xref: /dragonfly/sys/net/bpf.c (revision 6b47f3ea)
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/uio.h>
44 #include <sys/conf.h>
45 #include <sys/device.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/time.h>
49 #include <sys/proc.h>
50 #include <sys/signalvar.h>
51 #include <sys/filio.h>
52 #include <sys/sockio.h>
53 #include <sys/ttycom.h>
54 #include <sys/filedesc.h>
55 
56 #include <sys/event.h>
57 
58 #include <sys/socket.h>
59 #include <sys/vnode.h>
60 
61 #include <net/if.h>
62 #include <net/bpf.h>
63 #include <net/bpfdesc.h>
64 #include <net/netmsg2.h>
65 #include <net/netisr2.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/if_ether.h>
69 #include <sys/kernel.h>
70 #include <sys/sysctl.h>
71 
72 #include <netproto/802_11/ieee80211_dragonfly.h>
73 
74 #include <sys/devfs.h>
75 
76 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
77 DEVFS_DEFINE_CLONE_BITMAP(bpf);
78 
79 #if NBPF <= 1
80 #define BPF_PREALLOCATED_UNITS	4
81 #else
82 #define BPF_PREALLOCATED_UNITS	NBPF
83 #endif
84 
85 #if NBPF > 0
86 
87 struct netmsg_bpf_output {
88 	struct netmsg_base base;
89 	struct mbuf	*nm_mbuf;
90 	struct ifnet	*nm_ifp;
91 	struct sockaddr	*nm_dst;
92 	boolean_t 	nm_feedback;
93 };
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 	m->m_pkthdr.len = m->m_len = len;
232 	m->m_pkthdr.rcvif = NULL;
233 	*mp = m;
234 
235 	if (m->m_len < hlen) {
236 		error = EPERM;
237 		goto bad;
238 	}
239 
240 	error = uiomove(mtod(m, u_char *), len, uio);
241 	if (error)
242 		goto bad;
243 
244 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
245 	if (slen == 0) {
246 		error = EPERM;
247 		goto bad;
248 	}
249 
250 	/*
251 	 * Make room for link header, and copy it to sockaddr.
252 	 */
253 	if (hlen != 0) {
254 		if (sockp->sa_family == AF_IEEE80211) {
255 			/*
256 			 * Collect true length from the parameter header
257 			 * NB: sockp is known to be zero'd so if we do a
258 			 *     short copy unspecified parameters will be
259 			 *     zero.
260 			 * NB: packet may not be aligned after stripping
261 			 *     bpf params
262 			 * XXX check ibp_vers
263 			 */
264 			p = mtod(m, const struct ieee80211_bpf_params *);
265 			hlen = p->ibp_len;
266 			if (hlen > sizeof(sockp->sa_data)) {
267 				error = EINVAL;
268 				goto bad;
269 			}
270 		}
271 		bcopy(m->m_data, sockp->sa_data, hlen);
272 		m->m_pkthdr.len -= hlen;
273 		m->m_len -= hlen;
274 		m->m_data += hlen; /* XXX */
275 	}
276 	return (0);
277 bad:
278 	m_freem(m);
279 	return(error);
280 }
281 
282 /*
283  * Attach file to the bpf interface, i.e. make d listen on bp.
284  * Must be called at splimp.
285  */
286 static void
287 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
288 {
289 	/*
290 	 * Point d at bp, and add d to the interface's list of listeners.
291 	 * Finally, point the driver's bpf cookie at the interface so
292 	 * it will divert packets to bpf.
293 	 */
294 	lwkt_gettoken(&bpf_token);
295 	d->bd_bif = bp;
296 	SLIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
297 	*bp->bif_driverp = bp;
298 
299 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
300 	lwkt_reltoken(&bpf_token);
301 }
302 
303 /*
304  * Detach a file from its interface.
305  */
306 static void
307 bpf_detachd(struct bpf_d *d)
308 {
309 	int error;
310 	struct bpf_if *bp;
311 	struct ifnet *ifp;
312 
313 	lwkt_gettoken(&bpf_token);
314 	bp = d->bd_bif;
315 	ifp = bp->bif_ifp;
316 
317 	/* Remove d from the interface's descriptor list. */
318 	SLIST_REMOVE(&bp->bif_dlist, d, bpf_d, bd_next);
319 
320 	if (SLIST_EMPTY(&bp->bif_dlist)) {
321 		/*
322 		 * Let the driver know that there are no more listeners.
323 		 */
324 		*bp->bif_driverp = NULL;
325 	}
326 	d->bd_bif = NULL;
327 
328 	EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
329 
330 	/*
331 	 * Check if this descriptor had requested promiscuous mode.
332 	 * If so, turn it off.
333 	 */
334 	if (d->bd_promisc) {
335 		d->bd_promisc = 0;
336 		error = ifpromisc(ifp, 0);
337 		if (error != 0 && error != ENXIO) {
338 			/*
339 			 * ENXIO can happen if a pccard is unplugged,
340 			 * Something is really wrong if we were able to put
341 			 * the driver into promiscuous mode, but can't
342 			 * take it out.
343 			 */
344 			if_printf(ifp, "bpf_detach: ifpromisc failed(%d)\n",
345 				  error);
346 		}
347 	}
348 	lwkt_reltoken(&bpf_token);
349 }
350 
351 /*
352  * Open ethernet device.  Returns ENXIO for illegal minor device number,
353  * EBUSY if file is open by another process.
354  */
355 /* ARGSUSED */
356 static int
357 bpfopen(struct dev_open_args *ap)
358 {
359 	cdev_t dev = ap->a_head.a_dev;
360 	struct bpf_d *d;
361 
362 	lwkt_gettoken(&bpf_token);
363 	if (ap->a_cred->cr_prison) {
364 		lwkt_reltoken(&bpf_token);
365 		return(EPERM);
366 	}
367 
368 	d = dev->si_drv1;
369 	/*
370 	 * Each minor can be opened by only one process.  If the requested
371 	 * minor is in use, return EBUSY.
