1 /* $NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $ */
2 /*-
3 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 *
5 * Copyright (C) 1999-2000 by Maksim Yevmenkin <[email protected]>
6 * All rights reserved.
7 * Copyright (c) 2019 Kyle Evans <[email protected]>
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 <[email protected]>
35 * Nottingham University 1987.
36 *
37 * This source may be freely distributed, however I would be interested
38 * in any changes that are made.
39 *
40 * This driver takes packets off the IP i/f and hands them up to a
41 * user process to have its wicked way with. This driver has it's
42 * roots in a similar driver written by Phil Cockcroft (formerly) at
43 * UCL. This driver is based much more on read/write/poll mode of
44 * operation though.
45 *
46 * $FreeBSD$
47 */
48
49 #include "opt_inet.h"
50 #include "opt_inet6.h"
51
52 #include <sys/param.h>
53 #include <sys/lock.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/systm.h>
57 #include <sys/jail.h>
58 #include <sys/mbuf.h>
59 #include <sys/module.h>
60 #include <sys/socket.h>
61 #include <sys/eventhandler.h>
62 #include <sys/fcntl.h>
63 #include <sys/filio.h>
64 #include <sys/sockio.h>
65 #include <sys/sx.h>
66 #include <sys/syslog.h>
67 #include <sys/ttycom.h>
68 #include <sys/poll.h>
69 #include <sys/selinfo.h>
70 #include <sys/signalvar.h>
71 #include <sys/filedesc.h>
72 #include <sys/kernel.h>
73 #include <sys/sysctl.h>
74 #include <sys/conf.h>
75 #include <sys/uio.h>
76 #include <sys/malloc.h>
77 #include <sys/random.h>
78 #include <sys/ctype.h>
79
80 #include <net/ethernet.h>
81 #include <net/if.h>
82 #include <net/if_var.h>
83 #include <net/if_clone.h>
84 #include <net/if_dl.h>
85 #include <net/if_media.h>
86 #include <net/if_types.h>
87 #include <net/if_vlan_var.h>
88 #include <net/netisr.h>
89 #include <net/route.h>
90 #include <net/vnet.h>
91 #include <netinet/in.h>
92 #ifdef INET
93 #include <netinet/ip.h>
94 #endif
95 #ifdef INET6
96 #include <netinet/ip6.h>
97 #include <netinet6/ip6_var.h>
98 #endif
99 #include <netinet/udp.h>
100 #include <netinet/tcp.h>
101 #include <net/bpf.h>
102 #include <net/if_tap.h>
103 #include <net/if_tun.h>
104
105 #include <dev/virtio/network/virtio_net.h>
106
107 #include <sys/queue.h>
108 #include <sys/condvar.h>
109 #include <security/mac/mac_framework.h>
110
111 struct tuntap_driver;
112
113 /*
114 * tun_list is protected by global tunmtx. Other mutable fields are
115 * protected by tun->tun_mtx, or by their owning subsystem. tun_dev is
116 * static for the duration of a tunnel interface.
117 */
118 struct tuntap_softc {
119 TAILQ_ENTRY(tuntap_softc) tun_list;
120 struct cdev *tun_alias;
121 struct cdev *tun_dev;
122 u_short tun_flags; /* misc flags */
123 #define TUN_OPEN 0x0001
124 #define TUN_INITED 0x0002
125 #define TUN_UNUSED1 0x0008
126 #define TUN_UNUSED2 0x0010
127 #define TUN_LMODE 0x0020
128 #define TUN_RWAIT 0x0040
129 #define TUN_ASYNC 0x0080
130 #define TUN_IFHEAD 0x0100
131 #define TUN_DYING 0x0200
132 #define TUN_L2 0x0400
133 #define TUN_VMNET 0x0800
134
135 #define TUN_DRIVER_IDENT_MASK (TUN_L2 | TUN_VMNET)
136 #define TUN_READY (TUN_OPEN | TUN_INITED)
137
138 pid_t tun_pid; /* owning pid */
139 struct ifnet *tun_ifp; /* the interface */
140 struct sigio *tun_sigio; /* async I/O info */
141 struct tuntap_driver *tun_drv; /* appropriate driver */
142 struct selinfo tun_rsel; /* read select */
143 struct mtx tun_mtx; /* softc field mutex */
144 struct cv tun_cv; /* for ref'd dev destroy */
145 struct ether_addr tun_ether; /* remote address */
146 int tun_busy; /* busy count */
147 int tun_vhdrlen; /* virtio-net header length */
148 };
149 #define TUN2IFP(sc) ((sc)->tun_ifp)
150
151 #define TUNDEBUG if (tundebug) if_printf
152
153 #define TUN_LOCK(tp) mtx_lock(&(tp)->tun_mtx)
154 #define TUN_UNLOCK(tp) mtx_unlock(&(tp)->tun_mtx)
155 #define TUN_LOCK_ASSERT(tp) mtx_assert(&(tp)->tun_mtx, MA_OWNED);
156
157 #define TUN_VMIO_FLAG_MASK 0x0fff
158
159 /*
160 * Interface capabilities of a tap device that supports the virtio-net
161 * header.
162 */
163 #define TAP_VNET_HDR_CAPS (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6 \
164 | IFCAP_VLAN_HWCSUM \
165 | IFCAP_TSO | IFCAP_LRO \
166 | IFCAP_VLAN_HWTSO)
167
168 #define TAP_ALL_OFFLOAD (CSUM_TSO | CSUM_TCP | CSUM_UDP |\
169 CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
170
171 /*
172 * All mutable global variables in if_tun are locked using tunmtx, with
173 * the exception of tundebug, which is used unlocked, and the drivers' *clones,
174 * which are static after setup.
175 */
176 static struct mtx tunmtx;
177 static eventhandler_tag arrival_tag;
178 static eventhandler_tag clone_tag;
179 static const char tunname[] = "tun";
180 static const char tapname[] = "tap";
181 static const char vmnetname[] = "vmnet";
182 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
183 static int tundebug = 0;
184 static int tundclone = 1;
185 static int tap_allow_uopen = 0; /* allow user devfs cloning */
186 static int tapuponopen = 0; /* IFF_UP on open() */
187 static int tapdclone = 1; /* enable devfs cloning */
188
189 static TAILQ_HEAD(,tuntap_softc) tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
190 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
191
192 static struct sx tun_ioctl_sx;
193 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
194
195 SYSCTL_DECL(_net_link);
196 /* tun */
197 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
198 "IP tunnel software network interface");
199 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
200 "Enable legacy devfs interface creation");
201
202 /* tap */
203 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
204 "Ethernet tunnel software network interface");
205 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
206 "Enable legacy devfs interface creation for all users");
207 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
208 "Bring interface up when /dev/tap is opened");
209 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
210 "Enable legacy devfs interface creation");
211 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
212
213 static int tun_create_device(struct tuntap_driver *drv, int unit,
214 struct ucred *cr, struct cdev **dev, const char *name);
215 static int tun_busy_locked(struct tuntap_softc *tp);
216 static void tun_unbusy_locked(struct tuntap_softc *tp);
217 static int tun_busy(struct tuntap_softc *tp);
218 static void tun_unbusy(struct tuntap_softc *tp);
219
220 static int tuntap_name2info(const char *name, int *unit, int *flags);
221 static void tunclone(void *arg, struct ucred *cred, char *name,
222 int namelen, struct cdev **dev);
