xref: /dragonfly/sys/net/bpf.c (revision 07a2f99c)
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  */
42 
43 #include "use_bpf.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/conf.h>
48 #include <sys/device.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/filedesc.h>
58 
59 #include <sys/event.h>
60 
61 #include <sys/socket.h>
62 #include <sys/vnode.h>
63 
64 #include <sys/thread2.h>
65 
66 #include <net/if.h>
67 #include <net/bpf.h>
68 #include <net/bpfdesc.h>
69 #include <net/netmsg2.h>
70 
71 #include <netinet/in.h>
72 #include <netinet/if_ether.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 
76 #include <sys/devfs.h>
77 
78 struct netmsg_bpf_output {
79 	struct netmsg_base base;
80 	struct mbuf	*nm_mbuf;
81 	struct ifnet	*nm_ifp;
82 	struct sockaddr	*nm_dst;
83 };
84 
85 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
86 DEVFS_DECLARE_CLONE_BITMAP(bpf);
87 
88 #if NBPF <= 1
89 #define BPF_PREALLOCATED_UNITS	4
90 #else
91 #define BPF_PREALLOCATED_UNITS	NBPF
92 #endif
93 
94 #if NBPF > 0
95 
96 /*
97  * The default read buffer size is patchable.
98  */
99 static int bpf_bufsize = BPF_DEFAULTBUFSIZE;
100 SYSCTL_INT(_debug, OID_AUTO, bpf_bufsize, CTLFLAG_RW,
101    &bpf_bufsize, 0, "Current size of bpf buffer");
102 int bpf_maxbufsize = BPF_MAXBUFSIZE;
103 SYSCTL_INT(_debug, OID_AUTO, bpf_maxbufsize, CTLFLAG_RW,
104    &bpf_maxbufsize, 0, "Maximum size of bpf buffer");
105 
106 /*
107  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
108  */
109 static struct bpf_if	*bpf_iflist;
110 
111 static struct lwkt_token bpf_token = LWKT_TOKEN_INITIALIZER(bpf_token);
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", 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 	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 	lwkt_gettoken(&bpf_token);
268 	d->bd_bif = bp;
269 	SLIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
270 	*bp->bif_driverp = bp;
271 
272 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
273 	lwkt_reltoken(&bpf_token);
274 }
275 
276 /*
277  * Detach a file from its interface.
278  */
279 static void
280 bpf_detachd(struct bpf_d *d)
281 {
282 	int error;
283 	struct bpf_if *bp;
284 	struct ifnet *ifp;
285 
286 	lwkt_gettoken(&bpf_token);
287 	bp = d->bd_bif;
288 	ifp = bp->bif_ifp;
289 
290 	/* Remove d from the interface's descriptor list. */
291 	SLIST_REMOVE(&bp->bif_dlist, d, bpf_d, bd_next);
292 
293 	if (SLIST_EMPTY(&bp->bif_dlist)) {
294 		/*
295 		 * Let the driver know that there are no more listeners.
296 		 */
297 		*bp->bif_driverp = NULL;
298 	}
299 	d->bd_bif = NULL;
300 
301 	EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
302 
303 	/*
304 	 * Check if this descriptor had requested promiscuous mode.
305 	 * If so, turn it off.
306 	 */
307 	if (d->bd_promisc) {
308 		d->bd_promisc = 0;
309 		error = ifpromisc(ifp, 0);
310 		if (error != 0 && error != ENXIO) {
311 			/*
312 			 * ENXIO can happen if a pccard is unplugged,
313 			 * Something is really wrong if we were able to put
314 			 * the driver into promiscuous mode, but can't
315 			 * take it out.
316 			 */
317 			if_printf(ifp, "bpf_detach: ifpromisc failed(%d)\n",
318 				  error);
319 		}
320 	}
321 	lwkt_reltoken(&bpf_token);
322 }
323 
324 /*
325  * Open ethernet device.  Returns ENXIO for illegal minor device number,
326  * EBUSY if file is open by another process.
327  */
328 /* ARGSUSED */
329 static int
330 bpfopen(struct dev_open_args *ap)
331 {
332 	cdev_t dev = ap->a_head.a_dev;
333 	struct bpf_d *d;
334 
335 	lwkt_gettoken(&bpf_token);
336 	if (ap->a_cred->cr_prison) {
337 		lwkt_reltoken(&bpf_token);
338 		return(EPERM);
339 	}
340 
341 	d = dev->si_drv1;
342 	/*
343 	 * Each minor can be opened by only one process.  If the requested
344 	 * minor is in use, return EBUSY.
345 	 */
346 	if (d != NULL) {
347 		lwkt_reltoken(&bpf_token);
348 		return(EBUSY);
349 	}
350 
351 	d = kmalloc(sizeof *d, M_BPF, M_WAITOK | M_ZERO);
352 	dev->si_drv1 = d;
353 	d->bd_bufsize = bpf_bufsize;
354 	d->bd_sig = SIGIO;
355 	d->bd_seesent = 1;
356 	callout_init(&d->bd_callout);
357 	lwkt_reltoken(&bpf_token);
358 
359 	return(0);
360 }
361 
362 static int
363 bpfclone(struct dev_clone_args *ap)
364 {
365 	int unit;
366 
367 	unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(bpf), 0);
368 	ap->a_dev = make_only_dev(&bpf_ops, unit, 0, 0, 0600, "bpf%d", unit);
369 
370 	return 0;
371 }
372 
373 /*
374  * Close the descriptor by detaching it from its interface,
375  * deallocating its buffers, and marking it free.
376  */
377 /* ARGSUSED */
378 static int
379 bpfclose(struct dev_close_args *ap)
380 {
381 	cdev_t dev = ap->a_head.a_dev;
382 	struct bpf_d *d = dev->si_drv1;
383 
384 	lwkt_gettoken(&bpf_token);
385 	funsetown(&d->bd_sigio);
386 	crit_enter();
387 	if (d->bd_state == BPF_WAITING)
388 		callout_stop(&d->bd_callout);
389 	d->bd_state = BPF_IDLE;
390 	if (d->bd_bif != NULL)
391 		bpf_detachd(d);
392 	crit_exit();
393 	bpf_freed(d);
394 	dev->si_drv1 = NULL;
395 	if (dev->si_uminor >= BPF_PREALLOCATED_UNITS) {
396 		devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(bpf), dev->si_uminor);
397 		destroy_dev(dev);
398 	}
399 	kfree(d, M_BPF);
400 	lwkt_reltoken(&bpf_token);
401 
402 	return(0);
403 }
404 
405 /*
406  * Rotate the packet buffers in descriptor d.  Move the store buffer
407  * into the hold slot, and the free buffer into the store slot.
