xref: /dragonfly/sys/net/bpf.c (revision 0de090e1)
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. 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	8.2 (Berkeley) 3/28/94
35  *
36  * $FreeBSD: src/sys/net/bpf.c,v 1.59.2.12 2002/04/14 21:41:48 luigi Exp $
37  */
38 
39 #include "use_bpf.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/conf.h>
44 #include <sys/device.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/time.h>
48 #include <sys/proc.h>
49 #include <sys/signalvar.h>
50 #include <sys/filio.h>
51 #include <sys/sockio.h>
52 #include <sys/ttycom.h>
53 #include <sys/filedesc.h>
54 
55 #include <sys/event.h>
56 
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 
60 #include <sys/thread2.h>
61 
62 #include <net/if.h>
63 #include <net/bpf.h>
64 #include <net/bpfdesc.h>
65 #include <net/netmsg2.h>
66 #include <net/netisr2.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/if_ether.h>
70 #include <sys/kernel.h>
71 #include <sys/sysctl.h>
72 
73 #include <netproto/802_11/ieee80211_dragonfly.h>
74 
75 #include <sys/devfs.h>
76 
77 struct netmsg_bpf_output {
78 	struct netmsg_base base;
79 	struct mbuf	*nm_mbuf;
80 	struct ifnet	*nm_ifp;
81 	struct sockaddr	*nm_dst;
82 };
83 
84 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
85 DEVFS_DEFINE_CLONE_BITMAP(bpf);
86 
87 #if NBPF <= 1
88 #define BPF_PREALLOCATED_UNITS	4
89 #else
90 #define BPF_PREALLOCATED_UNITS	NBPF
91 #endif
92 
93 #if NBPF > 0
94 
95 /*
96  * The default read buffer size is patchable.
97  */
98 static int bpf_bufsize = BPF_DEFAULTBUFSIZE;
99 SYSCTL_INT(_debug, OID_AUTO, bpf_bufsize, CTLFLAG_RW,
100    &bpf_bufsize, 0, "Current size of bpf buffer");
101 int bpf_maxbufsize = BPF_MAXBUFSIZE;
102 SYSCTL_INT(_debug, OID_AUTO, bpf_maxbufsize, CTLFLAG_RW,
103    &bpf_maxbufsize, 0, "Maximum size of bpf buffer");
104 
105 /*
106  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
107  */
108 static struct bpf_if	*bpf_iflist;
109 
110 static struct lwkt_token bpf_token = LWKT_TOKEN_INITIALIZER(bpf_token);
111 
112 static int	bpf_allocbufs(struct bpf_d *);
113 static void	bpf_attachd(struct bpf_d *d, struct bpf_if *bp);
114 static void	bpf_detachd(struct bpf_d *d);
115 static void	bpf_resetd(struct bpf_d *);
116 static void	bpf_freed(struct bpf_d *);
117 static void	bpf_mcopy(volatile const void *, volatile void *, size_t);
118 static int	bpf_movein(struct uio *, int, struct mbuf **,
119 			   struct sockaddr *, int *, struct bpf_insn *);
120 static int	bpf_setif(struct bpf_d *, struct ifreq *);
121 static void	bpf_timed_out(void *);
122 static void	bpf_wakeup(struct bpf_d *);
123 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
124 			    void (*)(volatile const void *,
125 				     volatile 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", 0, D_MPSAFE },
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 	const struct ieee80211_bpf_params *p;
159 	struct mbuf *m;
160 	int error;
161 	int len;
162 	int hlen;
163 	int slen;
164 
165 	*datlen = 0;
166 	*mp = NULL;
167 
168 	/*
169 	 * Build a sockaddr based on the data link layer type.
170 	 * We do this at this level because the ethernet header
171 	 * is copied directly into the data field of the sockaddr.
172 	 * In the case of SLIP, there is no header and the packet
173 	 * is forwarded as is.
174 	 * Also, we are careful to leave room at the front of the mbuf
175 	 * for the link level header.
176 	 */
177 	switch (linktype) {
178 	case DLT_SLIP:
179 		sockp->sa_family = AF_INET;
180 		hlen = 0;
181 		break;
182 
183 	case DLT_EN10MB:
184 		sockp->sa_family = AF_UNSPEC;
185 		/* XXX Would MAXLINKHDR be better? */
186 		hlen = sizeof(struct ether_header);
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 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
211 		sockp->sa_family = AF_IEEE80211;
212 		hlen = 0;
213 		break;
214 
215 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
216 		sockp->sa_family = AF_IEEE80211;
217 		sockp->sa_len = 12;	/* XXX != 0 */
218 		hlen = sizeof(struct ieee80211_bpf_params);
219 		break;
220 
221 	default:
222 		return(EIO);
223 	}
224 
225 	len = uio->uio_resid;
226 	*datlen = len - hlen;
227 	if ((unsigned)len > MCLBYTES)
228 		return(EIO);
229 
230 	m = m_getl(len, M_WAITOK, MT_DATA, M_PKTHDR, NULL);
231 	if (m == NULL)
232 		return(ENOBUFS);
233 	m->m_pkthdr.len = m->m_len = len;
234 	m->m_pkthdr.rcvif = NULL;
235 	*mp = m;
236 
237 	if (m->m_len < hlen) {
238 		error = EPERM;
239 		goto bad;
240 	}
241 
242 	error = uiomove(mtod(m, u_char *), len, uio);
243 	if (error)
244 		goto bad;
245 
246 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
247 	if (slen == 0) {
248 		error = EPERM;
249 		goto bad;
250 	}
251 
252 	/*
253 	 * Make room for link header, and copy it to sockaddr.
254 	 */
255 	if (hlen != 0) {
256 		if (sockp->sa_family == AF_IEEE80211) {
257 			/*
258 			 * Collect true length from the parameter header
259 			 * NB: sockp is known to be zero'd so if we do a
260 			 *     short copy unspecified parameters will be
261 			 *     zero.
262 			 * NB: packet may not be aligned after stripping
263 			 *     bpf params
264 			 * XXX check ibp_vers
265 			 */
266 			p = mtod(m, const struct ieee80211_bpf_params *);
267 			hlen = p->ibp_len;
268 			if (hlen > sizeof(sockp->sa_data)) {
269 				error = EINVAL;
270 				goto bad;
271 			}
272 		}
273 		bcopy(m->m_data, sockp->sa_data, hlen);
274 		m->m_pkthdr.len -= hlen;
275 		m->m_len -= hlen;
276 		m->m_data += hlen; /* XXX */
277 	}
278 	return (0);
279 bad:
280 	m_freem(m);
281 	return(error);
282 }
283 
284 /*
285  * Attach file to the bpf interface, i.e. make d listen on bp.
