xref: /freebsd/sys/net/bpf.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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 #include <net/if_var.h>
72 #define	BPF_INTERNAL
73 #include <net/bpf.h>
74 #include <net/bpf_buffer.h>
75 #ifdef BPF_JITTER
76 #include <net/bpf_jitter.h>
77 #endif
78 #include <net/bpf_zerocopy.h>
79 #include <net/bpfdesc.h>
80 #include <net/vnet.h>
81 
82 #include <netinet/in.h>
83 #include <netinet/if_ether.h>
84 #include <sys/kernel.h>
85 #include <sys/sysctl.h>
86 
87 #include <net80211/ieee80211_freebsd.h>
88 
89 #include <security/mac/mac_framework.h>
90 
91 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
92 
93 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
94 
95 #define PRINET  26			/* interruptible */
96 
97 #define	SIZEOF_BPF_HDR(type)	\
98     (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
99 
100 #ifdef COMPAT_FREEBSD32
101 #include <sys/mount.h>
102 #include <compat/freebsd32/freebsd32.h>
103 #define BPF_ALIGNMENT32 sizeof(int32_t)
104 #define BPF_WORDALIGN32(x) (((x)+(BPF_ALIGNMENT32-1))&~(BPF_ALIGNMENT32-1))
105 
106 #ifndef BURN_BRIDGES
107 /*
108  * 32-bit version of structure prepended to each packet.  We use this header
109  * instead of the standard one for 32-bit streams.  We mark the a stream as
110  * 32-bit the first time we see a 32-bit compat ioctl request.
111  */
112 struct bpf_hdr32 {
113 	struct timeval32 bh_tstamp;	/* time stamp */
114 	uint32_t	bh_caplen;	/* length of captured portion */
115 	uint32_t	bh_datalen;	/* original length of packet */
116 	uint16_t	bh_hdrlen;	/* length of bpf header (this struct
117 					   plus alignment padding) */
118 };
119 #endif
120 
121 struct bpf_program32 {
122 	u_int bf_len;
123 	uint32_t bf_insns;
124 };
125 
126 struct bpf_dltlist32 {
127 	u_int	bfl_len;
128 	u_int	bfl_list;
129 };
130 
131 #define	BIOCSETF32	_IOW('B', 103, struct bpf_program32)
132 #define	BIOCSRTIMEOUT32	_IOW('B', 109, struct timeval32)
133 #define	BIOCGRTIMEOUT32	_IOR('B', 110, struct timeval32)
134 #define	BIOCGDLTLIST32	_IOWR('B', 121, struct bpf_dltlist32)
135 #define	BIOCSETWF32	_IOW('B', 123, struct bpf_program32)
136 #define	BIOCSETFNR32	_IOW('B', 130, struct bpf_program32)
137 #endif
138 
139 /*
140  * bpf_iflist is a list of BPF interface structures, each corresponding to a
141  * specific DLT.  The same network interface might have several BPF interface
142  * structures registered by different layers in the stack (i.e., 802.11
143  * frames, ethernet frames, etc).
144  */
145 static LIST_HEAD(, bpf_if)	bpf_iflist, bpf_freelist;
146 static struct mtx	bpf_mtx;		/* bpf global lock */
147 static int		bpf_bpfd_cnt;
148 
149 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
150 static void	bpf_detachd(struct bpf_d *);
151 static void	bpf_detachd_locked(struct bpf_d *);
152 static void	bpf_freed(struct bpf_d *);
153 static int	bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
154 		    struct sockaddr *, int *, struct bpf_insn *);
155 static int	bpf_setif(struct bpf_d *, struct ifreq *);
156 static void	bpf_timed_out(void *);
157 static __inline void
158 		bpf_wakeup(struct bpf_d *);
159 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
160 		    void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
161 		    struct bintime *);
162 static void	reset_d(struct bpf_d *);
163 static int	bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
164 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
165 static int	bpf_setdlt(struct bpf_d *, u_int);
166 static void	filt_bpfdetach(struct knote *);
167 static int	filt_bpfread(struct knote *, long);
168 static void	bpf_drvinit(void *);
169 static int	bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
170 
171 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
172 int bpf_maxinsns = BPF_MAXINSNS;
173 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
174     &bpf_maxinsns, 0, "Maximum bpf program instructions");
175 static int bpf_zerocopy_enable = 0;
176 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
177     &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
178 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
179     bpf_stats_sysctl, "bpf statistics portal");
180 
181 static VNET_DEFINE(int, bpf_optimize_writers) = 0;
182 #define	V_bpf_optimize_writers VNET(bpf_optimize_writers)
183 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW,
184     &VNET_NAME(bpf_optimize_writers), 0,
185     "Do not send packets until BPF program is set");
186 
187 static	d_open_t	bpfopen;
188 static	d_read_t	bpfread;
189 static	d_write_t	bpfwrite;
190 static	d_ioctl_t	bpfioctl;
191 static	d_poll_t	bpfpoll;
192 static	d_kqfilter_t	bpfkqfilter;
193 
194 static struct cdevsw bpf_cdevsw = {
195 	.d_version =	D_VERSION,
196 	.d_open =	bpfopen,
197 	.d_read =	bpfread,
198 	.d_write =	bpfwrite,
199 	.d_ioctl =	bpfioctl,
200 	.d_poll =	bpfpoll,
201 	.d_name =	"bpf",
202 	.d_kqfilter =	bpfkqfilter,
203 };
204 
205 static struct filterops bpfread_filtops = {
206 	.f_isfd = 1,
207 	.f_detach = filt_bpfdetach,
208 	.f_event = filt_bpfread,
209 };
210 
211 eventhandler_tag	bpf_ifdetach_cookie = NULL;
212 
213 /*
214  * LOCKING MODEL USED BY BPF:
215  * Locks:
216  * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
217  * some global counters and every bpf_if reference.
218  * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
219  * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
220  *   used by bpf_mtap code.
221  *
222  * Lock order:
223  *
224  * Global lock, interface lock, descriptor lock
225  *
226  * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
227  * working model. In many places (like bpf_detachd) we start with BPF descriptor
228  * (and we need to at least rlock it to get reliable interface pointer). This
229  * gives us potential LOR. As a result, we use global lock to protect from bpf_if
230  * change in every such place.
231  *
232  * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
233  * 3) descriptor main wlock.
234  * Reading bd_bif can be protected by any of these locks, typically global lock.
235  *
236  * Changing read/write BPF filter is protected by the same three locks,
237  * the same applies for reading.
238  *
239  * Sleeping in global lock is not allowed due to bpfdetach() using it.
240  */
241 
242 /*
243  * Wrapper functions for various buffering methods.  If the set of buffer
244  * modes expands, we will probably want to introduce a switch data structure
245  * similar to protosw, et.
246  */
247 static void
248 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
249     u_int len)
250 {
251 
252 	BPFD_LOCK_ASSERT(d);
253 
254 	switch (d->bd_bufmode) {
255 	case BPF_BUFMODE_BUFFER:
256 		return (bpf_buffer_append_bytes(d, buf, offset, src, len));
257 
258 	case BPF_BUFMODE_ZBUF:
259 		d->bd_zcopy++;
260 		return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
261 
262 	default:
263 		panic("bpf_buf_append_bytes");
264 	}
265 }
266 
267 static void
268 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
269     u_int len)
270 {
271 
272 	BPFD_LOCK_ASSERT(d);
273 
274 	switch (d->bd_bufmode) {
275 	case BPF_BUFMODE_BUFFER:
276 		return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
277 
278 	case BPF_BUFMODE_ZBUF:
279 		d->bd_zcopy++;
280 		return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
281 
282 	default:
283 		panic("bpf_buf_append_mbuf");
284 	}
285 }
286 
287 /*
288  * This function gets called when the free buffer is re-assigned.
289  */
290 static void
291 bpf_buf_reclaimed(struct bpf_d *d)
292 {
293 
294 	BPFD_LOCK_ASSERT(d);
295 
296 	switch (d->bd_bufmode) {
297 	case BPF_BUFMODE_BUFFER:
298 		return;
299 
300 	case BPF_BUFMODE_ZBUF:
301 		bpf_zerocopy_buf_reclaimed(d);
302 		return;
303 
304 	default:
305 		panic("bpf_buf_reclaimed");
306 	}
307 }
308 
309 /*
310  * If the buffer mechanism has a way to decide that a held buffer can be made
311  * free, then it is exposed via the bpf_canfreebuf() interface.  (1) is
312  * returned if the buffer can be discarded, (0) is returned if it cannot.
313  */
314 static int
315 bpf_canfreebuf(struct bpf_d *d)
316 {
317 
318 	BPFD_LOCK_ASSERT(d);
319 
320 	switch (d->bd_bufmode) {
321 	case BPF_BUFMODE_ZBUF:
322 		return (bpf_zerocopy_canfreebuf(d));
323 	}
324 	return (0);
325 }
326 
327 /*
328  * Allow the buffer model to indicate that the current store buffer is
329  * immutable, regardless of the appearance of space.  Return (1) if the
330  * buffer is writable, and (0) if not.
331  */
332 static int
333 bpf_canwritebuf(struct bpf_d *d)
334 {
335 	BPFD_LOCK_ASSERT(d);
336 
337 	switch (d->bd_bufmode) {
338 	case BPF_BUFMODE_ZBUF:
339 		return (bpf_zerocopy_canwritebuf(d));
340 	}
341 	return (1);
342 }
343 
344 /*
345  * Notify buffer model that an attempt to write to the store buffer has
346  * resulted in a dropped packet, in which case the buffer may be considered
347  * full.