372 	 */
373 	if (d != NULL) {
374 		lwkt_reltoken(&bpf_token);
375 		return(EBUSY);
376 	}
377 
378 	d = kmalloc(sizeof *d, M_BPF, M_WAITOK | M_ZERO);
379 	dev->si_drv1 = d;
380 	d->bd_bufsize = bpf_bufsize;
381 	d->bd_sig = SIGIO;
382 	d->bd_seesent = 1;
383 	d->bd_feedback = 0;
384 	callout_init(&d->bd_callout);
385 	lwkt_reltoken(&bpf_token);
386 
387 	return(0);
388 }
389 
390 static int
391 bpfclone(struct dev_clone_args *ap)
392 {
393 	int unit;
394 
395 	unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(bpf), 0);
396 	ap->a_dev = make_only_dev(&bpf_ops, unit, 0, 0, 0600, "bpf%d", unit);
397 
398 	return 0;
399 }
400 
401 /*
402  * Close the descriptor by detaching it from its interface,
403  * deallocating its buffers, and marking it free.
404  */
405 /* ARGSUSED */
406 static int
407 bpfclose(struct dev_close_args *ap)
408 {
409 	cdev_t dev = ap->a_head.a_dev;
410 	struct bpf_d *d = dev->si_drv1;
411 	int unit;
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 
423 	unit = dev->si_uminor;
424 	if (unit >= BPF_PREALLOCATED_UNITS) {
425 		destroy_dev(dev);
426 		devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(bpf), unit);
427 	}
428 	kfree(d, M_BPF);
429 	lwkt_reltoken(&bpf_token);
430 
431 	return(0);
432 }
433 
434 /*
435  * Rotate the packet buffers in descriptor d.  Move the store buffer
436  * into the hold slot, and the free buffer into the store slot.
437  * Zero the length of the new store buffer.
438  */
439 #define ROTATE_BUFFERS(d) \
440 	(d)->bd_hbuf = (d)->bd_sbuf; \
441 	(d)->bd_hlen = (d)->bd_slen; \
442 	(d)->bd_sbuf = (d)->bd_fbuf; \
443 	(d)->bd_slen = 0; \
444 	(d)->bd_fbuf = NULL;
445 /*
446  *  bpfread - read next chunk of packets from buffers
447  */
448 static int
449 bpfread(struct dev_read_args *ap)
450 {
451 	cdev_t dev = ap->a_head.a_dev;
452 	struct bpf_d *d = dev->si_drv1;
453 	int timed_out;
454 	int error;
455 
456 	lwkt_gettoken(&bpf_token);
457 	/*
458 	 * Restrict application to use a buffer the same size as
459 	 * as kernel buffers.
460 	 */
461 	if (ap->a_uio->uio_resid != d->bd_bufsize) {
462 		lwkt_reltoken(&bpf_token);
463 		return(EINVAL);
464 	}
465 
466 	if (d->bd_state == BPF_WAITING)
467 		callout_stop(&d->bd_callout);
468 	timed_out = (d->bd_state == BPF_TIMED_OUT);
469 	d->bd_state = BPF_IDLE;
470 	/*
471 	 * If the hold buffer is empty, then do a timed sleep, which
472 	 * ends when the timeout expires or when enough packets
473 	 * have arrived to fill the store buffer.
474 	 */
475 	while (d->bd_hbuf == NULL) {
476 		if ((d->bd_immediate || (ap->a_ioflag & IO_NDELAY) || timed_out)
477 		    && d->bd_slen != 0) {
478 			/*
479 			 * A packet(s) either arrived since the previous,
480 			 * We're in immediate mode, or are reading
481 			 * in non-blocking mode, and a packet(s)
482 			 * either arrived since the previous
483 			 * read or arrived while we were asleep.
484 			 * Rotate the buffers and return what's here.
485 			 */
486 			ROTATE_BUFFERS(d);
487 			break;
488 		}
489 
490 		/*
491 		 * No data is available, check to see if the bpf device
492 		 * is still pointed at a real interface.  If not, return
493 		 * ENXIO so that the userland process knows to rebind
494 		 * it before using it again.
495 		 */
496 		if (d->bd_bif == NULL) {
497 			lwkt_reltoken(&bpf_token);
498 			return(ENXIO);
499 		}
500 
501 		if (ap->a_ioflag & IO_NDELAY) {
502 			lwkt_reltoken(&bpf_token);
503 			return(EWOULDBLOCK);
504 		}
505 		error = tsleep(d, PCATCH, "bpf", d->bd_rtout);
506 		if (error == EINTR || error == ERESTART) {
507 			lwkt_reltoken(&bpf_token);
508 			return(error);
509 		}
510 		if (error == EWOULDBLOCK) {
511 			/*
512 			 * On a timeout, return what's in the buffer,
513 			 * which may be nothing.  If there is something
514 			 * in the store buffer, we can rotate the buffers.
515 			 */
516 			if (d->bd_hbuf)
517 				/*
518 				 * We filled up the buffer in between
519 				 * getting the timeout and arriving
520 				 * here, so we don't need to rotate.
521 				 */
522 				break;
523 
524 			if (d->bd_slen == 0) {
525 				lwkt_reltoken(&bpf_token);
526 				return(0);
527 			}
528 			ROTATE_BUFFERS(d);
529 			break;
530 		}
531 	}
532 	/*
533 	 * At this point, we know we have something in the hold slot.
534 	 */
535 
536 	/*
537 	 * Move data from hold buffer into user space.
538 	 * We know the entire buffer is transferred since
539 	 * we checked above that the read buffer is bpf_bufsize bytes.
540 	 */
541 	error = uiomove(d->bd_hbuf, d->bd_hlen, ap->a_uio);
542 
543 	d->bd_fbuf = d->bd_hbuf;
544 	d->bd_hbuf = NULL;
545 	d->bd_hlen = 0;
546 	lwkt_reltoken(&bpf_token);
547 
548 	return(error);
549 }
550 
551 
552 /*
553  * If there are processes sleeping on this descriptor, wake them up.