286  * Must be called at splimp.
287  */
288 static void
289 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
290 {
291 	/*
292 	 * Point d at bp, and add d to the interface's list of listeners.
293 	 * Finally, point the driver's bpf cookie at the interface so
294 	 * it will divert packets to bpf.
295 	 */
296 	lwkt_gettoken(&bpf_token);
297 	d->bd_bif = bp;
298 	SLIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
299 	*bp->bif_driverp = bp;
300 
301 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
302 	lwkt_reltoken(&bpf_token);
303 }
304 
305 /*
306  * Detach a file from its interface.
307  */
308 static void
309 bpf_detachd(struct bpf_d *d)
310 {
311 	int error;
312 	struct bpf_if *bp;
313 	struct ifnet *ifp;
314 
315 	lwkt_gettoken(&bpf_token);
316 	bp = d->bd_bif;
317 	ifp = bp->bif_ifp;
318 
319 	/* Remove d from the interface's descriptor list. */
320 	SLIST_REMOVE(&bp->bif_dlist, d, bpf_d, bd_next);
321 
322 	if (SLIST_EMPTY(&bp->bif_dlist)) {
323 		/*
324 		 * Let the driver know that there are no more listeners.
325 		 */
326 		*bp->bif_driverp = NULL;
327 	}
328 	d->bd_bif = NULL;
329 
330 	EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
331 
332 	/*
333 	 * Check if this descriptor had requested promiscuous mode.
334 	 * If so, turn it off.
335 	 */
336 	if (d->bd_promisc) {
337 		d->bd_promisc = 0;
338 		error = ifpromisc(ifp, 0);
339 		if (error != 0 && error != ENXIO) {
340 			/*
341 			 * ENXIO can happen if a pccard is unplugged,
342 			 * Something is really wrong if we were able to put
343 			 * the driver into promiscuous mode, but can't
344 			 * take it out.
345 			 */
346 			if_printf(ifp, "bpf_detach: ifpromisc failed(%d)\n",
347 				  error);
348 		}
349 	}
350 	lwkt_reltoken(&bpf_token);
351 }
352 
353 /*
354  * Open ethernet device.  Returns ENXIO for illegal minor device number,
355  * EBUSY if file is open by another process.
356  */
357 /* ARGSUSED */
358 static int
359 bpfopen(struct dev_open_args *ap)
360 {
361 	cdev_t dev = ap->a_head.a_dev;
362 	struct bpf_d *d;
363 
364 	lwkt_gettoken(&bpf_token);
365 	if (ap->a_cred->cr_prison) {
366 		lwkt_reltoken(&bpf_token);
367 		return(EPERM);
368 	}
369 
370 	d = dev->si_drv1;
371 	/*
372 	 * Each minor can be opened by only one process.  If the requested
373 	 * minor is in use, return EBUSY.
374 	 */
375 	if (d != NULL) {
376 		lwkt_reltoken(&bpf_token);
377 		return(EBUSY);
378 	}
379 
380 	d = kmalloc(sizeof *d, M_BPF, M_WAITOK | M_ZERO);
381 	dev->si_drv1 = d;
382 	d->bd_bufsize = bpf_bufsize;
383 	d->bd_sig = SIGIO;
384 	d->bd_seesent = 1;
385 	callout_init(&d->bd_callout);
386 	lwkt_reltoken(&bpf_token);
387 
388 	return(0);
389 }
390 
391 static int
392 bpfclone(struct dev_clone_args *ap)
393 {
394 	int unit;
395 
396 	unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(bpf), 0);
397 	ap->a_dev = make_only_dev(&bpf_ops, unit, 0, 0, 0600, "bpf%d", unit);
398 
399 	return 0;
400 }
401 
402 /*
403  * Close the descriptor by detaching it from its interface,
404  * deallocating its buffers, and marking it free.
405  */
406 /* ARGSUSED */
407 static int
408 bpfclose(struct dev_close_args *ap)
409 {
410 	cdev_t dev = ap->a_head.a_dev;
411 	struct bpf_d *d = dev->si_drv1;
412 
413 	lwkt_gettoken(&bpf_token);
414 	funsetown(&d->bd_sigio);
415 	if (d->bd_state == BPF_WAITING)
416 		callout_stop(&d->bd_callout);
417 	d->bd_state = BPF_IDLE;
418 	if (d->bd_bif != NULL)
419 		bpf_detachd(d);
420 	bpf_freed(d);
421 	dev->si_drv1 = NULL;
422 	if (dev->si_uminor >= BPF_PREALLOCATED_UNITS) {
423 		devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(bpf), dev->si_uminor);
424 		destroy_dev(dev);
425 	}
426 	kfree(d, M_BPF);
427 	lwkt_reltoken(&bpf_token);
428 
429 	return(0);
430 }
431 
432 /*
433  * Rotate the packet buffers in descriptor d.  Move the store buffer
434  * into the hold slot, and the free buffer into the store slot.
435  * Zero the length of the new store buffer.
436  */
437 #define ROTATE_BUFFERS(d) \
438 	(d)->bd_hbuf = (d)->bd_sbuf; \
439 	(d)->bd_hlen = (d)->bd_slen; \
440 	(d)->bd_sbuf = (d)->bd_fbuf; \
441 	(d)->bd_slen = 0; \
442 	(d)->bd_fbuf = NULL;
443 /*
444  *  bpfread - read next chunk of packets from buffers
445  */
446 static int
447 bpfread(struct dev_read_args *ap)
448 {
449 	cdev_t dev = ap->a_head.a_dev;
450 	struct bpf_d *d = dev->si_drv1;
451 	int timed_out;
452 	int error;
453 
454 	lwkt_gettoken(&bpf_token);
455 	/*
456 	 * Restrict application to use a buffer the same size as
457 	 * as kernel buffers.
458 	 */
459 	if (ap->a_uio->uio_resid != d->bd_bufsize) {
460 		lwkt_reltoken(&bpf_token);
461 		return(EINVAL);
462 	}
463 
464 	if (d->bd_state == BPF_WAITING)
465 		callout_stop(&d->bd_callout);
466 	timed_out = (d->bd_state == BPF_TIMED_OUT);
467 	d->bd_state = BPF_IDLE;
468 	/*
469 	 * If the hold buffer is empty, then do a timed sleep, which
470 	 * ends when the timeout expires or when enough packets
471 	 * have arrived to fill the store buffer.
472 	 */
473 	while (d->bd_hbuf == NULL) {
474 		if ((d->bd_immediate || (ap->a_ioflag & IO_NDELAY) || timed_out)
475 		    && d->bd_slen != 0) {
476 			/*
477 			 * A packet(s) either arrived since the previous,
478 			 * We're in immediate mode, or are reading
479 			 * in non-blocking mode, and a packet(s)
480 			 * either arrived since the previous
481 			 * read or arrived while we were asleep.
