1 /* $NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $ */
2 /*-
3 * SPDX-License-Identifier: BSD-2-Clause
4 *
5 * Copyright (C) 1999-2000 by Maksim Yevmenkin <m_evmenkin@yahoo.com>
6 * All rights reserved.
7 * Copyright (c) 2019 Kyle Evans <kevans@FreeBSD.org>
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 *
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * BASED ON:
32 * -------------------------------------------------------------------------
33 *
34 * Copyright (c) 1988, Julian Onions <jpo@cs.nott.ac.uk>
35 * Nottingham University 1987.
36 *
37 * This source may be freely distributed, however I would be interested
38 * in any changes that are made.
39 *
40 * This driver takes packets off the IP i/f and hands them up to a
41 * user process to have its wicked way with. This driver has it's
42 * roots in a similar driver written by Phil Cockcroft (formerly) at
43 * UCL. This driver is based much more on read/write/poll mode of
44 * operation though.
45 */
46
47 #include "opt_inet.h"
48 #include "opt_inet6.h"
49
50 #include <sys/param.h>
51 #include <sys/lock.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/systm.h>
55 #include <sys/jail.h>
56 #include <sys/mbuf.h>
57 #include <sys/module.h>
58 #include <sys/socket.h>
59 #include <sys/eventhandler.h>
60 #include <sys/fcntl.h>
61 #include <sys/filio.h>
62 #include <sys/sockio.h>
63 #include <sys/sx.h>
64 #include <sys/syslog.h>
65 #include <sys/ttycom.h>
66 #include <sys/poll.h>
67 #include <sys/selinfo.h>
68 #include <sys/signalvar.h>
69 #include <sys/filedesc.h>
70 #include <sys/kernel.h>
71 #include <sys/sysctl.h>
72 #include <sys/conf.h>
73 #include <sys/uio.h>
74 #include <sys/malloc.h>
75 #include <sys/random.h>
76 #include <sys/ctype.h>
77 #include <sys/osd.h>
78
79 #include <net/ethernet.h>
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/if_clone.h>
83 #include <net/if_dl.h>
84 #include <net/if_media.h>
85 #include <net/if_private.h>
86 #include <net/if_types.h>
87 #include <net/if_vlan_var.h>
88 #include <net/netisr.h>
89 #include <net/route.h>
90 #include <net/vnet.h>
91 #include <netinet/in.h>
92 #ifdef INET
93 #include <netinet/ip.h>
94 #endif
95 #ifdef INET6
96 #include <netinet/ip6.h>
97 #include <netinet6/ip6_var.h>
98 #endif
99 #include <netinet/udp.h>
100 #include <netinet/tcp.h>
101 #include <netinet/tcp_lro.h>
102 #include <net/bpf.h>
103 #include <net/if_tap.h>
104 #include <net/if_tun.h>
105
106 #include <dev/virtio/network/virtio_net.h>
107
108 #include <sys/queue.h>
109 #include <sys/condvar.h>
110 #include <security/mac/mac_framework.h>
111
112 struct tuntap_driver;
113
114 /*
115 * tun_list is protected by global tunmtx. Other mutable fields are
116 * protected by tun->tun_mtx, or by their owning subsystem. tun_dev is
117 * static for the duration of a tunnel interface.
118 */
119 struct tuntap_softc {
120 TAILQ_ENTRY(tuntap_softc) tun_list;
121 struct cdev *tun_alias;
122 struct cdev *tun_dev;
123 u_short tun_flags; /* misc flags */
124 #define TUN_OPEN 0x0001
125 #define TUN_INITED 0x0002
126 #define TUN_UNUSED1 0x0008
127 #define TUN_UNUSED2 0x0010
128 #define TUN_LMODE 0x0020
129 #define TUN_RWAIT 0x0040
130 #define TUN_ASYNC 0x0080
131 #define TUN_IFHEAD 0x0100
132 #define TUN_DYING 0x0200
133 #define TUN_L2 0x0400
134 #define TUN_VMNET 0x0800
135 #define TUN_TRANSIENT 0x1000
136
137 #define TUN_DRIVER_IDENT_MASK (TUN_L2 | TUN_VMNET)
138 #define TUN_READY (TUN_OPEN | TUN_INITED)
139
140 pid_t tun_pid; /* owning pid */
141 struct epoch_context tun_epoch_ctx;
142 struct ifnet *tun_ifp; /* the interface */
143 struct sigio *tun_sigio; /* async I/O info */
144 struct tuntap_driver *tun_drv; /* appropriate driver */
145 struct selinfo tun_rsel; /* read select */
146 struct mtx tun_mtx; /* softc field mutex */
147 struct cv tun_cv; /* for ref'd dev destroy */
148 struct ether_addr tun_ether; /* remote address */
149 int tun_busy; /* busy count */
150 int tun_vhdrlen; /* virtio-net header length */
151 struct lro_ctrl tun_lro; /* for TCP LRO */
152 bool tun_lro_ready; /* TCP LRO initialized */
153 };
154 #define TUN2IFP(sc) ((sc)->tun_ifp)
155
156 #define TUNDEBUG if (tundebug) if_printf
157
158 #define TUN_LOCK(tp) mtx_lock(&(tp)->tun_mtx)
159 #define TUN_UNLOCK(tp) mtx_unlock(&(tp)->tun_mtx)
160 #define TUN_LOCK_ASSERT(tp) mtx_assert(&(tp)->tun_mtx, MA_OWNED);
161
162 #define TUN_VMIO_FLAG_MASK 0x0fff
163
164 /*
165 * Interface capabilities of a tap device that supports the virtio-net
166 * header.
167 */
168 #define TAP_VNET_HDR_CAPS (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6 \
169 | IFCAP_VLAN_HWCSUM \
170 | IFCAP_TSO | IFCAP_LRO \
171 | IFCAP_VLAN_HWTSO)
172
173 #define TAP_ALL_OFFLOAD (CSUM_TSO | CSUM_TCP | CSUM_UDP |\
174 CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
175
176 /*
177 * All mutable global variables in if_tun are locked using tunmtx, with
178 * the exception of tundebug, which is used unlocked, and the drivers' *clones,
179 * which are static after setup.
180 */
181 static struct mtx tunmtx;
182 static eventhandler_tag rename_tag;
183 static eventhandler_tag clone_tag;
184 static int tuntap_osd_jail_slot;
185 static const char tunname[] = "tun";
186 static const char tapname[] = "tap";
187 static const char vmnetname[] = "vmnet";
188 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
189 static int tundebug = 0;
190 static int tundclone = 1;
191 static int tap_allow_uopen = 0; /* allow user devfs cloning */
192 static int tapuponopen = 0; /* IFF_UP on open() */
193 static int tapdclone = 1; /* enable devfs cloning */
194
195 static TAILQ_HEAD(,tuntap_softc) tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
196 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
197
198 static struct sx tun_ioctl_sx;
199 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
200
201 SYSCTL_DECL(_net_link);
202 /* tun */
203 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
204 "IP tunnel software network interface");
205 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
206 "Enable legacy devfs interface creation");
207
208 /* tap */
209 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
210 "Ethernet tunnel software network interface");
211 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
212 "Enable legacy devfs interface creation for all users");
213 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
214 "Bring interface up when /dev/tap is opened");
215 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
216 "Enable legacy devfs interface creation");
217 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
218
219 static int tun_create_device(struct tuntap_driver *drv, int unit,
220 struct ucred *cr, struct cdev **dev, const char *name);
221 static int tun_busy_locked(struct tuntap_softc *tp);
222 static void tun_unbusy_locked(struct tuntap_softc *tp);
223 static int tun_busy(struct tuntap_softc *tp);
224 static void tun_unbusy(struct tuntap_softc *tp);
225
226 static int tuntap_name2info(const char *name, int *unit, int *flags);
227 static void tunclone(void *arg, struct ucred *cred, char *name,
228 int namelen, struct cdev **dev);
229 static void tuncreate(struct cdev *dev);
230 static void tundtor(void *data);
231 static void tunrename(void *arg, struct ifnet *ifp);
232 static int tunifioctl(struct ifnet *, u_long, caddr_t);
233 static void tuninit(struct ifnet *);
234 static void tunifinit(void *xtp);
235 static int tuntapmodevent(module_t, int, void *);
236 static int tunoutput(struct ifnet *, struct mbuf *,
237 const struct sockaddr *, struct route *ro);
238 static void tunstart(struct ifnet *);
239 static void tunstart_l2(struct ifnet *);
240
241 static int tun_clone_match(struct if_clone *ifc, const char *name);
242 static int tap_clone_match(struct if_clone *ifc, const char *name);
243 static int vmnet_clone_match(struct if_clone *ifc, const char *name);
244 static int tun_clone_create(struct if_clone *, char *, size_t,
245 struct ifc_data *, struct ifnet **);
246 static int tun_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
247 static void tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
248
249 static d_open_t tunopen;
250 static d_read_t tunread;
251 static d_write_t tunwrite;
252 static d_ioctl_t tunioctl;
253 static d_poll_t tunpoll;
