_UNLOCK(tp);
1538 	if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1539 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1540 		m_freem (m0);
1541 		return (EHOSTDOWN);
1542 	}
1543 
1544 	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1545 		m_freem (m0);
1546 		return (EHOSTDOWN);
1547 	}
1548 
1549 	/* BPF writes need to be handled specially. */
1550 	if (dst->sa_family == AF_UNSPEC || dst->sa_family == pseudo_AF_HDRCMPLT)
1551 		bcopy(dst->sa_data, &af, sizeof(af));
1552 	else
1553 		af = RO_GET_FAMILY(ro, dst);
1554 
1555 	BPF_MTAP2(ifp, &af, sizeof(af), m0);
1556 
1557 	/* prepend sockaddr? this may abort if the mbuf allocation fails */
1558 	if (cached_tun_flags & TUN_LMODE) {
1559 		/* allocate space for sockaddr */
1560 		M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1561 
1562 		/* if allocation failed drop packet */
1563 		if (m0 == NULL) {
1564 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1565 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1566 			return (ENOBUFS);
1567 		} else {
1568 			bcopy(dst, m0->m_data, dst->sa_len);
1569 		}
1570 	}
1571 
1572 	if (cached_tun_flags & TUN_IFHEAD) {
1573 		/* Prepend the address family */
1574 		M_PREPEND(m0, 4, M_NOWAIT);
1575 
1576 		/* if allocation failed drop packet */
1577 		if (m0 == NULL) {
1578 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1579 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1580 			return (ENOBUFS);
1581 		} else
1582 			*(u_int32_t *)m0->m_data = htonl(af);
1583 	} else {
1584 #ifdef INET
1585 		if (af != AF_INET)
1586 #endif
1587 		{
1588 			m_freem(m0);
1589 			return (EAFNOSUPPORT);
1590 		}
1591 	}
1592 
1593 	error = (ifp->if_transmit)(ifp, m0);
1594 	if (error)
1595 		return (ENOBUFS);
1596 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1597 	return (0);
1598 }
1599 
1600 /*
1601  * the cdevsw interface is now pretty minimal.
1602  */
1603 static	int
tunioctl(struct cdev * dev,u_long cmd,caddr_t data,int flag,struct thread * td)1604 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1605     struct thread *td)
1606 {
1607 	struct ifreq ifr, *ifrp;
1608 	struct tuntap_softc *tp = dev->si_drv1;
1609 	struct ifnet *ifp = TUN2IFP(tp);
1610 	struct tuninfo *tunp;
1611 	int error, iflags, ival;
1612 	bool	l2tun;
1613 
1614 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1615 	if (l2tun) {
1616 		/* tap specific ioctls */
1617 		switch(cmd) {
1618 		/* VMware/VMnet port ioctl's */
1619 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1620     defined(COMPAT_FREEBSD4)
1621 		case _IO('V', 0):
1622 			ival = IOCPARM_IVAL(data);
1623 			data = (caddr_t)&ival;
1624 			/* FALLTHROUGH */
1625 #endif
1626 		case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1627 			iflags = *(int *)data;
1628 			iflags &= TUN_VMIO_FLAG_MASK;
1629 			iflags &= ~IFF_CANTCHANGE;
1630 			iflags |= IFF_UP;
1631 
1632 			TUN_LOCK(tp);
1633 			ifp->if_flags = iflags |
1634 			    (ifp->if_flags & IFF_CANTCHANGE);
1635 			TUN_UNLOCK(tp);
1636 
1637 			return (0);
1638 		case SIOCGIFADDR:	/* get MAC address of the remote side */
1639 			TUN_LOCK(tp);
1640 			bcopy(&tp->tun_ether.octet, data,