554  */
555 static void
556 bpf_wakeup(struct bpf_d *d)
557 {
558 	if (d->bd_state == BPF_WAITING) {
559 		callout_stop(&d->bd_callout);
560 		d->bd_state = BPF_IDLE;
561 	}
562 	wakeup(d);
563 	if (d->bd_async && d->bd_sig && d->bd_sigio)
564 		pgsigio(d->bd_sigio, d->bd_sig, 0);
565 
566 	KNOTE(&d->bd_kq.ki_note, 0);
567 }
568 
569 static void
570 bpf_timed_out(void *arg)
571 {
572 	struct bpf_d *d = (struct bpf_d *)arg;
573 
574 	if (d->bd_state == BPF_WAITING) {
575 		d->bd_state = BPF_TIMED_OUT;
576 		if (d->bd_slen != 0)
577 			bpf_wakeup(d);
578 	}
579 }
580 
581 static void
582 bpf_output_dispatch(netmsg_t msg)
583 {
584 	struct netmsg_bpf_output *bmsg = (struct netmsg_bpf_output *)msg;
585 	struct ifnet *ifp = bmsg->nm_ifp;
586 	struct mbuf *mc = NULL;
587 	int error;
588 
589 	if (bmsg->nm_feedback) {
590 		mc = m_dup(bmsg->nm_mbuf, M_NOWAIT);
591 		if (mc != NULL)
592 			mc->m_pkthdr.rcvif = ifp;
593 	}
594 
595 	/*
596 	 * The driver frees the mbuf.
597 	 */
598 	error = ifp->if_output(ifp, bmsg->nm_mbuf, bmsg->nm_dst, NULL);
599 	lwkt_replymsg(&msg->lmsg, error);
600 
601 	if (mc != NULL) {
602 		if (error == 0) {
603 			mc->m_flags &= ~M_HASH;
604 			(*ifp->if_input)(ifp, mc, NULL, -1);
605 		} else {
606 			m_freem(mc);
607 		}
608 	}
609 }
610 
611 static int
612 bpfwrite(struct dev_write_args *ap)
613 {
614 	cdev_t dev = ap->a_head.a_dev;
615 	struct bpf_d *d = dev->si_drv1;
616 	struct ifnet *ifp;
617 	struct mbuf *m;
618 	int error, ret;
619 	struct sockaddr dst;
620 	int datlen;
621 	struct netmsg_bpf_output bmsg;
622 
623 	lwkt_gettoken(&bpf_token);
624 	if (d->bd_bif == NULL) {
625 		lwkt_reltoken(&bpf_token);
626 		return(ENXIO);
627 	}
628 
629 	ifp = d->bd_bif->bif_ifp;
630 
631 	if (ap->a_uio->uio_resid == 0) {
632 		lwkt_reltoken(&bpf_token);
633 		return(0);
634 	}
635 
636 	error = bpf_movein(ap->a_uio, (int)d->bd_bif->bif_dlt, &m,
637 			   &dst, &datlen, d->bd_wfilter);
638 	if (error) {
639 		lwkt_reltoken(&bpf_token);
640 		return(error);
641 	}
642 
643 	if (datlen > ifp->if_mtu) {
644 		m_freem(m);
645 		lwkt_reltoken(&bpf_token);
646 		return(EMSGSIZE);
647 	}
648 
649 	if (d->bd_hdrcmplt)
650 		dst.sa_family = pseudo_AF_HDRCMPLT;
651 
652 	netmsg_init(&bmsg.base, NULL, &curthread->td_msgport,
653 		    0, bpf_output_dispatch);
654 	bmsg.nm_mbuf = m;
655 	bmsg.nm_ifp = ifp;
656 	bmsg.nm_dst = &dst;
657 
658 	if (d->bd_feedback)
659 		bmsg.nm_feedback = TRUE;
660 	else
661 		bmsg.nm_feedback = FALSE;
662 
663 	ret = lwkt_domsg(netisr_cpuport(0), &bmsg.base.lmsg, 0);
664 
665 	lwkt_reltoken(&bpf_token);
666 
667 	return ret;
668 }
669 
670 /*
671  * Reset a descriptor by flushing its packet buffer and clearing the
672  * receive and drop counts.  Should be called at splimp.
673  */
674 static void
675 bpf_resetd(struct bpf_d *d)
676 {
677 	if (d->bd_hbuf) {
678 		/* Free the hold buffer. */
679 		d->bd_fbuf = d->bd_hbuf;
680 		d->bd_hbuf = NULL;
681 	}
682 	d->bd_slen = 0;
683 	d->bd_hlen = 0;
684 	d->bd_rcount = 0;
685 	d->bd_dcount = 0;
686 }
687 
688 /*
689  *  FIONREAD		Check for read packet available.
690  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
691  *  BIOCGBLEN		Get buffer len [for read()].
692  *  BIOCSETF		Set ethernet read filter.
693  *  BIOCSETWF		Set ethernet write filter.
694  *  BIOCFLUSH		Flush read packet buffer.
695  *  BIOCPROMISC		Put interface into promiscuous mode.
696  *  BIOCGDLT		Get link layer type.
697  *  BIOCGETIF		Get interface name.
698  *  BIOCSETIF		Set interface.
699  *  BIOCSRTIMEOUT	Set read timeout.
700  *  BIOCGRTIMEOUT	Get read timeout.
701  *  BIOCGSTATS		Get packet stats.
702  *  BIOCIMMEDIATE	Set immediate mode.
703  *  BIOCVERSION		Get filter language version.
704  *  BIOCGHDRCMPLT	Get "header already complete" flag
705  *  BIOCSHDRCMPLT	Set "header already complete" flag
706  *  BIOCSFEEDBACK	Set packet feedback mode.
707  *  BIOCGFEEDBACK	Get packet feedback mode.
708  *  BIOCGSEESENT	Get "see packets sent" flag
709  *  BIOCSSEESENT	Set "see packets sent" flag
710  *  BIOCLOCK		Set "locked" flag
711  */
712 /* ARGSUSED */
713 static int
714 bpfioctl(struct dev_ioctl_args *ap)
715 {
716 	cdev_t dev = ap->a_head.a_dev;
717 	struct bpf_d *d = dev->si_drv1;
718 	int error = 0;
719 
720 	lwkt_gettoken(&bpf_token);
721 	if (d->bd_state == BPF_WAITING)
722 		callout_stop(&d->bd_callout);
723 	d->bd_state = BPF_IDLE;
724 
725 	if (d->bd_locked == 1) {
726 		switch (ap->a_cmd) {
727 		case BIOCGBLEN:
728 		case BIOCFLUSH:
729 		case BIOCGDLT:
730 		case BIOCGDLTLIST:
731 		case BIOCGETIF:
732 		case BIOCGRTIMEOUT:
733 		case BIOCGSTATS:
734 		case BIOCVERSION:
735 		case BIOCGRSIG:
736 		case BIOCGHDRCMPLT:
737 		case FIONREAD:
738 		case BIOCLOCK:
739 		case BIOCSRTIMEOUT:
740 		case BIOCIMMEDIATE:
741 		case TIOCGPGRP:
742 			break;
743 		default:
744 			lwkt_reltoken(&bpf_token);
745 			return (EPERM);
746 		}
747 	}
748 	switch (ap->a_cmd) {
749 	default:
750 		error = EINVAL;
751 		break;
752 
753 	/*
754 	 * Check for read packet available.