254 static d_kqfilter_t tunkqfilter;
255
256 static int tunkqread(struct knote *, long);
257 static int tunkqwrite(struct knote *, long);
258 static void tunkqdetach(struct knote *);
259
260 static const struct filterops tun_read_filterops = {
261 .f_isfd = 1,
262 .f_attach = NULL,
263 .f_detach = tunkqdetach,
264 .f_event = tunkqread,
265 .f_copy = knote_triv_copy,
266 };
267
268 static const struct filterops tun_write_filterops = {
269 .f_isfd = 1,
270 .f_attach = NULL,
271 .f_detach = tunkqdetach,
272 .f_event = tunkqwrite,
273 .f_copy = knote_triv_copy,
274 };
275
276 static struct tuntap_driver {
277 struct cdevsw cdevsw;
278 int ident_flags;
279 struct unrhdr *unrhdr;
280 struct clonedevs *clones;
281 ifc_match_f *clone_match_fn;
282 ifc_create_f *clone_create_fn;
283 ifc_destroy_f *clone_destroy_fn;
284 } tuntap_drivers[] = {
285 {
286 .ident_flags = 0,
287 .cdevsw = {
288 .d_version = D_VERSION,
289 .d_flags = D_NEEDMINOR,
290 .d_open = tunopen,
291 .d_read = tunread,
292 .d_write = tunwrite,
293 .d_ioctl = tunioctl,
294 .d_poll = tunpoll,
295 .d_kqfilter = tunkqfilter,
296 .d_name = tunname,
297 },
298 .clone_match_fn = tun_clone_match,
299 .clone_create_fn = tun_clone_create,
300 .clone_destroy_fn = tun_clone_destroy,
301 },
302 {
303 .ident_flags = TUN_L2,
304 .cdevsw = {
305 .d_version = D_VERSION,
306 .d_flags = D_NEEDMINOR,
307 .d_open = tunopen,
308 .d_read = tunread,
309 .d_write = tunwrite,
310 .d_ioctl = tunioctl,
311 .d_poll = tunpoll,
312 .d_kqfilter = tunkqfilter,
313 .d_name = tapname,
314 },
315 .clone_match_fn = tap_clone_match,
316 .clone_create_fn = tun_clone_create,
317 .clone_destroy_fn = tun_clone_destroy,
318 },
319 {
320 .ident_flags = TUN_L2 | TUN_VMNET,
321 .cdevsw = {
322 .d_version = D_VERSION,
323 .d_flags = D_NEEDMINOR,
324 .d_open = tunopen,
325 .d_read = tunread,
326 .d_write = tunwrite,
327 .d_ioctl = tunioctl,
328 .d_poll = tunpoll,
329 .d_kqfilter = tunkqfilter,
330 .d_name = vmnetname,
331 },
332 .clone_match_fn = vmnet_clone_match,
333 .clone_create_fn = tun_clone_create,
334 .clone_destroy_fn = tun_clone_destroy,
335 },
336 };
337 #define NDRV nitems(tuntap_drivers)
338
339 VNET_DEFINE_STATIC(struct if_clone *, tuntap_driver_cloners[NDRV]);
340 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners)
341
342 /*
343 * Mechanism for marking a tunnel device as busy so that we can safely do some
344 * orthogonal operations (such as operations on devices) without racing against
345 * tun_destroy. tun_destroy will wait on the condvar if we're at all busy or
346 * open, to be woken up when the condition is alleviated.
347 */
348 static int
tun_busy_locked(struct tuntap_softc * tp)349 tun_busy_locked(struct tuntap_softc *tp)
350 {
351
352 TUN_LOCK_ASSERT(tp);
353 if ((tp->tun_flags & TUN_DYING) != 0) {
354 /*
355 * Perhaps unintuitive, but the device is busy going away.
356 * Other interpretations of EBUSY from tun_busy make little
357 * sense, since making a busy device even more busy doesn't
358 * sound like a problem.
359 */
360 return (EBUSY);
361 }
362
363 ++tp->tun_busy;
364 return (0);
365 }
366
367 static void
tun_unbusy_locked(struct tuntap_softc * tp)368 tun_unbusy_locked(struct tuntap_softc *tp)
369 {
370
371 TUN_LOCK_ASSERT(tp);
372 KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
373
374 --tp->tun_busy;
375 /* Wake up anything that may be waiting on our busy tunnel. */
376 if (tp->tun_busy == 0)
377 cv_broadcast(&tp->tun_cv);
378 }
379
380 static int
tun_busy(struct tuntap_softc * tp)381 tun_busy(struct tuntap_softc *tp)
382 {
383 int ret;
384
385 TUN_LOCK(tp);
386 ret = tun_busy_locked(tp);
387 TUN_UNLOCK(tp);
388 return (ret);
389 }
390
391 static void
tun_unbusy(struct tuntap_softc * tp)392 tun_unbusy(struct tuntap_softc *tp)
393 {
394
395 TUN_LOCK(tp);
396 tun_unbusy_locked(tp);
397 TUN_UNLOCK(tp);
398 }
399
400 /*
401 * Sets unit and/or flags given the device name. Must be called with correct
402 * vnet context.
403 */
404 static int
tuntap_name2info(const char * name,int * outunit,int * outflags)405 tuntap_name2info(const char *name, int *outunit, int *outflags)
406 {
407 struct tuntap_driver *drv;
408 char *dname;
409 int flags, unit;
410 bool found;
411
412 if (name == NULL)
413 return (EINVAL);
414
415 /*
416 * Needed for dev_stdclone, but dev_stdclone will not modify, it just
417 * wants to be able to pass back a char * through the second param. We
418 * will always set that as NULL here, so we'll fake it.
419 */
420 dname = __DECONST(char *, name);
421 found = false;
422
423 for (u_int i = 0; i < NDRV; i++) {
424 drv = &tuntap_drivers[i];
425
426 if (strcmp(name, drv->cdevsw.d_name) == 0) {
427 found = true;
428 unit = -1;
429 flags = drv->ident_flags;
430 break;
431 }
432
433 if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
434 found = true;
435 flags = drv->ident_flags;
436 break;
437 }
438 }
439
440 if (!found)
441 return (ENXIO);
442
443 if (outunit != NULL)
444 *outunit = unit;
445 if (outflags != NULL)
446 *outflags = flags;
447 return (0);
448 }
449
450 static struct if_clone *
tuntap_cloner_from_flags(int tun_flags)451 tuntap_cloner_from_flags(int tun_flags)
452 {
453
454 for (u_int i = 0; i < NDRV; i++)
455 if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
456 tuntap_drivers[i].ident_flags)
457 return (V_tuntap_driver_cloners[i]);
458
459 return (NULL);
460 }
461
462 /*
463 * Get driver information from a set of flags specified. Masks the identifying
464 * part of the flags and compares it against all of the available
465 * tuntap_drivers.
466 */
467 static struct tuntap_driver *
tuntap_driver_from_flags(int tun_flags)468 tuntap_driver_from_flags(int tun_flags)
469 {
470
471 for (u_int i = 0; i < NDRV; i++)
472 if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
473 tuntap_drivers[i].ident_flags)
474 return (&tuntap_drivers[i]);
475
476 return (NULL);
477 }
478
479 static int
tun_clone_match(struct if_clone * ifc,const char * name)480 tun_clone_match(struct if_clone *ifc, const char *name)
481 {
482 int tunflags;
483
484 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
485 if ((tunflags & TUN_L2) == 0)
486 return (1);
487 }
488
489 return (0);
490 }
491
492 static int
tap_clone_match(struct if_clone * ifc,const char * name)493 tap_clone_match(struct if_clone *ifc, const char *name)
494 {
495 int tunflags;
496
497 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
498 if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
499 return (1);
500 }
501
502 return (0);
503 }
504
505 static int
vmnet_clone_match(struct if_clone * ifc,const char * name)506 vmnet_clone_match(struct if_clone *ifc, const char *name)
507 {
508 int tunflags;
509
510 if (tuntap_name2info(name, NULL, &tunflags) == 0) {
511 if ((tunflags & TUN_VMNET) != 0)
512 return (1);
513 }
514
515 return (0);
516 }
517
518 /*
519 * Create a clone via the ifnet cloning mechanism. Note that this is invoked
520 * indirectly by tunclone() below.
521 */
522 static int
tun_clone_create(struct if_clone * ifc,char * name,size_t len,struct ifc_data * ifd,struct ifnet ** ifpp)523 tun_clone_create(struct if_clone *ifc, char *name, size_t len,
524 struct ifc_data *ifd, struct ifnet **ifpp)
525 {
526 struct tuntap_driver *drv;
527 struct cdev *dev;
528 int err, i, tunflags, unit;
529
530 tunflags = 0;
531 /* The name here tells us exactly what we're creating */
532 err = tuntap_name2info(name, &unit, &tunflags);
533 if (err != 0)
534 return (err);
535
536 drv = tuntap_driver_from_flags(tunflags);
537 if (drv == NULL)
538 return (ENXIO);
539
540 if (unit != -1) {
541 /* If this unit number is still available that's okay. */
542 if (alloc_unr_specific(drv->unrhdr, unit) == -1)
543 return (EEXIST);
544 } else {
545 unit = alloc_unr(drv->unrhdr);
546 }
547
548 snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
549
550 /* find any existing device, or allocate new unit number */
551 dev = NULL;
552 i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
553 /* No preexisting struct cdev *, create one */
554 if (i != 0)
555 i = tun_create_device(drv, unit, NULL, &dev, name);
556 if (i == 0) {
557 struct tuntap_softc *tp;
558
559 tuncreate(dev);
560 tp = dev->si_drv1;
561 *ifpp = tp->tun_ifp;
562 }
563
564 return (i);
565 }
566
567 /*
568 * Create a clone via devfs access.