223 static void tuncreate(struct cdev *dev);
224 static void tundtor(void *data);
225 static void tunrename(void *arg, struct ifnet *ifp);
226 static int tunifioctl(struct ifnet *, u_long, caddr_t);
227 static void tuninit(struct ifnet *);
228 static void tunifinit(void *xtp);
229 static int tuntapmodevent(module_t, int, void *);
230 static int tunoutput(struct ifnet *, struct mbuf *,
231 const struct sockaddr *, struct route *ro);
232 static void tunstart(struct ifnet *);
233 static void tunstart_l2(struct ifnet *);
234
235 static int tun_clone_match(struct if_clone *ifc, const char *name);
236 static int tap_clone_match(struct if_clone *ifc, const char *name);
237 static int vmnet_clone_match(struct if_clone *ifc, const char *name);
238 static int tun_clone_create(struct if_clone *, char *, size_t, caddr_t);
239 static int tun_clone_destroy(struct if_clone *, struct ifnet *);
240 static void tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
241
242 static d_open_t tunopen;
243 static d_read_t tunread;
244 static d_write_t tunwrite;
245 static d_ioctl_t tunioctl;
246 static d_poll_t tunpoll;
247 static d_kqfilter_t tunkqfilter;
248
249 static int tunkqread(struct knote *, long);
250 static int tunkqwrite(struct knote *, long);
251 static void tunkqdetach(struct knote *);
252
253 static struct filterops tun_read_filterops = {
254 .f_isfd = 1,
255 .f_attach = NULL,
256 .f_detach = tunkqdetach,
257 .f_event = tunkqread,
258 };
259
260 static struct filterops tun_write_filterops = {
261 .f_isfd = 1,
262 .f_attach = NULL,
263 .f_detach = tunkqdetach,
264 .f_event = tunkqwrite,
265 };
266
267 static struct tuntap_driver {
268 struct cdevsw cdevsw;
269 int ident_flags;
270 struct unrhdr *unrhdr;
271 struct clonedevs *clones;
272 ifc_match_t *clone_match_fn;
273 ifc_create_t *clone_create_fn;
274 ifc_destroy_t *clone_destroy_fn;
275 } tuntap_drivers[] = {
276 {
277 .ident_flags = 0,
278 .cdevsw = {
279 .d_version = D_VERSION,
280 .d_flags = D_NEEDMINOR,
281 .d_open = tunopen,
282 .d_read = tunread,
283 .d_write = tunwrite,
284 .d_ioctl = tunioctl,
285 .d_poll = tunpoll,
286 .d_kqfilter = tunkqfilter,
287 .d_name = tunname,
288 },
289 .clone_match_fn = tun_clone_match,
290 .clone_create_fn = tun_clone_create,
291 .clone_destroy_fn = tun_clone_destroy,
292 },
293 {
294 .ident_flags = TUN_L2,
295 .cdevsw = {
296 .d_version = D_VERSION,
297 .d_flags = D_NEEDMINOR,
298 .d_open = tunopen,
299 .d_read = tunread,
300 .d_write = tunwrite,
301 .d_ioctl = tunioctl,
302 .d_poll = tunpoll,
303 .d_kqfilter = tunkqfilter,
304 .d_name = tapname,
305 },
306 .clone_match_fn = tap_clone_match,
307 .clone_create_fn = tun_clone_create,
308 .clone_destroy_fn = tun_clone_destroy,
309 },
310 {
311 .ident_flags = TUN_L2 | TUN_VMNET,
312 .cdevsw = {
313 .d_version = D_VERSION,
314 .d_flags = D_NEEDMINOR,
315 .d_open = tunopen,
316 .d_read = tunread,
317 .d_write = tunwrite,
318 .d_ioctl = tunioctl,
319 .d_poll = tunpoll,
320 .d_kqfilter = tunkqfilter,
321 .d_name = vmnetname,
322 },
323 .clone_match_fn = vmnet_clone_match,
324 .clone_create_fn = tun_clone_create,
325 .clone_destroy_fn = tun_clone_destroy,
326 },
327 };
328
329 struct tuntap_driver_cloner {
330 SLIST_ENTRY(tuntap_driver_cloner) link;
331 struct tuntap_driver *drv;
332 struct if_clone *cloner;
333 };
334
335 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
336 SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
337
338 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners)
339
340 /*
341 * Mechanism for marking a tunnel device as busy so that we can safely do some
342 * orthogonal operations (such as operations on devices) without racing against
343 * tun_destroy. tun_destroy will wait on the condvar if we're at all busy or
344 * open, to be woken up when the condition is alleviated.
345 */
346 static int
tun_busy_locked(struct tuntap_softc * tp)347 tun_busy_locked(struct tuntap_softc *tp)
348 {
349
350 TUN_LOCK_ASSERT(tp);
351 if ((tp->tun_flags & TUN_DYING) != 0) {
352 /*
353 * Perhaps unintuitive, but the device is busy going away.
354 * Other interpretations of EBUSY from tun_busy make little
355 * sense, since making a busy device even more busy doesn't
356 * sound like a problem.
357 */
358 return (EBUSY);
359 }
360
361 ++tp->tun_busy;
362 return (0);
363 }
364
365 static void
tun_unbusy_locked(struct tuntap_softc * tp)366 tun_unbusy_locked(struct tuntap_softc *tp)
367 {
368
369 TUN_LOCK_ASSERT(tp);
370 KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
371
372 --tp->tun_busy;
373 /* Wake up anything that may be waiting on our busy tunnel. */
374 if (tp->tun_busy == 0)
375 cv_broadcast(&tp->tun_cv);
376 }
377
378 static int
tun_busy(struct tuntap_softc * tp)379 tun_busy(struct tuntap_softc *tp)
380 {
381 int ret;
382
383 TUN_LOCK(tp);
384 ret = tun_busy_locked(tp);
385 TUN_UNLOCK(tp);
386 return (ret);
387 }
388
389 static void
tun_unbusy(struct tuntap_softc * tp)390 tun_unbusy(struct tuntap_softc *tp)
391 {
392
393 TUN_LOCK(tp);
394 tun_unbusy_locked(tp);
395 TUN_UNLOCK(tp);
396 }
397
398 /*
399 * Sets unit and/or flags given the device name. Must be called with correct
400 * vnet context.
401 */
402 static int
tuntap_name2info(const char * name,int * outunit,int * outflags)403 tuntap_name2info(const char *name, int *outunit, int *outflags)
404 {
405 struct tuntap_driver *drv;
406 struct tuntap_driver_cloner *drvc;
407 char *dname;
408 int flags, unit;
409 bool found;
410
411 if (name == NULL)
412 return (EINVAL);
413
414 /*
415 * Needed for dev_stdclone, but dev_stdclone will not modify, it just
416 * wants to be able to pass back a char * through the second param. We
417 * will always set that as NULL here, so we'll fake it.
418 */
419 dname = __DECONST(char *, name);
420 found = false;
421
422 KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
423 ("tuntap_driver_cloners failed to initialize"));
424 SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
425 KASSERT(drvc->drv != NULL,
426 ("tuntap_driver_cloners entry not properly initialized"));
427 drv = drvc->drv;
428
429 if (strcmp(name, drv->cdevsw.d_name) == 0) {
430 found = true;
431 unit = -1;
432 flags = drv->ident_flags;
433 break;
434 }
435
436 if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
437 found = true;
438 flags = drv->ident_flags;
439 break;
440 }
441 }
442
443 if (!found)
444 return (ENXIO);
445
446 if (outunit != NULL)
447 *outunit = unit;
448 if (outflags != NULL)
449 *outflags = flags;
450 return (0);
451 }
452
453 /*
454 * Get driver information from a set of flags specified. Masks the identifying
455 * part of the flags and compares it against all of the available
456 * tuntap_drivers. Must be called with correct vnet context.
