FreeBSD/Linux Kernel Cross Reference
sys/net/if_tap.c
1 /* $NetBSD: if_tap.c,v 1.24.2.1 2009/12/03 09:44:35 sborrill Exp $ */
2
3 /*
4 * Copyright (c) 2003, 2004 The NetBSD Foundation.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to the NetBSD Foundation
8 * by Quentin Garnier.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 /*
40 * tap(4) is a virtual Ethernet interface. It appears as a real Ethernet
41 * device to the system, but can also be accessed by userland through a
42 * character device interface, which allows reading and injecting frames.
43 */
44
45 #include <sys/cdefs.h>
46 __KERNEL_RCSID(0, "$NetBSD: if_tap.c,v 1.24.2.1 2009/12/03 09:44:35 sborrill Exp $");
47
48 #if defined(_KERNEL_OPT)
49 #include "bpfilter.h"
50 #endif
51
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kernel.h>
55 #include <sys/malloc.h>
56 #include <sys/conf.h>
57 #include <sys/device.h>
58 #include <sys/file.h>
59 #include <sys/filedesc.h>
60 #include <sys/ksyms.h>
61 #include <sys/poll.h>
62 #include <sys/select.h>
63 #include <sys/sockio.h>
64 #include <sys/sysctl.h>
65 #include <sys/kauth.h>
66
67 #include <net/if.h>
68 #include <net/if_dl.h>
69 #include <net/if_ether.h>
70 #include <net/if_media.h>
71 #include <net/if_tap.h>
72 #if NBPFILTER > 0
73 #include <net/bpf.h>
74 #endif
75
76 /*
77 * sysctl node management
78 *
79 * It's not really possible to use a SYSCTL_SETUP block with
80 * current LKM implementation, so it is easier to just define
81 * our own function.
82 *
83 * The handler function is a "helper" in Andrew Brown's sysctl
84 * framework terminology. It is used as a gateway for sysctl
85 * requests over the nodes.
86 *
87 * tap_log allows the module to log creations of nodes and
88 * destroy them all at once using sysctl_teardown.
89 */
90 static int tap_node;
91 static int tap_sysctl_handler(SYSCTLFN_PROTO);
92 SYSCTL_SETUP_PROTO(sysctl_tap_setup);
93
94 /*
95 * Since we're an Ethernet device, we need the 3 following
96 * components: a leading struct device, a struct ethercom,
97 * and also a struct ifmedia since we don't attach a PHY to
98 * ourselves. We could emulate one, but there's no real
99 * point.
100 */
101
102 struct tap_softc {
103 struct device sc_dev;
104 struct ifmedia sc_im;
105 struct ethercom sc_ec;
106 int sc_flags;
107 #define TAP_INUSE 0x00000001 /* tap device can only be opened once */
108 #define TAP_ASYNCIO 0x00000002 /* user is using async I/O (SIGIO) on the device */
109 #define TAP_NBIO 0x00000004 /* user wants calls to avoid blocking */
110 #define TAP_GOING 0x00000008 /* interface is being destroyed */
111 struct selinfo sc_rsel;
112 pid_t sc_pgid; /* For async. IO */
113 struct lock sc_rdlock;
114 struct simplelock sc_kqlock;
115 };
116
117 /* autoconf(9) glue */
118
119 void tapattach(int);
120
121 static int tap_match(struct device *, struct cfdata *, void *);
122 static void tap_attach(struct device *, struct device *, void *);
123 static int tap_detach(struct device*, int);
124
125 CFATTACH_DECL(tap, sizeof(struct tap_softc),
126 tap_match, tap_attach, tap_detach, NULL);
127 extern struct cfdriver tap_cd;
128
129 /* Real device access routines */
130 static int tap_dev_close(struct tap_softc *);
131 static int tap_dev_read(int, struct uio *, int);
132 static int tap_dev_write(int, struct uio *, int);
133 static int tap_dev_ioctl(int, u_long, caddr_t, struct lwp *);
134 static int tap_dev_poll(int, int, struct lwp *);
135 static int tap_dev_kqfilter(int, struct knote *);
136
137 /* Fileops access routines */
138 static int tap_fops_close(struct file *, struct lwp *);
139 static int tap_fops_read(struct file *, off_t *, struct uio *,
140 kauth_cred_t, int);
141 static int tap_fops_write(struct file *, off_t *, struct uio *,
142 kauth_cred_t, int);
143 static int tap_fops_ioctl(struct file *, u_long, void *,
144 struct lwp *);
145 static int tap_fops_poll(struct file *, int, struct lwp *);
146 static int tap_fops_kqfilter(struct file *, struct knote *);
147
148 static const struct fileops tap_fileops = {
149 tap_fops_read,
150 tap_fops_write,
151 tap_fops_ioctl,
152 fnullop_fcntl,
153 tap_fops_poll,
154 fbadop_stat,
155 tap_fops_close,
156 tap_fops_kqfilter,
157 };
158
159 /* Helper for cloning open() */
160 static int tap_dev_cloner(struct lwp *);
161
162 /* Character device routines */
163 static int tap_cdev_open(dev_t, int, int, struct lwp *);
164 static int tap_cdev_close(dev_t, int, int, struct lwp *);
165 static int tap_cdev_read(dev_t, struct uio *, int);
166 static int tap_cdev_write(dev_t, struct uio *, int);
167 static int tap_cdev_ioctl(dev_t, u_long, caddr_t, int, struct lwp *);
168 static int tap_cdev_poll(dev_t, int, struct lwp *);
169 static int tap_cdev_kqfilter(dev_t, struct knote *);
170
171 const struct cdevsw tap_cdevsw = {
172 tap_cdev_open, tap_cdev_close,
173 tap_cdev_read, tap_cdev_write,
174 tap_cdev_ioctl, nostop, notty,
175 tap_cdev_poll, nommap,
176 tap_cdev_kqfilter,
177 D_OTHER,
178 };
179
180 #define TAP_CLONER 0xfffff /* Maximal minor value */
181
182 /* kqueue-related routines */
183 static void tap_kqdetach(struct knote *);
184 static int tap_kqread(struct knote *, long);
185
186 /*
187 * Those are needed by the if_media interface.
