xref: /freebsd/sys/net/if_tuntap.c (revision 36138969847b231cd98f48272e2bdf88a8dc08dd)
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 ifnet		*tun_ifp;	/* the interface */
142 	struct sigio		*tun_sigio;	/* async I/O info */
143 	struct tuntap_driver	*tun_drv;	/* appropriate driver */
144 	struct selinfo		 tun_rsel;	/* read select */
145 	struct mtx		 tun_mtx;	/* softc field mutex */
146 	struct cv		 tun_cv;	/* for ref'd dev destroy */
147 	struct ether_addr	 tun_ether;	/* remote address */
148 	int			 tun_busy;	/* busy count */
149 	int			 tun_vhdrlen;	/* virtio-net header length */
150 	struct lro_ctrl		 tun_lro;	/* for TCP LRO */
151 	bool			 tun_lro_ready;	/* TCP LRO initialized */
152 };
153 #define	TUN2IFP(sc)	((sc)->tun_ifp)
154 
155 #define	TUNDEBUG	if (tundebug) if_printf
156 
157 #define	TUN_LOCK(tp)		mtx_lock(&(tp)->tun_mtx)
158 #define	TUN_UNLOCK(tp)		mtx_unlock(&(tp)->tun_mtx)
159 #define	TUN_LOCK_ASSERT(tp)	mtx_assert(&(tp)->tun_mtx, MA_OWNED);
160 
161 #define	TUN_VMIO_FLAG_MASK	0x0fff
162 
163 /*
164  * Interface capabilities of a tap device that supports the virtio-net
165  * header.
166  */
167 #define TAP_VNET_HDR_CAPS	(IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6	\
168 				| IFCAP_VLAN_HWCSUM			\
169 				| IFCAP_TSO | IFCAP_LRO			\
170 				| IFCAP_VLAN_HWTSO)
171 
172 #define TAP_ALL_OFFLOAD		(CSUM_TSO | CSUM_TCP | CSUM_UDP |\
173 				    CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
174 
175 /*
176  * All mutable global variables in if_tun are locked using tunmtx, with
177  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
178  * which are static after setup.
179  */
180 static struct mtx tunmtx;
181 static eventhandler_tag arrival_tag;
182 static eventhandler_tag clone_tag;
183 static int tuntap_osd_jail_slot;
184 static const char tunname[] = "tun";
185 static const char tapname[] = "tap";
186 static const char vmnetname[] = "vmnet";
187 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
188 static int tundebug = 0;
189 static int tundclone = 1;
190 static int tap_allow_uopen = 0;	/* allow user devfs cloning */
191 static int tapuponopen = 0;	/* IFF_UP on open() */
192 static int tapdclone = 1;	/* enable devfs cloning */
193 
194 static TAILQ_HEAD(,tuntap_softc)	tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
195 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
196 
197 static struct sx tun_ioctl_sx;
198 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
199 
200 SYSCTL_DECL(_net_link);
201 /* tun */
202 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
203     "IP tunnel software network interface");
204 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
205     "Enable legacy devfs interface creation");
206 
207 /* tap */
208 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
209     "Ethernet tunnel software network interface");
210 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
211     "Enable legacy devfs interface creation for all users");
212 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
213     "Bring interface up when /dev/tap is opened");
214 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
215     "Enable legacy devfs interface creation");
216 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
217 
218 static int	tun_create_device(struct tuntap_driver *drv, int unit,
219     struct ucred *cr, struct cdev **dev, const char *name);
220 static int	tun_busy_locked(struct tuntap_softc *tp);
221 static void	tun_unbusy_locked(struct tuntap_softc *tp);
222 static int	tun_busy(struct tuntap_softc *tp);
223 static void	tun_unbusy(struct tuntap_softc *tp);
224 
225 static int	tuntap_name2info(const char *name, int *unit, int *flags);
226 static void	tunclone(void *arg, struct ucred *cred, char *name,
227 		    int namelen, struct cdev **dev);
228 static void	tuncreate(struct cdev *dev);
229 static void	tundtor(void *data);
230 static void	tunrename(void *arg, struct ifnet *ifp);
231 static int	tunifioctl(struct ifnet *, u_long, caddr_t);
232 static void	tuninit(struct ifnet *);
233 static void	tunifinit(void *xtp);
234 static int	tuntapmodevent(module_t, int, void *);
235 static int	tunoutput(struct ifnet *, struct mbuf *,
236 		    const struct sockaddr *, struct route *ro);
237 static void	tunstart(struct ifnet *);
238 static void	tunstart_l2(struct ifnet *);
239 
240 static int	tun_clone_match(struct if_clone *ifc, const char *name);
241 static int	tap_clone_match(struct if_clone *ifc, const char *name);
242 static int	vmnet_clone_match(struct if_clone *ifc, const char *name);
243 static int	tun_clone_create(struct if_clone *, char *, size_t,
244 		    struct ifc_data *, struct ifnet **);
245 static int	tun_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
246 static void	tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
247 
248 static d_open_t		tunopen;
249 static d_read_t		tunread;
250 static d_write_t	tunwrite;
251 static d_ioctl_t	tunioctl;
252 static d_poll_t		tunpoll;
253 static d_kqfilter_t	tunkqfilter;
254 
255 static int		tunkqread(struct knote *, long);
256 static int		tunkqwrite(struct knote *, long);
257 static void		tunkqdetach(struct knote *);
258 
259 static const struct filterops tun_read_filterops = {
260 	.f_isfd =	1,
261 	.f_attach =	NULL,
262 	.f_detach =	tunkqdetach,
263 	.f_event =	tunkqread,
264 	.f_copy =	knote_triv_copy,
265 };
266 
267 static const struct filterops tun_write_filterops = {
268 	.f_isfd =	1,
269 	.f_attach =	NULL,
270 	.f_detach =	tunkqdetach,
271 	.f_event =	tunkqwrite,
272 	.f_copy =	knote_triv_copy,
273 };
274 
275 static struct tuntap_driver {
276 	struct cdevsw		 cdevsw;
277 	int			 ident_flags;
278 	struct unrhdr		*unrhdr;
279 	struct clonedevs	*clones;
280 	ifc_match_f		*clone_match_fn;
281 	ifc_create_f		*clone_create_fn;
282 	ifc_destroy_f		*clone_destroy_fn;
283 } tuntap_drivers[] = {
284 	{
285 		.ident_flags =	0,
286 		.cdevsw =	{
287 		    .d_version =	D_VERSION,
288 		    .d_flags =		D_NEEDMINOR,
289 		    .d_open =		tunopen,
290 		    .d_read =		tunread,
291 		    .d_write =		tunwrite,
292 		    .d_ioctl =		tunioctl,
293 		    .d_poll =		tunpoll,
294 		    .d_kqfilter =	tunkqfilter,
295 		    .d_name =		tunname,
296 		},
297 		.clone_match_fn =	tun_clone_match,
298 		.clone_create_fn =	tun_clone_create,
299 		.clone_destroy_fn =	tun_clone_destroy,
300 	},
301 	{
302 		.ident_flags =	TUN_L2,
303 		.cdevsw =	{
304 		    .d_version =	D_VERSION,
305 		    .d_flags =		D_NEEDMINOR,
306 		    .d_open =		tunopen,
307 		    .d_read =		tunread,
308 		    .d_write =		tunwrite,
309 		    .d_ioctl =		tunioctl,
310 		    .d_poll =		tunpoll,
311 		    .d_kqfilter =	tunkqfilter,
312 		    .d_name =		tapname,
313 		},
314 		.clone_match_fn =	tap_clone_match,
315 		.clone_create_fn =	tun_clone_create,
316 		.clone_destroy_fn =	tun_clone_destroy,
317 	},
318 	{
319 		.ident_flags =	TUN_L2 | TUN_VMNET,
320 		.cdevsw =	{
321 		    .d_version =	D_VERSION,
322 		    .d_flags =		D_NEEDMINOR,
323 		    .d_open =		tunopen,
324 		    .d_read =		tunread,
325 		    .d_write =		tunwrite,
326 		    .d_ioctl =		tunioctl,
327 		    .d_poll =		tunpoll,
328 		    .d_kqfilter =	tunkqfilter,
329 		    .d_name =		vmnetname,
330 		},
331 		.clone_match_fn =	vmnet_clone_match,
332 		.clone_create_fn =	tun_clone_create,
333 		.clone_destroy_fn =	tun_clone_destroy,
334 	},
335 };
336 #define	NDRV	nitems(tuntap_drivers)
337 
338 VNET_DEFINE_STATIC(struct if_clone *, tuntap_driver_cloners[NDRV]);
339 #define	V_tuntap_driver_cloners	VNET(tuntap_driver_cloners)
340 
341 /*
342  * Mechanism for marking a tunnel device as busy so that we can safely do some
343  * orthogonal operations (such as operations on devices) without racing against
344  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
345  * open, to be woken up when the condition is alleviated.
