xref: /freebsd/sys/net/bpf.c (revision 7cd2dcf07629713e5a3d60472cfe4701b705a167)
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  * 4. 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	8.4 (Berkeley) 1/9/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_bpf.h"
41 #include "opt_compat.h"
42 #include "opt_netgraph.h"
43 
44 #include <sys/types.h>
45 #include <sys/param.h>
46 #include <sys/lock.h>
47 #include <sys/rwlock.h>
48 #include <sys/systm.h>
49 #include <sys/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/jail.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/time.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/signalvar.h>
58 #include <sys/filio.h>
59 #include <sys/sockio.h>
60 #include <sys/ttycom.h>
61 #include <sys/uio.h>
62 
63 #include <sys/event.h>
64 #include <sys/file.h>
65 #include <sys/poll.h>
66 #include <sys/proc.h>
67 
68 #include <sys/socket.h>
69 
70 #include <net/if.h>
71 #define	BPF_INTERNAL
72 #include <net/bpf.h>
73 #include <net/bpf_buffer.h>
74 #ifdef BPF_JITTER
75 #include <net/bpf_jitter.h>
76 #endif
77 #include <net/bpf_zerocopy.h>
78 #include <net/bpfdesc.h>
79 #include <net/vnet.h>
80 
81 #include <netinet/in.h>
82 #include <netinet/if_ether.h>
83 #include <sys/kernel.h>
84 #include <sys/sysctl.h>
85 
86 #include <net80211/ieee80211_freebsd.h>
87 
88 #include <security/mac/mac_framework.h>
89 
90 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
91 
92 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
93 
94 #define PRINET  26			/* interruptible */
95 
96 #define	SIZEOF_BPF_HDR(type)	\
97     (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
98 
99 #ifdef COMPAT_FREEBSD32
100 #include <sys/mount.h>
101 #include <compat/freebsd32/freebsd32.h>
102 #define BPF_ALIGNMENT32 sizeof(int32_t)
103 #define BPF_WORDALIGN32(x) (((x)+(BPF_ALIGNMENT32-1))&~(BPF_ALIGNMENT32-1))
104 
105 #ifndef BURN_BRIDGES
106 /*
107  * 32-bit version of structure prepended to each packet.  We use this header
108  * instead of the standard one for 32-bit streams.  We mark the a stream as
109  * 32-bit the first time we see a 32-bit compat ioctl request.
110  */
111 struct bpf_hdr32 {
112 	struct timeval32 bh_tstamp;	/* time stamp */
113 	uint32_t	bh_caplen;	/* length of captured portion */
114 	uint32_t	bh_datalen;	/* original length of packet */
115 	uint16_t	bh_hdrlen;	/* length of bpf header (this struct
116 					   plus alignment padding) */
117 };
118 #endif
119 
120 struct bpf_program32 {
121 	u_int bf_len;
122 	uint32_t bf_insns;
123 };
124 
125 struct bpf_dltlist32 {
126 	u_int	bfl_len;
127 	u_int	bfl_list;
128 };
129 
130 #define	BIOCSETF32	_IOW('B', 103, struct bpf_program32)
131 #define	BIOCSRTIMEOUT32	_IOW('B', 109, struct timeval32)
132 #define	BIOCGRTIMEOUT32	_IOR('B', 110, struct timeval32)
133 #define	BIOCGDLTLIST32	_IOWR('B', 121, struct bpf_dltlist32)
134 #define	BIOCSETWF32	_IOW('B', 123, struct bpf_program32)
135 #define	BIOCSETFNR32	_IOW('B', 130, struct bpf_program32)
136 #endif
137 
138 /*
139  * bpf_iflist is a list of BPF interface structures, each corresponding to a
140  * specific DLT.  The same network interface might have several BPF interface
141  * structures registered by different layers in the stack (i.e., 802.11
142  * frames, ethernet frames, etc).
143  */
144 static LIST_HEAD(, bpf_if)	bpf_iflist;
145 static struct mtx	bpf_mtx;		/* bpf global lock */
146 static int		bpf_bpfd_cnt;
147 
148 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
149 static void	bpf_detachd(struct bpf_d *);
150 static void	bpf_detachd_locked(struct bpf_d *);
151 static void	bpf_freed(struct bpf_d *);
152 static int	bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
153 		    struct sockaddr *, int *, struct bpf_insn *);
154 static int	bpf_setif(struct bpf_d *, struct ifreq *);
155 static void	bpf_timed_out(void *);
156 static __inline void
157 		bpf_wakeup(struct bpf_d *);
158 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
159 		    void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
160 		    struct bintime *);
161 static void	reset_d(struct bpf_d *);
162 static int	bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
163 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
164 static int	bpf_setdlt(struct bpf_d *, u_int);
165 static void	filt_bpfdetach(struct knote *);
166 static int	filt_bpfread(struct knote *, long);
167 static void	bpf_drvinit(void *);
168 static int	bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
169 
170 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
171 int bpf_maxinsns = BPF_MAXINSNS;
172 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
173     &bpf_maxinsns, 0, "Maximum bpf program instructions");
174 static int bpf_zerocopy_enable = 0;
175 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
176     &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
177 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
178     bpf_stats_sysctl, "bpf statistics portal");
179 
180 static VNET_DEFINE(int, bpf_optimize_writers) = 0;
181 #define	V_bpf_optimize_writers VNET(bpf_optimize_writers)
182 SYSCTL_VNET_INT(_net_bpf, OID_AUTO, optimize_writers,
183     CTLFLAG_RW, &VNET_NAME(bpf_optimize_writers), 0,
184     "Do not send packets until BPF program is set");
185 
186 static	d_open_t	bpfopen;
187 static	d_read_t	bpfread;
188 static	d_write_t	bpfwrite;
189 static	d_ioctl_t	bpfioctl;
190 static	d_poll_t	bpfpoll;
191 static	d_kqfilter_t	bpfkqfilter;
192 
193 static struct cdevsw bpf_cdevsw = {
194 	.d_version =	D_VERSION,
195 	.d_open =	bpfopen,
196 	.d_read =	bpfread,
197 	.d_write =	bpfwrite,
198 	.d_ioctl =	bpfioctl,
199 	.d_poll =	bpfpoll,
200 	.d_name =	"bpf",
201 	.d_kqfilter =	bpfkqfilter,
202 };
203 
204 static struct filterops bpfread_filtops = {
205 	.f_isfd = 1,
206 	.f_detach = filt_bpfdetach,
207 	.f_event = filt_bpfread,
208 };
209 
210 eventhandler_tag	bpf_ifdetach_cookie = NULL;
211 
212 /*
213  * LOCKING MODEL USED BY BPF:
214  * Locks:
215  * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
216  * some global counters and every bpf_if reference.
217  * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
218  * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
219  *   used by bpf_mtap code.
220  *
221  * Lock order:
222  *
223  * Global lock, interface lock, descriptor lock
224  *
225  * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
226  * working model. In many places (like bpf_detachd) we start with BPF descriptor
227  * (and we need to at least rlock it to get reliable interface pointer). This
228  * gives us potential LOR. As a result, we use global lock to protect from bpf_if
229  * change in every such place.
230  *
231  * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
232  * 3) descriptor main wlock.
233  * Reading bd_bif can be protected by any of these locks, typically global lock.
234  *
235  * Changing read/write BPF filter is protected by the same three locks,
236  * the same applies for reading.
237  *
238  * Sleeping in global lock is not allowed due to bpfdetach() using it.
239  */
240 
241 /*
242  * Wrapper functions for various buffering methods.  If the set of buffer
243  * modes expands, we will probably want to introduce a switch data structure
244  * similar to protosw, et.
245  */
246 static void
247 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
248     u_int len)
249 {
250 
251 	BPFD_LOCK_ASSERT(d);
252 
253 	switch (d->bd_bufmode) {
254 	case BPF_BUFMODE_BUFFER:
255 		return (bpf_buffer_append_bytes(d, buf, offset, src, len));
256 
257 	case BPF_BUFMODE_ZBUF:
258 		d->bd_zcopy++;
259 		return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
260 
261 	default:
262 		panic("bpf_buf_append_bytes");
263 	}
264 }
265 
266 static void
267 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
268     u_int len)
269 {
270 
271 	BPFD_LOCK_ASSERT(d);
272 
273 	switch (d->bd_bufmode) {
274 	case BPF_BUFMODE_BUFFER:
275 		return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
276 
277 	case BPF_BUFMODE_ZBUF:
278 		d->bd_zcopy++;
279 		return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
280 
281 	default:
282 		panic("bpf_buf_append_mbuf");
283 	}
284 }
285 
286 /*
287  * This function gets called when the free buffer is re-assigned.
288  */
289 static void
290 bpf_buf_reclaimed(struct bpf_d *d)
291 {
292 
293 	BPFD_LOCK_ASSERT(d);
294 
295 	switch (d->bd_bufmode) {
296 	case BPF_BUFMODE_BUFFER:
297 		return;
298 
299 	case BPF_BUFMODE_ZBUF:
300 		bpf_zerocopy_buf_reclaimed(d);
301 		return;
302 
303 	default:
304 		panic("bpf_buf_reclaimed");
305 	}
306 }
307 
308 /*
309  * If the buffer mechanism has a way to decide that a held buffer can be made
310  * free, then it is exposed via the bpf_canfreebuf() interface.  (1) is
311  * returned if the buffer can be discarded, (0) is returned if it cannot.
312  */
313 static int
314 bpf_canfreebuf(struct bpf_d *d)
315 {
316 
317 	BPFD_LOCK_ASSERT(d);
318 
319 	switch (d->bd_bufmode) {
320 	case BPF_BUFMODE_ZBUF:
321 		return (bpf_zerocopy_canfreebuf(d));
322 	}
323 	return (0);
324 }
325 
326 /*
327  * Allow the buffer model to indicate that the current store buffer is
328  * immutable, regardless of the appearance of space.  Return (1) if the
329  * buffer is writable, and (0) if not.
330  */
331 static int
332 bpf_canwritebuf(struct bpf_d *d)
333 {
334 	BPFD_LOCK_ASSERT(d);
335 
336 	switch (d->bd_bufmode) {
337 	case BPF_BUFMODE_ZBUF:
338 		return (bpf_zerocopy_canwritebuf(d));
339 	}
340 	return (1);
341 }
342 
343 /*
344  * Notify buffer model that an attempt to write to the store buffer has
345  * resulted in a dropped packet, in which case the buffer may be considered
346  * full.
347  */
348 static void
349 bpf_buffull(struct bpf_d *d)
350 {
351 
352 	BPFD_LOCK_ASSERT(d);
353 
354 	switch (d->bd_bufmode) {
355 	case BPF_BUFMODE_ZBUF:
356 		bpf_zerocopy_buffull(d);
357 		break;
358 	}
359 }
360 
361 /*
362  * Notify the buffer model that a buffer has moved into the hold position.