1641 			    sizeof(tp->tun_ether.octet));
1642 			TUN_UNLOCK(tp);
1643 
1644 			return (0);
1645 		case SIOCSIFADDR:	/* set MAC address of the remote side */
1646 			TUN_LOCK(tp);
1647 			bcopy(data, &tp->tun_ether.octet,
1648 			    sizeof(tp->tun_ether.octet));
1649 			TUN_UNLOCK(tp);
1650 
1651 			return (0);
1652 		case TAPSVNETHDR:
1653 			ival = *(int *)data;
1654 			if (ival != 0 &&
1655 			    ival != sizeof(struct virtio_net_hdr) &&
1656 			    ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1657 				return (EINVAL);
1658 			}
1659 			TUN_LOCK(tp);
1660 			tun_vnethdr_set(ifp, ival);
1661 			TUN_UNLOCK(tp);
1662 
1663 			return (0);
1664 		case TAPGVNETHDR:
1665 			TUN_LOCK(tp);
1666 			*(int *)data = tp->tun_vhdrlen;
1667 			TUN_UNLOCK(tp);
1668 
1669 			return (0);
1670 		}
1671 
1672 		/* Fall through to the common ioctls if unhandled */
1673 	} else {
1674 		switch (cmd) {
1675 		case TUNSLMODE:
1676 			TUN_LOCK(tp);
1677 			if (*(int *)data) {
1678 				tp->tun_flags |= TUN_LMODE;
1679 				tp->tun_flags &= ~TUN_IFHEAD;
1680 			} else
1681 				tp->tun_flags &= ~TUN_LMODE;
1682 			TUN_UNLOCK(tp);
1683 
1684 			return (0);
1685 		case TUNSIFHEAD:
1686 			TUN_LOCK(tp);
1687 			if (*(int *)data) {
1688 				tp->tun_flags |= TUN_IFHEAD;
1689 				tp->tun_flags &= ~TUN_LMODE;
1690 			} else
1691 				tp->tun_flags &= ~TUN_IFHEAD;
1692 			TUN_UNLOCK(tp);
1693 
1694 			return (0);
1695 		case TUNGIFHEAD:
1696 			TUN_LOCK(tp);
1697 			*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1698 			TUN_UNLOCK(tp);
1699 
1700 			return (0);
1701 		case TUNSIFMODE:
1702 			/* deny this if UP */
1703 			if (TUN2IFP(tp)->if_flags & IFF_UP)
1704 				return (EBUSY);
1705 
1706 			switch (*(int *)data & ~IFF_MULTICAST) {
1707 			case IFF_POINTOPOINT:
1708 			case IFF_BROADCAST:
1709 				TUN_LOCK(tp);
1710 				TUN2IFP(tp)->if_flags &=
1711 				    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1712 				TUN2IFP(tp)->if_flags |= *(int *)data;
1713 				TUN_UNLOCK(tp);
1714 
1715 				break;
1716 			default:
1717 				return (EINVAL);
1718 			}
1719 
1720 			return (0);
1721 		case TUNSIFPID:
1722 			TUN_LOCK(tp);
1723 			tp->tun_pid = curthread->td_proc->p_pid;
1724 			TUN_UNLOCK(tp);
1725 
1726 			return (0);
1727 		}
1728 		/* Fall through to the common ioctls if unhandled */
1729 	}
1730 
1731 	switch (cmd) {
1732 	case TUNGIFNAME:
1733 		ifrp = (struct ifreq *)data;
1734 		strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1735 
1736 		return (0);
1737 	case TUNSIFINFO:
1738 		tunp = (struct tuninfo *)data;
1739 		if (TUN2IFP(tp)->if_type != tunp->type)
1740 			return (EPROTOTYPE);
1741 		TUN_LOCK(tp);
1742 		if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1743 			strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1744 			ifr.ifr_mtu = tunp->mtu;
1745 			CURVNET_SET(TUN2IFP(tp)->if_vnet);
1746 			error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1747 			    (caddr_t)&ifr, td);
1748 			CURVNET_RESTORE();
1749 			if (error) {
1750 				TUN_UNLOCK(tp);
1751 				return (error);
1752 			}
1753 		}
1754 		TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1755 		TUN_UNLOCK(tp);
1756 		break;
1757 	case TUNGIFINFO:
1758 		tunp = (struct tuninfo *)data;
1759 		TUN_LOCK(tp);
1760 		tunp->mtu = TUN2IFP(tp)->if_mtu;
1761 		tunp->type = TUN2IFP(tp)->if_type;
1762 		tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1763 		TUN_UNLOCK(tp);
1764 		break;
1765 	case TUNSDEBUG:
1766 		tundebug = *(int *)data;
1767 		break;
1768 	case TUNGDEBUG:
1769 		*(int *)data = tundebug;
1770 		break;
1771 	case TUNSTRANSIENT:
1772 		TUN_LOCK(tp);
1773 		if (*(int *)data)
1774 			tp->tun_flags |= TUN_TRANSIENT;
1775 		else
1776 			tp->tun_flags &= ~TUN_TRANSIENT;
1777 		TUN_UNLOCK(tp);
1778 		break;
1779 	case TUNGTRANSIENT:
1780 		TUN_LOCK(tp);
1781 		*(int *)data = (tp->tun_flags & TUN_TRANSIENT) != 0;
1782 		TUN_UNLOCK(tp);
1783 		break;
1784 	case FIONBIO:
1785 		break;
1786 	case FIOASYNC:
1787 		TUN_LOCK(tp);
1788 		if (*(int *)data)
1789 			tp->tun_flags |= TUN_ASYNC;
1790 		else
1791 			tp->tun_flags &= ~TUN_ASYNC;
1792 		TUN_UNLOCK(tp);
1793 		break;
1794 	case FIONREAD:
1795 		if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1796 			struct mbuf *mb;
1797 			IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1798 			IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1799 			for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1800 				*(int *)data += mb->m_len;
1801 			IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1802 		} else
1803 			*(int *)data = 0;
1804 		break;
1805 	case FIOSETOWN:
1806 		return (fsetown(*(int *)data, &tp->tun_sigio));
1807 
1808 	case FIOGETOWN:
1809 		*(int *)data = fgetown(&tp->tun_sigio);
1810 		return (0);
1811 
1812 	/* This is deprecated, FIOSETOWN should be used instead. */
1813 	case TIOCSPGRP:
1814 		return (fsetown(-(*(int *)data), &tp->tun_sigio));
1815 
1816 	/* This is deprecated, FIOGETOWN should be used instead. */
1817 	case TIOCGPGRP:
1818 		*(int *)data = -fgetown(&tp->tun_sigio);
1819 		return (0);
1820 
1821 	default:
1822 		return (ENOTTY);
1823 	}
1824 	return (0);
1825 }
1826 
1827 /*
1828  * The cdevsw read interface - reads a packet at a time, or at
1829  * least as much of a packet as can be read.
1830  */
1831 static	int
tunread(struct cdev * dev,struct uio * uio,int flag)1832 tunread(struct cdev *dev, struct uio *uio, int flag)
1833 {
1834 	struct tuntap_softc *tp = dev->si_drv1;
1835 	struct ifnet	*ifp = TUN2IFP(tp);
1836 	struct mbuf	*m;
1837 	size_t		len;
1838 	int		error = 0;
1839 
1840 	TUNDEBUG (ifp, "read\n");
1841 	TUN_LOCK(tp);
1842 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1843 		TUN_UNLOCK(tp);
1844 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1845 		return (EHOSTDOWN);
1846 	}
1847 
1848 	tp->tun_flags &= ~TUN_RWAIT;
1849 
1850 	for (;;) {
1851 		IFQ_DEQUEUE(&ifp->if_snd, m);
1852 		if (m != NULL)
1853 			break;
1854 		if (flag & O_NONBLOCK) {
1855 			TUN_UNLOCK(tp);
1856 			return (EWOULDBLOCK);
1857 		}
1858 		tp->tun_flags |= TUN_RWAIT;
1859 		error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | PZERO,
1860 		    "tunread", 0);
1861 		if (error != 0) {
1862 			TUN_UNLOCK(tp);
1863 			return (error);
1864 		}
1865 	}
1866 	TUN_UNLOCK(tp);
1867 
1868 	len = min(tp->tun_vhdrlen, uio->uio_resid);
1869 	if (len > 0) {