408  * Zero the length of the new store buffer.
409  */
410 #define ROTATE_BUFFERS(d) \
411 	(d)->bd_hbuf = (d)->bd_sbuf; \
412 	(d)->bd_hlen = (d)->bd_slen; \
413 	(d)->bd_sbuf = (d)->bd_fbuf; \
414 	(d)->bd_slen = 0; \
415 	(d)->bd_fbuf = NULL;
416 /*
417  *  bpfread - read next chunk of packets from buffers
418  */
419 static int
420 bpfread(struct dev_read_args *ap)
421 {
422 	cdev_t dev = ap->a_head.a_dev;
423 	struct bpf_d *d = dev->si_drv1;
424 	int timed_out;
425 	int error;
426 
427 	lwkt_gettoken(&bpf_token);
428 	/*
429 	 * Restrict application to use a buffer the same size as
430 	 * as kernel buffers.
431 	 */
432 	if (ap->a_uio->uio_resid != d->bd_bufsize) {
433 		lwkt_reltoken(&bpf_token);
434 		return(EINVAL);
435 	}
436 
437 	crit_enter();
438 	if (d->bd_state == BPF_WAITING)
439 		callout_stop(&d->bd_callout);
440 	timed_out = (d->bd_state == BPF_TIMED_OUT);
441 	d->bd_state = BPF_IDLE;
442 	/*
443 	 * If the hold buffer is empty, then do a timed sleep, which
444 	 * ends when the timeout expires or when enough packets
445 	 * have arrived to fill the store buffer.
446 	 */
447 	while (d->bd_hbuf == NULL) {
448 		if ((d->bd_immediate || (ap->a_ioflag & IO_NDELAY) || timed_out)
449 		    && d->bd_slen != 0) {
450 			/*
451 			 * A packet(s) either arrived since the previous,
452 			 * We're in immediate mode, or are reading
453 			 * in non-blocking mode, and a packet(s)
454 			 * either arrived since the previous
455 			 * read or arrived while we were asleep.
456 			 * Rotate the buffers and return what's here.
457 			 */
458 			ROTATE_BUFFERS(d);
459 			break;
460 		}
461 
462 		/*
463 		 * No data is available, check to see if the bpf device
464 		 * is still pointed at a real interface.  If not, return
465 		 * ENXIO so that the userland process knows to rebind
466 		 * it before using it again.
467 		 */
468 		if (d->bd_bif == NULL) {
469 			crit_exit();
470 			lwkt_reltoken(&bpf_token);
471 			return(ENXIO);
472 		}
473 
474 		if (ap->a_ioflag & IO_NDELAY) {
475 			crit_exit();
476 			lwkt_reltoken(&bpf_token);
477 			return(EWOULDBLOCK);
478 		}
479 		error = tsleep(d, PCATCH, "bpf", d->bd_rtout);
480 		if (error == EINTR || error == ERESTART) {
481 			crit_exit();
482 			lwkt_reltoken(&bpf_token);
483 			return(error);
484 		}
485 		if (error == EWOULDBLOCK) {
486 			/*
487 			 * On a timeout, return what's in the buffer,
488 			 * which may be nothing.  If there is something
489 			 * in the store buffer, we can rotate the buffers.
490 			 */
491 			if (d->bd_hbuf)
492 				/*
493 				 * We filled up the buffer in between
494 				 * getting the timeout and arriving
495 				 * here, so we don't need to rotate.
496 				 */
497 				break;
498 
499 			if (d->bd_slen == 0) {
500 				crit_exit();
501 				lwkt_reltoken(&bpf_token);
502 				return(0);
503 			}
504 			ROTATE_BUFFERS(d);
505 			break;
506 		}
507 	}
508 	/*
509 	 * At this point, we know we have something in the hold slot.
510 	 */
511 	crit_exit();
512 
513 	/*
514 	 * Move data from hold buffer into user space.
515 	 * We know the entire buffer is transferred since
516 	 * we checked above that the read buffer is bpf_bufsize bytes.
517 	 */
518 	error = uiomove(d->bd_hbuf, d->bd_hlen, ap->a_uio);
519 
520 	crit_enter();
521 	d->bd_fbuf = d->bd_hbuf;
522 	d->bd_hbuf = NULL;
523 	d->bd_hlen = 0;
524 	crit_exit();
525 	lwkt_reltoken(&bpf_token);
526 
527 	return(error);
528 }
529 
530 
531 /*
532  * If there are processes sleeping on this descriptor, wake them up.
533  */
534 static void
535 bpf_wakeup(struct bpf_d *d)
536 {
537 	if (d->bd_state == BPF_WAITING) {
538 		callout_stop(&d->bd_callout);
539 		d->bd_state = BPF_IDLE;
540 	}
541 	wakeup(d);
542 	if (d->bd_async && d->bd_sig && d->bd_sigio)
543 		pgsigio(d->bd_sigio, d->bd_sig, 0);
544 
545 	KNOTE(&d->bd_kq.ki_note, 0);
546 }
547 
548 static void
549 bpf_timed_out(void *arg)
550 {
551 	struct bpf_d *d = (struct bpf_d *)arg;
552 
553 	crit_enter();
554 	if (d->bd_state == BPF_WAITING) {
555 		d->bd_state = BPF_TIMED_OUT;
556 		if (d->bd_slen != 0)
557 			bpf_wakeup(d);
558 	}
559 	crit_exit();
560 }
561 
562 static void
563 bpf_output_dispatch(netmsg_t msg)
564 {
565 	struct netmsg_bpf_output *bmsg = (struct netmsg_bpf_output *)msg;
566 	struct ifnet *ifp = bmsg->nm_ifp;
567 	int error;
568 
569 	/*
570 	 * The driver frees the mbuf.