482 			 * Rotate the buffers and return what's here.
483 			 */
484 			ROTATE_BUFFERS(d);
485 			break;
486 		}
487 
488 		/*
489 		 * No data is available, check to see if the bpf device
490 		 * is still pointed at a real interface.  If not, return
491 		 * ENXIO so that the userland process knows to rebind
492 		 * it before using it again.
493 		 */
494 		if (d->bd_bif == NULL) {
495 			lwkt_reltoken(&bpf_token);
496 			return(ENXIO);
497 		}
498 
499 		if (ap->a_ioflag & IO_NDELAY) {
500 			lwkt_reltoken(&bpf_token);
501 			return(EWOULDBLOCK);
502 		}
503 		error = tsleep(d, PCATCH, "bpf", d->bd_rtout);
504 		if (error == EINTR || error == ERESTART) {
505 			lwkt_reltoken(&bpf_token);
506 			return(error);
507 		}
508 		if (error == EWOULDBLOCK) {
509 			/*
510 			 * On a timeout, return what's in the buffer,
511 			 * which may be nothing.  If there is something
512 			 * in the store buffer, we can rotate the buffers.
513 			 */
514 			if (d->bd_hbuf)
515 				/*
516 				 * We filled up the buffer in between
517 				 * getting the timeout and arriving
518 				 * here, so we don't need to rotate.
519 				 */
520 				break;
521 
522 			if (d->bd_slen == 0) {
523 				lwkt_reltoken(&bpf_token);
524 				return(0);
525 			}
526 			ROTATE_BUFFERS(d);
527 			break;
528 		}
529 	}
530 	/*
531 	 * At this point, we know we have something in the hold slot.
532 	 */
533 
534 	/*
535 	 * Move data from hold buffer into user space.
536 	 * We know the entire buffer is transferred since
537 	 * we checked above that the read buffer is bpf_bufsize bytes.
538 	 */
539 	error = uiomove(d->bd_hbuf, d->bd_hlen, ap->a_uio);
540 
541 	d->bd_fbuf = d->bd_hbuf;
542 	d->bd_hbuf = NULL;
543 	d->bd_hlen = 0;
544 	lwkt_reltoken(&bpf_token);
545 
546 	return(error);
547 }
548 
549 
550 /*
551  * If there are processes sleeping on this descriptor, wake them up.
552  */
553 static void
554 bpf_wakeup(struct bpf_d *d)
555 {
556 	if (d->bd_state == BPF_WAITING) {
557 		callout_stop(&d->bd_callout);
558 		d->bd_state = BPF_IDLE;
559 	}
560 	wakeup(d);
561 	if (d->bd_async && d->bd_sig && d->bd_sigio)
562 		pgsigio(d->bd_sigio, d->bd_sig, 0);
563 
564 	KNOTE(&d->bd_kq.ki_note, 0);
565 }
566 
567 static void
568 bpf_timed_out(void *arg)
569 {
570 	struct bpf_d *d = (struct bpf_d *)arg;
571 
572 	if (d->bd_state == BPF_WAITING) {
573 		d->bd_state = BPF_TIMED_OUT;
574 		if (d->bd_slen != 0)
575 			bpf_wakeup(d);
576 	}
577 }
578 
579 static void
580 bpf_output_dispatch(netmsg_t msg)
581 {
582 	struct netmsg_bpf_output *bmsg = (struct netmsg_bpf_output *)msg;
583 	struct ifnet *ifp = bmsg->nm_ifp;
584 	int error;
585 
586 	/*
587 	 * The driver frees the mbuf.
588 	 */
589 	error = ifp->if_output(ifp, bmsg->nm_mbuf, bmsg->nm_dst, NULL);
590 	lwkt_replymsg(&msg->lmsg, error);
591 }
592 
593 static int
594 bpfwrite(struct dev_write_args *ap)
595 {
596 	cdev_t dev = ap->a_head.a_dev;
597 	struct bpf_d *d = dev->si_drv1;
598 	struct ifnet *ifp;
599 	struct mbuf *m;
600 	int error, ret;
601 	struct sockaddr dst;
602 	int datlen;
603 	struct netmsg_bpf_output bmsg;
604 
605 	lwkt_gettoken(&bpf_token);
606 	if (d->bd_bif == NULL) {
607 		lwkt_reltoken(&bpf_token);
608 		return(ENXIO);
609 	}
610 
611 	ifp = d->bd_bif->bif_ifp;
612 
613 	if (ap->a_uio->uio_resid == 0) {
614 		lwkt_reltoken(&bpf_token);
615 		return(0);
616 	}
617 
618 	error = bpf_movein(ap->a_uio, (int)d->bd_bif->bif_dlt, &m,
619 			   &dst, &datlen, d->bd_wfilter);
620 	if (error) {
621 		lwkt_reltoken(&bpf_token);
622 		return(error);
623 	}
624 
625 	if (datlen > ifp->if_mtu) {
626 		m_freem(m);
627 		lwkt_reltoken(&bpf_token);
628 		return(EMSGSIZE);
629 	}
630 
631 	if (d->bd_hdrcmplt)
632 		dst.sa_family = pseudo_AF_HDRCMPLT;
633 
634 	netmsg_init(&bmsg.base, NULL, &curthread->td_msgport,
635 		    0, bpf_output_dispatch);
636 	bmsg.nm_mbuf = m;
637 	bmsg.nm_ifp = ifp;
638 	bmsg.nm_dst = &dst;
639 
640 	ret = lwkt_domsg(netisr_cpuport(0), &bmsg.base.lmsg, 0);
641 	lwkt_reltoken(&bpf_token);
642 
643 	return ret;
644 }
645 
646 /*
647  * Reset a descriptor by flushing its packet buffer and clearing the
648  * receive and drop counts.  Should be called at splimp.
649  */
650 static void
651 bpf_resetd(struct bpf_d *d)
652 {
653 	if (d->bd_hbuf) {
654 		/* Free the hold buffer. */
655 		d->bd_fbuf = d->bd_hbuf;
656 		d->bd_hbuf = NULL;
657 	}
658 	d->bd_slen = 0;
659 	d->bd_hlen = 0;
660 	d->bd_rcount = 0;
661 	d->bd_dcount = 0;
662 }
663 
664 /*
665  *  FIONREAD		Check for read packet available.
666  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
667  *  BIOCGBLEN		Get buffer len [for read()].
668  *  BIOCSETF		Set ethernet read filter.
