xref: /dragonfly/sys/net/bpf.c (revision 7c4f4eee)
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 <net/if.h>
61 #include <net/bpf.h>
62 #include <net/bpfdesc.h>
63 #include <net/netmsg2.h>
64 #include <net/netisr2.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/if_ether.h>
68 #include <sys/kernel.h>
69 #include <sys/sysctl.h>
70 
71 #include <netproto/802_11/ieee80211_dragonfly.h>
72 
73 #include <sys/devfs.h>
74 
75 struct netmsg_bpf_output {
76 	struct netmsg_base base;
77 	struct mbuf	*nm_mbuf;
78 	struct ifnet	*nm_ifp;
79 	struct sockaddr	*nm_dst;
80 	boolean_t 	nm_feedback;
81 };
82 
83 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
84 DEVFS_DEFINE_CLONE_BITMAP(bpf);
85 
86 #if NBPF <= 1
87 #define BPF_PREALLOCATED_UNITS	4
88 #else
89 #define BPF_PREALLOCATED_UNITS	NBPF
90 #endif
91 
92 #if NBPF > 0
93 
94 /*
95  * The default read buffer size is patchable.
96  */
97 static int bpf_bufsize = BPF_DEFAULTBUFSIZE;
98 SYSCTL_INT(_debug, OID_AUTO, bpf_bufsize, CTLFLAG_RW,
99    &bpf_bufsize, 0, "Current size of bpf buffer");
100 int bpf_maxbufsize = BPF_MAXBUFSIZE;
101 SYSCTL_INT(_debug, OID_AUTO, bpf_maxbufsize, CTLFLAG_RW,
102    &bpf_maxbufsize, 0, "Maximum size of bpf buffer");
103 
104 /*
105  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
106  */
107 static struct bpf_if	*bpf_iflist;
108 
109 static struct lwkt_token bpf_token = LWKT_TOKEN_INITIALIZER(bpf_token);
110 
111 static int	bpf_allocbufs(struct bpf_d *);
112 static void	bpf_attachd(struct bpf_d *d, struct bpf_if *bp);
113 static void	bpf_detachd(struct bpf_d *d);
114 static void	bpf_resetd(struct bpf_d *);
115 static void	bpf_freed(struct bpf_d *);
116 static void	bpf_mcopy(volatile const void *, volatile void *, size_t);
117 static int	bpf_movein(struct uio *, int, struct mbuf **,
118 			   struct sockaddr *, int *, struct bpf_insn *);
119 static int	bpf_setif(struct bpf_d *, struct ifreq *);
120 static void	bpf_timed_out(void *);
121 static void	bpf_wakeup(struct bpf_d *);
122 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
123 			    void (*)(volatile const void *,
124 				     volatile void *, size_t),
125 			    const struct timeval *);
126 static int	bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
127 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
128 static int	bpf_setdlt(struct bpf_d *, u_int);
129 static void	bpf_drvinit(void *unused);
130 static void	bpf_filter_detach(struct knote *kn);
131 static int	bpf_filter_read(struct knote *kn, long hint);
132 
133 static d_open_t		bpfopen;
134 static d_clone_t	bpfclone;
135 static d_close_t	bpfclose;
136 static d_read_t		bpfread;
137 static d_write_t	bpfwrite;
138 static d_ioctl_t	bpfioctl;
139 static d_kqfilter_t	bpfkqfilter;
140 
141 #define CDEV_MAJOR 23
142 static struct dev_ops bpf_ops = {
143 	{ "bpf", 0, D_MPSAFE },
144 	.d_open =	bpfopen,
145 	.d_close =	bpfclose,
146 	.d_read =	bpfread,
147 	.d_write =	bpfwrite,
148 	.d_ioctl =	bpfioctl,
149 	.d_kqfilter =	bpfkqfilter
150 };
151 
152 
153 static int
154 bpf_movein(struct uio *uio, int linktype, struct mbuf **mp,
155 	   struct sockaddr *sockp, int *datlen, struct bpf_insn *wfilter)
156 {
157 	const struct ieee80211_bpf_params *p;
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 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
210 		sockp->sa_family = AF_IEEE80211;
211 		hlen = 0;
212 		break;
213 
214 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
215 		sockp->sa_family = AF_IEEE80211;
216 		sockp->sa_len = 12;	/* XXX != 0 */
217 		hlen = sizeof(struct ieee80211_bpf_params);
218 		break;
219 
220 	default:
221 		return(EIO);
222 	}
223 
224 	len = uio->uio_resid;
225 	*datlen = len - hlen;
226 	if ((unsigned)len > MCLBYTES)
227 		return(EIO);
228 
229 	m = m_getl(len, M_WAITOK, MT_DATA, M_PKTHDR, NULL);
230 	if (m == NULL)
231 		return(ENOBUFS);
232 	m->m_pkthdr.len = m->m_len = len;
233 	m->m_pkthdr.rcvif = NULL;
234 	*mp = m;
235 
236 	if (m->m_len < hlen) {
237 		error = EPERM;
238 		goto bad;
239 	}
240 
241 	error = uiomove(mtod(m, u_char *), len, uio);
242 	if (error)
243 		goto bad;
244 
245 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
246 	if (slen == 0) {
247 		error = EPERM;
248 		goto bad;
249 	}
250 
251 	/*
252 	 * Make room for link header, and copy it to sockaddr.
253 	 */
254 	if (hlen != 0) {
255 		if (sockp->sa_family == AF_IEEE80211) {
256 			/*
257 			 * Collect true length from the parameter header
258 			 * NB: sockp is known to be zero'd so if we do a
259 			 *     short copy unspecified parameters will be
260 			 *     zero.
261 			 * NB: packet may not be aligned after stripping
262 			 *     bpf params
263 			 * XXX check ibp_vers
264 			 */
265 			p = mtod(m, const struct ieee80211_bpf_params *);
266 			hlen = p->ibp_len;
267 			if (hlen > sizeof(sockp->sa_data)) {
268 				error = EINVAL;
269 				goto bad;
270 			}
271 		}
272 		bcopy(m->m_data, sockp->sa_data, hlen);
273 		m->m_pkthdr.len -= hlen;
274 		m->m_len -= hlen;
275 		m->m_data += hlen; /* XXX */
276 	}
277 	return (0);
278 bad:
279 	m_freem(m);
280 	return(error);
281 }
282 
283 /*
284  * Attach file to the bpf interface, i.e. make d listen on bp.
285  * Must be called at splimp.
286  */
287 static void
288 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
289 {
290 	/*
291 	 * Point d at bp, and add d to the interface's list of listeners.
292 	 * Finally, point the driver's bpf cookie at the interface so
293 	 * it will divert packets to bpf.