188 */
189
190 static int tap_mediachange(struct ifnet *);
191 static void tap_mediastatus(struct ifnet *, struct ifmediareq *);
192
193 /*
194 * Those are needed by the ifnet interface, and would typically be
195 * there for any network interface driver.
196 * Some other routines are optional: watchdog and drain.
197 */
198
199 static void tap_start(struct ifnet *);
200 static void tap_stop(struct ifnet *, int);
201 static int tap_init(struct ifnet *);
202 static int tap_ioctl(struct ifnet *, u_long, caddr_t);
203
204 /* This is an internal function to keep tap_ioctl readable */
205 static int tap_lifaddr(struct ifnet *, u_long, struct ifaliasreq *);
206
207 /*
208 * tap is a clonable interface, although it is highly unrealistic for
209 * an Ethernet device.
210 *
211 * Here are the bits needed for a clonable interface.
212 */
213 static int tap_clone_create(struct if_clone *, int);
214 static int tap_clone_destroy(struct ifnet *);
215
216 struct if_clone tap_cloners = IF_CLONE_INITIALIZER("tap",
217 tap_clone_create,
218 tap_clone_destroy);
219
220 /* Helper functionis shared by the two cloning code paths */
221 static struct tap_softc * tap_clone_creator(int);
222 int tap_clone_destroyer(struct device *);
223
224 void
225 tapattach(int n)
226 {
227 int error;
228
229 error = config_cfattach_attach(tap_cd.cd_name, &tap_ca);
230 if (error) {
231 aprint_error("%s: unable to register cfattach\n",
232 tap_cd.cd_name);
233 (void)config_cfdriver_detach(&tap_cd);
234 return;
235 }
236
237 if_clone_attach(&tap_cloners);
238 }
239
240 /* Pretty much useless for a pseudo-device */
241 static int
242 tap_match(struct device *self, struct cfdata *cfdata,
243 void *arg)
244 {
245 return (1);
246 }
247
248 void
249 tap_attach(struct device *parent, struct device *self,
250 void *aux)
251 {
252 struct tap_softc *sc = (struct tap_softc *)self;
253 struct ifnet *ifp;
254 const struct sysctlnode *node;
255 u_int8_t enaddr[ETHER_ADDR_LEN] =
256 { 0xf2, 0x0b, 0xa4, 0xff, 0xff, 0xff };
257 char enaddrstr[3 * ETHER_ADDR_LEN];
258 struct timeval tv;
259 uint32_t ui;
260 int error;
261
262 aprint_normal("%s: faking Ethernet device\n",
263 self->dv_xname);
264
265 /*
266 * In order to obtain unique initial Ethernet address on a host,
267 * do some randomisation using the current uptime. It's not meant
268 * for anything but avoiding hard-coding an address.
269 */
270 getmicrouptime(&tv);
271 ui = (tv.tv_sec ^ tv.tv_usec) & 0xffffff;
272 memcpy(enaddr+3, (u_int8_t *)&ui, 3);
273
274 aprint_normal("%s: Ethernet address %s\n", sc->sc_dev.dv_xname,
275 ether_snprintf(enaddrstr, sizeof(enaddrstr), enaddr));
276
277 /*
278 * Why 1000baseT? Why not? You can add more.
279 *
280 * Note that there are 3 steps: init, one or several additions to
281 * list of supported media, and in the end, the selection of one
282 * of them.
283 */
284 ifmedia_init(&sc->sc_im, 0, tap_mediachange, tap_mediastatus);
285 ifmedia_add(&sc->sc_im, IFM_ETHER|IFM_1000_T, 0, NULL);
286 ifmedia_add(&sc->sc_im, IFM_ETHER|IFM_1000_T|IFM_FDX, 0, NULL);
287 ifmedia_add(&sc->sc_im, IFM_ETHER|IFM_100_TX, 0, NULL);
288 ifmedia_add(&sc->sc_im, IFM_ETHER|IFM_100_TX|IFM_FDX, 0, NULL);
289 ifmedia_add(&sc->sc_im, IFM_ETHER|IFM_10_T, 0, NULL);
290 ifmedia_add(&sc->sc_im, IFM_ETHER|IFM_10_T|IFM_FDX, 0, NULL);
291 ifmedia_add(&sc->sc_im, IFM_ETHER|IFM_AUTO, 0, NULL);
292 ifmedia_set(&sc->sc_im, IFM_ETHER|IFM_AUTO);
293
294 /*
295 * One should note that an interface must do multicast in order
296 * to support IPv6.
297 */
298 ifp = &sc->sc_ec.ec_if;
299 strcpy(ifp->if_xname, sc->sc_dev.dv_xname);
300 ifp->if_softc = sc;
301 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
302 ifp->if_ioctl = tap_ioctl;
303 ifp->if_start = tap_start;
304 ifp->if_stop = tap_stop;
305 ifp->if_init = tap_init;
306 IFQ_SET_READY(&ifp->if_snd);
307
308 sc->sc_ec.ec_capabilities = ETHERCAP_VLAN_MTU | ETHERCAP_JUMBO_MTU;
309
310 /* Those steps are mandatory for an Ethernet driver, the fisrt call
311 * being common to all network interface drivers. */
312 if_attach(ifp);
313 ether_ifattach(ifp, enaddr);
314
315 sc->sc_flags = 0;
316
317 /*
318 * Add a sysctl node for that interface.