755 	 */
756 	case FIONREAD:
757 		{
758 			int n;
759 
760 			n = d->bd_slen;
761 			if (d->bd_hbuf)
762 				n += d->bd_hlen;
763 
764 			*(int *)ap->a_data = n;
765 			break;
766 		}
767 
768 	case SIOCGIFADDR:
769 		{
770 			struct ifnet *ifp;
771 
772 			if (d->bd_bif == NULL) {
773 				error = EINVAL;
774 			} else {
775 				ifp = d->bd_bif->bif_ifp;
776 				ifnet_serialize_all(ifp);
777 				error = ifp->if_ioctl(ifp, ap->a_cmd,
778 						      ap->a_data, ap->a_cred);
779 				ifnet_deserialize_all(ifp);
780 			}
781 			break;
782 		}
783 
784 	/*
785 	 * Get buffer len [for read()].
786 	 */
787 	case BIOCGBLEN:
788 		*(u_int *)ap->a_data = d->bd_bufsize;
789 		break;
790 
791 	/*
792 	 * Set buffer length.
793 	 */
794 	case BIOCSBLEN:
795 		if (d->bd_bif != NULL) {
796 			error = EINVAL;
797 		} else {
798 			u_int size = *(u_int *)ap->a_data;
799 
800 			if (size > bpf_maxbufsize)
801 				*(u_int *)ap->a_data = size = bpf_maxbufsize;
802 			else if (size < BPF_MINBUFSIZE)
803 				*(u_int *)ap->a_data = size = BPF_MINBUFSIZE;
804 			d->bd_bufsize = size;
805 		}
806 		break;
807 
808 	/*
809 	 * Set link layer read filter.
810 	 */
811 	case BIOCSETF:
812 	case BIOCSETWF:
813 		error = bpf_setf(d, (struct bpf_program *)ap->a_data,
814 			ap->a_cmd);
815 		break;
816 
817 	/*
818 	 * Flush read packet buffer.
819 	 */
820 	case BIOCFLUSH:
821 		bpf_resetd(d);
822 		break;
823 
824 	/*
825 	 * Put interface into promiscuous mode.
826 	 */
827 	case BIOCPROMISC:
828 		if (d->bd_bif == NULL) {
829 			/*
830 			 * No interface attached yet.
831 			 */
832 			error = EINVAL;
833 			break;
834 		}
835 		if (d->bd_promisc == 0) {
836 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
837 			if (error == 0)
838 				d->bd_promisc = 1;
839 		}
840 		break;
841 
842 	/*
843 	 * Get device parameters.
844 	 */
845 	case BIOCGDLT:
846 		if (d->bd_bif == NULL)
847 			error = EINVAL;
848 		else
849 			*(u_int *)ap->a_data = d->bd_bif->bif_dlt;
850 		break;
851 
852 	/*
853 	 * Get a list of supported data link types.
854 	 */
855 	case BIOCGDLTLIST:
856 		if (d->bd_bif == NULL) {
857 			error = EINVAL;
858 		} else {
859 			error = bpf_getdltlist(d,
860 				(struct bpf_dltlist *)ap->a_data);
861 		}
862 		break;
863 
864 	/*
865 	 * Set data link type.
866 	 */
867 	case BIOCSDLT:
868 		if (d->bd_bif == NULL)
869 			error = EINVAL;
870 		else
871 			error = bpf_setdlt(d, *(u_int *)ap->a_data);
872 		break;
873 
874 	/*
875 	 * Get interface name.
876 	 */
877 	case BIOCGETIF:
878 		if (d->bd_bif == NULL) {
879 			error = EINVAL;
880 		} else {
881 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
882 			struct ifreq *const ifr = (struct ifreq *)ap->a_data;
883 
884 			strlcpy(ifr->ifr_name, ifp->if_xname,
885 				sizeof ifr->ifr_name);
886 		}
887 		break;
888 
889 	/*
890 	 * Set interface.
891 	 */
892 	case BIOCSETIF:
893 		error = bpf_setif(d, (struct ifreq *)ap->a_data);
894 		break;
895 
896 	/*
897 	 * Set read timeout.
898 	 */
899 	case BIOCSRTIMEOUT:
900 		{
901 			struct timeval *tv = (struct timeval *)ap->a_data;
902 
903 			/*
904 			 * Subtract 1 tick from tvtohz() since this isn't
905 			 * a one-shot timer.
906 			 */
907 			if ((error = itimerfix(tv)) == 0)
908 				d->bd_rtout = tvtohz_low(tv);
909 			break;
910 		}
911 
912 	/*
913 	 * Get read timeout.
914 	 */
915 	case BIOCGRTIMEOUT:
916 		{
917 			struct timeval *tv = (struct timeval *)ap->a_data;
918 
919 			tv->tv_sec = d->bd_rtout / hz;
920 			tv->tv_usec = (d->bd_rtout % hz) * ustick;
921 			break;
922 		}
923 
924 	/*
925 	 * Get packet stats.
926 	 */
927 	case BIOCGSTATS:
928 		{
929 			struct bpf_stat *bs = (struct bpf_stat *)ap->a_data;
930 
931 			bs->bs_recv = d->bd_rcount;
932 			bs->bs_drop = d->bd_dcount;
933 			break;
934 		}
935 
936 	/*
937 	 * Set immediate mode.