346  */
347 static int
tun_busy_locked(struct tuntap_softc * tp)348 tun_busy_locked(struct tuntap_softc *tp)
349 {
350 
351 	TUN_LOCK_ASSERT(tp);
352 	if ((tp->tun_flags & TUN_DYING) != 0) {
353 		/*
354 		 * Perhaps unintuitive, but the device is busy going away.
355 		 * Other interpretations of EBUSY from tun_busy make little
356 		 * sense, since making a busy device even more busy doesn't
357 		 * sound like a problem.
358 		 */
359 		return (EBUSY);
360 	}
361 
362 	++tp->tun_busy;
363 	return (0);
364 }
365 
366 static void
tun_unbusy_locked(struct tuntap_softc * tp)367 tun_unbusy_locked(struct tuntap_softc *tp)
368 {
369 
370 	TUN_LOCK_ASSERT(tp);
371 	KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
372 
373 	--tp->tun_busy;
374 	/* Wake up anything that may be waiting on our busy tunnel. */
375 	if (tp->tun_busy == 0)
376 		cv_broadcast(&tp->tun_cv);
377 }
378 
379 static int
tun_busy(struct tuntap_softc * tp)380 tun_busy(struct tuntap_softc *tp)
381 {
382 	int ret;
383 
384 	TUN_LOCK(tp);
385 	ret = tun_busy_locked(tp);
386 	TUN_UNLOCK(tp);
387 	return (ret);
388 }
389 
390 static void
tun_unbusy(struct tuntap_softc * tp)391 tun_unbusy(struct tuntap_softc *tp)
392 {
393 
394 	TUN_LOCK(tp);
395 	tun_unbusy_locked(tp);
396 	TUN_UNLOCK(tp);
397 }
398 
399 /*
400  * Sets unit and/or flags given the device name.  Must be called with correct
401  * vnet context.
402  */
403 static int
tuntap_name2info(const char * name,int * outunit,int * outflags)404 tuntap_name2info(const char *name, int *outunit, int *outflags)
405 {
406 	struct tuntap_driver *drv;
407 	char *dname;
408 	int flags, unit;
409 	bool found;
410 
411 	if (name == NULL)
412 		return (EINVAL);
413 
414 	/*
415 	 * Needed for dev_stdclone, but dev_stdclone will not modify, it just
416 	 * wants to be able to pass back a char * through the second param. We
417 	 * will always set that as NULL here, so we'll fake it.
418 	 */
419 	dname = __DECONST(char *, name);
420 	found = false;
421 
422 	for (u_int i = 0; i < NDRV; i++) {
423 		drv = &tuntap_drivers[i];
424 
425 		if (strcmp(name, drv->cdevsw.d_name) == 0) {
426 			found = true;
427 			unit = -1;
428 			flags = drv->ident_flags;
429 			break;
430 		}
431 
432 		if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
433 			found = true;
434 			flags = drv->ident_flags;
435 			break;
436 		}
437 	}
438 
439 	if (!found)
440 		return (ENXIO);
441 
442 	if (outunit != NULL)
443 		*outunit = unit;
444 	if (outflags != NULL)
445 		*outflags = flags;
446 	return (0);
447 }
448 
449 static struct if_clone *
tuntap_cloner_from_flags(int tun_flags)450 tuntap_cloner_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 (V_tuntap_driver_cloners[i]);
457 
458 	return (NULL);
459 }
460 
461 /*
462  * Get driver information from a set of flags specified.  Masks the identifying
463  * part of the flags and compares it against all of the available
464  * tuntap_drivers.
465  */
466 static struct tuntap_driver *
tuntap_driver_from_flags(int tun_flags)467 tuntap_driver_from_flags(int tun_flags)
468 {
469 
470 	for (u_int i = 0; i < NDRV; i++)
471 		if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
472 		    tuntap_drivers[i].ident_flags)
473 			return (&tuntap_drivers[i]);
474 
475 	return (NULL);
476 }
477 
478 static int
tun_clone_match(struct if_clone * ifc,const char * name)479 tun_clone_match(struct if_clone *ifc, const char *name)
480 {
481 	int tunflags;
482 
483 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
484 		if ((tunflags & TUN_L2) == 0)
485 			return (1);
486 	}
487 
488 	return (0);
489 }
490 
491 static int
tap_clone_match(struct if_clone * ifc,const char * name)492 tap_clone_match(struct if_clone *ifc, const char *name)
493 {
494 	int tunflags;
495 
496 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
497 		if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
498 			return (1);
499 	}
500 
501 	return (0);
502 }
503 
504 static int
vmnet_clone_match(struct if_clone * ifc,const char * name)505 vmnet_clone_match(struct if_clone *ifc, const char *name)
506 {
507 	int tunflags;
508 
509 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
510 		if ((tunflags & TUN_VMNET) != 0)
511 			return (1);
512 	}
513 
514 	return (0);
515 }
516 
517 /*
518  * Create a clone via the ifnet cloning mechanism.  Note that this is invoked
519  * indirectly by tunclone() below.
520  */
521 static int
tun_clone_create(struct if_clone * ifc,char * name,size_t len,struct ifc_data * ifd,struct ifnet ** ifpp)522 tun_clone_create(struct if_clone *ifc, char *name, size_t len,
523     struct ifc_data *ifd, struct ifnet **ifpp)
524 {
525 	struct tuntap_driver *drv;
526 	struct cdev *dev;
527 	int err, i, tunflags, unit;
528 
529 	tunflags = 0;
530 	/* The name here tells us exactly what we're creating */
531 	err = tuntap_name2info(name, &unit, &tunflags);
532 	if (err != 0)
533 		return (err);
534 
535 	drv = tuntap_driver_from_flags(tunflags);
536 	if (drv == NULL)
537 		return (ENXIO);
538 
539 	if (unit != -1) {
540 		/* If this unit number is still available that's okay. */
541 		if (alloc_unr_specific(drv->unrhdr, unit) == -1)
542 			return (EEXIST);
543 	} else {
544 		unit = alloc_unr(drv->unrhdr);
545 	}
546 
547 	snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
548 
549 	/* find any existing device, or allocate new unit number */
550 	dev = NULL;
551 	i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
552 	/* No preexisting struct cdev *, create one */
553 	if (i != 0)
554 		i = tun_create_device(drv, unit, NULL, &dev, name);
555 	if (i == 0) {
556 		struct tuntap_softc *tp;
557 
558 		tuncreate(dev);
559 		tp = dev->si_drv1;
560 		*ifpp = tp->tun_ifp;
561 	}
562 
563 	return (i);
564 }
565 
566 /*
567  * Create a clone via devfs access.
568  */
569 static void
tunclone(void * arg,struct ucred * cred,char * name,int namelen,struct cdev ** dev)570 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
571     struct cdev **dev)
572 {
573 	char devname[SPECNAMELEN + 1];
574 	struct tuntap_driver *drv;
575 	int append_unit, i, u, tunflags;
576 	bool mayclone;
577 
578 	if (*dev != NULL)
579 		return;
580 
581 	tunflags = 0;
582 	CURVNET_SET(CRED_TO_VNET(cred));
583 	if (tuntap_name2info(name, &u, &tunflags) != 0)
584 		goto out;	/* Not recognized */
585 
586 	if (u != -1 && u > IF_MAXUNIT)
587 		goto out;	/* Unit number too high */
588 
589 	mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
590 	if ((tunflags & TUN_L2) != 0) {
591 		/* tap/vmnet allow user open with a sysctl */
592 		mayclone = (mayclone || tap_allow_uopen) && tapdclone;
593 	} else {
594 		mayclone = mayclone && tundclone;
595 	}
596 
597 	/*
598 	 * If tun cloning is enabled, only the superuser can create an
599 	 * interface.
600 	 */
601 	if (!mayclone)
602 		goto out;
603 
604 	if (u == -1)
605 		append_unit = 1;
606 	else
607 		append_unit = 0;
608 
609 	drv = tuntap_driver_from_flags(tunflags);
610 	if (drv == NULL)
611 		goto out;
612 
613 	/* find any existing device, or allocate new unit number */
614 	i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
615 	if (i) {
616 		if (append_unit) {
617 			namelen = snprintf(devname, sizeof(devname), "%s%d",
618 			    name, u);
619 			name = devname;
620 		}
621 
622 		i = tun_create_device(drv, u, cred, dev, name);
623 	} else {
624 		/* Consumed by the dev_clone invoker. */
625 		dev_ref(*dev);
626 	}
627 	if (i == 0)
628 		if_clone_create(name, namelen, NULL);
629 out:
630 	CURVNET_RESTORE();
631 }
632 
633 static int
tun_destroy(struct tuntap_softc * tp,bool may_intr)634 tun_destroy(struct tuntap_softc *tp, bool may_intr)
635 {
636 	int error;
637 
638 	TUN_LOCK(tp);
639 
640 	/*
641 	 * Transient tunnels may have set TUN_DYING if we're being destroyed as
642 	 * a result of the last close, which we'll allow.