569 */
570 static void
tunclone(void * arg,struct ucred * cred,char * name,int namelen,struct cdev ** dev)571 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
572 struct cdev **dev)
573 {
574 char devname[SPECNAMELEN + 1];
575 struct tuntap_driver *drv;
576 int append_unit, i, u, tunflags;
577 bool mayclone;
578
579 if (*dev != NULL)
580 return;
581
582 tunflags = 0;
583 CURVNET_SET(CRED_TO_VNET(cred));
584 if (tuntap_name2info(name, &u, &tunflags) != 0)
585 goto out; /* Not recognized */
586
587 if (u != -1 && u > IF_MAXUNIT)
588 goto out; /* Unit number too high */
589
590 mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
591 if ((tunflags & TUN_L2) != 0) {
592 /* tap/vmnet allow user open with a sysctl */
593 mayclone = (mayclone || tap_allow_uopen) && tapdclone;
594 } else {
595 mayclone = mayclone && tundclone;
596 }
597
598 /*
599 * If tun cloning is enabled, only the superuser can create an
600 * interface.
601 */
602 if (!mayclone)
603 goto out;
604
605 if (u == -1)
606 append_unit = 1;
607 else
608 append_unit = 0;
609
610 drv = tuntap_driver_from_flags(tunflags);
611 if (drv == NULL)
612 goto out;
613
614 /* find any existing device, or allocate new unit number */
615 i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
616 if (i) {
617 if (append_unit) {
618 namelen = snprintf(devname, sizeof(devname), "%s%d",
619 name, u);
620 name = devname;
621 }
622
623 i = tun_create_device(drv, u, cred, dev, name);
624 } else {
625 /* Consumed by the dev_clone invoker. */
626 dev_ref(*dev);
627 }
628 if (i == 0)
629 if_clone_create(name, namelen, NULL);
630 out:
631 CURVNET_RESTORE();
632 }
633
634 static void
tunfree(struct epoch_context * ctx)635 tunfree(struct epoch_context *ctx)
636 {
637 struct tuntap_softc *tp;
638
639 tp = __containerof(ctx, struct tuntap_softc, tun_epoch_ctx);
640
641 /* Any remaining resources that would be needed by a concurrent open. */
642 mtx_destroy(&tp->tun_mtx);
643 free(tp, M_TUN);
644 }
645
646 static int
tun_destroy(struct tuntap_softc * tp,bool may_intr)647 tun_destroy(struct tuntap_softc *tp, bool may_intr)
648 {
649 int error;
650
651 TUN_LOCK(tp);
652
653 /*
654 * Transient tunnels may have set TUN_DYING if we're being destroyed as
655 * a result of the last close, which we'll allow.
656 */
657 MPASS((tp->tun_flags & (TUN_DYING | TUN_TRANSIENT)) != TUN_DYING);
658 tp->tun_flags |= TUN_DYING;
659 error = 0;
660 while (tp->tun_busy != 0) {
661 if (may_intr)
662 error = cv_wait_sig(&tp->tun_cv, &tp->tun_mtx);
663 else
664 cv_wait(&tp->tun_cv, &tp->tun_mtx);
665 if (error != 0 && tp->tun_busy != 0) {
666 tp->tun_flags &= ~TUN_DYING;
667 TUN_UNLOCK(tp);
668 return (error);
669 }
670 }
671 TUN_UNLOCK(tp);
672
673 CURVNET_SET(TUN2IFP(tp)->if_vnet);
674
675 mtx_lock(&tunmtx);
676 TAILQ_REMOVE(&tunhead, tp, tun_list);
677 mtx_unlock(&tunmtx);
678
679 /*
680 * destroy_dev will take care of any alias. For transient tunnels,
681 * we're being called from close(2) so we can't destroy it ourselves
682 * without deadlocking, but we already know that we can cleanup
683 * everything else and just continue to prevent it from being reopened.
684 */
685 if ((tp->tun_flags & TUN_TRANSIENT) != 0) {
686 atomic_store_ptr(&tp->tun_dev->si_drv1, tp->tun_dev);
687 destroy_dev_sched(tp->tun_dev);
688 } else {
689 destroy_dev(tp->tun_dev);
690 }
691 seldrain(&tp->tun_rsel);
692 knlist_clear(&tp->tun_rsel.si_note, 0);
693 knlist_destroy(&tp->tun_rsel.si_note);
694 if ((tp->tun_flags & TUN_L2) != 0) {
695 ether_ifdetach(TUN2IFP(tp));
696 } else {
697 bpfdetach(TUN2IFP(tp));
698 if_detach(TUN2IFP(tp));
699 }
700 sx_xlock(&tun_ioctl_sx);
701 TUN2IFP(tp)->if_softc = NULL;
702 sx_xunlock(&tun_ioctl_sx);
703 free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
704 if_free(TUN2IFP(tp));
705 cv_destroy(&tp->tun_cv);
706 NET_EPOCH_CALL(tunfree, &tp->tun_epoch_ctx);
707 CURVNET_RESTORE();
708
709 return (0);
710 }
711
712 static int
tun_clone_destroy(struct if_clone * ifc __unused,struct ifnet * ifp,uint32_t flags)713 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags)
714 {
715 struct tuntap_softc *tp = ifp->if_softc;
716
717 return (tun_destroy(tp, true));
718 }
719
720 static void
vnet_tun_init(const void * unused __unused)721 vnet_tun_init(const void *unused __unused)
722 {
723
724 for (u_int i = 0; i < NDRV; ++i) {
725 struct if_clone_addreq req = {
726 .match_f = tuntap_drivers[i].clone_match_fn,
727 .create_f = tuntap_drivers[i].clone_create_fn,
728 .destroy_f = tuntap_drivers[i].clone_destroy_fn,
729 };
730 V_tuntap_driver_cloners[i] =
731 ifc_attach_cloner(tuntap_drivers[i].cdevsw.d_name, &req);
732 };
733 }
734 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
735 vnet_tun_init, NULL);
736
737 static void
tun_uninit(const void * unused __unused)738 tun_uninit(const void *unused __unused)
739 {
740 struct tuntap_driver *drv;
741 struct tuntap_softc *tp;
742 int i;
743
744 EVENTHANDLER_DEREGISTER(ifnet_rename_event, rename_tag);
745 EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
746
747 CURVNET_SET(vnet0);
748 for (u_int i = 0; i < NDRV; i++) {
749 if_clone_detach(V_tuntap_driver_cloners[i]);
750 V_tuntap_driver_cloners[i] = NULL;
751 }
752 CURVNET_RESTORE();
753
754 if (tuntap_osd_jail_slot != 0)
755 osd_jail_deregister(tuntap_osd_jail_slot);
756
757 mtx_lock(&tunmtx);
758 while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
759 mtx_unlock(&tunmtx);
760 /* tun_destroy() will remove it from the tailq. */
761 tun_destroy(tp, false);
762 mtx_lock(&tunmtx);
763 }
764 mtx_unlock(&tunmtx);
765 for (i = 0; i < nitems(tuntap_drivers); ++i) {
766 drv = &tuntap_drivers[i];
767 destroy_dev_drain(&drv->cdevsw);
768 delete_unrhdr(drv->unrhdr);
769 clone_cleanup(&drv->clones);
770 }
771 NET_EPOCH_DRAIN_CALLBACKS();
772 mtx_destroy(&tunmtx);
773 }
774 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
775
776 static struct tuntap_driver *
tuntap_driver_from_ifnet(const struct ifnet * ifp)777 tuntap_driver_from_ifnet(const struct ifnet *ifp)
778 {
779 struct tuntap_driver *drv;
780 int i;
781
782 if (ifp == NULL)
783 return (NULL);
784
785 for (i = 0; i < nitems(tuntap_drivers); ++i) {
786 drv = &tuntap_drivers[i];
787 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
788 return (drv);
789 }
790
791 return (NULL);
792 }
793
794 /*
795 * Remove devices that were created by devfs cloning, as they hold references
796 * which prevent the prison from collapsing, in which state VNET sysuninits will
797 * not be invoked.