669  *  BIOCSETWF		Set ethernet write filter.
670  *  BIOCFLUSH		Flush read packet buffer.
671  *  BIOCPROMISC		Put interface into promiscuous mode.
672  *  BIOCGDLT		Get link layer type.
673  *  BIOCGETIF		Get interface name.
674  *  BIOCSETIF		Set interface.
675  *  BIOCSRTIMEOUT	Set read timeout.
676  *  BIOCGRTIMEOUT	Get read timeout.
677  *  BIOCGSTATS		Get packet stats.
678  *  BIOCIMMEDIATE	Set immediate mode.
679  *  BIOCVERSION		Get filter language version.
680  *  BIOCGHDRCMPLT	Get "header already complete" flag
681  *  BIOCSHDRCMPLT	Set "header already complete" flag
682  *  BIOCGSEESENT	Get "see packets sent" flag
683  *  BIOCSSEESENT	Set "see packets sent" flag
684  *  BIOCLOCK		Set "locked" flag
685  */
686 /* ARGSUSED */
687 static int
688 bpfioctl(struct dev_ioctl_args *ap)
689 {
690 	cdev_t dev = ap->a_head.a_dev;
691 	struct bpf_d *d = dev->si_drv1;
692 	int error = 0;
693 
694 	lwkt_gettoken(&bpf_token);
695 	if (d->bd_state == BPF_WAITING)
696 		callout_stop(&d->bd_callout);
697 	d->bd_state = BPF_IDLE;
698 
699 	if (d->bd_locked == 1) {
700 		switch (ap->a_cmd) {
701 		case BIOCGBLEN:
702 		case BIOCFLUSH:
703 		case BIOCGDLT:
704 		case BIOCGDLTLIST:
705 		case BIOCGETIF:
706 		case BIOCGRTIMEOUT:
707 		case BIOCGSTATS:
708 		case BIOCVERSION:
709 		case BIOCGRSIG:
710 		case BIOCGHDRCMPLT:
711 		case FIONREAD:
712 		case BIOCLOCK:
713 		case BIOCSRTIMEOUT:
714 		case BIOCIMMEDIATE:
715 		case TIOCGPGRP:
716 			break;
717 		default:
718 			lwkt_reltoken(&bpf_token);
719 			return (EPERM);
720 		}
721 	}
722 	switch (ap->a_cmd) {
723 	default:
724 		error = EINVAL;
725 		break;
726 
727 	/*
728 	 * Check for read packet available.
729 	 */
730 	case FIONREAD:
731 		{
732 			int n;
733 
734 			n = d->bd_slen;
735 			if (d->bd_hbuf)
736 				n += d->bd_hlen;
737 
738 			*(int *)ap->a_data = n;
739 			break;
740 		}
741 
742 	case SIOCGIFADDR:
743 		{
744 			struct ifnet *ifp;
745 
746 			if (d->bd_bif == NULL) {
747 				error = EINVAL;
748 			} else {
749 				ifp = d->bd_bif->bif_ifp;
750 				ifnet_serialize_all(ifp);
751 				error = ifp->if_ioctl(ifp, ap->a_cmd,
752 						      ap->a_data, ap->a_cred);
753 				ifnet_deserialize_all(ifp);
754 			}
755 			break;
756 		}
757 
758 	/*
759 	 * Get buffer len [for read()].
760 	 */
761 	case BIOCGBLEN:
762 		*(u_int *)ap->a_data = d->bd_bufsize;
763 		break;
764 
765 	/*
766 	 * Set buffer length.
767 	 */
768 	case BIOCSBLEN:
769 		if (d->bd_bif != NULL) {
770 			error = EINVAL;
771 		} else {
772 			u_int size = *(u_int *)ap->a_data;
773 
774 			if (size > bpf_maxbufsize)
775 				*(u_int *)ap->a_data = size = bpf_maxbufsize;
776 			else if (size < BPF_MINBUFSIZE)
777 				*(u_int *)ap->a_data = size = BPF_MINBUFSIZE;
778 			d->bd_bufsize = size;
779 		}
780 		break;
781 
782 	/*
783 	 * Set link layer read filter.
784 	 */
785 	case BIOCSETF:
786 	case BIOCSETWF:
787 		error = bpf_setf(d, (struct bpf_program *)ap->a_data,
788 			ap->a_cmd);
789 		break;
790 
791 	/*
792 	 * Flush read packet buffer.
793 	 */
794 	case BIOCFLUSH:
795 		bpf_resetd(d);
796 		break;
797 
798 	/*
799 	 * Put interface into promiscuous mode.
800 	 */
801 	case BIOCPROMISC:
802 		if (d->bd_bif == NULL) {
803 			/*
804 			 * No interface attached yet.
805 			 */
806 			error = EINVAL;
807 			break;
808 		}
809 		if (d->bd_promisc == 0) {
810 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
811 			if (error == 0)
812 				d->bd_promisc = 1;
813 		}
814 		break;
815 
816 	/*
817 	 * Get device parameters.
818 	 */
819 	case BIOCGDLT:
820 		if (d->bd_bif == NULL)
821 			error = EINVAL;
822 		else
823 			*(u_int *)ap->a_data = d->bd_bif->bif_dlt;
824 		break;
825 
826 	/*
827 	 * Get a list of supported data link types.
828 	 */
829 	case BIOCGDLTLIST:
830 		if (d->bd_bif == NULL) {
831 			error = EINVAL;
832 		} else {
833 			error = bpf_getdltlist(d,
834 				(struct bpf_dltlist *)ap->a_data);
835 		}
836 		break;
837 
838 	/*
839 	 * Set data link type.
840 	 */
841 	case BIOCSDLT:
842 		if (d->bd_bif == NULL)
843 			error = EINVAL;
844 		else
845 			error = bpf_setdlt(d, *(u_int *)ap->a_data);
846 		break;
847 
848 	/*
849 	 * Get interface name.
850 	 */
851 	case BIOCGETIF:
852 		if (d->bd_bif == NULL) {
853 			error = EINVAL;
854 		} else {
855 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
856 			struct ifreq *const ifr = (struct ifreq *)ap->a_data;
857 
858 			strlcpy(ifr->ifr_name, ifp->if_xname,
859 				sizeof ifr->ifr_name);
860 		}
861 		break;
862 
863 	/*
864 	 * Set interface.
865 	 */
866 	case BIOCSETIF:
867 		error = bpf_setif(d, (struct ifreq *)ap->a_data);
868 		break;
869 
870 	/*
871 	 * Set read timeout.
872 	 */
873 	case BIOCSRTIMEOUT:
874 		{
875 			struct timeval *tv = (struct timeval *)ap->a_data;
876 
877 			/*
878 			 * Subtract 1 tick from tvtohz() since this isn't
879 			 * a one-shot timer.