363  */
364 void
365 bpf_bufheld(struct bpf_d *d)
366 {
367 
368 	BPFD_LOCK_ASSERT(d);
369 
370 	switch (d->bd_bufmode) {
371 	case BPF_BUFMODE_ZBUF:
372 		bpf_zerocopy_bufheld(d);
373 		break;
374 	}
375 }
376 
377 static void
378 bpf_free(struct bpf_d *d)
379 {
380 
381 	switch (d->bd_bufmode) {
382 	case BPF_BUFMODE_BUFFER:
383 		return (bpf_buffer_free(d));
384 
385 	case BPF_BUFMODE_ZBUF:
386 		return (bpf_zerocopy_free(d));
387 
388 	default:
389 		panic("bpf_buf_free");
390 	}
391 }
392 
393 static int
394 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
395 {
396 
397 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
398 		return (EOPNOTSUPP);
399 	return (bpf_buffer_uiomove(d, buf, len, uio));
400 }
401 
402 static int
403 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
404 {
405 
406 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
407 		return (EOPNOTSUPP);
408 	return (bpf_buffer_ioctl_sblen(d, i));
409 }
410 
411 static int
412 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
413 {
414 
415 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
416 		return (EOPNOTSUPP);
417 	return (bpf_zerocopy_ioctl_getzmax(td, d, i));
418 }
419 
420 static int
421 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
422 {
423 
424 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
425 		return (EOPNOTSUPP);
426 	return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
427 }
428 
429 static int
430 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
431 {
432 
433 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
434 		return (EOPNOTSUPP);
435 	return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
436 }
437 
438 /*
439  * General BPF functions.
440  */
441 static int
442 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
443     struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter)
444 {
445 	const struct ieee80211_bpf_params *p;
446 	struct ether_header *eh;
447 	struct mbuf *m;
448 	int error;
449 	int len;
450 	int hlen;
451 	int slen;
452 
453 	/*
454 	 * Build a sockaddr based on the data link layer type.
455 	 * We do this at this level because the ethernet header
456 	 * is copied directly into the data field of the sockaddr.
457 	 * In the case of SLIP, there is no header and the packet
458 	 * is forwarded as is.
459 	 * Also, we are careful to leave room at the front of the mbuf
460 	 * for the link level header.
461 	 */
462 	switch (linktype) {
463 
464 	case DLT_SLIP:
465 		sockp->sa_family = AF_INET;
466 		hlen = 0;
467 		break;
468 
469 	case DLT_EN10MB:
470 		sockp->sa_family = AF_UNSPEC;
471 		/* XXX Would MAXLINKHDR be better? */
472 		hlen = ETHER_HDR_LEN;
473 		break;
474 
475 	case DLT_FDDI:
476 		sockp->sa_family = AF_IMPLINK;
477 		hlen = 0;
478 		break;
479 
480 	case DLT_RAW:
481 		sockp->sa_family = AF_UNSPEC;
482 		hlen = 0;
483 		break;
484 
485 	case DLT_NULL:
486 		/*
487 		 * null interface types require a 4 byte pseudo header which
488 		 * corresponds to the address family of the packet.
489 		 */
490 		sockp->sa_family = AF_UNSPEC;
491 		hlen = 4;
492 		break;
493 
494 	case DLT_ATM_RFC1483:
495 		/*
496 		 * en atm driver requires 4-byte atm pseudo header.
497 		 * though it isn't standard, vpi:vci needs to be
498 		 * specified anyway.
499 		 */
500 		sockp->sa_family = AF_UNSPEC;
501 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
502 		break;
503 
504 	case DLT_PPP:
505 		sockp->sa_family = AF_UNSPEC;
506 		hlen = 4;	/* This should match PPP_HDRLEN */
507 		break;
508 
509 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
510 		sockp->sa_family = AF_IEEE80211;
511 		hlen = 0;
512 		break;
513 
514 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
515 		sockp->sa_family = AF_IEEE80211;
516 		sockp->sa_len = 12;	/* XXX != 0 */
517 		hlen = sizeof(struct ieee80211_bpf_params);
518 		break;
519 
520 	default:
521 		return (EIO);
522 	}
523 
524 	len = uio->uio_resid;
525 
526 	if (len - hlen > ifp->if_mtu)
527 		return (EMSGSIZE);
528 
529 	if ((unsigned)len > MJUM16BYTES)
530 		return (EIO);
531 
532 	if (len <= MHLEN)
533 		MGETHDR(m, M_WAIT, MT_DATA);
534 	else if (len <= MCLBYTES)
535 		m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
536 	else
537 		m = m_getjcl(M_WAIT, MT_DATA, M_PKTHDR,
538 #if (MJUMPAGESIZE > MCLBYTES)
539 		    len <= MJUMPAGESIZE ? MJUMPAGESIZE :
540 #endif
541 		    (len <= MJUM9BYTES ? MJUM9BYTES : MJUM16BYTES));
542 	m->m_pkthdr.len = m->m_len = len;
543 	m->m_pkthdr.rcvif = NULL;
544 	*mp = m;
545 
546 	if (m->m_len < hlen) {
547 		error = EPERM;
548 		goto bad;
549 	}
550 
551 	error = uiomove(mtod(m, u_char *), len, uio);
552 	if (error)
553 		goto bad;
554 
555 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
556 	if (slen == 0) {
557 		error = EPERM;
558 		goto bad;
559 	}
560 
561 	/* Check for multicast destination */
562 	switch (linktype) {
563 	case DLT_EN10MB:
564 		eh = mtod(m, struct ether_header *);
565 		if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
566 			if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
567 			    ETHER_ADDR_LEN) == 0)
568 				m->m_flags |= M_BCAST;
569 			else
570 				m->m_flags |= M_MCAST;
571 		}
572 		break;
573 	}
574 
575 	/*
576 	 * Make room for link header, and copy it to sockaddr
577 	 */
578 	if (hlen != 0) {
579 		if (sockp->sa_family == AF_IEEE80211) {
580 			/*
581 			 * Collect true length from the parameter header
582 			 * NB: sockp is known to be zero'd so if we do a
583 			 *     short copy unspecified parameters will be
584 			 *     zero.
585 			 * NB: packet may not be aligned after stripping
586 			 *     bpf params
587 			 * XXX check ibp_vers
588 			 */
589 			p = mtod(m, const struct ieee80211_bpf_params *);
590 			hlen = p->ibp_len;
591 			if (hlen > sizeof(sockp->sa_data)) {
592 				error = EINVAL;
593 				goto bad;
594 			}
595 		}
596 		bcopy(m->m_data, sockp->sa_data, hlen);
597 	}
598 	*hdrlen = hlen;
599 
600 	return (0);
601 bad:
602 	m_freem(m);
603 	return (error);
604 }
605 
606 /*
607  * Attach file to the bpf interface, i.e. make d listen on bp.
608  */
609 static void
610 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
611 {
612 	int op_w;
613 
614 	BPF_LOCK_ASSERT();
615 
616 	/*
617 	 * Save sysctl value to protect from sysctl change
618 	 * between reads
619 	 */
620 	op_w = V_bpf_optimize_writers;
621 
622 	if (d->bd_bif != NULL)
623 		bpf_detachd_locked(d);
624 	/*
625 	 * Point d at bp, and add d to the interface's list.
626 	 * Since there are many applicaiotns using BPF for
627 	 * sending raw packets only (dhcpd, cdpd are good examples)
628 	 * we can delay adding d to the list of active listeners until
629 	 * some filter is configured.
630 	 */
631 
632 	BPFIF_WLOCK(bp);
633 	BPFD_LOCK(d);
634 
635 	d->bd_bif = bp;
636 
637 	if (op_w != 0) {
638 		/* Add to writers-only list */
639 		LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
640 		/*
641 		 * We decrement bd_writer on every filter set operation.
642 		 * First BIOCSETF is done by pcap_open_live() to set up
643 		 * snap length. After that appliation usually sets its own filter
644 		 */
645 		d->bd_writer = 2;
646 	} else
647 		LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
648 
649 	BPFD_UNLOCK(d);
650 	BPFIF_WUNLOCK(bp);
651 
652 	bpf_bpfd_cnt++;
653 
654 	CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
655 	    __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
656 
657 	if (op_w == 0)
658 		EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
659 }
660 
661 /*
662  * Add d to the list of active bp filters.
663  * Reuqires bpf_attachd() to be called before
664  */
665 static void
666 bpf_upgraded(struct bpf_d *d)
667 {
668 	struct bpf_if *bp;
669 
670 	BPF_LOCK_ASSERT();
671 
672 	bp = d->bd_bif;
673 
674 	/*
675 	 * Filter can be set several times without specifying interface.
676 	 * Mark d as reader and exit.
677 	 */
678 	if (bp == NULL) {
679 		BPFD_LOCK(d);
680 		d->bd_writer = 0;
681 		BPFD_UNLOCK(d);
682 		return;
683 	}
684 
685 	BPFIF_WLOCK(bp);
686 	BPFD_LOCK(d);
687 
688 	/* Remove from writers-only list */
689 	LIST_REMOVE(d, bd_next);
690 	LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
691 	/* Mark d as reader */
692 	d->bd_writer = 0;
693 
694 	BPFD_UNLOCK(d);
695 	BPFIF_WUNLOCK(bp);
696 
697 	CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
698 
699 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
700 }
701 
702 /*
703  * Detach a file from its interface.
704  */
705 static void
706 bpf_detachd(struct bpf_d *d)
707 {
708 	BPF_LOCK();
709 	bpf_detachd_locked(d);
710 	BPF_UNLOCK();
711 }
712 
713 static void
714 bpf_detachd_locked(struct bpf_d *d)
715 {
716 	int error;
717 	struct bpf_if *bp;
718 	struct ifnet *ifp;
719 
720 	CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
721 
722 	BPF_LOCK_ASSERT();
723 
724 	/* Check if descriptor is attached */
725 	if ((bp = d->bd_bif) == NULL)
726 		return;
727 
728 	BPFIF_WLOCK(bp);
729 	BPFD_LOCK(d);
730 
731 	/* Save bd_writer value */
732 	error = d->bd_writer;
733 
734 	/*
735 	 * Remove d from the interface's descriptor list.
736 	 */
737 	LIST_REMOVE(d, bd_next);
738 
739 	ifp = bp->bif_ifp;
740 	d->bd_bif = NULL;
741 	BPFD_UNLOCK(d);
742 	BPFIF_WUNLOCK(bp);
743 
744 	bpf_bpfd_cnt--;
745 
746 	/* Call event handler iff d is attached */
747 	if (error == 0)
748 		EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
749 
750 	/*
751 	 * Check if this descriptor had requested promiscuous mode.
752 	 * If so, turn it off.
753 	 */
754 	if (d->bd_promisc) {
755 		d->bd_promisc = 0;
756 		CURVNET_SET(ifp->if_vnet);
757 		error = ifpromisc(ifp, 0);
758 		CURVNET_RESTORE();
759 		if (error != 0 && error != ENXIO) {
760 			/*
761 			 * ENXIO can happen if a pccard is unplugged
762 			 * Something is really wrong if we were able to put
763 			 * the driver into promiscuous mode, but can't
764 			 * take it out.
765 			 */
766 			if_printf(bp->bif_ifp,
767 				"bpf_detach: ifpromisc failed (%d)\n", error);
768 		}
769 	}
770 }
771 
772 /*
773  * Close the descriptor by detaching it from its interface,
774  * deallocating its buffers, and marking it free.
