xref: /dragonfly/sys/net/bpf.c (revision 8a0bcd56)
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  * 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 University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *      @(#)bpf.c	8.2 (Berkeley) 3/28/94
39  *
40  * $FreeBSD: src/sys/net/bpf.c,v 1.59.2.12 2002/04/14 21:41:48 luigi Exp $
41  * $DragonFly: src/sys/net/bpf.c,v 1.50 2008/09/23 11:28:49 sephe Exp $
42  */
43 
44 #include "use_bpf.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/conf.h>
49 #include <sys/device.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/time.h>
53 #include <sys/proc.h>
54 #include <sys/signalvar.h>
55 #include <sys/filio.h>
56 #include <sys/sockio.h>
57 #include <sys/ttycom.h>
58 #include <sys/filedesc.h>
59 
60 #include <sys/event.h>
61 
62 #include <sys/socket.h>
63 #include <sys/vnode.h>
64 
65 #include <sys/thread2.h>
66 #include <sys/mplock2.h>
67 
68 #include <net/if.h>
69 #include <net/bpf.h>
70 #include <net/bpfdesc.h>
71 #include <net/netmsg2.h>
72 
73 #include <netinet/in.h>
74 #include <netinet/if_ether.h>
75 #include <sys/kernel.h>
76 #include <sys/sysctl.h>
77 
78 #include <sys/devfs.h>
79 
80 struct netmsg_bpf_output {
81 	struct netmsg_base base;
82 	struct mbuf	*nm_mbuf;
83 	struct ifnet	*nm_ifp;
84 	struct sockaddr	*nm_dst;
85 };
86 
87 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
88 DEVFS_DECLARE_CLONE_BITMAP(bpf);
89 
90 #if NBPF <= 1
91 #define BPF_PREALLOCATED_UNITS	4
92 #else
93 #define BPF_PREALLOCATED_UNITS	NBPF
94 #endif
95 
96 #if NBPF > 0
97 
98 /*
99  * The default read buffer size is patchable.
100  */
101 static int bpf_bufsize = BPF_DEFAULTBUFSIZE;
102 SYSCTL_INT(_debug, OID_AUTO, bpf_bufsize, CTLFLAG_RW,
103 	   &bpf_bufsize, 0, "");
104 int bpf_maxbufsize = BPF_MAXBUFSIZE;
105 SYSCTL_INT(_debug, OID_AUTO, bpf_maxbufsize, CTLFLAG_RW,
106 	   &bpf_maxbufsize, 0, "");
107 
108 /*
109  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
110  */
111 static struct bpf_if	*bpf_iflist;
112 
113 static int	bpf_allocbufs(struct bpf_d *);
114 static void	bpf_attachd(struct bpf_d *d, struct bpf_if *bp);
115 static void	bpf_detachd(struct bpf_d *d);
116 static void	bpf_resetd(struct bpf_d *);
117 static void	bpf_freed(struct bpf_d *);
118 static void	bpf_mcopy(const void *, void *, size_t);
119 static int	bpf_movein(struct uio *, int, struct mbuf **,
120 			   struct sockaddr *, int *, struct bpf_insn *);
121 static int	bpf_setif(struct bpf_d *, struct ifreq *);
122 static void	bpf_timed_out(void *);
123 static void	bpf_wakeup(struct bpf_d *);
124 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
125 			    void (*)(const void *, void *, size_t),
126 			    const struct timeval *);
127 static int	bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
128 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
129 static int	bpf_setdlt(struct bpf_d *, u_int);
130 static void	bpf_drvinit(void *unused);
131 static void	bpf_filter_detach(struct knote *kn);
132 static int	bpf_filter_read(struct knote *kn, long hint);
133 
134 static d_open_t		bpfopen;
135 static d_clone_t	bpfclone;
136 static d_close_t	bpfclose;
137 static d_read_t		bpfread;
138 static d_write_t	bpfwrite;
139 static d_ioctl_t	bpfioctl;
140 static d_kqfilter_t	bpfkqfilter;
141 
142 #define CDEV_MAJOR 23
143 static struct dev_ops bpf_ops = {
144 	{ "bpf", CDEV_MAJOR, 0 },
145 	.d_open =	bpfopen,
146 	.d_close =	bpfclose,
147 	.d_read =	bpfread,
148 	.d_write =	bpfwrite,
149 	.d_ioctl =	bpfioctl,
150 	.d_kqfilter =	bpfkqfilter
151 };
152 
153 
154 static int
155 bpf_movein(struct uio *uio, int linktype, struct mbuf **mp,
156 	   struct sockaddr *sockp, int *datlen, struct bpf_insn *wfilter)
157 {
158 	struct mbuf *m;
159 	int error;
160 	int len;
161 	int hlen;
162 	int slen;
163 
164 	*datlen = 0;
165 	*mp = NULL;
166 
167 	/*
168 	 * Build a sockaddr based on the data link layer type.
169 	 * We do this at this level because the ethernet header
170 	 * is copied directly into the data field of the sockaddr.
171 	 * In the case of SLIP, there is no header and the packet
172 	 * is forwarded as is.
173 	 * Also, we are careful to leave room at the front of the mbuf
174 	 * for the link level header.
175 	 */
176 	switch (linktype) {
177 	case DLT_SLIP:
178 		sockp->sa_family = AF_INET;
179 		hlen = 0;
180 		break;
181 
182 	case DLT_EN10MB:
183 		sockp->sa_family = AF_UNSPEC;
184 		/* XXX Would MAXLINKHDR be better? */
185 		hlen = sizeof(struct ether_header);
186 		break;
187 
188 	case DLT_RAW:
189 	case DLT_NULL:
190 		sockp->sa_family = AF_UNSPEC;
191 		hlen = 0;
192 		break;
193 
194 	case DLT_ATM_RFC1483:
195 		/*
196 		 * en atm driver requires 4-byte atm pseudo header.
197 		 * though it isn't standard, vpi:vci needs to be
198 		 * specified anyway.