294 	 */
295 	lwkt_gettoken(&bpf_token);
296 	d->bd_bif = bp;
297 	SLIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
298 	*bp->bif_driverp = bp;
299 
300 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
301 	lwkt_reltoken(&bpf_token);
302 }
303 
304 /*
305  * Detach a file from its interface.
306  */
307 static void
308 bpf_detachd(struct bpf_d *d)
309 {
310 	int error;
311 	struct bpf_if *bp;
312 	struct ifnet *ifp;
313 
314 	lwkt_gettoken(&bpf_token);
315 	bp = d->bd_bif;
316 	ifp = bp->bif_ifp;
317 
318 	/* Remove d from the interface's descriptor list. */
319 	SLIST_REMOVE(&bp->bif_dlist, d, bpf_d, bd_next);
320 
321 	if (SLIST_EMPTY(&bp->bif_dlist)) {
322 		/*
323 		 * Let the driver know that there are no more listeners.
324 		 */
325 		*bp->bif_driverp = NULL;
326 	}
327 	d->bd_bif = NULL;
328 
329 	EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
330 
331 	/*
332 	 * Check if this descriptor had requested promiscuous mode.
333 	 * If so, turn it off.
334 	 */
335 	if (d->bd_promisc) {
336 		d->bd_promisc = 0;
337 		error = ifpromisc(ifp, 0);
338 		if (error != 0 && error != ENXIO) {
339 			/*
340 			 * ENXIO can happen if a pccard is unplugged,
341 			 * Something is really wrong if we were able to put
342 			 * the driver into promiscuous mode, but can't
343 			 * take it out.
344 			 */
345 			if_printf(ifp, "bpf_detach: ifpromisc failed(%d)\n",
346 				  error);
347 		}
348 	}
349 	lwkt_reltoken(&bpf_token);
350 }
351 
352 /*
353  * Open ethernet device.  Returns ENXIO for illegal minor device number,
354  * EBUSY if file is open by another process.
355  */
356 /* ARGSUSED */
357 static int
358 bpfopen(struct dev_open_args *ap)
359 {
360 	cdev_t dev = ap->a_head.a_dev;
361 	struct bpf_d *d;
362 
363 	lwkt_gettoken(&bpf_token);
364 	if (ap->a_cred->cr_prison) {
365 		lwkt_reltoken(&bpf_token);
366 		return(EPERM);
367 	}
368 
369 	d = dev->si_drv1;
370 	/*
371 	 * Each minor can be opened by only one process.  If the requested
372 	 * minor is in use, return EBUSY.
373 	 */
374 	if (d != NULL) {
375 		lwkt_reltoken(&bpf_token);
376 		return(EBUSY);
377 	}
378 
379 	d = kmalloc(sizeof *d, M_BPF, M_WAITOK | M_ZERO);
380 	dev->si_drv1 = d;
381 	d->bd_bufsize = bpf_bufsize;
382 	d->bd_sig = SIGIO;
383 	d->bd_seesent = 1;
384 	d->bd_feedback = 0;
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 	int unit;
413 
414 	lwkt_gettoken(&bpf_token);
415 	funsetown(&d->bd_sigio);
416 	if (d->bd_state == BPF_WAITING)
417 		callout_stop(&d->bd_callout);
418 	d->bd_state = BPF_IDLE;
419 	if (d->bd_bif != NULL)
420 		bpf_detachd(d);
421 	bpf_freed(d);
422 	dev->si_drv1 = NULL;
423 
424 	unit = dev->si_uminor;
425 	if (unit >= BPF_PREALLOCATED_UNITS) {
426 		destroy_dev(dev);
427 		devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(bpf), unit);
428 	}
429 	kfree(d, M_BPF);
430 	lwkt_reltoken(&bpf_token);
431 
432 	return(0);
433 }
434 
435 /*
436  * Rotate the packet buffers in descriptor d.  Move the store buffer
437  * into the hold slot, and the free buffer into the store slot.
438  * Zero the length of the new store buffer.
439  */
440 #define ROTATE_BUFFERS(d) \
441 	(d)->bd_hbuf = (d)->bd_sbuf; \
442 	(d)->bd_hlen = (d)->bd_slen; \
443 	(d)->bd_sbuf = (d)->bd_fbuf; \
444 	(d)->bd_slen = 0; \
445 	(d)->bd_fbuf = NULL;
446 /*
447  *  bpfread - read next chunk of packets from buffers
448  */
449 static int
450 bpfread(struct dev_read_args *ap)
451 {
452 	cdev_t dev = ap->a_head.a_dev;
453 	struct bpf_d *d = dev->si_drv1;
454 	int timed_out;
455 	int error;
456 
457 	lwkt_gettoken(&bpf_token);
458 	/*
459 	 * Restrict application to use a buffer the same size as
460 	 * as kernel buffers.
461 	 */
462 	if (ap->a_uio->uio_resid != d->bd_bufsize) {
463 		lwkt_reltoken(&bpf_token);
464 		return(EINVAL);
465 	}
466 
467 	if (d->bd_state == BPF_WAITING)
468 		callout_stop(&d->bd_callout);
469 	timed_out = (d->bd_state == BPF_TIMED_OUT);
470 	d->bd_state = BPF_IDLE;
471 	/*
472 	 * If the hold buffer is empty, then do a timed sleep, which
473 	 * ends when the timeout expires or when enough packets
474 	 * have arrived to fill the store buffer.
475 	 */
476 	while (d->bd_hbuf == NULL) {
477 		if ((d->bd_immediate || (ap->a_ioflag & IO_NDELAY) || timed_out)
478 		    && d->bd_slen != 0) {
479 			/*
480 			 * A packet(s) either arrived since the previous,
481 			 * We're in immediate mode, or are reading
482 			 * in non-blocking mode, and a packet(s)
483 			 * either arrived since the previous
484 			 * read or arrived while we were asleep.
485 			 * Rotate the buffers and return what's here.
486 			 */
487 			ROTATE_BUFFERS(d);
488 			break;
489 		}
490 
491 		/*
492 		 * No data is available, check to see if the bpf device
493 		 * is still pointed at a real interface.  If not, return
494 		 * ENXIO so that the userland process knows to rebind
495 		 * it before using it again.
496 		 */
497 		if (d->bd_bif == NULL) {
498 			lwkt_reltoken(&bpf_token);
499 			return(ENXIO);
500 		}
501 
502 		if (ap->a_ioflag & IO_NDELAY) {
503 			lwkt_reltoken(&bpf_token);
504 			return(EWOULDBLOCK);
505 		}
506 		error = tsleep(d, PCATCH, "bpf", d->bd_rtout);
507 		if (error == EINTR || error == ERESTART) {
508 			lwkt_reltoken(&bpf_token);
509 			return(error);
510 		}
511 		if (error == EWOULDBLOCK) {
512 			/*
513 			 * On a timeout, return what's in the buffer,
514 			 * which may be nothing.  If there is something
515 			 * in the store buffer, we can rotate the buffers.