938 	 */
939 	case BIOCIMMEDIATE:
940 		d->bd_immediate = *(u_int *)ap->a_data;
941 		break;
942 
943 	case BIOCVERSION:
944 		{
945 			struct bpf_version *bv = (struct bpf_version *)ap->a_data;
946 
947 			bv->bv_major = BPF_MAJOR_VERSION;
948 			bv->bv_minor = BPF_MINOR_VERSION;
949 			break;
950 		}
951 
952 	/*
953 	 * Get "header already complete" flag
954 	 */
955 	case BIOCGHDRCMPLT:
956 		*(u_int *)ap->a_data = d->bd_hdrcmplt;
957 		break;
958 
959 	/*
960 	 * Set "header already complete" flag
961 	 */
962 	case BIOCSHDRCMPLT:
963 		d->bd_hdrcmplt = *(u_int *)ap->a_data ? 1 : 0;
964 		break;
965 
966 	/*
967 	 * Get "see sent packets" flag
968 	 */
969 	case BIOCGSEESENT:
970 		*(u_int *)ap->a_data = d->bd_seesent;
971 		break;
972 
973 	/*
974 	 * Set "see sent packets" flag
975 	 */
976 	case BIOCSSEESENT:
977 		d->bd_seesent = *(u_int *)ap->a_data;
978 		break;
979 
980 	case FIOASYNC:		/* Send signal on receive packets */
981 		d->bd_async = *(int *)ap->a_data;
982 		break;
983 
984 	/*
985 	 * Set "feed packets from bpf back to input" mode
986 	 */
987 	case BIOCSFEEDBACK:
988 		d->bd_feedback = *(int *)ap->a_data;
989 		break;
990 
991 	/*
992 	 * Get "feed packets from bpf back to input" mode
993 	 */
994 	case BIOCGFEEDBACK:
995 		*(u_int *)ap->a_data = d->bd_feedback;
996 		break;
997 
998 	case FIOSETOWN:
999 		error = fsetown(*(int *)ap->a_data, &d->bd_sigio);
1000 		break;
1001 
1002 	case FIOGETOWN:
1003 		*(int *)ap->a_data = fgetown(&d->bd_sigio);
1004 		break;
1005 
1006 	/* This is deprecated, FIOSETOWN should be used instead. */
1007 	case TIOCSPGRP:
1008 		error = fsetown(-(*(int *)ap->a_data), &d->bd_sigio);
1009 		break;
1010 
1011 	/* This is deprecated, FIOGETOWN should be used instead. */
1012 	case TIOCGPGRP:
1013 		*(int *)ap->a_data = -fgetown(&d->bd_sigio);
1014 		break;
1015 
1016 	case BIOCSRSIG:		/* Set receive signal */
1017 		{
1018 			u_int sig;
1019 
1020 			sig = *(u_int *)ap->a_data;
1021 
1022 			if (sig >= NSIG)
1023 				error = EINVAL;
1024 			else
1025 				d->bd_sig = sig;
1026 			break;
1027 		}
1028 	case BIOCGRSIG:
1029 		*(u_int *)ap->a_data = d->bd_sig;
1030 		break;
1031 	case BIOCLOCK:
1032 		d->bd_locked = 1;
1033 		break;
1034 	}
1035 	lwkt_reltoken(&bpf_token);
1036 
1037 	return(error);
1038 }
1039 
1040 /*
1041  * Set d's packet filter program to fp.  If this file already has a filter,
1042  * free it and replace it.  Returns EINVAL for bogus requests.
1043  */
1044 static int
1045 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1046 {
1047 	struct bpf_insn *fcode, *old;
1048 	u_int wfilter, flen, size;
1049 
1050 	if (cmd == BIOCSETWF) {
1051 		old = d->bd_wfilter;
1052 		wfilter = 1;
1053 	} else {
1054 		wfilter = 0;
1055 		old = d->bd_rfilter;
1056 	}
1057 	if (fp->bf_insns == NULL) {
1058 		if (fp->bf_len != 0)
1059 			return(EINVAL);
1060 		if (wfilter)
1061 			d->bd_wfilter = NULL;
1062 		else
1063 			d->bd_rfilter = NULL;
1064 		bpf_resetd(d);
1065 		if (old != NULL)
1066 			kfree(old, M_BPF);
1067 		return(0);
1068 	}
1069 	flen = fp->bf_len;
1070 	if (flen > BPF_MAXINSNS)
1071 		return(EINVAL);
1072 
1073 	size = flen * sizeof *fp->bf_insns;
1074 	fcode = (struct bpf_insn *)kmalloc(size, M_BPF, M_WAITOK);
1075 	if (copyin(fp->bf_insns, fcode, size) == 0 &&
1076 	    bpf_validate(fcode, (int)flen)) {
1077 		if (wfilter)
1078 			d->bd_wfilter = fcode;
1079 		else
1080 			d->bd_rfilter = fcode;
1081 		bpf_resetd(d);
1082 		if (old != NULL)
1083 			kfree(old, M_BPF);
1084 
1085 		return(0);
1086 	}
1087 	kfree(fcode, M_BPF);
1088 	return(EINVAL);
1089 }
1090 
1091 /*
1092  * Detach a file from its current interface (if attached at all) and attach
1093  * to the interface indicated by the name stored in ifr.
1094  * Return an errno or 0.
1095  */
1096 static int
1097 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1098 {
1099 	struct bpf_if *bp;
1100 	int error;
1101 	struct ifnet *theywant;
1102 
1103 	ifnet_lock();
1104 
1105 	theywant = ifunit(ifr->ifr_name);
1106 	if (theywant == NULL) {
1107 		ifnet_unlock();
1108 		return(ENXIO);
1109 	}
1110 
1111 	/*
1112 	 * Look through attached interfaces for the named one.
1113 	 */
1114 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1115 		struct ifnet *ifp = bp->bif_ifp;
1116 
1117 		if (ifp == NULL || ifp != theywant)
1118 			continue;
1119 		/* skip additional entry */
1120 		if (bp->bif_driverp != &ifp->if_bpf)
1121 			continue;
1122 		/*
1123 		 * We found the requested interface.
1124 		 * Allocate the packet buffers if we need to.
1125 		 * If we're already attached to requested interface,
1126 		 * just flush the buffer.
1127 		 */
1128 		if (d->bd_sbuf == NULL) {
1129 			error = bpf_allocbufs(d);
1130 			if (error != 0) {
1131 				ifnet_unlock();
1132 				return(error);
1133 			}
1134 		}
1135 		if (bp != d->bd_bif) {
1136 			if (d->bd_bif != NULL) {
1137 				/*
1138 				 * Detach if attached to something else.