775  */
776 static void
777 bpf_dtor(void *data)
778 {
779 	struct bpf_d *d = data;
780 
781 	BPFD_LOCK(d);
782 	if (d->bd_state == BPF_WAITING)
783 		callout_stop(&d->bd_callout);
784 	d->bd_state = BPF_IDLE;
785 	BPFD_UNLOCK(d);
786 	funsetown(&d->bd_sigio);
787 	bpf_detachd(d);
788 #ifdef MAC
789 	mac_bpfdesc_destroy(d);
790 #endif /* MAC */
791 	seldrain(&d->bd_sel);
792 	knlist_destroy(&d->bd_sel.si_note);
793 	callout_drain(&d->bd_callout);
794 	bpf_freed(d);
795 	free(d, M_BPF);
796 }
797 
798 /*
799  * Open ethernet device.  Returns ENXIO for illegal minor device number,
800  * EBUSY if file is open by another process.
801  */
802 /* ARGSUSED */
803 static	int
804 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
805 {
806 	struct bpf_d *d;
807 	int error, size;
808 
809 	d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
810 	error = devfs_set_cdevpriv(d, bpf_dtor);
811 	if (error != 0) {
812 		free(d, M_BPF);
813 		return (error);
814 	}
815 
816 	/*
817 	 * For historical reasons, perform a one-time initialization call to
818 	 * the buffer routines, even though we're not yet committed to a
819 	 * particular buffer method.
820 	 */
821 	bpf_buffer_init(d);
822 	d->bd_bufmode = BPF_BUFMODE_BUFFER;
823 	d->bd_sig = SIGIO;
824 	d->bd_direction = BPF_D_INOUT;
825 	BPF_PID_REFRESH(d, td);
826 #ifdef MAC
827 	mac_bpfdesc_init(d);
828 	mac_bpfdesc_create(td->td_ucred, d);
829 #endif
830 	mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
831 	callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
832 	knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
833 
834 	/* Allocate default buffers */
835 	size = d->bd_bufsize;
836 	bpf_buffer_ioctl_sblen(d, &size);
837 
838 	return (0);
839 }
840 
841 /*
842  *  bpfread - read next chunk of packets from buffers
843  */
844 static	int
845 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
846 {
847 	struct bpf_d *d;
848 	int error;
849 	int non_block;
850 	int timed_out;
851 
852 	error = devfs_get_cdevpriv((void **)&d);
853 	if (error != 0)
854 		return (error);
855 
856 	/*
857 	 * Restrict application to use a buffer the same size as
858 	 * as kernel buffers.
859 	 */
860 	if (uio->uio_resid != d->bd_bufsize)
861 		return (EINVAL);
862 
863 	non_block = ((ioflag & O_NONBLOCK) != 0);
864 
865 	BPFD_LOCK(d);
866 	BPF_PID_REFRESH_CUR(d);
867 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
868 		BPFD_UNLOCK(d);
869 		return (EOPNOTSUPP);
870 	}
871 	if (d->bd_state == BPF_WAITING)
872 		callout_stop(&d->bd_callout);
873 	timed_out = (d->bd_state == BPF_TIMED_OUT);
874 	d->bd_state = BPF_IDLE;
875 	/*
876 	 * If the hold buffer is empty, then do a timed sleep, which
877 	 * ends when the timeout expires or when enough packets
878 	 * have arrived to fill the store buffer.
879 	 */
880 	while (d->bd_hbuf == NULL) {
881 		if (d->bd_slen != 0) {
882 			/*
883 			 * A packet(s) either arrived since the previous
884 			 * read or arrived while we were asleep.
885 			 */
886 			if (d->bd_immediate || non_block || timed_out) {
887 				/*
888 				 * Rotate the buffers and return what's here
889 				 * if we are in immediate mode, non-blocking
890 				 * flag is set, or this descriptor timed out.
891 				 */
892 				ROTATE_BUFFERS(d);
893 				break;
894 			}
895 		}
896 
897 		/*
898 		 * No data is available, check to see if the bpf device
899 		 * is still pointed at a real interface.  If not, return
900 		 * ENXIO so that the userland process knows to rebind
901 		 * it before using it again.
902 		 */
903 		if (d->bd_bif == NULL) {
904 			BPFD_UNLOCK(d);
905 			return (ENXIO);
906 		}
907 
908 		if (non_block) {
909 			BPFD_UNLOCK(d);
910 			return (EWOULDBLOCK);
911 		}
912 		error = msleep(d, &d->bd_lock, PRINET|PCATCH,
913 		     "bpf", d->bd_rtout);
914 		if (error == EINTR || error == ERESTART) {
915 			BPFD_UNLOCK(d);
916 			return (error);
917 		}
918 		if (error == EWOULDBLOCK) {
919 			/*
920 			 * On a timeout, return what's in the buffer,
921 			 * which may be nothing.  If there is something
922 			 * in the store buffer, we can rotate the buffers.
923 			 */
924 			if (d->bd_hbuf)
925 				/*
926 				 * We filled up the buffer in between
927 				 * getting the timeout and arriving
928 				 * here, so we don't need to rotate.
929 				 */
930 				break;
931 
932 			if (d->bd_slen == 0) {
933 				BPFD_UNLOCK(d);
934 				return (0);
935 			}
936 			ROTATE_BUFFERS(d);
937 			break;
938 		}
939 	}
940 	/*
941 	 * At this point, we know we have something in the hold slot.
942 	 */
943 	BPFD_UNLOCK(d);
944 
945 	/*
946 	 * Move data from hold buffer into user space.
947 	 * We know the entire buffer is transferred since
948 	 * we checked above that the read buffer is bpf_bufsize bytes.
949 	 *
950 	 * XXXRW: More synchronization needed here: what if a second thread
951 	 * issues a read on the same fd at the same time?  Don't want this
952 	 * getting invalidated.
953 	 */
954 	error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
955 
956 	BPFD_LOCK(d);
957 	d->bd_fbuf = d->bd_hbuf;
958 	d->bd_hbuf = NULL;
959 	d->bd_hlen = 0;
960 	bpf_buf_reclaimed(d);
961 	BPFD_UNLOCK(d);
962 
963 	return (error);
964 }
965 
966 /*
967  * If there are processes sleeping on this descriptor, wake them up.
968  */
969 static __inline void
970 bpf_wakeup(struct bpf_d *d)
971 {
972 
973 	BPFD_LOCK_ASSERT(d);
974 	if (d->bd_state == BPF_WAITING) {
975 		callout_stop(&d->bd_callout);
976 		d->bd_state = BPF_IDLE;
977 	}
978 	wakeup(d);
979 	if (d->bd_async && d->bd_sig && d->bd_sigio)
980 		pgsigio(&d->bd_sigio, d->bd_sig, 0);
981 
982 	selwakeuppri(&d->bd_sel, PRINET);
983 	KNOTE_LOCKED(&d->bd_sel.si_note, 0);
984 }
985 
986 static void
987 bpf_timed_out(void *arg)
988 {
989 	struct bpf_d *d = (struct bpf_d *)arg;
990 
991 	BPFD_LOCK_ASSERT(d);
992 
993 	if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
994 		return;
995 	if (d->bd_state == BPF_WAITING) {
996 		d->bd_state = BPF_TIMED_OUT;
997 		if (d->bd_slen != 0)
998 			bpf_wakeup(d);
999 	}
1000 }
1001 
1002 static int
1003 bpf_ready(struct bpf_d *d)
1004 {
1005 
1006 	BPFD_LOCK_ASSERT(d);
1007 
1008 	if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1009 		return (1);
1010 	if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1011 	    d->bd_slen != 0)
1012 		return (1);
1013 	return (0);
1014 }
1015 
1016 static int
1017 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1018 {
1019 	struct bpf_d *d;
1020 	struct ifnet *ifp;
1021 	struct mbuf *m, *mc;
1022 	struct sockaddr dst;
1023 	int error, hlen;
1024 
1025 	error = devfs_get_cdevpriv((void **)&d);
1026 	if (error != 0)
1027 		return (error);
1028 
1029 	BPF_PID_REFRESH_CUR(d);
1030 	d->bd_wcount++;
1031 	/* XXX: locking required */
1032 	if (d->bd_bif == NULL) {
1033 		d->bd_wdcount++;
1034 		return (ENXIO);
1035 	}
1036 
1037 	ifp = d->bd_bif->bif_ifp;
1038 
1039 	if ((ifp->if_flags & IFF_UP) == 0) {
1040 		d->bd_wdcount++;
1041 		return (ENETDOWN);
1042 	}
1043 
1044 	if (uio->uio_resid == 0) {
1045 		d->bd_wdcount++;
1046 		return (0);
1047 	}
1048 
1049 	bzero(&dst, sizeof(dst));
1050 	m = NULL;
1051 	hlen = 0;
1052 	/* XXX: bpf_movein() can sleep */
1053 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1054 	    &m, &dst, &hlen, d->bd_wfilter);
1055 	if (error) {
1056 		d->bd_wdcount++;
1057 		return (error);
1058 	}
1059 	d->bd_wfcount++;
1060 	if (d->bd_hdrcmplt)
1061 		dst.sa_family = pseudo_AF_HDRCMPLT;
1062 
1063 	if (d->bd_feedback) {
1064 		mc = m_dup(m, M_DONTWAIT);
1065 		if (mc != NULL)
1066 			mc->m_pkthdr.rcvif = ifp;
1067 		/* Set M_PROMISC for outgoing packets to be discarded. */
1068 		if (d->bd_direction == BPF_D_INOUT)
1069 			m->m_flags |= M_PROMISC;
1070 	} else
1071 		mc = NULL;
1072 
1073 	m->m_pkthdr.len -= hlen;
1074 	m->m_len -= hlen;
1075 	m->m_data += hlen;	/* XXX */
1076 
1077 	CURVNET_SET(ifp->if_vnet);
1078 #ifdef MAC
1079 	BPFD_LOCK(d);
1080 	mac_bpfdesc_create_mbuf(d, m);
1081 	if (mc != NULL)
1082 		mac_bpfdesc_create_mbuf(d, mc);
1083 	BPFD_UNLOCK(d);
1084 #endif
1085 
1086 	error = (*ifp->if_output)(ifp, m, &dst, NULL);
1087 	if (error)
1088 		d->bd_wdcount++;
1089 
1090 	if (mc != NULL) {
1091 		if (error == 0)
1092 			(*ifp->if_input)(ifp, mc);
1093 		else
1094 			m_freem(mc);
1095 	}
1096 	CURVNET_RESTORE();
1097 
1098 	return (error);
1099 }
1100 
1101 /*
1102  * Reset a descriptor by flushing its packet buffer and clearing the receive
1103  * and drop counts.  This is doable for kernel-only buffers, but with
1104  * zero-copy buffers, we can't write to (or rotate) buffers that are
1105  * currently owned by userspace.  It would be nice if we could encapsulate
1106  * this logic in the buffer code rather than here.
1107  */
1108 static void
1109 reset_d(struct bpf_d *d)
1110 {
1111 
1112 	BPFD_LOCK_ASSERT(d);
1113 
1114 	if ((d->bd_hbuf != NULL) &&
1115 	    (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1116 		/* Free the hold buffer. */
1117 		d->bd_fbuf = d->bd_hbuf;
1118 		d->bd_hbuf = NULL;
1119 		d->bd_hlen = 0;
1120 		bpf_buf_reclaimed(d);
1121 	}
1122 	if (bpf_canwritebuf(d))
1123 		d->bd_slen = 0;
1124 	d->bd_rcount = 0;
1125 	d->bd_dcount = 0;
1126 	d->bd_fcount = 0;
1127 	d->bd_wcount = 0;
1128 	d->bd_wfcount = 0;
1129 	d->bd_wdcount = 0;
1130 	d->bd_zcopy = 0;
1131 }
1132 
1133 /*
1134  *  FIONREAD		Check for read packet available.