199 		 */
200 		sockp->sa_family = AF_UNSPEC;
201 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
202 		break;
203 
204 	case DLT_PPP:
205 		sockp->sa_family = AF_UNSPEC;
206 		hlen = 4;	/* This should match PPP_HDRLEN */
207 		break;
208 
209 	default:
210 		return(EIO);
211 	}
212 
213 	len = uio->uio_resid;
214 	*datlen = len - hlen;
215 	if ((unsigned)len > MCLBYTES)
216 		return(EIO);
217 
218 	m = m_getl(len, MB_WAIT, MT_DATA, M_PKTHDR, NULL);
219 	if (m == NULL)
220 		return(ENOBUFS);
221 	m->m_pkthdr.len = m->m_len = len;
222 	m->m_pkthdr.rcvif = NULL;
223 	*mp = m;
224 
225 	if (m->m_len < hlen) {
226 		error = EPERM;
227 		goto bad;
228 	}
229 
230 	error = uiomove(mtod(m, u_char *), len, uio);
231 	if (error)
232 		goto bad;
233 
234 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
235 	if (slen == 0) {
236 		error = EPERM;
237 		goto bad;
238 	}
239 
240 	/*
241 	 * Make room for link header, and copy it to sockaddr.
242 	 */
243 	if (hlen != 0) {
244 		bcopy(m->m_data, sockp->sa_data, hlen);
245 		m->m_pkthdr.len -= hlen;
246 		m->m_len -= hlen;
247 		m->m_data += hlen; /* XXX */
248 	}
249 	return (0);
250 bad:
251 	m_freem(m);
252 	return(error);
253 }
254 
255 /*
256  * Attach file to the bpf interface, i.e. make d listen on bp.
257  * Must be called at splimp.
258  */
259 static void
260 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
261 {
262 	/*
263 	 * Point d at bp, and add d to the interface's list of listeners.
264 	 * Finally, point the driver's bpf cookie at the interface so
265 	 * it will divert packets to bpf.
266 	 */
267 	d->bd_bif = bp;
268 	SLIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
269 	*bp->bif_driverp = bp;
270 
271 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
272 }
273 
274 /*
275  * Detach a file from its interface.
276  */
277 static void
278 bpf_detachd(struct bpf_d *d)
279 {
280 	int error;
281 	struct bpf_if *bp;
282 	struct ifnet *ifp;
283 
284 	bp = d->bd_bif;
285 	ifp = bp->bif_ifp;
286 
287 	/* Remove d from the interface's descriptor list. */
288 	SLIST_REMOVE(&bp->bif_dlist, d, bpf_d, bd_next);
289 
290 	if (SLIST_EMPTY(&bp->bif_dlist)) {
291 		/*
292 		 * Let the driver know that there are no more listeners.
293 		 */
294 		*bp->bif_driverp = NULL;
295 	}
296 	d->bd_bif = NULL;
297 
298 	EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
299 
300 	/*
301 	 * Check if this descriptor had requested promiscuous mode.
302 	 * If so, turn it off.
303 	 */
304 	if (d->bd_promisc) {
305 		d->bd_promisc = 0;
306 		error = ifpromisc(ifp, 0);
307 		if (error != 0 && error != ENXIO) {
308 			/*
309 			 * ENXIO can happen if a pccard is unplugged,
310 			 * Something is really wrong if we were able to put
311 			 * the driver into promiscuous mode, but can't
312 			 * take it out.
313 			 */
314 			if_printf(ifp, "bpf_detach: ifpromisc failed(%d)\n",
315 				  error);
316 		}
317 	}
318 }
319 
320 /*
321  * Open ethernet device.  Returns ENXIO for illegal minor device number,
322  * EBUSY if file is open by another process.
323  */
324 /* ARGSUSED */
325 static int
326 bpfopen(struct dev_open_args *ap)
327 {
328 	cdev_t dev = ap->a_head.a_dev;
329 	struct bpf_d *d;
330 
331 	if (ap->a_cred->cr_prison)
332 		return(EPERM);
333 
334 	d = dev->si_drv1;
335 	/*
336 	 * Each minor can be opened by only one process.  If the requested
337 	 * minor is in use, return EBUSY.
338 	 */
339 	if (d != NULL)
340 		return(EBUSY);
341 
342 	MALLOC(d, struct bpf_d *, sizeof *d, M_BPF, M_WAITOK | M_ZERO);
343 	dev->si_drv1 = d;
344 	d->bd_bufsize = bpf_bufsize;
345 	d->bd_sig = SIGIO;
346 	d->bd_seesent = 1;
347 	callout_init(&d->bd_callout);
348 	return(0);
349 }
350 
351 static int
352 bpfclone(struct dev_clone_args *ap)
353 {
354 	int unit;
355 
356 	unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(bpf), 0);
357 	ap->a_dev = make_only_dev(&bpf_ops, unit, 0, 0, 0600, "bpf%d", unit);
358 
359 	return 0;
360 }
361 
362 /*
363  * Close the descriptor by detaching it from its interface,
364  * deallocating its buffers, and marking it free.
365  */
366 /* ARGSUSED */
367 static int
368 bpfclose(struct dev_close_args *ap)
369 {
370 	cdev_t dev = ap->a_head.a_dev;
371 	struct bpf_d *d = dev->si_drv1;
372 
373 	funsetown(d->bd_sigio);
374 	crit_enter();
375 	if (d->bd_state == BPF_WAITING)
376 		callout_stop(&d->bd_callout);
377 	d->bd_state = BPF_IDLE;
378 	if (d->bd_bif != NULL)
379 		bpf_detachd(d);
380 	crit_exit();
381 	bpf_freed(d);
382 	dev->si_drv1 = NULL;
383 	if (dev->si_uminor >= BPF_PREALLOCATED_UNITS) {
384 		devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(bpf), dev->si_uminor);
385 		destroy_dev(dev);
386 	}
387 	kfree(d, M_BPF);
388 	return(0);
389 }
390 
391 /*
392  * Rotate the packet buffers in descriptor d.  Move the store buffer
393  * into the hold slot, and the free buffer into the store slot.
394  * Zero the length of the new store buffer.
395  */
396 #define ROTATE_BUFFERS(d) \
397 	(d)->bd_hbuf = (d)->bd_sbuf; \
398 	(d)->bd_hlen = (d)->bd_slen; \
399 	(d)->bd_sbuf = (d)->bd_fbuf; \
400 	(d)->bd_slen = 0; \
401 	(d)->bd_fbuf = NULL;
402 /*
403  *  bpfread - read next chunk of packets from buffers
404  */
405 static int
406 bpfread(struct dev_read_args *ap)
407 {
408 	cdev_t dev = ap->a_head.a_dev;
409 	struct bpf_d *d = dev->si_drv1;
410 	int timed_out;
411 	int error;
412 
413 	/*
414 	 * Restrict application to use a buffer the same size as
415 	 * as kernel buffers.