457 */
458 static struct tuntap_driver *
tuntap_driver_from_flags(int tun_flags)459 tuntap_driver_from_flags(int tun_flags)
460 {
461 struct tuntap_driver *drv;
462 struct tuntap_driver_cloner *drvc;
463
464 KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
465 ("tuntap_driver_cloners failed to initialize"));
466 SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
467 KASSERT(drvc->drv != NULL,
468 ("tuntap_driver_cloners entry not properly initialized"));
469 drv = drvc->drv;
470 if ((tun_flags & TUN_DRIVER_IDENT_MASK) == drv->ident_flags)
471 return (drv);
472 }
473
474 return (NULL);
475 }
476
477 static int
tun_clone_match(struct if_clone * ifc,const char * name)478 tun_clone_match(struct if_clone *ifc, const char *name)
479 {
480 int tunflags;
481
482 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
483 if ((tunflags & TUN_L2) == 0)
484 return (1);
485 }
486
487 return (0);
488 }
489
490 static int
tap_clone_match(struct if_clone * ifc,const char * name)491 tap_clone_match(struct if_clone *ifc, const char *name)
492 {
493 int tunflags;
494
495 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
496 if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
497 return (1);
498 }
499
500 return (0);
501 }
502
503 static int
vmnet_clone_match(struct if_clone * ifc,const char * name)504 vmnet_clone_match(struct if_clone *ifc, const char *name)
505 {
506 int tunflags;
507
508 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
509 if ((tunflags & TUN_VMNET) != 0)
510 return (1);
511 }
512
513 return (0);
514 }
515
516 static int
tun_clone_create(struct if_clone * ifc,char * name,size_t len,caddr_t params)517 tun_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
518 {
519 struct tuntap_driver *drv;
520 struct cdev *dev;
521 int err, i, tunflags, unit;
522
523 tunflags = 0;
524 /* The name here tells us exactly what we're creating */
525 err = tuntap_name2info(name, &unit, &tunflags);
526 if (err != 0)
527 return (err);
528
529 drv = tuntap_driver_from_flags(tunflags);
530 if (drv == NULL)
531 return (ENXIO);
532
533 if (unit != -1) {
534 /* If this unit number is still available that's okay. */
535 if (alloc_unr_specific(drv->unrhdr, unit) == -1)
536 return (EEXIST);
537 } else {
538 unit = alloc_unr(drv->unrhdr);
539 }
540
541 snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
542
543 /* find any existing device, or allocate new unit number */
544 dev = NULL;
545 i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
546 /* No preexisting struct cdev *, create one */
547 if (i != 0)
548 i = tun_create_device(drv, unit, NULL, &dev, name);
549 if (i == 0)
550 tuncreate(dev);
551
552 return (i);
553 }
554
555 static void
tunclone(void * arg,struct ucred * cred,char * name,int namelen,struct cdev ** dev)556 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
557 struct cdev **dev)
558 {
559 char devname[SPECNAMELEN + 1];
560 struct tuntap_driver *drv;
561 int append_unit, i, u, tunflags;
562 bool mayclone;
563
564 if (*dev != NULL)
565 return;
566
567 tunflags = 0;
568 CURVNET_SET(CRED_TO_VNET(cred));
569 if (tuntap_name2info(name, &u, &tunflags) != 0)
570 goto out; /* Not recognized */
571
572 if (u != -1 && u > IF_MAXUNIT)
573 goto out; /* Unit number too high */
574
575 mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
576 if ((tunflags & TUN_L2) != 0) {
577 /* tap/vmnet allow user open with a sysctl */
578 mayclone = (mayclone || tap_allow_uopen) && tapdclone;
579 } else {
580 mayclone = mayclone && tundclone;
581 }
582
583 /*
584 * If tun cloning is enabled, only the superuser can create an
585 * interface.
586 */
587 if (!mayclone)
588 goto out;
589
590 if (u == -1)
591 append_unit = 1;
592 else
593 append_unit = 0;
594
595 drv = tuntap_driver_from_flags(tunflags);
596 if (drv == NULL)
597 goto out;
598
599 /* find any existing device, or allocate new unit number */
600 i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
601 if (i) {
602 if (append_unit) {
603 namelen = snprintf(devname, sizeof(devname), "%s%d",
604 name, u);
605 name = devname;
606 }
607
608 i = tun_create_device(drv, u, cred, dev, name);
609 }
610 if (i == 0)
611 if_clone_create(name, namelen, NULL);
612 out:
613 CURVNET_RESTORE();
614 }
615
616 static void
tun_destroy(struct tuntap_softc * tp)617 tun_destroy(struct tuntap_softc *tp)
618 {
619
620 TUN_LOCK(tp);
621 tp->tun_flags |= TUN_DYING;
622 if (tp->tun_busy != 0)
623 cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
624 else
625 TUN_UNLOCK(tp);
626
627 CURVNET_SET(TUN2IFP(tp)->if_vnet);
628
629 /* destroy_dev will take care of any alias. */
630 destroy_dev(tp->tun_dev);
631 seldrain(&tp->tun_rsel);
632 knlist_clear(&tp->tun_rsel.si_note, 0);
633 knlist_destroy(&tp->tun_rsel.si_note);
634 if ((tp->tun_flags & TUN_L2) != 0) {
635 ether_ifdetach(TUN2IFP(tp));
636 } else {
637 bpfdetach(TUN2IFP(tp));
638 if_detach(TUN2IFP(tp));
639 }
640 sx_xlock(&tun_ioctl_sx);
641 TUN2IFP(tp)->if_softc = NULL;
642 sx_xunlock(&tun_ioctl_sx);
643 free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
644 if_free(TUN2IFP(tp));
645 mtx_destroy(&tp->tun_mtx);
646 cv_destroy(&tp->tun_cv);
647 free(tp, M_TUN);
648 CURVNET_RESTORE();
649 }
650
651 static int
tun_clone_destroy(struct if_clone * ifc __unused,struct ifnet * ifp)652 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp)
653 {
654 struct tuntap_softc *tp = ifp->if_softc;
655
656 mtx_lock(&tunmtx);
657 TAILQ_REMOVE(&tunhead, tp, tun_list);
658 mtx_unlock(&tunmtx);
659 tun_destroy(tp);
660
661 return (0);
662 }
663
664 static void
vnet_tun_init(const void * unused __unused)665 vnet_tun_init(const void *unused __unused)
666 {
667 struct tuntap_driver *drv;
668 struct tuntap_driver_cloner *drvc;
669 int i;
670
671 for (i = 0; i < nitems(tuntap_drivers); ++i) {
672 drv = &tuntap_drivers[i];
673 drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
674
675 drvc->drv = drv;
676 drvc->cloner = if_clone_advanced(drv->cdevsw.d_name, 0,
677 drv->clone_match_fn, drv->clone_create_fn,
678 drv->clone_destroy_fn);
679 SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
680 };
681 }
682 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
683 vnet_tun_init, NULL);
684
685 static void
vnet_tun_uninit(const void * unused __unused)686 vnet_tun_uninit(const void *unused __unused)
687 {
688 struct tuntap_driver_cloner *drvc;
689
690 while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
691 drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
692 SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
693
694 if_clone_detach(drvc->cloner);
695 free(drvc, M_TUN);
696 }
697 }
698 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
699 vnet_tun_uninit, NULL);
700
701 static void
tun_uninit(const void * unused __unused)702 tun_uninit(const void *unused __unused)
703 {
704 struct tuntap_driver *drv;
705 struct tuntap_softc *tp;
706 int i;
707
708 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
709 EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
710 drain_dev_clone_events();
711
712 mtx_lock(&tunmtx);
713 while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
714 TAILQ_REMOVE(&tunhead, tp, tun_list);
715 mtx_unlock(&tunmtx);
716 tun_destroy(tp);
717 mtx_lock(&tunmtx);
718 }
719 mtx_unlock(&tunmtx);
720 for (i = 0; i < nitems(tuntap_drivers); ++i) {
721 drv = &tuntap_drivers[i];
722 delete_unrhdr(drv->unrhdr);
723 clone_cleanup(&drv->clones);
724 }
725 mtx_destroy(&tunmtx);
726 }
727 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
728
729 static struct tuntap_driver *
tuntap_driver_from_ifnet(const struct ifnet * ifp)730 tuntap_driver_from_ifnet(const struct ifnet *ifp)
731 {
732 struct tuntap_driver *drv;
733 int i;
734
735 if (ifp == NULL)
736 return (NULL);
737
738 for (i = 0; i < nitems(tuntap_drivers); ++i) {
739 drv = &tuntap_drivers[i];
740 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
741 return (drv);
742 }
743
744 return (NULL);
745 }
746
747 static int
tuntapmodevent(module_t mod,int type,void * data)748 tuntapmodevent(module_t mod, int type, void *data)