1870 		struct virtio_net_hdr_mrg_rxbuf vhdr;
1871 
1872 		bzero(&vhdr, sizeof(vhdr));
1873 		if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1874 			m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1875 		}
1876 
1877 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1878 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1879 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1880 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1881 		    vhdr.hdr.csum_offset);
1882 		error = uiomove(&vhdr, len, uio);
1883 	}
1884 	if (error == 0)
1885 		error = m_mbuftouio(uio, m, 0);
1886 	m_freem(m);
1887 	return (error);
1888 }
1889 
1890 static int
tunwrite_l2(struct tuntap_softc * tp,struct mbuf * m,struct virtio_net_hdr_mrg_rxbuf * vhdr)1891 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1892 	    struct virtio_net_hdr_mrg_rxbuf *vhdr)
1893 {
1894 	struct epoch_tracker et;
1895 	struct ether_header *eh;
1896 	struct ifnet *ifp;
1897 
1898 	ifp = TUN2IFP(tp);
1899 
1900 	/*
1901 	 * Only pass a unicast frame to ether_input(), if it would
1902 	 * actually have been received by non-virtual hardware.
1903 	 */
1904 	if (m->m_len < sizeof(struct ether_header)) {
1905 		m_freem(m);
1906 		return (0);
1907 	}
1908 
1909 	eh = mtod(m, struct ether_header *);
1910 
1911 	if ((ifp->if_flags & IFF_PROMISC) == 0 &&
1912 	    !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1913 	    bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1914 		m_freem(m);
1915 		return (0);
1916 	}
1917 
1918 	if (vhdr != NULL) {
1919 		if (virtio_net_rx_csum(m, &vhdr->hdr)) {
1920 			m_freem(m);
1921 			return (0);
1922 		}
1923 	} else {
1924 		switch (ntohs(eh->ether_type)) {
1925 #ifdef INET
1926 		case ETHERTYPE_IP:
1927 			if (ifp->if_capenable & IFCAP_RXCSUM) {
1928 				m->m_pkthdr.csum_flags |=
1929 				    CSUM_IP_CHECKED | CSUM_IP_VALID |
1930 				    CSUM_DATA_VALID | CSUM_SCTP_VALID |
1931 				    CSUM_PSEUDO_HDR;
1932 				m->m_pkthdr.csum_data = 0xffff;
1933 			}
1934 			break;
1935 #endif
1936 #ifdef INET6
1937 		case ETHERTYPE_IPV6:
1938 			if (ifp->if_capenable & IFCAP_RXCSUM_IPV6) {
1939 				m->m_pkthdr.csum_flags |=
1940 				    CSUM_DATA_VALID_IPV6 | CSUM_SCTP_VALID |
1941 				    CSUM_PSEUDO_HDR;
1942 				m->m_pkthdr.csum_data = 0xffff;
1943 			}
1944 			break;
1945 #endif
1946 		}
1947 	}
1948 
1949 	/* Pass packet up to parent. */
1950 	CURVNET_SET(ifp->if_vnet);
1951 	NET_EPOCH_ENTER(et);
1952 #if defined(INET) || defined(INET6)
1953 	if (tp->tun_lro_ready && ifp->if_capenable & IFCAP_LRO &&
1954 	    tcp_lro_rx(&tp->tun_lro, m, 0) == 0)
1955 		tcp_lro_flush_all(&tp->tun_lro);
1956 	else
1957 #endif
1958 		(*ifp->if_input)(ifp, m);
1959 	NET_EPOCH_EXIT(et);
1960 	CURVNET_RESTORE();
1961 	/* ibytes are counted in parent */
1962 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1963 	return (0);
1964 }
1965 
1966 static int
tunwrite_l3(struct tuntap_softc * tp,struct mbuf * m)1967 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1968 {
1969 	struct epoch_tracker et;
1970 	struct ifnet *ifp;
1971 	int family, isr;
1972 
1973 	ifp = TUN2IFP(tp);
1974 	/* Could be unlocked read? */
1975 	TUN_LOCK(tp);
