xref: /freebsd/sys/net/bpf.c (revision d429ea332342fcb98d27a350d0c4944bf9aec3f9)
1 /*-
2  * Copyright (c) 1990, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from the Stanford/CMU enet packet filter,
6  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
7  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
8  * Berkeley Laboratory.
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *      @(#)bpf.c	8.4 (Berkeley) 1/9/95
35  *
36  * $FreeBSD$
37  */
38 
39 #include "opt_bpf.h"
40 #include "opt_mac.h"
41 #include "opt_netgraph.h"
42 
43 #include <sys/types.h>
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/conf.h>
47 #include <sys/fcntl.h>
48 #include <sys/mac.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/time.h>
52 #include <sys/proc.h>
53 #include <sys/signalvar.h>
54 #include <sys/filio.h>
55 #include <sys/sockio.h>
56 #include <sys/ttycom.h>
57 #include <sys/uio.h>
58 
59 #include <sys/event.h>
60 #include <sys/file.h>
61 #include <sys/poll.h>
62 #include <sys/proc.h>
63 
64 #include <sys/socket.h>
65 
66 #include <net/if.h>
67 #include <net/bpf.h>
68 #include <net/bpfdesc.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/if_ether.h>
72 #include <sys/kernel.h>
73 #include <sys/sysctl.h>
74 
75 static MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
76 
77 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
78 
79 #define PRINET  26			/* interruptible */
80 
81 /*
82  * The default read buffer size is patchable.
83  */
84 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
85 static int bpf_bufsize = 4096;
86 SYSCTL_INT(_net_bpf, OID_AUTO, bufsize, CTLFLAG_RW,
87     &bpf_bufsize, 0, "");
88 static int bpf_maxbufsize = BPF_MAXBUFSIZE;
89 SYSCTL_INT(_net_bpf, OID_AUTO, maxbufsize, CTLFLAG_RW,
90     &bpf_maxbufsize, 0, "");
91 static int bpf_maxinsns = BPF_MAXINSNS;
92 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
93     &bpf_maxinsns, 0, "Maximum bpf program instructions");
94 
95 /*
96  * bpf_iflist is a list of BPF interface structures, each corresponding to a
97  * specific DLT.  The same network interface might have several BPF interface
98  * structures registered by different layers in the stack (i.e., 802.11
99  * frames, ethernet frames, etc).
100  */
101 static LIST_HEAD(, bpf_if)	bpf_iflist;
102 static struct mtx	bpf_mtx;		/* bpf global lock */
103 
104 static int	bpf_allocbufs(struct bpf_d *);
105 static void	bpf_attachd(struct bpf_d *d, struct bpf_if *bp);
106 static void	bpf_detachd(struct bpf_d *d);
107 static void	bpf_freed(struct bpf_d *);
108 static void	bpf_mcopy(const void *, void *, size_t);
109 static int	bpf_movein(struct uio *, int,
110 		    struct mbuf **, struct sockaddr *, int *);
111 static int	bpf_setif(struct bpf_d *, struct ifreq *);
112 static void	bpf_timed_out(void *);
113 static __inline void
114 		bpf_wakeup(struct bpf_d *);
115 static void	catchpacket(struct bpf_d *, u_char *, u_int,
116 		    u_int, void (*)(const void *, void *, size_t));
117 static void	reset_d(struct bpf_d *);
118 static int	 bpf_setf(struct bpf_d *, struct bpf_program *);
119 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
120 static int	bpf_setdlt(struct bpf_d *, u_int);
121 static void	filt_bpfdetach(struct knote *);
122 static int	filt_bpfread(struct knote *, long);
123 static void	bpf_drvinit(void *);
124 static void	bpf_clone(void *, char *, int, struct cdev **);
125 
126 static	d_open_t	bpfopen;
127 static	d_close_t	bpfclose;
128 static	d_read_t	bpfread;
129 static	d_write_t	bpfwrite;
130 static	d_ioctl_t	bpfioctl;
131 static	d_poll_t	bpfpoll;
132 static	d_kqfilter_t	bpfkqfilter;
133 
134 static struct cdevsw bpf_cdevsw = {
135 	.d_version =	D_VERSION,
136 	.d_flags =	D_NEEDGIANT,
137 	.d_open =	bpfopen,
138 	.d_close =	bpfclose,
139 	.d_read =	bpfread,
140 	.d_write =	bpfwrite,
141 	.d_ioctl =	bpfioctl,
142 	.d_poll =	bpfpoll,
143 	.d_name =	"bpf",
144 	.d_kqfilter =	bpfkqfilter,
145 };
146 
147 static struct filterops bpfread_filtops =
148 	{ 1, NULL, filt_bpfdetach, filt_bpfread };
149 
150 static int
151 bpf_movein(uio, linktype, mp, sockp, datlen)
152 	struct uio *uio;
153 	int linktype, *datlen;
154 	struct mbuf **mp;
155 	struct sockaddr *sockp;
156 {
157 	struct mbuf *m;
158 	int error;
159 	int len;
160 	int hlen;
161 
162 	/*
163 	 * Build a sockaddr based on the data link layer type.
164 	 * We do this at this level because the ethernet header
165 	 * is copied directly into the data field of the sockaddr.
166 	 * In the case of SLIP, there is no header and the packet
167 	 * is forwarded as is.
168 	 * Also, we are careful to leave room at the front of the mbuf
169 	 * for the link level header.
170 	 */
171 	switch (linktype) {
172 
173 	case DLT_SLIP:
174 		sockp->sa_family = AF_INET;
175 		hlen = 0;
176 		break;
177 
178 	case DLT_EN10MB:
179 		sockp->sa_family = AF_UNSPEC;
180 		/* XXX Would MAXLINKHDR be better? */
181 		hlen = ETHER_HDR_LEN;
182 		break;
183 
184 	case DLT_FDDI:
185 		sockp->sa_family = AF_IMPLINK;
186 		hlen = 0;
187 		break;
188 
189 	case DLT_RAW:
190 	case DLT_NULL:
191 		sockp->sa_family = AF_UNSPEC;
192 		hlen = 0;
193 		break;
194 
195 	case DLT_ATM_RFC1483:
196 		/*
197 		 * en atm driver requires 4-byte atm pseudo header.
198 		 * though it isn't standard, vpi:vci needs to be
199 		 * specified anyway.
200 		 */
201 		sockp->sa_family = AF_UNSPEC;
202 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
203 		break;
204 
205 	case DLT_PPP:
206 		sockp->sa_family = AF_UNSPEC;
207 		hlen = 4;	/* This should match PPP_HDRLEN */
208 		break;
209 
210 	default:
211 		return (EIO);
212 	}
213 
214 	len = uio->uio_resid;
215 	*datlen = len - hlen;
216 	if ((unsigned)len > MCLBYTES)
217 		return (EIO);
218 
219 	if (len > MHLEN) {
220 		m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
221 	} else {
222 		MGETHDR(m, M_TRYWAIT, MT_DATA);
223 	}
224 	if (m == NULL)
225 		return (ENOBUFS);
226 	m->m_pkthdr.len = m->m_len = len;
227 	m->m_pkthdr.rcvif = NULL;
228 	*mp = m;
229 
230 	/*
231 	 * Make room for link header.