319 *
320 * The pointer transmitted is not a string, but instead a pointer to
321 * the softc structure, which we can use to build the string value on
322 * the fly in the helper function of the node. See the comments for
323 * tap_sysctl_handler for details.
324 *
325 * Usually sysctl_createv is called with CTL_CREATE as the before-last
326 * component. However, we can allocate a number ourselves, as we are
327 * the only consumer of the net.link.<iface> node. In this case, the
328 * unit number is conveniently used to number the node. CTL_CREATE
329 * would just work, too.
330 */
331 if ((error = sysctl_createv(NULL, 0, NULL,
332 &node, CTLFLAG_READWRITE,
333 CTLTYPE_STRING, sc->sc_dev.dv_xname, NULL,
334 tap_sysctl_handler, 0, sc, 18,
335 CTL_NET, AF_LINK, tap_node, device_unit(&sc->sc_dev),
336 CTL_EOL)) != 0)
337 aprint_error("%s: sysctl_createv returned %d, ignoring\n",
338 sc->sc_dev.dv_xname, error);
339
340 /*
341 * Initialize the two locks for the device.
342 *
343 * We need a lock here because even though the tap device can be
344 * opened only once, the file descriptor might be passed to another
345 * process, say a fork(2)ed child.
346 *
347 * The Giant saves us from most of the hassle, but since the read
348 * operation can sleep, we don't want two processes to wake up at
349 * the same moment and both try and dequeue a single packet.
350 *
351 * The queue for event listeners (used by kqueue(9), see below) has
352 * to be protected, too, but we don't need the same level of
353 * complexity for that lock, so a simple spinning lock is fine.
354 */
355 lockinit(&sc->sc_rdlock, PSOCK|PCATCH, "tapl", 0, LK_SLEEPFAIL);
356 simple_lock_init(&sc->sc_kqlock);
357 }
358
359 /*
360 * When detaching, we do the inverse of what is done in the attach
361 * routine, in reversed order.
362 */
363 static int
364 tap_detach(struct device* self, int flags)
365 {
366 struct tap_softc *sc = (struct tap_softc *)self;
367 struct ifnet *ifp = &sc->sc_ec.ec_if;
368 int error, s;
369
370 /*
371 * Some processes might be sleeping on "tap", so we have to make
372 * them release their hold on the device.
373 *
374 * The LK_DRAIN operation will wait for every locked process to
375 * release their hold.
376 */
377 sc->sc_flags |= TAP_GOING;
378 s = splnet();
379 tap_stop(ifp, 1);
380 if_down(ifp);
381 splx(s);
382 lockmgr(&sc->sc_rdlock, LK_DRAIN, NULL);
383
384 /*
385 * Destroying a single leaf is a very straightforward operation using
386 * sysctl_destroyv. One should be sure to always end the path with
387 * CTL_EOL.
388 */
389 if ((error = sysctl_destroyv(NULL, CTL_NET, AF_LINK, tap_node,
390 device_unit(&sc->sc_dev), CTL_EOL)) != 0)
391 aprint_error("%s: sysctl_destroyv returned %d, ignoring\n",
392 sc->sc_dev.dv_xname, error);
393 ether_ifdetach(ifp);
394 if_detach(ifp);
395 ifmedia_delete_instance(&sc->sc_im, IFM_INST_ANY);
396
397 return (0);
398 }
399
400 /*
401 * This function is called by the ifmedia layer to notify the driver
402 * that the user requested a media change. A real driver would
403 * reconfigure the hardware.
404 */
405 static int
406 tap_mediachange(struct ifnet *ifp)
407 {
408 return (0);
409 }
410
411 /*
412 * Here the user asks for the currently used media.
413 */
414 static void
415 tap_mediastatus(struct ifnet *ifp, struct ifmediareq *imr)
416 {
417 struct tap_softc *sc = (struct tap_softc *)ifp->if_softc;
418 imr->ifm_active = sc->sc_im.ifm_cur->ifm_media;
419 }
420
421 /*
422 * This is the function where we SEND packets.
423 *
424 * There is no 'receive' equivalent. A typical driver will get
425 * interrupts from the hardware, and from there will inject new packets
426 * into the network stack.
427 *
428 * Once handled, a packet must be freed. A real driver might not be able
429 * to fit all the pending packets into the hardware, and is allowed to
430 * return before having sent all the packets. It should then use the
431 * if_flags flag IFF_OACTIVE to notify the upper layer.
432 *
433 * There are also other flags one should check, such as IFF_PAUSE.
434 *
435 * It is our duty to make packets available to BPF listeners.
436 *
437 * You should be aware that this function is called by the Ethernet layer
438 * at splnet().
439 *
440 * When the device is opened, we have to pass the packet(s) to the
441 * userland. For that we stay in OACTIVE mode while the userland gets
442 * the packets, and we send a signal to the processes waiting to read.
443 *
444 * wakeup(sc) is the counterpart to the tsleep call in
445 * tap_dev_read, while selnotify() is used for kevent(2) and
446 * poll(2) (which includes select(2)) listeners.
447 */
448 static void
449 tap_start(struct ifnet *ifp)
450 {
451 struct tap_softc *sc = (struct tap_softc *)ifp->if_softc;
452 struct mbuf *m0;
453
454 if ((sc->sc_flags & TAP_INUSE) == 0) {
455 /* Simply drop packets */
456 for(;;) {
457 IFQ_DEQUEUE(&ifp->if_snd, m0);
458 if (m0 == NULL)
459 return;
460
461 ifp->if_opackets++;
462 #if NBPFILTER > 0
463 if (ifp->if_bpf)
464 bpf_mtap(ifp->if_bpf, m0);
465 #endif
466
467 m_freem(m0);
468 }
469 } else if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
470 ifp->if_flags |= IFF_OACTIVE;
471 wakeup(sc);
472 selnotify(&sc->sc_rsel, 1);
473 if (sc->sc_flags & TAP_ASYNCIO)
474 fownsignal(sc->sc_pgid, SIGIO, POLL_IN,
475 POLLIN|POLLRDNORM, NULL);
476 }
477 }
478
479 /*
480 * A typical driver will only contain the following handlers for
481 * ioctl calls, except SIOCSIFPHYADDR.