416 	 */
417 	if (ap->a_uio->uio_resid != d->bd_bufsize)
418 		return(EINVAL);
419 
420 	crit_enter();
421 	if (d->bd_state == BPF_WAITING)
422 		callout_stop(&d->bd_callout);
423 	timed_out = (d->bd_state == BPF_TIMED_OUT);
424 	d->bd_state = BPF_IDLE;
425 	/*
426 	 * If the hold buffer is empty, then do a timed sleep, which
427 	 * ends when the timeout expires or when enough packets
428 	 * have arrived to fill the store buffer.
429 	 */
430 	while (d->bd_hbuf == NULL) {
431 		if ((d->bd_immediate || (ap->a_ioflag & IO_NDELAY) || timed_out)
432 		    && d->bd_slen != 0) {
433 			/*
434 			 * A packet(s) either arrived since the previous,
435 			 * We're in immediate mode, or are reading
436 			 * in non-blocking mode, and a packet(s)
437 			 * either arrived since the previous
438 			 * read or arrived while we were asleep.
439 			 * Rotate the buffers and return what's here.
440 			 */
441 			ROTATE_BUFFERS(d);
442 			break;
443 		}
444 
445 		/*
446 		 * No data is available, check to see if the bpf device
447 		 * is still pointed at a real interface.  If not, return
448 		 * ENXIO so that the userland process knows to rebind
449 		 * it before using it again.
450 		 */
451 		if (d->bd_bif == NULL) {
452 			crit_exit();
453 			return(ENXIO);
454 		}
455 
456 		if (ap->a_ioflag & IO_NDELAY) {
457 			crit_exit();
458 			return(EWOULDBLOCK);
459 		}
460 		error = tsleep(d, PCATCH, "bpf", d->bd_rtout);
461 		if (error == EINTR || error == ERESTART) {
462 			crit_exit();
463 			return(error);
464 		}
465 		if (error == EWOULDBLOCK) {
466 			/*
467 			 * On a timeout, return what's in the buffer,
468 			 * which may be nothing.  If there is something
469 			 * in the store buffer, we can rotate the buffers.
470 			 */
471 			if (d->bd_hbuf)
472 				/*
473 				 * We filled up the buffer in between
474 				 * getting the timeout and arriving
475 				 * here, so we don't need to rotate.
476 				 */
477 				break;
478 
479 			if (d->bd_slen == 0) {
480 				crit_exit();
481 				return(0);
482 			}
483 			ROTATE_BUFFERS(d);
484 			break;
485 		}
486 	}
487 	/*
488 	 * At this point, we know we have something in the hold slot.
489 	 */
490 	crit_exit();
491 
492 	/*
493 	 * Move data from hold buffer into user space.
494 	 * We know the entire buffer is transferred since
495 	 * we checked above that the read buffer is bpf_bufsize bytes.
496 	 */
497 	error = uiomove(d->bd_hbuf, d->bd_hlen, ap->a_uio);
498 
499 	crit_enter();
500 	d->bd_fbuf = d->bd_hbuf;
501 	d->bd_hbuf = NULL;
502 	d->bd_hlen = 0;
503 	crit_exit();
504 
505 	return(error);
506 }
507 
508 
509 /*
510  * If there are processes sleeping on this descriptor, wake them up.
511  */
512 static void
513 bpf_wakeup(struct bpf_d *d)
514 {
515 	if (d->bd_state == BPF_WAITING) {
516 		callout_stop(&d->bd_callout);
517 		d->bd_state = BPF_IDLE;
518 	}
519 	wakeup(d);
520 	if (d->bd_async && d->bd_sig && d->bd_sigio)
521 		pgsigio(d->bd_sigio, d->bd_sig, 0);
522 
523 	get_mplock();
524 	KNOTE(&d->bd_kq.ki_note, 0);
525 	rel_mplock();
526 }
527 
528 static void
529 bpf_timed_out(void *arg)
530 {
531 	struct bpf_d *d = (struct bpf_d *)arg;
532 
533 	crit_enter();
534 	if (d->bd_state == BPF_WAITING) {
535 		d->bd_state = BPF_TIMED_OUT;
536 		if (d->bd_slen != 0)
537 			bpf_wakeup(d);
538 	}
539 	crit_exit();
540 }
541 
542 static void
543 bpf_output_dispatch(netmsg_t msg)
544 {
545 	struct netmsg_bpf_output *bmsg = (struct netmsg_bpf_output *)msg;
546 	struct ifnet *ifp = bmsg->nm_ifp;
547 	int error;
548 
549 	/*
550 	 * The driver frees the mbuf.
551 	 */
552 	error = ifp->if_output(ifp, bmsg->nm_mbuf, bmsg->nm_dst, NULL);
553 	lwkt_replymsg(&msg->lmsg, error);
554 }
555 
556 static int
557 bpfwrite(struct dev_write_args *ap)
558 {
559 	cdev_t dev = ap->a_head.a_dev;
560 	struct bpf_d *d = dev->si_drv1;
561 	struct ifnet *ifp;
562 	struct mbuf *m;
563 	int error;
564 	struct sockaddr dst;
565 	int datlen;
566 	struct netmsg_bpf_output bmsg;
567 
568 	if (d->bd_bif == NULL)
569 		return(ENXIO);
570 
571 	ifp = d->bd_bif->bif_ifp;
572 
573 	if (ap->a_uio->uio_resid == 0)
574 		return(0);
575 
576 	error = bpf_movein(ap->a_uio, (int)d->bd_bif->bif_dlt, &m,
577 			   &dst, &datlen, d->bd_wfilter);
578 	if (error)
579 		return(error);
580 
581 	if (datlen > ifp->if_mtu) {
582 		m_freem(m);
583 		return(EMSGSIZE);
584 	}
585 
586 	if (d->bd_hdrcmplt)
587 		dst.sa_family = pseudo_AF_HDRCMPLT;
588 
589 	netmsg_init(&bmsg.base, NULL, &curthread->td_msgport,
590 		    0, bpf_output_dispatch);
591 	bmsg.nm_mbuf = m;
592 	bmsg.nm_ifp = ifp;
593 	bmsg.nm_dst = &dst;
594 
595 	return lwkt_domsg(cpu_portfn(0), &bmsg.base.lmsg, 0);
596 }
597 
598 /*
599  * Reset a descriptor by flushing its packet buffer and clearing the
600  * receive and drop counts.  Should be called at splimp.