232 	 */
233 	if (hlen != 0) {
234 		m->m_pkthdr.len -= hlen;
235 		m->m_len -= hlen;
236 #if BSD >= 199103
237 		m->m_data += hlen; /* XXX */
238 #else
239 		m->m_off += hlen;
240 #endif
241 		error = uiomove(sockp->sa_data, hlen, uio);
242 		if (error)
243 			goto bad;
244 	}
245 	error = uiomove(mtod(m, void *), len - hlen, uio);
246 	if (!error)
247 		return (0);
248 bad:
249 	m_freem(m);
250 	return (error);
251 }
252 
253 /*
254  * Attach file to the bpf interface, i.e. make d listen on bp.
255  */
256 static void
257 bpf_attachd(d, bp)
258 	struct bpf_d *d;
259 	struct bpf_if *bp;
260 {
261 	/*
262 	 * Point d at bp, and add d to the interface's list of listeners.
263 	 * Finally, point the driver's bpf cookie at the interface so
264 	 * it will divert packets to bpf.
265 	 */
266 	BPFIF_LOCK(bp);
267 	d->bd_bif = bp;
268 	LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
269 
270 	*bp->bif_driverp = bp;
271 	BPFIF_UNLOCK(bp);
272 }
273 
274 /*
275  * Detach a file from its interface.
276  */
277 static void
278 bpf_detachd(d)
279 	struct bpf_d *d;
280 {
281 	int error;
282 	struct bpf_if *bp;
283 	struct ifnet *ifp;
284 
285 	bp = d->bd_bif;
286 	BPFIF_LOCK(bp);
287 	BPFD_LOCK(d);
288 	ifp = d->bd_bif->bif_ifp;
289 
290 	/*
291 	 * Remove d from the interface's descriptor list.
292 	 */
293 	LIST_REMOVE(d, bd_next);
294 
295 	/*
296 	 * Let the driver know that there are no more listeners.
297 	 */
298 	if (LIST_EMPTY(&bp->bif_dlist))
299 		*bp->bif_driverp = NULL;
300 
301 	d->bd_bif = NULL;
302 	BPFD_UNLOCK(d);
303 	BPFIF_UNLOCK(bp);
304 
305 	/*
306 	 * Check if this descriptor had requested promiscuous mode.
307 	 * If so, turn it off.
308 	 */
309 	if (d->bd_promisc) {
310 		d->bd_promisc = 0;
311 		error = ifpromisc(ifp, 0);
312 		if (error != 0 && error != ENXIO) {
313 			/*
314 			 * ENXIO can happen if a pccard is unplugged
315 			 * Something is really wrong if we were able to put
316 			 * the driver into promiscuous mode, but can't
317 			 * take it out.
318 			 */
319 			if_printf(bp->bif_ifp,
320 				"bpf_detach: ifpromisc failed (%d)\n", error);
321 		}
322 	}
323 }
324 
325 /*
326  * Open ethernet device.  Returns ENXIO for illegal minor device number,
327  * EBUSY if file is open by another process.
328  */
329 /* ARGSUSED */
330 static	int
331 bpfopen(dev, flags, fmt, td)
332 	struct cdev *dev;
333 	int flags;
334 	int fmt;
335 	struct thread *td;
336 {
337 	struct bpf_d *d;
338 
339 	mtx_lock(&bpf_mtx);
340 	d = dev->si_drv1;
341 	/*
342 	 * Each minor can be opened by only one process.  If the requested
343 	 * minor is in use, return EBUSY.
344 	 */
345 	if (d != NULL) {
346 		mtx_unlock(&bpf_mtx);
347 		return (EBUSY);
348 	}
349 	dev->si_drv1 = (struct bpf_d *)~0;	/* mark device in use */
350 	mtx_unlock(&bpf_mtx);
351 
352 	if ((dev->si_flags & SI_NAMED) == 0)
353 		make_dev(&bpf_cdevsw, minor(dev), UID_ROOT, GID_WHEEL, 0600,
354 		    "bpf%d", dev2unit(dev));
355 	MALLOC(d, struct bpf_d *, sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
356 	dev->si_drv1 = d;
357 	d->bd_bufsize = bpf_bufsize;
358 	d->bd_sig = SIGIO;
359 	d->bd_seesent = 1;
360 #ifdef MAC
361 	mac_init_bpfdesc(d);
362 	mac_create_bpfdesc(td->td_ucred, d);
363 #endif
364 	mtx_init(&d->bd_mtx, devtoname(dev), "bpf cdev lock", MTX_DEF);
365 	callout_init(&d->bd_callout, NET_CALLOUT_MPSAFE);
366 	knlist_init(&d->bd_sel.si_note, &d->bd_mtx);
367 
368 	return (0);
369 }
370 
371 /*
372  * Close the descriptor by detaching it from its interface,
373  * deallocating its buffers, and marking it free.
374  */
375 /* ARGSUSED */
376 static	int
377 bpfclose(dev, flags, fmt, td)
378 	struct cdev *dev;
379 	int flags;
380 	int fmt;
381 	struct thread *td;
382 {
383 	struct bpf_d *d = dev->si_drv1;
384 
385 	BPFD_LOCK(d);
386 	if (d->bd_state == BPF_WAITING)
387 		callout_stop(&d->bd_callout);
388 	d->bd_state = BPF_IDLE;
389 	BPFD_UNLOCK(d);
390 	funsetown(&d->bd_sigio);
391 	mtx_lock(&bpf_mtx);
392 	if (d->bd_bif)
393 		bpf_detachd(d);
394 	mtx_unlock(&bpf_mtx);
395 	selwakeuppri(&d->bd_sel, PRINET);
396 #ifdef MAC
397 	mac_destroy_bpfdesc(d);
398 #endif /* MAC */
399 	knlist_destroy(&d->bd_sel.si_note);
400 	bpf_freed(d);
401 	dev->si_drv1 = NULL;
402 	free(d, M_BPF);
403 
404 	return (0);
405 }
406 
407 
408 /*
409  * Rotate the packet buffers in descriptor d.  Move the store buffer
410  * into the hold slot, and the free buffer into the store slot.
411  * Zero the length of the new store buffer.