798 */
799 static int
tuntap_prison_remove(void * obj,void * data __unused)800 tuntap_prison_remove(void *obj, void *data __unused)
801 {
802 #ifdef VIMAGE
803 struct prison *pr;
804
805 pr = obj;
806 if (prison_owns_vnet(pr)) {
807 CURVNET_SET(pr->pr_vnet);
808 for (u_int i = 0; i < NDRV; i++) {
809 if_clone_detach(V_tuntap_driver_cloners[i]);
810 V_tuntap_driver_cloners[i] = NULL;
811 }
812 CURVNET_RESTORE();
813 }
814 #endif
815 return (0);
816 }
817
818 static int
tuntapmodevent(module_t mod,int type,void * data)819 tuntapmodevent(module_t mod, int type, void *data)
820 {
821 struct tuntap_driver *drv;
822 int i;
823
824 switch (type) {
825 case MOD_LOAD:
826 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
827 for (i = 0; i < nitems(tuntap_drivers); ++i) {
828 drv = &tuntap_drivers[i];
829 clone_setup(&drv->clones);
830 drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
831 }
832 osd_method_t methods[PR_MAXMETHOD] = {
833 [PR_METHOD_REMOVE] = tuntap_prison_remove,
834 };
835 tuntap_osd_jail_slot = osd_jail_register(NULL, methods);
836 rename_tag = EVENTHANDLER_REGISTER(ifnet_rename_event,
837 tunrename, NULL, EVENTHANDLER_PRI_ANY);
838 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, NULL,
839 EVENTHANDLER_PRI_ANY);
840 break;
841 case MOD_UNLOAD:
842 /* See tun_uninit(). */
843 break;
844 default:
845 return EOPNOTSUPP;
846 }
847 return 0;
848 }
849
850 static moduledata_t tuntap_mod = {
851 "if_tuntap",
852 tuntapmodevent,
853 0
854 };
855
856 /* We'll only ever have these two, so no need for a macro. */
857 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
858 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
859
860 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
861 MODULE_VERSION(if_tuntap, 1);
862 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
863 MODULE_VERSION(if_tun, 1);
864 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
865 MODULE_VERSION(if_tap, 1);
866
867 static int
tun_create_device(struct tuntap_driver * drv,int unit,struct ucred * cr,struct cdev ** dev,const char * name)868 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
869 struct cdev **dev, const char *name)
870 {
871 struct make_dev_args args;
872 struct tuntap_softc *tp;
873 int error;
874
875 tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
876 mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
877 cv_init(&tp->tun_cv, "tun_condvar");
878 tp->tun_flags = drv->ident_flags;
879 tp->tun_drv = drv;
880
881 make_dev_args_init(&args);
882 if (cr != NULL)
883 args.mda_flags = MAKEDEV_REF | MAKEDEV_CHECKNAME;
884 args.mda_devsw = &drv->cdevsw;
885 args.mda_cr = cr;
886 args.mda_uid = UID_UUCP;
887 args.mda_gid = GID_DIALER;
888 args.mda_mode = 0600;
889 args.mda_unit = unit;
890 args.mda_si_drv1 = tp;
891 error = make_dev_s(&args, dev, "%s", name);
892 if (error != 0) {
893 mtx_destroy(&tp->tun_mtx);
894 cv_destroy(&tp->tun_cv);
895 free(tp, M_TUN);
896 return (error);
897 }
898
899 KASSERT((*dev)->si_drv1 != NULL,
900 ("Failed to set si_drv1 at %s creation", name));
901 tp->tun_dev = *dev;
902 knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
903 mtx_lock(&tunmtx);
904 TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
905 mtx_unlock(&tunmtx);
906 return (0);
907 }
908
909 static void
tunstart(struct ifnet * ifp)910 tunstart(struct ifnet *ifp)
911 {
912 struct tuntap_softc *tp = ifp->if_softc;
913 struct mbuf *m;
914
915 TUNDEBUG(ifp, "starting\n");
916 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
917 IFQ_LOCK(&ifp->if_snd);
918 IFQ_POLL_NOLOCK(&ifp->if_snd, m);
919 if (m == NULL) {
920 IFQ_UNLOCK(&ifp->if_snd);
921 return;
922 }
923 IFQ_UNLOCK(&ifp->if_snd);
924 }
925
926 TUN_LOCK(tp);
927 if (tp->tun_flags & TUN_RWAIT) {
928 tp->tun_flags &= ~TUN_RWAIT;
929 wakeup(tp);
930 }
931 selwakeuppri(&tp->tun_rsel, PZERO);
932 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
933 if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
934 TUN_UNLOCK(tp);
935 pgsigio(&tp->tun_sigio, SIGIO, 0);
936 } else
937 TUN_UNLOCK(tp);
938 }
939
940 /*
941 * tunstart_l2
942 *
943 * queue packets from higher level ready to put out
944 */
945 static void
tunstart_l2(struct ifnet * ifp)946 tunstart_l2(struct ifnet *ifp)
947 {
948 struct tuntap_softc *tp = ifp->if_softc;
949
950 TUNDEBUG(ifp, "starting\n");
951
952 /*
953 * do not junk pending output if we are in VMnet mode.
954 * XXX: can this do any harm because of queue overflow?
955 */
956
957 TUN_LOCK(tp);
958 if (((tp->tun_flags & TUN_VMNET) == 0) &&
959 ((tp->tun_flags & TUN_READY) != TUN_READY)) {
960 struct mbuf *m;
961
962 /* Unlocked read. */
963 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
964
965 for (;;) {
966 IF_DEQUEUE(&ifp->if_snd, m);
967 if (m != NULL) {
968 m_freem(m);
969 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
970 } else
971 break;
972 }
973 TUN_UNLOCK(tp);
974
975 return;
976 }
977
978 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
979
980 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
981 if (tp->tun_flags & TUN_RWAIT) {
982 tp->tun_flags &= ~TUN_RWAIT;
983 wakeup(tp);
984 }
985
986 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
987 TUN_UNLOCK(tp);
988 pgsigio(&tp->tun_sigio, SIGIO, 0);
989 TUN_LOCK(tp);
990 }
991
992 selwakeuppri(&tp->tun_rsel, PZERO);
993 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
994 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
995 }
996
997 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
998 TUN_UNLOCK(tp);
999 } /* tunstart_l2 */
1000
1001 static int
tap_transmit(struct ifnet * ifp,struct mbuf * m)1002 tap_transmit(struct ifnet *ifp, struct mbuf *m)
1003 {
1004 int error;
1005
1006 BPF_MTAP(ifp, m);
1007 IFQ_HANDOFF(ifp, m, error);
1008 return (error);
1009 }
1010
1011 static void
tuncreate(struct cdev * dev)1012 tuncreate(struct cdev *dev)
1013 {
1014 struct tuntap_driver *drv;
1015 struct tuntap_softc *tp;
1016 struct ifnet *ifp;
1017 struct ether_addr eaddr;
1018 int iflags;
1019 u_char type;
1020
1021 tp = dev->si_drv1;
1022 KASSERT(tp != NULL,
1023 ("si_drv1 should have been initialized at creation"));
1024
1025 drv = tp->tun_drv;
1026 iflags = IFF_MULTICAST;
1027 if ((tp->tun_flags & TUN_L2) != 0) {
1028 type = IFT_ETHER;
1029 iflags |= IFF_BROADCAST | IFF_SIMPLEX;
1030 } else {
1031 type = IFT_PPP;
1032 iflags |= IFF_POINTOPOINT;
1033 }
1034 ifp = tp->tun_ifp = if_alloc(type);
1035 ifp->if_softc = tp;
1036 if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
1037 ifp->if_ioctl = tunifioctl;
1038 ifp->if_flags = iflags;
1039 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
1040 ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_MEXTPG;
1041 if ((tp->tun_flags & TUN_L2) != 0)
1042 ifp->if_capabilities |=
1043 IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO;
1044 ifp->if_capenable |= IFCAP_LINKSTATE | IFCAP_MEXTPG;
1045
1046 if ((tp->tun_flags & TUN_L2) != 0) {
1047 ifp->if_init = tunifinit;
1048 ifp->if_start = tunstart_l2;
1049 ifp->if_transmit = tap_transmit;
1050 ifp->if_qflush = if_qflush;
1051
1052 ether_gen_addr(ifp, &eaddr);
1053 ether_ifattach(ifp, eaddr.octet);
1054 } else {
1055 ifp->if_mtu = TUNMTU;
1056 ifp->if_start = tunstart;
1057 ifp->if_output = tunoutput;
1058
1059 ifp->if_snd.ifq_drv_maxlen = 0;
1060 IFQ_SET_READY(&ifp->if_snd);
1061
1062 if_attach(ifp);
1063 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
1064 }
1065
1066 TUN_LOCK(tp);
1067 tp->tun_flags |= TUN_INITED;
1068 TUN_UNLOCK(tp);
1069
1070 TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
1071 ifp->if_xname, dev2unit(dev));
1072 }
1073
1074 static void
tunrename(void * arg __unused,struct ifnet * ifp)1075 tunrename(void *arg __unused, struct ifnet *ifp)
1076 {
1077 struct tuntap_softc *tp;
1078 int error;
1079
1080 if (tuntap_driver_from_ifnet(ifp) == NULL)
1081 return;
1082
1083 /*
1084 * We need to grab the ioctl sx long enough to make sure the softc is
1085 * still there. If it is, we can safely try to busy the tun device.
1086 * The busy may fail if the device is currently dying, in which case
1087 * we do nothing. If it doesn't fail, the busy count stops the device
1088 * from dying until we've created the alias (that will then be
1089 * subsequently destroyed).
1090 */
1091 sx_xlock(&tun_ioctl_sx);
1092 tp = ifp->if_softc;
1093 if (tp == NULL) {
1094 sx_xunlock(&tun_ioctl_sx);
1095 return;
1096 }
1097 error = tun_busy(tp);
1098 sx_xunlock(&tun_ioctl_sx);
1099 if (error != 0)
1100 return;
1101 if (tp->tun_alias != NULL) {
1102 destroy_dev(tp->tun_alias);
1103 tp->tun_alias = NULL;
1104 }
1105
1106 if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1107 goto out;
1108
1109 /*
1110 * Failure's ok, aliases are created on a best effort basis. If a
1111 * tun user/consumer decides to rename the interface to conflict with
1112 * another device (non-ifnet) on the system, we will assume they know
1113 * what they are doing. make_dev_alias_p won't touch tun_alias on
1114 * failure, so we use it but ignore the return value.