601  */
602 static void
603 bpf_resetd(struct bpf_d *d)
604 {
605 	if (d->bd_hbuf) {
606 		/* Free the hold buffer. */
607 		d->bd_fbuf = d->bd_hbuf;
608 		d->bd_hbuf = NULL;
609 	}
610 	d->bd_slen = 0;
611 	d->bd_hlen = 0;
612 	d->bd_rcount = 0;
613 	d->bd_dcount = 0;
614 }
615 
616 /*
617  *  FIONREAD		Check for read packet available.
618  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
619  *  BIOCGBLEN		Get buffer len [for read()].
620  *  BIOCSETF		Set ethernet read filter.
621  *  BIOCSETWF		Set ethernet write filter.
622  *  BIOCFLUSH		Flush read packet buffer.
623  *  BIOCPROMISC		Put interface into promiscuous mode.
624  *  BIOCGDLT		Get link layer type.
625  *  BIOCGETIF		Get interface name.
626  *  BIOCSETIF		Set interface.
627  *  BIOCSRTIMEOUT	Set read timeout.
628  *  BIOCGRTIMEOUT	Get read timeout.
629  *  BIOCGSTATS		Get packet stats.
630  *  BIOCIMMEDIATE	Set immediate mode.
631  *  BIOCVERSION		Get filter language version.
632  *  BIOCGHDRCMPLT	Get "header already complete" flag
633  *  BIOCSHDRCMPLT	Set "header already complete" flag
634  *  BIOCGSEESENT	Get "see packets sent" flag
635  *  BIOCSSEESENT	Set "see packets sent" flag
636  *  BIOCLOCK		Set "locked" flag
637  */
638 /* ARGSUSED */
639 static int
640 bpfioctl(struct dev_ioctl_args *ap)
641 {
642 	cdev_t dev = ap->a_head.a_dev;
643 	struct bpf_d *d = dev->si_drv1;
644 	int error = 0;
645 
646 	crit_enter();
647 	if (d->bd_state == BPF_WAITING)
648 		callout_stop(&d->bd_callout);
649 	d->bd_state = BPF_IDLE;
650 	crit_exit();
651 
652 	if (d->bd_locked == 1) {
653 		switch (ap->a_cmd) {
654 		case BIOCGBLEN:
655 		case BIOCFLUSH:
656 		case BIOCGDLT:
657 		case BIOCGDLTLIST:
658 		case BIOCGETIF:
659 		case BIOCGRTIMEOUT:
660 		case BIOCGSTATS:
661 		case BIOCVERSION:
662 		case BIOCGRSIG:
663 		case BIOCGHDRCMPLT:
664 		case FIONREAD:
665 		case BIOCLOCK:
666 		case BIOCSRTIMEOUT:
667 		case BIOCIMMEDIATE:
668 		case TIOCGPGRP:
669 			break;
670 		default:
671 			return (EPERM);
672 		}
673 	}
674 	switch (ap->a_cmd) {
675 	default:
676 		error = EINVAL;
677 		break;
678 
679 	/*
680 	 * Check for read packet available.
681 	 */
682 	case FIONREAD:
683 		{
684 			int n;
685 
686 			crit_enter();
687 			n = d->bd_slen;
688 			if (d->bd_hbuf)
689 				n += d->bd_hlen;
690 			crit_exit();
691 
692 			*(int *)ap->a_data = n;
693 			break;
694 		}
695 
696 	case SIOCGIFADDR:
697 		{
698 			struct ifnet *ifp;
699 
700 			if (d->bd_bif == NULL) {
701 				error = EINVAL;
702 			} else {
703 				ifp = d->bd_bif->bif_ifp;
704 				ifnet_serialize_all(ifp);
705 				error = ifp->if_ioctl(ifp, ap->a_cmd,
706 						      ap->a_data, ap->a_cred);
707 				ifnet_deserialize_all(ifp);
708 			}
709 			break;
710 		}
711 
712 	/*
713 	 * Get buffer len [for read()].
714 	 */
715 	case BIOCGBLEN:
716 		*(u_int *)ap->a_data = d->bd_bufsize;
717 		break;
718 
719 	/*
720 	 * Set buffer length.
721 	 */
722 	case BIOCSBLEN:
723 		if (d->bd_bif != NULL) {
724 			error = EINVAL;
725 		} else {
726 			u_int size = *(u_int *)ap->a_data;
727 
728 			if (size > bpf_maxbufsize)
729 				*(u_int *)ap->a_data = size = bpf_maxbufsize;
730 			else if (size < BPF_MINBUFSIZE)
731 				*(u_int *)ap->a_data = size = BPF_MINBUFSIZE;
732 			d->bd_bufsize = size;
733 		}
734 		break;
735 
736 	/*
737 	 * Set link layer read filter.
738 	 */
739 	case BIOCSETF:
740 	case BIOCSETWF:
741 		error = bpf_setf(d, (struct bpf_program *)ap->a_data,
742 			ap->a_cmd);
743 		break;
744 
745 	/*
746 	 * Flush read packet buffer.
747 	 */
748 	case BIOCFLUSH:
749 		crit_enter();
750 		bpf_resetd(d);
751 		crit_exit();
752 		break;
753 
754 	/*
755 	 * Put interface into promiscuous mode.
756 	 */
757 	case BIOCPROMISC:
758 		if (d->bd_bif == NULL) {
759 			/*
760 			 * No interface attached yet.
761 			 */
762 			error = EINVAL;
763 			break;
764 		}
765 		crit_enter();
766 		if (d->bd_promisc == 0) {
767 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
768 			if (error == 0)
769 				d->bd_promisc = 1;
770 		}
771 		crit_exit();
772 		break;
773 
774 	/*
775 	 * Get device parameters.
776 	 */
777 	case BIOCGDLT:
778 		if (d->bd_bif == NULL)
779 			error = EINVAL;
780 		else
781 			*(u_int *)ap->a_data = d->bd_bif->bif_dlt;
782 		break;
783 
784 	/*
785 	 * Get a list of supported data link types.