412  */
413 #define ROTATE_BUFFERS(d) \
414 	(d)->bd_hbuf = (d)->bd_sbuf; \
415 	(d)->bd_hlen = (d)->bd_slen; \
416 	(d)->bd_sbuf = (d)->bd_fbuf; \
417 	(d)->bd_slen = 0; \
418 	(d)->bd_fbuf = NULL;
419 /*
420  *  bpfread - read next chunk of packets from buffers
421  */
422 static	int
423 bpfread(dev, uio, ioflag)
424 	struct cdev *dev;
425 	struct uio *uio;
426 	int ioflag;
427 {
428 	struct bpf_d *d = dev->si_drv1;
429 	int timed_out;
430 	int error;
431 
432 	/*
433 	 * Restrict application to use a buffer the same size as
434 	 * as kernel buffers.
435 	 */
436 	if (uio->uio_resid != d->bd_bufsize)
437 		return (EINVAL);
438 
439 	BPFD_LOCK(d);
440 	if (d->bd_state == BPF_WAITING)
441 		callout_stop(&d->bd_callout);
442 	timed_out = (d->bd_state == BPF_TIMED_OUT);
443 	d->bd_state = BPF_IDLE;
444 	/*
445 	 * If the hold buffer is empty, then do a timed sleep, which
446 	 * ends when the timeout expires or when enough packets
447 	 * have arrived to fill the store buffer.
448 	 */
449 	while (d->bd_hbuf == NULL) {
450 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
451 			/*
452 			 * A packet(s) either arrived since the previous
453 			 * read or arrived while we were asleep.
454 			 * Rotate the buffers and return what's here.
455 			 */
456 			ROTATE_BUFFERS(d);
457 			break;
458 		}
459 
460 		/*
461 		 * No data is available, check to see if the bpf device
462 		 * is still pointed at a real interface.  If not, return
463 		 * ENXIO so that the userland process knows to rebind
464 		 * it before using it again.
465 		 */
466 		if (d->bd_bif == NULL) {
467 			BPFD_UNLOCK(d);
468 			return (ENXIO);
469 		}
470 
471 		if (ioflag & O_NONBLOCK) {
472 			BPFD_UNLOCK(d);
473 			return (EWOULDBLOCK);
474 		}
475 		error = msleep(d, &d->bd_mtx, PRINET|PCATCH,
476 		     "bpf", d->bd_rtout);
477 		if (error == EINTR || error == ERESTART) {
478 			BPFD_UNLOCK(d);
479 			return (error);
480 		}
481 		if (error == EWOULDBLOCK) {
482 			/*
483 			 * On a timeout, return what's in the buffer,
484 			 * which may be nothing.  If there is something
485 			 * in the store buffer, we can rotate the buffers.
486 			 */
487 			if (d->bd_hbuf)
488 				/*
489 				 * We filled up the buffer in between
490 				 * getting the timeout and arriving
491 				 * here, so we don't need to rotate.
492 				 */
493 				break;
494 
495 			if (d->bd_slen == 0) {
496 				BPFD_UNLOCK(d);
497 				return (0);
498 			}
499 			ROTATE_BUFFERS(d);
500 			break;
501 		}
502 	}
503 	/*
504 	 * At this point, we know we have something in the hold slot.
505 	 */
506 	BPFD_UNLOCK(d);
507 
508 	/*
509 	 * Move data from hold buffer into user space.
510 	 * We know the entire buffer is transferred since
511 	 * we checked above that the read buffer is bpf_bufsize bytes.
512 	 */
513 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
514 
515 	BPFD_LOCK(d);
516 	d->bd_fbuf = d->bd_hbuf;
517 	d->bd_hbuf = NULL;
518 	d->bd_hlen = 0;
519 	BPFD_UNLOCK(d);
520 
521 	return (error);
522 }
523 
524 
525 /*
526  * If there are processes sleeping on this descriptor, wake them up.
527  */
528 static __inline void
529 bpf_wakeup(d)
530 	struct bpf_d *d;
531 {
532 
533 	BPFD_LOCK_ASSERT(d);
534 	if (d->bd_state == BPF_WAITING) {
535 		callout_stop(&d->bd_callout);
536 		d->bd_state = BPF_IDLE;
537 	}
538 	wakeup(d);
539 	if (d->bd_async && d->bd_sig && d->bd_sigio)
540 		pgsigio(&d->bd_sigio, d->bd_sig, 0);
541 
542 	selwakeuppri(&d->bd_sel, PRINET);
543 	KNOTE_LOCKED(&d->bd_sel.si_note, 0);
544 }
545 
546 static void
547 bpf_timed_out(arg)
548 	void *arg;
549 {
550 	struct bpf_d *d = (struct bpf_d *)arg;
551 
552 	BPFD_LOCK(d);
553 	if (d->bd_state == BPF_WAITING) {
554 		d->bd_state = BPF_TIMED_OUT;
555 		if (d->bd_slen != 0)
556 			bpf_wakeup(d);
557 	}
558 	BPFD_UNLOCK(d);
559 }
560 
561 static	int
562 bpfwrite(dev, uio, ioflag)
563 	struct cdev *dev;
564 	struct uio *uio;
565 	int ioflag;
566 {
567 	struct bpf_d *d = dev->si_drv1;
568 	struct ifnet *ifp;
569 	struct mbuf *m;
570 	int error;
571 	struct sockaddr dst;
572 	int datlen;
573 
574 	if (d->bd_bif == NULL)
575 		return (ENXIO);
576 
577 	ifp = d->bd_bif->bif_ifp;
578 
579 	if ((ifp->if_flags & IFF_UP) == 0)
580 		return (ENETDOWN);
581 
582 	if (uio->uio_resid == 0)
583 		return (0);
584 
585 	bzero(&dst, sizeof(dst));
586 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, &dst, &datlen);
587 	if (error)
588 		return (error);
589 
590 	if (datlen > ifp->if_mtu) {
591 		m_freem(m);
592 		return (EMSGSIZE);
593 	}
594 
595 	if (d->bd_hdrcmplt)
596 		dst.sa_family = pseudo_AF_HDRCMPLT;
597 
598 #ifdef MAC
599 	BPFD_LOCK(d);
600 	mac_create_mbuf_from_bpfdesc(d, m);
601 	BPFD_UNLOCK(d);
602 #endif
603 	NET_LOCK_GIANT();
604 	error = (*ifp->if_output)(ifp, m, &dst, NULL);
605 	NET_UNLOCK_GIANT();
606 	/*
607 	 * The driver frees the mbuf.
608 	 */
609 	return (error);
610 }
611 
612 /*
613  * Reset a descriptor by flushing its packet buffer and clearing the
614  * receive and drop counts.
615  */
616 static void
617 reset_d(d)
618 	struct bpf_d *d;
619 {
620 
621 	mtx_assert(&d->bd_mtx, MA_OWNED);
622 	if (d->bd_hbuf) {
623 		/* Free the hold buffer. */
624 		d->bd_fbuf = d->bd_hbuf;
625 		d->bd_hbuf = NULL;
626 	}
627 	d->bd_slen = 0;
628 	d->bd_hlen = 0;
629 	d->bd_rcount = 0;
630 	d->bd_dcount = 0;
631 }
632 
633 /*
634  *  FIONREAD		Check for read packet available.