516 			 */
517 			if (d->bd_hbuf)
518 				/*
519 				 * We filled up the buffer in between
520 				 * getting the timeout and arriving
521 				 * here, so we don't need to rotate.
522 				 */
523 				break;
524 
525 			if (d->bd_slen == 0) {
526 				lwkt_reltoken(&bpf_token);
527 				return(0);
528 			}
529 			ROTATE_BUFFERS(d);
530 			break;
531 		}
532 	}
533 	/*
534 	 * At this point, we know we have something in the hold slot.
535 	 */
536 
537 	/*
538 	 * Move data from hold buffer into user space.
539 	 * We know the entire buffer is transferred since
540 	 * we checked above that the read buffer is bpf_bufsize bytes.
541 	 */
542 	error = uiomove(d->bd_hbuf, d->bd_hlen, ap->a_uio);
543 
544 	d->bd_fbuf = d->bd_hbuf;
545 	d->bd_hbuf = NULL;
546 	d->bd_hlen = 0;
547 	lwkt_reltoken(&bpf_token);
548 
549 	return(error);
550 }
551 
552 
553 /*
554  * If there are processes sleeping on this descriptor, wake them up.
555  */
556 static void
557 bpf_wakeup(struct bpf_d *d)
558 {
559 	if (d->bd_state == BPF_WAITING) {
560 		callout_stop(&d->bd_callout);
561 		d->bd_state = BPF_IDLE;
562 	}
563 	wakeup(d);
564 	if (d->bd_async && d->bd_sig && d->bd_sigio)
565 		pgsigio(d->bd_sigio, d->bd_sig, 0);
566 
567 	KNOTE(&d->bd_kq.ki_note, 0);
568 }
569 
570 static void
571 bpf_timed_out(void *arg)
572 {
573 	struct bpf_d *d = (struct bpf_d *)arg;
574 
575 	if (d->bd_state == BPF_WAITING) {
576 		d->bd_state = BPF_TIMED_OUT;
577 		if (d->bd_slen != 0)
578 			bpf_wakeup(d);
579 	}
580 }
581 
582 static void
583 bpf_output_dispatch(netmsg_t msg)
584 {
585 	struct netmsg_bpf_output *bmsg = (struct netmsg_bpf_output *)msg;
586 	struct ifnet *ifp = bmsg->nm_ifp;
587 	struct mbuf *mc = NULL;
588 	int error;
589 
590 	if (bmsg->nm_feedback) {
591 		mc = m_dup(bmsg->nm_mbuf, M_NOWAIT);
592 		if (mc != NULL)
593 			mc->m_pkthdr.rcvif = ifp;
594 	}
595 
596 	/*
597 	 * The driver frees the mbuf.
598 	 */
599 	error = ifp->if_output(ifp, bmsg->nm_mbuf, bmsg->nm_dst, NULL);
600 	lwkt_replymsg(&msg->lmsg, error);
601 
602 	if (mc != NULL) {
603 		if (error == 0) {
604 			mc->m_flags &= ~M_HASH;
605 			(*ifp->if_input)(ifp, mc, NULL, -1);
606 		} else {
607 			m_freem(mc);
608 		}
609 	}
610 }
611 
612 static int
613 bpfwrite(struct dev_write_args *ap)
614 {
615 	cdev_t dev = ap->a_head.a_dev;
616 	struct bpf_d *d = dev->si_drv1;
617 	struct ifnet *ifp;
618 	struct mbuf *m;
619 	int error, ret;
620 	struct sockaddr dst;
621 	int datlen;
622 	struct netmsg_bpf_output bmsg;
623 
624 	lwkt_gettoken(&bpf_token);
625 	if (d->bd_bif == NULL) {
626 		lwkt_reltoken(&bpf_token);
627 		return(ENXIO);
628 	}
629 
630 	ifp = d->bd_bif->bif_ifp;
631 
632 	if (ap->a_uio->uio_resid == 0) {
633 		lwkt_reltoken(&bpf_token);
634 		return(0);
635 	}
636 
637 	error = bpf_movein(ap->a_uio, (int)d->bd_bif->bif_dlt, &m,
638 			   &dst, &datlen, d->bd_wfilter);
639 	if (error) {
640 		lwkt_reltoken(&bpf_token);
641 		return(error);
642 	}
643 
644 	if (datlen > ifp->if_mtu) {
645 		m_freem(m);
646 		lwkt_reltoken(&bpf_token);
647 		return(EMSGSIZE);
648 	}
649 
650 	if (d->bd_hdrcmplt)
651 		dst.sa_family = pseudo_AF_HDRCMPLT;
652 
653 	netmsg_init(&bmsg.base, NULL, &curthread->td_msgport,
654 		    0, bpf_output_dispatch);
655 	bmsg.nm_mbuf = m;
656 	bmsg.nm_ifp = ifp;
657 	bmsg.nm_dst = &dst;
658 
659 	if (d->bd_feedback)
660 		bmsg.nm_feedback = TRUE;
661 	else
662 		bmsg.nm_feedback = FALSE;
663 
664 	ret = lwkt_domsg(netisr_cpuport(0), &bmsg.base.lmsg, 0);
665 
666 	lwkt_reltoken(&bpf_token);
667 
668 	return ret;
669 }
670 
671 /*
672  * Reset a descriptor by flushing its packet buffer and clearing the
673  * receive and drop counts.  Should be called at splimp.
674  */
675 static void
676 bpf_resetd(struct bpf_d *d)
677 {
678 	if (d->bd_hbuf) {
679 		/* Free the hold buffer. */
680 		d->bd_fbuf = d->bd_hbuf;
681 		d->bd_hbuf = NULL;
682 	}
683 	d->bd_slen = 0;
684 	d->bd_hlen = 0;
685 	d->bd_rcount = 0;
686 	d->bd_dcount = 0;
687 }
688 
689 /*
690  *  FIONREAD		Check for read packet available.
691  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
692  *  BIOCGBLEN		Get buffer len [for read()].
693  *  BIOCSETF		Set ethernet read filter.
694  *  BIOCSETWF		Set ethernet write filter.
695  *  BIOCFLUSH		Flush read packet buffer.
696  *  BIOCPROMISC		Put interface into promiscuous mode.
697  *  BIOCGDLT		Get link layer type.
698  *  BIOCGETIF		Get interface name.
699  *  BIOCSETIF		Set interface.
700  *  BIOCSRTIMEOUT	Set read timeout.
701  *  BIOCGRTIMEOUT	Get read timeout.