348  */
349 static void
350 bpf_buffull(struct bpf_d *d)
351 {
352 
353 	BPFD_LOCK_ASSERT(d);
354 
355 	switch (d->bd_bufmode) {
356 	case BPF_BUFMODE_ZBUF:
357 		bpf_zerocopy_buffull(d);
358 		break;
359 	}
360 }
361 
362 /*
363  * Notify the buffer model that a buffer has moved into the hold position.
364  */
365 void
366 bpf_bufheld(struct bpf_d *d)
367 {
368 
369 	BPFD_LOCK_ASSERT(d);
370 
371 	switch (d->bd_bufmode) {
372 	case BPF_BUFMODE_ZBUF:
373 		bpf_zerocopy_bufheld(d);
374 		break;
375 	}
376 }
377 
378 static void
379 bpf_free(struct bpf_d *d)
380 {
381 
382 	switch (d->bd_bufmode) {
383 	case BPF_BUFMODE_BUFFER:
384 		return (bpf_buffer_free(d));
385 
386 	case BPF_BUFMODE_ZBUF:
387 		return (bpf_zerocopy_free(d));
388 
389 	default:
390 		panic("bpf_buf_free");
391 	}
392 }
393 
394 static int
395 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
396 {
397 
398 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
399 		return (EOPNOTSUPP);
400 	return (bpf_buffer_uiomove(d, buf, len, uio));
401 }
402 
403 static int
404 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
405 {
406 
407 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
408 		return (EOPNOTSUPP);
409 	return (bpf_buffer_ioctl_sblen(d, i));
410 }
411 
412 static int
413 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
414 {
415 
416 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
417 		return (EOPNOTSUPP);
418 	return (bpf_zerocopy_ioctl_getzmax(td, d, i));
419 }
420 
421 static int
422 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
423 {
424 
425 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
426 		return (EOPNOTSUPP);
427 	return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
428 }
429 
430 static int
431 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
432 {
433 
434 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
435 		return (EOPNOTSUPP);
436 	return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
437 }
438 
439 /*
440  * General BPF functions.
441  */
442 static int
443 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
444     struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter)
445 {
446 	const struct ieee80211_bpf_params *p;
447 	struct ether_header *eh;
448 	struct mbuf *m;
449 	int error;
450 	int len;
451 	int hlen;
452 	int slen;
453 
454 	/*
455 	 * Build a sockaddr based on the data link layer type.
456 	 * We do this at this level because the ethernet header
457 	 * is copied directly into the data field of the sockaddr.
458 	 * In the case of SLIP, there is no header and the packet
459 	 * is forwarded as is.
460 	 * Also, we are careful to leave room at the front of the mbuf
461 	 * for the link level header.
462 	 */
463 	switch (linktype) {
464 
465 	case DLT_SLIP:
466 		sockp->sa_family = AF_INET;
467 		hlen = 0;
468 		break;
469 
470 	case DLT_EN10MB:
471 		sockp->sa_family = AF_UNSPEC;
472 		/* XXX Would MAXLINKHDR be better? */
473 		hlen = ETHER_HDR_LEN;
474 		break;
475 
476 	case DLT_FDDI:
477 		sockp->sa_family = AF_IMPLINK;
478 		hlen = 0;
479 		break;
480 
481 	case DLT_RAW:
482 		sockp->sa_family = AF_UNSPEC;
483 		hlen = 0;
484 		break;
485 
486 	case DLT_NULL:
487 		/*
488 		 * null interface types require a 4 byte pseudo header which
489 		 * corresponds to the address family of the packet.
490 		 */
491 		sockp->sa_family = AF_UNSPEC;
492 		hlen = 4;
493 		break;
494 
495 	case DLT_ATM_RFC1483:
496 		/*
497 		 * en atm driver requires 4-byte atm pseudo header.
498 		 * though it isn't standard, vpi:vci needs to be
499 		 * specified anyway.
500 		 */
501 		sockp->sa_family = AF_UNSPEC;
502 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
503 		break;
504 
505 	case DLT_PPP:
506 		sockp->sa_family = AF_UNSPEC;
507 		hlen = 4;	/* This should match PPP_HDRLEN */
508 		break;
509 
510 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
511 		sockp->sa_family = AF_IEEE80211;
512 		hlen = 0;
513 		break;
514 
515 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
516 		sockp->sa_family = AF_IEEE80211;
517 		sockp->sa_len = 12;	/* XXX != 0 */
518 		hlen = sizeof(struct ieee80211_bpf_params);
519 		break;
520 
521 	default:
522 		return (EIO);
523 	}
524 
525 	len = uio->uio_resid;
526 	if (len < hlen || len - hlen > ifp->if_mtu)
527 		return (EMSGSIZE);
528 
529 	m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
530 	if (m == NULL)
531 		return (EIO);
532 	m->m_pkthdr.len = m->m_len = len;
533 	*mp = m;
534 
535 	error = uiomove(mtod(m, u_char *), len, uio);
536 	if (error)
537 		goto bad;
538 
539 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
540 	if (slen == 0) {
541 		error = EPERM;
542 		goto bad;
543 	}
544 
545 	/* Check for multicast destination */
546 	switch (linktype) {
547 	case DLT_EN10MB:
548 		eh = mtod(m, struct ether_header *);
549 		if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
550 			if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
551 			    ETHER_ADDR_LEN) == 0)
552 				m->m_flags |= M_BCAST;
553 			else
554 				m->m_flags |= M_MCAST;
555 		}
556 		break;
557 	}
558 
559 	/*
560 	 * Make room for link header, and copy it to sockaddr
561 	 */
562 	if (hlen != 0) {
563 		if (sockp->sa_family == AF_IEEE80211) {
564 			/*
565 			 * Collect true length from the parameter header
566 			 * NB: sockp is known to be zero'd so if we do a
567 			 *     short copy unspecified parameters will be
568 			 *     zero.
569 			 * NB: packet may not be aligned after stripping
570 			 *     bpf params
571 			 * XXX check ibp_vers
572 			 */
573 			p = mtod(m, const struct ieee80211_bpf_params *);
574 			hlen = p->ibp_len;
575 			if (hlen > sizeof(sockp->sa_data)) {
576 				error = EINVAL;
577 				goto bad;
578 			}
579 		}
580 		bcopy(mtod(m, const void *), sockp->sa_data, hlen);
581 	}
582 	*hdrlen = hlen;
583 
584 	return (0);
585 bad:
586 	m_freem(m);
587 	return (error);
588 }
589 
590 /*
591  * Attach file to the bpf interface, i.e. make d listen on bp.
592  */
593 static void
594 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
595 {
596 	int op_w;
597 
598 	BPF_LOCK_ASSERT();
599 
600 	/*
601 	 * Save sysctl value to protect from sysctl change
602 	 * between reads
603 	 */
604 	op_w = V_bpf_optimize_writers;
605 
606 	if (d->bd_bif != NULL)
607 		bpf_detachd_locked(d);
608 	/*
609 	 * Point d at bp, and add d to the interface's list.
610 	 * Since there are many applicaiotns using BPF for
611 	 * sending raw packets only (dhcpd, cdpd are good examples)
612 	 * we can delay adding d to the list of active listeners until
613 	 * some filter is configured.
614 	 */
615 
616 	BPFIF_WLOCK(bp);
617 	BPFD_LOCK(d);
618 
619 	d->bd_bif = bp;
620 
621 	if (op_w != 0) {
622 		/* Add to writers-only list */
623 		LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
624 		/*
625 		 * We decrement bd_writer on every filter set operation.
626 		 * First BIOCSETF is done by pcap_open_live() to set up
627 		 * snap length. After that appliation usually sets its own filter
628 		 */
629 		d->bd_writer = 2;
630 	} else
631 		LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
632 
633 	BPFD_UNLOCK(d);
634 	BPFIF_WUNLOCK(bp);
635 
636 	bpf_bpfd_cnt++;
637 
638 	CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
639 	    __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
640 
641 	if (op_w == 0)
642 		EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
643 }
644 
645 /*
646  * Check if we need to upgrade our descriptor @d from write-only mode.
647  */
648 static int
649 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen)
650 {
651 	int is_snap, need_upgrade;
652 
653 	/*
654 	 * Check if we've already upgraded or new filter is empty.
655 	 */
656 	if (d->bd_writer == 0 || fcode == NULL)
657 		return (0);
658 
659 	need_upgrade = 0;
660 
661 	/*
662 	 * Check if cmd looks like snaplen setting from
663 	 * pcap_bpf.c:pcap_open_live().
664 	 * Note we're not checking .k value here:
665 	 * while pcap_open_live() definitely sets to to non-zero value,
666 	 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
667 	 * do not consider upgrading immediately
668 	 */
669 	if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K))
670 		is_snap = 1;
671 	else
672 		is_snap = 0;
673 
674 	if (is_snap == 0) {
675 		/*
676 		 * We're setting first filter and it doesn't look like
677 		 * setting snaplen.  We're probably using bpf directly.
678 		 * Upgrade immediately.
679 		 */
680 		need_upgrade = 1;
681 	} else {
682 		/*
683 		 * Do not require upgrade by first BIOCSETF
684 		 * (used to set snaplen) by pcap_open_live().