880 			 */
881 			if ((error = itimerfix(tv)) == 0)
882 				d->bd_rtout = tvtohz_low(tv);
883 			break;
884 		}
885 
886 	/*
887 	 * Get read timeout.
888 	 */
889 	case BIOCGRTIMEOUT:
890 		{
891 			struct timeval *tv = (struct timeval *)ap->a_data;
892 
893 			tv->tv_sec = d->bd_rtout / hz;
894 			tv->tv_usec = (d->bd_rtout % hz) * ustick;
895 			break;
896 		}
897 
898 	/*
899 	 * Get packet stats.
900 	 */
901 	case BIOCGSTATS:
902 		{
903 			struct bpf_stat *bs = (struct bpf_stat *)ap->a_data;
904 
905 			bs->bs_recv = d->bd_rcount;
906 			bs->bs_drop = d->bd_dcount;
907 			break;
908 		}
909 
910 	/*
911 	 * Set immediate mode.
912 	 */
913 	case BIOCIMMEDIATE:
914 		d->bd_immediate = *(u_int *)ap->a_data;
915 		break;
916 
917 	case BIOCVERSION:
918 		{
919 			struct bpf_version *bv = (struct bpf_version *)ap->a_data;
920 
921 			bv->bv_major = BPF_MAJOR_VERSION;
922 			bv->bv_minor = BPF_MINOR_VERSION;
923 			break;
924 		}
925 
926 	/*
927 	 * Get "header already complete" flag
928 	 */
929 	case BIOCGHDRCMPLT:
930 		*(u_int *)ap->a_data = d->bd_hdrcmplt;
931 		break;
932 
933 	/*
934 	 * Set "header already complete" flag
935 	 */
936 	case BIOCSHDRCMPLT:
937 		d->bd_hdrcmplt = *(u_int *)ap->a_data ? 1 : 0;
938 		break;
939 
940 	/*
941 	 * Get "see sent packets" flag
942 	 */
943 	case BIOCGSEESENT:
944 		*(u_int *)ap->a_data = d->bd_seesent;
945 		break;
946 
947 	/*
948 	 * Set "see sent packets" flag
949 	 */
950 	case BIOCSSEESENT:
951 		d->bd_seesent = *(u_int *)ap->a_data;
952 		break;
953 
954 	case FIOASYNC:		/* Send signal on receive packets */
955 		d->bd_async = *(int *)ap->a_data;
956 		break;
957 
958 	case FIOSETOWN:
959 		error = fsetown(*(int *)ap->a_data, &d->bd_sigio);
960 		break;
961 
962 	case FIOGETOWN:
963 		*(int *)ap->a_data = fgetown(&d->bd_sigio);
964 		break;
965 
966 	/* This is deprecated, FIOSETOWN should be used instead. */
967 	case TIOCSPGRP:
968 		error = fsetown(-(*(int *)ap->a_data), &d->bd_sigio);
969 		break;
970 
971 	/* This is deprecated, FIOGETOWN should be used instead. */
972 	case TIOCGPGRP:
973 		*(int *)ap->a_data = -fgetown(&d->bd_sigio);
974 		break;
975 
976 	case BIOCSRSIG:		/* Set receive signal */
977 		{
978 			u_int sig;
979 
980 			sig = *(u_int *)ap->a_data;
981 
982 			if (sig >= NSIG)
983 				error = EINVAL;
984 			else
985 				d->bd_sig = sig;
986 			break;
987 		}
988 	case BIOCGRSIG:
989 		*(u_int *)ap->a_data = d->bd_sig;
990 		break;
991 	case BIOCLOCK:
992 		d->bd_locked = 1;
993 		break;
994 	}
995 	lwkt_reltoken(&bpf_token);
996 
997 	return(error);
998 }
999 
1000 /*
1001  * Set d's packet filter program to fp.  If this file already has a filter,
1002  * free it and replace it.  Returns EINVAL for bogus requests.
1003  */
1004 static int
1005 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1006 {
1007 	struct bpf_insn *fcode, *old;
1008 	u_int wfilter, flen, size;
1009 
1010 	if (cmd == BIOCSETWF) {
1011 		old = d->bd_wfilter;
1012 		wfilter = 1;
1013 	} else {
1014 		wfilter = 0;
1015 		old = d->bd_rfilter;
1016 	}
1017 	if (fp->bf_insns == NULL) {
1018 		if (fp->bf_len != 0)
1019 			return(EINVAL);
1020 		if (wfilter)
1021 			d->bd_wfilter = NULL;
1022 		else
1023 			d->bd_rfilter = NULL;
1024 		bpf_resetd(d);
1025 		if (old != NULL)
1026 			kfree(old, M_BPF);
1027 		return(0);
1028 	}
1029 	flen = fp->bf_len;
1030 	if (flen > BPF_MAXINSNS)
1031 		return(EINVAL);
1032 
1033 	size = flen * sizeof *fp->bf_insns;
1034 	fcode = (struct bpf_insn *)kmalloc(size, M_BPF, M_WAITOK);
1035 	if (copyin(fp->bf_insns, fcode, size) == 0 &&
1036 	    bpf_validate(fcode, (int)flen)) {
1037 		if (wfilter)
1038 			d->bd_wfilter = fcode;
1039 		else
1040 			d->bd_rfilter = fcode;
1041 		bpf_resetd(d);
1042 		if (old != NULL)
1043 			kfree(old, M_BPF);
1044 
1045 		return(0);
1046 	}
1047 	kfree(fcode, M_BPF);
1048 	return(EINVAL);
1049 }
1050 
1051 /*
1052  * Detach a file from its current interface (if attached at all) and attach
1053  * to the interface indicated by the name stored in ifr.
1054  * Return an errno or 0.
1055  */
1056 static int
1057 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1058 {
1059 	struct bpf_if *bp;
1060 	int error;
1061 	struct ifnet *theywant;
1062 
1063 	ifnet_lock();
1064 
1065 	theywant = ifunit(ifr->ifr_name);
1066 	if (theywant == NULL) {
1067 		ifnet_unlock();
1068 		return(ENXIO);
1069 	}
1070 
1071 	/*
1072 	 * Look through attached interfaces for the named one.
1073 	 */
1074 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1075 		struct ifnet *ifp = bp->bif_ifp;
1076 
1077 		if (ifp == NULL || ifp != theywant)
1078 			continue;
1079 		/* skip additional entry */
1080 		if (bp->bif_driverp != &ifp->if_bpf)
1081 			continue;
1082 		/*
1083 		 * We found the requested interface.
1084 		 * Allocate the packet buffers if we need to.