1135  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
1136  *  BIOCGBLEN		Get buffer len [for read()].
1137  *  BIOCSETF		Set read filter.
1138  *  BIOCSETFNR		Set read filter without resetting descriptor.
1139  *  BIOCSETWF		Set write filter.
1140  *  BIOCFLUSH		Flush read packet buffer.
1141  *  BIOCPROMISC		Put interface into promiscuous mode.
1142  *  BIOCGDLT		Get link layer type.
1143  *  BIOCGETIF		Get interface name.
1144  *  BIOCSETIF		Set interface.
1145  *  BIOCSRTIMEOUT	Set read timeout.
1146  *  BIOCGRTIMEOUT	Get read timeout.
1147  *  BIOCGSTATS		Get packet stats.
1148  *  BIOCIMMEDIATE	Set immediate mode.
1149  *  BIOCVERSION		Get filter language version.
1150  *  BIOCGHDRCMPLT	Get "header already complete" flag
1151  *  BIOCSHDRCMPLT	Set "header already complete" flag
1152  *  BIOCGDIRECTION	Get packet direction flag
1153  *  BIOCSDIRECTION	Set packet direction flag
1154  *  BIOCGTSTAMP		Get time stamp format and resolution.
1155  *  BIOCSTSTAMP		Set time stamp format and resolution.
1156  *  BIOCLOCK		Set "locked" flag
1157  *  BIOCFEEDBACK	Set packet feedback mode.
1158  *  BIOCSETZBUF		Set current zero-copy buffer locations.
1159  *  BIOCGETZMAX		Get maximum zero-copy buffer size.
1160  *  BIOCROTZBUF		Force rotation of zero-copy buffer
1161  *  BIOCSETBUFMODE	Set buffer mode.
1162  *  BIOCGETBUFMODE	Get current buffer mode.
1163  */
1164 /* ARGSUSED */
1165 static	int
1166 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1167     struct thread *td)
1168 {
1169 	struct bpf_d *d;
1170 	int error;
1171 
1172 	error = devfs_get_cdevpriv((void **)&d);
1173 	if (error != 0)
1174 		return (error);
1175 
1176 	/*
1177 	 * Refresh PID associated with this descriptor.
1178 	 */
1179 	BPFD_LOCK(d);
1180 	BPF_PID_REFRESH(d, td);
1181 	if (d->bd_state == BPF_WAITING)
1182 		callout_stop(&d->bd_callout);
1183 	d->bd_state = BPF_IDLE;
1184 	BPFD_UNLOCK(d);
1185 
1186 	if (d->bd_locked == 1) {
1187 		switch (cmd) {
1188 		case BIOCGBLEN:
1189 		case BIOCFLUSH:
1190 		case BIOCGDLT:
1191 		case BIOCGDLTLIST:
1192 #ifdef COMPAT_FREEBSD32
1193 		case BIOCGDLTLIST32:
1194 #endif
1195 		case BIOCGETIF:
1196 		case BIOCGRTIMEOUT:
1197 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1198 		case BIOCGRTIMEOUT32:
1199 #endif
1200 		case BIOCGSTATS:
1201 		case BIOCVERSION:
1202 		case BIOCGRSIG:
1203 		case BIOCGHDRCMPLT:
1204 		case BIOCSTSTAMP:
1205 		case BIOCFEEDBACK:
1206 		case FIONREAD:
1207 		case BIOCLOCK:
1208 		case BIOCSRTIMEOUT:
1209 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1210 		case BIOCSRTIMEOUT32:
1211 #endif
1212 		case BIOCIMMEDIATE:
1213 		case TIOCGPGRP:
1214 		case BIOCROTZBUF:
1215 			break;
1216 		default:
1217 			return (EPERM);
1218 		}
1219 	}
1220 #ifdef COMPAT_FREEBSD32
1221 	/*
1222 	 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1223 	 * that it will get 32-bit packet headers.
1224 	 */
1225 	switch (cmd) {
1226 	case BIOCSETF32:
1227 	case BIOCSETFNR32:
1228 	case BIOCSETWF32:
1229 	case BIOCGDLTLIST32:
1230 	case BIOCGRTIMEOUT32:
1231 	case BIOCSRTIMEOUT32:
1232 		BPFD_LOCK(d);
1233 		d->bd_compat32 = 1;
1234 		BPFD_UNLOCK(d);
1235 	}
1236 #endif
1237 
1238 	CURVNET_SET(TD_TO_VNET(td));
1239 	switch (cmd) {
1240 
1241 	default:
1242 		error = EINVAL;
1243 		break;
1244 
1245 	/*
1246 	 * Check for read packet available.
1247 	 */
1248 	case FIONREAD:
1249 		{
1250 			int n;
1251 
1252 			BPFD_LOCK(d);
1253 			n = d->bd_slen;
1254 			if (d->bd_hbuf)
1255 				n += d->bd_hlen;
1256 			BPFD_UNLOCK(d);
1257 
1258 			*(int *)addr = n;
1259 			break;
1260 		}
1261 
1262 	case SIOCGIFADDR:
1263 		{
1264 			struct ifnet *ifp;
1265 
1266 			if (d->bd_bif == NULL)
1267 				error = EINVAL;
1268 			else {
1269 				ifp = d->bd_bif->bif_ifp;
1270 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
1271 			}
1272 			break;
1273 		}
1274 
1275 	/*
1276 	 * Get buffer len [for read()].
1277 	 */
1278 	case BIOCGBLEN:
1279 		BPFD_LOCK(d);
1280 		*(u_int *)addr = d->bd_bufsize;
1281 		BPFD_UNLOCK(d);
1282 		break;
1283 
1284 	/*
1285 	 * Set buffer length.
1286 	 */
1287 	case BIOCSBLEN:
1288 		error = bpf_ioctl_sblen(d, (u_int *)addr);
1289 		break;
1290 
1291 	/*
1292 	 * Set link layer read filter.
1293 	 */
1294 	case BIOCSETF:
1295 	case BIOCSETFNR:
1296 	case BIOCSETWF:
1297 #ifdef COMPAT_FREEBSD32
1298 	case BIOCSETF32:
1299 	case BIOCSETFNR32:
1300 	case BIOCSETWF32:
1301 #endif
1302 		error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1303 		break;
1304 
1305 	/*
1306 	 * Flush read packet buffer.
1307 	 */
1308 	case BIOCFLUSH:
1309 		BPFD_LOCK(d);
1310 		reset_d(d);
1311 		BPFD_UNLOCK(d);
1312 		break;
1313 
1314 	/*
1315 	 * Put interface into promiscuous mode.
1316 	 */
1317 	case BIOCPROMISC:
1318 		if (d->bd_bif == NULL) {
1319 			/*
1320 			 * No interface attached yet.
1321 			 */
1322 			error = EINVAL;
1323 			break;
1324 		}
1325 		if (d->bd_promisc == 0) {
1326 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1327 			if (error == 0)
1328 				d->bd_promisc = 1;
1329 		}
1330 		break;
1331 
1332 	/*
1333 	 * Get current data link type.
1334 	 */
1335 	case BIOCGDLT:
1336 		BPF_LOCK();
1337 		if (d->bd_bif == NULL)
1338 			error = EINVAL;
1339 		else
1340 			*(u_int *)addr = d->bd_bif->bif_dlt;
1341 		BPF_UNLOCK();
1342 		break;
1343 
1344 	/*
1345 	 * Get a list of supported data link types.
1346 	 */
1347 #ifdef COMPAT_FREEBSD32
1348 	case BIOCGDLTLIST32:
1349 		{
1350 			struct bpf_dltlist32 *list32;
1351 			struct bpf_dltlist dltlist;
1352 
1353 			list32 = (struct bpf_dltlist32 *)addr;
1354 			dltlist.bfl_len = list32->bfl_len;
1355 			dltlist.bfl_list = PTRIN(list32->bfl_list);
1356 			BPF_LOCK();
1357 			if (d->bd_bif == NULL)
1358 				error = EINVAL;
1359 			else {
1360 				error = bpf_getdltlist(d, &dltlist);
1361 				if (error == 0)
1362 					list32->bfl_len = dltlist.bfl_len;
1363 			}
1364 			BPF_UNLOCK();
1365 			break;
1366 		}
1367 #endif
1368 
1369 	case BIOCGDLTLIST:
1370 		BPF_LOCK();
1371 		if (d->bd_bif == NULL)
1372 			error = EINVAL;
1373 		else
1374 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1375 		BPF_UNLOCK();
1376 		break;
1377 
1378 	/*
1379 	 * Set data link type.
1380 	 */
1381 	case BIOCSDLT:
1382 		BPF_LOCK();
1383 		if (d->bd_bif == NULL)
1384 			error = EINVAL;
1385 		else
1386 			error = bpf_setdlt(d, *(u_int *)addr);
1387 		BPF_UNLOCK();
1388 		break;
1389 
1390 	/*
1391 	 * Get interface name.
1392 	 */
1393 	case BIOCGETIF:
1394 		BPF_LOCK();
1395 		if (d->bd_bif == NULL)
1396 			error = EINVAL;
1397 		else {
1398 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
1399 			struct ifreq *const ifr = (struct ifreq *)addr;
1400 
1401 			strlcpy(ifr->ifr_name, ifp->if_xname,
1402 			    sizeof(ifr->ifr_name));
1403 		}
1404 		BPF_UNLOCK();
1405 		break;
1406 
1407 	/*
1408 	 * Set interface.
1409 	 */
1410 	case BIOCSETIF:
1411 		BPF_LOCK();
1412 		error = bpf_setif(d, (struct ifreq *)addr);
1413 		BPF_UNLOCK();
1414 		break;
1415 
1416 	/*
1417 	 * Set read timeout.
1418 	 */
1419 	case BIOCSRTIMEOUT:
1420 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1421 	case BIOCSRTIMEOUT32:
1422 #endif
1423 		{
1424 			struct timeval *tv = (struct timeval *)addr;
1425 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1426 			struct timeval32 *tv32;
1427 			struct timeval tv64;
1428 
1429 			if (cmd == BIOCSRTIMEOUT32) {
1430 				tv32 = (struct timeval32 *)addr;
1431 				tv = &tv64;
1432 				tv->tv_sec = tv32->tv_sec;
1433 				tv->tv_usec = tv32->tv_usec;
1434 			} else
1435 #endif
1436 				tv = (struct timeval *)addr;
1437 
1438 			/*
1439 			 * Subtract 1 tick from tvtohz() since this isn't
1440 			 * a one-shot timer.
1441 			 */
1442 			if ((error = itimerfix(tv)) == 0)
1443 				d->bd_rtout = tvtohz(tv) - 1;
1444 			break;
1445 		}
1446 
1447 	/*
1448 	 * Get read timeout.
1449 	 */
1450 	case BIOCGRTIMEOUT:
1451 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1452 	case BIOCGRTIMEOUT32:
1453 #endif
1454 		{
1455 			struct timeval *tv;
1456 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1457 			struct timeval32 *tv32;
1458 			struct timeval tv64;
1459 
1460 			if (cmd == BIOCGRTIMEOUT32)
1461 				tv = &tv64;
1462 			else
1463 #endif
1464 				tv = (struct timeval *)addr;
1465 
1466 			tv->tv_sec = d->bd_rtout / hz;
1467 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1468 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1469 			if (cmd == BIOCGRTIMEOUT32) {
1470 				tv32 = (struct timeval32 *)addr;
1471 				tv32->tv_sec = tv->tv_sec;
1472 				tv32->tv_usec = tv->tv_usec;
1473 			}
1474 #endif
1475 
1476 			break;
1477 		}
1478 
1479 	/*
1480 	 * Get packet stats.