702  *  BIOCGSTATS		Get packet stats.
703  *  BIOCIMMEDIATE	Set immediate mode.
704  *  BIOCVERSION		Get filter language version.
705  *  BIOCGHDRCMPLT	Get "header already complete" flag
706  *  BIOCSHDRCMPLT	Set "header already complete" flag
707  *  BIOCSFEEDBACK	Set packet feedback mode.
708  *  BIOCGFEEDBACK	Get packet feedback mode.
709  *  BIOCGSEESENT	Get "see packets sent" flag
710  *  BIOCSSEESENT	Set "see packets sent" flag
711  *  BIOCLOCK		Set "locked" flag
712  */
713 /* ARGSUSED */
714 static int
715 bpfioctl(struct dev_ioctl_args *ap)
716 {
717 	cdev_t dev = ap->a_head.a_dev;
718 	struct bpf_d *d = dev->si_drv1;
719 	int error = 0;
720 
721 	lwkt_gettoken(&bpf_token);
722 	if (d->bd_state == BPF_WAITING)
723 		callout_stop(&d->bd_callout);
724 	d->bd_state = BPF_IDLE;
725 
726 	if (d->bd_locked == 1) {
727 		switch (ap->a_cmd) {
728 		case BIOCGBLEN:
729 		case BIOCFLUSH:
730 		case BIOCGDLT:
731 		case BIOCGDLTLIST:
732 		case BIOCGETIF:
733 		case BIOCGRTIMEOUT:
734 		case BIOCGSTATS:
735 		case BIOCVERSION:
736 		case BIOCGRSIG:
737 		case BIOCGHDRCMPLT:
738 		case FIONREAD:
739 		case BIOCLOCK:
740 		case BIOCSRTIMEOUT:
741 		case BIOCIMMEDIATE:
742 		case TIOCGPGRP:
743 			break;
744 		default:
745 			lwkt_reltoken(&bpf_token);
746 			return (EPERM);
747 		}
748 	}
749 	switch (ap->a_cmd) {
750 	default:
751 		error = EINVAL;
752 		break;
753 
754 	/*
755 	 * Check for read packet available.
756 	 */
757 	case FIONREAD:
758 		{
759 			int n;
760 
761 			n = d->bd_slen;
762 			if (d->bd_hbuf)
763 				n += d->bd_hlen;
764 
765 			*(int *)ap->a_data = n;
766 			break;
767 		}
768 
769 	case SIOCGIFADDR:
770 		{
771 			struct ifnet *ifp;
772 
773 			if (d->bd_bif == NULL) {
774 				error = EINVAL;
775 			} else {
776 				ifp = d->bd_bif->bif_ifp;
777 				ifnet_serialize_all(ifp);
778 				error = ifp->if_ioctl(ifp, ap->a_cmd,
779 						      ap->a_data, ap->a_cred);
780 				ifnet_deserialize_all(ifp);
781 			}
782 			break;
783 		}
784 
785 	/*
786 	 * Get buffer len [for read()].
787 	 */
788 	case BIOCGBLEN:
789 		*(u_int *)ap->a_data = d->bd_bufsize;
790 		break;
791 
792 	/*
793 	 * Set buffer length.
794 	 */
795 	case BIOCSBLEN:
796 		if (d->bd_bif != NULL) {
797 			error = EINVAL;
798 		} else {
799 			u_int size = *(u_int *)ap->a_data;
800 
801 			if (size > bpf_maxbufsize)
802 				*(u_int *)ap->a_data = size = bpf_maxbufsize;
803 			else if (size < BPF_MINBUFSIZE)
804 				*(u_int *)ap->a_data = size = BPF_MINBUFSIZE;
805 			d->bd_bufsize = size;
806 		}
807 		break;
808 
809 	/*
810 	 * Set link layer read filter.
811 	 */
812 	case BIOCSETF:
813 	case BIOCSETWF:
814 		error = bpf_setf(d, (struct bpf_program *)ap->a_data,
815 			ap->a_cmd);
816 		break;
817 
818 	/*
819 	 * Flush read packet buffer.
820 	 */
821 	case BIOCFLUSH:
822 		bpf_resetd(d);
823 		break;
824 
825 	/*
826 	 * Put interface into promiscuous mode.
827 	 */
828 	case BIOCPROMISC:
829 		if (d->bd_bif == NULL) {
830 			/*
831 			 * No interface attached yet.
832 			 */
833 			error = EINVAL;
834 			break;
835 		}
836 		if (d->bd_promisc == 0) {
837 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
838 			if (error == 0)
839 				d->bd_promisc = 1;
840 		}
841 		break;
842 
843 	/*
844 	 * Get device parameters.
845 	 */
846 	case BIOCGDLT:
847 		if (d->bd_bif == NULL)
848 			error = EINVAL;
849 		else
850 			*(u_int *)ap->a_data = d->bd_bif->bif_dlt;
851 		break;
852 
853 	/*
854 	 * Get a list of supported data link types.
855 	 */
856 	case BIOCGDLTLIST:
857 		if (d->bd_bif == NULL) {
858 			error = EINVAL;
859 		} else {
860 			error = bpf_getdltlist(d,
861 				(struct bpf_dltlist *)ap->a_data);
862 		}
863 		break;
864 
865 	/*
866 	 * Set data link type.
867 	 */
868 	case BIOCSDLT:
869 		if (d->bd_bif == NULL)
870 			error = EINVAL;
871 		else
872 			error = bpf_setdlt(d, *(u_int *)ap->a_data);
873 		break;
874 
875 	/*
876 	 * Get interface name.
877 	 */
878 	case BIOCGETIF:
879 		if (d->bd_bif == NULL) {
880 			error = EINVAL;
881 		} else {
882 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
883 			struct ifreq *const ifr = (struct ifreq *)ap->a_data;
884 
885 			strlcpy(ifr->ifr_name, ifp->if_xname,
886 				sizeof ifr->ifr_name);
887 		}
888 		break;
889 
890 	/*
891 	 * Set interface.
892 	 */
893 	case BIOCSETIF:
894 		error = bpf_setif(d, (struct ifreq *)ap->a_data);
895 		break;
896 
897 	/*
898 	 * Set read timeout.
899 	 */
900 	case BIOCSRTIMEOUT:
901 		{
902 			struct timeval *tv = (struct timeval *)ap->a_data;
903 
904 			/*
905 			 * Subtract 1 tick from tvtohz() since this isn't
906 			 * a one-shot timer.
907 			 */
908 			if ((error = itimerfix(tv)) == 0)
909 				d->bd_rtout = tvtohz_low(tv);
910 			break;
911 		}
912 
913 	/*
914 	 * Get read timeout.