571 	 */
572 	error = ifp->if_output(ifp, bmsg->nm_mbuf, bmsg->nm_dst, NULL);
573 	lwkt_replymsg(&msg->lmsg, error);
574 }
575 
576 static int
577 bpfwrite(struct dev_write_args *ap)
578 {
579 	cdev_t dev = ap->a_head.a_dev;
580 	struct bpf_d *d = dev->si_drv1;
581 	struct ifnet *ifp;
582 	struct mbuf *m;
583 	int error, ret;
584 	struct sockaddr dst;
585 	int datlen;
586 	struct netmsg_bpf_output bmsg;
587 
588 	lwkt_gettoken(&bpf_token);
589 	if (d->bd_bif == NULL) {
590 		lwkt_reltoken(&bpf_token);
591 		return(ENXIO);
592 	}
593 
594 	ifp = d->bd_bif->bif_ifp;
595 
596 	if (ap->a_uio->uio_resid == 0) {
597 		lwkt_reltoken(&bpf_token);
598 		return(0);
599 	}
600 
601 	error = bpf_movein(ap->a_uio, (int)d->bd_bif->bif_dlt, &m,
602 			   &dst, &datlen, d->bd_wfilter);
603 	if (error) {
604 		lwkt_reltoken(&bpf_token);
605 		return(error);
606 	}
607 
608 	if (datlen > ifp->if_mtu) {
609 		m_freem(m);
610 		lwkt_reltoken(&bpf_token);
611 		return(EMSGSIZE);
612 	}
613 
614 	if (d->bd_hdrcmplt)
615 		dst.sa_family = pseudo_AF_HDRCMPLT;
616 
617 	netmsg_init(&bmsg.base, NULL, &curthread->td_msgport,
618 		    0, bpf_output_dispatch);
619 	bmsg.nm_mbuf = m;
620 	bmsg.nm_ifp = ifp;
621 	bmsg.nm_dst = &dst;
622 
623 	ret = lwkt_domsg(netisr_portfn(0), &bmsg.base.lmsg, 0);
624 	lwkt_reltoken(&bpf_token);
625 
626 	return ret;
627 }
628 
629 /*
630  * Reset a descriptor by flushing its packet buffer and clearing the
631  * receive and drop counts.  Should be called at splimp.
632  */
633 static void
634 bpf_resetd(struct bpf_d *d)
635 {
636 	if (d->bd_hbuf) {
637 		/* Free the hold buffer. */
638 		d->bd_fbuf = d->bd_hbuf;
639 		d->bd_hbuf = NULL;
640 	}
641 	d->bd_slen = 0;
642 	d->bd_hlen = 0;
643 	d->bd_rcount = 0;
644 	d->bd_dcount = 0;
645 }
646 
647 /*
648  *  FIONREAD		Check for read packet available.
649  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
650  *  BIOCGBLEN		Get buffer len [for read()].
651  *  BIOCSETF		Set ethernet read filter.
652  *  BIOCSETWF		Set ethernet write filter.
653  *  BIOCFLUSH		Flush read packet buffer.
654  *  BIOCPROMISC		Put interface into promiscuous mode.
655  *  BIOCGDLT		Get link layer type.
656  *  BIOCGETIF		Get interface name.
657  *  BIOCSETIF		Set interface.
658  *  BIOCSRTIMEOUT	Set read timeout.
659  *  BIOCGRTIMEOUT	Get read timeout.
660  *  BIOCGSTATS		Get packet stats.
661  *  BIOCIMMEDIATE	Set immediate mode.
662  *  BIOCVERSION		Get filter language version.
663  *  BIOCGHDRCMPLT	Get "header already complete" flag
664  *  BIOCSHDRCMPLT	Set "header already complete" flag
665  *  BIOCGSEESENT	Get "see packets sent" flag
666  *  BIOCSSEESENT	Set "see packets sent" flag
667  *  BIOCLOCK		Set "locked" flag
668  */
669 /* ARGSUSED */
670 static int
671 bpfioctl(struct dev_ioctl_args *ap)
672 {
673 	cdev_t dev = ap->a_head.a_dev;
674 	struct bpf_d *d = dev->si_drv1;
675 	int error = 0;
676 
677 	lwkt_gettoken(&bpf_token);
678 	crit_enter();
679 	if (d->bd_state == BPF_WAITING)
680 		callout_stop(&d->bd_callout);
681 	d->bd_state = BPF_IDLE;
682 	crit_exit();
683 
684 	if (d->bd_locked == 1) {
685 		switch (ap->a_cmd) {
686 		case BIOCGBLEN:
687 		case BIOCFLUSH:
688 		case BIOCGDLT:
689 		case BIOCGDLTLIST:
690 		case BIOCGETIF:
691 		case BIOCGRTIMEOUT:
692 		case BIOCGSTATS:
693 		case BIOCVERSION:
694 		case BIOCGRSIG:
695 		case BIOCGHDRCMPLT:
696 		case FIONREAD:
697 		case BIOCLOCK:
698 		case BIOCSRTIMEOUT:
699 		case BIOCIMMEDIATE:
700 		case TIOCGPGRP:
701 			break;
702 		default:
703 			lwkt_reltoken(&bpf_token);
704 			return (EPERM);
705 		}
706 	}
707 	switch (ap->a_cmd) {
708 	default:
709 		error = EINVAL;
710 		break;
711 
712 	/*
713 	 * Check for read packet available.
714 	 */
715 	case FIONREAD:
716 		{
717 			int n;
718 
719 			crit_enter();
720 			n = d->bd_slen;
721 			if (d->bd_hbuf)
722 				n += d->bd_hlen;
723 			crit_exit();
724 
725 			*(int *)ap->a_data = n;
726 			break;
727 		}
728 
729 	case SIOCGIFADDR:
730 		{
731 			struct ifnet *ifp;
732 
733 			if (d->bd_bif == NULL) {
734 				error = EINVAL;
735 			} else {
736 				ifp = d->bd_bif->bif_ifp;
737 				ifnet_serialize_all(ifp);
738 				error = ifp->if_ioctl(ifp, ap->a_cmd,
739 						      ap->a_data, ap->a_cred);
740 				ifnet_deserialize_all(ifp);
741 			}
742 			break;
743 		}
744 
745 	/*
746 	 * Get buffer len [for read()].