1481 	 */
1482 	case BIOCGSTATS:
1483 		{
1484 			struct bpf_stat *bs = (struct bpf_stat *)addr;
1485 
1486 			/* XXXCSJP overflow */
1487 			bs->bs_recv = d->bd_rcount;
1488 			bs->bs_drop = d->bd_dcount;
1489 			break;
1490 		}
1491 
1492 	/*
1493 	 * Set immediate mode.
1494 	 */
1495 	case BIOCIMMEDIATE:
1496 		BPFD_LOCK(d);
1497 		d->bd_immediate = *(u_int *)addr;
1498 		BPFD_UNLOCK(d);
1499 		break;
1500 
1501 	case BIOCVERSION:
1502 		{
1503 			struct bpf_version *bv = (struct bpf_version *)addr;
1504 
1505 			bv->bv_major = BPF_MAJOR_VERSION;
1506 			bv->bv_minor = BPF_MINOR_VERSION;
1507 			break;
1508 		}
1509 
1510 	/*
1511 	 * Get "header already complete" flag
1512 	 */
1513 	case BIOCGHDRCMPLT:
1514 		BPFD_LOCK(d);
1515 		*(u_int *)addr = d->bd_hdrcmplt;
1516 		BPFD_UNLOCK(d);
1517 		break;
1518 
1519 	/*
1520 	 * Set "header already complete" flag
1521 	 */
1522 	case BIOCSHDRCMPLT:
1523 		BPFD_LOCK(d);
1524 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1525 		BPFD_UNLOCK(d);
1526 		break;
1527 
1528 	/*
1529 	 * Get packet direction flag
1530 	 */
1531 	case BIOCGDIRECTION:
1532 		BPFD_LOCK(d);
1533 		*(u_int *)addr = d->bd_direction;
1534 		BPFD_UNLOCK(d);
1535 		break;
1536 
1537 	/*
1538 	 * Set packet direction flag
1539 	 */
1540 	case BIOCSDIRECTION:
1541 		{
1542 			u_int	direction;
1543 
1544 			direction = *(u_int *)addr;
1545 			switch (direction) {
1546 			case BPF_D_IN:
1547 			case BPF_D_INOUT:
1548 			case BPF_D_OUT:
1549 				BPFD_LOCK(d);
1550 				d->bd_direction = direction;
1551 				BPFD_UNLOCK(d);
1552 				break;
1553 			default:
1554 				error = EINVAL;
1555 			}
1556 		}
1557 		break;
1558 
1559 	/*
1560 	 * Get packet timestamp format and resolution.
1561 	 */
1562 	case BIOCGTSTAMP:
1563 		BPFD_LOCK(d);
1564 		*(u_int *)addr = d->bd_tstamp;
1565 		BPFD_UNLOCK(d);
1566 		break;
1567 
1568 	/*
1569 	 * Set packet timestamp format and resolution.
1570 	 */
1571 	case BIOCSTSTAMP:
1572 		{
1573 			u_int	func;
1574 
1575 			func = *(u_int *)addr;
1576 			if (BPF_T_VALID(func))
1577 				d->bd_tstamp = func;
1578 			else
1579 				error = EINVAL;
1580 		}
1581 		break;
1582 
1583 	case BIOCFEEDBACK:
1584 		BPFD_LOCK(d);
1585 		d->bd_feedback = *(u_int *)addr;
1586 		BPFD_UNLOCK(d);
1587 		break;
1588 
1589 	case BIOCLOCK:
1590 		BPFD_LOCK(d);
1591 		d->bd_locked = 1;
1592 		BPFD_UNLOCK(d);
1593 		break;
1594 
1595 	case FIONBIO:		/* Non-blocking I/O */
1596 		break;
1597 
1598 	case FIOASYNC:		/* Send signal on receive packets */
1599 		BPFD_LOCK(d);
1600 		d->bd_async = *(int *)addr;
1601 		BPFD_UNLOCK(d);
1602 		break;
1603 
1604 	case FIOSETOWN:
1605 		/*
1606 		 * XXX: Add some sort of locking here?
1607 		 * fsetown() can sleep.
1608 		 */
1609 		error = fsetown(*(int *)addr, &d->bd_sigio);
1610 		break;
1611 
1612 	case FIOGETOWN:
1613 		BPFD_LOCK(d);
1614 		*(int *)addr = fgetown(&d->bd_sigio);
1615 		BPFD_UNLOCK(d);
1616 		break;
1617 
1618 	/* This is deprecated, FIOSETOWN should be used instead. */
1619 	case TIOCSPGRP:
1620 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
1621 		break;
1622 
1623 	/* This is deprecated, FIOGETOWN should be used instead. */
1624 	case TIOCGPGRP:
1625 		*(int *)addr = -fgetown(&d->bd_sigio);
1626 		break;
1627 
1628 	case BIOCSRSIG:		/* Set receive signal */
1629 		{
1630 			u_int sig;
1631 
1632 			sig = *(u_int *)addr;
1633 
1634 			if (sig >= NSIG)
1635 				error = EINVAL;
1636 			else {
1637 				BPFD_LOCK(d);
1638 				d->bd_sig = sig;
1639 				BPFD_UNLOCK(d);
1640 			}
1641 			break;
1642 		}
1643 	case BIOCGRSIG:
1644 		BPFD_LOCK(d);
1645 		*(u_int *)addr = d->bd_sig;
1646 		BPFD_UNLOCK(d);
1647 		break;
1648 
1649 	case BIOCGETBUFMODE:
1650 		BPFD_LOCK(d);
1651 		*(u_int *)addr = d->bd_bufmode;
1652 		BPFD_UNLOCK(d);
1653 		break;
1654 
1655 	case BIOCSETBUFMODE:
1656 		/*
1657 		 * Allow the buffering mode to be changed as long as we
1658 		 * haven't yet committed to a particular mode.  Our
1659 		 * definition of commitment, for now, is whether or not a
1660 		 * buffer has been allocated or an interface attached, since
1661 		 * that's the point where things get tricky.
1662 		 */
1663 		switch (*(u_int *)addr) {
1664 		case BPF_BUFMODE_BUFFER:
1665 			break;
1666 
1667 		case BPF_BUFMODE_ZBUF:
1668 			if (bpf_zerocopy_enable)
1669 				break;
1670 			/* FALLSTHROUGH */
1671 
1672 		default:
1673 			CURVNET_RESTORE();
1674 			return (EINVAL);
1675 		}
1676 
1677 		BPFD_LOCK(d);
1678 		if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1679 		    d->bd_fbuf != NULL || d->bd_bif != NULL) {
1680 			BPFD_UNLOCK(d);
1681 			CURVNET_RESTORE();
1682 			return (EBUSY);
1683 		}
1684 		d->bd_bufmode = *(u_int *)addr;
1685 		BPFD_UNLOCK(d);
1686 		break;
1687 
1688 	case BIOCGETZMAX:
1689 		error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1690 		break;
1691 
1692 	case BIOCSETZBUF:
1693 		error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1694 		break;
1695 
1696 	case BIOCROTZBUF:
1697 		error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1698 		break;
1699 	}
1700 	CURVNET_RESTORE();
1701 	return (error);
1702 }
1703 
1704 /*
1705  * Set d's packet filter program to fp.  If this file already has a filter,
1706  * free it and replace it.  Returns EINVAL for bogus requests.
1707  *
1708  * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1709  * since reading d->bd_bif can't be protected by d or interface lock due to
1710  * lock order.
1711  *
1712  * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1713  * interface read lock to read all filers.
1714  *
1715  */
1716 static int
1717 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1718 {
1719 #ifdef COMPAT_FREEBSD32
1720 	struct bpf_program fp_swab;
1721 	struct bpf_program32 *fp32;
1722 #endif
1723 	struct bpf_insn *fcode, *old;
1724 #ifdef BPF_JITTER
1725 	bpf_jit_filter *jfunc, *ofunc;
1726 #endif
1727 	size_t size;
1728 	u_int flen;
1729 	int need_upgrade;
1730 
1731 #ifdef COMPAT_FREEBSD32
1732 	switch (cmd) {
1733 	case BIOCSETF32:
1734 	case BIOCSETWF32:
1735 	case BIOCSETFNR32:
1736 		fp32 = (struct bpf_program32 *)fp;
1737 		fp_swab.bf_len = fp32->bf_len;
1738 		fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1739 		fp = &fp_swab;
1740 		switch (cmd) {
1741 		case BIOCSETF32:
1742 			cmd = BIOCSETF;
1743 			break;
1744 		case BIOCSETWF32:
1745 			cmd = BIOCSETWF;
1746 			break;
1747 		}
1748 		break;
1749 	}
1750 #endif
1751 
1752 	fcode = NULL;
1753 #ifdef BPF_JITTER
1754 	jfunc = ofunc = NULL;
1755 #endif
1756 	need_upgrade = 0;
1757 
1758 	/*
1759 	 * Check new filter validness before acquiring any locks.
1760 	 * Allocate memory for new filter, if needed.
1761 	 */
1762 	flen = fp->bf_len;
1763 	if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1764 		return (EINVAL);
1765 	size = flen * sizeof(*fp->bf_insns);
1766 	if (size > 0) {
1767 		/* We're setting up new filter.  Copy and check actual data. */
1768 		fcode = malloc(size, M_BPF, M_WAITOK);
1769 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1770 		    !bpf_validate(fcode, flen)) {
1771 			free(fcode, M_BPF);
1772 			return (EINVAL);
1773 		}
1774 #ifdef BPF_JITTER
1775 		/* Filter is copied inside fcode and is perfectly valid. */
1776 		jfunc = bpf_jitter(fcode, flen);
1777 #endif
1778 	}
1779 
1780 	BPF_LOCK();
1781 
1782 	/*
1783 	 * Set up new filter.
1784 	 * Protect filter change by interface lock.
1785 	 * Additionally, we are protected by global lock here.
1786 	 */
1787 	if (d->bd_bif != NULL)
1788 		BPFIF_WLOCK(d->bd_bif);
1789 	BPFD_LOCK(d);
1790 	if (cmd == BIOCSETWF) {
1791 		old = d->bd_wfilter;
1792 		d->bd_wfilter = fcode;
1793 	} else {
1794 		old = d->bd_rfilter;
1795 		d->bd_rfilter = fcode;
1796 #ifdef BPF_JITTER
1797 		ofunc = d->bd_bfilter;
1798 		d->bd_bfilter = jfunc;
1799 #endif
1800 		if (cmd == BIOCSETF)
1801 			reset_d(d);
1802 
1803 		if (fcode != NULL) {
1804 			/*
1805 			 * Do not require upgrade by first BIOCSETF
1806 			 * (used to set snaplen) by pcap_open_live().