1139 				 */
1140 				bpf_detachd(d);
1141 			}
1142 
1143 			bpf_attachd(d, bp);
1144 		}
1145 		bpf_resetd(d);
1146 
1147 		ifnet_unlock();
1148 		return(0);
1149 	}
1150 
1151 	ifnet_unlock();
1152 
1153 	/* Not found. */
1154 	return(ENXIO);
1155 }
1156 
1157 static struct filterops bpf_read_filtops =
1158 	{ FILTEROP_ISFD, NULL, bpf_filter_detach, bpf_filter_read };
1159 
1160 static int
1161 bpfkqfilter(struct dev_kqfilter_args *ap)
1162 {
1163 	cdev_t dev = ap->a_head.a_dev;
1164 	struct knote *kn = ap->a_kn;
1165 	struct klist *klist;
1166 	struct bpf_d *d;
1167 
1168 	lwkt_gettoken(&bpf_token);
1169 	d = dev->si_drv1;
1170 	if (d->bd_bif == NULL) {
1171 		ap->a_result = 1;
1172 		lwkt_reltoken(&bpf_token);
1173 		return (0);
1174 	}
1175 
1176 	ap->a_result = 0;
1177 	switch (kn->kn_filter) {
1178 	case EVFILT_READ:
1179 		kn->kn_fop = &bpf_read_filtops;
1180 		kn->kn_hook = (caddr_t)d;
1181 		break;
1182 	default:
1183 		ap->a_result = EOPNOTSUPP;
1184 		lwkt_reltoken(&bpf_token);
1185 		return (0);
1186 	}
1187 
1188 	klist = &d->bd_kq.ki_note;
1189 	knote_insert(klist, kn);
1190 	lwkt_reltoken(&bpf_token);
1191 
1192 	return (0);
1193 }
1194 
1195 static void
1196 bpf_filter_detach(struct knote *kn)
1197 {
1198 	struct klist *klist;
1199 	struct bpf_d *d;
1200 
1201 	d = (struct bpf_d *)kn->kn_hook;
1202 	klist = &d->bd_kq.ki_note;
1203 	knote_remove(klist, kn);
1204 }
1205 
1206 static int
1207 bpf_filter_read(struct knote *kn, long hint)
1208 {
1209 	struct bpf_d *d;
1210 	int ready = 0;
1211 
1212 	d = (struct bpf_d *)kn->kn_hook;
1213 	if (d->bd_hlen != 0 ||
1214 	    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1215 	    d->bd_slen != 0)) {
1216 		ready = 1;
1217 	} else {
1218 		/* Start the read timeout if necessary. */
1219 		if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1220 			callout_reset(&d->bd_callout, d->bd_rtout,
1221 			    bpf_timed_out, d);
1222 			d->bd_state = BPF_WAITING;
1223 		}
1224 	}
1225 
1226 	return (ready);
1227 }
1228 
1229 
1230 /*
1231  * Process the packet pkt of length pktlen.  The packet is parsed
1232  * by each listener's filter, and if accepted, stashed into the
1233  * corresponding buffer.
1234  */
1235 void
1236 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1237 {
1238 	struct bpf_d *d;
1239 	struct timeval tv;
1240 	int gottime = 0;
1241 	u_int slen;
1242 
1243 	lwkt_gettoken(&bpf_token);
1244 	/* Re-check */
1245 	if (bp == NULL) {
1246 		lwkt_reltoken(&bpf_token);
1247 		return;
1248 	}
1249 
1250 	/*
1251 	 * Note that the ipl does not have to be raised at this point.
1252 	 * The only problem that could arise here is that if two different
1253 	 * interfaces shared any data.  This is not the case.
1254 	 */
1255 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1256 		++d->bd_rcount;
1257 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1258 		if (slen != 0) {
1259 			if (!gottime) {
1260 				microtime(&tv);
1261 				gottime = 1;
1262 			}
1263 			catchpacket(d, pkt, pktlen, slen, _bcopy, &tv);
1264 		}
1265 	}
1266 	lwkt_reltoken(&bpf_token);
1267 }
1268 
1269 /*
1270  * Copy data from an mbuf chain into a buffer.  This code is derived
1271  * from m_copydata in sys/uipc_mbuf.c.
1272  */
1273 static void
1274 bpf_mcopy(volatile const void *src_arg, volatile void *dst_arg, size_t len)
1275 {
1276 	volatile const struct mbuf *m;
1277 	u_int count;
1278 	volatile u_char *dst;
1279 
1280 	m = src_arg;
1281 	dst = dst_arg;
1282 	while (len > 0) {
1283 		if (m == NULL)
1284 			panic("bpf_mcopy");
1285 		count = min(m->m_len, len);
1286 		bcopy(mtod(m, void *), dst, count);
1287 		m = m->m_next;
1288 		dst += count;
1289 		len -= count;
1290 	}
1291 }
1292 
1293 /*
1294  * Process the packet in the mbuf chain m.  The packet is parsed by each
1295  * listener's filter, and if accepted, stashed into the corresponding
1296  * buffer.
1297  */
1298 void
1299 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1300 {
1301 	struct bpf_d *d;
1302 	u_int pktlen, slen;
1303 	struct timeval tv;
1304 	int gottime = 0;
1305 
1306 	lwkt_gettoken(&bpf_token);
1307 	/* Re-check */
1308 	if (bp == NULL) {
1309 		lwkt_reltoken(&bpf_token);
1310 		return;
1311 	}
1312 
1313 	/* Don't compute pktlen, if no descriptor is attached. */
1314 	if (SLIST_EMPTY(&bp->bif_dlist)) {
1315 		lwkt_reltoken(&bpf_token);
1316 		return;
1317 	}
1318 
1319 	pktlen = m_lengthm(m, NULL);
1320 
1321 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1322 		if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL))
1323 			continue;
1324 		++d->bd_rcount;
1325 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1326 		if (slen != 0) {
1327 			if (!gottime) {
1328 				microtime(&tv);
1329 				gottime = 1;
1330 			}
1331 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy,
1332 				    &tv);
1333 		}
1334 	}
1335 	lwkt_reltoken(&bpf_token);
1336 }
1337 
1338 /*
1339  * Incoming linkage from device drivers, where we have a mbuf chain
1340  * but need to prepend some arbitrary header from a linear buffer.
1341  *
1342  * Con up a minimal dummy header to pacify bpf.  Allocate (only) a
1343  * struct m_hdr on the stack.  This is safe as bpf only reads from the
1344  * fields in this header that we initialize, and will not try to free
1345  * it or keep a pointer to it.