747 	 */
748 	case BIOCGBLEN:
749 		*(u_int *)ap->a_data = d->bd_bufsize;
750 		break;
751 
752 	/*
753 	 * Set buffer length.
754 	 */
755 	case BIOCSBLEN:
756 		if (d->bd_bif != NULL) {
757 			error = EINVAL;
758 		} else {
759 			u_int size = *(u_int *)ap->a_data;
760 
761 			if (size > bpf_maxbufsize)
762 				*(u_int *)ap->a_data = size = bpf_maxbufsize;
763 			else if (size < BPF_MINBUFSIZE)
764 				*(u_int *)ap->a_data = size = BPF_MINBUFSIZE;
765 			d->bd_bufsize = size;
766 		}
767 		break;
768 
769 	/*
770 	 * Set link layer read filter.
771 	 */
772 	case BIOCSETF:
773 	case BIOCSETWF:
774 		error = bpf_setf(d, (struct bpf_program *)ap->a_data,
775 			ap->a_cmd);
776 		break;
777 
778 	/*
779 	 * Flush read packet buffer.
780 	 */
781 	case BIOCFLUSH:
782 		crit_enter();
783 		bpf_resetd(d);
784 		crit_exit();
785 		break;
786 
787 	/*
788 	 * Put interface into promiscuous mode.
789 	 */
790 	case BIOCPROMISC:
791 		if (d->bd_bif == NULL) {
792 			/*
793 			 * No interface attached yet.
794 			 */
795 			error = EINVAL;
796 			break;
797 		}
798 		crit_enter();
799 		if (d->bd_promisc == 0) {
800 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
801 			if (error == 0)
802 				d->bd_promisc = 1;
803 		}
804 		crit_exit();
805 		break;
806 
807 	/*
808 	 * Get device parameters.
809 	 */
810 	case BIOCGDLT:
811 		if (d->bd_bif == NULL)
812 			error = EINVAL;
813 		else
814 			*(u_int *)ap->a_data = d->bd_bif->bif_dlt;
815 		break;
816 
817 	/*
818 	 * Get a list of supported data link types.
819 	 */
820 	case BIOCGDLTLIST:
821 		if (d->bd_bif == NULL) {
822 			error = EINVAL;
823 		} else {
824 			error = bpf_getdltlist(d,
825 				(struct bpf_dltlist *)ap->a_data);
826 		}
827 		break;
828 
829 	/*
830 	 * Set data link type.
831 	 */
832 	case BIOCSDLT:
833 		if (d->bd_bif == NULL)
834 			error = EINVAL;
835 		else
836 			error = bpf_setdlt(d, *(u_int *)ap->a_data);
837 		break;
838 
839 	/*
840 	 * Get interface name.
841 	 */
842 	case BIOCGETIF:
843 		if (d->bd_bif == NULL) {
844 			error = EINVAL;
845 		} else {
846 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
847 			struct ifreq *const ifr = (struct ifreq *)ap->a_data;
848 
849 			strlcpy(ifr->ifr_name, ifp->if_xname,
850 				sizeof ifr->ifr_name);
851 		}
852 		break;
853 
854 	/*
855 	 * Set interface.
856 	 */
857 	case BIOCSETIF:
858 		error = bpf_setif(d, (struct ifreq *)ap->a_data);
859 		break;
860 
861 	/*
862 	 * Set read timeout.
863 	 */
864 	case BIOCSRTIMEOUT:
865 		{
866 			struct timeval *tv = (struct timeval *)ap->a_data;
867 
868 			/*
869 			 * Subtract 1 tick from tvtohz() since this isn't
870 			 * a one-shot timer.
871 			 */
872 			if ((error = itimerfix(tv)) == 0)
873 				d->bd_rtout = tvtohz_low(tv);
874 			break;
875 		}
876 
877 	/*
878 	 * Get read timeout.
879 	 */
880 	case BIOCGRTIMEOUT:
881 		{
882 			struct timeval *tv = (struct timeval *)ap->a_data;
883 
884 			tv->tv_sec = d->bd_rtout / hz;
885 			tv->tv_usec = (d->bd_rtout % hz) * ustick;
886 			break;
887 		}
888 
889 	/*
890 	 * Get packet stats.
891 	 */
892 	case BIOCGSTATS:
893 		{
894 			struct bpf_stat *bs = (struct bpf_stat *)ap->a_data;
895 
896 			bs->bs_recv = d->bd_rcount;
897 			bs->bs_drop = d->bd_dcount;
898 			break;
899 		}
900 
901 	/*
902 	 * Set immediate mode.