1807 			 */
1808 			if (d->bd_writer != 0 && --d->bd_writer == 0)
1809 				need_upgrade = 1;
1810 			CTR4(KTR_NET, "%s: filter function set by pid %d, "
1811 			    "bd_writer counter %d, need_upgrade %d",
1812 			    __func__, d->bd_pid, d->bd_writer, need_upgrade);
1813 		}
1814 	}
1815 	BPFD_UNLOCK(d);
1816 	if (d->bd_bif != NULL)
1817 		BPFIF_WUNLOCK(d->bd_bif);
1818 	if (old != NULL)
1819 		free(old, M_BPF);
1820 #ifdef BPF_JITTER
1821 	if (ofunc != NULL)
1822 		bpf_destroy_jit_filter(ofunc);
1823 #endif
1824 
1825 	/* Move d to active readers list. */
1826 	if (need_upgrade)
1827 		bpf_upgraded(d);
1828 
1829 	BPF_UNLOCK();
1830 	return (0);
1831 }
1832 
1833 /*
1834  * Detach a file from its current interface (if attached at all) and attach
1835  * to the interface indicated by the name stored in ifr.
1836  * Return an errno or 0.
1837  */
1838 static int
1839 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1840 {
1841 	struct bpf_if *bp;
1842 	struct ifnet *theywant;
1843 
1844 	BPF_LOCK_ASSERT();
1845 
1846 	theywant = ifunit(ifr->ifr_name);
1847 	if (theywant == NULL || theywant->if_bpf == NULL)
1848 		return (ENXIO);
1849 
1850 	bp = theywant->if_bpf;
1851 
1852 	/* Check if interface is not being detached from BPF */
1853 	BPFIF_RLOCK(bp);
1854 	if (bp->flags & BPFIF_FLAG_DYING) {
1855 		BPFIF_RUNLOCK(bp);
1856 		return (ENXIO);
1857 	}
1858 	BPFIF_RUNLOCK(bp);
1859 
1860 	/*
1861 	 * Behavior here depends on the buffering model.  If we're using
1862 	 * kernel memory buffers, then we can allocate them here.  If we're
1863 	 * using zero-copy, then the user process must have registered
1864 	 * buffers by the time we get here.  If not, return an error.
1865 	 */
1866 	switch (d->bd_bufmode) {
1867 	case BPF_BUFMODE_BUFFER:
1868 	case BPF_BUFMODE_ZBUF:
1869 		if (d->bd_sbuf == NULL)
1870 			return (EINVAL);
1871 		break;
1872 
1873 	default:
1874 		panic("bpf_setif: bufmode %d", d->bd_bufmode);
1875 	}
1876 	if (bp != d->bd_bif)
1877 		bpf_attachd(d, bp);
1878 	BPFD_LOCK(d);
1879 	reset_d(d);
1880 	BPFD_UNLOCK(d);
1881 	return (0);
1882 }
1883 
1884 /*
1885  * Support for select() and poll() system calls
1886  *
1887  * Return true iff the specific operation will not block indefinitely.
1888  * Otherwise, return false but make a note that a selwakeup() must be done.
1889  */
1890 static int
1891 bpfpoll(struct cdev *dev, int events, struct thread *td)
1892 {
1893 	struct bpf_d *d;
1894 	int revents;
1895 
1896 	if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1897 		return (events &
1898 		    (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1899 
1900 	/*
1901 	 * Refresh PID associated with this descriptor.
1902 	 */
1903 	revents = events & (POLLOUT | POLLWRNORM);
1904 	BPFD_LOCK(d);
1905 	BPF_PID_REFRESH(d, td);
1906 	if (events & (POLLIN | POLLRDNORM)) {
1907 		if (bpf_ready(d))
1908 			revents |= events & (POLLIN | POLLRDNORM);
1909 		else {
1910 			selrecord(td, &d->bd_sel);
1911 			/* Start the read timeout if necessary. */
1912 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1913 				callout_reset(&d->bd_callout, d->bd_rtout,
1914 				    bpf_timed_out, d);
1915 				d->bd_state = BPF_WAITING;
1916 			}
1917 		}
1918 	}
1919 	BPFD_UNLOCK(d);
1920 	return (revents);
1921 }
1922 
1923 /*
1924  * Support for kevent() system call.  Register EVFILT_READ filters and
1925  * reject all others.
1926  */
1927 int
1928 bpfkqfilter(struct cdev *dev, struct knote *kn)
1929 {
1930 	struct bpf_d *d;
1931 
1932 	if (devfs_get_cdevpriv((void **)&d) != 0 ||
1933 	    kn->kn_filter != EVFILT_READ)
1934 		return (1);
1935 
1936 	/*
1937 	 * Refresh PID associated with this descriptor.
1938 	 */
1939 	BPFD_LOCK(d);
1940 	BPF_PID_REFRESH_CUR(d);
1941 	kn->kn_fop = &bpfread_filtops;
1942 	kn->kn_hook = d;
1943 	knlist_add(&d->bd_sel.si_note, kn, 1);
1944 	BPFD_UNLOCK(d);
1945 
1946 	return (0);
1947 }
1948 
1949 static void
1950 filt_bpfdetach(struct knote *kn)
1951 {
1952 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1953 
1954 	knlist_remove(&d->bd_sel.si_note, kn, 0);
1955 }
1956 
1957 static int
1958 filt_bpfread(struct knote *kn, long hint)
1959 {
1960 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1961 	int ready;
1962 
1963 	BPFD_LOCK_ASSERT(d);
1964 	ready = bpf_ready(d);
1965 	if (ready) {
1966 		kn->kn_data = d->bd_slen;
1967 		if (d->bd_hbuf)
1968 			kn->kn_data += d->bd_hlen;
1969 	} else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1970 		callout_reset(&d->bd_callout, d->bd_rtout,
1971 		    bpf_timed_out, d);
1972 		d->bd_state = BPF_WAITING;
1973 	}
1974 
1975 	return (ready);
1976 }
1977 
1978 #define	BPF_TSTAMP_NONE		0
1979 #define	BPF_TSTAMP_FAST		1
1980 #define	BPF_TSTAMP_NORMAL	2
1981 #define	BPF_TSTAMP_EXTERN	3
1982 
1983 static int
1984 bpf_ts_quality(int tstype)
1985 {
1986 
1987 	if (tstype == BPF_T_NONE)
1988 		return (BPF_TSTAMP_NONE);
1989 	if ((tstype & BPF_T_FAST) != 0)
1990 		return (BPF_TSTAMP_FAST);
1991 
1992 	return (BPF_TSTAMP_NORMAL);
1993 }
1994 
1995 static int
1996 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
1997 {
1998 	struct m_tag *tag;
1999 	int quality;
2000 
2001 	quality = bpf_ts_quality(tstype);
2002 	if (quality == BPF_TSTAMP_NONE)
2003 		return (quality);
2004 
2005 	if (m != NULL) {
2006 		tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2007 		if (tag != NULL) {
2008 			*bt = *(struct bintime *)(tag + 1);
2009 			return (BPF_TSTAMP_EXTERN);
2010 		}
2011 	}
2012 	if (quality == BPF_TSTAMP_NORMAL)
2013 		binuptime(bt);
2014 	else
2015 		getbinuptime(bt);
2016 
2017 	return (quality);
2018 }
2019 
2020 /*
2021  * Incoming linkage from device drivers.  Process the packet pkt, of length
2022  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
2023  * by each process' filter, and if accepted, stashed into the corresponding
2024  * buffer.
2025  */
2026 void
2027 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2028 {
2029 	struct bintime bt;
2030 	struct bpf_d *d;
2031 #ifdef BPF_JITTER
2032 	bpf_jit_filter *bf;
2033 #endif
2034 	u_int slen;
2035 	int gottime;
2036 
2037 	gottime = BPF_TSTAMP_NONE;
2038 
2039 	BPFIF_RLOCK(bp);
2040 
2041 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2042 		/*
2043 		 * We are not using any locks for d here because:
2044 		 * 1) any filter change is protected by interface
2045 		 * write lock
2046 		 * 2) destroying/detaching d is protected by interface
2047 		 * write lock, too
2048 		 */
2049 
2050 		/* XXX: Do not protect counter for the sake of performance. */
2051 		++d->bd_rcount;
2052 		/*
2053 		 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2054 		 * way for the caller to indiciate to us whether this packet
2055 		 * is inbound or outbound.  In the bpf_mtap() routines, we use
2056 		 * the interface pointers on the mbuf to figure it out.
2057 		 */
2058 #ifdef BPF_JITTER
2059 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2060 		if (bf != NULL)
2061 			slen = (*(bf->func))(pkt, pktlen, pktlen);
2062 		else
2063 #endif
2064 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2065 		if (slen != 0) {
2066 			/*
2067 			 * Filter matches. Let's to acquire write lock.
2068 			 */
2069 			BPFD_LOCK(d);
2070 
2071 			d->bd_fcount++;
2072 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2073 				gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2074 #ifdef MAC
2075 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2076 #endif
2077 				catchpacket(d, pkt, pktlen, slen,
2078 				    bpf_append_bytes, &bt);
2079 			BPFD_UNLOCK(d);
2080 		}
2081 	}
2082 	BPFIF_RUNLOCK(bp);
2083 }
2084 
2085 #define	BPF_CHECK_DIRECTION(d, r, i)				\
2086 	    (((d)->bd_direction == BPF_D_IN && (r) != (i)) ||	\
2087 	    ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2088 
2089 /*
2090  * Incoming linkage from device drivers, when packet is in an mbuf chain.
2091  * Locking model is explained in bpf_tap().
2092  */
2093 void
2094 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2095 {
2096 	struct bintime bt;
2097 	struct bpf_d *d;
2098 #ifdef BPF_JITTER
2099 	bpf_jit_filter *bf;
2100 #endif
2101 	u_int pktlen, slen;
2102 	int gottime;
2103 
2104 	/* Skip outgoing duplicate packets. */
2105 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2106 		m->m_flags &= ~M_PROMISC;
2107 		return;
2108 	}
2109 
2110 	pktlen = m_length(m, NULL);
2111 	gottime = BPF_TSTAMP_NONE;
2112 
2113 	BPFIF_RLOCK(bp);
2114 
2115 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2116 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2117 			continue;
2118 		++d->bd_rcount;
2119 #ifdef BPF_JITTER
2120 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2121 		/* XXX We cannot handle multiple mbufs. */
2122 		if (bf != NULL && m->m_next == NULL)
2123 			slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2124 		else
2125 #endif
2126 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2127 		if (slen != 0) {
2128 			BPFD_LOCK(d);
2129 
2130 			d->bd_fcount++;
2131 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2132 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2133 #ifdef MAC
2134 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2135 #endif
2136 				catchpacket(d, (u_char *)m, pktlen, slen,
2137 				    bpf_append_mbuf, &bt);
2138 			BPFD_UNLOCK(d);
2139 		}
2140 	}
2141 	BPFIF_RUNLOCK(bp);
2142 }
2143 
2144 /*
2145  * Incoming linkage from device drivers, when packet is in
2146  * an mbuf chain and to be prepended by a contiguous header.
2147  */
2148 void
2149 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2150 {
2151 	struct bintime bt;
2152 	struct mbuf mb;
2153 	struct bpf_d *d;
2154 	u_int pktlen, slen;
2155 	int gottime;
2156 
2157 	/* Skip outgoing duplicate packets. */
2158 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2159 		m->m_flags &= ~M_PROMISC;
2160 		return;
2161 	}
2162 
2163 	pktlen = m_length(m, NULL);
2164 	/*
2165 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
2166 	 * Note that we cut corners here; we only setup what's
2167 	 * absolutely needed--this mbuf should never go anywhere else.