635  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
636  *  BIOCGBLEN		Get buffer len [for read()].
637  *  BIOCSETF		Set ethernet read filter.
638  *  BIOCFLUSH		Flush read packet buffer.
639  *  BIOCPROMISC		Put interface into promiscuous mode.
640  *  BIOCGDLT		Get link layer type.
641  *  BIOCGETIF		Get interface name.
642  *  BIOCSETIF		Set interface.
643  *  BIOCSRTIMEOUT	Set read timeout.
644  *  BIOCGRTIMEOUT	Get read timeout.
645  *  BIOCGSTATS		Get packet stats.
646  *  BIOCIMMEDIATE	Set immediate mode.
647  *  BIOCVERSION		Get filter language version.
648  *  BIOCGHDRCMPLT	Get "header already complete" flag
649  *  BIOCSHDRCMPLT	Set "header already complete" flag
650  *  BIOCGSEESENT	Get "see packets sent" flag
651  *  BIOCSSEESENT	Set "see packets sent" flag
652  */
653 /* ARGSUSED */
654 static	int
655 bpfioctl(dev, cmd, addr, flags, td)
656 	struct cdev *dev;
657 	u_long cmd;
658 	caddr_t addr;
659 	int flags;
660 	struct thread *td;
661 {
662 	struct bpf_d *d = dev->si_drv1;
663 	int error = 0;
664 
665 	BPFD_LOCK(d);
666 	if (d->bd_state == BPF_WAITING)
667 		callout_stop(&d->bd_callout);
668 	d->bd_state = BPF_IDLE;
669 	BPFD_UNLOCK(d);
670 
671 	switch (cmd) {
672 
673 	default:
674 		error = EINVAL;
675 		break;
676 
677 	/*
678 	 * Check for read packet available.
679 	 */
680 	case FIONREAD:
681 		{
682 			int n;
683 
684 			BPFD_LOCK(d);
685 			n = d->bd_slen;
686 			if (d->bd_hbuf)
687 				n += d->bd_hlen;
688 			BPFD_UNLOCK(d);
689 
690 			*(int *)addr = n;
691 			break;
692 		}
693 
694 	case SIOCGIFADDR:
695 		{
696 			struct ifnet *ifp;
697 
698 			if (d->bd_bif == NULL)
699 				error = EINVAL;
700 			else {
701 				ifp = d->bd_bif->bif_ifp;
702 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
703 			}
704 			break;
705 		}
706 
707 	/*
708 	 * Get buffer len [for read()].
709 	 */
710 	case BIOCGBLEN:
711 		*(u_int *)addr = d->bd_bufsize;
712 		break;
713 
714 	/*
715 	 * Set buffer length.
716 	 */
717 	case BIOCSBLEN:
718 		if (d->bd_bif != NULL)
719 			error = EINVAL;
720 		else {
721 			u_int size = *(u_int *)addr;
722 
723 			if (size > bpf_maxbufsize)
724 				*(u_int *)addr = size = bpf_maxbufsize;
725 			else if (size < BPF_MINBUFSIZE)
726 				*(u_int *)addr = size = BPF_MINBUFSIZE;
727 			d->bd_bufsize = size;
728 		}
729 		break;
730 
731 	/*
732 	 * Set link layer read filter.
733 	 */
734 	case BIOCSETF:
735 		error = bpf_setf(d, (struct bpf_program *)addr);
736 		break;
737 
738 	/*
739 	 * Flush read packet buffer.
740 	 */
741 	case BIOCFLUSH:
742 		BPFD_LOCK(d);
743 		reset_d(d);
744 		BPFD_UNLOCK(d);
745 		break;
746 
747 	/*
748 	 * Put interface into promiscuous mode.
749 	 */
750 	case BIOCPROMISC:
751 		if (d->bd_bif == NULL) {
752 			/*
753 			 * No interface attached yet.
754 			 */
755 			error = EINVAL;
756 			break;
757 		}
758 		if (d->bd_promisc == 0) {
759 			mtx_lock(&Giant);
760 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
761 			mtx_unlock(&Giant);
762 			if (error == 0)
763 				d->bd_promisc = 1;
764 		}
765 		break;
766 
767 	/*
768 	 * Get current data link type.
769 	 */
770 	case BIOCGDLT:
771 		if (d->bd_bif == NULL)
772 			error = EINVAL;
773 		else
774 			*(u_int *)addr = d->bd_bif->bif_dlt;
775 		break;
776 
777 	/*
778 	 * Get a list of supported data link types.
779 	 */
780 	case BIOCGDLTLIST:
781 		if (d->bd_bif == NULL)
782 			error = EINVAL;
783 		else
784 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
785 		break;
786 
787 	/*
788 	 * Set data link type.
789 	 */
790 	case BIOCSDLT:
791 		if (d->bd_bif == NULL)
792 			error = EINVAL;
793 		else
794 			error = bpf_setdlt(d, *(u_int *)addr);
795 		break;
796 
797 	/*
798 	 * Get interface name.
799 	 */
800 	case BIOCGETIF:
801 		if (d->bd_bif == NULL)
802 			error = EINVAL;
803 		else {
804 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
805 			struct ifreq *const ifr = (struct ifreq *)addr;
806 
807 			strlcpy(ifr->ifr_name, ifp->if_xname,
808 			    sizeof(ifr->ifr_name));
809 		}
810 		break;
811 
812 	/*
813 	 * Set interface.
814 	 */
815 	case BIOCSETIF:
816 		error = bpf_setif(d, (struct ifreq *)addr);
817 		break;
818 
819 	/*
820 	 * Set read timeout.
821 	 */
822 	case BIOCSRTIMEOUT:
823 		{
824 			struct timeval *tv = (struct timeval *)addr;
825 
826 			/*
827 			 * Subtract 1 tick from tvtohz() since this isn't
828 			 * a one-shot timer.
829 			 */
830 			if ((error = itimerfix(tv)) == 0)
831 				d->bd_rtout = tvtohz(tv) - 1;
832 			break;
833 		}
834 
835 	/*
836 	 * Get read timeout.
837 	 */
838 	case BIOCGRTIMEOUT:
839 		{
840 			struct timeval *tv = (struct timeval *)addr;
841 
842 			tv->tv_sec = d->bd_rtout / hz;
843 			tv->tv_usec = (d->bd_rtout % hz) * tick;
844 			break;
845 		}
846 
847 	/*
848 	 * Get packet stats.
849 	 */
850 	case BIOCGSTATS:
851 		{
852 			struct bpf_stat *bs = (struct bpf_stat *)addr;
853 
854 			bs->bs_recv = d->bd_rcount;
855 			bs->bs_drop = d->bd_dcount;
856 			break;
857 		}
858 
859 	/*
860 	 * Set immediate mode.