1115 */
1116 make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1117 ifp->if_xname);
1118 out:
1119 tun_unbusy(tp);
1120 }
1121
1122 static int
tunopen(struct cdev * dev,int flag,int mode,struct thread * td)1123 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1124 {
1125 struct epoch_tracker et;
1126 struct ifnet *ifp;
1127 struct tuntap_softc *tp;
1128 void *p;
1129 int error __diagused, tunflags;
1130
1131 /*
1132 * Transient tunnels do deferred destroy of the tun device but want
1133 * to immediately cleanup state, so they clobber si_drv1 to avoid a
1134 * use-after-free in case someone does happen to open it in the interim.
1135 * We avoid using NULL to be able to distinguish from an uninitialized
1136 * cdev.
1137 *
1138 * We use the net epoch here to let a concurrent tun_destroy() schedule
1139 * freeing our tuntap_softc, in case we entered here and loaded si_drv1
1140 * before it was swapped out. If we managed to load this while it was
1141 * still a softc, then the concurrent tun_destroy() hasn't yet scheduled
1142 * it to be free- that will take place sometime after the epoch we just
1143 * entered, so we can safely use it.
1144 */
1145 NET_EPOCH_ENTER(et);
1146 p = atomic_load_ptr(&dev->si_drv1);
1147 if (p == dev) {
1148 NET_EPOCH_EXIT(et);
1149 return (ENXIO);
1150 }
1151
1152 tunflags = 0;
1153 CURVNET_SET(TD_TO_VNET(td));
1154 error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1155 if (error != 0) {
1156 CURVNET_RESTORE();
1157 NET_EPOCH_EXIT(et);
1158 return (error); /* Shouldn't happen */
1159 }
1160
1161 tp = p;
1162 KASSERT(tp != NULL,
1163 ("si_drv1 should have been initialized at creation"));
1164
1165 TUN_LOCK(tp);
1166 if ((tp->tun_flags & TUN_INITED) == 0) {
1167 TUN_UNLOCK(tp);
1168 CURVNET_RESTORE();
1169 NET_EPOCH_EXIT(et);
1170 return (ENXIO);
1171 }
1172 if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1173 TUN_UNLOCK(tp);
1174 CURVNET_RESTORE();
1175 NET_EPOCH_EXIT(et);
1176 return (EBUSY);
1177 }
1178
1179 NET_EPOCH_EXIT(et);
1180 error = tun_busy_locked(tp);
1181 KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1182 ifp = TUN2IFP(tp);
1183
1184 if ((tp->tun_flags & TUN_L2) != 0) {
1185 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1186 sizeof(tp->tun_ether.octet));
1187
1188 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1189 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1190
1191 if (tapuponopen)
1192 ifp->if_flags |= IFF_UP;
1193 }
1194
1195 tp->tun_pid = td->td_proc->p_pid;
1196 tp->tun_flags |= TUN_OPEN;
1197
1198 if_link_state_change(ifp, LINK_STATE_UP);
1199 TUNDEBUG(ifp, "open\n");
1200 TUN_UNLOCK(tp);
1201
1202 /*
1203 * This can fail with either ENOENT or EBUSY. This is in the middle of
1204 * d_open, so ENOENT should not be possible. EBUSY is possible, but
1205 * the only cdevpriv dtor being set will be tundtor and the softc being
1206 * passed is constant for a given cdev. We ignore the possible error
1207 * because of this as either "unlikely" or "not actually a problem."
1208 */
1209 (void)devfs_set_cdevpriv(tp, tundtor);
1210 CURVNET_RESTORE();
1211 return (0);
1212 }
1213
1214 /*
1215 * tundtor - tear down the device - mark i/f down & delete
1216 * routing info
1217 */
1218 static void
tundtor(void * data)1219 tundtor(void *data)
1220 {
1221 struct proc *p;
1222 struct tuntap_softc *tp;
1223 struct ifnet *ifp;
1224 bool l2tun;
1225
1226 tp = data;
1227 p = curproc;
1228 ifp = TUN2IFP(tp);
1229
1230 TUN_LOCK(tp);
1231
1232 /*
1233 * Realistically, we can't be obstinate here. This only means that the
1234 * tuntap device was closed out of order, and the last closer wasn't the
1235 * controller. These are still good to know about, though, as software
1236 * should avoid multiple processes with a tuntap device open and
1237 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1238 * parent).
1239 */
1240 if (p->p_pid != tp->tun_pid) {
1241 log(LOG_INFO,
1242 "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1243 p->p_pid, p->p_comm, tp->tun_dev->si_name);
1244 }
1245
1246 /*
1247 * junk all pending output
1248 */
1249 CURVNET_SET(ifp->if_vnet);
1250
1251 l2tun = false;
1252 if ((tp->tun_flags & TUN_L2) != 0) {
1253 l2tun = true;
1254 IF_DRAIN(&ifp->if_snd);
1255 } else {
1256 IFQ_PURGE(&ifp->if_snd);
1257 }
1258
1259 /* For vmnet, we won't do most of the address/route bits */
1260 if ((tp->tun_flags & TUN_VMNET) != 0 ||
1261 (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1262 goto out;
1263 #if defined(INET) || defined(INET6)
1264 if (l2tun && tp->tun_lro_ready) {
1265 TUNDEBUG (ifp, "LRO disabled\n");
1266 tcp_lro_free(&tp->tun_lro);
1267 tp->tun_lro_ready = false;
1268 }
1269 #endif
1270 if (ifp->if_flags & IFF_UP) {
1271 TUN_UNLOCK(tp);
1272 if_down(ifp);
1273 TUN_LOCK(tp);
1274 }
1275
1276 /* Delete all addresses and routes which reference this interface. */
1277 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1278 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1279 TUN_UNLOCK(tp);
1280 if_purgeaddrs(ifp);
1281 TUN_LOCK(tp);
1282 }
1283
1284 out:
1285 if_link_state_change(ifp, LINK_STATE_DOWN);
1286 CURVNET_RESTORE();
1287
1288 funsetown(&tp->tun_sigio);
1289 selwakeuppri(&tp->tun_rsel, PZERO);
1290 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1291 TUNDEBUG (ifp, "closed\n");
1292 tp->tun_flags &= ~TUN_OPEN;
1293 tp->tun_pid = 0;
1294 tun_vnethdr_set(ifp, 0);
1295
1296 tun_unbusy_locked(tp);
1297 if ((tp->tun_flags & TUN_TRANSIENT) != 0) {
1298 struct if_clone *cloner;
1299 int error __diagused;
1300
1301 /* Mark it busy so that nothing can re-open it. */
1302 tp->tun_flags |= TUN_DYING;
1303 TUN_UNLOCK(tp);
1304
1305 CURVNET_SET_QUIET(ifp->if_home_vnet);
1306 cloner = tuntap_cloner_from_flags(tp->tun_flags);
1307 CURVNET_RESTORE();
1308
1309 error = if_clone_destroyif(cloner, ifp);
1310 MPASS(error == 0 || error == EINTR || error == ERESTART);
1311 return;
1312 }
1313
1314 TUN_UNLOCK(tp);
1315 }
1316
1317 static void
tuninit(struct ifnet * ifp)1318 tuninit(struct ifnet *ifp)
1319 {
1320 struct tuntap_softc *tp = ifp->if_softc;
1321
1322 TUNDEBUG(ifp, "tuninit\n");
1323
1324 TUN_LOCK(tp);
1325 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1326 if ((tp->tun_flags & TUN_L2) == 0) {
1327 ifp->if_flags |= IFF_UP;
1328 getmicrotime(&ifp->if_lastchange);
1329 TUN_UNLOCK(tp);
1330 } else {
1331 #if defined(INET) || defined(INET6)
1332 if (tcp_lro_init(&tp->tun_lro) == 0) {
1333 TUNDEBUG(ifp, "LRO enabled\n");
1334 tp->tun_lro.ifp = ifp;
1335 tp->tun_lro_ready = true;
1336 } else {
1337 TUNDEBUG(ifp, "Could not enable LRO\n");
1338 tp->tun_lro_ready = false;
1339 }
1340 #endif
1341 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1342 TUN_UNLOCK(tp);
1343 /* attempt to start output */
1344 tunstart_l2(ifp);
1345 }
1346
1347 }
1348
1349 /*
1350 * Used only for l2 tunnel.
1351 */
1352 static void
tunifinit(void * xtp)1353 tunifinit(void *xtp)
1354 {
1355 struct tuntap_softc *tp;
1356
1357 tp = (struct tuntap_softc *)xtp;
1358 tuninit(tp->tun_ifp);
1359 }
1360
1361 /*
1362 * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1363 * current value of ifp->if_capenable.