482 * The latter is a hack I used to set the Ethernet address of the
483 * faked device.
484 *
485 * Note that both ifmedia_ioctl() and ether_ioctl() have to be
486 * called under splnet().
487 */
488 static int
489 tap_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
490 {
491 struct tap_softc *sc = (struct tap_softc *)ifp->if_softc;
492 struct ifreq *ifr = (struct ifreq *)data;
493 int s, error;
494
495 s = splnet();
496
497 switch (cmd) {
498 case SIOCSIFMEDIA:
499 case SIOCGIFMEDIA:
500 error = ifmedia_ioctl(ifp, ifr, &sc->sc_im, cmd);
501 break;
502 case SIOCSIFPHYADDR:
503 error = tap_lifaddr(ifp, cmd, (struct ifaliasreq *)data);
504 break;
505 default:
506 error = ether_ioctl(ifp, cmd, data);
507 if (error == ENETRESET)
508 error = 0;
509 break;
510 }
511
512 splx(s);
513
514 return (error);
515 }
516
517 /*
518 * Helper function to set Ethernet address. This shouldn't be done there,
519 * and should actually be available to all Ethernet drivers, real or not.
520 */
521 static int
522 tap_lifaddr(struct ifnet *ifp, u_long cmd, struct ifaliasreq *ifra)
523 {
524 struct sockaddr *sa = (struct sockaddr *)&ifra->ifra_addr;
525
526 if (sa->sa_family != AF_LINK)
527 return (EINVAL);
528
529 memcpy(LLADDR(ifp->if_sadl), sa->sa_data, ETHER_ADDR_LEN);
530
531 return (0);
532 }
533
534 /*
535 * _init() would typically be called when an interface goes up,
536 * meaning it should configure itself into the state in which it
537 * can send packets.
538 */
539 static int
540 tap_init(struct ifnet *ifp)
541 {
542 ifp->if_flags |= IFF_RUNNING;
543
544 tap_start(ifp);
545
546 return (0);
547 }
548
549 /*
550 * _stop() is called when an interface goes down. It is our
551 * responsability to validate that state by clearing the
552 * IFF_RUNNING flag.
553 *
554 * We have to wake up all the sleeping processes to have the pending
555 * read requests cancelled.
556 */
557 static void
558 tap_stop(struct ifnet *ifp, int disable)
559 {
560 struct tap_softc *sc = (struct tap_softc *)ifp->if_softc;
561
562 ifp->if_flags &= ~IFF_RUNNING;
563 wakeup(sc);
564 selnotify(&sc->sc_rsel, 1);
565 if (sc->sc_flags & TAP_ASYNCIO)
566 fownsignal(sc->sc_pgid, SIGIO, POLL_HUP, 0, NULL);
567 }
568
569 /*
570 * The 'create' command of ifconfig can be used to create
571 * any numbered instance of a given device. Thus we have to
572 * make sure we have enough room in cd_devs to create the
573 * user-specified instance. config_attach_pseudo will do this
574 * for us.
575 */
576 static int
577 tap_clone_create(struct if_clone *ifc, int unit)
578 {
579 if (tap_clone_creator(unit) == NULL) {
580 aprint_error("%s%d: unable to attach an instance\n",
581 tap_cd.cd_name, unit);
582 return (ENXIO);
583 }
584
585 return (0);
586 }
587
588 /*
589 * tap(4) can be cloned by two ways:
590 * using 'ifconfig tap0 create', which will use the network
591 * interface cloning API, and call tap_clone_create above.
592 * opening the cloning device node, whose minor number is TAP_CLONER.
593 * See below for an explanation on how this part work.
594 *
595 * config_attach_pseudo can be called with unit = DVUNIT_ANY to have
596 * autoconf(9) choose a unit number for us. This is what happens when
597 * the cloner is openend, while the ifcloner interface creates a device
598 * with a specific unit number.
599 */
600 static struct tap_softc *
601 tap_clone_creator(int unit)
602 {
603 struct cfdata *cf;
604
605 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
606 cf->cf_name = tap_cd.cd_name;
607 cf->cf_atname = tap_ca.ca_name;
608 cf->cf_unit = unit;
609 cf->cf_fstate = FSTATE_STAR;
610
611 return (struct tap_softc *)config_attach_pseudo(cf);
612 }
613
614 /*
615 * The clean design of if_clone and autoconf(9) makes that part
616 * really straightforward. The second argument of config_detach
617 * means neither QUIET nor FORCED.
618 */
619 static int
620 tap_clone_destroy(struct ifnet *ifp)
621 {
622 return tap_clone_destroyer((struct device *)ifp->if_softc);
623 }
624
625 int
626 tap_clone_destroyer(struct device *dev)
627 {
628 struct cfdata *cf = device_cfdata(dev);
629 int error;
630
631 if ((error = config_detach(dev, 0)) != 0)
632 aprint_error("%s: unable to detach instance\n",
633 dev->dv_xname);
634 free(cf, M_DEVBUF);
635
636 return (error);
637 }
638
639 /*
640 * tap(4) is a bit of an hybrid device. It can be used in two different
641 * ways:
642 * 1. ifconfig tapN create, then use /dev/tapN to read/write off it.
643 * 2. open /dev/tap, get a new interface created and read/write off it.
644 * That interface is destroyed when the process that had it created exits.