861 	 */
862 	case BIOCIMMEDIATE:
863 		d->bd_immediate = *(u_int *)addr;
864 		break;
865 
866 	case BIOCVERSION:
867 		{
868 			struct bpf_version *bv = (struct bpf_version *)addr;
869 
870 			bv->bv_major = BPF_MAJOR_VERSION;
871 			bv->bv_minor = BPF_MINOR_VERSION;
872 			break;
873 		}
874 
875 	/*
876 	 * Get "header already complete" flag
877 	 */
878 	case BIOCGHDRCMPLT:
879 		*(u_int *)addr = d->bd_hdrcmplt;
880 		break;
881 
882 	/*
883 	 * Set "header already complete" flag
884 	 */
885 	case BIOCSHDRCMPLT:
886 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
887 		break;
888 
889 	/*
890 	 * Get "see sent packets" flag
891 	 */
892 	case BIOCGSEESENT:
893 		*(u_int *)addr = d->bd_seesent;
894 		break;
895 
896 	/*
897 	 * Set "see sent packets" flag
898 	 */
899 	case BIOCSSEESENT:
900 		d->bd_seesent = *(u_int *)addr;
901 		break;
902 
903 	case FIONBIO:		/* Non-blocking I/O */
904 		break;
905 
906 	case FIOASYNC:		/* Send signal on receive packets */
907 		d->bd_async = *(int *)addr;
908 		break;
909 
910 	case FIOSETOWN:
911 		error = fsetown(*(int *)addr, &d->bd_sigio);
912 		break;
913 
914 	case FIOGETOWN:
915 		*(int *)addr = fgetown(&d->bd_sigio);
916 		break;
917 
918 	/* This is deprecated, FIOSETOWN should be used instead. */
919 	case TIOCSPGRP:
920 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
921 		break;
922 
923 	/* This is deprecated, FIOGETOWN should be used instead. */
924 	case TIOCGPGRP:
925 		*(int *)addr = -fgetown(&d->bd_sigio);
926 		break;
927 
928 	case BIOCSRSIG:		/* Set receive signal */
929 		{
930 			u_int sig;
931 
932 			sig = *(u_int *)addr;
933 
934 			if (sig >= NSIG)
935 				error = EINVAL;
936 			else
937 				d->bd_sig = sig;
938 			break;
939 		}
940 	case BIOCGRSIG:
941 		*(u_int *)addr = d->bd_sig;
942 		break;
943 	}
944 	return (error);
945 }
946 
947 /*
948  * Set d's packet filter program to fp.  If this file already has a filter,
949  * free it and replace it.  Returns EINVAL for bogus requests.
950  */
951 static int
952 bpf_setf(d, fp)
953 	struct bpf_d *d;
954 	struct bpf_program *fp;
955 {
956 	struct bpf_insn *fcode, *old;
957 	u_int flen, size;
958 
959 	if (fp->bf_insns == NULL) {
960 		if (fp->bf_len != 0)
961 			return (EINVAL);
962 		BPFD_LOCK(d);
963 		old = d->bd_filter;
964 		d->bd_filter = NULL;
965 		reset_d(d);
966 		BPFD_UNLOCK(d);
967 		if (old != NULL)
968 			free((caddr_t)old, M_BPF);
969 		return (0);
970 	}
971 	flen = fp->bf_len;
972 	if (flen > bpf_maxinsns)
973 		return (EINVAL);
974 
975 	size = flen * sizeof(*fp->bf_insns);
976 	fcode = (struct bpf_insn *)malloc(size, M_BPF, M_WAITOK);
977 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
978 	    bpf_validate(fcode, (int)flen)) {
979 		BPFD_LOCK(d);
980 		old = d->bd_filter;
981 		d->bd_filter = fcode;
982 		reset_d(d);
983 		BPFD_UNLOCK(d);
984 		if (old != NULL)
985 			free((caddr_t)old, M_BPF);
986 
987 		return (0);
988 	}
989 	free((caddr_t)fcode, M_BPF);
990 	return (EINVAL);
991 }
992 
993 /*
994  * Detach a file from its current interface (if attached at all) and attach
995  * to the interface indicated by the name stored in ifr.
996  * Return an errno or 0.
997  */
998 static int
999 bpf_setif(d, ifr)
1000 	struct bpf_d *d;
1001 	struct ifreq *ifr;
1002 {
1003 	struct bpf_if *bp;
1004 	int error;
1005 	struct ifnet *theywant;
1006 
1007 	theywant = ifunit(ifr->ifr_name);
1008 	if (theywant == NULL)
1009 		return ENXIO;
1010 
1011 	/*
1012 	 * Look through attached interfaces for the named one.
1013 	 */
1014 	mtx_lock(&bpf_mtx);
1015 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
1016 		struct ifnet *ifp = bp->bif_ifp;
1017 
1018 		if (ifp == NULL || ifp != theywant)
1019 			continue;
1020 		/* skip additional entry */
1021 		if (bp->bif_driverp != &ifp->if_bpf)
1022 			continue;
1023 
1024 		mtx_unlock(&bpf_mtx);
1025 		/*
1026 		 * We found the requested interface.
1027 		 * Allocate the packet buffers if we need to.
1028 		 * If we're already attached to requested interface,
1029 		 * just flush the buffer.
1030 		 */
1031 		if (d->bd_sbuf == NULL) {
1032 			error = bpf_allocbufs(d);
1033 			if (error != 0)
1034 				return (error);
1035 		}
1036 		if (bp != d->bd_bif) {
1037 			if (d->bd_bif)
1038 				/*
1039 				 * Detach if attached to something else.
1040 				 */
1041 				bpf_detachd(d);
1042 
1043 			bpf_attachd(d, bp);
1044 		}
1045 		BPFD_LOCK(d);
1046 		reset_d(d);
1047 		BPFD_UNLOCK(d);
1048 		return (0);
1049 	}
1050 	mtx_unlock(&bpf_mtx);
1051 	/* Not found. */
1052 	return (ENXIO);
1053 }
1054 
1055 /*
1056  * Support for select() and poll() system calls
1057  *
1058  * Return true iff the specific operation will not block indefinitely.
1059  * Otherwise, return false but make a note that a selwakeup() must be done.
1060  */
1061 static int
1062 bpfpoll(dev, events, td)
1063 	struct cdev *dev;
1064 	int events;
1065 	struct thread *td;
1066 {
1067 	struct bpf_d *d;
1068 	int revents;
1069 
1070 	d = dev->si_drv1;
1071 	if (d->bd_bif == NULL)
1072 		return (ENXIO);
1073 
1074 	revents = events & (POLLOUT | POLLWRNORM);
1075 	BPFD_LOCK(d);
1076 	if (events & (POLLIN | POLLRDNORM)) {
1077 		if (bpf_ready(d))
1078 			revents |= events & (POLLIN | POLLRDNORM);
1079 		else {
1080 			selrecord(td, &d->bd_sel);
1081 			/* Start the read timeout if necessary. */
1082 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1083 				callout_reset(&d->bd_callout, d->bd_rtout,
1084 				    bpf_timed_out, d);
1085 				d->bd_state = BPF_WAITING;
1086 			}
1087 		}
1088 	}
1089 	BPFD_UNLOCK(d);
1090 	return (revents);
1091 }
1092 
1093 /*
1094  * Support for kevent() system call.  Register EVFILT_READ filters and
1095  * reject all others.