1364 */
1365 static void
tun_caps_changed(struct ifnet * ifp)1366 tun_caps_changed(struct ifnet *ifp)
1367 {
1368 uint64_t hwassist = 0;
1369
1370 TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1371 if (ifp->if_capenable & IFCAP_TXCSUM)
1372 hwassist |= CSUM_TCP | CSUM_UDP;
1373 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1374 hwassist |= CSUM_TCP_IPV6
1375 | CSUM_UDP_IPV6;
1376 if (ifp->if_capenable & IFCAP_TSO4)
1377 hwassist |= CSUM_IP_TSO;
1378 if (ifp->if_capenable & IFCAP_TSO6)
1379 hwassist |= CSUM_IP6_TSO;
1380 ifp->if_hwassist = hwassist;
1381 }
1382
1383 /*
1384 * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1385 * if_capabilities and if_capenable as needed.
1386 */
1387 static void
tun_vnethdr_set(struct ifnet * ifp,int vhdrlen)1388 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1389 {
1390 struct tuntap_softc *tp = ifp->if_softc;
1391
1392 TUN_LOCK_ASSERT(tp);
1393
1394 if (tp->tun_vhdrlen == vhdrlen)
1395 return;
1396
1397 /*
1398 * Update if_capabilities to reflect the
1399 * functionalities offered by the virtio-net
1400 * header.
1401 */
1402 if (vhdrlen != 0)
1403 ifp->if_capabilities |=
1404 TAP_VNET_HDR_CAPS;
1405 else
1406 ifp->if_capabilities &=
1407 ~TAP_VNET_HDR_CAPS;
1408 /*
1409 * Disable any capabilities that we don't
1410 * support anymore.
1411 */
1412 ifp->if_capenable &= ifp->if_capabilities;
1413 tun_caps_changed(ifp);
1414 tp->tun_vhdrlen = vhdrlen;
1415
1416 TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1417 vhdrlen, ifp->if_capabilities);
1418 }
1419
1420 /*
1421 * Process an ioctl request.
1422 */
1423 static int
tunifioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1424 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1425 {
1426 struct ifreq *ifr = (struct ifreq *)data;
1427 struct tuntap_softc *tp;
1428 struct ifstat *ifs;
1429 struct ifmediareq *ifmr;
1430 int dummy, error = 0;
1431 bool l2tun;
1432
1433 ifmr = NULL;
1434 sx_xlock(&tun_ioctl_sx);
1435 tp = ifp->if_softc;
1436 if (tp == NULL) {
1437 error = ENXIO;
1438 goto bad;
1439 }
1440 l2tun = (tp->tun_flags & TUN_L2) != 0;
1441 switch(cmd) {
1442 case SIOCGIFSTATUS:
1443 ifs = (struct ifstat *)data;
1444 TUN_LOCK(tp);
1445 if (tp->tun_pid)
1446 snprintf(ifs->ascii, sizeof(ifs->ascii),
1447 "\tOpened by PID %d\n", tp->tun_pid);
1448 else
1449 ifs->ascii[0] = '\0';
1450 TUN_UNLOCK(tp);
1451 break;
1452 case SIOCSIFADDR:
1453 if (l2tun)
1454 error = ether_ioctl(ifp, cmd, data);
1455 else
1456 tuninit(ifp);
1457 if (error == 0)
1458 TUNDEBUG(ifp, "address set\n");
1459 break;
1460 case SIOCSIFMTU:
1461 ifp->if_mtu = ifr->ifr_mtu;
1462 TUNDEBUG(ifp, "mtu set\n");
1463 break;
1464 case SIOCSIFFLAGS:
1465 case SIOCADDMULTI:
1466 case SIOCDELMULTI:
1467 break;
1468 case SIOCGIFMEDIA:
1469 if (!l2tun) {
1470 error = EINVAL;
1471 break;
1472 }
1473
1474 ifmr = (struct ifmediareq *)data;
1475 dummy = ifmr->ifm_count;
1476 ifmr->ifm_count = 1;
1477 ifmr->ifm_status = IFM_AVALID;
1478 ifmr->ifm_active = IFM_ETHER | IFM_FDX | IFM_1000_T;
1479 if (tp->tun_flags & TUN_OPEN)
1480 ifmr->ifm_status |= IFM_ACTIVE;
1481 ifmr->ifm_current = ifmr->ifm_active;
1482 if (dummy >= 1) {
1483 int media = IFM_ETHER;
1484 error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1485 }
1486 break;
1487 case SIOCSIFCAP:
1488 TUN_LOCK(tp);
1489 ifp->if_capenable = ifr->ifr_reqcap;
1490 tun_caps_changed(ifp);
1491 TUN_UNLOCK(tp);
1492 VLAN_CAPABILITIES(ifp);
1493 break;
1494 default:
1495 if (l2tun) {
1496 error = ether_ioctl(ifp, cmd, data);
1497 } else {
1498 error = EINVAL;
1499 }
1500 }
1501 bad:
1502 sx_xunlock(&tun_ioctl_sx);
1503 return (error);
1504 }
1505
1506 /*
1507 * tunoutput - queue packets from higher level ready to put out.
1508 */
1509 static int
tunoutput(struct ifnet * ifp,struct mbuf * m0,const struct sockaddr * dst,struct route * ro)1510 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1511 struct route *ro)
1512 {
1513 struct tuntap_softc *tp = ifp->if_softc;
1514 u_short cached_tun_flags;
1515 int error;
1516 u_int32_t af;
1517
1518 TUNDEBUG (ifp, "tunoutput\n");
1519
1520 #ifdef MAC
1521 error = mac_ifnet_check_transmit(ifp, m0);
1522 if (error) {
1523 m_freem(m0);
1524 return (error);
1525 }
1526 #endif
1527
1528 /* Could be unlocked read? */
1529 TUN_LOCK(tp);
1530 cached_tun_flags = tp->tun_flags;
1531 TUN_UNLOCK(tp);
1532 if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1533 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1534 m_freem (m0);
1535 return (EHOSTDOWN);
1536 }
1537
1538 if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1539 m_freem (m0);
1540 return (EHOSTDOWN);
1541 }
1542
1543 /* BPF writes need to be handled specially. */
1544 if (dst->sa_family == AF_UNSPEC || dst->sa_family == pseudo_AF_HDRCMPLT)
1545 bcopy(dst->sa_data, &af, sizeof(af));
1546 else
1547 af = RO_GET_FAMILY(ro, dst);
1548
1549 BPF_MTAP2(ifp, &af, sizeof(af), m0);
1550
1551 /* prepend sockaddr? this may abort if the mbuf allocation fails */
1552 if (cached_tun_flags & TUN_LMODE) {
1553 /* allocate space for sockaddr */
1554 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1555
1556 /* if allocation failed drop packet */
1557 if (m0 == NULL) {
1558 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1559 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1560 return (ENOBUFS);
1561 } else {
1562 bcopy(dst, m0->m_data, dst->sa_len);
1563 }
1564 }
1565
1566 if (cached_tun_flags & TUN_IFHEAD) {
1567 /* Prepend the address family */
1568 M_PREPEND(m0, 4, M_NOWAIT);
1569
1570 /* if allocation failed drop packet */
1571 if (m0 == NULL) {
1572 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1573 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1574 return (ENOBUFS);
1575 } else
1576 *(u_int32_t *)m0->m_data = htonl(af);
1577 } else {
1578 #ifdef INET
1579 if (af != AF_INET)
1580 #endif
1581 {
1582 m_freem(m0);
1583 return (EAFNOSUPPORT);
1584 }
1585 }
1586
1587 error = (ifp->if_transmit)(ifp, m0);
1588 if (error)
1589 return (ENOBUFS);
1590 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1591 return (0);
1592 }
1593
1594 /*
1595 * the cdevsw interface is now pretty minimal.