786 	 */
787 	case BIOCGDLTLIST:
788 		if (d->bd_bif == NULL) {
789 			error = EINVAL;
790 		} else {
791 			error = bpf_getdltlist(d,
792 				(struct bpf_dltlist *)ap->a_data);
793 		}
794 		break;
795 
796 	/*
797 	 * Set data link type.
798 	 */
799 	case BIOCSDLT:
800 		if (d->bd_bif == NULL)
801 			error = EINVAL;
802 		else
803 			error = bpf_setdlt(d, *(u_int *)ap->a_data);
804 		break;
805 
806 	/*
807 	 * Get interface name.
808 	 */
809 	case BIOCGETIF:
810 		if (d->bd_bif == NULL) {
811 			error = EINVAL;
812 		} else {
813 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
814 			struct ifreq *const ifr = (struct ifreq *)ap->a_data;
815 
816 			strlcpy(ifr->ifr_name, ifp->if_xname,
817 				sizeof ifr->ifr_name);
818 		}
819 		break;
820 
821 	/*
822 	 * Set interface.
823 	 */
824 	case BIOCSETIF:
825 		error = bpf_setif(d, (struct ifreq *)ap->a_data);
826 		break;
827 
828 	/*
829 	 * Set read timeout.
830 	 */
831 	case BIOCSRTIMEOUT:
832 		{
833 			struct timeval *tv = (struct timeval *)ap->a_data;
834 
835 			/*
836 			 * Subtract 1 tick from tvtohz() since this isn't
837 			 * a one-shot timer.
838 			 */
839 			if ((error = itimerfix(tv)) == 0)
840 				d->bd_rtout = tvtohz_low(tv);
841 			break;
842 		}
843 
844 	/*
845 	 * Get read timeout.
846 	 */
847 	case BIOCGRTIMEOUT:
848 		{
849 			struct timeval *tv = (struct timeval *)ap->a_data;
850 
851 			tv->tv_sec = d->bd_rtout / hz;
852 			tv->tv_usec = (d->bd_rtout % hz) * ustick;
853 			break;
854 		}
855 
856 	/*
857 	 * Get packet stats.
858 	 */
859 	case BIOCGSTATS:
860 		{
861 			struct bpf_stat *bs = (struct bpf_stat *)ap->a_data;
862 
863 			bs->bs_recv = d->bd_rcount;
864 			bs->bs_drop = d->bd_dcount;
865 			break;
866 		}
867 
868 	/*
869 	 * Set immediate mode.
870 	 */
871 	case BIOCIMMEDIATE:
872 		d->bd_immediate = *(u_int *)ap->a_data;
873 		break;
874 
875 	case BIOCVERSION:
876 		{
877 			struct bpf_version *bv = (struct bpf_version *)ap->a_data;
878 
879 			bv->bv_major = BPF_MAJOR_VERSION;
880 			bv->bv_minor = BPF_MINOR_VERSION;
881 			break;
882 		}
883 
884 	/*
885 	 * Get "header already complete" flag
886 	 */
887 	case BIOCGHDRCMPLT:
888 		*(u_int *)ap->a_data = d->bd_hdrcmplt;
889 		break;
890 
891 	/*
892 	 * Set "header already complete" flag
893 	 */
894 	case BIOCSHDRCMPLT:
895 		d->bd_hdrcmplt = *(u_int *)ap->a_data ? 1 : 0;
896 		break;
897 
898 	/*
899 	 * Get "see sent packets" flag
900 	 */
901 	case BIOCGSEESENT:
902 		*(u_int *)ap->a_data = d->bd_seesent;
903 		break;
904 
905 	/*
906 	 * Set "see sent packets" flag
907 	 */
908 	case BIOCSSEESENT:
909 		d->bd_seesent = *(u_int *)ap->a_data;
910 		break;
911 
912 	case FIOASYNC:		/* Send signal on receive packets */
913 		d->bd_async = *(int *)ap->a_data;
914 		break;
915 
916 	case FIOSETOWN:
917 		error = fsetown(*(int *)ap->a_data, &d->bd_sigio);
918 		break;
919 
920 	case FIOGETOWN:
921 		*(int *)ap->a_data = fgetown(d->bd_sigio);
922 		break;
923 
924 	/* This is deprecated, FIOSETOWN should be used instead. */
925 	case TIOCSPGRP:
926 		error = fsetown(-(*(int *)ap->a_data), &d->bd_sigio);
927 		break;
928 
929 	/* This is deprecated, FIOGETOWN should be used instead. */
930 	case TIOCGPGRP:
931 		*(int *)ap->a_data = -fgetown(d->bd_sigio);
932 		break;
933 
934 	case BIOCSRSIG:		/* Set receive signal */
935 		{
936 			u_int sig;
937 
938 			sig = *(u_int *)ap->a_data;
939 
940 			if (sig >= NSIG)
941 				error = EINVAL;
942 			else
943 				d->bd_sig = sig;
944 			break;
945 		}
946 	case BIOCGRSIG:
947 		*(u_int *)ap->a_data = d->bd_sig;
948 		break;
949 	case BIOCLOCK:
950 		d->bd_locked = 1;
951 		break;
952 	}
953 	return(error);
954 }
955 
956 /*
957  * Set d's packet filter program to fp.  If this file already has a filter,
958  * free it and replace it.  Returns EINVAL for bogus requests.
959  */
960 static int
961 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
962 {
963 	struct bpf_insn *fcode, *old;
964 	u_int wfilter, flen, size;
965 
966 	if (cmd == BIOCSETWF) {
967 		old = d->bd_wfilter;
968 		wfilter = 1;
969 	} else {
970 		wfilter = 0;
971 		old = d->bd_rfilter;
972 	}
973 	if (fp->bf_insns == NULL) {
974 		if (fp->bf_len != 0)
975 			return(EINVAL);
976 		crit_enter();
977 		if (wfilter)
978 			d->bd_wfilter = NULL;
979 		else
980 			d->bd_rfilter = NULL;
981 		bpf_resetd(d);
982 		crit_exit();
983 		if (old != NULL)
984 			kfree(old, M_BPF);
985 		return(0);
986 	}
987 	flen = fp->bf_len;
988 	if (flen > BPF_MAXINSNS)
989 		return(EINVAL);
990 
991 	size = flen * sizeof *fp->bf_insns;
992 	fcode = (struct bpf_insn *)kmalloc(size, M_BPF, M_WAITOK);
993 	if (copyin(fp->bf_insns, fcode, size) == 0 &&
994 	    bpf_validate(fcode, (int)flen)) {
995 		crit_enter();
996 		if (wfilter)
997 			d->bd_wfilter = fcode;
998 		else
999 			d->bd_rfilter = fcode;
1000 		bpf_resetd(d);
1001 		crit_exit();
1002 		if (old != NULL)
1003 			kfree(old, M_BPF);
1004 
1005 		return(0);
1006 	}
1007 	kfree(fcode, M_BPF);
1008 	return(EINVAL);
1009 }
1010 
1011 /*
1012  * Detach a file from its current interface (if attached at all) and attach
1013  * to the interface indicated by the name stored in ifr.