1096  */
1097 int
1098 bpfkqfilter(dev, kn)
1099 	struct cdev *dev;
1100 	struct knote *kn;
1101 {
1102 	struct bpf_d *d = (struct bpf_d *)dev->si_drv1;
1103 
1104 	if (kn->kn_filter != EVFILT_READ)
1105 		return (1);
1106 
1107 	kn->kn_fop = &bpfread_filtops;
1108 	kn->kn_hook = d;
1109 	knlist_add(&d->bd_sel.si_note, kn, 0);
1110 
1111 	return (0);
1112 }
1113 
1114 static void
1115 filt_bpfdetach(kn)
1116 	struct knote *kn;
1117 {
1118 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1119 
1120 	knlist_remove(&d->bd_sel.si_note, kn, 0);
1121 }
1122 
1123 static int
1124 filt_bpfread(kn, hint)
1125 	struct knote *kn;
1126 	long hint;
1127 {
1128 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1129 	int ready;
1130 
1131 	BPFD_LOCK_ASSERT(d);
1132 	ready = bpf_ready(d);
1133 	if (ready) {
1134 		kn->kn_data = d->bd_slen;
1135 		if (d->bd_hbuf)
1136 			kn->kn_data += d->bd_hlen;
1137 	}
1138 	else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1139 		callout_reset(&d->bd_callout, d->bd_rtout,
1140 		    bpf_timed_out, d);
1141 		d->bd_state = BPF_WAITING;
1142 	}
1143 
1144 	return (ready);
1145 }
1146 
1147 /*
1148  * Incoming linkage from device drivers.  Process the packet pkt, of length
1149  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1150  * by each process' filter, and if accepted, stashed into the corresponding
1151  * buffer.
1152  */
1153 void
1154 bpf_tap(bp, pkt, pktlen)
1155 	struct bpf_if *bp;
1156 	u_char *pkt;
1157 	u_int pktlen;
1158 {
1159 	struct bpf_d *d;
1160 	u_int slen;
1161 
1162 	/*
1163 	 * Lockless read to avoid cost of locking the interface if there are
1164 	 * no descriptors attached.
1165 	 */
1166 	if (LIST_EMPTY(&bp->bif_dlist))
1167 		return;
1168 
1169 	BPFIF_LOCK(bp);
1170 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1171 		BPFD_LOCK(d);
1172 		++d->bd_rcount;
1173 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
1174 		if (slen != 0) {
1175 #ifdef MAC
1176 			if (mac_check_bpfdesc_receive(d, bp->bif_ifp) == 0)
1177 #endif
1178 				catchpacket(d, pkt, pktlen, slen, bcopy);
1179 		}
1180 		BPFD_UNLOCK(d);
1181 	}
1182 	BPFIF_UNLOCK(bp);
1183 }
1184 
1185 /*
1186  * Copy data from an mbuf chain into a buffer.  This code is derived
1187  * from m_copydata in sys/uipc_mbuf.c.
1188  */
1189 static void
1190 bpf_mcopy(src_arg, dst_arg, len)
1191 	const void *src_arg;
1192 	void *dst_arg;
1193 	size_t len;
1194 {
1195 	const struct mbuf *m;
1196 	u_int count;
1197 	u_char *dst;
1198 
1199 	m = src_arg;
1200 	dst = dst_arg;
1201 	while (len > 0) {
1202 		if (m == NULL)
1203 			panic("bpf_mcopy");
1204 		count = min(m->m_len, len);
1205 		bcopy(mtod(m, void *), dst, count);
1206 		m = m->m_next;
1207 		dst += count;
1208 		len -= count;
1209 	}
1210 }
1211 
1212 /*
1213  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1214  */
1215 void
1216 bpf_mtap(bp, m)
1217 	struct bpf_if *bp;
1218 	struct mbuf *m;
1219 {
1220 	struct bpf_d *d;
1221 	u_int pktlen, slen;
1222 
1223 	/*
1224 	 * Lockless read to avoid cost of locking the interface if there are
1225 	 * no descriptors attached.
1226 	 */
1227 	if (LIST_EMPTY(&bp->bif_dlist))
1228 		return;
1229 
1230 	pktlen = m_length(m, NULL);
1231 	if (pktlen == m->m_len) {
1232 		bpf_tap(bp, mtod(m, u_char *), pktlen);
1233 		return;
1234 	}
1235 
1236 	BPFIF_LOCK(bp);
1237 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1238 		if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL))
1239 			continue;
1240 		BPFD_LOCK(d);
1241 		++d->bd_rcount;
1242 		slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
1243 		if (slen != 0)
1244 #ifdef MAC
1245 			if (mac_check_bpfdesc_receive(d, bp->bif_ifp) == 0)
1246 #endif
1247 				catchpacket(d, (u_char *)m, pktlen, slen,
1248 				    bpf_mcopy);
1249 		BPFD_UNLOCK(d);
1250 	}
1251 	BPFIF_UNLOCK(bp);
1252 }
1253 
1254 /*
1255  * Incoming linkage from device drivers, when packet is in
1256  * an mbuf chain and to be prepended by a contiguous header.
1257  */
1258 void
1259 bpf_mtap2(bp, data, dlen, m)
1260 	struct bpf_if *bp;
1261 	void *data;
1262 	u_int dlen;
1263 	struct mbuf *m;
1264 {
1265 	struct mbuf mb;
1266 	struct bpf_d *d;
1267 	u_int pktlen, slen;
1268 
1269 	/*
1270 	 * Lockless read to avoid cost of locking the interface if there are
1271 	 * no descriptors attached.
1272 	 */
1273 	if (LIST_EMPTY(&bp->bif_dlist))
1274 		return;
1275 
1276 	pktlen = m_length(m, NULL);
1277 	/*
1278 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1279 	 * Note that we cut corners here; we only setup what's
1280 	 * absolutely needed--this mbuf should never go anywhere else.