903 	 */
904 	case BIOCIMMEDIATE:
905 		d->bd_immediate = *(u_int *)ap->a_data;
906 		break;
907 
908 	case BIOCVERSION:
909 		{
910 			struct bpf_version *bv = (struct bpf_version *)ap->a_data;
911 
912 			bv->bv_major = BPF_MAJOR_VERSION;
913 			bv->bv_minor = BPF_MINOR_VERSION;
914 			break;
915 		}
916 
917 	/*
918 	 * Get "header already complete" flag
919 	 */
920 	case BIOCGHDRCMPLT:
921 		*(u_int *)ap->a_data = d->bd_hdrcmplt;
922 		break;
923 
924 	/*
925 	 * Set "header already complete" flag
926 	 */
927 	case BIOCSHDRCMPLT:
928 		d->bd_hdrcmplt = *(u_int *)ap->a_data ? 1 : 0;
929 		break;
930 
931 	/*
932 	 * Get "see sent packets" flag
933 	 */
934 	case BIOCGSEESENT:
935 		*(u_int *)ap->a_data = d->bd_seesent;
936 		break;
937 
938 	/*
939 	 * Set "see sent packets" flag
940 	 */
941 	case BIOCSSEESENT:
942 		d->bd_seesent = *(u_int *)ap->a_data;
943 		break;
944 
945 	case FIOASYNC:		/* Send signal on receive packets */
946 		d->bd_async = *(int *)ap->a_data;
947 		break;
948 
949 	case FIOSETOWN:
950 		error = fsetown(*(int *)ap->a_data, &d->bd_sigio);
951 		break;
952 
953 	case FIOGETOWN:
954 		*(int *)ap->a_data = fgetown(&d->bd_sigio);
955 		break;
956 
957 	/* This is deprecated, FIOSETOWN should be used instead. */
958 	case TIOCSPGRP:
959 		error = fsetown(-(*(int *)ap->a_data), &d->bd_sigio);
960 		break;
961 
962 	/* This is deprecated, FIOGETOWN should be used instead. */
963 	case TIOCGPGRP:
964 		*(int *)ap->a_data = -fgetown(&d->bd_sigio);
965 		break;
966 
967 	case BIOCSRSIG:		/* Set receive signal */
968 		{
969 			u_int sig;
970 
971 			sig = *(u_int *)ap->a_data;
972 
973 			if (sig >= NSIG)
974 				error = EINVAL;
975 			else
976 				d->bd_sig = sig;
977 			break;
978 		}
979 	case BIOCGRSIG:
980 		*(u_int *)ap->a_data = d->bd_sig;
981 		break;
982 	case BIOCLOCK:
983 		d->bd_locked = 1;
984 		break;
985 	}
986 	lwkt_reltoken(&bpf_token);
987 
988 	return(error);
989 }
990 
991 /*
992  * Set d's packet filter program to fp.  If this file already has a filter,
993  * free it and replace it.  Returns EINVAL for bogus requests.
994  */
995 static int
996 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
997 {
998 	struct bpf_insn *fcode, *old;
999 	u_int wfilter, flen, size;
1000 
1001 	if (cmd == BIOCSETWF) {
1002 		old = d->bd_wfilter;
1003 		wfilter = 1;
1004 	} else {
1005 		wfilter = 0;
1006 		old = d->bd_rfilter;
1007 	}
1008 	if (fp->bf_insns == NULL) {
1009 		if (fp->bf_len != 0)
1010 			return(EINVAL);
1011 		crit_enter();
1012 		if (wfilter)
1013 			d->bd_wfilter = NULL;
1014 		else
1015 			d->bd_rfilter = NULL;
1016 		bpf_resetd(d);
1017 		crit_exit();
1018 		if (old != NULL)
1019 			kfree(old, M_BPF);
1020 		return(0);
1021 	}
1022 	flen = fp->bf_len;
1023 	if (flen > BPF_MAXINSNS)
1024 		return(EINVAL);
1025 
1026 	size = flen * sizeof *fp->bf_insns;
1027 	fcode = (struct bpf_insn *)kmalloc(size, M_BPF, M_WAITOK);
1028 	if (copyin(fp->bf_insns, fcode, size) == 0 &&
1029 	    bpf_validate(fcode, (int)flen)) {
1030 		crit_enter();
1031 		if (wfilter)
1032 			d->bd_wfilter = fcode;
1033 		else
1034 			d->bd_rfilter = fcode;
1035 		bpf_resetd(d);
1036 		crit_exit();
1037 		if (old != NULL)
1038 			kfree(old, M_BPF);
1039 
1040 		return(0);
1041 	}
1042 	kfree(fcode, M_BPF);
1043 	return(EINVAL);
1044 }
1045 
1046 /*
1047  * Detach a file from its current interface (if attached at all) and attach
1048  * to the interface indicated by the name stored in ifr.
1049  * Return an errno or 0.
1050  */
1051 static int
1052 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1053 {
1054 	struct bpf_if *bp;
1055 	int error;
1056 	struct ifnet *theywant;
1057 
1058 	theywant = ifunit(ifr->ifr_name);
1059 	if (theywant == NULL)
1060 		return(ENXIO);
1061 
1062 	/*
1063 	 * Look through attached interfaces for the named one.
1064 	 */
1065 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1066 		struct ifnet *ifp = bp->bif_ifp;
1067 
1068 		if (ifp == NULL || ifp != theywant)
1069 			continue;
1070 		/* skip additional entry */
1071 		if (bp->bif_driverp != &ifp->if_bpf)
1072 			continue;
1073 		/*
1074 		 * We found the requested interface.
1075 		 * Allocate the packet buffers if we need to.
1076 		 * If we're already attached to requested interface,
1077 		 * just flush the buffer.
1078 		 */
1079 		if (d->bd_sbuf == NULL) {
1080 			error = bpf_allocbufs(d);
1081 			if (error != 0)
1082 				return(error);
1083 		}
1084 		crit_enter();
1085 		if (bp != d->bd_bif) {
1086 			if (d->bd_bif != NULL) {
1087 				/*
1088 				 * Detach if attached to something else.