749 {
750 struct tuntap_driver *drv;
751 int i;
752
753 switch (type) {
754 case MOD_LOAD:
755 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
756 for (i = 0; i < nitems(tuntap_drivers); ++i) {
757 drv = &tuntap_drivers[i];
758 clone_setup(&drv->clones);
759 drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
760 }
761 arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
762 tunrename, 0, 1000);
763 if (arrival_tag == NULL)
764 return (ENOMEM);
765 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
766 if (clone_tag == NULL)
767 return (ENOMEM);
768 break;
769 case MOD_UNLOAD:
770 /* See tun_uninit, so it's done after the vnet_sysuninit() */
771 break;
772 default:
773 return EOPNOTSUPP;
774 }
775 return 0;
776 }
777
778 static moduledata_t tuntap_mod = {
779 "if_tuntap",
780 tuntapmodevent,
781 0
782 };
783
784 /* We'll only ever have these two, so no need for a macro. */
785 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
786 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
787
788 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
789 MODULE_VERSION(if_tuntap, 1);
790 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
791 MODULE_VERSION(if_tun, 1);
792 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
793 MODULE_VERSION(if_tap, 1);
794
795 static int
tun_create_device(struct tuntap_driver * drv,int unit,struct ucred * cr,struct cdev ** dev,const char * name)796 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
797 struct cdev **dev, const char *name)
798 {
799 struct make_dev_args args;
800 struct tuntap_softc *tp;
801 int error;
802
803 tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
804 mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
805 cv_init(&tp->tun_cv, "tun_condvar");
806 tp->tun_flags = drv->ident_flags;
807 tp->tun_drv = drv;
808
809 make_dev_args_init(&args);
810 if (cr != NULL)
811 args.mda_flags = MAKEDEV_REF;
812 args.mda_devsw = &drv->cdevsw;
813 args.mda_cr = cr;
814 args.mda_uid = UID_UUCP;
815 args.mda_gid = GID_DIALER;
816 args.mda_mode = 0600;
817 args.mda_unit = unit;
818 args.mda_si_drv1 = tp;
819 error = make_dev_s(&args, dev, "%s", name);
820 if (error != 0) {
821 free(tp, M_TUN);
822 return (error);
823 }
824
825 KASSERT((*dev)->si_drv1 != NULL,
826 ("Failed to set si_drv1 at %s creation", name));
827 tp->tun_dev = *dev;
828 knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
829 mtx_lock(&tunmtx);
830 TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
831 mtx_unlock(&tunmtx);
832 return (0);
833 }
834
835 static void
tunstart(struct ifnet * ifp)836 tunstart(struct ifnet *ifp)
837 {
838 struct tuntap_softc *tp = ifp->if_softc;
839 struct mbuf *m;
840
841 TUNDEBUG(ifp, "starting\n");
842 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
843 IFQ_LOCK(&ifp->if_snd);
844 IFQ_POLL_NOLOCK(&ifp->if_snd, m);
845 if (m == NULL) {
846 IFQ_UNLOCK(&ifp->if_snd);
847 return;
848 }
849 IFQ_UNLOCK(&ifp->if_snd);
850 }
851
852 TUN_LOCK(tp);
853 if (tp->tun_flags & TUN_RWAIT) {
854 tp->tun_flags &= ~TUN_RWAIT;
855 wakeup(tp);
856 }
857 selwakeuppri(&tp->tun_rsel, PZERO + 1);
858 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
859 if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
860 TUN_UNLOCK(tp);
861 pgsigio(&tp->tun_sigio, SIGIO, 0);
862 } else
863 TUN_UNLOCK(tp);
864 }
865
866 /*
867 * tunstart_l2
868 *
869 * queue packets from higher level ready to put out
870 */
871 static void
tunstart_l2(struct ifnet * ifp)872 tunstart_l2(struct ifnet *ifp)
873 {
874 struct tuntap_softc *tp = ifp->if_softc;
875
876 TUNDEBUG(ifp, "starting\n");
877
878 /*
879 * do not junk pending output if we are in VMnet mode.
880 * XXX: can this do any harm because of queue overflow?
881 */
882
883 TUN_LOCK(tp);
884 if (((tp->tun_flags & TUN_VMNET) == 0) &&
885 ((tp->tun_flags & TUN_READY) != TUN_READY)) {
886 struct mbuf *m;
887
888 /* Unlocked read. */
889 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
890
891 for (;;) {
892 IF_DEQUEUE(&ifp->if_snd, m);
893 if (m != NULL) {
894 m_freem(m);
895 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
896 } else
897 break;
898 }
899 TUN_UNLOCK(tp);
900
901 return;
902 }
903
904 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
905
906 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
907 if (tp->tun_flags & TUN_RWAIT) {
908 tp->tun_flags &= ~TUN_RWAIT;
909 wakeup(tp);
910 }
911
912 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
913 TUN_UNLOCK(tp);
914 pgsigio(&tp->tun_sigio, SIGIO, 0);
915 TUN_LOCK(tp);
916 }
917
918 selwakeuppri(&tp->tun_rsel, PZERO+1);
919 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
920 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
921 }
922
923 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
924 TUN_UNLOCK(tp);
925 } /* tunstart_l2 */
926
927 /* XXX: should return an error code so it can fail. */
928 static void
tuncreate(struct cdev * dev)929 tuncreate(struct cdev *dev)
930 {
931 struct tuntap_driver *drv;
932 struct tuntap_softc *tp;
933 struct ifnet *ifp;
934 struct ether_addr eaddr;
935 int iflags;
936 u_char type;
937
938 tp = dev->si_drv1;
939 KASSERT(tp != NULL,
940 ("si_drv1 should have been initialized at creation"));
941
942 drv = tp->tun_drv;
943 iflags = IFF_MULTICAST;
944 if ((tp->tun_flags & TUN_L2) != 0) {
945 type = IFT_ETHER;
946 iflags |= IFF_BROADCAST | IFF_SIMPLEX;
947 } else {
948 type = IFT_PPP;
949 iflags |= IFF_POINTOPOINT;
950 }
951 ifp = tp->tun_ifp = if_alloc(type);
952 if (ifp == NULL)
953 panic("%s%d: failed to if_alloc() interface.\n",
954 drv->cdevsw.d_name, dev2unit(dev));
955 ifp->if_softc = tp;
956 if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
957 ifp->if_ioctl = tunifioctl;
958 ifp->if_flags = iflags;
959 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
960 ifp->if_capabilities |= IFCAP_LINKSTATE;
961 ifp->if_capenable |= IFCAP_LINKSTATE;
962
963 if ((tp->tun_flags & TUN_L2) != 0) {
964 ifp->if_init = tunifinit;
965 ifp->if_start = tunstart_l2;
966
967 ether_gen_addr(ifp, &eaddr);
968 ether_ifattach(ifp, eaddr.octet);
969 } else {
970 ifp->if_mtu = TUNMTU;
971 ifp->if_start = tunstart;
972 ifp->if_output = tunoutput;
973
974 ifp->if_snd.ifq_drv_maxlen = 0;
975 IFQ_SET_READY(&ifp->if_snd);
976
977 if_attach(ifp);
978 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
979 }
980
981 TUN_LOCK(tp);
982 tp->tun_flags |= TUN_INITED;
983 TUN_UNLOCK(tp);
984
985 TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
986 ifp->if_xname, dev2unit(dev));
987 }
988
989 static void
tunrename(void * arg __unused,struct ifnet * ifp)990 tunrename(void *arg __unused, struct ifnet *ifp)
991 {
992 struct tuntap_softc *tp;
993 int error;
994
995 if ((ifp->if_flags & IFF_RENAMING) == 0)
996 return;
997
998 if (tuntap_driver_from_ifnet(ifp) == NULL)
999 return;
1000
1001 /*
1002 * We need to grab the ioctl sx long enough to make sure the softc is
1003 * still there. If it is, we can safely try to busy the tun device.
1004 * The busy may fail if the device is currently dying, in which case
1005 * we do nothing. If it doesn't fail, the busy count stops the device
1006 * from dying until we've created the alias (that will then be
1007 * subsequently destroyed).
1008 */
1009 sx_xlock(&tun_ioctl_sx);
1010 tp = ifp->if_softc;
1011 if (tp == NULL) {
1012 sx_xunlock(&tun_ioctl_sx);
1013 return;
1014 }
1015 error = tun_busy(tp);
1016 sx_xunlock(&tun_ioctl_sx);
1017 if (error != 0)
1018 return;
1019 if (tp->tun_alias != NULL) {
1020 destroy_dev(tp->tun_alias);
1021 tp->tun_alias = NULL;
1022 }
1023
1024 if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1025 goto out;
1026
1027 /*
1028 * Failure's ok, aliases are created on a best effort basis. If a
1029 * tun user/consumer decides to rename the interface to conflict with
1030 * another device (non-ifnet) on the system, we will assume they know
1031 * what they are doing. make_dev_alias_p won't touch tun_alias on
1032 * failure, so we use it but ignore the return value.