1281 	 */
1282 	mb.m_next = m;
1283 	mb.m_data = data;
1284 	mb.m_len = dlen;
1285 	pktlen += dlen;
1286 
1287 	BPFIF_LOCK(bp);
1288 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1289 		if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL))
1290 			continue;
1291 		BPFD_LOCK(d);
1292 		++d->bd_rcount;
1293 		slen = bpf_filter(d->bd_filter, (u_char *)&mb, pktlen, 0);
1294 		if (slen != 0)
1295 #ifdef MAC
1296 			if (mac_check_bpfdesc_receive(d, bp->bif_ifp) == 0)
1297 #endif
1298 				catchpacket(d, (u_char *)&mb, pktlen, slen,
1299 				    bpf_mcopy);
1300 		BPFD_UNLOCK(d);
1301 	}
1302 	BPFIF_UNLOCK(bp);
1303 }
1304 
1305 /*
1306  * Move the packet data from interface memory (pkt) into the
1307  * store buffer.  "cpfn" is the routine called to do the actual data
1308  * transfer.  bcopy is passed in to copy contiguous chunks, while
1309  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1310  * pkt is really an mbuf.
1311  */
1312 static void
1313 catchpacket(d, pkt, pktlen, snaplen, cpfn)
1314 	struct bpf_d *d;
1315 	u_char *pkt;
1316 	u_int pktlen, snaplen;
1317 	void (*cpfn)(const void *, void *, size_t);
1318 {
1319 	struct bpf_hdr *hp;
1320 	int totlen, curlen;
1321 	int hdrlen = d->bd_bif->bif_hdrlen;
1322 	int do_wakeup = 0;
1323 
1324 	BPFD_LOCK_ASSERT(d);
1325 	/*
1326 	 * Figure out how many bytes to move.  If the packet is
1327 	 * greater or equal to the snapshot length, transfer that
1328 	 * much.  Otherwise, transfer the whole packet (unless
1329 	 * we hit the buffer size limit).
1330 	 */
1331 	totlen = hdrlen + min(snaplen, pktlen);
1332 	if (totlen > d->bd_bufsize)
1333 		totlen = d->bd_bufsize;
1334 
1335 	/*
1336 	 * Round up the end of the previous packet to the next longword.
1337 	 */
1338 	curlen = BPF_WORDALIGN(d->bd_slen);
1339 	if (curlen + totlen > d->bd_bufsize) {
1340 		/*
1341 		 * This packet will overflow the storage buffer.
1342 		 * Rotate the buffers if we can, then wakeup any
1343 		 * pending reads.
1344 		 */
1345 		if (d->bd_fbuf == NULL) {
1346 			/*
1347 			 * We haven't completed the previous read yet,
1348 			 * so drop the packet.
1349 			 */
1350 			++d->bd_dcount;
1351 			return;
1352 		}
1353 		ROTATE_BUFFERS(d);
1354 		do_wakeup = 1;
1355 		curlen = 0;
1356 	}
1357 	else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
1358 		/*
1359 		 * Immediate mode is set, or the read timeout has
1360 		 * already expired during a select call.  A packet
1361 		 * arrived, so the reader should be woken up.
1362 		 */
1363 		do_wakeup = 1;
1364 
1365 	/*
1366 	 * Append the bpf header.
1367 	 */
1368 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1369 	microtime(&hp->bh_tstamp);
1370 	hp->bh_datalen = pktlen;
1371 	hp->bh_hdrlen = hdrlen;
1372 	/*
1373 	 * Copy the packet data into the store buffer and update its length.
1374 	 */
1375 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1376 	d->bd_slen = curlen + totlen;
1377 
1378 	if (do_wakeup)
1379 		bpf_wakeup(d);
1380 }
1381 
1382 /*
1383  * Initialize all nonzero fields of a descriptor.
1384  */
1385 static int
1386 bpf_allocbufs(d)
1387 	struct bpf_d *d;
1388 {
1389 	d->bd_fbuf = (caddr_t)malloc(d->bd_bufsize, M_BPF, M_WAITOK);
1390 	if (d->bd_fbuf == NULL)
1391 		return (ENOBUFS);
1392 
1393 	d->bd_sbuf = (caddr_t)malloc(d->bd_bufsize, M_BPF, M_WAITOK);
1394 	if (d->bd_sbuf == NULL) {
1395 		free(d->bd_fbuf, M_BPF);
1396 		return (ENOBUFS);
1397 	}
1398 	d->bd_slen = 0;
1399 	d->bd_hlen = 0;
1400 	return (0);
1401 }
1402 
1403 /*
1404  * Free buffers currently in use by a descriptor.
1405  * Called on close.
1406  */
1407 static void
1408 bpf_freed(d)
1409 	struct bpf_d *d;
1410 {
1411 	/*
1412 	 * We don't need to lock out interrupts since this descriptor has
1413 	 * been detached from its interface and it yet hasn't been marked
1414 	 * free.
1415 	 */
1416 	if (d->bd_sbuf != NULL) {
1417 		free(d->bd_sbuf, M_BPF);
1418 		if (d->bd_hbuf != NULL)
1419 			free(d->bd_hbuf, M_BPF);
1420 		if (d->bd_fbuf != NULL)
1421 			free(d->bd_fbuf, M_BPF);
1422 	}
1423 	if (d->bd_filter)
1424 		free((caddr_t)d->bd_filter, M_BPF);
1425 	mtx_destroy(&d->bd_mtx);
1426 }
1427 
1428 /*
1429  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
1430  * fixed size of the link header (variable length headers not yet supported).
1431  */
1432 void
1433 bpfattach(ifp, dlt, hdrlen)
1434 	struct ifnet *ifp;
1435 	u_int dlt, hdrlen;
1436 {
1437 
1438 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
1439 }
1440 
1441 /*
1442  * Attach an interface to bpf.  ifp is a pointer to the structure
1443  * defining the interface to be attached, dlt is the link layer type,
1444  * and hdrlen is the fixed size of the link header (variable length
1445  * headers are not yet supporrted).
1446  */
1447 void
1448 bpfattach2(ifp, dlt, hdrlen, driverp)
1449 	struct ifnet *ifp;
1450 	u_int dlt, hdrlen;
1451 	struct bpf_if **driverp;
1452 {
1453 	struct bpf_if *bp;
1454 	bp = (struct bpf_if *)malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
1455 	if (bp == NULL)
1456 		panic("bpfattach");
1457 
1458 	LIST_INIT(&bp->bif_dlist);
1459 	bp->bif_driverp = driverp;
1460 	bp->bif_ifp = ifp;
1461 	bp->bif_dlt = dlt;
1462 	mtx_init(&bp->bif_mtx, "bpf interface lock", NULL, MTX_DEF);
1463 
1464 	mtx_lock(&bpf_mtx);
1465 	LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
1466 	mtx_unlock(&bpf_mtx);
1467 
1468 	*bp->bif_driverp = NULL;
1469 
1470 	/*
1471 	 * Compute the length of the bpf header.  This is not necessarily
1472 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1473 	 * that the network layer header begins on a longword boundary (for
1474 	 * performance reasons and to alleviate alignment restrictions).
1475 	 */
1476 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1477 
1478 	if (bootverbose)
1479 		if_printf(ifp, "bpf attached\n");
1480 }
1481 
1482 /*
1483  * Detach bpf from an interface.  This involves detaching each descriptor
1484  * associated with the interface, and leaving bd_bif NULL.  Notify each
1485  * descriptor as it's detached so that any sleepers wake up and get
1486  * ENXIO.