645 *
646 * The first way is managed by the cdevsw structure, and you access interfaces
647 * through a (major, minor) mapping: tap4 is obtained by the minor number
648 * 4. The entry points for the cdevsw interface are prefixed by tap_cdev_.
649 *
650 * The second way is the so-called "cloning" device. It's a special minor
651 * number (chosen as the maximal number, to allow as much tap devices as
652 * possible). The user first opens the cloner (e.g., /dev/tap), and that
653 * call ends in tap_cdev_open. The actual place where it is handled is
654 * tap_dev_cloner.
655 *
656 * An tap device cannot be opened more than once at a time, so the cdevsw
657 * part of open() does nothing but noting that the interface is being used and
658 * hence ready to actually handle packets.
659 */
660
661 static int
662 tap_cdev_open(dev_t dev, int flags, int fmt, struct lwp *l)
663 {
664 struct tap_softc *sc;
665
666 if (minor(dev) == TAP_CLONER)
667 return tap_dev_cloner(l);
668
669 sc = (struct tap_softc *)device_lookup(&tap_cd, minor(dev));
670 if (sc == NULL)
671 return (ENXIO);
672
673 /* The device can only be opened once */
674 if (sc->sc_flags & TAP_INUSE)
675 return (EBUSY);
676 sc->sc_flags |= TAP_INUSE;
677 return (0);
678 }
679
680 /*
681 * There are several kinds of cloning devices, and the most simple is the one
682 * tap(4) uses. What it does is change the file descriptor with a new one,
683 * with its own fileops structure (which maps to the various read, write,
684 * ioctl functions). It starts allocating a new file descriptor with falloc,
685 * then actually creates the new tap devices.
686 *
687 * Once those two steps are successful, we can re-wire the existing file
688 * descriptor to its new self. This is done with fdclone(): it fills the fp
689 * structure as needed (notably f_data gets filled with the fifth parameter
690 * passed, the unit of the tap device which will allows us identifying the
691 * device later), and returns EMOVEFD.
692 *
693 * That magic value is interpreted by sys_open() which then replaces the
694 * current file descriptor by the new one (through a magic member of struct
695 * lwp, l_dupfd).
696 *
697 * The tap device is flagged as being busy since it otherwise could be
698 * externally accessed through the corresponding device node with the cdevsw
699 * interface.
700 */
701
702 static int
703 tap_dev_cloner(struct lwp *l)
704 {
705 struct tap_softc *sc;
706 struct file *fp;
707 int error, fd;
708
709 if ((error = falloc(l, &fp, &fd)) != 0)
710 return (error);
711
712 if ((sc = tap_clone_creator(DVUNIT_ANY)) == NULL) {
713 FILE_UNUSE(fp, l);
714 ffree(fp);
715 return (ENXIO);
716 }
717
718 sc->sc_flags |= TAP_INUSE;
719
720 return fdclone(l, fp, fd, FREAD|FWRITE, &tap_fileops,
721 (void *)(intptr_t)device_unit(&sc->sc_dev));
722 }
723
724 /*
725 * While all other operations (read, write, ioctl, poll and kqfilter) are
726 * really the same whether we are in cdevsw or fileops mode, the close()
727 * function is slightly different in the two cases.
728 *
729 * As for the other, the core of it is shared in tap_dev_close. What
730 * it does is sufficient for the cdevsw interface, but the cloning interface
731 * needs another thing: the interface is destroyed when the processes that
732 * created it closes it.
733 */
734 static int
735 tap_cdev_close(dev_t dev, int flags, int fmt,
736 struct lwp *l)
737 {
738 struct tap_softc *sc =
739 (struct tap_softc *)device_lookup(&tap_cd, minor(dev));
740
741 if (sc == NULL)
742 return (ENXIO);
743
744 return tap_dev_close(sc);
745 }
746
747 /*
748 * It might happen that the administrator used ifconfig to externally destroy
749 * the interface. In that case, tap_fops_close will be called while
750 * tap_detach is already happening. If we called it again from here, we
751 * would dead lock. TAP_GOING ensures that this situation doesn't happen.
752 */
753 static int
754 tap_fops_close(struct file *fp, struct lwp *l)
755 {
756 int unit = (intptr_t)fp->f_data;
757 struct tap_softc *sc;
758 int error;
759
760 sc = (struct tap_softc *)device_lookup(&tap_cd, unit);
761 if (sc == NULL)
762 return (ENXIO);
763
764 /* tap_dev_close currently always succeeds, but it might not
765 * always be the case. */
766 if ((error = tap_dev_close(sc)) != 0)
767 return (error);
768
769 /* Destroy the device now that it is no longer useful,
770 * unless it's already being destroyed. */
771 if ((sc->sc_flags & TAP_GOING) != 0)
772 return (0);
773
774 return tap_clone_destroyer((struct device *)sc);
775 }
776
777 static int
778 tap_dev_close(struct tap_softc *sc)
779 {
780 struct ifnet *ifp;
781 int s;
782
783 s = splnet();
784 /* Let tap_start handle packets again */
785 ifp = &sc->sc_ec.ec_if;
786 ifp->if_flags &= ~IFF_OACTIVE;
787
788 /* Purge output queue */
789 if (!(IFQ_IS_EMPTY(&ifp->if_snd))) {
790 struct mbuf *m;
791
792 for (;;) {
793 IFQ_DEQUEUE(&ifp->if_snd, m);
794 if (m == NULL)
795 break;
796
797 ifp->if_opackets++;
798 #if NBPFILTER > 0
799 if (ifp->if_bpf)
800 bpf_mtap(ifp->if_bpf, m);
801 #endif
802 m_freem(m);
803 }
804 }
805 splx(s);
806
807 sc->sc_flags &= ~(TAP_INUSE | TAP_ASYNCIO);
808
809 return (0);
810 }
811
812 static int
813 tap_cdev_read(dev_t dev, struct uio *uio, int flags)
814 {
815 return tap_dev_read(minor(dev), uio, flags);
816 }
817
818 static int
819 tap_fops_read(struct file *fp, off_t *offp, struct uio *uio,
820 kauth_cred_t cred, int flags)
821 {
822 return tap_dev_read((intptr_t)fp->f_data, uio, flags);
823 }
824
825 static int
826 tap_dev_read(int unit, struct uio *uio, int flags)
827 {
828 struct tap_softc *sc =
829 (struct tap_softc *)device_lookup(&tap_cd, unit);
830 struct ifnet *ifp;
831 struct mbuf *m, *n;
832 int error = 0, s;
833
834 if (sc == NULL)
835 return (ENXIO);
836
837 ifp = &sc->sc_ec.ec_if;
838 if ((ifp->if_flags & IFF_UP) == 0)
839 return (EHOSTDOWN);
840
841 /*
842 * In the TAP_NBIO case, we have to make sure we won't be sleeping
843 */
844 if ((sc->sc_flags & TAP_NBIO) &&
845 lockstatus(&sc->sc_rdlock) == LK_EXCLUSIVE)
846 return (EWOULDBLOCK);
847 error = lockmgr(&sc->sc_rdlock, LK_EXCLUSIVE, NULL);
848 if (error != 0)
849 return (error);
850
851 s = splnet();
852 if (IFQ_IS_EMPTY(&ifp->if_snd)) {
853 ifp->if_flags &= ~IFF_OACTIVE;
854 splx(s);
855 /*
856 * We must release the lock before sleeping, and re-acquire it
857 * after.