1596 */
1597 static int
tunioctl(struct cdev * dev,u_long cmd,caddr_t data,int flag,struct thread * td)1598 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1599 struct thread *td)
1600 {
1601 struct ifreq ifr, *ifrp;
1602 struct tuntap_softc *tp = dev->si_drv1;
1603 struct ifnet *ifp = TUN2IFP(tp);
1604 struct tuninfo *tunp;
1605 int error, iflags, ival;
1606 bool l2tun;
1607
1608 l2tun = (tp->tun_flags & TUN_L2) != 0;
1609 if (l2tun) {
1610 /* tap specific ioctls */
1611 switch(cmd) {
1612 /* VMware/VMnet port ioctl's */
1613 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1614 defined(COMPAT_FREEBSD4)
1615 case _IO('V', 0):
1616 ival = IOCPARM_IVAL(data);
1617 data = (caddr_t)&ival;
1618 /* FALLTHROUGH */
1619 #endif
1620 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1621 iflags = *(int *)data;
1622 iflags &= TUN_VMIO_FLAG_MASK;
1623 iflags &= ~IFF_CANTCHANGE;
1624 iflags |= IFF_UP;
1625
1626 TUN_LOCK(tp);
1627 ifp->if_flags = iflags |
1628 (ifp->if_flags & IFF_CANTCHANGE);
1629 TUN_UNLOCK(tp);
1630
1631 return (0);
1632 case SIOCGIFADDR: /* get MAC address of the remote side */
1633 TUN_LOCK(tp);
1634 bcopy(&tp->tun_ether.octet, data,
1635 sizeof(tp->tun_ether.octet));
1636 TUN_UNLOCK(tp);
1637
1638 return (0);
1639 case SIOCSIFADDR: /* set MAC address of the remote side */
1640 TUN_LOCK(tp);
1641 bcopy(data, &tp->tun_ether.octet,
1642 sizeof(tp->tun_ether.octet));
1643 TUN_UNLOCK(tp);
1644
1645 return (0);
1646 case TAPSVNETHDR:
1647 ival = *(int *)data;
1648 if (ival != 0 &&
1649 ival != sizeof(struct virtio_net_hdr) &&
1650 ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1651 return (EINVAL);
1652 }
1653 TUN_LOCK(tp);
1654 tun_vnethdr_set(ifp, ival);
1655 TUN_UNLOCK(tp);
1656
1657 return (0);
1658 case TAPGVNETHDR:
1659 TUN_LOCK(tp);
1660 *(int *)data = tp->tun_vhdrlen;
1661 TUN_UNLOCK(tp);
1662
1663 return (0);
1664 }
1665
1666 /* Fall through to the common ioctls if unhandled */
1667 } else {
1668 switch (cmd) {
1669 case TUNSLMODE:
1670 TUN_LOCK(tp);
1671 if (*(int *)data) {
1672 tp->tun_flags |= TUN_LMODE;
1673 tp->tun_flags &= ~TUN_IFHEAD;
1674 } else
1675 tp->tun_flags &= ~TUN_LMODE;
1676 TUN_UNLOCK(tp);
1677
1678 return (0);
1679 case TUNSIFHEAD:
1680 TUN_LOCK(tp);
1681 if (*(int *)data) {
1682 tp->tun_flags |= TUN_IFHEAD;
1683 tp->tun_flags &= ~TUN_LMODE;
1684 } else
1685 tp->tun_flags &= ~TUN_IFHEAD;
1686 TUN_UNLOCK(tp);
1687
1688 return (0);
1689 case TUNGIFHEAD:
1690 TUN_LOCK(tp);
1691 *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1692 TUN_UNLOCK(tp);
1693
1694 return (0);
1695 case TUNSIFMODE:
1696 /* deny this if UP */
1697 if (TUN2IFP(tp)->if_flags & IFF_UP)
1698 return (EBUSY);
1699
1700 switch (*(int *)data & ~IFF_MULTICAST) {
1701 case IFF_POINTOPOINT:
1702 case IFF_BROADCAST:
1703 TUN_LOCK(tp);
1704 TUN2IFP(tp)->if_flags &=
1705 ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1706 TUN2IFP(tp)->if_flags |= *(int *)data;
1707 TUN_UNLOCK(tp);
1708
1709 break;
1710 default:
1711 return (EINVAL);
1712 }
1713
1714 return (0);
1715 case TUNSIFPID:
1716 TUN_LOCK(tp);
1717 tp->tun_pid = curthread->td_proc->p_pid;
1718 TUN_UNLOCK(tp);
1719
1720 return (0);
1721 }
1722 /* Fall through to the common ioctls if unhandled */
1723 }
1724
1725 switch (cmd) {
1726 case TUNGIFNAME:
1727 ifrp = (struct ifreq *)data;
1728 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1729
1730 return (0);
1731 case TUNSIFINFO:
1732 tunp = (struct tuninfo *)data;
1733 if (TUN2IFP(tp)->if_type != tunp->type)
1734 return (EPROTOTYPE);
1735 TUN_LOCK(tp);
1736 if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1737 strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1738 ifr.ifr_mtu = tunp->mtu;
1739 CURVNET_SET(TUN2IFP(tp)->if_vnet);
1740 error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1741 (caddr_t)&ifr, td);
1742 CURVNET_RESTORE();
1743 if (error) {
1744 TUN_UNLOCK(tp);
1745 return (error);
1746 }
1747 }
1748 TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1749 TUN_UNLOCK(tp);
1750 break;
1751 case TUNGIFINFO:
1752 tunp = (struct tuninfo *)data;
1753 TUN_LOCK(tp);
1754 tunp->mtu = TUN2IFP(tp)->if_mtu;
1755 tunp->type = TUN2IFP(tp)->if_type;
1756 tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1757 TUN_UNLOCK(tp);
1758 break;
1759 case TUNSDEBUG:
1760 tundebug = *(int *)data;
1761 break;
1762 case TUNGDEBUG:
1763 *(int *)data = tundebug;
1764 break;
1765 case TUNSTRANSIENT:
1766 TUN_LOCK(tp);
1767 if (*(int *)data)
1768 tp->tun_flags |= TUN_TRANSIENT;
1769 else
1770 tp->tun_flags &= ~TUN_TRANSIENT;
1771 TUN_UNLOCK(tp);
1772 break;
1773 case TUNGTRANSIENT:
1774 TUN_LOCK(tp);
1775 *(int *)data = (tp->tun_flags & TUN_TRANSIENT) != 0;
1776 TUN_UNLOCK(tp);
1777 break;
1778 case FIONBIO:
1779 break;
1780 case FIOASYNC:
1781 TUN_LOCK(tp);
1782 if (*(int *)data)
1783 tp->tun_flags |= TUN_ASYNC;
1784 else
1785 tp->tun_flags &= ~TUN_ASYNC;
1786 TUN_UNLOCK(tp);
1787 break;
1788 case FIONREAD:
1789 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1790 struct mbuf *mb;
1791 IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1792 IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1793 for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1794 *(int *)data += mb->m_len;
1795 IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1796 } else
1797 *(int *)data = 0;
1798 break;
1799 case FIOSETOWN:
1800 return (fsetown(*(int *)data, &tp->tun_sigio));
1801
1802 case FIOGETOWN:
1803 *(int *)data = fgetown(&tp->tun_sigio);
1804 return (0);
1805
1806 /* This is deprecated, FIOSETOWN should be used instead. */
1807 case TIOCSPGRP:
1808 return (fsetown(-(*(int *)data), &tp->tun_sigio));
1809
1810 /* This is deprecated, FIOGETOWN should be used instead. */
1811 case TIOCGPGRP:
1812 *(int *)data = -fgetown(&tp->tun_sigio);
1813 return (0);
1814
1815 default:
1816 return (ENOTTY);
1817 }
1818 return (0);
1819 }
1820
1821 /*
1822 * The cdevsw read interface - reads a packet at a time, or at
1823 * least as much of a packet as can be read.
1824 */
1825 static int
tunread(struct cdev * dev,struct uio * uio,int flag)1826 tunread(struct cdev *dev, struct uio *uio, int flag)
1827 {
1828 struct tuntap_softc *tp = dev->si_drv1;
1829 struct ifnet *ifp = TUN2IFP(tp);
1830 struct mbuf *m;
1831 size_t len;
1832 int error = 0;
1833
1834 TUNDEBUG (ifp, "read\n");
1835 TUN_LOCK(tp);
1836 if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1837 TUN_UNLOCK(tp);
1838 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1839 return (EHOSTDOWN);
1840 }
1841
1842 tp->tun_flags &= ~TUN_RWAIT;
1843
1844 for (;;) {
1845 IFQ_DEQUEUE(&ifp->if_snd, m);
1846 if (m != NULL)
1847 break;
1848 if (flag & O_NONBLOCK) {
1849 TUN_UNLOCK(tp);
1850 return (EWOULDBLOCK);
1851 }
1852 tp->tun_flags |= TUN_RWAIT;
1853 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | PZERO,
1854 "tunread", 0);
1855 if (error != 0) {
1856 TUN_UNLOCK(tp);
1857 return (error);
1858 }
1859 }
1860 TUN_UNLOCK(tp);
1861
1862 len = min(tp->tun_vhdrlen, uio->uio_resid);
1863 if (len > 0) {
1864 struct virtio_net_hdr_mrg_rxbuf vhdr;
1865
1866 bzero(&vhdr, sizeof(vhdr));
1867 if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1868 m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1869 }
1870
1871 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1872 "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1873 vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1874 vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1875 vhdr.hdr.csum_offset);
1876 error = uiomove(&vhdr, len, uio);
1877 }
1878 if (error == 0)
1879 error = m_mbuftouio(uio, m, 0);
1880 m_freem(m);
1881 return (error);
1882 }
1883
1884 static int
tunwrite_l2(struct tuntap_softc * tp,struct mbuf * m,struct virtio_net_hdr_mrg_rxbuf * vhdr)1885 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1886 struct virtio_net_hdr_mrg_rxbuf *vhdr)
1887 {
1888 struct epoch_tracker et;
1889 struct ether_header *eh;
1890 struct ifnet *ifp;
1891
1892 ifp = TUN2IFP(tp);
1893
1894 /*
1895 * Only pass a unicast frame to ether_input(), if it would
1896 * actually have been received by non-virtual hardware.