1033 */
1034 make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1035 ifp->if_xname);
1036 out:
1037 tun_unbusy(tp);
1038 }
1039
1040 static int
tunopen(struct cdev * dev,int flag,int mode,struct thread * td)1041 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1042 {
1043 struct ifnet *ifp;
1044 struct tuntap_softc *tp;
1045 int error, tunflags;
1046
1047 tunflags = 0;
1048 CURVNET_SET(TD_TO_VNET(td));
1049 error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1050 if (error != 0) {
1051 CURVNET_RESTORE();
1052 return (error); /* Shouldn't happen */
1053 }
1054
1055 tp = dev->si_drv1;
1056 KASSERT(tp != NULL,
1057 ("si_drv1 should have been initialized at creation"));
1058
1059 TUN_LOCK(tp);
1060 if ((tp->tun_flags & TUN_INITED) == 0) {
1061 TUN_UNLOCK(tp);
1062 CURVNET_RESTORE();
1063 return (ENXIO);
1064 }
1065 if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1066 TUN_UNLOCK(tp);
1067 CURVNET_RESTORE();
1068 return (EBUSY);
1069 }
1070
1071 error = tun_busy_locked(tp);
1072 KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1073 ifp = TUN2IFP(tp);
1074
1075 if ((tp->tun_flags & TUN_L2) != 0) {
1076 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1077 sizeof(tp->tun_ether.octet));
1078
1079 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1080 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1081
1082 if (tapuponopen)
1083 ifp->if_flags |= IFF_UP;
1084 }
1085
1086 tp->tun_pid = td->td_proc->p_pid;
1087 tp->tun_flags |= TUN_OPEN;
1088
1089 if_link_state_change(ifp, LINK_STATE_UP);
1090 TUNDEBUG(ifp, "open\n");
1091 TUN_UNLOCK(tp);
1092
1093 /*
1094 * This can fail with either ENOENT or EBUSY. This is in the middle of
1095 * d_open, so ENOENT should not be possible. EBUSY is possible, but
1096 * the only cdevpriv dtor being set will be tundtor and the softc being
1097 * passed is constant for a given cdev. We ignore the possible error
1098 * because of this as either "unlikely" or "not actually a problem."
1099 */
1100 (void)devfs_set_cdevpriv(tp, tundtor);
1101 CURVNET_RESTORE();
1102 return (0);
1103 }
1104
1105 /*
1106 * tundtor - tear down the device - mark i/f down & delete
1107 * routing info
1108 */
1109 static void
tundtor(void * data)1110 tundtor(void *data)
1111 {
1112 struct proc *p;
1113 struct tuntap_softc *tp;
1114 struct ifnet *ifp;
1115 bool l2tun;
1116
1117 tp = data;
1118 p = curproc;
1119 ifp = TUN2IFP(tp);
1120
1121 TUN_LOCK(tp);
1122
1123 /*
1124 * Realistically, we can't be obstinate here. This only means that the
1125 * tuntap device was closed out of order, and the last closer wasn't the
1126 * controller. These are still good to know about, though, as software
1127 * should avoid multiple processes with a tuntap device open and
1128 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1129 * parent).
1130 */
1131 if (p->p_pid != tp->tun_pid) {
1132 log(LOG_INFO,
1133 "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1134 p->p_pid, p->p_comm, tp->tun_dev->si_name);
1135 }
1136
1137 /*
1138 * junk all pending output
1139 */
1140 CURVNET_SET(ifp->if_vnet);
1141
1142 l2tun = false;
1143 if ((tp->tun_flags & TUN_L2) != 0) {
1144 l2tun = true;
1145 IF_DRAIN(&ifp->if_snd);
1146 } else {
1147 IFQ_PURGE(&ifp->if_snd);
1148 }
1149
1150 /* For vmnet, we won't do most of the address/route bits */
1151 if ((tp->tun_flags & TUN_VMNET) != 0 ||
1152 (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1153 goto out;
1154
1155 if (ifp->if_flags & IFF_UP) {
1156 TUN_UNLOCK(tp);
1157 if_down(ifp);
1158 TUN_LOCK(tp);
1159 }
1160
1161 /* Delete all addresses and routes which reference this interface. */
1162 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1163 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1164 TUN_UNLOCK(tp);
1165 if_purgeaddrs(ifp);
1166 TUN_LOCK(tp);
1167 }
1168
1169 out:
1170 if_link_state_change(ifp, LINK_STATE_DOWN);
1171 CURVNET_RESTORE();
1172
1173 funsetown(&tp->tun_sigio);
1174 selwakeuppri(&tp->tun_rsel, PZERO + 1);
1175 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1176 TUNDEBUG (ifp, "closed\n");
1177 tp->tun_flags &= ~TUN_OPEN;
1178 tp->tun_pid = 0;
1179 tun_vnethdr_set(ifp, 0);
1180
1181 tun_unbusy_locked(tp);
1182 TUN_UNLOCK(tp);
1183 }
1184
1185 static void
tuninit(struct ifnet * ifp)1186 tuninit(struct ifnet *ifp)
1187 {
1188 struct tuntap_softc *tp = ifp->if_softc;
1189
1190 TUNDEBUG(ifp, "tuninit\n");
1191
1192 TUN_LOCK(tp);
1193 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1194 if ((tp->tun_flags & TUN_L2) == 0) {
1195 ifp->if_flags |= IFF_UP;
1196 getmicrotime(&ifp->if_lastchange);
1197 TUN_UNLOCK(tp);
1198 } else {
1199 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1200 TUN_UNLOCK(tp);
1201 /* attempt to start output */
1202 tunstart_l2(ifp);
1203 }
1204
1205 }
1206
1207 /*
1208 * Used only for l2 tunnel.
1209 */
1210 static void
tunifinit(void * xtp)1211 tunifinit(void *xtp)
1212 {
1213 struct tuntap_softc *tp;
1214
1215 tp = (struct tuntap_softc *)xtp;
1216 tuninit(tp->tun_ifp);
1217 }
1218
1219 /*
1220 * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1221 * current value of ifp->if_capenable.
1222 */
1223 static void
tun_caps_changed(struct ifnet * ifp)1224 tun_caps_changed(struct ifnet *ifp)
1225 {
1226 uint64_t hwassist = 0;
1227
1228 TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1229 if (ifp->if_capenable & IFCAP_TXCSUM)
1230 hwassist |= CSUM_TCP | CSUM_UDP;
1231 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1232 hwassist |= CSUM_TCP_IPV6
1233 | CSUM_UDP_IPV6;
1234 if (ifp->if_capenable & IFCAP_TSO4)
1235 hwassist |= CSUM_IP_TSO;
1236 if (ifp->if_capenable & IFCAP_TSO6)
1237 hwassist |= CSUM_IP6_TSO;
1238 ifp->if_hwassist = hwassist;
1239 }
1240
1241 /*
1242 * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1243 * if_capabilities and if_capenable as needed.
1244 */
1245 static void
tun_vnethdr_set(struct ifnet * ifp,int vhdrlen)1246 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1247 {
1248 struct tuntap_softc *tp = ifp->if_softc;
1249
1250 TUN_LOCK_ASSERT(tp);
1251
1252 if (tp->tun_vhdrlen == vhdrlen)
1253 return;
1254
1255 /*
1256 * Update if_capabilities to reflect the
1257 * functionalities offered by the virtio-net
1258 * header.
1259 */
1260 if (vhdrlen != 0)
1261 ifp->if_capabilities |=
1262 TAP_VNET_HDR_CAPS;
1263 else
1264 ifp->if_capabilities &=
1265 ~TAP_VNET_HDR_CAPS;
1266 /*
1267 * Disable any capabilities that we don't
1268 * support anymore.