858 */
859 (void)lockmgr(&sc->sc_rdlock, LK_RELEASE, NULL);
860 if (sc->sc_flags & TAP_NBIO)
861 error = EWOULDBLOCK;
862 else
863 error = tsleep(sc, PSOCK|PCATCH, "tap", 0);
864
865 if (error != 0)
866 return (error);
867 /* The device might have been downed */
868 if ((ifp->if_flags & IFF_UP) == 0)
869 return (EHOSTDOWN);
870 if ((sc->sc_flags & TAP_NBIO) &&
871 lockstatus(&sc->sc_rdlock) == LK_EXCLUSIVE)
872 return (EWOULDBLOCK);
873 error = lockmgr(&sc->sc_rdlock, LK_EXCLUSIVE, NULL);
874 if (error != 0)
875 return (error);
876 s = splnet();
877 }
878
879 IFQ_DEQUEUE(&ifp->if_snd, m);
880 ifp->if_flags &= ~IFF_OACTIVE;
881 splx(s);
882 if (m == NULL) {
883 error = 0;
884 goto out;
885 }
886
887 ifp->if_opackets++;
888 #if NBPFILTER > 0
889 if (ifp->if_bpf)
890 bpf_mtap(ifp->if_bpf, m);
891 #endif
892
893 /*
894 * One read is one packet.
895 */
896 do {
897 error = uiomove(mtod(m, caddr_t),
898 min(m->m_len, uio->uio_resid), uio);
899 MFREE(m, n);
900 m = n;
901 } while (m != NULL && uio->uio_resid > 0 && error == 0);
902
903 if (m != NULL)
904 m_freem(m);
905
906 out:
907 (void)lockmgr(&sc->sc_rdlock, LK_RELEASE, NULL);
908 return (error);
909 }
910
911 static int
912 tap_cdev_write(dev_t dev, struct uio *uio, int flags)
913 {
914 return tap_dev_write(minor(dev), uio, flags);
915 }
916
917 static int
918 tap_fops_write(struct file *fp, off_t *offp, struct uio *uio,
919 kauth_cred_t cred, int flags)
920 {
921 return tap_dev_write((intptr_t)fp->f_data, uio, flags);
922 }
923
924 static int
925 tap_dev_write(int unit, struct uio *uio, int flags)
926 {
927 struct tap_softc *sc =
928 (struct tap_softc *)device_lookup(&tap_cd, unit);
929 struct ifnet *ifp;
930 struct mbuf *m, **mp;
931 int error = 0;
932 int s;
933
934 if (sc == NULL)
935 return (ENXIO);
936
937 ifp = &sc->sc_ec.ec_if;
938
939 /* One write, one packet, that's the rule */
940 MGETHDR(m, M_DONTWAIT, MT_DATA);
941 if (m == NULL) {
942 ifp->if_ierrors++;
943 return (ENOBUFS);
944 }
945 m->m_pkthdr.len = uio->uio_resid;
946
947 mp = &m;
948 while (error == 0 && uio->uio_resid > 0) {
949 if (*mp != m) {
950 MGET(*mp, M_DONTWAIT, MT_DATA);
951 if (*mp == NULL) {
952 error = ENOBUFS;
953 break;
954 }
955 }
956 (*mp)->m_len = min(MHLEN, uio->uio_resid);
957 error = uiomove(mtod(*mp, caddr_t), (*mp)->m_len, uio);
958 mp = &(*mp)->m_next;
959 }
960 if (error) {
961 ifp->if_ierrors++;
962 m_freem(m);
963 return (error);
964 }
965
966 ifp->if_ipackets++;
967 m->m_pkthdr.rcvif = ifp;
968
969 #if NBPFILTER > 0
970 if (ifp->if_bpf)
971 bpf_mtap(ifp->if_bpf, m);
972 #endif
973 s =splnet();
974 (*ifp->if_input)(ifp, m);
975 splx(s);
976
977 return (0);
978 }
979
980 static int
981 tap_cdev_ioctl(dev_t dev, u_long cmd, caddr_t data, int flags,
982 struct lwp *l)
983 {
984 return tap_dev_ioctl(minor(dev), cmd, data, l);
985 }
986
987 static int
988 tap_fops_ioctl(struct file *fp, u_long cmd, void *data, struct lwp *l)
989 {
990 return tap_dev_ioctl((intptr_t)fp->f_data, cmd, (caddr_t)data, l);
991 }
992
993 static int
994 tap_dev_ioctl(int unit, u_long cmd, caddr_t data, struct lwp *l)
995 {
996 struct tap_softc *sc =
997 (struct tap_softc *)device_lookup(&tap_cd, unit);
998 int error = 0;
999
1000 if (sc == NULL)
1001 return (ENXIO);
1002
1003 switch (cmd) {
1004 case FIONREAD:
1005 {
1006 struct ifnet *ifp = &sc->sc_ec.ec_if;
1007 struct mbuf *m;
1008 int s;
1009
1010 s = splnet();
1011 IFQ_POLL(&ifp->if_snd, m);
1012
1013 if (m == NULL)
1014 *(int *)data = 0;
1015 else
1016 *(int *)data = m->m_pkthdr.len;
1017 splx(s);
1018 } break;
1019 case TIOCSPGRP:
1020 case FIOSETOWN:
1021 error = fsetown(l->l_proc, &sc->sc_pgid, cmd, data);
1022 break;
1023 case TIOCGPGRP:
1024 case FIOGETOWN:
1025 error = fgetown(l->l_proc, sc->sc_pgid, cmd, data);
1026 break;
1027 case FIOASYNC:
1028 if (*(int *)data)
1029 sc->sc_flags |= TAP_ASYNCIO;
1030 else
1031 sc->sc_flags &= ~TAP_ASYNCIO;
1032 break;
1033 case FIONBIO:
1034 if (*(int *)data)
1035 sc->sc_flags |= TAP_NBIO;
1036 else
1037 sc->sc_flags &= ~TAP_NBIO;
1038 break;
1039 case TAPGIFNAME:
1040 {
1041 struct ifreq *ifr = (struct ifreq *)data;
1042 struct ifnet *ifp = &sc->sc_ec.ec_if;
1043
1044 strlcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1045 } break;
1046 default:
1047 error = ENOTTY;
1048 break;
1049 }
1050
1051 return (0);
1052 }
1053
1054 static int
1055 tap_cdev_poll(dev_t dev, int events, struct lwp *l)
1056 {
1057 return tap_dev_poll(minor(dev), events, l);
1058 }
1059
1060 static int
1061 tap_fops_poll(struct file *fp, int events, struct lwp *l)
1062 {
1063 return tap_dev_poll((intptr_t)fp->f_data, events, l);
1064 }
1065
1066 static int
1067 tap_dev_poll(int unit, int events, struct lwp *l)
1068 {
1069 struct tap_softc *sc =
1070 (struct tap_softc *)device_lookup(&tap_cd, unit);
1071 int revents = 0;
1072
1073 if (sc == NULL)
1074 return (ENXIO);
1075
1076 if (events & (POLLIN|POLLRDNORM)) {
1077 struct ifnet *ifp = &sc->sc_ec.ec_if;
1078 struct mbuf *m;
1079 int s;
1080
1081 s = splnet();
1082 IFQ_POLL(&ifp->if_snd, m);
1083 splx(s);
1084
1085 if (m != NULL)
1086 revents |= events & (POLLIN|POLLRDNORM);
1087 else {
1088 simple_lock(&sc->sc_kqlock);
1089 selrecord(l, &sc->sc_rsel);
1090 simple_unlock(&sc->sc_kqlock);
1091 }
1092 }
1093 revents |= events & (POLLOUT|POLLWRNORM);
1094
1095 return (revents);
1096 }
1097
1098 static struct filterops tap_read_filterops = { 1, NULL, tap_kqdetach,
1099 tap_kqread };
1100 static struct filterops tap_seltrue_filterops = { 1, NULL, tap_kqdetach,
1101 filt_seltrue };
1102
1103 static int
1104 tap_cdev_kqfilter(dev_t dev, struct knote *kn)
1105 {
1106 return tap_dev_kqfilter(minor(dev), kn);
1107 }
1108
1109 static int
1110 tap_fops_kqfilter(struct file *fp, struct knote *kn)
1111 {
1112 return tap_dev_kqfilter((intptr_t)fp->f_data, kn);
1113 }
1114
1115 static int
1116 tap_dev_kqfilter(int unit, struct knote *kn)
1117 {
1118 struct tap_softc *sc =
1119 (struct tap_softc *)device_lookup(&tap_cd, unit);
1120
1121 if (sc == NULL)
1122 return (ENXIO);
1123
1124 switch(kn->kn_filter) {
1125 case EVFILT_READ:
1126 kn->kn_fop = &tap_read_filterops;
1127 break;
1128 case EVFILT_WRITE:
1129 kn->kn_fop = &tap_seltrue_filterops;
1130 break;
1131 default:
1132 return (1);
1133 }
1134
1135 kn->kn_hook = sc;
1136 simple_lock(&sc->sc_kqlock);
1137 SLIST_INSERT_HEAD(&sc->sc_rsel.sel_klist, kn, kn_selnext);
1138 simple_unlock(&sc->sc_kqlock);
1139 return (0);
1140 }
1141
1142 static void
1143 tap_kqdetach(struct knote *kn)
1144 {
1145 struct tap_softc *sc = (struct tap_softc *)kn->kn_hook;
1146
1147 simple_lock(&sc->sc_kqlock);
1148 SLIST_REMOVE(&sc->sc_rsel.sel_klist, kn, knote, kn_selnext);
1149 simple_unlock(&sc->sc_kqlock);
1150 }
1151
1152 static int
1153 tap_kqread(struct knote *kn, long hint)
1154 {
1155 struct tap_softc *sc = (struct tap_softc *)kn->kn_hook;
1156 struct ifnet *ifp = &sc->sc_ec.ec_if;
1157 struct mbuf *m;
1158 int s;
1159
1160 s = splnet();
1161 IFQ_POLL(&ifp->if_snd, m);
1162
1163 if (m == NULL)
1164 kn->kn_data = 0;
1165 else
1166 kn->kn_data = m->m_pkthdr.len;
1167 splx(s);
1168 return (kn->kn_data != 0 ? 1 : 0);
1169 }
1170
1171 /*
1172 * sysctl management routines
1173 * You can set the address of an interface through:
1174 * net.link.tap.tap<number>
1175 *
1176 * Note the consistent use of tap_log in order to use
1177 * sysctl_teardown at unload time.