915 	 */
916 	case BIOCGRTIMEOUT:
917 		{
918 			struct timeval *tv = (struct timeval *)ap->a_data;
919 
920 			tv->tv_sec = d->bd_rtout / hz;
921 			tv->tv_usec = (d->bd_rtout % hz) * ustick;
922 			break;
923 		}
924 
925 	/*
926 	 * Get packet stats.
927 	 */
928 	case BIOCGSTATS:
929 		{
930 			struct bpf_stat *bs = (struct bpf_stat *)ap->a_data;
931 
932 			bs->bs_recv = d->bd_rcount;
933 			bs->bs_drop = d->bd_dcount;
934 			break;
935 		}
936 
937 	/*
938 	 * Set immediate mode.
939 	 */
940 	case BIOCIMMEDIATE:
941 		d->bd_immediate = *(u_int *)ap->a_data;
942 		break;
943 
944 	case BIOCVERSION:
945 		{
946 			struct bpf_version *bv = (struct bpf_version *)ap->a_data;
947 
948 			bv->bv_major = BPF_MAJOR_VERSION;
949 			bv->bv_minor = BPF_MINOR_VERSION;
950 			break;
951 		}
952 
953 	/*
954 	 * Get "header already complete" flag
955 	 */
956 	case BIOCGHDRCMPLT:
957 		*(u_int *)ap->a_data = d->bd_hdrcmplt;
958 		break;
959 
960 	/*
961 	 * Set "header already complete" flag
962 	 */
963 	case BIOCSHDRCMPLT:
964 		d->bd_hdrcmplt = *(u_int *)ap->a_data ? 1 : 0;
965 		break;
966 
967 	/*
968 	 * Get "see sent packets" flag
969 	 */
970 	case BIOCGSEESENT:
971 		*(u_int *)ap->a_data = d->bd_seesent;
972 		break;
973 
974 	/*
975 	 * Set "see sent packets" flag
976 	 */
977 	case BIOCSSEESENT:
978 		d->bd_seesent = *(u_int *)ap->a_data;
979 		break;
980 
981 	case FIOASYNC:		/* Send signal on receive packets */
982 		d->bd_async = *(int *)ap->a_data;
983 		break;
984 
985 	/*
986 	 * Set "feed packets from bpf back to input" mode
987 	 */
988 	case BIOCSFEEDBACK:
989 		d->bd_feedback = *(int *)ap->a_data;
990 		break;
991 
992 	/*
993 	 * Get "feed packets from bpf back to input" mode
994 	 */
995 	case BIOCGFEEDBACK:
996 		*(u_int *)ap->a_data = d->bd_feedback;
997 		break;
998 
999 	case FIOSETOWN:
1000 		error = fsetown(*(int *)ap->a_data, &d->bd_sigio);
1001 		break;
1002 
1003 	case FIOGETOWN:
1004 		*(int *)ap->a_data = fgetown(&d->bd_sigio);
1005 		break;
1006 
1007 	/* This is deprecated, FIOSETOWN should be used instead. */
1008 	case TIOCSPGRP:
1009 		error = fsetown(-(*(int *)ap->a_data), &d->bd_sigio);
1010 		break;
1011 
1012 	/* This is deprecated, FIOGETOWN should be used instead. */
1013 	case TIOCGPGRP:
1014 		*(int *)ap->a_data = -fgetown(&d->bd_sigio);
1015 		break;
1016 
1017 	case BIOCSRSIG:		/* Set receive signal */
1018 		{
1019 			u_int sig;
1020 
1021 			sig = *(u_int *)ap->a_data;
1022 
1023 			if (sig >= NSIG)
1024 				error = EINVAL;
1025 			else
1026 				d->bd_sig = sig;
1027 			break;
1028 		}
1029 	case BIOCGRSIG:
1030 		*(u_int *)ap->a_data = d->bd_sig;
1031 		break;
1032 	case BIOCLOCK:
1033 		d->bd_locked = 1;
1034 		break;
1035 	}
1036 	lwkt_reltoken(&bpf_token);
1037 
1038 	return(error);
1039 }
1040 
1041 /*
1042  * Set d's packet filter program to fp.  If this file already has a filter,
1043  * free it and replace it.  Returns EINVAL for bogus requests.
1044  */
1045 static int
1046 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1047 {
1048 	struct bpf_insn *fcode, *old;
1049 	u_int wfilter, flen, size;
1050 
1051 	if (cmd == BIOCSETWF) {
1052 		old = d->bd_wfilter;
1053 		wfilter = 1;
1054 	} else {
1055 		wfilter = 0;
1056 		old = d->bd_rfilter;
1057 	}
1058 	if (fp->bf_insns == NULL) {
1059 		if (fp->bf_len != 0)
1060 			return(EINVAL);
1061 		if (wfilter)
1062 			d->bd_wfilter = NULL;
1063 		else
1064 			d->bd_rfilter = NULL;
1065 		bpf_resetd(d);
1066 		if (old != NULL)
1067 			kfree(old, M_BPF);
1068 		return(0);
1069 	}
1070 	flen = fp->bf_len;
1071 	if (flen > BPF_MAXINSNS)
1072 		return(EINVAL);
1073 
1074 	size = flen * sizeof *fp->bf_insns;
1075 	fcode = (struct bpf_insn *)kmalloc(size, M_BPF, M_WAITOK);
1076 	if (copyin(fp->bf_insns, fcode, size) == 0 &&
1077 	    bpf_validate(fcode, (int)flen)) {
1078 		if (wfilter)
1079 			d->bd_wfilter = fcode;
1080 		else
1081 			d->bd_rfilter = fcode;
1082 		bpf_resetd(d);
1083 		if (old != NULL)
1084 			kfree(old, M_BPF);
1085 
1086 		return(0);
1087 	}
1088 	kfree(fcode, M_BPF);
1089 	return(EINVAL);
1090 }
1091 
1092 /*
1093  * Detach a file from its current interface (if attached at all) and attach
1094  * to the interface indicated by the name stored in ifr.
1095  * Return an errno or 0.
1096  */
1097 static int
1098 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1099 {
1100 	struct bpf_if *bp;
1101 	int error;
1102 	struct ifnet *theywant;
1103 
1104 	ifnet_lock();
1105 
1106 	theywant = ifunit(ifr->ifr_name);
1107 	if (theywant == NULL) {
1108 		ifnet_unlock();
1109 		return(ENXIO);
1110 	}
1111 
1112 	/*
1113 	 * Look through attached interfaces for the named one.
1114 	 */
1115 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1116 		struct ifnet *ifp = bp->bif_ifp;
1117 
1118 		if (ifp == NULL || ifp != theywant)
1119 			continue;
1120 		/* skip additional entry */
1121 		if (bp->bif_driverp != &ifp->if_bpf)
1122 			continue;
1123 		/*
1124 		 * We found the requested interface.
