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