1178 *
1179 * In the kernel you will find a lot of SYSCTL_SETUP blocks. Those
1180 * blocks register a function in a special section of the kernel
1181 * (called a link set) which is used at init_sysctl() time to cycle
1182 * through all those functions to create the kernel's sysctl tree.
1183 *
1184 * It is not (currently) possible to use link sets in a LKM, so the
1185 * easiest is to simply call our own setup routine at load time.
1186 *
1187 * In the SYSCTL_SETUP blocks you find in the kernel, nodes have the
1188 * CTLFLAG_PERMANENT flag, meaning they cannot be removed. Once the
1189 * whole kernel sysctl tree is built, it is not possible to add any
1190 * permanent node.
1191 *
1192 * It should be noted that we're not saving the sysctlnode pointer
1193 * we are returned when creating the "tap" node. That structure
1194 * cannot be trusted once out of the calling function, as it might
1195 * get reused. So we just save the MIB number, and always give the
1196 * full path starting from the root for later calls to sysctl_createv
1197 * and sysctl_destroyv.
1198 */
1199 SYSCTL_SETUP(sysctl_tap_setup, "sysctl net.link.tap subtree setup")
1200 {
1201 const struct sysctlnode *node;
1202 int error = 0;
1203
1204 if ((error = sysctl_createv(clog, 0, NULL, NULL,
1205 CTLFLAG_PERMANENT,
1206 CTLTYPE_NODE, "net", NULL,
1207 NULL, 0, NULL, 0,
1208 CTL_NET, CTL_EOL)) != 0)
1209 return;
1210
1211 if ((error = sysctl_createv(clog, 0, NULL, NULL,
1212 CTLFLAG_PERMANENT,
1213 CTLTYPE_NODE, "link", NULL,
1214 NULL, 0, NULL, 0,
1215 CTL_NET, AF_LINK, CTL_EOL)) != 0)
1216 return;
1217
1218 /*
1219 * The first four parameters of sysctl_createv are for management.
1220 *
1221 * The four that follows, here starting with a '' for the flags,
1222 * describe the node.
1223 *
1224 * The next series of four set its value, through various possible
1225 * means.
1226 *
1227 * Last but not least, the path to the node is described. That path
1228 * is relative to the given root (third argument). Here we're
1229 * starting from the root.
1230 */
1231 if ((error = sysctl_createv(clog, 0, NULL, &node,
1232 CTLFLAG_PERMANENT,
1233 CTLTYPE_NODE, "tap", NULL,
1234 NULL, 0, NULL, 0,
1235 CTL_NET, AF_LINK, CTL_CREATE, CTL_EOL)) != 0)
1236 return;
1237 tap_node = node->sysctl_num;
1238 }
1239
1240 /*
1241 * The helper functions make Andrew Brown's interface really
1242 * shine. It makes possible to create value on the fly whether
1243 * the sysctl value is read or written.
1244 *
1245 * As shown as an example in the man page, the first step is to
1246 * create a copy of the node to have sysctl_lookup work on it.
1247 *
1248 * Here, we have more work to do than just a copy, since we have
1249 * to create the string. The first step is to collect the actual
1250 * value of the node, which is a convenient pointer to the softc
1251 * of the interface. From there we create the string and use it
1252 * as the value, but only for the *copy* of the node.
1253 *
1254 * Then we let sysctl_lookup do the magic, which consists in
1255 * setting oldp and newp as required by the operation. When the
1256 * value is read, that means that the string will be copied to
1257 * the user, and when it is written, the new value will be copied
1258 * over in the addr array.
1259 *
1260 * If newp is NULL, the user was reading the value, so we don't
1261 * have anything else to do. If a new value was written, we
1262 * have to check it.
1263 *
1264 * If it is incorrect, we can return an error and leave 'node' as
1265 * it is: since it is a copy of the actual node, the change will
1266 * be forgotten.
1267 *
1268 * Upon a correct input, we commit the change to the ifnet
1269 * structure of our interface.
1270 */
1271 static int
1272 tap_sysctl_handler(SYSCTLFN_ARGS)
1273 {
1274 struct sysctlnode node;
1275 struct tap_softc *sc;
1276 struct ifnet *ifp;
1277 int error;
1278 size_t len;
1279 char addr[3 * ETHER_ADDR_LEN];
1280
1281 node = *rnode;
1282 sc = node.sysctl_data;
1283 ifp = &sc->sc_ec.ec_if;
1284 (void)ether_snprintf(addr, sizeof(addr), LLADDR(ifp->if_sadl));
1285 node.sysctl_data = addr;
1286 error = sysctl_lookup(SYSCTLFN_CALL(&node));
1287 if (error || newp == NULL)
1288 return (error);
1289
1290 len = strlen(addr);
1291 if (len < 11 || len > 17)
1292 return (EINVAL);
1293
1294 /* Commit change */
1295 if (ether_nonstatic_aton(LLADDR(ifp->if_sadl), addr) != 0)
1296 return (EINVAL);
1297 return (error);
1298 }
Cache object: c8ee0ac764cf19463a05cc7899149025
|