643 	 */
644 	MPASS((tp->tun_flags & (TUN_DYING | TUN_TRANSIENT)) != TUN_DYING);
645 	tp->tun_flags |= TUN_DYING;
646 	error = 0;
647 	while (tp->tun_busy != 0) {
648 		if (may_intr)
649 			error = cv_wait_sig(&tp->tun_cv, &tp->tun_mtx);
650 		else
651 			cv_wait(&tp->tun_cv, &tp->tun_mtx);
652 		if (error != 0) {
653 			tp->tun_flags &= ~TUN_DYING;
654 			TUN_UNLOCK(tp);
655 			return (error);
656 		}
657 	}
658 	TUN_UNLOCK(tp);
659 
660 	CURVNET_SET(TUN2IFP(tp)->if_vnet);
661 
662 	mtx_lock(&tunmtx);
663 	TAILQ_REMOVE(&tunhead, tp, tun_list);
664 	mtx_unlock(&tunmtx);
665 
666 	/* destroy_dev will take care of any alias. */
667 	destroy_dev(tp->tun_dev);
668 	seldrain(&tp->tun_rsel);
669 	knlist_clear(&tp->tun_rsel.si_note, 0);
670 	knlist_destroy(&tp->tun_rsel.si_note);
671 	if ((tp->tun_flags & TUN_L2) != 0) {
672 		ether_ifdetach(TUN2IFP(tp));
673 	} else {
674 		bpfdetach(TUN2IFP(tp));
675 		if_detach(TUN2IFP(tp));
676 	}
677 	sx_xlock(&tun_ioctl_sx);
678 	TUN2IFP(tp)->if_softc = NULL;
679 	sx_xunlock(&tun_ioctl_sx);
680 	free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
681 	if_free(TUN2IFP(tp));
682 	mtx_destroy(&tp->tun_mtx);
683 	cv_destroy(&tp->tun_cv);
684 	free(tp, M_TUN);
685 	CURVNET_RESTORE();
686 
687 	return (0);
688 }
689 
690 static int
tun_clone_destroy(struct if_clone * ifc __unused,struct ifnet * ifp,uint32_t flags)691 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags)
692 {
693 	struct tuntap_softc *tp = ifp->if_softc;
694 
695 	return (tun_destroy(tp, true));
696 }
697 
698 static void
vnet_tun_init(const void * unused __unused)699 vnet_tun_init(const void *unused __unused)
700 {
701 
702 	for (u_int i = 0; i < NDRV; ++i) {
703 		struct if_clone_addreq req = {
704 			.match_f = tuntap_drivers[i].clone_match_fn,
705 			.create_f = tuntap_drivers[i].clone_create_fn,
706 			.destroy_f = tuntap_drivers[i].clone_destroy_fn,
707 		};
708 		V_tuntap_driver_cloners[i] =
709 		    ifc_attach_cloner(tuntap_drivers[i].cdevsw.d_name, &req);
710 	};
711 }
712 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
713 		vnet_tun_init, NULL);
714 
715 static void
tun_uninit(const void * unused __unused)716 tun_uninit(const void *unused __unused)
717 {
718 	struct tuntap_driver *drv;
719 	struct tuntap_softc *tp;
720 	int i;
721 
722 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
723 	EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
724 
725 	CURVNET_SET(vnet0);
726 	for (u_int i = 0; i < NDRV; i++) {
727 		if_clone_detach(V_tuntap_driver_cloners[i]);
728 		V_tuntap_driver_cloners[i] = NULL;
729 	}
730 	CURVNET_RESTORE();
731 
732 	if (tuntap_osd_jail_slot != 0)
733 		osd_jail_deregister(tuntap_osd_jail_slot);
734 
735 	mtx_lock(&tunmtx);
736 	while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
737 		mtx_unlock(&tunmtx);
738 		/* tun_destroy() will remove it from the tailq. */
739 		tun_destroy(tp, false);
740 		mtx_lock(&tunmtx);
741 	}
742 	mtx_unlock(&tunmtx);
743 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
744 		drv = &tuntap_drivers[i];
745 		delete_unrhdr(drv->unrhdr);
746 		clone_cleanup(&drv->clones);
747 	}
748 	mtx_destroy(&tunmtx);
749 }
750 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
751 
752 static struct tuntap_driver *
tuntap_driver_from_ifnet(const struct ifnet * ifp)753 tuntap_driver_from_ifnet(const struct ifnet *ifp)
754 {
755 	struct tuntap_driver *drv;
756 	int i;
757 
758 	if (ifp == NULL)
759 		return (NULL);
760 
761 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
762 		drv = &tuntap_drivers[i];
763 		if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
764 			return (drv);
765 	}
766 
767 	return (NULL);
768 }
769 
770 /*
771  * Remove devices that were created by devfs cloning, as they hold references
772  * which prevent the prison from collapsing, in which state VNET sysuninits will
773  * not be invoked.
774  */
775 static int
tuntap_prison_remove(void * obj,void * data __unused)776 tuntap_prison_remove(void *obj, void *data __unused)
777 {
778 #ifdef VIMAGE
779 	struct prison *pr;
780 
781 	pr = obj;
782 	if (prison_owns_vnet(pr)) {
783 		CURVNET_SET(pr->pr_vnet);
784 		for (u_int i = 0; i < NDRV; i++) {
785 			if_clone_detach(V_tuntap_driver_cloners[i]);
786 			V_tuntap_driver_cloners[i] = NULL;
787 		}
788 		CURVNET_RESTORE();
789 	}
790 #endif
791 	return (0);
792 }
793 
794 static int
tuntapmodevent(module_t mod,int type,void * data)795 tuntapmodevent(module_t mod, int type, void *data)
796 {
797 	struct tuntap_driver *drv;
798 	int i;
799 
800 	switch (type) {
801 	case MOD_LOAD:
802 		mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
803 		for (i = 0; i < nitems(tuntap_drivers); ++i) {
804 			drv = &tuntap_drivers[i];
805 			clone_setup(&drv->clones);
806 			drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
807 		}
808 		osd_method_t methods[PR_MAXMETHOD] = {
809 			[PR_METHOD_REMOVE] = tuntap_prison_remove,
810 		};
811 		tuntap_osd_jail_slot = osd_jail_register(NULL, methods);
812 		arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
813 		    tunrename, 0, 1000);
814 		if (arrival_tag == NULL)
815 			return (ENOMEM);
816 		clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
817 		if (clone_tag == NULL)
818 			return (ENOMEM);
819 		break;
820 	case MOD_UNLOAD:
821 		/* See tun_uninit(). */
822 		break;
823 	default:
824 		return EOPNOTSUPP;
825 	}
826 	return 0;
827 }
828 
829 static moduledata_t tuntap_mod = {
830 	"if_tuntap",
831 	tuntapmodevent,
832 	0
833 };
834 
835 /* We'll only ever have these two, so no need for a macro. */
836 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
837 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
838 
839 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
840 MODULE_VERSION(if_tuntap, 1);
841 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
842 MODULE_VERSION(if_tun, 1);
843 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
844 MODULE_VERSION(if_tap, 1);
845 
846 static int
tun_create_device(struct tuntap_driver * drv,int unit,struct ucred * cr,struct cdev ** dev,const char * name)847 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
848     struct cdev **dev, const char *name)
849 {
850 	struct make_dev_args args;
851 	struct tuntap_softc *tp;
852 	int error;
853 
854 	tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
855 	mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
856 	cv_init(&tp->tun_cv, "tun_condvar");
857 	tp->tun_flags = drv->ident_flags;
858 	tp->tun_drv = drv;
859 
860 	make_dev_args_init(&args);
861 	if (cr != NULL)
862 		args.mda_flags = MAKEDEV_REF | MAKEDEV_CHECKNAME;
863 	args.mda_devsw = &drv->cdevsw;
864 	args.mda_cr = cr;
865 	args.mda_uid = UID_UUCP;
866 	args.mda_gid = GID_DIALER;
867 	args.mda_mode = 0600;
868 	args.mda_unit = unit;
869 	args.mda_si_drv1 = tp;
870 	error = make_dev_s(&args, dev, "%s", name);
871 	if (error != 0) {
872 		mtx_destroy(&tp->tun_mtx);
873 		cv_destroy(&tp->tun_cv);
874 		free(tp, M_TUN);
875 		return (error);
876 	}
877 
878 	KASSERT((*dev)->si_drv1 != NULL,
879 	    ("Failed to set si_drv1 at %s creation", name));
880 	tp->tun_dev = *dev;
881 	knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
882 	mtx_lock(&tunmtx);
883 	TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
884 	mtx_unlock(&tunmtx);
885 	return (0);
886 }
887 
888 static void
tunstart(struct ifnet * ifp)889 tunstart(struct ifnet *ifp)
890 {
891 	struct tuntap_softc *tp = ifp->if_softc;
892 	struct mbuf *m;
893 
894 	TUNDEBUG(ifp, "starting\n");
895 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
896 		IFQ_LOCK(&ifp->if_snd);
897 		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
898 		if (m == NULL) {
899 			IFQ_UNLOCK(&ifp->if_snd);
900 			return;
901 		}
902 		IFQ_UNLOCK(&ifp->if_snd);
903 	}
904 
905 	TUN_LOCK(tp);
906 	if (tp->tun_flags & TUN_RWAIT) {
907 		tp->tun_flags &= ~TUN_RWAIT;
908 		wakeup(tp);
909 	}
910 	selwakeuppri(&tp->tun_rsel, PZERO);
911 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
912 	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
913 		TUN_UNLOCK(tp);
914 		pgsigio(&tp->tun_sigio, SIGIO, 0);
915 	} else
916 		TUN_UNLOCK(tp);
917 }
918 
919 /*
920  * tunstart_l2
921  *
922  * queue packets from higher level ready to put out
923  */
924 static void
tunstart_l2(struct ifnet * ifp)925 tunstart_l2(struct ifnet *ifp)
926 {
927 	struct tuntap_softc	*tp = ifp->if_softc;
928 
929 	TUNDEBUG(ifp, "starting\n");
930 
931 	/*
932 	 * do not junk pending output if we are in VMnet mode.