1269 */
1270 ifp->if_capenable &= ifp->if_capabilities;
1271 tun_caps_changed(ifp);
1272 tp->tun_vhdrlen = vhdrlen;
1273
1274 TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1275 vhdrlen, ifp->if_capabilities);
1276 }
1277
1278 /*
1279 * Process an ioctl request.
1280 */
1281 static int
tunifioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1282 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1283 {
1284 struct ifreq *ifr = (struct ifreq *)data;
1285 struct tuntap_softc *tp;
1286 struct ifstat *ifs;
1287 struct ifmediareq *ifmr;
1288 int dummy, error = 0;
1289 bool l2tun;
1290
1291 ifmr = NULL;
1292 sx_xlock(&tun_ioctl_sx);
1293 tp = ifp->if_softc;
1294 if (tp == NULL) {
1295 error = ENXIO;
1296 goto bad;
1297 }
1298 l2tun = (tp->tun_flags & TUN_L2) != 0;
1299 switch(cmd) {
1300 case SIOCGIFSTATUS:
1301 ifs = (struct ifstat *)data;
1302 TUN_LOCK(tp);
1303 if (tp->tun_pid)
1304 snprintf(ifs->ascii, sizeof(ifs->ascii),
1305 "\tOpened by PID %d\n", tp->tun_pid);
1306 else
1307 ifs->ascii[0] = '\0';
1308 TUN_UNLOCK(tp);
1309 break;
1310 case SIOCSIFADDR:
1311 if (l2tun)
1312 error = ether_ioctl(ifp, cmd, data);
1313 else
1314 tuninit(ifp);
1315 if (error == 0)
1316 TUNDEBUG(ifp, "address set\n");
1317 break;
1318 case SIOCSIFMTU:
1319 ifp->if_mtu = ifr->ifr_mtu;
1320 TUNDEBUG(ifp, "mtu set\n");
1321 break;
1322 case SIOCSIFFLAGS:
1323 case SIOCADDMULTI:
1324 case SIOCDELMULTI:
1325 break;
1326 case SIOCGIFMEDIA:
1327 if (!l2tun) {
1328 error = EINVAL;
1329 break;
1330 }
1331
1332 ifmr = (struct ifmediareq *)data;
1333 dummy = ifmr->ifm_count;
1334 ifmr->ifm_count = 1;
1335 ifmr->ifm_status = IFM_AVALID;
1336 ifmr->ifm_active = IFM_ETHER;
1337 if (tp->tun_flags & TUN_OPEN)
1338 ifmr->ifm_status |= IFM_ACTIVE;
1339 ifmr->ifm_current = ifmr->ifm_active;
1340 if (dummy >= 1) {
1341 int media = IFM_ETHER;
1342 error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1343 }
1344 break;
1345 case SIOCSIFCAP:
1346 TUN_LOCK(tp);
1347 ifp->if_capenable = ifr->ifr_reqcap;
1348 tun_caps_changed(ifp);
1349 TUN_UNLOCK(tp);
1350 VLAN_CAPABILITIES(ifp);
1351 break;
1352 default:
1353 if (l2tun) {
1354 error = ether_ioctl(ifp, cmd, data);
1355 } else {
1356 error = EINVAL;
1357 }
1358 }
1359 bad:
1360 sx_xunlock(&tun_ioctl_sx);
1361 return (error);
1362 }
1363
1364 /*
1365 * tunoutput - queue packets from higher level ready to put out.
1366 */
1367 static int
tunoutput(struct ifnet * ifp,struct mbuf * m0,const struct sockaddr * dst,struct route * ro)1368 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1369 struct route *ro)
1370 {
1371 struct tuntap_softc *tp = ifp->if_softc;
1372 u_short cached_tun_flags;
1373 int error;
1374 u_int32_t af;
1375
1376 TUNDEBUG (ifp, "tunoutput\n");
1377
1378 #ifdef MAC
1379 error = mac_ifnet_check_transmit(ifp, m0);
1380 if (error) {
1381 m_freem(m0);
1382 return (error);
1383 }
1384 #endif
1385
1386 /* Could be unlocked read? */
1387 TUN_LOCK(tp);
1388 cached_tun_flags = tp->tun_flags;
1389 TUN_UNLOCK(tp);
1390 if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1391 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1392 m_freem (m0);
1393 return (EHOSTDOWN);
1394 }
1395
1396 if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1397 m_freem (m0);
1398 return (EHOSTDOWN);
1399 }
1400
1401 /* BPF writes need to be handled specially. */
1402 if (dst->sa_family == AF_UNSPEC)
1403 bcopy(dst->sa_data, &af, sizeof(af));
1404 else
1405 af = dst->sa_family;
1406
1407 if (bpf_peers_present(ifp->if_bpf))
1408 bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1409
1410 /* prepend sockaddr? this may abort if the mbuf allocation fails */
1411 if (cached_tun_flags & TUN_LMODE) {
1412 /* allocate space for sockaddr */
1413 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1414
1415 /* if allocation failed drop packet */
1416 if (m0 == NULL) {
1417 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1418 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1419 return (ENOBUFS);
1420 } else {
1421 bcopy(dst, m0->m_data, dst->sa_len);
1422 }
1423 }
1424
1425 if (cached_tun_flags & TUN_IFHEAD) {
1426 /* Prepend the address family */
1427 M_PREPEND(m0, 4, M_NOWAIT);
1428
1429 /* if allocation failed drop packet */
1430 if (m0 == NULL) {
1431 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1432 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1433 return (ENOBUFS);
1434 } else
1435 *(u_int32_t *)m0->m_data = htonl(af);
1436 } else {
1437 #ifdef INET
1438 if (af != AF_INET)
1439 #endif
1440 {
1441 m_freem(m0);
1442 return (EAFNOSUPPORT);
1443 }
1444 }
1445
1446 error = (ifp->if_transmit)(ifp, m0);
1447 if (error)
1448 return (ENOBUFS);
1449 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1450 return (0);
1451 }
1452
1453 /*
1454 * the cdevsw interface is now pretty minimal.