685 		 */
686 
687 		if (--d->bd_writer == 0) {
688 			/*
689 			 * First snaplen filter has already
690 			 * been set. This is probably catch-all
691 			 * filter
692 			 */
693 			need_upgrade = 1;
694 		}
695 	}
696 
697 	CTR5(KTR_NET,
698 	    "%s: filter function set by pid %d, "
699 	    "bd_writer counter %d, snap %d upgrade %d",
700 	    __func__, d->bd_pid, d->bd_writer,
701 	    is_snap, need_upgrade);
702 
703 	return (need_upgrade);
704 }
705 
706 /*
707  * Add d to the list of active bp filters.
708  * Reuqires bpf_attachd() to be called before
709  */
710 static void
711 bpf_upgraded(struct bpf_d *d)
712 {
713 	struct bpf_if *bp;
714 
715 	BPF_LOCK_ASSERT();
716 
717 	bp = d->bd_bif;
718 
719 	/*
720 	 * Filter can be set several times without specifying interface.
721 	 * Mark d as reader and exit.
722 	 */
723 	if (bp == NULL) {
724 		BPFD_LOCK(d);
725 		d->bd_writer = 0;
726 		BPFD_UNLOCK(d);
727 		return;
728 	}
729 
730 	BPFIF_WLOCK(bp);
731 	BPFD_LOCK(d);
732 
733 	/* Remove from writers-only list */
734 	LIST_REMOVE(d, bd_next);
735 	LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
736 	/* Mark d as reader */
737 	d->bd_writer = 0;
738 
739 	BPFD_UNLOCK(d);
740 	BPFIF_WUNLOCK(bp);
741 
742 	CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
743 
744 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
745 }
746 
747 /*
748  * Detach a file from its interface.
749  */
750 static void
751 bpf_detachd(struct bpf_d *d)
752 {
753 	BPF_LOCK();
754 	bpf_detachd_locked(d);
755 	BPF_UNLOCK();
756 }
757 
758 static void
759 bpf_detachd_locked(struct bpf_d *d)
760 {
761 	int error;
762 	struct bpf_if *bp;
763 	struct ifnet *ifp;
764 
765 	CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
766 
767 	BPF_LOCK_ASSERT();
768 
769 	/* Check if descriptor is attached */
770 	if ((bp = d->bd_bif) == NULL)
771 		return;
772 
773 	BPFIF_WLOCK(bp);
774 	BPFD_LOCK(d);
775 
776 	/* Save bd_writer value */
777 	error = d->bd_writer;
778 
779 	/*
780 	 * Remove d from the interface's descriptor list.
781 	 */
782 	LIST_REMOVE(d, bd_next);
783 
784 	ifp = bp->bif_ifp;
785 	d->bd_bif = NULL;
786 	BPFD_UNLOCK(d);
787 	BPFIF_WUNLOCK(bp);
788 
789 	bpf_bpfd_cnt--;
790 
791 	/* Call event handler iff d is attached */
792 	if (error == 0)
793 		EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
794 
795 	/*
796 	 * Check if this descriptor had requested promiscuous mode.
797 	 * If so, turn it off.
798 	 */
799 	if (d->bd_promisc) {
800 		d->bd_promisc = 0;
801 		CURVNET_SET(ifp->if_vnet);
802 		error = ifpromisc(ifp, 0);
803 		CURVNET_RESTORE();
804 		if (error != 0 && error != ENXIO) {
805 			/*
806 			 * ENXIO can happen if a pccard is unplugged
807 			 * Something is really wrong if we were able to put
808 			 * the driver into promiscuous mode, but can't
809 			 * take it out.
810 			 */
811 			if_printf(bp->bif_ifp,
812 				"bpf_detach: ifpromisc failed (%d)\n", error);
813 		}
814 	}
815 }
816 
817 /*
818  * Close the descriptor by detaching it from its interface,
819  * deallocating its buffers, and marking it free.
820  */
821 static void
822 bpf_dtor(void *data)
823 {
824 	struct bpf_d *d = data;
825 
826 	BPFD_LOCK(d);
827 	if (d->bd_state == BPF_WAITING)
828 		callout_stop(&d->bd_callout);
829 	d->bd_state = BPF_IDLE;
830 	BPFD_UNLOCK(d);
831 	funsetown(&d->bd_sigio);
832 	bpf_detachd(d);
833 #ifdef MAC
834 	mac_bpfdesc_destroy(d);
835 #endif /* MAC */
836 	seldrain(&d->bd_sel);
837 	knlist_destroy(&d->bd_sel.si_note);
838 	callout_drain(&d->bd_callout);
839 	bpf_freed(d);
840 	free(d, M_BPF);
841 }
842 
843 /*
844  * Open ethernet device.  Returns ENXIO for illegal minor device number,
845  * EBUSY if file is open by another process.
846  */
847 /* ARGSUSED */
848 static	int
849 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
850 {
851 	struct bpf_d *d;
852 	int error, size;
853 
854 	d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
855 	error = devfs_set_cdevpriv(d, bpf_dtor);
856 	if (error != 0) {
857 		free(d, M_BPF);
858 		return (error);
859 	}
860 
861 	/*
862 	 * For historical reasons, perform a one-time initialization call to
863 	 * the buffer routines, even though we're not yet committed to a
864 	 * particular buffer method.
865 	 */
866 	bpf_buffer_init(d);
867 	d->bd_hbuf_in_use = 0;
868 	d->bd_bufmode = BPF_BUFMODE_BUFFER;
869 	d->bd_sig = SIGIO;
870 	d->bd_direction = BPF_D_INOUT;
871 	BPF_PID_REFRESH(d, td);
872 #ifdef MAC
873 	mac_bpfdesc_init(d);
874 	mac_bpfdesc_create(td->td_ucred, d);
875 #endif
876 	mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
877 	callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
878 	knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
879 
880 	/* Allocate default buffers */
881 	size = d->bd_bufsize;
882 	bpf_buffer_ioctl_sblen(d, &size);
883 
884 	return (0);
885 }
886 
887 /*
888  *  bpfread - read next chunk of packets from buffers
889  */
890 static	int
891 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
892 {
893 	struct bpf_d *d;
894 	int error;
895 	int non_block;
896 	int timed_out;
897 
898 	error = devfs_get_cdevpriv((void **)&d);
899 	if (error != 0)
900 		return (error);
901 
902 	/*
903 	 * Restrict application to use a buffer the same size as
904 	 * as kernel buffers.
905 	 */
906 	if (uio->uio_resid != d->bd_bufsize)
907 		return (EINVAL);
908 
909 	non_block = ((ioflag & O_NONBLOCK) != 0);
910 
911 	BPFD_LOCK(d);
912 	BPF_PID_REFRESH_CUR(d);
913 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
914 		BPFD_UNLOCK(d);
915 		return (EOPNOTSUPP);
916 	}
917 	if (d->bd_state == BPF_WAITING)
918 		callout_stop(&d->bd_callout);
919 	timed_out = (d->bd_state == BPF_TIMED_OUT);
920 	d->bd_state = BPF_IDLE;
921 	while (d->bd_hbuf_in_use) {
922 		error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
923 		    PRINET|PCATCH, "bd_hbuf", 0);
924 		if (error != 0) {
925 			BPFD_UNLOCK(d);
926 			return (error);
927 		}
928 	}
929 	/*
930 	 * If the hold buffer is empty, then do a timed sleep, which
931 	 * ends when the timeout expires or when enough packets
932 	 * have arrived to fill the store buffer.
933 	 */
934 	while (d->bd_hbuf == NULL) {
935 		if (d->bd_slen != 0) {
936 			/*
937 			 * A packet(s) either arrived since the previous
938 			 * read or arrived while we were asleep.
939 			 */
940 			if (d->bd_immediate || non_block || timed_out) {
941 				/*
942 				 * Rotate the buffers and return what's here
943 				 * if we are in immediate mode, non-blocking
944 				 * flag is set, or this descriptor timed out.
945 				 */
946 				ROTATE_BUFFERS(d);
947 				break;
948 			}
949 		}
950 
951 		/*
952 		 * No data is available, check to see if the bpf device
953 		 * is still pointed at a real interface.  If not, return
954 		 * ENXIO so that the userland process knows to rebind
955 		 * it before using it again.
956 		 */
957 		if (d->bd_bif == NULL) {
958 			BPFD_UNLOCK(d);
959 			return (ENXIO);
960 		}
961 
962 		if (non_block) {
963 			BPFD_UNLOCK(d);
964 			return (EWOULDBLOCK);
965 		}
966 		error = msleep(d, &d->bd_lock, PRINET|PCATCH,
967 		     "bpf", d->bd_rtout);
968 		if (error == EINTR || error == ERESTART) {
969 			BPFD_UNLOCK(d);
970 			return (error);
971 		}
972 		if (error == EWOULDBLOCK) {
973 			/*
974 			 * On a timeout, return what's in the buffer,
975 			 * which may be nothing.  If there is something
976 			 * in the store buffer, we can rotate the buffers.
977 			 */
978 			if (d->bd_hbuf)
979 				/*
980 				 * We filled up the buffer in between
981 				 * getting the timeout and arriving
982 				 * here, so we don't need to rotate.
983 				 */
984 				break;
985 
986 			if (d->bd_slen == 0) {
987 				BPFD_UNLOCK(d);
988 				return (0);
989 			}
990 			ROTATE_BUFFERS(d);
991 			break;
992 		}
993 	}
994 	/*
995 	 * At this point, we know we have something in the hold slot.