1897 */
1898 if (m->m_len < sizeof(struct ether_header)) {
1899 m_freem(m);
1900 return (0);
1901 }
1902
1903 eh = mtod(m, struct ether_header *);
1904
1905 if ((ifp->if_flags & IFF_PROMISC) == 0 &&
1906 !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1907 bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1908 m_freem(m);
1909 return (0);
1910 }
1911
1912 if (vhdr != NULL) {
1913 if (virtio_net_rx_csum(m, &vhdr->hdr)) {
1914 m_freem(m);
1915 return (0);
1916 }
1917 } else {
1918 switch (ntohs(eh->ether_type)) {
1919 #ifdef INET
1920 case ETHERTYPE_IP:
1921 if (ifp->if_capenable & IFCAP_RXCSUM) {
1922 m->m_pkthdr.csum_flags |=
1923 CSUM_IP_CHECKED | CSUM_IP_VALID |
1924 CSUM_DATA_VALID | CSUM_SCTP_VALID |
1925 CSUM_PSEUDO_HDR;
1926 m->m_pkthdr.csum_data = 0xffff;
1927 }
1928 break;
1929 #endif
1930 #ifdef INET6
1931 case ETHERTYPE_IPV6:
1932 if (ifp->if_capenable & IFCAP_RXCSUM_IPV6) {
1933 m->m_pkthdr.csum_flags |=
1934 CSUM_DATA_VALID_IPV6 | CSUM_SCTP_VALID |
1935 CSUM_PSEUDO_HDR;
1936 m->m_pkthdr.csum_data = 0xffff;
1937 }
1938 break;
1939 #endif
1940 }
1941 }
1942
1943 /* Pass packet up to parent. */
1944 CURVNET_SET(ifp->if_vnet);
1945 NET_EPOCH_ENTER(et);
1946 #if defined(INET) || defined(INET6)
1947 if (tp->tun_lro_ready && ifp->if_capenable & IFCAP_LRO &&
1948 tcp_lro_rx(&tp->tun_lro, m, 0) == 0)
1949 tcp_lro_flush_all(&tp->tun_lro);
1950 else
1951 #endif
1952 (*ifp->if_input)(ifp, m);
1953 NET_EPOCH_EXIT(et);
1954 CURVNET_RESTORE();
1955 /* ibytes are counted in parent */
1956 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1957 return (0);
1958 }
1959
1960 static int
tunwrite_l3(struct tuntap_softc * tp,struct mbuf * m)1961 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1962 {
1963 struct epoch_tracker et;
1964 struct ifnet *ifp;
1965 int family, isr;
1966
1967 ifp = TUN2IFP(tp);
1968 /* Could be unlocked read? */
1969 TUN_LOCK(tp);
1970 if (tp->tun_flags & TUN_IFHEAD) {
1971 TUN_UNLOCK(tp);
1972 if (m->m_len < sizeof(family) &&
1973 (m = m_pullup(m, sizeof(family))) == NULL)
1974 return (ENOBUFS);
1975 family = ntohl(*mtod(m, u_int32_t *));
1976 m_adj(m, sizeof(family));
1977 } else {
1978 TUN_UNLOCK(tp);
1979 family = AF_INET;
1980 }
1981
1982 BPF_MTAP2(ifp, &family, sizeof(family), m);
1983
1984 switch (family) {
1985 #ifdef INET
1986 case AF_INET:
1987 isr = NETISR_IP;
1988 break;
1989 #endif
1990 #ifdef INET6
1991 case AF_INET6:
1992 isr = NETISR_IPV6;
1993 break;
1994 #endif
1995 default:
1996 m_freem(m);
1997 return (EAFNOSUPPORT);
1998 }
1999 random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
2000 if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
2001 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
2002 CURVNET_SET(ifp->if_vnet);
2003 M_SETFIB(m, ifp->if_fib);
2004 NET_EPOCH_ENTER(et);
2005 netisr_dispatch(isr, m);
2006 NET_EPOCH_EXIT(et);
2007 CURVNET_RESTORE();
2008 return (0);
2009 }
2010
2011 /*
2012 * the cdevsw write interface - an atomic write is a packet - or else!
2013 */
2014 static int
tunwrite(struct cdev * dev,struct uio * uio,int flag)2015 tunwrite(struct cdev *dev, struct uio *uio, int flag)
2016 {
2017 struct virtio_net_hdr_mrg_rxbuf vhdr;
2018 struct tuntap_softc *tp;
2019 struct ifnet *ifp;
2020 struct mbuf *m;
2021 uint32_t mru;
2022 int align, vhdrlen, error;
2023 bool l2tun;
2024
2025 tp = dev->si_drv1;
2026 ifp = TUN2IFP(tp);
2027 TUNDEBUG(ifp, "tunwrite\n");
2028 if ((ifp->if_flags & IFF_UP) != IFF_UP)
2029 /* ignore silently */
2030 return (0);
2031
2032 if (uio->uio_resid == 0)
2033 return (0);
2034
2035 l2tun = (tp->tun_flags & TUN_L2) != 0;
2036 mru = l2tun ? TAPMRU : TUNMRU;
2037 vhdrlen = tp->tun_vhdrlen;
2038 align = 0;
2039 if (l2tun) {
2040 align = ETHER_ALIGN;
2041 mru += vhdrlen;
2042 } else if ((tp->tun_flags & TUN_IFHEAD) != 0)
2043 mru += sizeof(uint32_t); /* family */
2044 if (uio->uio_resid < 0 || uio->uio_resid > mru) {
2045 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
2046 return (EIO);
2047 }
2048
2049 if (vhdrlen > 0) {
2050 error = uiomove(&vhdr, vhdrlen, uio);
2051 if (error != 0)
2052 return (error);
2053 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
2054 "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
2055 vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
2056 vhdr.hdr.gso_size, vhdr.hdr.csum_start,
2057 vhdr.hdr.csum_offset);
2058 }
2059
2060 if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
2061 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2062 return (ENOBUFS);
2063 }
2064
2065 m->m_pkthdr.rcvif = ifp;
2066 #ifdef MAC
2067 mac_ifnet_create_mbuf(ifp, m);
2068 #endif
2069
2070 if (l2tun)
2071 return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
2072
2073 return (tunwrite_l3(tp, m));
2074 }
2075
2076 /*
2077 * tunpoll - the poll interface, this is only useful on reads
2078 * really. The write detect always returns true, write never blocks
2079 * anyway, it either accepts the packet or drops it.
2080 */
2081 static int
tunpoll(struct cdev * dev,int events,struct thread * td)2082 tunpoll(struct cdev *dev, int events, struct thread *td)
2083 {
2084 struct tuntap_softc *tp = dev->si_drv1;
2085 struct ifnet *ifp = TUN2IFP(tp);
2086 int revents = 0;
2087
2088 TUNDEBUG(ifp, "tunpoll\n");
2089
2090 if (events & (POLLIN | POLLRDNORM)) {
2091 IFQ_LOCK(&ifp->if_snd);
2092 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
2093 TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
2094 revents |= events & (POLLIN | POLLRDNORM);
2095 } else {
2096 TUNDEBUG(ifp, "tunpoll waiting\n");
2097 selrecord(td, &tp->tun_rsel);
2098 }
2099 IFQ_UNLOCK(&ifp->if_snd);
2100 }
2101 revents |= events & (POLLOUT | POLLWRNORM);
2102
2103 return (revents);
2104 }
2105
2106 /*
2107 * tunkqfilter - support for the kevent() system call.
2108 */
2109 static int
tunkqfilter(struct cdev * dev,struct knote * kn)2110 tunkqfilter(struct cdev *dev, struct knote *kn)
2111 {
2112 struct tuntap_softc *tp = dev->si_drv1;
2113 struct ifnet *ifp = TUN2IFP(tp);
2114
2115 switch(kn->kn_filter) {
2116 case EVFILT_READ:
2117 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
2118 ifp->if_xname, dev2unit(dev));
2119 kn->kn_fop = &tun_read_filterops;
2120 break;
2121
2122 case EVFILT_WRITE:
2123 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
2124 ifp->if_xname, dev2unit(dev));
2125 kn->kn_fop = &tun_write_filterops;
2126 break;
2127
2128 default:
2129 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
2130 ifp->if_xname, dev2unit(dev));
2131 return(EINVAL);
2132 }
2133
2134 kn->kn_hook = tp;
2135 knlist_add(&tp->tun_rsel.si_note, kn, 0);
2136
2137 return (0);
2138 }
2139
2140 /*
2141 * Return true of there is data in the interface queue.
2142 */
2143 static int
tunkqread(struct knote * kn,long hint)2144 tunkqread(struct knote *kn, long hint)
2145 {
2146 int ret;
2147 struct tuntap_softc *tp = kn->kn_hook;
2148 struct cdev *dev = tp->tun_dev;
2149 struct ifnet *ifp = TUN2IFP(tp);
2150
2151 if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
2152 TUNDEBUG(ifp,
2153 "%s have data in the queue. Len = %d, minor = %#x\n",
2154 ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
2155 ret = 1;
2156 } else {
2157 TUNDEBUG(ifp,
2158 "%s waiting for data, minor = %#x\n", ifp->if_xname,
2159 dev2unit(dev));
2160 ret = 0;
2161 }
2162
2163 return (ret);
2164 }
2165
2166 /*
2167 * Always can write, always return MTU in kn->data.
2168 */
2169 static int
tunkqwrite(struct knote * kn,long hint)2170 tunkqwrite(struct knote *kn, long hint)
2171 {
2172 struct tuntap_softc *tp = kn->kn_hook;
2173 struct ifnet *ifp = TUN2IFP(tp);
2174
2175 kn->kn_data = ifp->if_mtu;
2176
2177 return (1);
2178 }
2179
2180 static void
tunkqdetach(struct knote * kn)2181 tunkqdetach(struct knote *kn)
2182 {
2183 struct tuntap_softc *tp = kn->kn_hook;
2184
2185 knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2186 }
2187