1976 	if (tp->tun_flags & TUN_IFHEAD) {
1977 		TUN_UNLOCK(tp);
1978 		if (m->m_len < sizeof(family) &&
1979 		(m = m_pullup(m, sizeof(family))) == NULL)
1980 			return (ENOBUFS);
1981 		family = ntohl(*mtod(m, u_int32_t *));
1982 		m_adj(m, sizeof(family));
1983 	} else {
1984 		TUN_UNLOCK(tp);
1985 		family = AF_INET;
1986 	}
1987 
1988 	BPF_MTAP2(ifp, &family, sizeof(family), m);
1989 
1990 	switch (family) {
1991 #ifdef INET
1992 	case AF_INET:
1993 		isr = NETISR_IP;
1994 		break;
1995 #endif
1996 #ifdef INET6
1997 	case AF_INET6:
1998 		isr = NETISR_IPV6;
1999 		break;
2000 #endif
2001 	default:
2002 		m_freem(m);
2003 		return (EAFNOSUPPORT);
2004 	}
2005 	random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
2006 	if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
2007 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
2008 	CURVNET_SET(ifp->if_vnet);
2009 	M_SETFIB(m, ifp->if_fib);
2010 	NET_EPOCH_ENTER(et);
2011 	netisr_dispatch(isr, m);
2012 	NET_EPOCH_EXIT(et);
2013 	CURVNET_RESTORE();
2014 	return (0);
2015 }
2016 
2017 /*
2018  * the cdevsw write interface - an atomic write is a packet - or else!
2019  */
2020 static	int
tunwrite(struct cdev * dev,struct uio * uio,int flag)2021 tunwrite(struct cdev *dev, struct uio *uio, int flag)
2022 {
2023 	struct virtio_net_hdr_mrg_rxbuf vhdr;
2024 	struct tuntap_softc *tp;
2025 	struct ifnet	*ifp;
2026 	struct mbuf	*m;
2027 	uint32_t	mru;
2028 	int		align, vhdrlen, error;
2029 	bool		l2tun;
2030 
2031 	tp = dev->si_drv1;
2032 	ifp = TUN2IFP(tp);
2033 	TUNDEBUG(ifp, "tunwrite\n");
2034 	if ((ifp->if_flags & IFF_UP) != IFF_UP)
2035 		/* ignore silently */
2036 		return (0);
2037 
2038 	if (uio->uio_resid == 0)
2039 		return (0);
2040 
2041 	l2tun = (tp->tun_flags & TUN_L2) != 0;
2042 	mru = l2tun ? TAPMRU : TUNMRU;
2043 	vhdrlen = tp->tun_vhdrlen;
2044 	align = 0;
2045 	if (l2tun) {
2046 		align = ETHER_ALIGN;
2047 		mru += vhdrlen;
2048 	} else if ((tp->tun_flags & TUN_IFHEAD) != 0)
2049 		mru += sizeof(uint32_t);	/* family */
2050 	if (uio->uio_resid < 0 || uio->uio_resid > mru) {
2051 		TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
2052 		return (EIO);
2053 	}
2054 
2055 	if (vhdrlen > 0) {
2056 		error = uiomove(&vhdr, vhdrlen, uio);
2057 		if (error != 0)
2058 			return (error);
2059 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
2060 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
2061 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
2062 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
2063 		    vhdr.hdr.csum_offset);
2064 	}
2065 
2066 	if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
2067 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2068 		return (ENOBUFS);
2069 	}
2070 
2071 	m->m_pkthdr.rcvif = ifp;
2072 #ifdef MAC
2073 	mac_ifnet_create_mbuf(ifp, m);
2074 #endif
2075 
2076 	if (l2tun)
2077 		return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
2078 
2079 	return (tunwrite_l3(tp, m));
2080 }
2081 
2082 /*
2083  * tunpoll - the poll interface, this is only useful on reads
2084  * really. The write detect always returns true, write never blocks
2085  * anyway, it either accepts the packet or drops it.