1014  * Return an errno or 0.
1015  */
1016 static int
1017 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1018 {
1019 	struct bpf_if *bp;
1020 	int error;
1021 	struct ifnet *theywant;
1022 
1023 	theywant = ifunit(ifr->ifr_name);
1024 	if (theywant == NULL)
1025 		return(ENXIO);
1026 
1027 	/*
1028 	 * Look through attached interfaces for the named one.
1029 	 */
1030 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1031 		struct ifnet *ifp = bp->bif_ifp;
1032 
1033 		if (ifp == NULL || ifp != theywant)
1034 			continue;
1035 		/* skip additional entry */
1036 		if (bp->bif_driverp != &ifp->if_bpf)
1037 			continue;
1038 		/*
1039 		 * We found the requested interface.
1040 		 * Allocate the packet buffers if we need to.
1041 		 * If we're already attached to requested interface,
1042 		 * just flush the buffer.
1043 		 */
1044 		if (d->bd_sbuf == NULL) {
1045 			error = bpf_allocbufs(d);
1046 			if (error != 0)
1047 				return(error);
1048 		}
1049 		crit_enter();
1050 		if (bp != d->bd_bif) {
1051 			if (d->bd_bif != NULL) {
1052 				/*
1053 				 * Detach if attached to something else.
1054 				 */
1055 				bpf_detachd(d);
1056 			}
1057 
1058 			bpf_attachd(d, bp);
1059 		}
1060 		bpf_resetd(d);
1061 		crit_exit();
1062 		return(0);
1063 	}
1064 
1065 	/* Not found. */
1066 	return(ENXIO);
1067 }
1068 
1069 static struct filterops bpf_read_filtops =
1070 	{ FILTEROP_ISFD, NULL, bpf_filter_detach, bpf_filter_read };
1071 
1072 static int
1073 bpfkqfilter(struct dev_kqfilter_args *ap)
1074 {
1075 	cdev_t dev = ap->a_head.a_dev;
1076 	struct knote *kn = ap->a_kn;
1077 	struct klist *klist;
1078 	struct bpf_d *d;
1079 
1080 	d = dev->si_drv1;
1081 	if (d->bd_bif == NULL) {
1082 		ap->a_result = 1;
1083 		return (0);
1084 	}
1085 
1086 	ap->a_result = 0;
1087 	switch (kn->kn_filter) {
1088 	case EVFILT_READ:
1089 		kn->kn_fop = &bpf_read_filtops;
1090 		kn->kn_hook = (caddr_t)d;
1091 		break;
1092 	default:
1093 		ap->a_result = EOPNOTSUPP;
1094 		return (0);
1095 	}
1096 
1097 	klist = &d->bd_kq.ki_note;
1098 	knote_insert(klist, kn);
1099 
1100 	return (0);
1101 }
1102 
1103 static void
1104 bpf_filter_detach(struct knote *kn)
1105 {
1106 	struct klist *klist;
1107 	struct bpf_d *d;
1108 
1109 	d = (struct bpf_d *)kn->kn_hook;
1110 	klist = &d->bd_kq.ki_note;
1111 	knote_remove(klist, kn);
1112 }
1113 
1114 static int
1115 bpf_filter_read(struct knote *kn, long hint)
1116 {
1117 	struct bpf_d *d;
1118 	int ready = 0;
1119 
1120 	crit_enter();
1121 	d = (struct bpf_d *)kn->kn_hook;
1122 	if (d->bd_hlen != 0 ||
1123 	    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1124 	    d->bd_slen != 0)) {
1125 		ready = 1;
1126 	} else {
1127 		/* Start the read timeout if necessary. */
1128 		if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1129 			callout_reset(&d->bd_callout, d->bd_rtout,
1130 			    bpf_timed_out, d);
1131 			d->bd_state = BPF_WAITING;
1132 		}
1133 	}
1134 	crit_exit();
1135 
1136 	return (ready);
1137 }
1138 
1139 
1140 /*
1141  * Process the packet pkt of length pktlen.  The packet is parsed
1142  * by each listener's filter, and if accepted, stashed into the
1143  * corresponding buffer.
1144  */
1145 void
1146 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1147 {
1148 	struct bpf_d *d;
1149 	struct timeval tv;
1150 	int gottime = 0;
1151 	u_int slen;
1152 
1153 	get_mplock();
1154 
1155 	/* Re-check */
1156 	if (bp == NULL) {
1157 		rel_mplock();
1158 		return;
1159 	}
1160 
1161 	/*
1162 	 * Note that the ipl does not have to be raised at this point.
1163 	 * The only problem that could arise here is that if two different
1164 	 * interfaces shared any data.  This is not the case.
1165 	 */
1166 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1167 		++d->bd_rcount;
1168 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1169 		if (slen != 0) {
1170 			if (!gottime) {
1171 				microtime(&tv);
1172 				gottime = 1;
1173 			}
1174 			catchpacket(d, pkt, pktlen, slen, ovbcopy, &tv);
1175 		}
1176 	}
1177 
1178 	rel_mplock();
1179 }
1180 
1181 /*
1182  * Copy data from an mbuf chain into a buffer.  This code is derived
1183  * from m_copydata in sys/uipc_mbuf.c.
1184  */
1185 static void
1186 bpf_mcopy(const void *src_arg, void *dst_arg, size_t len)
1187 {
1188 	const struct mbuf *m;
1189 	u_int count;
1190 	u_char *dst;
1191 
1192 	m = src_arg;
1193 	dst = dst_arg;
1194 	while (len > 0) {
1195 		if (m == NULL)
1196 			panic("bpf_mcopy");
1197 		count = min(m->m_len, len);
1198 		bcopy(mtod(m, void *), dst, count);
1199 		m = m->m_next;
1200 		dst += count;
1201 		len -= count;
1202 	}
1203 }
1204 
1205 /*
1206  * Process the packet in the mbuf chain m.  The packet is parsed by each
1207  * listener's filter, and if accepted, stashed into the corresponding
1208  * buffer.