1089 				 */
1090 				bpf_detachd(d);
1091 			}
1092 
1093 			bpf_attachd(d, bp);
1094 		}
1095 		bpf_resetd(d);
1096 		crit_exit();
1097 		return(0);
1098 	}
1099 
1100 	/* Not found. */
1101 	return(ENXIO);
1102 }
1103 
1104 static struct filterops bpf_read_filtops =
1105 	{ FILTEROP_ISFD, NULL, bpf_filter_detach, bpf_filter_read };
1106 
1107 static int
1108 bpfkqfilter(struct dev_kqfilter_args *ap)
1109 {
1110 	cdev_t dev = ap->a_head.a_dev;
1111 	struct knote *kn = ap->a_kn;
1112 	struct klist *klist;
1113 	struct bpf_d *d;
1114 
1115 	lwkt_gettoken(&bpf_token);
1116 	d = dev->si_drv1;
1117 	if (d->bd_bif == NULL) {
1118 		ap->a_result = 1;
1119 		lwkt_reltoken(&bpf_token);
1120 		return (0);
1121 	}
1122 
1123 	ap->a_result = 0;
1124 	switch (kn->kn_filter) {
1125 	case EVFILT_READ:
1126 		kn->kn_fop = &bpf_read_filtops;
1127 		kn->kn_hook = (caddr_t)d;
1128 		break;
1129 	default:
1130 		ap->a_result = EOPNOTSUPP;
1131 		lwkt_reltoken(&bpf_token);
1132 		return (0);
1133 	}
1134 
1135 	klist = &d->bd_kq.ki_note;
1136 	knote_insert(klist, kn);
1137 	lwkt_reltoken(&bpf_token);
1138 
1139 	return (0);
1140 }
1141 
1142 static void
1143 bpf_filter_detach(struct knote *kn)
1144 {
1145 	struct klist *klist;
1146 	struct bpf_d *d;
1147 
1148 	d = (struct bpf_d *)kn->kn_hook;
1149 	klist = &d->bd_kq.ki_note;
1150 	knote_remove(klist, kn);
1151 }
1152 
1153 static int
1154 bpf_filter_read(struct knote *kn, long hint)
1155 {
1156 	struct bpf_d *d;
1157 	int ready = 0;
1158 
1159 	crit_enter();
1160 	d = (struct bpf_d *)kn->kn_hook;
1161 	if (d->bd_hlen != 0 ||
1162 	    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1163 	    d->bd_slen != 0)) {
1164 		ready = 1;
1165 	} else {
1166 		/* Start the read timeout if necessary. */
1167 		if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1168 			callout_reset(&d->bd_callout, d->bd_rtout,
1169 			    bpf_timed_out, d);
1170 			d->bd_state = BPF_WAITING;
1171 		}
1172 	}
1173 	crit_exit();
1174 
1175 	return (ready);
1176 }
1177 
1178 
1179 /*
1180  * Process the packet pkt of length pktlen.  The packet is parsed
1181  * by each listener's filter, and if accepted, stashed into the
1182  * corresponding buffer.
1183  */
1184 void
1185 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1186 {
1187 	struct bpf_d *d;
1188 	struct timeval tv;
1189 	int gottime = 0;
1190 	u_int slen;
1191 
1192 	lwkt_gettoken(&bpf_token);
1193 	/* Re-check */
1194 	if (bp == NULL) {
1195 		lwkt_reltoken(&bpf_token);
1196 		return;
1197 	}
1198 
1199 	/*
1200 	 * Note that the ipl does not have to be raised at this point.
1201 	 * The only problem that could arise here is that if two different
1202 	 * interfaces shared any data.  This is not the case.
1203 	 */
1204 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1205 		++d->bd_rcount;
1206 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1207 		if (slen != 0) {
1208 			if (!gottime) {
1209 				microtime(&tv);
1210 				gottime = 1;
1211 			}
1212 			catchpacket(d, pkt, pktlen, slen, ovbcopy, &tv);
1213 		}
1214 	}
1215 	lwkt_reltoken(&bpf_token);
1216 }
1217 
1218 /*
1219  * Copy data from an mbuf chain into a buffer.  This code is derived
1220  * from m_copydata in sys/uipc_mbuf.c.
1221  */
1222 static void
1223 bpf_mcopy(const void *src_arg, void *dst_arg, size_t len)
1224 {
1225 	const struct mbuf *m;
1226 	u_int count;
1227 	u_char *dst;
1228 
1229 	m = src_arg;
1230 	dst = dst_arg;
1231 	while (len > 0) {
1232 		if (m == NULL)
1233 			panic("bpf_mcopy");
1234 		count = min(m->m_len, len);
1235 		bcopy(mtod(m, void *), dst, count);
1236 		m = m->m_next;
1237 		dst += count;
1238 		len -= count;
1239 	}
1240 }
1241 
1242 /*
1243  * Process the packet in the mbuf chain m.  The packet is parsed by each
1244  * listener's filter, and if accepted, stashed into the corresponding
1245  * buffer.
1246  */
1247 void
1248 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1249 {
1250 	struct bpf_d *d;
1251 	u_int pktlen, slen;
1252 	struct timeval tv;
1253 	int gottime = 0;
1254 
1255 	lwkt_gettoken(&bpf_token);
1256 	/* Re-check */
1257 	if (bp == NULL) {
1258 		lwkt_reltoken(&bpf_token);
1259 		return;
1260 	}
1261 
1262 	/* Don't compute pktlen, if no descriptor is attached. */
1263 	if (SLIST_EMPTY(&bp->bif_dlist)) {
1264 		lwkt_reltoken(&bpf_token);
1265 		return;
1266 	}
1267 
1268 	pktlen = m_lengthm(m, NULL);
1269 
1270 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1271 		if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL))
1272 			continue;
1273 		++d->bd_rcount;
1274 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1275 		if (slen != 0) {
1276 			if (!gottime) {
1277 				microtime(&tv);
1278 				gottime = 1;
1279 			}
1280 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy,
1281 				    &tv);
1282 		}
1283 	}
1284 	lwkt_reltoken(&bpf_token);
1285 }
1286 
1287 /*
1288  * Incoming linkage from device drivers, where we have a mbuf chain
1289  * but need to prepend some arbitrary header from a linear buffer.
1290  *
1291  * Con up a minimal dummy header to pacify bpf.  Allocate (only) a
1292  * struct m_hdr on the stack.  This is safe as bpf only reads from the
1293  * fields in this header that we initialize, and will not try to free
1294  * it or keep a pointer to it.
1295  */
1296 void
1297 bpf_mtap_hdr(struct bpf_if *arg, caddr_t data, u_int dlen, struct mbuf *m, u_int direction)
1298 {
1299 	struct m_hdr mh;
1300 
1301 	mh.mh_flags = 0;
1302 	mh.mh_next = m;
1303 	mh.mh_len = dlen;
1304 	mh.mh_data = data;
1305 
1306 	return bpf_mtap(arg, (struct mbuf *) &mh);
1307 }
1308 
1309 void
1310 bpf_mtap_family(struct bpf_if *bp, struct mbuf *m, sa_family_t family)
1311 {
1312 	u_int family4;
1313 
1314 	KKASSERT(family != AF_UNSPEC);
1315 
1316 	family4 = (u_int)family;
1317 	bpf_ptap(bp, m, &family4, sizeof(family4));
1318 }
1319 
1320 /*
1321  * Process the packet in the mbuf chain m with the header in m prepended.