1125 		 * Allocate the packet buffers if we need to.
1126 		 * If we're already attached to requested interface,
1127 		 * just flush the buffer.
1128 		 */
1129 		if (d->bd_sbuf == NULL) {
1130 			error = bpf_allocbufs(d);
1131 			if (error != 0) {
1132 				ifnet_unlock();
1133 				return(error);
1134 			}
1135 		}
1136 		if (bp != d->bd_bif) {
1137 			if (d->bd_bif != NULL) {
1138 				/*
1139 				 * Detach if attached to something else.
1140 				 */
1141 				bpf_detachd(d);
1142 			}
1143 
1144 			bpf_attachd(d, bp);
1145 		}
1146 		bpf_resetd(d);
1147 
1148 		ifnet_unlock();
1149 		return(0);
1150 	}
1151 
1152 	ifnet_unlock();
1153 
1154 	/* Not found. */
1155 	return(ENXIO);
1156 }
1157 
1158 static struct filterops bpf_read_filtops =
1159 	{ FILTEROP_ISFD, NULL, bpf_filter_detach, bpf_filter_read };
1160 
1161 static int
1162 bpfkqfilter(struct dev_kqfilter_args *ap)
1163 {
1164 	cdev_t dev = ap->a_head.a_dev;
1165 	struct knote *kn = ap->a_kn;
1166 	struct klist *klist;
1167 	struct bpf_d *d;
1168 
1169 	lwkt_gettoken(&bpf_token);
1170 	d = dev->si_drv1;
1171 	if (d->bd_bif == NULL) {
1172 		ap->a_result = 1;
1173 		lwkt_reltoken(&bpf_token);
1174 		return (0);
1175 	}
1176 
1177 	ap->a_result = 0;
1178 	switch (kn->kn_filter) {
1179 	case EVFILT_READ:
1180 		kn->kn_fop = &bpf_read_filtops;
1181 		kn->kn_hook = (caddr_t)d;
1182 		break;
1183 	default:
1184 		ap->a_result = EOPNOTSUPP;
1185 		lwkt_reltoken(&bpf_token);
1186 		return (0);
1187 	}
1188 
1189 	klist = &d->bd_kq.ki_note;
1190 	knote_insert(klist, kn);
1191 	lwkt_reltoken(&bpf_token);
1192 
1193 	return (0);
1194 }
1195 
1196 static void
1197 bpf_filter_detach(struct knote *kn)
1198 {
1199 	struct klist *klist;
1200 	struct bpf_d *d;
1201 
1202 	d = (struct bpf_d *)kn->kn_hook;
1203 	klist = &d->bd_kq.ki_note;
1204 	knote_remove(klist, kn);
1205 }
1206 
1207 static int
1208 bpf_filter_read(struct knote *kn, long hint)
1209 {
1210 	struct bpf_d *d;
1211 	int ready = 0;
1212 
1213 	d = (struct bpf_d *)kn->kn_hook;
1214 	if (d->bd_hlen != 0 ||
1215 	    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1216 	    d->bd_slen != 0)) {
1217 		ready = 1;
1218 	} else {
1219 		/* Start the read timeout if necessary. */
1220 		if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1221 			callout_reset(&d->bd_callout, d->bd_rtout,
1222 			    bpf_timed_out, d);
1223 			d->bd_state = BPF_WAITING;
1224 		}
1225 	}
1226 
1227 	return (ready);
1228 }
1229 
1230 
1231 /*
1232  * Process the packet pkt of length pktlen.  The packet is parsed
1233  * by each listener's filter, and if accepted, stashed into the
1234  * corresponding buffer.
1235  */
1236 void
1237 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1238 {
1239 	struct bpf_d *d;
1240 	struct timeval tv;
1241 	int gottime = 0;
1242 	u_int slen;
1243 
1244 	lwkt_gettoken(&bpf_token);
1245 	/* Re-check */
1246 	if (bp == NULL) {
1247 		lwkt_reltoken(&bpf_token);
1248 		return;
1249 	}
1250 
1251 	/*
1252 	 * Note that the ipl does not have to be raised at this point.
1253 	 * The only problem that could arise here is that if two different
1254 	 * interfaces shared any data.  This is not the case.
1255 	 */
1256 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1257 		++d->bd_rcount;
1258 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1259 		if (slen != 0) {
1260 			if (!gottime) {
1261 				microtime(&tv);
1262 				gottime = 1;
1263 			}
1264 			catchpacket(d, pkt, pktlen, slen, _bcopy, &tv);
1265 		}
1266 	}
1267 	lwkt_reltoken(&bpf_token);
1268 }
1269 
1270 /*
1271  * Copy data from an mbuf chain into a buffer.  This code is derived
1272  * from m_copydata in sys/uipc_mbuf.c.
1273  */
1274 static void
1275 bpf_mcopy(volatile const void *src_arg, volatile void *dst_arg, size_t len)
1276 {
1277 	volatile const struct mbuf *m;
1278 	u_int count;
1279 	volatile u_char *dst;
1280 
1281 	m = src_arg;
1282 	dst = dst_arg;
1283 	while (len > 0) {
1284 		if (m == NULL)
1285 			panic("bpf_mcopy");
1286 		count = min(m->m_len, len);
1287 		bcopy(mtod(m, void *), dst, count);
1288 		m = m->m_next;
1289 		dst += count;
1290 		len -= count;
1291 	}
1292 }
1293 
1294 /*
1295  * Process the packet in the mbuf chain m.  The packet is parsed by each
1296  * listener's filter, and if accepted, stashed into the corresponding
1297  * buffer.
1298  */
1299 void
1300 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1301 {
1302 	struct bpf_d *d;
1303 	u_int pktlen, slen;
1304 	struct timeval tv;
1305 	int gottime = 0;
1306 
1307 	lwkt_gettoken(&bpf_token);
1308 	/* Re-check */
1309 	if (bp == NULL) {
1310 		lwkt_reltoken(&bpf_token);
1311 		return;
1312 	}
1313 
1314 	/* Don't compute pktlen, if no descriptor is attached. */
1315 	if (SLIST_EMPTY(&bp->bif_dlist)) {
1316 		lwkt_reltoken(&bpf_token);
1317 		return;
1318 	}
1319 
1320 	pktlen = m_lengthm(m, NULL);
1321 
1322 	SLIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1323 		if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL))
1324 			continue;
1325 		++d->bd_rcount;
1326 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1327 		if (slen != 0) {
1328 			if (!gottime) {
1329 				microtime(&tv);
1330 				gottime = 1;
1331 			}
1332 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy,
1333 				    &tv);
1334 		}
1335 	}
1336 	lwkt_reltoken(&bpf_token);
1337 }
1338 
1339 /*
1340  * Incoming linkage from device drivers, where we have a mbuf chain
1341  * but need to prepend some arbitrary header from a linear buffer.