996 	 */
997 	d->bd_hbuf_in_use = 1;
998 	BPFD_UNLOCK(d);
999 
1000 	/*
1001 	 * Move data from hold buffer into user space.
1002 	 * We know the entire buffer is transferred since
1003 	 * we checked above that the read buffer is bpf_bufsize bytes.
1004   	 *
1005 	 * We do not have to worry about simultaneous reads because
1006 	 * we waited for sole access to the hold buffer above.
1007 	 */
1008 	error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1009 
1010 	BPFD_LOCK(d);
1011 	KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1012 	d->bd_fbuf = d->bd_hbuf;
1013 	d->bd_hbuf = NULL;
1014 	d->bd_hlen = 0;
1015 	bpf_buf_reclaimed(d);
1016 	d->bd_hbuf_in_use = 0;
1017 	wakeup(&d->bd_hbuf_in_use);
1018 	BPFD_UNLOCK(d);
1019 
1020 	return (error);
1021 }
1022 
1023 /*
1024  * If there are processes sleeping on this descriptor, wake them up.
1025  */
1026 static __inline void
1027 bpf_wakeup(struct bpf_d *d)
1028 {
1029 
1030 	BPFD_LOCK_ASSERT(d);
1031 	if (d->bd_state == BPF_WAITING) {
1032 		callout_stop(&d->bd_callout);
1033 		d->bd_state = BPF_IDLE;
1034 	}
1035 	wakeup(d);
1036 	if (d->bd_async && d->bd_sig && d->bd_sigio)
1037 		pgsigio(&d->bd_sigio, d->bd_sig, 0);
1038 
1039 	selwakeuppri(&d->bd_sel, PRINET);
1040 	KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1041 }
1042 
1043 static void
1044 bpf_timed_out(void *arg)
1045 {
1046 	struct bpf_d *d = (struct bpf_d *)arg;
1047 
1048 	BPFD_LOCK_ASSERT(d);
1049 
1050 	if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
1051 		return;
1052 	if (d->bd_state == BPF_WAITING) {
1053 		d->bd_state = BPF_TIMED_OUT;
1054 		if (d->bd_slen != 0)
1055 			bpf_wakeup(d);
1056 	}
1057 }
1058 
1059 static int
1060 bpf_ready(struct bpf_d *d)
1061 {
1062 
1063 	BPFD_LOCK_ASSERT(d);
1064 
1065 	if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1066 		return (1);
1067 	if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1068 	    d->bd_slen != 0)
1069 		return (1);
1070 	return (0);
1071 }
1072 
1073 static int
1074 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1075 {
1076 	struct bpf_d *d;
1077 	struct ifnet *ifp;
1078 	struct mbuf *m, *mc;
1079 	struct sockaddr dst;
1080 	int error, hlen;
1081 
1082 	error = devfs_get_cdevpriv((void **)&d);
1083 	if (error != 0)
1084 		return (error);
1085 
1086 	BPF_PID_REFRESH_CUR(d);
1087 	d->bd_wcount++;
1088 	/* XXX: locking required */
1089 	if (d->bd_bif == NULL) {
1090 		d->bd_wdcount++;
1091 		return (ENXIO);
1092 	}
1093 
1094 	ifp = d->bd_bif->bif_ifp;
1095 
1096 	if ((ifp->if_flags & IFF_UP) == 0) {
1097 		d->bd_wdcount++;
1098 		return (ENETDOWN);
1099 	}
1100 
1101 	if (uio->uio_resid == 0) {
1102 		d->bd_wdcount++;
1103 		return (0);
1104 	}
1105 
1106 	bzero(&dst, sizeof(dst));
1107 	m = NULL;
1108 	hlen = 0;
1109 	/* XXX: bpf_movein() can sleep */
1110 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1111 	    &m, &dst, &hlen, d->bd_wfilter);
1112 	if (error) {
1113 		d->bd_wdcount++;
1114 		return (error);
1115 	}
1116 	d->bd_wfcount++;
1117 	if (d->bd_hdrcmplt)
1118 		dst.sa_family = pseudo_AF_HDRCMPLT;
1119 
1120 	if (d->bd_feedback) {
1121 		mc = m_dup(m, M_NOWAIT);
1122 		if (mc != NULL)
1123 			mc->m_pkthdr.rcvif = ifp;
1124 		/* Set M_PROMISC for outgoing packets to be discarded. */
1125 		if (d->bd_direction == BPF_D_INOUT)
1126 			m->m_flags |= M_PROMISC;
1127 	} else
1128 		mc = NULL;
1129 
1130 	m->m_pkthdr.len -= hlen;
1131 	m->m_len -= hlen;
1132 	m->m_data += hlen;	/* XXX */
1133 
1134 	CURVNET_SET(ifp->if_vnet);
1135 #ifdef MAC
1136 	BPFD_LOCK(d);
1137 	mac_bpfdesc_create_mbuf(d, m);
1138 	if (mc != NULL)
1139 		mac_bpfdesc_create_mbuf(d, mc);
1140 	BPFD_UNLOCK(d);
1141 #endif
1142 
1143 	error = (*ifp->if_output)(ifp, m, &dst, NULL);
1144 	if (error)
1145 		d->bd_wdcount++;
1146 
1147 	if (mc != NULL) {
1148 		if (error == 0)
1149 			(*ifp->if_input)(ifp, mc);
1150 		else
1151 			m_freem(mc);
1152 	}
1153 	CURVNET_RESTORE();
1154 
1155 	return (error);
1156 }
1157 
1158 /*
1159  * Reset a descriptor by flushing its packet buffer and clearing the receive
1160  * and drop counts.  This is doable for kernel-only buffers, but with
1161  * zero-copy buffers, we can't write to (or rotate) buffers that are
1162  * currently owned by userspace.  It would be nice if we could encapsulate
1163  * this logic in the buffer code rather than here.
1164  */
1165 static void
1166 reset_d(struct bpf_d *d)
1167 {
1168 
1169 	BPFD_LOCK_ASSERT(d);
1170 
1171 	while (d->bd_hbuf_in_use)
1172 		mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1173 		    "bd_hbuf", 0);
1174 	if ((d->bd_hbuf != NULL) &&
1175 	    (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1176 		/* Free the hold buffer. */
1177 		d->bd_fbuf = d->bd_hbuf;
1178 		d->bd_hbuf = NULL;
1179 		d->bd_hlen = 0;
1180 		bpf_buf_reclaimed(d);
1181 	}
1182 	if (bpf_canwritebuf(d))
1183 		d->bd_slen = 0;
1184 	d->bd_rcount = 0;
1185 	d->bd_dcount = 0;
1186 	d->bd_fcount = 0;
1187 	d->bd_wcount = 0;
1188 	d->bd_wfcount = 0;
1189 	d->bd_wdcount = 0;
1190 	d->bd_zcopy = 0;
1191 }
1192 
1193 /*
1194  *  FIONREAD		Check for read packet available.
1195  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
1196  *  BIOCGBLEN		Get buffer len [for read()].
1197  *  BIOCSETF		Set read filter.
1198  *  BIOCSETFNR		Set read filter without resetting descriptor.
1199  *  BIOCSETWF		Set write filter.
1200  *  BIOCFLUSH		Flush read packet buffer.
1201  *  BIOCPROMISC		Put interface into promiscuous mode.
1202  *  BIOCGDLT		Get link layer type.
1203  *  BIOCGETIF		Get interface name.
1204  *  BIOCSETIF		Set interface.
1205  *  BIOCSRTIMEOUT	Set read timeout.
1206  *  BIOCGRTIMEOUT	Get read timeout.
1207  *  BIOCGSTATS		Get packet stats.
1208  *  BIOCIMMEDIATE	Set immediate mode.
1209  *  BIOCVERSION		Get filter language version.
1210  *  BIOCGHDRCMPLT	Get "header already complete" flag
1211  *  BIOCSHDRCMPLT	Set "header already complete" flag
1212  *  BIOCGDIRECTION	Get packet direction flag
1213  *  BIOCSDIRECTION	Set packet direction flag
1214  *  BIOCGTSTAMP		Get time stamp format and resolution.
1215  *  BIOCSTSTAMP		Set time stamp format and resolution.
1216  *  BIOCLOCK		Set "locked" flag
1217  *  BIOCFEEDBACK	Set packet feedback mode.
1218  *  BIOCSETZBUF		Set current zero-copy buffer locations.
1219  *  BIOCGETZMAX		Get maximum zero-copy buffer size.
1220  *  BIOCROTZBUF		Force rotation of zero-copy buffer
1221  *  BIOCSETBUFMODE	Set buffer mode.
1222  *  BIOCGETBUFMODE	Get current buffer mode.
1223  */
1224 /* ARGSUSED */
1225 static	int
1226 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1227     struct thread *td)
1228 {
1229 	struct bpf_d *d;
1230 	int error;
1231 
1232 	error = devfs_get_cdevpriv((void **)&d);
1233 	if (error != 0)
1234 		return (error);
1235 
1236 	/*
1237 	 * Refresh PID associated with this descriptor.