1209  */
1210 void
1211 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1212 {
1213 	struct bpf_d *d;
1214 	u_int pktlen, slen;
1215 	struct timeval tv;
1216 	int gottime = 0;
1217 
1218 	get_mplock();
1219 
1220 	/* Re-check */
1221 	if (bp == NULL) {
1222 		rel_mplock();
1223 		return;
1224 	}
1225 
1226 	/* Don't compute pktlen, if no descriptor is attached. */
1227 	if (SLIST_EMPTY(&bp->bif_dlist)) {
1228 		rel_mplock();
1229 		return;
1230 	}
1231 
1232 	pktlen = m_lengthm(m, NULL);
1233 
1234 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1235 		if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL))
1236 			continue;
1237 		++d->bd_rcount;
1238 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1239 		if (slen != 0) {
1240 			if (!gottime) {
1241 				microtime(&tv);
1242 				gottime = 1;
1243 			}
1244 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy,
1245 				    &tv);
1246 		}
1247 	}
1248 
1249 	rel_mplock();
1250 }
1251 
1252 /*
1253  * Incoming linkage from device drivers, where we have a mbuf chain
1254  * but need to prepend some arbitrary header from a linear buffer.
1255  *
1256  * Con up a minimal dummy header to pacify bpf.  Allocate (only) a
1257  * struct m_hdr on the stack.  This is safe as bpf only reads from the
1258  * fields in this header that we initialize, and will not try to free
1259  * it or keep a pointer to it.
1260  */
1261 void
1262 bpf_mtap_hdr(struct bpf_if *arg, caddr_t data, u_int dlen, struct mbuf *m, u_int direction)
1263 {
1264 	struct m_hdr mh;
1265 
1266 	mh.mh_flags = 0;
1267 	mh.mh_next = m;
1268 	mh.mh_len = dlen;
1269 	mh.mh_data = data;
1270 
1271 	return bpf_mtap(arg, (struct mbuf *) &mh);
1272 }
1273 
1274 void
1275 bpf_mtap_family(struct bpf_if *bp, struct mbuf *m, sa_family_t family)
1276 {
1277 	u_int family4;
1278 
1279 	KKASSERT(family != AF_UNSPEC);
1280 
1281 	family4 = (u_int)family;
1282 	bpf_ptap(bp, m, &family4, sizeof(family4));
1283 }
1284 
1285 /*
1286  * Process the packet in the mbuf chain m with the header in m prepended.
1287  * The packet is parsed by each listener's filter, and if accepted,
1288  * stashed into the corresponding buffer.
1289  */
1290 void
1291 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1292 {
1293 	struct mbuf mb;
1294 
1295 	/*
1296 	 * Craft on-stack mbuf suitable for passing to bpf_mtap.
1297 	 * Note that we cut corners here; we only setup what's
1298 	 * absolutely needed--this mbuf should never go anywhere else.
1299 	 */
1300 	mb.m_next = m;
1301 	mb.m_data = __DECONST(void *, data); /* LINTED */
1302 	mb.m_len = dlen;
1303 	mb.m_pkthdr.rcvif = m->m_pkthdr.rcvif;
1304 
1305 	bpf_mtap(bp, &mb);
1306 }
1307 
1308 /*
1309  * Move the packet data from interface memory (pkt) into the
1310  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1311  * otherwise 0.  "copy" is the routine called to do the actual data
1312  * transfer.  bcopy is passed in to copy contiguous chunks, while
1313  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1314  * pkt is really an mbuf.
1315  */
1316 static void
1317 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1318 	    void (*cpfn)(const void *, void *, size_t),
1319 	    const struct timeval *tv)
1320 {
1321 	struct bpf_hdr *hp;
1322 	int totlen, curlen;
1323 	int hdrlen = d->bd_bif->bif_hdrlen;
1324 	int wakeup = 0;
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 		wakeup = 1;
1355 		curlen = 0;
1356 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1357 		/*
1358 		 * Immediate mode is set, or the read timeout has
1359 		 * already expired during a select call.  A packet
1360 		 * arrived, so the reader should be woken up.
1361 		 */
1362 		wakeup = 1;
1363 	}
1364 
1365 	/*
1366 	 * Append the bpf header.
1367 	 */
1368 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1369 	hp->bh_tstamp = *tv;
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 (wakeup)
1379 		bpf_wakeup(d);
1380 }
1381 
1382 /*
1383  * Initialize all nonzero fields of a descriptor.
1384  */
1385 static int
1386 bpf_allocbufs(struct bpf_d *d)
1387 {
1388 	d->bd_fbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1389 	d->bd_sbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1390 	d->bd_slen = 0;
1391 	d->bd_hlen = 0;
1392 	return(0);
1393 }
1394 
1395 /*
1396  * Free buffers and packet filter program currently in use by a descriptor.
1397  * Called on close.
1398  */
1399 static void
1400 bpf_freed(struct bpf_d *d)
1401 {
1402 	/*
1403 	 * We don't need to lock out interrupts since this descriptor has
1404 	 * been detached from its interface and it yet hasn't been marked
1405 	 * free.
1406 	 */
1407 	if (d->bd_sbuf != NULL) {
1408 		kfree(d->bd_sbuf, M_BPF);
1409 		if (d->bd_hbuf != NULL)
1410 			kfree(d->bd_hbuf, M_BPF);
1411 		if (d->bd_fbuf != NULL)
1412 			kfree(d->bd_fbuf, M_BPF);
1413 	}
1414 	if (d->bd_rfilter)
1415 		kfree(d->bd_rfilter, M_BPF);
1416 	if (d->bd_wfilter)
1417 		kfree(d->bd_wfilter, M_BPF);
1418 }
1419 
1420 /*
1421  * Attach an interface to bpf.  ifp is a pointer to the structure
1422  * defining the interface to be attached, dlt is the link layer type,
1423  * and hdrlen is the fixed size of the link header (variable length
1424  * headers are not yet supported).