1487  */
1488 void
1489 bpfdetach(ifp)
1490 	struct ifnet *ifp;
1491 {
1492 	struct bpf_if	*bp;
1493 	struct bpf_d	*d;
1494 
1495 	/* Locate BPF interface information */
1496 	mtx_lock(&bpf_mtx);
1497 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
1498 		if (ifp == bp->bif_ifp)
1499 			break;
1500 	}
1501 
1502 	/* Interface wasn't attached */
1503 	if ((bp == NULL) || (bp->bif_ifp == NULL)) {
1504 		mtx_unlock(&bpf_mtx);
1505 		printf("bpfdetach: %s was not attached\n", ifp->if_xname);
1506 		return;
1507 	}
1508 
1509 	LIST_REMOVE(bp, bif_next);
1510 	mtx_unlock(&bpf_mtx);
1511 
1512 	while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
1513 		bpf_detachd(d);
1514 		BPFD_LOCK(d);
1515 		bpf_wakeup(d);
1516 		BPFD_UNLOCK(d);
1517 	}
1518 
1519 	mtx_destroy(&bp->bif_mtx);
1520 	free(bp, M_BPF);
1521 }
1522 
1523 /*
1524  * Get a list of available data link type of the interface.
1525  */
1526 static int
1527 bpf_getdltlist(d, bfl)
1528 	struct bpf_d *d;
1529 	struct bpf_dltlist *bfl;
1530 {
1531 	int n, error;
1532 	struct ifnet *ifp;
1533 	struct bpf_if *bp;
1534 
1535 	ifp = d->bd_bif->bif_ifp;
1536 	n = 0;
1537 	error = 0;
1538 	mtx_lock(&bpf_mtx);
1539 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
1540 		if (bp->bif_ifp != ifp)
1541 			continue;
1542 		if (bfl->bfl_list != NULL) {
1543 			if (n >= bfl->bfl_len) {
1544 				mtx_unlock(&bpf_mtx);
1545 				return (ENOMEM);
1546 			}
1547 			error = copyout(&bp->bif_dlt,
1548 			    bfl->bfl_list + n, sizeof(u_int));
1549 		}
1550 		n++;
1551 	}
1552 	mtx_unlock(&bpf_mtx);
1553 	bfl->bfl_len = n;
1554 	return (error);
1555 }
1556 
1557 /*
1558  * Set the data link type of a BPF instance.
1559  */
1560 static int
1561 bpf_setdlt(d, dlt)
1562 	struct bpf_d *d;
1563 	u_int dlt;
1564 {
1565 	int error, opromisc;
1566 	struct ifnet *ifp;
1567 	struct bpf_if *bp;
1568 
1569 	if (d->bd_bif->bif_dlt == dlt)
1570 		return (0);
1571 	ifp = d->bd_bif->bif_ifp;
1572 	mtx_lock(&bpf_mtx);
1573 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
1574 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1575 			break;
1576 	}
1577 	mtx_unlock(&bpf_mtx);
1578 	if (bp != NULL) {
1579 		opromisc = d->bd_promisc;
1580 		bpf_detachd(d);
1581 		bpf_attachd(d, bp);
1582 		BPFD_LOCK(d);
1583 		reset_d(d);
1584 		BPFD_UNLOCK(d);
1585 		if (opromisc) {
1586 			error = ifpromisc(bp->bif_ifp, 1);
1587 			if (error)
1588 				if_printf(bp->bif_ifp,
1589 					"bpf_setdlt: ifpromisc failed (%d)\n",
1590 					error);
1591 			else
1592 				d->bd_promisc = 1;
1593 		}
1594 	}
1595 	return (bp == NULL ? EINVAL : 0);
1596 }
1597 
1598 static void
1599 bpf_clone(arg, name, namelen, dev)
1600 	void *arg;
1601 	char *name;
1602 	int namelen;
1603 	struct cdev **dev;
1604 {
1605 	int u;
1606 
1607 	if (*dev != NULL)
1608 		return;
1609 	if (dev_stdclone(name, NULL, "bpf", &u) != 1)
1610 		return;
1611 	*dev = make_dev(&bpf_cdevsw, unit2minor(u), UID_ROOT, GID_WHEEL, 0600,
1612 	    "bpf%d", u);
1613 	dev_ref(*dev);
1614 	(*dev)->si_flags |= SI_CHEAPCLONE;
1615 	return;
1616 }
1617 
1618 static void
1619 bpf_drvinit(unused)
1620 	void *unused;
1621 {
1622 
1623 	mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
1624 	LIST_INIT(&bpf_iflist);
1625 	EVENTHANDLER_REGISTER(dev_clone, bpf_clone, 0, 1000);
1626 }
1627 
1628 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL)
1629 
1630 #else /* !DEV_BPF && !NETGRAPH_BPF */
1631 /*
1632  * NOP stubs to allow bpf-using drivers to load and function.
1633  *
1634  * A 'better' implementation would allow the core bpf functionality
1635  * to be loaded at runtime.
1636  */
1637 
1638 void
1639 bpf_tap(bp, pkt, pktlen)
1640 	struct bpf_if *bp;
1641 	u_char *pkt;
1642 	u_int pktlen;
1643 {
1644 }
1645 
1646 void
1647 bpf_mtap(bp, m)
1648 	struct bpf_if *bp;
1649 	struct mbuf *m;
1650 {
1651 }
1652 
1653 void
1654 bpf_mtap2(bp, d, l, m)
1655 	struct bpf_if *bp;
1656 	void *d;
1657 	u_int l;
1658 	struct mbuf *m;
1659 {
1660 }
1661 
1662 void
1663 bpfattach(ifp, dlt, hdrlen)
1664 	struct ifnet *ifp;
1665 	u_int dlt, hdrlen;
1666 {
1667 }
1668 
1669 void
1670 bpfattach2(ifp, dlt, hdrlen, driverp)
1671 	struct ifnet *ifp;
1672 	u_int dlt, hdrlen;
1673 	struct bpf_if **driverp;
1674 {
1675 }
1676 
1677 void
1678 bpfdetach(ifp)
1679 	struct ifnet *ifp;
1680 {
1681 }
1682 
1683 u_int
1684 bpf_filter(pc, p, wirelen, buflen)
1685 	const struct bpf_insn *pc;
1686 	u_char *p;
1687 	u_int wirelen;
1688 	u_int buflen;
1689 {
1690 	return -1;	/* "no filter" behaviour */
1691 }
1692 
1693 int
1694 bpf_validate(f, len)
1695 	const struct bpf_insn *f;
1696 	int len;
1697 {
1698 	return 0;		/* false */
1699 }
1700 
1701 #endif /* !DEV_BPF && !NETGRAPH_BPF */
1702