1238 	 */
1239 	BPFD_LOCK(d);
1240 	BPF_PID_REFRESH(d, td);
1241 	if (d->bd_state == BPF_WAITING)
1242 		callout_stop(&d->bd_callout);
1243 	d->bd_state = BPF_IDLE;
1244 	BPFD_UNLOCK(d);
1245 
1246 	if (d->bd_locked == 1) {
1247 		switch (cmd) {
1248 		case BIOCGBLEN:
1249 		case BIOCFLUSH:
1250 		case BIOCGDLT:
1251 		case BIOCGDLTLIST:
1252 #ifdef COMPAT_FREEBSD32
1253 		case BIOCGDLTLIST32:
1254 #endif
1255 		case BIOCGETIF:
1256 		case BIOCGRTIMEOUT:
1257 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1258 		case BIOCGRTIMEOUT32:
1259 #endif
1260 		case BIOCGSTATS:
1261 		case BIOCVERSION:
1262 		case BIOCGRSIG:
1263 		case BIOCGHDRCMPLT:
1264 		case BIOCSTSTAMP:
1265 		case BIOCFEEDBACK:
1266 		case FIONREAD:
1267 		case BIOCLOCK:
1268 		case BIOCSRTIMEOUT:
1269 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1270 		case BIOCSRTIMEOUT32:
1271 #endif
1272 		case BIOCIMMEDIATE:
1273 		case TIOCGPGRP:
1274 		case BIOCROTZBUF:
1275 			break;
1276 		default:
1277 			return (EPERM);
1278 		}
1279 	}
1280 #ifdef COMPAT_FREEBSD32
1281 	/*
1282 	 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1283 	 * that it will get 32-bit packet headers.
1284 	 */
1285 	switch (cmd) {
1286 	case BIOCSETF32:
1287 	case BIOCSETFNR32:
1288 	case BIOCSETWF32:
1289 	case BIOCGDLTLIST32:
1290 	case BIOCGRTIMEOUT32:
1291 	case BIOCSRTIMEOUT32:
1292 		BPFD_LOCK(d);
1293 		d->bd_compat32 = 1;
1294 		BPFD_UNLOCK(d);
1295 	}
1296 #endif
1297 
1298 	CURVNET_SET(TD_TO_VNET(td));
1299 	switch (cmd) {
1300 
1301 	default:
1302 		error = EINVAL;
1303 		break;
1304 
1305 	/*
1306 	 * Check for read packet available.
1307 	 */
1308 	case FIONREAD:
1309 		{
1310 			int n;
1311 
1312 			BPFD_LOCK(d);
1313 			n = d->bd_slen;
1314 			while (d->bd_hbuf_in_use)
1315 				mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1316 				    PRINET, "bd_hbuf", 0);
1317 			if (d->bd_hbuf)
1318 				n += d->bd_hlen;
1319 			BPFD_UNLOCK(d);
1320 
1321 			*(int *)addr = n;
1322 			break;
1323 		}
1324 
1325 	case SIOCGIFADDR:
1326 		{
1327 			struct ifnet *ifp;
1328 
1329 			if (d->bd_bif == NULL)
1330 				error = EINVAL;
1331 			else {
1332 				ifp = d->bd_bif->bif_ifp;
1333 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
1334 			}
1335 			break;
1336 		}
1337 
1338 	/*
1339 	 * Get buffer len [for read()].
1340 	 */
1341 	case BIOCGBLEN:
1342 		BPFD_LOCK(d);
1343 		*(u_int *)addr = d->bd_bufsize;
1344 		BPFD_UNLOCK(d);
1345 		break;
1346 
1347 	/*
1348 	 * Set buffer length.
1349 	 */
1350 	case BIOCSBLEN:
1351 		error = bpf_ioctl_sblen(d, (u_int *)addr);
1352 		break;
1353 
1354 	/*
1355 	 * Set link layer read filter.
1356 	 */
1357 	case BIOCSETF:
1358 	case BIOCSETFNR:
1359 	case BIOCSETWF:
1360 #ifdef COMPAT_FREEBSD32
1361 	case BIOCSETF32:
1362 	case BIOCSETFNR32:
1363 	case BIOCSETWF32:
1364 #endif
1365 		error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1366 		break;
1367 
1368 	/*
1369 	 * Flush read packet buffer.
1370 	 */
1371 	case BIOCFLUSH:
1372 		BPFD_LOCK(d);
1373 		reset_d(d);
1374 		BPFD_UNLOCK(d);
1375 		break;
1376 
1377 	/*
1378 	 * Put interface into promiscuous mode.
1379 	 */
1380 	case BIOCPROMISC:
1381 		if (d->bd_bif == NULL) {
1382 			/*
1383 			 * No interface attached yet.
1384 			 */
1385 			error = EINVAL;
1386 			break;
1387 		}
1388 		if (d->bd_promisc == 0) {
1389 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1390 			if (error == 0)
1391 				d->bd_promisc = 1;
1392 		}
1393 		break;
1394 
1395 	/*
1396 	 * Get current data link type.
1397 	 */
1398 	case BIOCGDLT:
1399 		BPF_LOCK();
1400 		if (d->bd_bif == NULL)
1401 			error = EINVAL;
1402 		else
1403 			*(u_int *)addr = d->bd_bif->bif_dlt;
1404 		BPF_UNLOCK();
1405 		break;
1406 
1407 	/*
1408 	 * Get a list of supported data link types.
1409 	 */
1410 #ifdef COMPAT_FREEBSD32
1411 	case BIOCGDLTLIST32:
1412 		{
1413 			struct bpf_dltlist32 *list32;
1414 			struct bpf_dltlist dltlist;
1415 
1416 			list32 = (struct bpf_dltlist32 *)addr;
1417 			dltlist.bfl_len = list32->bfl_len;
1418 			dltlist.bfl_list = PTRIN(list32->bfl_list);
1419 			BPF_LOCK();
1420 			if (d->bd_bif == NULL)
1421 				error = EINVAL;
1422 			else {
1423 				error = bpf_getdltlist(d, &dltlist);
1424 				if (error == 0)
1425 					list32->bfl_len = dltlist.bfl_len;
1426 			}
1427 			BPF_UNLOCK();
1428 			break;
1429 		}
1430 #endif
1431 
1432 	case BIOCGDLTLIST:
1433 		BPF_LOCK();
1434 		if (d->bd_bif == NULL)
1435 			error = EINVAL;
1436 		else
1437 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1438 		BPF_UNLOCK();
1439 		break;
1440 
1441 	/*
1442 	 * Set data link type.
1443 	 */
1444 	case BIOCSDLT:
1445 		BPF_LOCK();
1446 		if (d->bd_bif == NULL)
1447 			error = EINVAL;
1448 		else
1449 			error = bpf_setdlt(d, *(u_int *)addr);
1450 		BPF_UNLOCK();
1451 		break;
1452 
1453 	/*
1454 	 * Get interface name.
1455 	 */
1456 	case BIOCGETIF:
1457 		BPF_LOCK();
1458 		if (d->bd_bif == NULL)
1459 			error = EINVAL;
1460 		else {
1461 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
1462 			struct ifreq *const ifr = (struct ifreq *)addr;
1463 
1464 			strlcpy(ifr->ifr_name, ifp->if_xname,
1465 			    sizeof(ifr->ifr_name));
1466 		}
1467 		BPF_UNLOCK();
1468 		break;
1469 
1470 	/*
1471 	 * Set interface.
1472 	 */
1473 	case BIOCSETIF:
1474 		BPF_LOCK();
1475 		error = bpf_setif(d, (struct ifreq *)addr);
1476 		BPF_UNLOCK();
1477 		break;
1478 
1479 	/*
1480 	 * Set read timeout.
1481 	 */
1482 	case BIOCSRTIMEOUT:
1483 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1484 	case BIOCSRTIMEOUT32:
1485 #endif
1486 		{
1487 			struct timeval *tv = (struct timeval *)addr;
1488 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1489 			struct timeval32 *tv32;
1490 			struct timeval tv64;
1491 
1492 			if (cmd == BIOCSRTIMEOUT32) {
1493 				tv32 = (struct timeval32 *)addr;
1494 				tv = &tv64;
1495 				tv->tv_sec = tv32->tv_sec;
1496 				tv->tv_usec = tv32->tv_usec;
1497 			} else
1498 #endif
1499 				tv = (struct timeval *)addr;
1500 
1501 			/*
1502 			 * Subtract 1 tick from tvtohz() since this isn't
1503 			 * a one-shot timer.
1504 			 */
1505 			if ((error = itimerfix(tv)) == 0)
1506 				d->bd_rtout = tvtohz(tv) - 1;
1507 			break;
1508 		}
1509 
1510 	/*
1511 	 * Get read timeout.
1512 	 */
1513 	case BIOCGRTIMEOUT:
1514 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1515 	case BIOCGRTIMEOUT32:
1516 #endif
1517 		{
1518 			struct timeval *tv;
1519 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1520 			struct timeval32 *tv32;
1521 			struct timeval tv64;
1522 
1523 			if (cmd == BIOCGRTIMEOUT32)
1524 				tv = &tv64;
1525 			else
1526 #endif
1527 				tv = (struct timeval *)addr;
1528 
1529 			tv->tv_sec = d->bd_rtout / hz;
1530 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1531 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1532 			if (cmd == BIOCGRTIMEOUT32) {
1533 				tv32 = (struct timeval32 *)addr;
1534 				tv32->tv_sec = tv->tv_sec;
1535 				tv32->tv_usec = tv->tv_usec;
1536 			}
1537 #endif
1538 
1539 			break;
1540 		}
1541 
1542 	/*
1543 	 * Get packet stats.