933 	 * XXX: can this do any harm because of queue overflow?
934 	 */
935 
936 	TUN_LOCK(tp);
937 	if (((tp->tun_flags & TUN_VMNET) == 0) &&
938 	    ((tp->tun_flags & TUN_READY) != TUN_READY)) {
939 		struct mbuf *m;
940 
941 		/* Unlocked read. */
942 		TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
943 
944 		for (;;) {
945 			IF_DEQUEUE(&ifp->if_snd, m);
946 			if (m != NULL) {
947 				m_freem(m);
948 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
949 			} else
950 				break;
951 		}
952 		TUN_UNLOCK(tp);
953 
954 		return;
955 	}
956 
957 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
958 
959 	if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
960 		if (tp->tun_flags & TUN_RWAIT) {
961 			tp->tun_flags &= ~TUN_RWAIT;
962 			wakeup(tp);
963 		}
964 
965 		if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
966 			TUN_UNLOCK(tp);
967 			pgsigio(&tp->tun_sigio, SIGIO, 0);
968 			TUN_LOCK(tp);
969 		}
970 
971 		selwakeuppri(&tp->tun_rsel, PZERO);
972 		KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
973 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
974 	}
975 
976 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
977 	TUN_UNLOCK(tp);
978 } /* tunstart_l2 */
979 
980 static int
tap_transmit(struct ifnet * ifp,struct mbuf * m)981 tap_transmit(struct ifnet *ifp, struct mbuf *m)
982 {
983 	int error;
984 
985 	BPF_MTAP(ifp, m);
986 	IFQ_HANDOFF(ifp, m, error);
987 	return (error);
988 }
989 
990 static void
tuncreate(struct cdev * dev)991 tuncreate(struct cdev *dev)
992 {
993 	struct tuntap_driver *drv;
994 	struct tuntap_softc *tp;
995 	struct ifnet *ifp;
996 	struct ether_addr eaddr;
997 	int iflags;
998 	u_char type;
999 
1000 	tp = dev->si_drv1;
1001 	KASSERT(tp != NULL,
1002 	    ("si_drv1 should have been initialized at creation"));
1003 
1004 	drv = tp->tun_drv;
1005 	iflags = IFF_MULTICAST;
1006 	if ((tp->tun_flags & TUN_L2) != 0) {
1007 		type = IFT_ETHER;
1008 		iflags |= IFF_BROADCAST | IFF_SIMPLEX;
1009 	} else {
1010 		type = IFT_PPP;
1011 		iflags |= IFF_POINTOPOINT;
1012 	}
1013 	ifp = tp->tun_ifp = if_alloc(type);
1014 	ifp->if_softc = tp;
1015 	if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
1016 	ifp->if_ioctl = tunifioctl;
1017 	ifp->if_flags = iflags;
1018 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
1019 	ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_MEXTPG;
1020 	if ((tp->tun_flags & TUN_L2) != 0)
1021 		ifp->if_capabilities |=
1022 		    IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO;
1023 	ifp->if_capenable |= IFCAP_LINKSTATE | IFCAP_MEXTPG;
1024 
1025 	if ((tp->tun_flags & TUN_L2) != 0) {
1026 		ifp->if_init = tunifinit;
1027 		ifp->if_start = tunstart_l2;
1028 		ifp->if_transmit = tap_transmit;
1029 		ifp->if_qflush = if_qflush;
1030 
1031 		ether_gen_addr(ifp, &eaddr);
1032 		ether_ifattach(ifp, eaddr.octet);
1033 	} else {
1034 		ifp->if_mtu = TUNMTU;
1035 		ifp->if_start = tunstart;
1036 		ifp->if_output = tunoutput;
1037 
1038 		ifp->if_snd.ifq_drv_maxlen = 0;
1039 		IFQ_SET_READY(&ifp->if_snd);
1040 
1041 		if_attach(ifp);
1042 		bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
1043 	}
1044 
1045 	TUN_LOCK(tp);
1046 	tp->tun_flags |= TUN_INITED;
1047 	TUN_UNLOCK(tp);
1048 
1049 	TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
1050 	    ifp->if_xname, dev2unit(dev));
1051 }
1052 
1053 static void
tunrename(void * arg __unused,struct ifnet * ifp)1054 tunrename(void *arg __unused, struct ifnet *ifp)
1055 {
1056 	struct tuntap_softc *tp;
1057 	int error;
1058 
1059 	if ((ifp->if_flags & IFF_RENAMING) == 0)
1060 		return;
1061 
1062 	if (tuntap_driver_from_ifnet(ifp) == NULL)
1063 		return;
1064 
1065 	/*
1066 	 * We need to grab the ioctl sx long enough to make sure the softc is
1067 	 * still there.  If it is, we can safely try to busy the tun device.
1068 	 * The busy may fail if the device is currently dying, in which case
1069 	 * we do nothing.  If it doesn't fail, the busy count stops the device
1070 	 * from dying until we've created the alias (that will then be
1071 	 * subsequently destroyed).
1072 	 */
1073 	sx_xlock(&tun_ioctl_sx);
1074 	tp = ifp->if_softc;
1075 	if (tp == NULL) {
1076 		sx_xunlock(&tun_ioctl_sx);
1077 		return;
1078 	}
1079 	error = tun_busy(tp);
1080 	sx_xunlock(&tun_ioctl_sx);
1081 	if (error != 0)
1082 		return;
1083 	if (tp->tun_alias != NULL) {
1084 		destroy_dev(tp->tun_alias);
1085 		tp->tun_alias = NULL;
1086 	}
1087 
1088 	if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1089 		goto out;
1090 
1091 	/*
1092 	 * Failure's ok, aliases are created on a best effort basis.  If a
1093 	 * tun user/consumer decides to rename the interface to conflict with
1094 	 * another device (non-ifnet) on the system, we will assume they know
1095 	 * what they are doing.  make_dev_alias_p won't touch tun_alias on
1096 	 * failure, so we use it but ignore the return value.
1097 	 */
1098 	make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1099 	    ifp->if_xname);
1100 out:
1101 	tun_unbusy(tp);
1102 }
1103 
1104 static int
tunopen(struct cdev * dev,int flag,int mode,struct thread * td)1105 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1106 {
1107 	struct ifnet	*ifp;
1108 	struct tuntap_softc *tp;
1109 	int error __diagused, tunflags;
1110 
1111 	tunflags = 0;
1112 	CURVNET_SET(TD_TO_VNET(td));
1113 	error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1114 	if (error != 0) {
1115 		CURVNET_RESTORE();
1116 		return (error);	/* Shouldn't happen */
1117 	}
1118 
1119 	tp = dev->si_drv1;
1120 	KASSERT(tp != NULL,
1121 	    ("si_drv1 should have been initialized at creation"));
1122 
1123 	TUN_LOCK(tp);
1124 	if ((tp->tun_flags & TUN_INITED) == 0) {
1125 		TUN_UNLOCK(tp);
1126 		CURVNET_RESTORE();
1127 		return (ENXIO);
1128 	}
1129 	if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1130 		TUN_UNLOCK(tp);
1131 		CURVNET_RESTORE();
1132 		return (EBUSY);
1133 	}
1134 
1135 	error = tun_busy_locked(tp);
1136 	KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1137 	ifp = TUN2IFP(tp);
1138 
1139 	if ((tp->tun_flags & TUN_L2) != 0) {
1140 		bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1141 		    sizeof(tp->tun_ether.octet));
1142 
1143 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
1144 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1145 
1146 		if (tapuponopen)
1147 			ifp->if_flags |= IFF_UP;
1148 	}
1149 
1150 	tp->tun_pid = td->td_proc->p_pid;
1151 	tp->tun_flags |= TUN_OPEN;
1152 
1153 	if_link_state_change(ifp, LINK_STATE_UP);
1154 	TUNDEBUG(ifp, "open\n");
1155 	TUN_UNLOCK(tp);
1156 
1157 	/*
1158 	 * This can fail with either ENOENT or EBUSY.  This is in the middle of
1159 	 * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1160 	 * the only cdevpriv dtor being set will be tundtor and the softc being
1161 	 * passed is constant for a given cdev.  We ignore the possible error
1162 	 * because of this as either "unlikely" or "not actually a problem."