2086  */
2087 static	int
tunpoll(struct cdev * dev,int events,struct thread * td)2088 tunpoll(struct cdev *dev, int events, struct thread *td)
2089 {
2090 	struct tuntap_softc *tp = dev->si_drv1;
2091 	struct ifnet	*ifp = TUN2IFP(tp);
2092 	int		revents = 0;
2093 
2094 	TUNDEBUG(ifp, "tunpoll\n");
2095 
2096 	if (events & (POLLIN | POLLRDNORM)) {
2097 		IFQ_LOCK(&ifp->if_snd);
2098 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
2099 			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
2100 			revents |= events & (POLLIN | POLLRDNORM);
2101 		} else {
2102 			TUNDEBUG(ifp, "tunpoll waiting\n");
2103 			selrecord(td, &tp->tun_rsel);
2104 		}
2105 		IFQ_UNLOCK(&ifp->if_snd);
2106 	}
2107 	revents |= events & (POLLOUT | POLLWRNORM);
2108 
2109 	return (revents);
2110 }
2111 
2112 /*
2113  * tunkqfilter - support for the kevent() system call.
2114  */
2115 static int
tunkqfilter(struct cdev * dev,struct knote * kn)2116 tunkqfilter(struct cdev *dev, struct knote *kn)
2117 {
2118 	struct tuntap_softc	*tp = dev->si_drv1;
2119 	struct ifnet	*ifp = TUN2IFP(tp);
2120 
2121 	switch(kn->kn_filter) {
2122 	case EVFILT_READ:
2123 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
2124 		    ifp->if_xname, dev2unit(dev));
2125 		kn->kn_fop = &tun_read_filterops;
2126 		break;
2127 
2128 	case EVFILT_WRITE:
2129 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
2130 		    ifp->if_xname, dev2unit(dev));
2131 		kn->kn_fop = &tun_write_filterops;
2132 		break;
2133 
2134 	default:
2135 		TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
2136 		    ifp->if_xname, dev2unit(dev));
2137 		return(EINVAL);
2138 	}
2139 
2140 	kn->kn_hook = tp;
2141 	knlist_add(&tp->tun_rsel.si_note, kn, 0);
2142 
2143 	return (0);
2144 }
2145 
2146 /*
2147  * Return true of there is data in the interface queue.
2148  */
2149 static int
tunkqread(struct knote * kn,long hint)2150 tunkqread(struct knote *kn, long hint)
2151 {
2152 	int			ret;
2153 	struct tuntap_softc	*tp = kn->kn_hook;
2154 	struct cdev		*dev = tp->tun_dev;
2155 	struct ifnet	*ifp = TUN2IFP(tp);
2156 
2157 	if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
2158 		TUNDEBUG(ifp,
2159 		    "%s have data in the queue.  Len = %d, minor = %#x\n",
2160 		    ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
2161 		ret = 1;
2162 	} else {
2163 		TUNDEBUG(ifp,
2164 		    "%s waiting for data, minor = %#x\n", ifp->if_xname,
2165 		    dev2unit(dev));
2166 		ret = 0;
2167 	}
2168 
2169 	return (ret);
2170 }
2171 
2172 /*
2173  * Always can write, always return MTU in kn->data.
2174  */
2175 static int
tunkqwrite(struct knote * kn,long hint)2176 tunkqwrite(struct knote *kn, long hint)
2177 {
2178 	struct tuntap_softc	*tp = kn->kn_hook;
2179 	struct ifnet	*ifp = TUN2IFP(tp);
2180 
2181 	kn->kn_data = ifp->if_mtu;
2182 
2183 	return (1);
2184 }
2185 
2186 static void
tunkqdetach(struct knote * kn)2187 tunkqdetach(struct knote *kn)
2188 {
2189 	struct tuntap_softc	*tp = kn->kn_hook;
2190 
2191 	knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2192 }
2193