1544 	 */
1545 	case BIOCGSTATS:
1546 		{
1547 			struct bpf_stat *bs = (struct bpf_stat *)addr;
1548 
1549 			/* XXXCSJP overflow */
1550 			bs->bs_recv = d->bd_rcount;
1551 			bs->bs_drop = d->bd_dcount;
1552 			break;
1553 		}
1554 
1555 	/*
1556 	 * Set immediate mode.
1557 	 */
1558 	case BIOCIMMEDIATE:
1559 		BPFD_LOCK(d);
1560 		d->bd_immediate = *(u_int *)addr;
1561 		BPFD_UNLOCK(d);
1562 		break;
1563 
1564 	case BIOCVERSION:
1565 		{
1566 			struct bpf_version *bv = (struct bpf_version *)addr;
1567 
1568 			bv->bv_major = BPF_MAJOR_VERSION;
1569 			bv->bv_minor = BPF_MINOR_VERSION;
1570 			break;
1571 		}
1572 
1573 	/*
1574 	 * Get "header already complete" flag
1575 	 */
1576 	case BIOCGHDRCMPLT:
1577 		BPFD_LOCK(d);
1578 		*(u_int *)addr = d->bd_hdrcmplt;
1579 		BPFD_UNLOCK(d);
1580 		break;
1581 
1582 	/*
1583 	 * Set "header already complete" flag
1584 	 */
1585 	case BIOCSHDRCMPLT:
1586 		BPFD_LOCK(d);
1587 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1588 		BPFD_UNLOCK(d);
1589 		break;
1590 
1591 	/*
1592 	 * Get packet direction flag
1593 	 */
1594 	case BIOCGDIRECTION:
1595 		BPFD_LOCK(d);
1596 		*(u_int *)addr = d->bd_direction;
1597 		BPFD_UNLOCK(d);
1598 		break;
1599 
1600 	/*
1601 	 * Set packet direction flag
1602 	 */
1603 	case BIOCSDIRECTION:
1604 		{
1605 			u_int	direction;
1606 
1607 			direction = *(u_int *)addr;
1608 			switch (direction) {
1609 			case BPF_D_IN:
1610 			case BPF_D_INOUT:
1611 			case BPF_D_OUT:
1612 				BPFD_LOCK(d);
1613 				d->bd_direction = direction;
1614 				BPFD_UNLOCK(d);
1615 				break;
1616 			default:
1617 				error = EINVAL;
1618 			}
1619 		}
1620 		break;
1621 
1622 	/*
1623 	 * Get packet timestamp format and resolution.
1624 	 */
1625 	case BIOCGTSTAMP:
1626 		BPFD_LOCK(d);
1627 		*(u_int *)addr = d->bd_tstamp;
1628 		BPFD_UNLOCK(d);
1629 		break;
1630 
1631 	/*
1632 	 * Set packet timestamp format and resolution.
1633 	 */
1634 	case BIOCSTSTAMP:
1635 		{
1636 			u_int	func;
1637 
1638 			func = *(u_int *)addr;
1639 			if (BPF_T_VALID(func))
1640 				d->bd_tstamp = func;
1641 			else
1642 				error = EINVAL;
1643 		}
1644 		break;
1645 
1646 	case BIOCFEEDBACK:
1647 		BPFD_LOCK(d);
1648 		d->bd_feedback = *(u_int *)addr;
1649 		BPFD_UNLOCK(d);
1650 		break;
1651 
1652 	case BIOCLOCK:
1653 		BPFD_LOCK(d);
1654 		d->bd_locked = 1;
1655 		BPFD_UNLOCK(d);
1656 		break;
1657 
1658 	case FIONBIO:		/* Non-blocking I/O */
1659 		break;
1660 
1661 	case FIOASYNC:		/* Send signal on receive packets */
1662 		BPFD_LOCK(d);
1663 		d->bd_async = *(int *)addr;
1664 		BPFD_UNLOCK(d);
1665 		break;
1666 
1667 	case FIOSETOWN:
1668 		/*
1669 		 * XXX: Add some sort of locking here?
1670 		 * fsetown() can sleep.
1671 		 */
1672 		error = fsetown(*(int *)addr, &d->bd_sigio);
1673 		break;
1674 
1675 	case FIOGETOWN:
1676 		BPFD_LOCK(d);
1677 		*(int *)addr = fgetown(&d->bd_sigio);
1678 		BPFD_UNLOCK(d);
1679 		break;
1680 
1681 	/* This is deprecated, FIOSETOWN should be used instead. */
1682 	case TIOCSPGRP:
1683 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
1684 		break;
1685 
1686 	/* This is deprecated, FIOGETOWN should be used instead. */
1687 	case TIOCGPGRP:
1688 		*(int *)addr = -fgetown(&d->bd_sigio);
1689 		break;
1690 
1691 	case BIOCSRSIG:		/* Set receive signal */
1692 		{
1693 			u_int sig;
1694 
1695 			sig = *(u_int *)addr;
1696 
1697 			if (sig >= NSIG)
1698 				error = EINVAL;
1699 			else {
1700 				BPFD_LOCK(d);
1701 				d->bd_sig = sig;
1702 				BPFD_UNLOCK(d);
1703 			}
1704 			break;
1705 		}
1706 	case BIOCGRSIG:
1707 		BPFD_LOCK(d);
1708 		*(u_int *)addr = d->bd_sig;
1709 		BPFD_UNLOCK(d);
1710 		break;
1711 
1712 	case BIOCGETBUFMODE:
1713 		BPFD_LOCK(d);
1714 		*(u_int *)addr = d->bd_bufmode;
1715 		BPFD_UNLOCK(d);
1716 		break;
1717 
1718 	case BIOCSETBUFMODE:
1719 		/*
1720 		 * Allow the buffering mode to be changed as long as we
1721 		 * haven't yet committed to a particular mode.  Our
1722 		 * definition of commitment, for now, is whether or not a
1723 		 * buffer has been allocated or an interface attached, since
1724 		 * that's the point where things get tricky.
1725 		 */
1726 		switch (*(u_int *)addr) {
1727 		case BPF_BUFMODE_BUFFER:
1728 			break;
1729 
1730 		case BPF_BUFMODE_ZBUF:
1731 			if (bpf_zerocopy_enable)
1732 				break;
1733 			/* FALLSTHROUGH */
1734 
1735 		default:
1736 			CURVNET_RESTORE();
1737 			return (EINVAL);
1738 		}
1739 
1740 		BPFD_LOCK(d);
1741 		if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1742 		    d->bd_fbuf != NULL || d->bd_bif != NULL) {
1743 			BPFD_UNLOCK(d);
1744 			CURVNET_RESTORE();
1745 			return (EBUSY);
1746 		}
1747 		d->bd_bufmode = *(u_int *)addr;
1748 		BPFD_UNLOCK(d);
1749 		break;
1750 
1751 	case BIOCGETZMAX:
1752 		error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1753 		break;
1754 
1755 	case BIOCSETZBUF:
1756 		error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1757 		break;
1758 
1759 	case BIOCROTZBUF:
1760 		error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1761 		break;
1762 	}
1763 	CURVNET_RESTORE();
1764 	return (error);
1765 }
1766 
1767 /*
1768  * Set d's packet filter program to fp.  If this file already has a filter,
1769  * free it and replace it.  Returns EINVAL for bogus requests.
1770  *
1771  * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1772  * since reading d->bd_bif can't be protected by d or interface lock due to
1773  * lock order.
1774  *
1775  * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1776  * interface read lock to read all filers.
1777  *
1778  */
1779 static int
1780 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1781 {
1782 #ifdef COMPAT_FREEBSD32
1783 	struct bpf_program fp_swab;
1784 	struct bpf_program32 *fp32;
1785 #endif
1786 	struct bpf_insn *fcode, *old;
1787 #ifdef BPF_JITTER
1788 	bpf_jit_filter *jfunc, *ofunc;
1789 #endif
1790 	size_t size;
1791 	u_int flen;
1792 	int need_upgrade;
1793 
1794 #ifdef COMPAT_FREEBSD32
1795 	switch (cmd) {
1796 	case BIOCSETF32:
1797 	case BIOCSETWF32:
1798 	case BIOCSETFNR32:
1799 		fp32 = (struct bpf_program32 *)fp;
1800 		fp_swab.bf_len = fp32->bf_len;
1801 		fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1802 		fp = &fp_swab;
1803 		switch (cmd) {
1804 		case BIOCSETF32:
1805 			cmd = BIOCSETF;
1806 			break;
1807 		case BIOCSETWF32:
1808 			cmd = BIOCSETWF;
1809 			break;
1810 		}
1811 		break;
1812 	}
1813 #endif
1814 
1815 	fcode = NULL;
1816 #ifdef BPF_JITTER
1817 	jfunc = ofunc = NULL;
1818 #endif
1819 	need_upgrade = 0;
1820 
1821 	/*
1822 	 * Check new filter validness before acquiring any locks.
1823 	 * Allocate memory for new filter, if needed.