1163 	 */
1164 	(void)devfs_set_cdevpriv(tp, tundtor);
1165 	CURVNET_RESTORE();
1166 	return (0);
1167 }
1168 
1169 /*
1170  * tundtor - tear down the device - mark i/f down & delete
1171  * routing info
1172  */
1173 static void
tundtor(void * data)1174 tundtor(void *data)
1175 {
1176 	struct proc *p;
1177 	struct tuntap_softc *tp;
1178 	struct ifnet *ifp;
1179 	bool l2tun;
1180 
1181 	tp = data;
1182 	p = curproc;
1183 	ifp = TUN2IFP(tp);
1184 
1185 	TUN_LOCK(tp);
1186 
1187 	/*
1188 	 * Realistically, we can't be obstinate here.  This only means that the
1189 	 * tuntap device was closed out of order, and the last closer wasn't the
1190 	 * controller.  These are still good to know about, though, as software
1191 	 * should avoid multiple processes with a tuntap device open and
1192 	 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1193 	 * parent).
1194 	 */
1195 	if (p->p_pid != tp->tun_pid) {
1196 		log(LOG_INFO,
1197 		    "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1198 		    p->p_pid, p->p_comm, tp->tun_dev->si_name);
1199 	}
1200 
1201 	/*
1202 	 * junk all pending output
1203 	 */
1204 	CURVNET_SET(ifp->if_vnet);
1205 
1206 	l2tun = false;
1207 	if ((tp->tun_flags & TUN_L2) != 0) {
1208 		l2tun = true;
1209 		IF_DRAIN(&ifp->if_snd);
1210 	} else {
1211 		IFQ_PURGE(&ifp->if_snd);
1212 	}
1213 
1214 	/* For vmnet, we won't do most of the address/route bits */
1215 	if ((tp->tun_flags & TUN_VMNET) != 0 ||
1216 	    (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1217 		goto out;
1218 #if defined(INET) || defined(INET6)
1219 	if (l2tun && tp->tun_lro_ready) {
1220 		TUNDEBUG (ifp, "LRO disabled\n");
1221 		tcp_lro_free(&tp->tun_lro);
1222 		tp->tun_lro_ready = false;
1223 	}
1224 #endif
1225 	if (ifp->if_flags & IFF_UP) {
1226 		TUN_UNLOCK(tp);
1227 		if_down(ifp);
1228 		TUN_LOCK(tp);
1229 	}
1230 
1231 	/* Delete all addresses and routes which reference this interface. */
1232 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1233 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1234 		TUN_UNLOCK(tp);
1235 		if_purgeaddrs(ifp);
1236 		TUN_LOCK(tp);
1237 	}
1238 
1239 out:
1240 	if_link_state_change(ifp, LINK_STATE_DOWN);
1241 	CURVNET_RESTORE();
1242 
1243 	funsetown(&tp->tun_sigio);
1244 	selwakeuppri(&tp->tun_rsel, PZERO);
1245 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1246 	TUNDEBUG (ifp, "closed\n");
1247 	tp->tun_flags &= ~TUN_OPEN;
1248 	tp->tun_pid = 0;
1249 	tun_vnethdr_set(ifp, 0);
1250 
1251 	tun_unbusy_locked(tp);
1252 	if ((tp->tun_flags & TUN_TRANSIENT) != 0) {
1253 		struct if_clone *cloner;
1254 		int error __diagused;
1255 
1256 		/* Mark it busy so that nothing can re-open it. */
1257 		tp->tun_flags |= TUN_DYING;
1258 		TUN_UNLOCK(tp);
1259 
1260 		CURVNET_SET_QUIET(ifp->if_home_vnet);
1261 		cloner = tuntap_cloner_from_flags(tp->tun_flags);
1262 		CURVNET_RESTORE();
1263 
1264 		error = if_clone_destroyif(cloner, ifp);
1265 		MPASS(error == 0 || error == EINTR || error == ERESTART);
1266 		return;
1267 	}
1268 
1269 	TUN_UNLOCK(tp);
1270 }
1271 
1272 static void
tuninit(struct ifnet * ifp)1273 tuninit(struct ifnet *ifp)
1274 {
1275 	struct tuntap_softc *tp = ifp->if_softc;
1276 
1277 	TUNDEBUG(ifp, "tuninit\n");
1278 
1279 	TUN_LOCK(tp);
1280 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1281 	if ((tp->tun_flags & TUN_L2) == 0) {
1282 		ifp->if_flags |= IFF_UP;
1283 		getmicrotime(&ifp->if_lastchange);
1284 		TUN_UNLOCK(tp);
1285 	} else {
1286 #if defined(INET) || defined(INET6)
1287 		if (tcp_lro_init(&tp->tun_lro) == 0) {
1288 			TUNDEBUG(ifp, "LRO enabled\n");
1289 			tp->tun_lro.ifp = ifp;
1290 			tp->tun_lro_ready = true;
1291 		} else {
1292 			TUNDEBUG(ifp, "Could not enable LRO\n");
1293 			tp->tun_lro_ready = false;
1294 		}
1295 #endif
1296 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1297 		TUN_UNLOCK(tp);
1298 		/* attempt to start output */
1299 		tunstart_l2(ifp);
1300 	}
1301 
1302 }
1303 
1304 /*
1305  * Used only for l2 tunnel.
1306  */
1307 static void
tunifinit(void * xtp)1308 tunifinit(void *xtp)
1309 {
1310 	struct tuntap_softc *tp;
1311 
1312 	tp = (struct tuntap_softc *)xtp;
1313 	tuninit(tp->tun_ifp);
1314 }
1315 
1316 /*
1317  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1318  * current value of ifp->if_capenable.
1319  */
1320 static void
tun_caps_changed(struct ifnet * ifp)1321 tun_caps_changed(struct ifnet *ifp)
1322 {
1323 	uint64_t hwassist = 0;
1324 
1325 	TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1326 	if (ifp->if_capenable & IFCAP_TXCSUM)
1327 		hwassist |= CSUM_TCP | CSUM_UDP;
1328 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1329 		hwassist |= CSUM_TCP_IPV6
1330 		    | CSUM_UDP_IPV6;
1331 	if (ifp->if_capenable & IFCAP_TSO4)
1332 		hwassist |= CSUM_IP_TSO;
1333 	if (ifp->if_capenable & IFCAP_TSO6)
1334 		hwassist |= CSUM_IP6_TSO;
1335 	ifp->if_hwassist = hwassist;
1336 }
1337 
1338 /*
1339  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1340  * if_capabilities and if_capenable as needed.
1341  */
1342 static void
tun_vnethdr_set(struct ifnet * ifp,int vhdrlen)1343 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1344 {
1345 	struct tuntap_softc *tp = ifp->if_softc;
1346 
1347 	TUN_LOCK_ASSERT(tp);
1348 
1349 	if (tp->tun_vhdrlen == vhdrlen)
1350 		return;
1351 
1352 	/*
1353 	 * Update if_capabilities to reflect the
1354 	 * functionalities offered by the virtio-net
1355 	 * header.
1356 	 */
1357 	if (vhdrlen != 0)
1358 		ifp->if_capabilities |=
1359 			TAP_VNET_HDR_CAPS;
1360 	else
1361 		ifp->if_capabilities &=
1362 			~TAP_VNET_HDR_CAPS;
1363 	/*
1364 	 * Disable any capabilities that we don't
1365 	 * support anymore.
1366 	 */
1367 	ifp->if_capenable &= ifp->if_capabilities;
1368 	tun_caps_changed(ifp);
1369 	tp->tun_vhdrlen = vhdrlen;
1370 
1371 	TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1372 	    vhdrlen, ifp->if_capabilities);
1373 }
1374 
1375 /*
1376  * Process an ioctl request.