1085 		 * If we're already attached to requested interface,
1086 		 * just flush the buffer.
1087 		 */
1088 		if (d->bd_sbuf == NULL) {
1089 			error = bpf_allocbufs(d);
1090 			if (error != 0) {
1091 				ifnet_unlock();
1092 				return(error);
1093 			}
1094 		}
1095 		if (bp != d->bd_bif) {
1096 			if (d->bd_bif != NULL) {
1097 				/*
1098 				 * Detach if attached to something else.
1099 				 */
1100 				bpf_detachd(d);
1101 			}
1102 
1103 			bpf_attachd(d, bp);
1104 		}
1105 		bpf_resetd(d);
1106 
1107 		ifnet_unlock();
1108 		return(0);
1109 	}
1110 
1111 	ifnet_unlock();
1112 
1113 	/* Not found. */
1114 	return(ENXIO);
1115 }
1116 
1117 static struct filterops bpf_read_filtops =
1118 	{ FILTEROP_ISFD, NULL, bpf_filter_detach, bpf_filter_read };
1119 
1120 static int
1121 bpfkqfilter(struct dev_kqfilter_args *ap)
1122 {
1123 	cdev_t dev = ap->a_head.a_dev;
1124 	struct knote *kn = ap->a_kn;
1125 	struct klist *klist;
1126 	struct bpf_d *d;
1127 
1128 	lwkt_gettoken(&bpf_token);
1129 	d = dev->si_drv1;
1130 	if (d->bd_bif == NULL) {
1131 		ap->a_result = 1;
1132 		lwkt_reltoken(&bpf_token);
1133 		return (0);
1134 	}
1135 
1136 	ap->a_result = 0;
1137 	switch (kn->kn_filter) {
1138 	case EVFILT_READ:
1139 		kn->kn_fop = &bpf_read_filtops;
1140 		kn->kn_hook = (caddr_t)d;
1141 		break;
1142 	default:
1143 		ap->a_result = EOPNOTSUPP;
1144 		lwkt_reltoken(&bpf_token);
1145 		return (0);
1146 	}
1147 
1148 	klist = &d->bd_kq.ki_note;
1149 	knote_insert(klist, kn);
1150 	lwkt_reltoken(&bpf_token);
1151 
1152 	return (0);
1153 }
1154 
1155 static void
1156 bpf_filter_detach(struct knote *kn)
1157 {
1158 	struct klist *klist;
1159 	struct bpf_d *d;
1160 
1161 	d = (struct bpf_d *)kn->kn_hook;
1162 	klist = &d->bd_kq.ki_note;
1163 	knote_remove(klist, kn);
1164 }
1165 
1166 static int
1167 bpf_filter_read(struct knote *kn, long hint)
1168 {
1169 	struct bpf_d *d;
1170 	int ready = 0;
1171 
1172 	d = (struct bpf_d *)kn->kn_hook;
1173 	if (d->bd_hlen != 0 ||
1174 	    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1175 	    d->bd_slen != 0)) {
1176 		ready = 1;
1177 	} else {
1178 		/* Start the read timeout if necessary. */
1179 		if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1180 			callout_reset(&d->bd_callout, d->bd_rtout,
1181 			    bpf_timed_out, d);
1182 			d->bd_state = BPF_WAITING;
1183 		}
1184 	}
1185 
1186 	return (ready);
1187 }
1188 
1189 
1190 /*
1191  * Process the packet pkt of length pktlen.  The packet is parsed
1192  * by each listener's filter, and if accepted, stashed into the
1193  * corresponding buffer.
1194  */
1195 void
1196 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1197 {
1198 	struct bpf_d *d;
1199 	struct timeval tv;
1200 	int gottime = 0;
1201 	u_int slen;
1202 
1203 	lwkt_gettoken(&bpf_token);
1204 	/* Re-check */
1205 	if (bp == NULL) {
1206 		lwkt_reltoken(&bpf_token);
1207 		return;
1208 	}
1209 
1210 	/*
1211 	 * Note that the ipl does not have to be raised at this point.
1212 	 * The only problem that could arise here is that if two different
1213 	 * interfaces shared any data.  This is not the case.
1214 	 */
1215 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1216 		++d->bd_rcount;
1217 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1218 		if (slen != 0) {
1219 			if (!gottime) {
1220 				microtime(&tv);
1221 				gottime = 1;
1222 			}
1223 			catchpacket(d, pkt, pktlen, slen, bcopy, &tv);
1224 		}
1225 	}
1226 	lwkt_reltoken(&bpf_token);
1227 }
1228 
1229 /*
1230  * Copy data from an mbuf chain into a buffer.  This code is derived
1231  * from m_copydata in sys/uipc_mbuf.c.
1232  */
1233 static void
1234 bpf_mcopy(volatile const void *src_arg, volatile void *dst_arg, size_t len)
1235 {
1236 	volatile const struct mbuf *m;
1237 	u_int count;
1238 	volatile u_char *dst;
1239 
1240 	m = src_arg;
1241 	dst = dst_arg;
1242 	while (len > 0) {
1243 		if (m == NULL)
1244 			panic("bpf_mcopy");
1245 		count = min(m->m_len, len);
1246 		bcopy(mtod(m, void *), dst, count);
1247 		m = m->m_next;
1248 		dst += count;
1249 		len -= count;
1250 	}
1251 }
1252 
1253 /*
1254  * Process the packet in the mbuf chain m.  The packet is parsed by each
1255  * listener's filter, and if accepted, stashed into the corresponding
1256  * buffer.
1257  */
1258 void
1259 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1260 {
1261 	struct bpf_d *d;
1262 	u_int pktlen, slen;
1263 	struct timeval tv;
1264 	int gottime = 0;
1265 
1266 	lwkt_gettoken(&bpf_token);
1267 	/* Re-check */
1268 	if (bp == NULL) {
1269 		lwkt_reltoken(&bpf_token);
1270 		return;
1271 	}
1272 
1273 	/* Don't compute pktlen, if no descriptor is attached. */
1274 	if (SLIST_EMPTY(&bp->bif_dlist)) {
1275 		lwkt_reltoken(&bpf_token);
1276 		return;
1277 	}
1278 
1279 	pktlen = m_lengthm(m, NULL);
1280 
1281 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1282 		if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL))
1283 			continue;
1284 		++d->bd_rcount;
1285 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1286 		if (slen != 0) {
1287 			if (!gottime) {
1288 				microtime(&tv);
1289 				gottime = 1;
1290 			}
1291 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy,
1292 				    &tv);
1293 		}
1294 	}
1295 	lwkt_reltoken(&bpf_token);
1296 }
1297 
1298 /*
1299  * Incoming linkage from device drivers, where we have a mbuf chain
1300  * but need to prepend some arbitrary header from a linear buffer.