2168 	 */
2169 	mb.m_next = m;
2170 	mb.m_data = data;
2171 	mb.m_len = dlen;
2172 	pktlen += dlen;
2173 
2174 	gottime = BPF_TSTAMP_NONE;
2175 
2176 	BPFIF_RLOCK(bp);
2177 
2178 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2179 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2180 			continue;
2181 		++d->bd_rcount;
2182 		slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2183 		if (slen != 0) {
2184 			BPFD_LOCK(d);
2185 
2186 			d->bd_fcount++;
2187 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2188 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2189 #ifdef MAC
2190 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2191 #endif
2192 				catchpacket(d, (u_char *)&mb, pktlen, slen,
2193 				    bpf_append_mbuf, &bt);
2194 			BPFD_UNLOCK(d);
2195 		}
2196 	}
2197 	BPFIF_RUNLOCK(bp);
2198 }
2199 
2200 #undef	BPF_CHECK_DIRECTION
2201 
2202 #undef	BPF_TSTAMP_NONE
2203 #undef	BPF_TSTAMP_FAST
2204 #undef	BPF_TSTAMP_NORMAL
2205 #undef	BPF_TSTAMP_EXTERN
2206 
2207 static int
2208 bpf_hdrlen(struct bpf_d *d)
2209 {
2210 	int hdrlen;
2211 
2212 	hdrlen = d->bd_bif->bif_hdrlen;
2213 #ifndef BURN_BRIDGES
2214 	if (d->bd_tstamp == BPF_T_NONE ||
2215 	    BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2216 #ifdef COMPAT_FREEBSD32
2217 		if (d->bd_compat32)
2218 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2219 		else
2220 #endif
2221 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2222 	else
2223 #endif
2224 		hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2225 #ifdef COMPAT_FREEBSD32
2226 	if (d->bd_compat32)
2227 		hdrlen = BPF_WORDALIGN32(hdrlen);
2228 	else
2229 #endif
2230 		hdrlen = BPF_WORDALIGN(hdrlen);
2231 
2232 	return (hdrlen - d->bd_bif->bif_hdrlen);
2233 }
2234 
2235 static void
2236 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2237 {
2238 	struct bintime bt2;
2239 	struct timeval tsm;
2240 	struct timespec tsn;
2241 
2242 	if ((tstype & BPF_T_MONOTONIC) == 0) {
2243 		bt2 = *bt;
2244 		bintime_add(&bt2, &boottimebin);
2245 		bt = &bt2;
2246 	}
2247 	switch (BPF_T_FORMAT(tstype)) {
2248 	case BPF_T_MICROTIME:
2249 		bintime2timeval(bt, &tsm);
2250 		ts->bt_sec = tsm.tv_sec;
2251 		ts->bt_frac = tsm.tv_usec;
2252 		break;
2253 	case BPF_T_NANOTIME:
2254 		bintime2timespec(bt, &tsn);
2255 		ts->bt_sec = tsn.tv_sec;
2256 		ts->bt_frac = tsn.tv_nsec;
2257 		break;
2258 	case BPF_T_BINTIME:
2259 		ts->bt_sec = bt->sec;
2260 		ts->bt_frac = bt->frac;
2261 		break;
2262 	}
2263 }
2264 
2265 /*
2266  * Move the packet data from interface memory (pkt) into the
2267  * store buffer.  "cpfn" is the routine called to do the actual data
2268  * transfer.  bcopy is passed in to copy contiguous chunks, while
2269  * bpf_append_mbuf is passed in to copy mbuf chains.  In the latter case,
2270  * pkt is really an mbuf.
2271  */
2272 static void
2273 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2274     void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2275     struct bintime *bt)
2276 {
2277 	struct bpf_xhdr hdr;
2278 #ifndef BURN_BRIDGES
2279 	struct bpf_hdr hdr_old;
2280 #ifdef COMPAT_FREEBSD32
2281 	struct bpf_hdr32 hdr32_old;
2282 #endif
2283 #endif
2284 	int caplen, curlen, hdrlen, totlen;
2285 	int do_wakeup = 0;
2286 	int do_timestamp;
2287 	int tstype;
2288 
2289 	BPFD_LOCK_ASSERT(d);
2290 
2291 	/*
2292 	 * Detect whether user space has released a buffer back to us, and if
2293 	 * so, move it from being a hold buffer to a free buffer.  This may
2294 	 * not be the best place to do it (for example, we might only want to
2295 	 * run this check if we need the space), but for now it's a reliable
2296 	 * spot to do it.
2297 	 */
2298 	if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2299 		d->bd_fbuf = d->bd_hbuf;
2300 		d->bd_hbuf = NULL;
2301 		d->bd_hlen = 0;
2302 		bpf_buf_reclaimed(d);
2303 	}
2304 
2305 	/*
2306 	 * Figure out how many bytes to move.  If the packet is
2307 	 * greater or equal to the snapshot length, transfer that
2308 	 * much.  Otherwise, transfer the whole packet (unless
2309 	 * we hit the buffer size limit).
2310 	 */
2311 	hdrlen = bpf_hdrlen(d);
2312 	totlen = hdrlen + min(snaplen, pktlen);
2313 	if (totlen > d->bd_bufsize)
2314 		totlen = d->bd_bufsize;
2315 
2316 	/*
2317 	 * Round up the end of the previous packet to the next longword.
2318 	 *
2319 	 * Drop the packet if there's no room and no hope of room
2320 	 * If the packet would overflow the storage buffer or the storage
2321 	 * buffer is considered immutable by the buffer model, try to rotate
2322 	 * the buffer and wakeup pending processes.
2323 	 */
2324 #ifdef COMPAT_FREEBSD32
2325 	if (d->bd_compat32)
2326 		curlen = BPF_WORDALIGN32(d->bd_slen);
2327 	else
2328 #endif
2329 		curlen = BPF_WORDALIGN(d->bd_slen);
2330 	if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2331 		if (d->bd_fbuf == NULL) {
2332 			/*
2333 			 * There's no room in the store buffer, and no
2334 			 * prospect of room, so drop the packet.  Notify the
2335 			 * buffer model.
2336 			 */
2337 			bpf_buffull(d);
2338 			++d->bd_dcount;
2339 			return;
2340 		}
2341 		ROTATE_BUFFERS(d);
2342 		do_wakeup = 1;
2343 		curlen = 0;
2344 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2345 		/*
2346 		 * Immediate mode is set, or the read timeout has already
2347 		 * expired during a select call.  A packet arrived, so the
2348 		 * reader should be woken up.
2349 		 */
2350 		do_wakeup = 1;
2351 	caplen = totlen - hdrlen;
2352 	tstype = d->bd_tstamp;
2353 	do_timestamp = tstype != BPF_T_NONE;
2354 #ifndef BURN_BRIDGES
2355 	if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2356 		struct bpf_ts ts;
2357 		if (do_timestamp)
2358 			bpf_bintime2ts(bt, &ts, tstype);
2359 #ifdef COMPAT_FREEBSD32
2360 		if (d->bd_compat32) {
2361 			bzero(&hdr32_old, sizeof(hdr32_old));
2362 			if (do_timestamp) {
2363 				hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2364 				hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2365 			}
2366 			hdr32_old.bh_datalen = pktlen;
2367 			hdr32_old.bh_hdrlen = hdrlen;
2368 			hdr32_old.bh_caplen = caplen;
2369 			bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2370 			    sizeof(hdr32_old));
2371 			goto copy;
2372 		}
2373 #endif
2374 		bzero(&hdr_old, sizeof(hdr_old));
2375 		if (do_timestamp) {
2376 			hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2377 			hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2378 		}
2379 		hdr_old.bh_datalen = pktlen;
2380 		hdr_old.bh_hdrlen = hdrlen;
2381 		hdr_old.bh_caplen = caplen;
2382 		bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2383 		    sizeof(hdr_old));
2384 		goto copy;
2385 	}
2386 #endif
2387 
2388 	/*
2389 	 * Append the bpf header.  Note we append the actual header size, but
2390 	 * move forward the length of the header plus padding.
2391 	 */
2392 	bzero(&hdr, sizeof(hdr));
2393 	if (do_timestamp)
2394 		bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2395 	hdr.bh_datalen = pktlen;
2396 	hdr.bh_hdrlen = hdrlen;
2397 	hdr.bh_caplen = caplen;
2398 	bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2399 
2400 	/*
2401 	 * Copy the packet data into the store buffer and update its length.
2402 	 */
2403 #ifndef BURN_BRIDGES
2404 copy:
2405 #endif
2406 	(*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2407 	d->bd_slen = curlen + totlen;
2408 
2409 	if (do_wakeup)
2410 		bpf_wakeup(d);
2411 }
2412 
2413 /*
2414  * Free buffers currently in use by a descriptor.
2415  * Called on close.
2416  */
2417 static void
2418 bpf_freed(struct bpf_d *d)
2419 {
2420 
2421 	/*
2422 	 * We don't need to lock out interrupts since this descriptor has
2423 	 * been detached from its interface and it yet hasn't been marked
2424 	 * free.
2425 	 */
2426 	bpf_free(d);
2427 	if (d->bd_rfilter != NULL) {
2428 		free((caddr_t)d->bd_rfilter, M_BPF);
2429 #ifdef BPF_JITTER
2430 		if (d->bd_bfilter != NULL)
2431 			bpf_destroy_jit_filter(d->bd_bfilter);
2432 #endif
2433 	}
2434 	if (d->bd_wfilter != NULL)
2435 		free((caddr_t)d->bd_wfilter, M_BPF);
2436 	mtx_destroy(&d->bd_lock);
2437 }
2438 
2439 /*
2440  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
2441  * fixed size of the link header (variable length headers not yet supported).
2442  */
2443 void
2444 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2445 {
2446 
2447 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2448 }
2449 
2450 /*
2451  * Attach an interface to bpf.  ifp is a pointer to the structure
2452  * defining the interface to be attached, dlt is the link layer type,
2453  * and hdrlen is the fixed size of the link header (variable length
2454  * headers are not yet supporrted).
2455  */
2456 void
2457 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2458 {
2459 	struct bpf_if *bp;
2460 
2461 	bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2462 	if (bp == NULL)
2463 		panic("bpfattach");
2464 
2465 	LIST_INIT(&bp->bif_dlist);
2466 	LIST_INIT(&bp->bif_wlist);
2467 	bp->bif_ifp = ifp;
2468 	bp->bif_dlt = dlt;
2469 	rw_init(&bp->bif_lock, "bpf interface lock");
2470 	KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2471 	*driverp = bp;
2472 
2473 	BPF_LOCK();
2474 	LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2475 	BPF_UNLOCK();
2476 
2477 	bp->bif_hdrlen = hdrlen;
2478 
2479 	if (bootverbose)
2480 		if_printf(ifp, "bpf attached\n");
2481 }
2482 
2483 /*
2484  * Detach bpf from an interface. This involves detaching each descriptor
2485  * associated with the interface. Notify each descriptor as it's detached
2486  * so that any sleepers wake up and get ENXIO.