1346  */
1347 void
1348 bpf_mtap_hdr(struct bpf_if *arg, caddr_t data, u_int dlen, struct mbuf *m,
1349     u_int direction)
1350 {
1351 	struct m_hdr mh;
1352 
1353 	mh.mh_flags = 0;
1354 	mh.mh_next = m;
1355 	mh.mh_len = dlen;
1356 	mh.mh_data = data;
1357 
1358 	bpf_mtap(arg, (struct mbuf *) &mh);
1359 }
1360 
1361 void
1362 bpf_mtap_family(struct bpf_if *bp, struct mbuf *m, sa_family_t family)
1363 {
1364 	u_int family4;
1365 
1366 	KKASSERT(family != AF_UNSPEC);
1367 
1368 	family4 = (u_int)family;
1369 	bpf_ptap(bp, m, &family4, sizeof(family4));
1370 }
1371 
1372 /*
1373  * Process the packet in the mbuf chain m with the header in m prepended.
1374  * The packet is parsed by each listener's filter, and if accepted,
1375  * stashed into the corresponding buffer.
1376  */
1377 void
1378 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1379 {
1380 	struct mbuf mb;
1381 
1382 	/*
1383 	 * Craft on-stack mbuf suitable for passing to bpf_mtap.
1384 	 * Note that we cut corners here; we only setup what's
1385 	 * absolutely needed--this mbuf should never go anywhere else.
1386 	 */
1387 	mb.m_next = m;
1388 	mb.m_data = __DECONST(void *, data); /* LINTED */
1389 	mb.m_len = dlen;
1390 	mb.m_pkthdr.rcvif = m->m_pkthdr.rcvif;
1391 
1392 	bpf_mtap(bp, &mb);
1393 }
1394 
1395 /*
1396  * Move the packet data from interface memory (pkt) into the
1397  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1398  * otherwise 0.  "copy" is the routine called to do the actual data
1399  * transfer.  bcopy is passed in to copy contiguous chunks, while
1400  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1401  * pkt is really an mbuf.
1402  */
1403 static void
1404 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1405 	    void (*cpfn)(volatile const void *, volatile void *, size_t),
1406 	    const struct timeval *tv)
1407 {
1408 	struct bpf_hdr *hp;
1409 	int totlen, curlen;
1410 	int hdrlen = d->bd_bif->bif_hdrlen;
1411 	int wakeup = 0;
1412 	/*
1413 	 * Figure out how many bytes to move.  If the packet is
1414 	 * greater or equal to the snapshot length, transfer that
1415 	 * much.  Otherwise, transfer the whole packet (unless
1416 	 * we hit the buffer size limit).
1417 	 */
1418 	totlen = hdrlen + min(snaplen, pktlen);
1419 	if (totlen > d->bd_bufsize)
1420 		totlen = d->bd_bufsize;
1421 
1422 	/*
1423 	 * Round up the end of the previous packet to the next longword.
1424 	 */
1425 	curlen = BPF_WORDALIGN(d->bd_slen);
1426 	if (curlen + totlen > d->bd_bufsize) {
1427 		/*
1428 		 * This packet will overflow the storage buffer.
1429 		 * Rotate the buffers if we can, then wakeup any
1430 		 * pending reads.
1431 		 */
1432 		if (d->bd_fbuf == NULL) {
1433 			/*
1434 			 * We haven't completed the previous read yet,
1435 			 * so drop the packet.
1436 			 */
1437 			++d->bd_dcount;
1438 			return;
1439 		}
1440 		ROTATE_BUFFERS(d);
1441 		wakeup = 1;
1442 		curlen = 0;
1443 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1444 		/*
1445 		 * Immediate mode is set, or the read timeout has
1446 		 * already expired during a select call.  A packet
1447 		 * arrived, so the reader should be woken up.
1448 		 */
1449 		wakeup = 1;
1450 	}
1451 
1452 	/*
1453 	 * Append the bpf header.
1454 	 */
1455 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1456 	hp->bh_tstamp = *tv;
1457 	hp->bh_datalen = pktlen;
1458 	hp->bh_hdrlen = hdrlen;
1459 	/*
1460 	 * Copy the packet data into the store buffer and update its length.
1461 	 */
1462 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1463 	d->bd_slen = curlen + totlen;
1464 
1465 	if (wakeup)
1466 		bpf_wakeup(d);
1467 }
1468 
1469 /*
1470  * Initialize all nonzero fields of a descriptor.
1471  */
1472 static int
1473 bpf_allocbufs(struct bpf_d *d)
1474 {
1475 	d->bd_fbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1476 	d->bd_sbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1477 	d->bd_slen = 0;
1478 	d->bd_hlen = 0;
1479 	return(0);
1480 }
1481 
1482 /*
1483  * Free buffers and packet filter program currently in use by a descriptor.
1484  * Called on close.
1485  */
1486 static void
1487 bpf_freed(struct bpf_d *d)
1488 {
1489 	/*
1490 	 * We don't need to lock out interrupts since this descriptor has
1491 	 * been detached from its interface and it yet hasn't been marked
1492 	 * free.
1493 	 */
1494 	if (d->bd_sbuf != NULL) {
1495 		kfree(d->bd_sbuf, M_BPF);
1496 		if (d->bd_hbuf != NULL)
1497 			kfree(d->bd_hbuf, M_BPF);
1498 		if (d->bd_fbuf != NULL)
1499 			kfree(d->bd_fbuf, M_BPF);
1500 	}
1501 	if (d->bd_rfilter)
1502 		kfree(d->bd_rfilter, M_BPF);
1503 	if (d->bd_wfilter)
1504 		kfree(d->bd_wfilter, M_BPF);
1505 }
1506 
1507 /*
1508  * Attach an interface to bpf.  ifp is a pointer to the structure
1509  * defining the interface to be attached, dlt is the link layer type,
1510  * and hdrlen is the fixed size of the link header (variable length
1511  * headers are not yet supported).
1512  */
1513 void
1514 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1515 {
1516 	bpfattach_dlt(ifp, dlt, hdrlen, &ifp->if_bpf);
1517 }
1518 
1519 void
1520 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1521 {
1522 	struct bpf_if *bp;
1523 
1524 	bp = kmalloc(sizeof *bp, M_BPF, M_WAITOK | M_ZERO);
1525 
1526 	lwkt_gettoken(&bpf_token);
1527 
1528 	SLIST_INIT(&bp->bif_dlist);
1529 	bp->bif_ifp = ifp;
1530 	bp->bif_dlt = dlt;
1531 	bp->bif_driverp = driverp;
1532 	*bp->bif_driverp = NULL;
1533 
1534 	bp->bif_next = bpf_iflist;
1535 	bpf_iflist = bp;
1536 
1537 	/*
1538 	 * Compute the length of the bpf header.  This is not necessarily
1539 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1540 	 * that the network layer header begins on a longword boundary (for
1541 	 * performance reasons and to alleviate alignment restrictions).