1301  *
1302  * Con up a minimal dummy header to pacify bpf.  Allocate (only) a
1303  * struct m_hdr on the stack.  This is safe as bpf only reads from the
1304  * fields in this header that we initialize, and will not try to free
1305  * it or keep a pointer to it.
1306  */
1307 void
1308 bpf_mtap_hdr(struct bpf_if *arg, caddr_t data, u_int dlen, struct mbuf *m,
1309     u_int direction)
1310 {
1311 	struct m_hdr mh;
1312 
1313 	mh.mh_flags = 0;
1314 	mh.mh_next = m;
1315 	mh.mh_len = dlen;
1316 	mh.mh_data = data;
1317 
1318 	bpf_mtap(arg, (struct mbuf *) &mh);
1319 }
1320 
1321 void
1322 bpf_mtap_family(struct bpf_if *bp, struct mbuf *m, sa_family_t family)
1323 {
1324 	u_int family4;
1325 
1326 	KKASSERT(family != AF_UNSPEC);
1327 
1328 	family4 = (u_int)family;
1329 	bpf_ptap(bp, m, &family4, sizeof(family4));
1330 }
1331 
1332 /*
1333  * Process the packet in the mbuf chain m with the header in m prepended.
1334  * The packet is parsed by each listener's filter, and if accepted,
1335  * stashed into the corresponding buffer.
1336  */
1337 void
1338 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1339 {
1340 	struct mbuf mb;
1341 
1342 	/*
1343 	 * Craft on-stack mbuf suitable for passing to bpf_mtap.
1344 	 * Note that we cut corners here; we only setup what's
1345 	 * absolutely needed--this mbuf should never go anywhere else.
1346 	 */
1347 	mb.m_next = m;
1348 	mb.m_data = __DECONST(void *, data); /* LINTED */
1349 	mb.m_len = dlen;
1350 	mb.m_pkthdr.rcvif = m->m_pkthdr.rcvif;
1351 
1352 	bpf_mtap(bp, &mb);
1353 }
1354 
1355 /*
1356  * Move the packet data from interface memory (pkt) into the
1357  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1358  * otherwise 0.  "copy" is the routine called to do the actual data
1359  * transfer.  bcopy is passed in to copy contiguous chunks, while
1360  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1361  * pkt is really an mbuf.
1362  */
1363 static void
1364 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1365 	    void (*cpfn)(volatile const void *, volatile void *, size_t),
1366 	    const struct timeval *tv)
1367 {
1368 	struct bpf_hdr *hp;
1369 	int totlen, curlen;
1370 	int hdrlen = d->bd_bif->bif_hdrlen;
1371 	int wakeup = 0;
1372 	/*
1373 	 * Figure out how many bytes to move.  If the packet is
1374 	 * greater or equal to the snapshot length, transfer that
1375 	 * much.  Otherwise, transfer the whole packet (unless
1376 	 * we hit the buffer size limit).
1377 	 */
1378 	totlen = hdrlen + min(snaplen, pktlen);
1379 	if (totlen > d->bd_bufsize)
1380 		totlen = d->bd_bufsize;
1381 
1382 	/*
1383 	 * Round up the end of the previous packet to the next longword.
1384 	 */
1385 	curlen = BPF_WORDALIGN(d->bd_slen);
1386 	if (curlen + totlen > d->bd_bufsize) {
1387 		/*
1388 		 * This packet will overflow the storage buffer.
1389 		 * Rotate the buffers if we can, then wakeup any
1390 		 * pending reads.
1391 		 */
1392 		if (d->bd_fbuf == NULL) {
1393 			/*
1394 			 * We haven't completed the previous read yet,
1395 			 * so drop the packet.
1396 			 */
1397 			++d->bd_dcount;
1398 			return;
1399 		}
1400 		ROTATE_BUFFERS(d);
1401 		wakeup = 1;
1402 		curlen = 0;
1403 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1404 		/*
1405 		 * Immediate mode is set, or the read timeout has
1406 		 * already expired during a select call.  A packet
1407 		 * arrived, so the reader should be woken up.
1408 		 */
1409 		wakeup = 1;
1410 	}
1411 
1412 	/*
1413 	 * Append the bpf header.
1414 	 */
1415 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1416 	hp->bh_tstamp = *tv;
1417 	hp->bh_datalen = pktlen;
1418 	hp->bh_hdrlen = hdrlen;
1419 	/*
1420 	 * Copy the packet data into the store buffer and update its length.
1421 	 */
1422 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1423 	d->bd_slen = curlen + totlen;
1424 
1425 	if (wakeup)
1426 		bpf_wakeup(d);
1427 }
1428 
1429 /*
1430  * Initialize all nonzero fields of a descriptor.
1431  */
1432 static int
1433 bpf_allocbufs(struct bpf_d *d)
1434 {
1435 	d->bd_fbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1436 	d->bd_sbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1437 	d->bd_slen = 0;
1438 	d->bd_hlen = 0;
1439 	return(0);
1440 }
1441 
1442 /*
1443  * Free buffers and packet filter program currently in use by a descriptor.
1444  * Called on close.
1445  */
1446 static void
1447 bpf_freed(struct bpf_d *d)
1448 {
1449 	/*
1450 	 * We don't need to lock out interrupts since this descriptor has
1451 	 * been detached from its interface and it yet hasn't been marked
1452 	 * free.
1453 	 */
1454 	if (d->bd_sbuf != NULL) {
1455 		kfree(d->bd_sbuf, M_BPF);
1456 		if (d->bd_hbuf != NULL)
1457 			kfree(d->bd_hbuf, M_BPF);
1458 		if (d->bd_fbuf != NULL)
1459 			kfree(d->bd_fbuf, M_BPF);
1460 	}
1461 	if (d->bd_rfilter)
1462 		kfree(d->bd_rfilter, M_BPF);
1463 	if (d->bd_wfilter)
1464 		kfree(d->bd_wfilter, M_BPF);
1465 }
1466 
1467 /*
1468  * Attach an interface to bpf.  ifp is a pointer to the structure
1469  * defining the interface to be attached, dlt is the link layer type,
1470  * and hdrlen is the fixed size of the link header (variable length
1471  * headers are not yet supported).