2487  */
2488 void
2489 bpfdetach(struct ifnet *ifp)
2490 {
2491 	struct bpf_if	*bp;
2492 	struct bpf_d	*d;
2493 #ifdef INVARIANTS
2494 	int ndetached;
2495 
2496 	ndetached = 0;
2497 #endif
2498 
2499 	BPF_LOCK();
2500 	/* Find all bpf_if struct's which reference ifp and detach them. */
2501 	do {
2502 		LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2503 			if (ifp == bp->bif_ifp)
2504 				break;
2505 		}
2506 		if (bp != NULL)
2507 			LIST_REMOVE(bp, bif_next);
2508 
2509 		if (bp != NULL) {
2510 #ifdef INVARIANTS
2511 			ndetached++;
2512 #endif
2513 			while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2514 				bpf_detachd_locked(d);
2515 				BPFD_LOCK(d);
2516 				bpf_wakeup(d);
2517 				BPFD_UNLOCK(d);
2518 			}
2519 			/* Free writer-only descriptors */
2520 			while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2521 				bpf_detachd_locked(d);
2522 				BPFD_LOCK(d);
2523 				bpf_wakeup(d);
2524 				BPFD_UNLOCK(d);
2525 			}
2526 
2527 			/*
2528 			 * Delay freing bp till interface is detached
2529 			 * and all routes through this interface are removed.
2530 			 * Mark bp as detached to restrict new consumers.
2531 			 */
2532 			BPFIF_WLOCK(bp);
2533 			bp->flags |= BPFIF_FLAG_DYING;
2534 			BPFIF_WUNLOCK(bp);
2535 		}
2536 	} while (bp != NULL);
2537 	BPF_UNLOCK();
2538 
2539 #ifdef INVARIANTS
2540 	if (ndetached == 0)
2541 		printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2542 #endif
2543 }
2544 
2545 /*
2546  * Interface departure handler.
2547  * Note departure event does not guarantee interface is going down.
2548  */
2549 static void
2550 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2551 {
2552 	struct bpf_if *bp;
2553 
2554 	BPF_LOCK();
2555 	if ((bp = ifp->if_bpf) == NULL) {
2556 		BPF_UNLOCK();
2557 		return;
2558 	}
2559 
2560 	/* Check if bpfdetach() was called previously */
2561 	if ((bp->flags & BPFIF_FLAG_DYING) == 0) {
2562 		BPF_UNLOCK();
2563 		return;
2564 	}
2565 
2566 	CTR3(KTR_NET, "%s: freing BPF instance %p for interface %p",
2567 	    __func__, bp, ifp);
2568 
2569 	ifp->if_bpf = NULL;
2570 	BPF_UNLOCK();
2571 
2572 	rw_destroy(&bp->bif_lock);
2573 	free(bp, M_BPF);
2574 }
2575 
2576 /*
2577  * Get a list of available data link type of the interface.
2578  */
2579 static int
2580 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2581 {
2582 	int n, error;
2583 	struct ifnet *ifp;
2584 	struct bpf_if *bp;
2585 
2586 	BPF_LOCK_ASSERT();
2587 
2588 	ifp = d->bd_bif->bif_ifp;
2589 	n = 0;
2590 	error = 0;
2591 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2592 		if (bp->bif_ifp != ifp)
2593 			continue;
2594 		if (bfl->bfl_list != NULL) {
2595 			if (n >= bfl->bfl_len)
2596 				return (ENOMEM);
2597 			error = copyout(&bp->bif_dlt,
2598 			    bfl->bfl_list + n, sizeof(u_int));
2599 		}
2600 		n++;
2601 	}
2602 	bfl->bfl_len = n;
2603 	return (error);
2604 }
2605 
2606 /*
2607  * Set the data link type of a BPF instance.
2608  */
2609 static int
2610 bpf_setdlt(struct bpf_d *d, u_int dlt)
2611 {
2612 	int error, opromisc;
2613 	struct ifnet *ifp;
2614 	struct bpf_if *bp;
2615 
2616 	BPF_LOCK_ASSERT();
2617 
2618 	if (d->bd_bif->bif_dlt == dlt)
2619 		return (0);
2620 	ifp = d->bd_bif->bif_ifp;
2621 
2622 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2623 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2624 			break;
2625 	}
2626 
2627 	if (bp != NULL) {
2628 		opromisc = d->bd_promisc;
2629 		bpf_attachd(d, bp);
2630 		BPFD_LOCK(d);
2631 		reset_d(d);
2632 		BPFD_UNLOCK(d);
2633 		if (opromisc) {
2634 			error = ifpromisc(bp->bif_ifp, 1);
2635 			if (error)
2636 				if_printf(bp->bif_ifp,
2637 					"bpf_setdlt: ifpromisc failed (%d)\n",
2638 					error);
2639 			else
2640 				d->bd_promisc = 1;
2641 		}
2642 	}
2643 	return (bp == NULL ? EINVAL : 0);
2644 }
2645 
2646 static void
2647 bpf_drvinit(void *unused)
2648 {
2649 	struct cdev *dev;
2650 
2651 	mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2652 	LIST_INIT(&bpf_iflist);
2653 
2654 	dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2655 	/* For compatibility */
2656 	make_dev_alias(dev, "bpf0");
2657 
2658 	/* Register interface departure handler */
2659 	bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2660 		    ifnet_departure_event, bpf_ifdetach, NULL,
2661 		    EVENTHANDLER_PRI_ANY);
2662 }
2663 
2664 /*
2665  * Zero out the various packet counters associated with all of the bpf
2666  * descriptors.  At some point, we will probably want to get a bit more
2667  * granular and allow the user to specify descriptors to be zeroed.
2668  */
2669 static void
2670 bpf_zero_counters(void)
2671 {
2672 	struct bpf_if *bp;
2673 	struct bpf_d *bd;
2674 
2675 	BPF_LOCK();
2676 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2677 		BPFIF_RLOCK(bp);
2678 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2679 			BPFD_LOCK(bd);
2680 			bd->bd_rcount = 0;
2681 			bd->bd_dcount = 0;
2682 			bd->bd_fcount = 0;
2683 			bd->bd_wcount = 0;
2684 			bd->bd_wfcount = 0;
2685 			bd->bd_zcopy = 0;
2686 			BPFD_UNLOCK(bd);
2687 		}
2688 		BPFIF_RUNLOCK(bp);
2689 	}
2690 	BPF_UNLOCK();
2691 }
2692 
2693 /*
2694  * Fill filter statistics
2695  */
2696 static void
2697 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2698 {
2699 
2700 	bzero(d, sizeof(*d));
2701 	BPFD_LOCK_ASSERT(bd);
2702 	d->bd_structsize = sizeof(*d);
2703 	/* XXX: reading should be protected by global lock */
2704 	d->bd_immediate = bd->bd_immediate;
2705 	d->bd_promisc = bd->bd_promisc;
2706 	d->bd_hdrcmplt = bd->bd_hdrcmplt;
2707 	d->bd_direction = bd->bd_direction;
2708 	d->bd_feedback = bd->bd_feedback;
2709 	d->bd_async = bd->bd_async;
2710 	d->bd_rcount = bd->bd_rcount;
2711 	d->bd_dcount = bd->bd_dcount;
2712 	d->bd_fcount = bd->bd_fcount;
2713 	d->bd_sig = bd->bd_sig;
2714 	d->bd_slen = bd->bd_slen;
2715 	d->bd_hlen = bd->bd_hlen;
2716 	d->bd_bufsize = bd->bd_bufsize;
2717 	d->bd_pid = bd->bd_pid;
2718 	strlcpy(d->bd_ifname,
2719 	    bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2720 	d->bd_locked = bd->bd_locked;
2721 	d->bd_wcount = bd->bd_wcount;
2722 	d->bd_wdcount = bd->bd_wdcount;
2723 	d->bd_wfcount = bd->bd_wfcount;
2724 	d->bd_zcopy = bd->bd_zcopy;
2725 	d->bd_bufmode = bd->bd_bufmode;
2726 }
2727 
2728 /*
2729  * Handle `netstat -B' stats request
2730  */
2731 static int
2732 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2733 {
2734 	struct xbpf_d *xbdbuf, *xbd, zerostats;
2735 	int index, error;
2736 	struct bpf_if *bp;
2737 	struct bpf_d *bd;
2738 
2739 	/*
2740 	 * XXX This is not technically correct. It is possible for non
2741 	 * privileged users to open bpf devices. It would make sense
2742 	 * if the users who opened the devices were able to retrieve
2743 	 * the statistics for them, too.
2744 	 */
2745 	error = priv_check(req->td, PRIV_NET_BPF);
2746 	if (error)
2747 		return (error);
2748 	/*
2749 	 * Check to see if the user is requesting that the counters be
2750 	 * zeroed out.  Explicitly check that the supplied data is zeroed,
2751 	 * as we aren't allowing the user to set the counters currently.
2752 	 */
2753 	if (req->newptr != NULL) {
2754 		if (req->newlen != sizeof(zerostats))
2755 			return (EINVAL);
2756 		bzero(&zerostats, sizeof(zerostats));
2757 		xbd = req->newptr;
2758 		if (bcmp(xbd, &zerostats, sizeof(*xbd)) != 0)
2759 			return (EINVAL);
2760 		bpf_zero_counters();
2761 		return (0);
2762 	}
2763 	if (req->oldptr == NULL)
2764 		return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2765 	if (bpf_bpfd_cnt == 0)
2766 		return (SYSCTL_OUT(req, 0, 0));
2767 	xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2768 	BPF_LOCK();
2769 	if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2770 		BPF_UNLOCK();
2771 		free(xbdbuf, M_BPF);
2772 		return (ENOMEM);
2773 	}
2774 	index = 0;
2775 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2776 		BPFIF_RLOCK(bp);
2777 		/* Send writers-only first */
2778 		LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2779 			xbd = &xbdbuf[index++];
2780 			BPFD_LOCK(bd);
2781 			bpfstats_fill_xbpf(xbd, bd);
2782 			BPFD_UNLOCK(bd);
2783 		}
2784 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2785 			xbd = &xbdbuf[index++];
2786 			BPFD_LOCK(bd);
2787 			bpfstats_fill_xbpf(xbd, bd);
2788 			BPFD_UNLOCK(bd);
2789 		}
2790 		BPFIF_RUNLOCK(bp);
2791 	}
2792 	BPF_UNLOCK();
2793 	error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2794 	free(xbdbuf, M_BPF);
2795 	return (error);
2796 }
2797 
2798 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2799 
2800 #else /* !DEV_BPF && !NETGRAPH_BPF */
2801 /*
2802  * NOP stubs to allow bpf-using drivers to load and function.
2803  *
2804  * A 'better' implementation would allow the core bpf functionality
2805  * to be loaded at runtime.
2806  */
2807 static struct bpf_if bp_null;
2808 
2809 void
2810 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2811 {
2812 }
2813 
2814 void
2815 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2816 {
2817 }
2818 
2819 void
2820 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2821 {
2822 }
2823 
2824 void
2825 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2826 {
2827 
2828 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2829 }
2830 
2831 void
2832 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2833 {
2834 
2835 	*driverp = &bp_null;
2836 }
2837 
2838 void
2839 bpfdetach(struct ifnet *ifp)
2840 {
2841 }
2842 
2843 u_int
2844 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2845 {
2846 	return -1;	/* "no filter" behaviour */
2847 }
2848 
2849 int
2850 bpf_validate(const struct bpf_insn *f, int len)
2851 {
2852 	return 0;		/* false */
2853 }
2854 
2855 #endif /* !DEV_BPF && !NETGRAPH_BPF */
2856