1322  * The packet is parsed by each listener's filter, and if accepted,
1323  * stashed into the corresponding buffer.
1324  */
1325 void
1326 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1327 {
1328 	struct mbuf mb;
1329 
1330 	/*
1331 	 * Craft on-stack mbuf suitable for passing to bpf_mtap.
1332 	 * Note that we cut corners here; we only setup what's
1333 	 * absolutely needed--this mbuf should never go anywhere else.
1334 	 */
1335 	mb.m_next = m;
1336 	mb.m_data = __DECONST(void *, data); /* LINTED */
1337 	mb.m_len = dlen;
1338 	mb.m_pkthdr.rcvif = m->m_pkthdr.rcvif;
1339 
1340 	bpf_mtap(bp, &mb);
1341 }
1342 
1343 /*
1344  * Move the packet data from interface memory (pkt) into the
1345  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1346  * otherwise 0.  "copy" is the routine called to do the actual data
1347  * transfer.  bcopy is passed in to copy contiguous chunks, while
1348  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1349  * pkt is really an mbuf.
1350  */
1351 static void
1352 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1353 	    void (*cpfn)(const void *, void *, size_t),
1354 	    const struct timeval *tv)
1355 {
1356 	struct bpf_hdr *hp;
1357 	int totlen, curlen;
1358 	int hdrlen = d->bd_bif->bif_hdrlen;
1359 	int wakeup = 0;
1360 	/*
1361 	 * Figure out how many bytes to move.  If the packet is
1362 	 * greater or equal to the snapshot length, transfer that
1363 	 * much.  Otherwise, transfer the whole packet (unless
1364 	 * we hit the buffer size limit).
1365 	 */
1366 	totlen = hdrlen + min(snaplen, pktlen);
1367 	if (totlen > d->bd_bufsize)
1368 		totlen = d->bd_bufsize;
1369 
1370 	/*
1371 	 * Round up the end of the previous packet to the next longword.
1372 	 */
1373 	curlen = BPF_WORDALIGN(d->bd_slen);
1374 	if (curlen + totlen > d->bd_bufsize) {
1375 		/*
1376 		 * This packet will overflow the storage buffer.
1377 		 * Rotate the buffers if we can, then wakeup any
1378 		 * pending reads.
1379 		 */
1380 		if (d->bd_fbuf == NULL) {
1381 			/*
1382 			 * We haven't completed the previous read yet,
1383 			 * so drop the packet.
1384 			 */
1385 			++d->bd_dcount;
1386 			return;
1387 		}
1388 		ROTATE_BUFFERS(d);
1389 		wakeup = 1;
1390 		curlen = 0;
1391 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1392 		/*
1393 		 * Immediate mode is set, or the read timeout has
1394 		 * already expired during a select call.  A packet
1395 		 * arrived, so the reader should be woken up.
1396 		 */
1397 		wakeup = 1;
1398 	}
1399 
1400 	/*
1401 	 * Append the bpf header.
1402 	 */
1403 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1404 	hp->bh_tstamp = *tv;
1405 	hp->bh_datalen = pktlen;
1406 	hp->bh_hdrlen = hdrlen;
1407 	/*
1408 	 * Copy the packet data into the store buffer and update its length.
1409 	 */
1410 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1411 	d->bd_slen = curlen + totlen;
1412 
1413 	if (wakeup)
1414 		bpf_wakeup(d);
1415 }
1416 
1417 /*
1418  * Initialize all nonzero fields of a descriptor.
1419  */
1420 static int
1421 bpf_allocbufs(struct bpf_d *d)
1422 {
1423 	d->bd_fbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1424 	d->bd_sbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1425 	d->bd_slen = 0;
1426 	d->bd_hlen = 0;
1427 	return(0);
1428 }
1429 
1430 /*
1431  * Free buffers and packet filter program currently in use by a descriptor.
1432  * Called on close.
1433  */
1434 static void
1435 bpf_freed(struct bpf_d *d)
1436 {
1437 	/*
1438 	 * We don't need to lock out interrupts since this descriptor has
1439 	 * been detached from its interface and it yet hasn't been marked
1440 	 * free.
1441 	 */
1442 	if (d->bd_sbuf != NULL) {
1443 		kfree(d->bd_sbuf, M_BPF);
1444 		if (d->bd_hbuf != NULL)
1445 			kfree(d->bd_hbuf, M_BPF);
1446 		if (d->bd_fbuf != NULL)
1447 			kfree(d->bd_fbuf, M_BPF);
1448 	}
1449 	if (d->bd_rfilter)
1450 		kfree(d->bd_rfilter, M_BPF);
1451 	if (d->bd_wfilter)
1452 		kfree(d->bd_wfilter, M_BPF);
1453 }
1454 
1455 /*
1456  * Attach an interface to bpf.  ifp is a pointer to the structure
1457  * defining the interface to be attached, dlt is the link layer type,
1458  * and hdrlen is the fixed size of the link header (variable length
1459  * headers are not yet supported).
1460  */
1461 void
1462 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1463 {
1464 	bpfattach_dlt(ifp, dlt, hdrlen, &ifp->if_bpf);
1465 }
1466 
1467 void
1468 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1469 {
1470 	struct bpf_if *bp;
1471 
1472 	bp = kmalloc(sizeof *bp, M_BPF, M_WAITOK | M_ZERO);
1473 
1474 	lwkt_gettoken(&bpf_token);
1475 
1476 	SLIST_INIT(&bp->bif_dlist);
1477 	bp->bif_ifp = ifp;
1478 	bp->bif_dlt = dlt;
1479 	bp->bif_driverp = driverp;
1480 	*bp->bif_driverp = NULL;
1481 
1482 	bp->bif_next = bpf_iflist;
1483 	bpf_iflist = bp;
1484 
1485 	/*
1486 	 * Compute the length of the bpf header.  This is not necessarily
1487 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1488 	 * that the network layer header begins on a longword boundary (for
1489 	 * performance reasons and to alleviate alignment restrictions).