1377  */
1378 static int
tunifioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1379 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1380 {
1381 	struct ifreq *ifr = (struct ifreq *)data;
1382 	struct tuntap_softc *tp;
1383 	struct ifstat *ifs;
1384 	struct ifmediareq	*ifmr;
1385 	int		dummy, error = 0;
1386 	bool		l2tun;
1387 
1388 	ifmr = NULL;
1389 	sx_xlock(&tun_ioctl_sx);
1390 	tp = ifp->if_softc;
1391 	if (tp == NULL) {
1392 		error = ENXIO;
1393 		goto bad;
1394 	}
1395 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1396 	switch(cmd) {
1397 	case SIOCGIFSTATUS:
1398 		ifs = (struct ifstat *)data;
1399 		TUN_LOCK(tp);
1400 		if (tp->tun_pid)
1401 			snprintf(ifs->ascii, sizeof(ifs->ascii),
1402 			    "\tOpened by PID %d\n", tp->tun_pid);
1403 		else
1404 			ifs->ascii[0] = '\0';
1405 		TUN_UNLOCK(tp);
1406 		break;
1407 	case SIOCSIFADDR:
1408 		if (l2tun)
1409 			error = ether_ioctl(ifp, cmd, data);
1410 		else
1411 			tuninit(ifp);
1412 		if (error == 0)
1413 		    TUNDEBUG(ifp, "address set\n");
1414 		break;
1415 	case SIOCSIFMTU:
1416 		ifp->if_mtu = ifr->ifr_mtu;
1417 		TUNDEBUG(ifp, "mtu set\n");
1418 		break;
1419 	case SIOCSIFFLAGS:
1420 	case SIOCADDMULTI:
1421 	case SIOCDELMULTI:
1422 		break;
1423 	case SIOCGIFMEDIA:
1424 		if (!l2tun) {
1425 			error = EINVAL;
1426 			break;
1427 		}
1428 
1429 		ifmr = (struct ifmediareq *)data;
1430 		dummy = ifmr->ifm_count;
1431 		ifmr->ifm_count = 1;
1432 		ifmr->ifm_status = IFM_AVALID;
1433 		ifmr->ifm_active = IFM_ETHER | IFM_FDX | IFM_1000_T;
1434 		if (tp->tun_flags & TUN_OPEN)
1435 			ifmr->ifm_status |= IFM_ACTIVE;
1436 		ifmr->ifm_current = ifmr->ifm_active;
1437 		if (dummy >= 1) {
1438 			int media = IFM_ETHER;
1439 			error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1440 		}
1441 		break;
1442 	case SIOCSIFCAP:
1443 		TUN_LOCK(tp);
1444 		ifp->if_capenable = ifr->ifr_reqcap;
1445 		tun_caps_changed(ifp);
1446 		TUN_UNLOCK(tp);
1447 		VLAN_CAPABILITIES(ifp);
1448 		break;
1449 	default:
1450 		if (l2tun) {
1451 			error = ether_ioctl(ifp, cmd, data);
1452 		} else {
1453 			error = EINVAL;
1454 		}
1455 	}
1456 bad:
1457 	sx_xunlock(&tun_ioctl_sx);
1458 	return (error);
1459 }
1460 
1461 /*
1462  * tunoutput - queue packets from higher level ready to put out.
1463  */
1464 static int
tunoutput(struct ifnet * ifp,struct mbuf * m0,const struct sockaddr * dst,struct route * ro)1465 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1466     struct route *ro)
1467 {
1468 	struct tuntap_softc *tp = ifp->if_softc;
1469 	u_short cached_tun_flags;
1470 	int error;
1471 	u_int32_t af;
1472 
1473 	TUNDEBUG (ifp, "tunoutput\n");
1474 
1475 #ifdef MAC
1476 	error = mac_ifnet_check_transmit(ifp, m0);
1477 	if (error) {
1478 		m_freem(m0);
1479 		return (error);
1480 	}
1481 #endif
1482 
1483 	/* Could be unlocked read? */
1484 	TUN_LOCK(tp);
1485 	cached_tun_flags = tp->tun_flags;
1486 	TUN_UNLOCK(tp);
1487 	if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1488 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1489 		m_freem (m0);
1490 		return (EHOSTDOWN);
1491 	}
1492 
1493 	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1494 		m_freem (m0);
1495 		return (EHOSTDOWN);
1496 	}
1497 
1498 	/* BPF writes need to be handled specially. */
1499 	if (dst->sa_family == AF_UNSPEC || dst->sa_family == pseudo_AF_HDRCMPLT)
1500 		bcopy(dst->sa_data, &af, sizeof(af));
1501 	else
1502 		af = RO_GET_FAMILY(ro, dst);
1503 
1504 	BPF_MTAP2(ifp, &af, sizeof(af), m0);
1505 
1506 	/* prepend sockaddr? this may abort if the mbuf allocation fails */
1507 	if (cached_tun_flags & TUN_LMODE) {
1508 		/* allocate space for sockaddr */
1509 		M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1510 
1511 		/* if allocation failed drop packet */
1512 		if (m0 == NULL) {
1513 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1514 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1515 			return (ENOBUFS);
1516 		} else {
1517 			bcopy(dst, m0->m_data, dst->sa_len);
1518 		}
1519 	}
1520 
1521 	if (cached_tun_flags & TUN_IFHEAD) {
1522 		/* Prepend the address family */
1523 		M_PREPEND(m0, 4, M_NOWAIT);
1524 
1525 		/* if allocation failed drop packet */
1526 		if (m0 == NULL) {
1527 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1528 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1529 			return (ENOBUFS);
1530 		} else
1531 			*(u_int32_t *)m0->m_data = htonl(af);
1532 	} else {
1533 #ifdef INET
1534 		if (af != AF_INET)
1535 #endif
1536 		{
1537 			m_freem(m0);
1538 			return (EAFNOSUPPORT);
1539 		}
1540 	}
1541 
1542 	error = (ifp->if_transmit)(ifp, m0);
1543 	if (error)
1544 		return (ENOBUFS);
1545 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1546 	return (0);
1547 }
1548 
1549 /*
1550  * the cdevsw interface is now pretty minimal.
1551  */
1552 static	int
tunioctl(struct cdev * dev,u_long cmd,caddr_t data,int flag,struct thread * td)1553 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1554     struct thread *td)
1555 {
1556 	struct ifreq ifr, *ifrp;
1557 	struct tuntap_softc *tp = dev->si_drv1;
1558 	struct ifnet *ifp = TUN2IFP(tp);
1559 	struct tuninfo *tunp;
1560 	int error, iflags, ival;
1561 	bool	l2tun;
1562 
1563 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1564 	if (l2tun) {
1565 		/* tap specific ioctls */
1566 		switch(cmd) {
1567 		/* VMware/VMnet port ioctl's */
1568 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1569     defined(COMPAT_FREEBSD4)
1570 		case _IO('V', 0):
1571 			ival = IOCPARM_IVAL(data);
1572 			data = (caddr_t)&ival;
1573 			/* FALLTHROUGH */
1574 #endif
1575 		case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1576 			iflags = *(int *)data;
1577 			iflags &= TUN_VMIO_FLAG_MASK;
1578 			iflags &= ~IFF_CANTCHANGE;
1579 			iflags |= IFF_UP;
1580 
1581 			TUN_LOCK(tp);
1582 			ifp->if_flags = iflags |
1583 			    (ifp->if_flags & IFF_CANTCHANGE);
1584 			TUN_UNLOCK(tp);
1585 
1586 			return (0);
1587 		case SIOCGIFADDR:	/* get MAC address of the remote side */
1588 			TUN_LOCK(tp);
1589 			bcopy(&tp->tun_ether.octet, data,
1590 			    sizeof(tp->tun_ether.octet));
1591 			TUN_UNLOCK(tp);
1592 
1593 			return (0);
1594 		case SIOCSIFADDR:	/* set MAC address of the remote side */
1595 			TUN_LOCK(tp);
1596 			bcopy(data, &tp->tun_ether.octet,
1597 			    sizeof(tp->tun_ether.octet));
1598 			TUN_UNLOCK(tp);
1599 
1600 			return (0);
1601 		case TAPSVNETHDR:
1602 			ival = *(int *)data;
1603 			if (ival != 0 &&
1604 			    ival != sizeof(struct virtio_net_hdr) &&
1605 			    ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1606 				return (EINVAL);
1607 			}
1608 			TUN_LOCK(tp);
1609 			tun_vnethdr_set(ifp, ival);
1610 			TUN_UNLOCK(tp);
1611 
1612 			return (0);
1613 		case TAPGVNETHDR:
1614 			TUN_LOCK(tp);