1542 	 */
1543 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1544 
1545 	lwkt_reltoken(&bpf_token);
1546 
1547 	if (bootverbose)
1548 		if_printf(ifp, "bpf attached\n");
1549 }
1550 
1551 /*
1552  * Detach bpf from an interface.  This involves detaching each descriptor
1553  * associated with the interface, and leaving bd_bif NULL.  Notify each
1554  * descriptor as it's detached so that any sleepers wake up and get
1555  * ENXIO.
1556  */
1557 void
1558 bpfdetach(struct ifnet *ifp)
1559 {
1560 	struct bpf_if *bp, *bp_prev;
1561 	struct bpf_d *d;
1562 
1563 	lwkt_gettoken(&bpf_token);
1564 
1565 	/* Locate BPF interface information */
1566 	bp_prev = NULL;
1567 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1568 		if (ifp == bp->bif_ifp)
1569 			break;
1570 		bp_prev = bp;
1571 	}
1572 
1573 	/* Interface wasn't attached */
1574 	if (bp->bif_ifp == NULL) {
1575 		lwkt_reltoken(&bpf_token);
1576 		kprintf("bpfdetach: %s was not attached\n", ifp->if_xname);
1577 		return;
1578 	}
1579 
1580 	while ((d = SLIST_FIRST(&bp->bif_dlist)) != NULL) {
1581 		bpf_detachd(d);
1582 		bpf_wakeup(d);
1583 	}
1584 
1585 	if (bp_prev != NULL)
1586 		bp_prev->bif_next = bp->bif_next;
1587 	else
1588 		bpf_iflist = bp->bif_next;
1589 
1590 	kfree(bp, M_BPF);
1591 
1592 	lwkt_reltoken(&bpf_token);
1593 }
1594 
1595 /*
1596  * Get a list of available data link type of the interface.
1597  */
1598 static int
1599 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1600 {
1601 	int n, error;
1602 	struct ifnet *ifp;
1603 	struct bpf_if *bp;
1604 
1605 	ifp = d->bd_bif->bif_ifp;
1606 	n = 0;
1607 	error = 0;
1608 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1609 		if (bp->bif_ifp != ifp)
1610 			continue;
1611 		if (bfl->bfl_list != NULL) {
1612 			if (n >= bfl->bfl_len) {
1613 				return (ENOMEM);
1614 			}
1615 			error = copyout(&bp->bif_dlt,
1616 			    bfl->bfl_list + n, sizeof(u_int));
1617 		}
1618 		n++;
1619 	}
1620 	bfl->bfl_len = n;
1621 	return(error);
1622 }
1623 
1624 /*
1625  * Set the data link type of a BPF instance.
1626  */
1627 static int
1628 bpf_setdlt(struct bpf_d *d, u_int dlt)
1629 {
1630 	int error, opromisc;
1631 	struct ifnet *ifp;
1632 	struct bpf_if *bp;
1633 
1634 	if (d->bd_bif->bif_dlt == dlt)
1635 		return (0);
1636 	ifp = d->bd_bif->bif_ifp;
1637 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1638 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1639 			break;
1640 	}
1641 	if (bp != NULL) {
1642 		opromisc = d->bd_promisc;
1643 		bpf_detachd(d);
1644 		bpf_attachd(d, bp);
1645 		bpf_resetd(d);
1646 		if (opromisc) {
1647 			error = ifpromisc(bp->bif_ifp, 1);
1648 			if (error) {
1649 				if_printf(bp->bif_ifp,
1650 					"bpf_setdlt: ifpromisc failed (%d)\n",
1651 					error);
1652 			} else {
1653 				d->bd_promisc = 1;
1654 			}
1655 		}
1656 	}
1657 	return(bp == NULL ? EINVAL : 0);
1658 }
1659 
1660 void
1661 bpf_gettoken(void)
1662 {
1663 	lwkt_gettoken(&bpf_token);
1664 }
1665 
1666 void
1667 bpf_reltoken(void)
1668 {
1669 	lwkt_reltoken(&bpf_token);
1670 }
1671 
1672 static void
1673 bpf_drvinit(void *unused)
1674 {
1675 	int i;
1676 
1677 	make_autoclone_dev(&bpf_ops, &DEVFS_CLONE_BITMAP(bpf),
1678 		bpfclone, 0, 0, 0600, "bpf");
1679 	for (i = 0; i < BPF_PREALLOCATED_UNITS; i++) {
1680 		make_dev(&bpf_ops, i, 0, 0, 0600, "bpf%d", i);
1681 		devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(bpf), i);
1682 	}
1683 }
1684 
1685 static void
1686 bpf_drvuninit(void *unused)
1687 {
1688 	devfs_clone_handler_del("bpf");
1689 	dev_ops_remove_all(&bpf_ops);
1690 	devfs_clone_bitmap_uninit(&DEVFS_CLONE_BITMAP(bpf));
1691 }
1692 
1693 SYSINIT(bpfdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR, bpf_drvinit, NULL);
1694 SYSUNINIT(bpfdev, SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvuninit, NULL);
1695 
1696 #else /* !BPF */
1697 /*
1698  * NOP stubs to allow bpf-using drivers to load and function.
1699  *
1700  * A 'better' implementation would allow the core bpf functionality
1701  * to be loaded at runtime.
1702  */
1703 
1704 void
1705 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1706 {
1707 }
1708 
1709 void
1710 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1711 {
1712 }
1713 
1714 void
1715 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1716 {
1717 }
1718 
1719 void
1720 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1721 {
1722 }
1723 
1724 void
1725 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1726 {
1727 }
1728 
1729 void
1730 bpfdetach(struct ifnet *ifp)
1731 {
1732 }
1733 
1734 u_int
1735 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
1736 {
1737 	return -1;	/* "no filter" behaviour */
1738 }
1739 
1740 void
1741 bpf_gettoken(void)
1742 {
1743 }
1744 
1745 void
1746 bpf_reltoken(void)
1747 {
1748 }
1749 
1750 #endif /* !BPF */
1751