1490 	 */
1491 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1492 
1493 	lwkt_reltoken(&bpf_token);
1494 
1495 	if (bootverbose)
1496 		if_printf(ifp, "bpf attached\n");
1497 }
1498 
1499 /*
1500  * Detach bpf from an interface.  This involves detaching each descriptor
1501  * associated with the interface, and leaving bd_bif NULL.  Notify each
1502  * descriptor as it's detached so that any sleepers wake up and get
1503  * ENXIO.
1504  */
1505 void
1506 bpfdetach(struct ifnet *ifp)
1507 {
1508 	struct bpf_if *bp, *bp_prev;
1509 	struct bpf_d *d;
1510 
1511 	lwkt_gettoken(&bpf_token);
1512 	crit_enter();
1513 
1514 	/* Locate BPF interface information */
1515 	bp_prev = NULL;
1516 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1517 		if (ifp == bp->bif_ifp)
1518 			break;
1519 		bp_prev = bp;
1520 	}
1521 
1522 	/* Interface wasn't attached */
1523 	if (bp->bif_ifp == NULL) {
1524 		crit_exit();
1525 		lwkt_reltoken(&bpf_token);
1526 		kprintf("bpfdetach: %s was not attached\n", ifp->if_xname);
1527 		return;
1528 	}
1529 
1530 	while ((d = SLIST_FIRST(&bp->bif_dlist)) != NULL) {
1531 		bpf_detachd(d);
1532 		bpf_wakeup(d);
1533 	}
1534 
1535 	if (bp_prev != NULL)
1536 		bp_prev->bif_next = bp->bif_next;
1537 	else
1538 		bpf_iflist = bp->bif_next;
1539 
1540 	kfree(bp, M_BPF);
1541 
1542 	crit_exit();
1543 	lwkt_reltoken(&bpf_token);
1544 }
1545 
1546 /*
1547  * Get a list of available data link type of the interface.
1548  */
1549 static int
1550 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1551 {
1552 	int n, error;
1553 	struct ifnet *ifp;
1554 	struct bpf_if *bp;
1555 
1556 	ifp = d->bd_bif->bif_ifp;
1557 	n = 0;
1558 	error = 0;
1559 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1560 		if (bp->bif_ifp != ifp)
1561 			continue;
1562 		if (bfl->bfl_list != NULL) {
1563 			if (n >= bfl->bfl_len) {
1564 				return (ENOMEM);
1565 			}
1566 			error = copyout(&bp->bif_dlt,
1567 			    bfl->bfl_list + n, sizeof(u_int));
1568 		}
1569 		n++;
1570 	}
1571 	bfl->bfl_len = n;
1572 	return(error);
1573 }
1574 
1575 /*
1576  * Set the data link type of a BPF instance.
1577  */
1578 static int
1579 bpf_setdlt(struct bpf_d *d, u_int dlt)
1580 {
1581 	int error, opromisc;
1582 	struct ifnet *ifp;
1583 	struct bpf_if *bp;
1584 
1585 	if (d->bd_bif->bif_dlt == dlt)
1586 		return (0);
1587 	ifp = d->bd_bif->bif_ifp;
1588 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1589 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1590 			break;
1591 	}
1592 	if (bp != NULL) {
1593 		opromisc = d->bd_promisc;
1594 		crit_enter();
1595 		bpf_detachd(d);
1596 		bpf_attachd(d, bp);
1597 		bpf_resetd(d);
1598 		if (opromisc) {
1599 			error = ifpromisc(bp->bif_ifp, 1);
1600 			if (error) {
1601 				if_printf(bp->bif_ifp,
1602 					"bpf_setdlt: ifpromisc failed (%d)\n",
1603 					error);
1604 			} else {
1605 				d->bd_promisc = 1;
1606 			}
1607 		}
1608 		crit_exit();
1609 	}
1610 	return(bp == NULL ? EINVAL : 0);
1611 }
1612 
1613 void
1614 bpf_gettoken(void)
1615 {
1616 	lwkt_gettoken(&bpf_token);
1617 }
1618 
1619 void
1620 bpf_reltoken(void)
1621 {
1622 	lwkt_reltoken(&bpf_token);
1623 }
1624 
1625 static void
1626 bpf_drvinit(void *unused)
1627 {
1628 	int i;
1629 
1630 	make_autoclone_dev(&bpf_ops, &DEVFS_CLONE_BITMAP(bpf),
1631 		bpfclone, 0, 0, 0600, "bpf");
1632 	for (i = 0; i < BPF_PREALLOCATED_UNITS; i++) {
1633 		make_dev(&bpf_ops, i, 0, 0, 0600, "bpf%d", i);
1634 		devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(bpf), i);
1635 	}
1636 }
1637 
1638 static void
1639 bpf_drvuninit(void *unused)
1640 {
1641 	devfs_clone_handler_del("bpf");
1642 	dev_ops_remove_all(&bpf_ops);
1643 	devfs_clone_bitmap_uninit(&DEVFS_CLONE_BITMAP(bpf));
1644 }
1645 
1646 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvinit,NULL)
1647 SYSUNINIT(bpfdev, SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvuninit, NULL);
1648 
1649 #else /* !BPF */
1650 /*
1651  * NOP stubs to allow bpf-using drivers to load and function.
1652  *
1653  * A 'better' implementation would allow the core bpf functionality
1654  * to be loaded at runtime.
1655  */
1656 
1657 void
1658 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1659 {
1660 }
1661 
1662 void
1663 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1664 {
1665 }
1666 
1667 void
1668 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1669 {
1670 }
1671 
1672 void
1673 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1674 {
1675 }
1676 
1677 void
1678 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1679 {
1680 }
1681 
1682 void
1683 bpfdetach(struct ifnet *ifp)
1684 {
1685 }
1686 
1687 u_int
1688 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
1689 {
1690 	return -1;	/* "no filter" behaviour */
1691 }
1692 
1693 void
1694 bpf_gettoken(void)
1695 {
1696 }
1697 
1698 void
1699 bpf_reltoken(void)
1700 {
1701 }
1702 
1703 #endif /* !BPF */
1704