1342  *
1343  * Con up a minimal dummy header to pacify bpf.  Allocate (only) a
1344  * struct m_hdr on the stack.  This is safe as bpf only reads from the
1345  * fields in this header that we initialize, and will not try to free
1346  * it or keep a pointer to it.
1347  */
1348 void
1349 bpf_mtap_hdr(struct bpf_if *arg, caddr_t data, u_int dlen, struct mbuf *m,
1350     u_int direction)
1351 {
1352 	struct m_hdr mh;
1353 
1354 	mh.mh_flags = 0;
1355 	mh.mh_next = m;
1356 	mh.mh_len = dlen;
1357 	mh.mh_data = data;
1358 
1359 	bpf_mtap(arg, (struct mbuf *) &mh);
1360 }
1361 
1362 void
1363 bpf_mtap_family(struct bpf_if *bp, struct mbuf *m, sa_family_t family)
1364 {
1365 	u_int family4;
1366 
1367 	KKASSERT(family != AF_UNSPEC);
1368 
1369 	family4 = (u_int)family;
1370 	bpf_ptap(bp, m, &family4, sizeof(family4));
1371 }
1372 
1373 /*
1374  * Process the packet in the mbuf chain m with the header in m prepended.
1375  * The packet is parsed by each listener's filter, and if accepted,
1376  * stashed into the corresponding buffer.
1377  */
1378 void
1379 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1380 {
1381 	struct mbuf mb;
1382 
1383 	/*
1384 	 * Craft on-stack mbuf suitable for passing to bpf_mtap.
1385 	 * Note that we cut corners here; we only setup what's
1386 	 * absolutely needed--this mbuf should never go anywhere else.
1387 	 */
1388 	mb.m_next = m;
1389 	mb.m_data = __DECONST(void *, data); /* LINTED */
1390 	mb.m_len = dlen;
1391 	mb.m_pkthdr.rcvif = m->m_pkthdr.rcvif;
1392 
1393 	bpf_mtap(bp, &mb);
1394 }
1395 
1396 /*
1397  * Move the packet data from interface memory (pkt) into the
1398  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1399  * otherwise 0.  "copy" is the routine called to do the actual data
1400  * transfer.  bcopy is passed in to copy contiguous chunks, while
1401  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1402  * pkt is really an mbuf.
1403  */
1404 static void
1405 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1406 	    void (*cpfn)(volatile const void *, volatile void *, size_t),
1407 	    const struct timeval *tv)
1408 {
1409 	struct bpf_hdr *hp;
1410 	int totlen, curlen;
1411 	int hdrlen = d->bd_bif->bif_hdrlen;
1412 	int wakeup = 0;
1413 	/*
1414 	 * Figure out how many bytes to move.  If the packet is
1415 	 * greater or equal to the snapshot length, transfer that
1416 	 * much.  Otherwise, transfer the whole packet (unless
1417 	 * we hit the buffer size limit).
1418 	 */
1419 	totlen = hdrlen + min(snaplen, pktlen);
1420 	if (totlen > d->bd_bufsize)
1421 		totlen = d->bd_bufsize;
1422 
1423 	/*
1424 	 * Round up the end of the previous packet to the next longword.
1425 	 */
1426 	curlen = BPF_WORDALIGN(d->bd_slen);
1427 	if (curlen + totlen > d->bd_bufsize) {
1428 		/*
1429 		 * This packet will overflow the storage buffer.
1430 		 * Rotate the buffers if we can, then wakeup any
1431 		 * pending reads.
1432 		 */
1433 		if (d->bd_fbuf == NULL) {
1434 			/*
1435 			 * We haven't completed the previous read yet,
1436 			 * so drop the packet.
1437 			 */
1438 			++d->bd_dcount;
1439 			return;
1440 		}
1441 		ROTATE_BUFFERS(d);
1442 		wakeup = 1;
1443 		curlen = 0;
1444 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1445 		/*
1446 		 * Immediate mode is set, or the read timeout has
1447 		 * already expired during a select call.  A packet
1448 		 * arrived, so the reader should be woken up.
1449 		 */
1450 		wakeup = 1;
1451 	}
1452 
1453 	/*
1454 	 * Append the bpf header.
1455 	 */
1456 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1457 	hp->bh_tstamp = *tv;
1458 	hp->bh_datalen = pktlen;
1459 	hp->bh_hdrlen = hdrlen;
1460 	/*
1461 	 * Copy the packet data into the store buffer and update its length.
1462 	 */
1463 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1464 	d->bd_slen = curlen + totlen;
1465 
1466 	if (wakeup)
1467 		bpf_wakeup(d);
1468 }
1469 
1470 /*
1471  * Initialize all nonzero fields of a descriptor.
1472  */
1473 static int
1474 bpf_allocbufs(struct bpf_d *d)
1475 {
1476 	d->bd_fbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1477 	d->bd_sbuf = kmalloc(d->bd_bufsize, M_BPF, M_WAITOK);
1478 	d->bd_slen = 0;
1479 	d->bd_hlen = 0;
1480 	return(0);
1481 }
1482 
1483 /*
1484  * Free buffers and packet filter program currently in use by a descriptor.
1485  * Called on close.
1486  */
1487 static void
1488 bpf_freed(struct bpf_d *d)
1489 {
1490 	/*
1491 	 * We don't need to lock out interrupts since this descriptor has
1492 	 * been detached from its interface and it yet hasn't been marked
1493 	 * free.
1494 	 */
1495 	if (d->bd_sbuf != NULL) {
1496 		kfree(d->bd_sbuf, M_BPF);
1497 		if (d->bd_hbuf != NULL)
1498 			kfree(d->bd_hbuf, M_BPF);
1499 		if (d->bd_fbuf != NULL)
1500 			kfree(d->bd_fbuf, M_BPF);
1501 	}
1502 	if (d->bd_rfilter)
1503 		kfree(d->bd_rfilter, M_BPF);
1504 	if (d->bd_wfilter)
1505 		kfree(d->bd_wfilter, M_BPF);
1506 }
1507 
1508 /*
1509  * Attach an interface to bpf.  ifp is a pointer to the structure
1510  * defining the interface to be attached, dlt is the link layer type,
1511  * and hdrlen is the fixed size of the link header (variable length
1512  * headers are not yet supported).