1425  */
1426 void
1427 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1428 {
1429 	bpfattach_dlt(ifp, dlt, hdrlen, &ifp->if_bpf);
1430 }
1431 
1432 void
1433 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1434 {
1435 	struct bpf_if *bp;
1436 
1437 	bp = kmalloc(sizeof *bp, M_BPF, M_WAITOK | M_ZERO);
1438 
1439 	SLIST_INIT(&bp->bif_dlist);
1440 	bp->bif_ifp = ifp;
1441 	bp->bif_dlt = dlt;
1442 	bp->bif_driverp = driverp;
1443 	*bp->bif_driverp = NULL;
1444 
1445 	bp->bif_next = bpf_iflist;
1446 	bpf_iflist = bp;
1447 
1448 	/*
1449 	 * Compute the length of the bpf header.  This is not necessarily
1450 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1451 	 * that the network layer header begins on a longword boundary (for
1452 	 * performance reasons and to alleviate alignment restrictions).
1453 	 */
1454 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1455 
1456 	if (bootverbose)
1457 		if_printf(ifp, "bpf attached\n");
1458 }
1459 
1460 /*
1461  * Detach bpf from an interface.  This involves detaching each descriptor
1462  * associated with the interface, and leaving bd_bif NULL.  Notify each
1463  * descriptor as it's detached so that any sleepers wake up and get
1464  * ENXIO.
1465  */
1466 void
1467 bpfdetach(struct ifnet *ifp)
1468 {
1469 	struct bpf_if *bp, *bp_prev;
1470 	struct bpf_d *d;
1471 
1472 	crit_enter();
1473 
1474 	/* Locate BPF interface information */
1475 	bp_prev = NULL;
1476 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1477 		if (ifp == bp->bif_ifp)
1478 			break;
1479 		bp_prev = bp;
1480 	}
1481 
1482 	/* Interface wasn't attached */
1483 	if (bp->bif_ifp == NULL) {
1484 		crit_exit();
1485 		kprintf("bpfdetach: %s was not attached\n", ifp->if_xname);
1486 		return;
1487 	}
1488 
1489 	while ((d = SLIST_FIRST(&bp->bif_dlist)) != NULL) {
1490 		bpf_detachd(d);
1491 		bpf_wakeup(d);
1492 	}
1493 
1494 	if (bp_prev != NULL)
1495 		bp_prev->bif_next = bp->bif_next;
1496 	else
1497 		bpf_iflist = bp->bif_next;
1498 
1499 	kfree(bp, M_BPF);
1500 
1501 	crit_exit();
1502 }
1503 
1504 /*
1505  * Get a list of available data link type of the interface.
1506  */
1507 static int
1508 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1509 {
1510 	int n, error;
1511 	struct ifnet *ifp;
1512 	struct bpf_if *bp;
1513 
1514 	ifp = d->bd_bif->bif_ifp;
1515 	n = 0;
1516 	error = 0;
1517 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1518 		if (bp->bif_ifp != ifp)
1519 			continue;
1520 		if (bfl->bfl_list != NULL) {
1521 			if (n >= bfl->bfl_len) {
1522 				return (ENOMEM);
1523 			}
1524 			error = copyout(&bp->bif_dlt,
1525 			    bfl->bfl_list + n, sizeof(u_int));
1526 		}
1527 		n++;
1528 	}
1529 	bfl->bfl_len = n;
1530 	return(error);
1531 }
1532 
1533 /*
1534  * Set the data link type of a BPF instance.
1535  */
1536 static int
1537 bpf_setdlt(struct bpf_d *d, u_int dlt)
1538 {
1539 	int error, opromisc;
1540 	struct ifnet *ifp;
1541 	struct bpf_if *bp;
1542 
1543 	if (d->bd_bif->bif_dlt == dlt)
1544 		return (0);
1545 	ifp = d->bd_bif->bif_ifp;
1546 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1547 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1548 			break;
1549 	}
1550 	if (bp != NULL) {
1551 		opromisc = d->bd_promisc;
1552 		crit_enter();
1553 		bpf_detachd(d);
1554 		bpf_attachd(d, bp);
1555 		bpf_resetd(d);
1556 		if (opromisc) {
1557 			error = ifpromisc(bp->bif_ifp, 1);
1558 			if (error) {
1559 				if_printf(bp->bif_ifp,
1560 					"bpf_setdlt: ifpromisc failed (%d)\n",
1561 					error);
1562 			} else {
1563 				d->bd_promisc = 1;
1564 			}
1565 		}
1566 		crit_exit();
1567 	}
1568 	return(bp == NULL ? EINVAL : 0);
1569 }
1570 
1571 static void
1572 bpf_drvinit(void *unused)
1573 {
1574 	int i;
1575 
1576 	make_autoclone_dev(&bpf_ops, &DEVFS_CLONE_BITMAP(bpf),
1577 		bpfclone, 0, 0, 0600, "bpf");
1578 	for (i = 0; i < BPF_PREALLOCATED_UNITS; i++) {
1579 		make_dev(&bpf_ops, i, 0, 0, 0600, "bpf%d", i);
1580 		devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(bpf), i);
1581 	}
1582 }
1583 
1584 static void
1585 bpf_drvuninit(void *unused)
1586 {
1587 	devfs_clone_handler_del("bpf");
1588 	dev_ops_remove_all(&bpf_ops);
1589 	devfs_clone_bitmap_uninit(&DEVFS_CLONE_BITMAP(bpf));
1590 }
1591 
1592 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvinit,NULL)
1593 SYSUNINIT(bpfdev, SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvuninit, NULL);
1594 
1595 #else /* !BPF */
1596 /*
1597  * NOP stubs to allow bpf-using drivers to load and function.
1598  *
1599  * A 'better' implementation would allow the core bpf functionality
1600  * to be loaded at runtime.
1601  */
1602 
1603 void
1604 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1605 {
1606 }
1607 
1608 void
1609 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1610 {
1611 }
1612 
1613 void
1614 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1615 {
1616 }
1617 
1618 void
1619 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1620 {
1621 }
1622 
1623 void
1624 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1625 {
1626 }
1627 
1628 void
1629 bpfdetach(struct ifnet *ifp)
1630 {
1631 }
1632 
1633 u_int
1634 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
1635 {
1636 	return -1;	/* "no filter" behaviour */
1637 }
1638 
1639 #endif /* !BPF */
1640