1824 	 */
1825 	flen = fp->bf_len;
1826 	if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1827 		return (EINVAL);
1828 	size = flen * sizeof(*fp->bf_insns);
1829 	if (size > 0) {
1830 		/* We're setting up new filter.  Copy and check actual data. */
1831 		fcode = malloc(size, M_BPF, M_WAITOK);
1832 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1833 		    !bpf_validate(fcode, flen)) {
1834 			free(fcode, M_BPF);
1835 			return (EINVAL);
1836 		}
1837 #ifdef BPF_JITTER
1838 		/* Filter is copied inside fcode and is perfectly valid. */
1839 		jfunc = bpf_jitter(fcode, flen);
1840 #endif
1841 	}
1842 
1843 	BPF_LOCK();
1844 
1845 	/*
1846 	 * Set up new filter.
1847 	 * Protect filter change by interface lock.
1848 	 * Additionally, we are protected by global lock here.
1849 	 */
1850 	if (d->bd_bif != NULL)
1851 		BPFIF_WLOCK(d->bd_bif);
1852 	BPFD_LOCK(d);
1853 	if (cmd == BIOCSETWF) {
1854 		old = d->bd_wfilter;
1855 		d->bd_wfilter = fcode;
1856 	} else {
1857 		old = d->bd_rfilter;
1858 		d->bd_rfilter = fcode;
1859 #ifdef BPF_JITTER
1860 		ofunc = d->bd_bfilter;
1861 		d->bd_bfilter = jfunc;
1862 #endif
1863 		if (cmd == BIOCSETF)
1864 			reset_d(d);
1865 
1866 		need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen);
1867 	}
1868 	BPFD_UNLOCK(d);
1869 	if (d->bd_bif != NULL)
1870 		BPFIF_WUNLOCK(d->bd_bif);
1871 	if (old != NULL)
1872 		free(old, M_BPF);
1873 #ifdef BPF_JITTER
1874 	if (ofunc != NULL)
1875 		bpf_destroy_jit_filter(ofunc);
1876 #endif
1877 
1878 	/* Move d to active readers list. */
1879 	if (need_upgrade != 0)
1880 		bpf_upgraded(d);
1881 
1882 	BPF_UNLOCK();
1883 	return (0);
1884 }
1885 
1886 /*
1887  * Detach a file from its current interface (if attached at all) and attach
1888  * to the interface indicated by the name stored in ifr.
1889  * Return an errno or 0.
1890  */
1891 static int
1892 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1893 {
1894 	struct bpf_if *bp;
1895 	struct ifnet *theywant;
1896 
1897 	BPF_LOCK_ASSERT();
1898 
1899 	theywant = ifunit(ifr->ifr_name);
1900 	if (theywant == NULL || theywant->if_bpf == NULL)
1901 		return (ENXIO);
1902 
1903 	bp = theywant->if_bpf;
1904 
1905 	/* Check if interface is not being detached from BPF */
1906 	BPFIF_RLOCK(bp);
1907 	if (bp->flags & BPFIF_FLAG_DYING) {
1908 		BPFIF_RUNLOCK(bp);
1909 		return (ENXIO);
1910 	}
1911 	BPFIF_RUNLOCK(bp);
1912 
1913 	/*
1914 	 * Behavior here depends on the buffering model.  If we're using
1915 	 * kernel memory buffers, then we can allocate them here.  If we're
1916 	 * using zero-copy, then the user process must have registered
1917 	 * buffers by the time we get here.  If not, return an error.
1918 	 */
1919 	switch (d->bd_bufmode) {
1920 	case BPF_BUFMODE_BUFFER:
1921 	case BPF_BUFMODE_ZBUF:
1922 		if (d->bd_sbuf == NULL)
1923 			return (EINVAL);
1924 		break;
1925 
1926 	default:
1927 		panic("bpf_setif: bufmode %d", d->bd_bufmode);
1928 	}
1929 	if (bp != d->bd_bif)
1930 		bpf_attachd(d, bp);
1931 	BPFD_LOCK(d);
1932 	reset_d(d);
1933 	BPFD_UNLOCK(d);
1934 	return (0);
1935 }
1936 
1937 /*
1938  * Support for select() and poll() system calls
1939  *
1940  * Return true iff the specific operation will not block indefinitely.
1941  * Otherwise, return false but make a note that a selwakeup() must be done.
1942  */
1943 static int
1944 bpfpoll(struct cdev *dev, int events, struct thread *td)
1945 {
1946 	struct bpf_d *d;
1947 	int revents;
1948 
1949 	if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1950 		return (events &
1951 		    (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1952 
1953 	/*
1954 	 * Refresh PID associated with this descriptor.
1955 	 */
1956 	revents = events & (POLLOUT | POLLWRNORM);
1957 	BPFD_LOCK(d);
1958 	BPF_PID_REFRESH(d, td);
1959 	if (events & (POLLIN | POLLRDNORM)) {
1960 		if (bpf_ready(d))
1961 			revents |= events & (POLLIN | POLLRDNORM);
1962 		else {
1963 			selrecord(td, &d->bd_sel);
1964 			/* Start the read timeout if necessary. */
1965 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1966 				callout_reset(&d->bd_callout, d->bd_rtout,
1967 				    bpf_timed_out, d);
1968 				d->bd_state = BPF_WAITING;
1969 			}
1970 		}
1971 	}
1972 	BPFD_UNLOCK(d);
1973 	return (revents);
1974 }
1975 
1976 /*
1977  * Support for kevent() system call.  Register EVFILT_READ filters and
1978  * reject all others.
1979  */
1980 int
1981 bpfkqfilter(struct cdev *dev, struct knote *kn)
1982 {
1983 	struct bpf_d *d;
1984 
1985 	if (devfs_get_cdevpriv((void **)&d) != 0 ||
1986 	    kn->kn_filter != EVFILT_READ)
1987 		return (1);
1988 
1989 	/*
1990 	 * Refresh PID associated with this descriptor.
1991 	 */
1992 	BPFD_LOCK(d);
1993 	BPF_PID_REFRESH_CUR(d);
1994 	kn->kn_fop = &bpfread_filtops;
1995 	kn->kn_hook = d;
1996 	knlist_add(&d->bd_sel.si_note, kn, 1);
1997 	BPFD_UNLOCK(d);
1998 
1999 	return (0);
2000 }
2001 
2002 static void
2003 filt_bpfdetach(struct knote *kn)
2004 {
2005 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2006 
2007 	knlist_remove(&d->bd_sel.si_note, kn, 0);
2008 }
2009 
2010 static int
2011 filt_bpfread(struct knote *kn, long hint)
2012 {
2013 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2014 	int ready;
2015 
2016 	BPFD_LOCK_ASSERT(d);
2017 	ready = bpf_ready(d);
2018 	if (ready) {
2019 		kn->kn_data = d->bd_slen;
2020 		while (d->bd_hbuf_in_use)
2021 			mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2022 			    PRINET, "bd_hbuf", 0);
2023 		if (d->bd_hbuf)
2024 			kn->kn_data += d->bd_hlen;
2025 	} else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2026 		callout_reset(&d->bd_callout, d->bd_rtout,
2027 		    bpf_timed_out, d);
2028 		d->bd_state = BPF_WAITING;
2029 	}
2030 
2031 	return (ready);
2032 }
2033 
2034 #define	BPF_TSTAMP_NONE		0
2035 #define	BPF_TSTAMP_FAST		1
2036 #define	BPF_TSTAMP_NORMAL	2
2037 #define	BPF_TSTAMP_EXTERN	3
2038 
2039 static int
2040 bpf_ts_quality(int tstype)
2041 {
2042 
2043 	if (tstype == BPF_T_NONE)
2044 		return (BPF_TSTAMP_NONE);
2045 	if ((tstype & BPF_T_FAST) != 0)
2046 		return (BPF_TSTAMP_FAST);
2047 
2048 	return (BPF_TSTAMP_NORMAL);
2049 }
2050 
2051 static int
2052 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2053 {
2054 	struct m_tag *tag;
2055 	int quality;
2056 
2057 	quality = bpf_ts_quality(tstype);
2058 	if (quality == BPF_TSTAMP_NONE)
2059 		return (quality);
2060 
2061 	if (m != NULL) {
2062 		tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2063 		if (tag != NULL) {
2064 			*bt = *(struct bintime *)(tag + 1);
2065 			return (BPF_TSTAMP_EXTERN);
2066 		}
2067 	}
2068 	if (quality == BPF_TSTAMP_NORMAL)
2069 		binuptime(bt);
2070 	else
2071 		getbinuptime(bt);
2072 
2073 	return (quality);
2074 }
2075 
2076 /*
2077  * Incoming linkage from device drivers.  Process the packet pkt, of length
2078  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
2079  * by each process' filter, and if accepted, stashed into the corresponding
2080  * buffer.
2081  */
2082 void
2083 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2084 {
2085 	struct bintime bt;
2086 	struct bpf_d *d;
2087 #ifdef BPF_JITTER
2088 	bpf_jit_filter *bf;
2089 #endif
2090 	u_int slen;
2091 	int gottime;
2092 
2093 	gottime = BPF_TSTAMP_NONE;
2094 
2095 	BPFIF_RLOCK(bp);
2096 
2097 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2098 		/*
2099 		 * We are not using any locks for d here because:
2100 		 * 1) any filter change is protected by interface
2101 		 * write lock
2102 		 * 2) destroying/detaching d is protected by interface
2103 		 * write lock, too
2104 		 */
2105 
2106 		/* XXX: Do not protect counter for the sake of performance. */
2107 		++d->bd_rcount;
2108 		/*
2109 		 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2110 		 * way for the caller to indiciate to us whether this packet
2111 		 * is inbound or outbound.  In the bpf_mtap() routines, we use
2112 		 * the interface pointers on the mbuf to figure it out.