1615 			*(int *)data = tp->tun_vhdrlen;
1616 			TUN_UNLOCK(tp);
1617 
1618 			return (0);
1619 		}
1620 
1621 		/* Fall through to the common ioctls if unhandled */
1622 	} else {
1623 		switch (cmd) {
1624 		case TUNSLMODE:
1625 			TUN_LOCK(tp);
1626 			if (*(int *)data) {
1627 				tp->tun_flags |= TUN_LMODE;
1628 				tp->tun_flags &= ~TUN_IFHEAD;
1629 			} else
1630 				tp->tun_flags &= ~TUN_LMODE;
1631 			TUN_UNLOCK(tp);
1632 
1633 			return (0);
1634 		case TUNSIFHEAD:
1635 			TUN_LOCK(tp);
1636 			if (*(int *)data) {
1637 				tp->tun_flags |= TUN_IFHEAD;
1638 				tp->tun_flags &= ~TUN_LMODE;
1639 			} else
1640 				tp->tun_flags &= ~TUN_IFHEAD;
1641 			TUN_UNLOCK(tp);
1642 
1643 			return (0);
1644 		case TUNGIFHEAD:
1645 			TUN_LOCK(tp);
1646 			*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1647 			TUN_UNLOCK(tp);
1648 
1649 			return (0);
1650 		case TUNSIFMODE:
1651 			/* deny this if UP */
1652 			if (TUN2IFP(tp)->if_flags & IFF_UP)
1653 				return (EBUSY);
1654 
1655 			switch (*(int *)data & ~IFF_MULTICAST) {
1656 			case IFF_POINTOPOINT:
1657 			case IFF_BROADCAST:
1658 				TUN_LOCK(tp);
1659 				TUN2IFP(tp)->if_flags &=
1660 				    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1661 				TUN2IFP(tp)->if_flags |= *(int *)data;
1662 				TUN_UNLOCK(tp);
1663 
1664 				break;
1665 			default:
1666 				return (EINVAL);
1667 			}
1668 
1669 			return (0);
1670 		case TUNSIFPID:
1671 			TUN_LOCK(tp);
1672 			tp->tun_pid = curthread->td_proc->p_pid;
1673 			TUN_UNLOCK(tp);
1674 
1675 			return (0);
1676 		}
1677 		/* Fall through to the common ioctls if unhandled */
1678 	}
1679 
1680 	switch (cmd) {
1681 	case TUNGIFNAME:
1682 		ifrp = (struct ifreq *)data;
1683 		strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1684 
1685 		return (0);
1686 	case TUNSIFINFO:
1687 		tunp = (struct tuninfo *)data;
1688 		if (TUN2IFP(tp)->if_type != tunp->type)
1689 			return (EPROTOTYPE);
1690 		TUN_LOCK(tp);
1691 		if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1692 			strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1693 			ifr.ifr_mtu = tunp->mtu;
1694 			CURVNET_SET(TUN2IFP(tp)->if_vnet);
1695 			error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1696 			    (caddr_t)&ifr, td);
1697 			CURVNET_RESTORE();
1698 			if (error) {
1699 				TUN_UNLOCK(tp);
1700 				return (error);
1701 			}
1702 		}
1703 		TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1704 		TUN_UNLOCK(tp);
1705 		break;
1706 	case TUNGIFINFO:
1707 		tunp = (struct tuninfo *)data;
1708 		TUN_LOCK(tp);
1709 		tunp->mtu = TUN2IFP(tp)->if_mtu;
1710 		tunp->type = TUN2IFP(tp)->if_type;
1711 		tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1712 		TUN_UNLOCK(tp);
1713 		break;
1714 	case TUNSDEBUG:
1715 		tundebug = *(int *)data;
1716 		break;
1717 	case TUNGDEBUG:
1718 		*(int *)data = tundebug;
1719 		break;
1720 	case TUNSTRANSIENT:
1721 		TUN_LOCK(tp);
1722 		if (*(int *)data)
1723 			tp->tun_flags |= TUN_TRANSIENT;
1724 		else
1725 			tp->tun_flags &= ~TUN_TRANSIENT;
1726 		TUN_UNLOCK(tp);
1727 		break;
1728 	case TUNGTRANSIENT:
1729 		TUN_LOCK(tp);
1730 		*(int *)data = (tp->tun_flags & TUN_TRANSIENT) != 0;
1731 		TUN_UNLOCK(tp);
1732 		break;
1733 	case FIONBIO:
1734 		break;
1735 	case FIOASYNC:
1736 		TUN_LOCK(tp);
1737 		if (*(int *)data)
1738 			tp->tun_flags |= TUN_ASYNC;
1739 		else
1740 			tp->tun_flags &= ~TUN_ASYNC;
1741 		TUN_UNLOCK(tp);
1742 		break;
1743 	case FIONREAD:
1744 		if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1745 			struct mbuf *mb;
1746 			IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1747 			IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1748 			for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1749 				*(int *)data += mb->m_len;
1750 			IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1751 		} else
1752 			*(int *)data = 0;
1753 		break;
1754 	case FIOSETOWN:
1755 		return (fsetown(*(int *)data, &tp->tun_sigio));
1756 
1757 	case FIOGETOWN:
1758 		*(int *)data = fgetown(&tp->tun_sigio);
1759 		return (0);
1760 
1761 	/* This is deprecated, FIOSETOWN should be used instead. */
1762 	case TIOCSPGRP:
1763 		return (fsetown(-(*(int *)data), &tp->tun_sigio));
1764 
1765 	/* This is deprecated, FIOGETOWN should be used instead. */
1766 	case TIOCGPGRP:
1767 		*(int *)data = -fgetown(&tp->tun_sigio);
1768 		return (0);
1769 
1770 	default:
1771 		return (ENOTTY);
1772 	}
1773 	return (0);
1774 }
1775 
1776 /*
1777  * The cdevsw read interface - reads a packet at a time, or at
1778  * least as much of a packet as can be read.
1779  */
1780 static	int
tunread(struct cdev * dev,struct uio * uio,int flag)1781 tunread(struct cdev *dev, struct uio *uio, int flag)
1782 {
1783 	struct tuntap_softc *tp = dev->si_drv1;
1784 	struct ifnet	*ifp = TUN2IFP(tp);
1785 	struct mbuf	*m;
1786 	size_t		len;
1787 	int		error = 0;
1788 
1789 	TUNDEBUG (ifp, "read\n");
1790 	TUN_LOCK(tp);
1791 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1792 		TUN_UNLOCK(tp);
1793 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1794 		return (EHOSTDOWN);
1795 	}
1796 
1797 	tp->tun_flags &= ~TUN_RWAIT;
1798 
1799 	for (;;) {
1800 		IFQ_DEQUEUE(&ifp->if_snd, m);
1801 		if (m != NULL)
1802 			break;
1803 		if (flag & O_NONBLOCK) {
1804 			TUN_UNLOCK(tp);
1805 			return (EWOULDBLOCK);
1806 		}
1807 		tp->tun_flags |= TUN_RWAIT;
1808 		error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | PZERO,
1809 		    "tunread", 0);
1810 		if (error != 0) {
1811 			TUN_UNLOCK(tp);
1812 			return (error);
1813 		}
1814 	}
1815 	TUN_UNLOCK(tp);
1816 
1817 	len = min(tp->tun_vhdrlen, uio->uio_resid);
1818 	if (len > 0) {
1819 		struct virtio_net_hdr_mrg_rxbuf vhdr;
1820 
1821 		bzero(&vhdr, sizeof(vhdr));
1822 		if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1823 			m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1824 		}
1825 
1826 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1827 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1828 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1829 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1830 		    vhdr.hdr.csum_offset);
1831 		error = uiomove(&vhdr, len, uio);
1832 	}
1833 	if (error == 0)
1834 		error = m_mbuftouio(uio, m, 0);
1835 	m_freem(m);
1836 	return (error);
1837 }
1838 
1839 static int
tunwrite_l2(struct tuntap_softc * tp,struct mbuf * m,struct virtio_net_hdr_mrg_rxbuf * vhdr)1840 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1841 	    struct virtio_net_hdr_mrg_rxbuf *vhdr)
1842 {
1843 	struct epoch_tracker et;
1844 	struct ether_header *eh;
1845 	struct ifnet *ifp;
1846 
1847 	ifp = TUN2IFP(tp);
1848 
1849 	/*
1850 	 * Only pass a unicast frame to ether_input(), if it would
1851 	 * actually have been received by non-virtual hardware.