1472  */
1473 void
1474 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1475 {
1476 	bpfattach_dlt(ifp, dlt, hdrlen, &ifp->if_bpf);
1477 }
1478 
1479 void
1480 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1481 {
1482 	struct bpf_if *bp;
1483 
1484 	bp = kmalloc(sizeof *bp, M_BPF, M_WAITOK | M_ZERO);
1485 
1486 	lwkt_gettoken(&bpf_token);
1487 
1488 	SLIST_INIT(&bp->bif_dlist);
1489 	bp->bif_ifp = ifp;
1490 	bp->bif_dlt = dlt;
1491 	bp->bif_driverp = driverp;
1492 	*bp->bif_driverp = NULL;
1493 
1494 	bp->bif_next = bpf_iflist;
1495 	bpf_iflist = bp;
1496 
1497 	/*
1498 	 * Compute the length of the bpf header.  This is not necessarily
1499 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1500 	 * that the network layer header begins on a longword boundary (for
1501 	 * performance reasons and to alleviate alignment restrictions).
1502 	 */
1503 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1504 
1505 	lwkt_reltoken(&bpf_token);
1506 
1507 	if (bootverbose)
1508 		if_printf(ifp, "bpf attached\n");
1509 }
1510 
1511 /*
1512  * Detach bpf from an interface.  This involves detaching each descriptor
1513  * associated with the interface, and leaving bd_bif NULL.  Notify each
1514  * descriptor as it's detached so that any sleepers wake up and get
1515  * ENXIO.
1516  */
1517 void
1518 bpfdetach(struct ifnet *ifp)
1519 {
1520 	struct bpf_if *bp, *bp_prev;
1521 	struct bpf_d *d;
1522 
1523 	lwkt_gettoken(&bpf_token);
1524 
1525 	/* Locate BPF interface information */
1526 	bp_prev = NULL;
1527 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1528 		if (ifp == bp->bif_ifp)
1529 			break;
1530 		bp_prev = bp;
1531 	}
1532 
1533 	/* Interface wasn't attached */
1534 	if (bp->bif_ifp == NULL) {
1535 		lwkt_reltoken(&bpf_token);
1536 		kprintf("bpfdetach: %s was not attached\n", ifp->if_xname);
1537 		return;
1538 	}
1539 
1540 	while ((d = SLIST_FIRST(&bp->bif_dlist)) != NULL) {
1541 		bpf_detachd(d);
1542 		bpf_wakeup(d);
1543 	}
1544 
1545 	if (bp_prev != NULL)
1546 		bp_prev->bif_next = bp->bif_next;
1547 	else
1548 		bpf_iflist = bp->bif_next;
1549 
1550 	kfree(bp, M_BPF);
1551 
1552 	lwkt_reltoken(&bpf_token);
1553 }
1554 
1555 /*
1556  * Get a list of available data link type of the interface.
1557  */
1558 static int
1559 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1560 {
1561 	int n, error;
1562 	struct ifnet *ifp;
1563 	struct bpf_if *bp;
1564 
1565 	ifp = d->bd_bif->bif_ifp;
1566 	n = 0;
1567 	error = 0;
1568 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1569 		if (bp->bif_ifp != ifp)
1570 			continue;
1571 		if (bfl->bfl_list != NULL) {
1572 			if (n >= bfl->bfl_len) {
1573 				return (ENOMEM);
1574 			}
1575 			error = copyout(&bp->bif_dlt,
1576 			    bfl->bfl_list + n, sizeof(u_int));
1577 		}
1578 		n++;
1579 	}
1580 	bfl->bfl_len = n;
1581 	return(error);
1582 }
1583 
1584 /*
1585  * Set the data link type of a BPF instance.
1586  */
1587 static int
1588 bpf_setdlt(struct bpf_d *d, u_int dlt)
1589 {
1590 	int error, opromisc;
1591 	struct ifnet *ifp;
1592 	struct bpf_if *bp;
1593 
1594 	if (d->bd_bif->bif_dlt == dlt)
1595 		return (0);
1596 	ifp = d->bd_bif->bif_ifp;
1597 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1598 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1599 			break;
1600 	}
1601 	if (bp != NULL) {
1602 		opromisc = d->bd_promisc;
1603 		bpf_detachd(d);
1604 		bpf_attachd(d, bp);
1605 		bpf_resetd(d);
1606 		if (opromisc) {
1607 			error = ifpromisc(bp->bif_ifp, 1);
1608 			if (error) {
1609 				if_printf(bp->bif_ifp,
1610 					"bpf_setdlt: ifpromisc failed (%d)\n",
1611 					error);
1612 			} else {
1613 				d->bd_promisc = 1;
1614 			}
1615 		}
1616 	}
1617 	return(bp == NULL ? EINVAL : 0);
1618 }
1619 
1620 void
1621 bpf_gettoken(void)
1622 {
1623 	lwkt_gettoken(&bpf_token);
1624 }
1625 
1626 void
1627 bpf_reltoken(void)
1628 {
1629 	lwkt_reltoken(&bpf_token);
1630 }
1631 
1632 static void
1633 bpf_drvinit(void *unused)
1634 {
1635 	int i;
1636 
1637 	make_autoclone_dev(&bpf_ops, &DEVFS_CLONE_BITMAP(bpf),
1638 		bpfclone, 0, 0, 0600, "bpf");
1639 	for (i = 0; i < BPF_PREALLOCATED_UNITS; i++) {
1640 		make_dev(&bpf_ops, i, 0, 0, 0600, "bpf%d", i);
1641 		devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(bpf), i);
1642 	}
1643 }
1644 
1645 static void
1646 bpf_drvuninit(void *unused)
1647 {
1648 	devfs_clone_handler_del("bpf");
1649 	dev_ops_remove_all(&bpf_ops);
1650 	devfs_clone_bitmap_uninit(&DEVFS_CLONE_BITMAP(bpf));
1651 }
1652 
1653 SYSINIT(bpfdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR, bpf_drvinit, NULL);
1654 SYSUNINIT(bpfdev, SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvuninit, NULL);
1655 
1656 #else /* !BPF */
1657 /*
1658  * NOP stubs to allow bpf-using drivers to load and function.
1659  *
1660  * A 'better' implementation would allow the core bpf functionality
1661  * to be loaded at runtime.
1662  */
1663 
1664 void
1665 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1666 {
1667 }
1668 
1669 void
1670 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1671 {
1672 }
1673 
1674 void
1675 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1676 {
1677 }
1678 
1679 void
1680 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1681 {
1682 }
1683 
1684 void
1685 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1686 {
1687 }
1688 
1689 void
1690 bpfdetach(struct ifnet *ifp)
1691 {
1692 }
1693 
1694 u_int
1695 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
1696 {
1697 	return -1;	/* "no filter" behaviour */
1698 }
1699 
1700 void
1701 bpf_gettoken(void)
1702 {
1703 }
1704 
1705 void
1706 bpf_reltoken(void)
1707 {
1708 }
1709 
1710 #endif /* !BPF */
1711