1455 */
1456 static int
tunioctl(struct cdev * dev,u_long cmd,caddr_t data,int flag,struct thread * td)1457 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1458 struct thread *td)
1459 {
1460 struct ifreq ifr, *ifrp;
1461 struct tuntap_softc *tp = dev->si_drv1;
1462 struct ifnet *ifp = TUN2IFP(tp);
1463 struct tuninfo *tunp;
1464 int error, iflags, ival;
1465 bool l2tun;
1466
1467 l2tun = (tp->tun_flags & TUN_L2) != 0;
1468 if (l2tun) {
1469 /* tap specific ioctls */
1470 switch(cmd) {
1471 /* VMware/VMnet port ioctl's */
1472 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1473 defined(COMPAT_FREEBSD4)
1474 case _IO('V', 0):
1475 ival = IOCPARM_IVAL(data);
1476 data = (caddr_t)&ival;
1477 /* FALLTHROUGH */
1478 #endif
1479 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1480 iflags = *(int *)data;
1481 iflags &= TUN_VMIO_FLAG_MASK;
1482 iflags &= ~IFF_CANTCHANGE;
1483 iflags |= IFF_UP;
1484
1485 TUN_LOCK(tp);
1486 ifp->if_flags = iflags |
1487 (ifp->if_flags & IFF_CANTCHANGE);
1488 TUN_UNLOCK(tp);
1489
1490 return (0);
1491 case SIOCGIFADDR: /* get MAC address of the remote side */
1492 TUN_LOCK(tp);
1493 bcopy(&tp->tun_ether.octet, data,
1494 sizeof(tp->tun_ether.octet));
1495 TUN_UNLOCK(tp);
1496
1497 return (0);
1498 case SIOCSIFADDR: /* set MAC address of the remote side */
1499 TUN_LOCK(tp);
1500 bcopy(data, &tp->tun_ether.octet,
1501 sizeof(tp->tun_ether.octet));
1502 TUN_UNLOCK(tp);
1503
1504 return (0);
1505 case TAPSVNETHDR:
1506 ival = *(int *)data;
1507 if (ival != 0 &&
1508 ival != sizeof(struct virtio_net_hdr) &&
1509 ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1510 return (EINVAL);
1511 }
1512 TUN_LOCK(tp);
1513 tun_vnethdr_set(ifp, ival);
1514 TUN_UNLOCK(tp);
1515
1516 return (0);
1517 case TAPGVNETHDR:
1518 TUN_LOCK(tp);
1519 *(int *)data = tp->tun_vhdrlen;
1520 TUN_UNLOCK(tp);
1521
1522 return (0);
1523 }
1524
1525 /* Fall through to the common ioctls if unhandled */
1526 } else {
1527 switch (cmd) {
1528 case TUNSLMODE:
1529 TUN_LOCK(tp);
1530 if (*(int *)data) {
1531 tp->tun_flags |= TUN_LMODE;
1532 tp->tun_flags &= ~TUN_IFHEAD;
1533 } else
1534 tp->tun_flags &= ~TUN_LMODE;
1535 TUN_UNLOCK(tp);
1536
1537 return (0);
1538 case TUNSIFHEAD:
1539 TUN_LOCK(tp);
1540 if (*(int *)data) {
1541 tp->tun_flags |= TUN_IFHEAD;
1542 tp->tun_flags &= ~TUN_LMODE;
1543 } else
1544 tp->tun_flags &= ~TUN_IFHEAD;
1545 TUN_UNLOCK(tp);
1546
1547 return (0);
1548 case TUNGIFHEAD:
1549 TUN_LOCK(tp);
1550 *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1551 TUN_UNLOCK(tp);
1552
1553 return (0);
1554 case TUNSIFMODE:
1555 /* deny this if UP */
1556 if (TUN2IFP(tp)->if_flags & IFF_UP)
1557 return (EBUSY);
1558
1559 switch (*(int *)data & ~IFF_MULTICAST) {
1560 case IFF_POINTOPOINT:
1561 case IFF_BROADCAST:
1562 TUN_LOCK(tp);
1563 TUN2IFP(tp)->if_flags &=
1564 ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1565 TUN2IFP(tp)->if_flags |= *(int *)data;
1566 TUN_UNLOCK(tp);
1567
1568 break;
1569 default:
1570 return (EINVAL);
1571 }
1572
1573 return (0);
1574 case TUNSIFPID:
1575 TUN_LOCK(tp);
1576 tp->tun_pid = curthread->td_proc->p_pid;
1577 TUN_UNLOCK(tp);
1578
1579 return (0);
1580 }
1581 /* Fall through to the common ioctls if unhandled */
1582 }
1583
1584 switch (cmd) {
1585 case TUNGIFNAME:
1586 ifrp = (struct ifreq *)data;
1587 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1588
1589 return (0);
1590 case TUNSIFINFO:
1591 tunp = (struct tuninfo *)data;
1592 if (TUN2IFP(tp)->if_type != tunp->type)
1593 return (EPROTOTYPE);
1594 TUN_LOCK(tp);
1595 if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1596 strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1597 ifr.ifr_mtu = tunp->mtu;
1598 CURVNET_SET(TUN2IFP(tp)->if_vnet);
1599 error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1600 (caddr_t)&ifr, td);
1601 CURVNET_RESTORE();
1602 if (error) {
1603 TUN_UNLOCK(tp);
1604 return (error);
1605 }
1606 }
1607 TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1608 TUN_UNLOCK(tp);
1609 break;
1610 case TUNGIFINFO:
1611 tunp = (struct tuninfo *)data;
1612 TUN_LOCK(tp);
1613 tunp->mtu = TUN2IFP(tp)->if_mtu;
1614 tunp->type = TUN2IFP(tp)->if_type;
1615 tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1616 TUN_UNLOCK(tp);
1617 break;
1618 case TUNSDEBUG:
1619 tundebug = *(int *)data;
1620 break;
1621 case TUNGDEBUG:
1622 *(int *)data = tundebug;
1623 break;
1624 case FIONBIO:
1625 break;
1626 case FIOASYNC:
1627 TUN_LOCK(tp);
1628 if (*(int *)data)
1629 tp->tun_flags |= TUN_ASYNC;
1630 else
1631 tp->tun_flags &= ~TUN_ASYNC;
1632 TUN_UNLOCK(tp);
1633 break;
1634 case FIONREAD:
1635 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1636 struct mbuf *mb;
1637 IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1638 IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1639 for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1640 *(int *)data += mb->m_len;
1641 IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1642 } else
1643 *(int *)data = 0;
1644 break;
1645 case FIOSETOWN:
1646 return (fsetown(*(int *)data, &tp->tun_sigio));
1647
1648 case FIOGETOWN:
1649 *(int *)data = fgetown(&tp->tun_sigio);
1650 return (0);
1651
1652 /* This is deprecated, FIOSETOWN should be used instead. */
1653 case TIOCSPGRP:
1654 return (fsetown(-(*(int *)data), &tp->tun_sigio));
1655
1656 /* This is deprecated, FIOGETOWN should be used instead. */
1657 case TIOCGPGRP:
1658 *(int *)data = -fgetown(&tp->tun_sigio);
1659 return (0);
1660
1661 default:
1662 return (ENOTTY);
1663 }
1664 return (0);
1665 }
1666
1667 /*
1668 * The cdevsw read interface - reads a packet at a time, or at
1669 * least as much of a packet as can be read.
1670 */
1671 static int
tunread(struct cdev * dev,struct uio * uio,int flag)1672 tunread(struct cdev *dev, struct uio *uio, int flag)
1673 {
1674 struct tuntap_softc *tp = dev->si_drv1;
1675 struct ifnet *ifp = TUN2IFP(tp);
1676 struct mbuf *m;
1677 size_t len;
1678 int error = 0;
1679
1680 TUNDEBUG (ifp, "read\n");
1681 TUN_LOCK(tp);
1682 if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1683 TUN_UNLOCK(tp);
1684 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1685 return (EHOSTDOWN);
1686 }
1687
1688 tp->tun_flags &= ~TUN_RWAIT;
1689
1690 for (;;) {
1691 IFQ_DEQUEUE(&ifp->if_snd, m);
1692 if (m != NULL)
1693 break;
1694 if (flag & O_NONBLOCK) {
1695 TUN_UNLOCK(tp);
1696 return (EWOULDBLOCK);
1697 }
1698 tp->tun_flags |= TUN_RWAIT;
1699 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1700 "tunread", 0);
1701 if (error != 0) {
1702 TUN_UNLOCK(tp);
1703 return (error);
1704 }
1705 }
1706 TUN_UNLOCK(tp);
1707
1708 if ((tp->tun_flags & TUN_L2) != 0)
1709 BPF_MTAP(ifp, m);
1710
1711 len = min(tp->tun_vhdrlen, uio->uio_resid);
1712 if (len > 0) {
1713 struct virtio_net_hdr_mrg_rxbuf vhdr;
1714
1715 bzero(&vhdr, sizeof(vhdr));
1716 if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1717 m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1718 }
1719
1720 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1721 "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1722 vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1723 vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1724 vhdr.hdr.csum_offset);
1725 error = uiomove(&vhdr, len, uio);
1726 }
1727
1728 while (m && uio->uio_resid > 0 && error == 0) {
1729 len = min(uio->uio_resid, m->m_len);
1730 if (len != 0)
1731 error = uiomove(mtod(m, void *), len, uio);
1732 m = m_free(m);
1733 }
1734
1735 if (m) {
1736 TUNDEBUG(ifp, "Dropping mbuf\n");
1737 m_freem(m);
1738 }
1739 return (error);
1740 }
1741
1742 static int
tunwrite_l2(struct tuntap_softc * tp,struct mbuf * m,struct virtio_net_hdr_mrg_rxbuf * vhdr)1743 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1744 struct virtio_net_hdr_mrg_rxbuf *vhdr)
1745 {
1746 struct epoch_tracker et;
1747 struct ether_header *eh;
1748 struct ifnet *ifp;
1749
1750 ifp = TUN2IFP(tp);
1751
1752 /*
1753 * Only pass a unicast frame to ether_input(), if it would
1754 * actually have been received by non-virtual hardware.