1513  */
1514 void
1515 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1516 {
1517 	bpfattach_dlt(ifp, dlt, hdrlen, &ifp->if_bpf);
1518 }
1519 
1520 void
1521 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1522 {
1523 	struct bpf_if *bp;
1524 
1525 	bp = kmalloc(sizeof *bp, M_BPF, M_WAITOK | M_ZERO);
1526 
1527 	lwkt_gettoken(&bpf_token);
1528 
1529 	SLIST_INIT(&bp->bif_dlist);
1530 	bp->bif_ifp = ifp;
1531 	bp->bif_dlt = dlt;
1532 	bp->bif_driverp = driverp;
1533 	*bp->bif_driverp = NULL;
1534 
1535 	bp->bif_next = bpf_iflist;
1536 	bpf_iflist = bp;
1537 
1538 	/*
1539 	 * Compute the length of the bpf header.  This is not necessarily
1540 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1541 	 * that the network layer header begins on a longword boundary (for
1542 	 * performance reasons and to alleviate alignment restrictions).
1543 	 */
1544 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1545 
1546 	lwkt_reltoken(&bpf_token);
1547 
1548 	if (bootverbose)
1549 		if_printf(ifp, "bpf attached\n");
1550 }
1551 
1552 /*
1553  * Detach bpf from an interface.  This involves detaching each descriptor
1554  * associated with the interface, and leaving bd_bif NULL.  Notify each
1555  * descriptor as it's detached so that any sleepers wake up and get
1556  * ENXIO.
1557  */
1558 void
1559 bpfdetach(struct ifnet *ifp)
1560 {
1561 	struct bpf_if *bp, *bp_prev;
1562 	struct bpf_d *d;
1563 
1564 	lwkt_gettoken(&bpf_token);
1565 
1566 	/* Locate BPF interface information */
1567 	bp_prev = NULL;
1568 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1569 		if (ifp == bp->bif_ifp)
1570 			break;
1571 		bp_prev = bp;
1572 	}
1573 
1574 	/* Interface wasn't attached */
1575 	if (bp->bif_ifp == NULL) {
1576 		lwkt_reltoken(&bpf_token);
1577 		kprintf("bpfdetach: %s was not attached\n", ifp->if_xname);
1578 		return;
1579 	}
1580 
1581 	while ((d = SLIST_FIRST(&bp->bif_dlist)) != NULL) {
1582 		bpf_detachd(d);
1583 		bpf_wakeup(d);
1584 	}
1585 
1586 	if (bp_prev != NULL)
1587 		bp_prev->bif_next = bp->bif_next;
1588 	else
1589 		bpf_iflist = bp->bif_next;
1590 
1591 	kfree(bp, M_BPF);
1592 
1593 	lwkt_reltoken(&bpf_token);
1594 }
1595 
1596 /*
1597  * Get a list of available data link type of the interface.
1598  */
1599 static int
1600 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1601 {
1602 	int n, error;
1603 	struct ifnet *ifp;
1604 	struct bpf_if *bp;
1605 
1606 	ifp = d->bd_bif->bif_ifp;
1607 	n = 0;
1608 	error = 0;
1609 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1610 		if (bp->bif_ifp != ifp)
1611 			continue;
1612 		if (bfl->bfl_list != NULL) {
1613 			if (n >= bfl->bfl_len) {
1614 				return (ENOMEM);
1615 			}
1616 			error = copyout(&bp->bif_dlt,
1617 			    bfl->bfl_list + n, sizeof(u_int));
1618 		}
1619 		n++;
1620 	}
1621 	bfl->bfl_len = n;
1622 	return(error);
1623 }
1624 
1625 /*
1626  * Set the data link type of a BPF instance.
1627  */
1628 static int
1629 bpf_setdlt(struct bpf_d *d, u_int dlt)
1630 {
1631 	int error, opromisc;
1632 	struct ifnet *ifp;
1633 	struct bpf_if *bp;
1634 
1635 	if (d->bd_bif->bif_dlt == dlt)
1636 		return (0);
1637 	ifp = d->bd_bif->bif_ifp;
1638 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1639 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1640 			break;
1641 	}
1642 	if (bp != NULL) {
1643 		opromisc = d->bd_promisc;
1644 		bpf_detachd(d);
1645 		bpf_attachd(d, bp);
1646 		bpf_resetd(d);
1647 		if (opromisc) {
1648 			error = ifpromisc(bp->bif_ifp, 1);
1649 			if (error) {
1650 				if_printf(bp->bif_ifp,
1651 					"bpf_setdlt: ifpromisc failed (%d)\n",
1652 					error);
1653 			} else {
1654 				d->bd_promisc = 1;
1655 			}
1656 		}
1657 	}
1658 	return(bp == NULL ? EINVAL : 0);
1659 }
1660 
1661 void
1662 bpf_gettoken(void)
1663 {
1664 	lwkt_gettoken(&bpf_token);
1665 }
1666 
1667 void
1668 bpf_reltoken(void)
1669 {
1670 	lwkt_reltoken(&bpf_token);
1671 }
1672 
1673 static void
1674 bpf_drvinit(void *unused)
1675 {
1676 	int i;
1677 
1678 	make_autoclone_dev(&bpf_ops, &DEVFS_CLONE_BITMAP(bpf),
1679 		bpfclone, 0, 0, 0600, "bpf");
1680 	for (i = 0; i < BPF_PREALLOCATED_UNITS; i++) {
1681 		make_dev(&bpf_ops, i, 0, 0, 0600, "bpf%d", i);
1682 		devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(bpf), i);
1683 	}
1684 }
1685 
1686 static void
1687 bpf_drvuninit(void *unused)
1688 {
1689 	devfs_clone_handler_del("bpf");
1690 	dev_ops_remove_all(&bpf_ops);
1691 	devfs_clone_bitmap_uninit(&DEVFS_CLONE_BITMAP(bpf));
1692 }
1693 
1694 SYSINIT(bpfdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE+CDEV_MAJOR, bpf_drvinit, NULL);
1695 SYSUNINIT(bpfdev, SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvuninit, NULL);
1696 
1697 #else /* !BPF */
1698 /*
1699  * NOP stubs to allow bpf-using drivers to load and function.
1700  *
1701  * A 'better' implementation would allow the core bpf functionality
1702  * to be loaded at runtime.
1703  */
1704 
1705 void
1706 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1707 {
1708 }
1709 
1710 void
1711 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1712 {
1713 }
1714 
1715 void
1716 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen)
1717 {
1718 }
1719 
1720 void
1721 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1722 {
1723 }
1724 
1725 void
1726 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1727 {
1728 }
1729 
1730 void
1731 bpfdetach(struct ifnet *ifp)
1732 {
1733 }
1734 
1735 u_int
1736 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
1737 {
1738 	return -1;	/* "no filter" behaviour */
1739 }
1740 
1741 void
1742 bpf_gettoken(void)
1743 {
1744 }
1745 
1746 void
1747 bpf_reltoken(void)
1748 {
1749 }
1750 
1751 #endif /* !BPF */
1752