1852 	 */
1853 	if (m->m_len < sizeof(struct ether_header)) {
1854 		m_freem(m);
1855 		return (0);
1856 	}
1857 
1858 	eh = mtod(m, struct ether_header *);
1859 
1860 	if ((ifp->if_flags & IFF_PROMISC) == 0 &&
1861 	    !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1862 	    bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1863 		m_freem(m);
1864 		return (0);
1865 	}
1866 
1867 	if (vhdr != NULL) {
1868 		if (virtio_net_rx_csum(m, &vhdr->hdr)) {
1869 			m_freem(m);
1870 			return (0);
1871 		}
1872 	} else {
1873 		switch (ntohs(eh->ether_type)) {
1874 #ifdef INET
1875 		case ETHERTYPE_IP:
1876 			if (ifp->if_capenable & IFCAP_RXCSUM) {
1877 				m->m_pkthdr.csum_flags |=
1878 				    CSUM_IP_CHECKED | CSUM_IP_VALID |
1879 				    CSUM_DATA_VALID | CSUM_SCTP_VALID |
1880 				    CSUM_PSEUDO_HDR;
1881 				m->m_pkthdr.csum_data = 0xffff;
1882 			}
1883 			break;
1884 #endif
1885 #ifdef INET6
1886 		case ETHERTYPE_IPV6:
1887 			if (ifp->if_capenable & IFCAP_RXCSUM_IPV6) {
1888 				m->m_pkthdr.csum_flags |=
1889 				    CSUM_DATA_VALID_IPV6 | CSUM_SCTP_VALID |
1890 				    CSUM_PSEUDO_HDR;
1891 				m->m_pkthdr.csum_data = 0xffff;
1892 			}
1893 			break;
1894 #endif
1895 		}
1896 	}
1897 
1898 	/* Pass packet up to parent. */
1899 	CURVNET_SET(ifp->if_vnet);
1900 	NET_EPOCH_ENTER(et);
1901 #if defined(INET) || defined(INET6)
1902 	if (tp->tun_lro_ready && ifp->if_capenable & IFCAP_LRO &&
1903 	    tcp_lro_rx(&tp->tun_lro, m, 0) == 0)
1904 		tcp_lro_flush_all(&tp->tun_lro);
1905 	else
1906 #endif
1907 		(*ifp->if_input)(ifp, m);
1908 	NET_EPOCH_EXIT(et);
1909 	CURVNET_RESTORE();
1910 	/* ibytes are counted in parent */
1911 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1912 	return (0);
1913 }
1914 
1915 static int
tunwrite_l3(struct tuntap_softc * tp,struct mbuf * m)1916 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1917 {
1918 	struct epoch_tracker et;
1919 	struct ifnet *ifp;
1920 	int family, isr;
1921 
1922 	ifp = TUN2IFP(tp);
1923 	/* Could be unlocked read? */
1924 	TUN_LOCK(tp);
1925 	if (tp->tun_flags & TUN_IFHEAD) {
1926 		TUN_UNLOCK(tp);
1927 		if (m->m_len < sizeof(family) &&
1928 		(m = m_pullup(m, sizeof(family))) == NULL)
1929 			return (ENOBUFS);
1930 		family = ntohl(*mtod(m, u_int32_t *));
1931 		m_adj(m, sizeof(family));
1932 	} else {
1933 		TUN_UNLOCK(tp);
1934 		family = AF_INET;
1935 	}
1936 
1937 	BPF_MTAP2(ifp, &family, sizeof(family), m);
1938 
1939 	switch (family) {
1940 #ifdef INET
1941 	case AF_INET:
1942 		isr = NETISR_IP;
1943 		break;
1944 #endif
1945 #ifdef INET6
1946 	case AF_INET6:
1947 		isr = NETISR_IPV6;
1948 		break;
1949 #endif
1950 	default:
1951 		m_freem(m);
1952 		return (EAFNOSUPPORT);
1953 	}
1954 	random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1955 	if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1956 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1957 	CURVNET_SET(ifp->if_vnet);
1958 	M_SETFIB(m, ifp->if_fib);
1959 	NET_EPOCH_ENTER(et);
1960 	netisr_dispatch(isr, m);
1961 	NET_EPOCH_EXIT(et);
1962 	CURVNET_RESTORE();
1963 	return (0);
1964 }
1965 
1966 /*
1967  * the cdevsw write interface - an atomic write is a packet - or else!
1968  */
1969 static	int
tunwrite(struct cdev * dev,struct uio * uio,int flag)1970 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1971 {
1972 	struct virtio_net_hdr_mrg_rxbuf vhdr;
1973 	struct tuntap_softc *tp;
1974 	struct ifnet	*ifp;
1975 	struct mbuf	*m;
1976 	uint32_t	mru;
1977 	int		align, vhdrlen, error;
1978 	bool		l2tun;
1979 
1980 	tp = dev->si_drv1;
1981 	ifp = TUN2IFP(tp);
1982 	TUNDEBUG(ifp, "tunwrite\n");
1983 	if ((ifp->if_flags & IFF_UP) != IFF_UP)
1984 		/* ignore silently */
1985 		return (0);
1986 
1987 	if (uio->uio_resid == 0)
1988 		return (0);
1989 
1990 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1991 	mru = l2tun ? TAPMRU : TUNMRU;
1992 	vhdrlen = tp->tun_vhdrlen;
1993 	align = 0;
1994 	if (l2tun) {
1995 		align = ETHER_ALIGN;
1996 		mru += vhdrlen;
1997 	} else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1998 		mru += sizeof(uint32_t);	/* family */
1999 	if (uio->uio_resid < 0 || uio->uio_resid > mru) {
2000 		TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
2001 		return (EIO);
2002 	}
2003 
2004 	if (vhdrlen > 0) {
2005 		error = uiomove(&vhdr, vhdrlen, uio);
2006 		if (error != 0)
2007 			return (error);
2008 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
2009 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
2010 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
2011 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
2012 		    vhdr.hdr.csum_offset);
2013 	}
2014 
2015 	if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
2016 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2017 		return (ENOBUFS);
2018 	}
2019 
2020 	m->m_pkthdr.rcvif = ifp;
2021 #ifdef MAC
2022 	mac_ifnet_create_mbuf(ifp, m);
2023 #endif
2024 
2025 	if (l2tun)
2026 		return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
2027 
2028 	return (tunwrite_l3(tp, m));
2029 }
2030 
2031 /*
2032  * tunpoll - the poll interface, this is only useful on reads
2033  * really. The write detect always returns true, write never blocks
2034  * anyway, it either accepts the packet or drops it.
2035  */
2036 static	int
tunpoll(struct cdev * dev,int events,struct thread * td)2037 tunpoll(struct cdev *dev, int events, struct thread *td)
2038 {
2039 	struct tuntap_softc *tp = dev->si_drv1;
2040 	struct ifnet	*ifp = TUN2IFP(tp);
2041 	int		revents = 0;
2042 
2043 	TUNDEBUG(ifp, "tunpoll\n");
2044 
2045 	if (events & (POLLIN | POLLRDNORM)) {
2046 		IFQ_LOCK(&ifp->if_snd);
2047 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
2048 			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
2049 			revents |= events & (POLLIN | POLLRDNORM);
2050 		} else {
2051 			TUNDEBUG(ifp, "tunpoll waiting\n");
2052 			selrecord(td, &tp->tun_rsel);
2053 		}
2054 		IFQ_UNLOCK(&ifp->if_snd);
2055 	}
2056 	revents |= events & (POLLOUT | POLLWRNORM);
2057 
2058 	return (revents);
2059 }
2060 
2061 /*
2062  * tunkqfilter - support for the kevent() system call.
2063  */
2064 static int
tunkqfilter(struct cdev * dev,struct knote * kn)2065 tunkqfilter(struct cdev *dev, struct knote *kn)
2066 {
2067 	struct tuntap_softc	*tp = dev->si_drv1;
2068 	struct ifnet	*ifp = TUN2IFP(tp);
2069 
2070 	switch(kn->kn_filter) {
2071 	case EVFILT_READ:
2072 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
2073 		    ifp->if_xname, dev2unit(dev));
2074 		kn->kn_fop = &tun_read_filterops;
2075 		break;
2076 
2077 	case EVFILT_WRITE:
2078 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
2079 		    ifp->if_xname, dev2unit(dev));
2080 		kn->kn_fop = &tun_write_filterops;
2081 		break;
2082 
2083 	default:
2084 		TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
2085 		    ifp->if_xname, dev2unit(dev));
2086 		return(EINVAL);
2087 	}
2088 
2089 	kn->kn_hook = tp;
2090 	knlist_add(&tp->tun_rsel.si_note, kn, 0);
2091 
2092 	return (0);
2093 }
2094 
2095 /*
2096  * Return true of there is data in the interface queue.
2097  */
2098 static int
tunkqread(struct knote * kn,long hint)2099 tunkqread(struct knote *kn, long hint)
2100 {
2101 	int			ret;
2102 	struct tuntap_softc	*tp = kn->kn_hook;
2103 	struct cdev		*dev = tp->tun_dev;
2104 	struct ifnet	*ifp = TUN2IFP(tp);
2105 
2106 	if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
2107 		TUNDEBUG(ifp,
2108 		    "%s have data in the queue.  Len = %d, minor = %#x\n",
2109 		    ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
2110 		ret = 1;
2111 	} else {
2112 		TUNDEBUG(ifp,
2113 		    "%s waiting for data, minor = %#x\n", ifp->if_xname,
2114 		    dev2unit(dev));
2115 		ret = 0;
2116 	}
2117 
2118 	return (ret);
2119 }
2120 
2121 /*
2122  * Always can write, always return MTU in kn->data.
2123  */
2124 static int
tunkqwrite(struct knote * kn,long hint)2125 tunkqwrite(struct knote *kn, long hint)
2126 {
2127 	struct tuntap_softc	*tp = kn->kn_hook;
2128 	struct ifnet	*ifp = TUN2IFP(tp);
2129 
2130 	kn->kn_data = ifp->if_mtu;
2131 
2132 	return (1);
2133 }
2134 
2135 static void
tunkqdetach(struct knote * kn)2136 tunkqdetach(struct knote *kn)
2137 {
2138 	struct tuntap_softc	*tp = kn->kn_hook;
2139 
2140 	knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2141 }
2142