1755 */
1756 if (m->m_len < sizeof(struct ether_header)) {
1757 m_freem(m);
1758 return (0);
1759 }
1760
1761 eh = mtod(m, struct ether_header *);
1762
1763 if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1764 !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1765 bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1766 m_freem(m);
1767 return (0);
1768 }
1769
1770 if (vhdr != NULL && virtio_net_rx_csum(m, &vhdr->hdr)) {
1771 m_freem(m);
1772 return (0);
1773 }
1774
1775 /* Pass packet up to parent. */
1776 CURVNET_SET(ifp->if_vnet);
1777 NET_EPOCH_ENTER(et);
1778 (*ifp->if_input)(ifp, m);
1779 NET_EPOCH_EXIT(et);
1780 CURVNET_RESTORE();
1781 /* ibytes are counted in parent */
1782 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1783 return (0);
1784 }
1785
1786 static int
tunwrite_l3(struct tuntap_softc * tp,struct mbuf * m)1787 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1788 {
1789 struct epoch_tracker et;
1790 struct ifnet *ifp;
1791 int family, isr;
1792
1793 ifp = TUN2IFP(tp);
1794 /* Could be unlocked read? */
1795 TUN_LOCK(tp);
1796 if (tp->tun_flags & TUN_IFHEAD) {
1797 TUN_UNLOCK(tp);
1798 if (m->m_len < sizeof(family) &&
1799 (m = m_pullup(m, sizeof(family))) == NULL)
1800 return (ENOBUFS);
1801 family = ntohl(*mtod(m, u_int32_t *));
1802 m_adj(m, sizeof(family));
1803 } else {
1804 TUN_UNLOCK(tp);
1805 family = AF_INET;
1806 }
1807
1808 BPF_MTAP2(ifp, &family, sizeof(family), m);
1809
1810 switch (family) {
1811 #ifdef INET
1812 case AF_INET:
1813 isr = NETISR_IP;
1814 break;
1815 #endif
1816 #ifdef INET6
1817 case AF_INET6:
1818 isr = NETISR_IPV6;
1819 break;
1820 #endif
1821 default:
1822 m_freem(m);
1823 return (EAFNOSUPPORT);
1824 }
1825 random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1826 if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1827 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1828 CURVNET_SET(ifp->if_vnet);
1829 M_SETFIB(m, ifp->if_fib);
1830 NET_EPOCH_ENTER(et);
1831 netisr_dispatch(isr, m);
1832 NET_EPOCH_EXIT(et);
1833 CURVNET_RESTORE();
1834 return (0);
1835 }
1836
1837 /*
1838 * the cdevsw write interface - an atomic write is a packet - or else!
1839 */
1840 static int
tunwrite(struct cdev * dev,struct uio * uio,int flag)1841 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1842 {
1843 struct virtio_net_hdr_mrg_rxbuf vhdr;
1844 struct tuntap_softc *tp;
1845 struct ifnet *ifp;
1846 struct mbuf *m;
1847 uint32_t mru;
1848 int align, vhdrlen, error;
1849 bool l2tun;
1850
1851 tp = dev->si_drv1;
1852 ifp = TUN2IFP(tp);
1853 TUNDEBUG(ifp, "tunwrite\n");
1854 if ((ifp->if_flags & IFF_UP) != IFF_UP)
1855 /* ignore silently */
1856 return (0);
1857
1858 if (uio->uio_resid == 0)
1859 return (0);
1860
1861 l2tun = (tp->tun_flags & TUN_L2) != 0;
1862 mru = l2tun ? TAPMRU : TUNMRU;
1863 vhdrlen = tp->tun_vhdrlen;
1864 align = 0;
1865 if (l2tun) {
1866 align = ETHER_ALIGN;
1867 mru += vhdrlen;
1868 } else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1869 mru += sizeof(uint32_t); /* family */
1870 if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1871 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1872 return (EIO);
1873 }
1874
1875 if (vhdrlen > 0) {
1876 error = uiomove(&vhdr, vhdrlen, uio);
1877 if (error != 0)
1878 return (error);
1879 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1880 "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1881 vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1882 vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1883 vhdr.hdr.csum_offset);
1884 }
1885
1886 if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1887 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1888 return (ENOBUFS);
1889 }
1890
1891 m->m_pkthdr.rcvif = ifp;
1892 #ifdef MAC
1893 mac_ifnet_create_mbuf(ifp, m);
1894 #endif
1895
1896 if (l2tun)
1897 return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1898
1899 return (tunwrite_l3(tp, m));
1900 }
1901
1902 /*
1903 * tunpoll - the poll interface, this is only useful on reads
1904 * really. The write detect always returns true, write never blocks
1905 * anyway, it either accepts the packet or drops it.
1906 */
1907 static int
tunpoll(struct cdev * dev,int events,struct thread * td)1908 tunpoll(struct cdev *dev, int events, struct thread *td)
1909 {
1910 struct tuntap_softc *tp = dev->si_drv1;
1911 struct ifnet *ifp = TUN2IFP(tp);
1912 int revents = 0;
1913
1914 TUNDEBUG(ifp, "tunpoll\n");
1915
1916 if (events & (POLLIN | POLLRDNORM)) {
1917 IFQ_LOCK(&ifp->if_snd);
1918 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1919 TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1920 revents |= events & (POLLIN | POLLRDNORM);
1921 } else {
1922 TUNDEBUG(ifp, "tunpoll waiting\n");
1923 selrecord(td, &tp->tun_rsel);
1924 }
1925 IFQ_UNLOCK(&ifp->if_snd);
1926 }
1927 revents |= events & (POLLOUT | POLLWRNORM);
1928
1929 return (revents);
1930 }
1931
1932 /*
1933 * tunkqfilter - support for the kevent() system call.
1934 */
1935 static int
tunkqfilter(struct cdev * dev,struct knote * kn)1936 tunkqfilter(struct cdev *dev, struct knote *kn)
1937 {
1938 struct tuntap_softc *tp = dev->si_drv1;
1939 struct ifnet *ifp = TUN2IFP(tp);
1940
1941 switch(kn->kn_filter) {
1942 case EVFILT_READ:
1943 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1944 ifp->if_xname, dev2unit(dev));
1945 kn->kn_fop = &tun_read_filterops;
1946 break;
1947
1948 case EVFILT_WRITE:
1949 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1950 ifp->if_xname, dev2unit(dev));
1951 kn->kn_fop = &tun_write_filterops;
1952 break;
1953
1954 default:
1955 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1956 ifp->if_xname, dev2unit(dev));
1957 return(EINVAL);
1958 }
1959
1960 kn->kn_hook = tp;
1961 knlist_add(&tp->tun_rsel.si_note, kn, 0);
1962
1963 return (0);
1964 }
1965
1966 /*
1967 * Return true of there is data in the interface queue.
1968 */
1969 static int
tunkqread(struct knote * kn,long hint)1970 tunkqread(struct knote *kn, long hint)
1971 {
1972 int ret;
1973 struct tuntap_softc *tp = kn->kn_hook;
1974 struct cdev *dev = tp->tun_dev;
1975 struct ifnet *ifp = TUN2IFP(tp);
1976
1977 if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
1978 TUNDEBUG(ifp,
1979 "%s have data in the queue. Len = %d, minor = %#x\n",
1980 ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
1981 ret = 1;
1982 } else {
1983 TUNDEBUG(ifp,
1984 "%s waiting for data, minor = %#x\n", ifp->if_xname,
1985 dev2unit(dev));
1986 ret = 0;
1987 }
1988
1989 return (ret);
1990 }
1991
1992 /*
1993 * Always can write, always return MTU in kn->data.
1994 */
1995 static int
tunkqwrite(struct knote * kn,long hint)1996 tunkqwrite(struct knote *kn, long hint)
1997 {
1998 struct tuntap_softc *tp = kn->kn_hook;
1999 struct ifnet *ifp = TUN2IFP(tp);
2000
2001 kn->kn_data = ifp->if_mtu;
2002
2003 return (1);
2004 }
2005
2006 static void
tunkqdetach(struct knote * kn)2007 tunkqdetach(struct knote *kn)
2008 {
2009 struct tuntap_softc *tp = kn->kn_hook;
2010
2011 knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2012 }
2013