2113 		 */
2114 #ifdef BPF_JITTER
2115 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2116 		if (bf != NULL)
2117 			slen = (*(bf->func))(pkt, pktlen, pktlen);
2118 		else
2119 #endif
2120 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2121 		if (slen != 0) {
2122 			/*
2123 			 * Filter matches. Let's to acquire write lock.
2124 			 */
2125 			BPFD_LOCK(d);
2126 
2127 			d->bd_fcount++;
2128 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2129 				gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2130 #ifdef MAC
2131 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2132 #endif
2133 				catchpacket(d, pkt, pktlen, slen,
2134 				    bpf_append_bytes, &bt);
2135 			BPFD_UNLOCK(d);
2136 		}
2137 	}
2138 	BPFIF_RUNLOCK(bp);
2139 }
2140 
2141 #define	BPF_CHECK_DIRECTION(d, r, i)				\
2142 	    (((d)->bd_direction == BPF_D_IN && (r) != (i)) ||	\
2143 	    ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2144 
2145 /*
2146  * Incoming linkage from device drivers, when packet is in an mbuf chain.
2147  * Locking model is explained in bpf_tap().
2148  */
2149 void
2150 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2151 {
2152 	struct bintime bt;
2153 	struct bpf_d *d;
2154 #ifdef BPF_JITTER
2155 	bpf_jit_filter *bf;
2156 #endif
2157 	u_int pktlen, slen;
2158 	int gottime;
2159 
2160 	/* Skip outgoing duplicate packets. */
2161 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2162 		m->m_flags &= ~M_PROMISC;
2163 		return;
2164 	}
2165 
2166 	pktlen = m_length(m, NULL);
2167 	gottime = BPF_TSTAMP_NONE;
2168 
2169 	BPFIF_RLOCK(bp);
2170 
2171 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2172 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2173 			continue;
2174 		++d->bd_rcount;
2175 #ifdef BPF_JITTER
2176 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2177 		/* XXX We cannot handle multiple mbufs. */
2178 		if (bf != NULL && m->m_next == NULL)
2179 			slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2180 		else
2181 #endif
2182 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, 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 *)m, pktlen, slen,
2193 				    bpf_append_mbuf, &bt);
2194 			BPFD_UNLOCK(d);
2195 		}
2196 	}
2197 	BPFIF_RUNLOCK(bp);
2198 }
2199 
2200 /*
2201  * Incoming linkage from device drivers, when packet is in
2202  * an mbuf chain and to be prepended by a contiguous header.
2203  */
2204 void
2205 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2206 {
2207 	struct bintime bt;
2208 	struct mbuf mb;
2209 	struct bpf_d *d;
2210 	u_int pktlen, slen;
2211 	int gottime;
2212 
2213 	/* Skip outgoing duplicate packets. */
2214 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2215 		m->m_flags &= ~M_PROMISC;
2216 		return;
2217 	}
2218 
2219 	pktlen = m_length(m, NULL);
2220 	/*
2221 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
2222 	 * Note that we cut corners here; we only setup what's
2223 	 * absolutely needed--this mbuf should never go anywhere else.
2224 	 */
2225 	mb.m_next = m;
2226 	mb.m_data = data;
2227 	mb.m_len = dlen;
2228 	pktlen += dlen;
2229 
2230 	gottime = BPF_TSTAMP_NONE;
2231 
2232 	BPFIF_RLOCK(bp);
2233 
2234 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2235 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2236 			continue;
2237 		++d->bd_rcount;
2238 		slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2239 		if (slen != 0) {
2240 			BPFD_LOCK(d);
2241 
2242 			d->bd_fcount++;
2243 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2244 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2245 #ifdef MAC
2246 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2247 #endif
2248 				catchpacket(d, (u_char *)&mb, pktlen, slen,
2249 				    bpf_append_mbuf, &bt);
2250 			BPFD_UNLOCK(d);
2251 		}
2252 	}
2253 	BPFIF_RUNLOCK(bp);
2254 }
2255 
2256 #undef	BPF_CHECK_DIRECTION
2257 
2258 #undef	BPF_TSTAMP_NONE
2259 #undef	BPF_TSTAMP_FAST
2260 #undef	BPF_TSTAMP_NORMAL
2261 #undef	BPF_TSTAMP_EXTERN
2262 
2263 static int
2264 bpf_hdrlen(struct bpf_d *d)
2265 {
2266 	int hdrlen;
2267 
2268 	hdrlen = d->bd_bif->bif_hdrlen;
2269 #ifndef BURN_BRIDGES
2270 	if (d->bd_tstamp == BPF_T_NONE ||
2271 	    BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2272 #ifdef COMPAT_FREEBSD32
2273 		if (d->bd_compat32)
2274 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2275 		else
2276 #endif
2277 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2278 	else
2279 #endif
2280 		hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2281 #ifdef COMPAT_FREEBSD32
2282 	if (d->bd_compat32)
2283 		hdrlen = BPF_WORDALIGN32(hdrlen);
2284 	else
2285 #endif
2286 		hdrlen = BPF_WORDALIGN(hdrlen);
2287 
2288 	return (hdrlen - d->bd_bif->bif_hdrlen);
2289 }
2290 
2291 static void
2292 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2293 {
2294 	struct bintime bt2;
2295 	struct timeval tsm;
2296 	struct timespec tsn;
2297 
2298 	if ((tstype & BPF_T_MONOTONIC) == 0) {
2299 		bt2 = *bt;
2300 		bintime_add(&bt2, &boottimebin);
2301 		bt = &bt2;
2302 	}
2303 	switch (BPF_T_FORMAT(tstype)) {
2304 	case BPF_T_MICROTIME:
2305 		bintime2timeval(bt, &tsm);
2306 		ts->bt_sec = tsm.tv_sec;
2307 		ts->bt_frac = tsm.tv_usec;
2308 		break;
2309 	case BPF_T_NANOTIME:
2310 		bintime2timespec(bt, &tsn);
2311 		ts->bt_sec = tsn.tv_sec;
2312 		ts->bt_frac = tsn.tv_nsec;
2313 		break;
2314 	case BPF_T_BINTIME:
2315 		ts->bt_sec = bt->sec;
2316 		ts->bt_frac = bt->frac;
2317 		break;
2318 	}
2319 }
2320 
2321 /*
2322  * Move the packet data from interface memory (pkt) into the
2323  * store buffer.  "cpfn" is the routine called to do the actual data
2324  * transfer.  bcopy is passed in to copy contiguous chunks, while
2325  * bpf_append_mbuf is passed in to copy mbuf chains.  In the latter case,
2326  * pkt is really an mbuf.
2327  */
2328 static void
2329 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2330     void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2331     struct bintime *bt)
2332 {
2333 	struct bpf_xhdr hdr;
2334 #ifndef BURN_BRIDGES
2335 	struct bpf_hdr hdr_old;
2336 #ifdef COMPAT_FREEBSD32
2337 	struct bpf_hdr32 hdr32_old;
2338 #endif
2339 #endif
2340 	int caplen, curlen, hdrlen, totlen;
2341 	int do_wakeup = 0;
2342 	int do_timestamp;
2343 	int tstype;
2344 
2345 	BPFD_LOCK_ASSERT(d);
2346 
2347 	/*
2348 	 * Detect whether user space has released a buffer back to us, and if
2349 	 * so, move it from being a hold buffer to a free buffer.  This may
2350 	 * not be the best place to do it (for example, we might only want to
2351 	 * run this check if we need the space), but for now it's a reliable
2352 	 * spot to do it.
2353 	 */
2354 	if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2355 		while (d->bd_hbuf_in_use)
2356 			mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2357 			    PRINET, "bd_hbuf", 0);
2358 		d->bd_fbuf = d->bd_hbuf;
2359 		d->bd_hbuf = NULL;
2360 		d->bd_hlen = 0;
2361 		bpf_buf_reclaimed(d);
2362 	}
2363 
2364 	/*
2365 	 * Figure out how many bytes to move.  If the packet is
2366 	 * greater or equal to the snapshot length, transfer that
2367 	 * much.  Otherwise, transfer the whole packet (unless
2368 	 * we hit the buffer size limit).
2369 	 */
2370 	hdrlen = bpf_hdrlen(d);
2371 	totlen = hdrlen + min(snaplen, pktlen);
2372 	if (totlen > d->bd_bufsize)
2373 		totlen = d->bd_bufsize;
2374 
2375 	/*
2376 	 * Round up the end of the previous packet to the next longword.
2377 	 *
2378 	 * Drop the packet if there's no room and no hope of room
2379 	 * If the packet would overflow the storage buffer or the storage
2380 	 * buffer is considered immutable by the buffer model, try to rotate
2381 	 * the buffer and wakeup pending processes.
2382 	 */
2383 #ifdef COMPAT_FREEBSD32
2384 	if (d->bd_compat32)
2385 		curlen = BPF_WORDALIGN32(d->bd_slen);
2386 	else
2387 #endif
2388 		curlen = BPF_WORDALIGN(d->bd_slen);
2389 	if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2390 		if (d->bd_fbuf == NULL) {
2391 			/*
2392 			 * There's no room in the store buffer, and no
2393 			 * prospect of room, so drop the packet.  Notify the
2394 			 * buffer model.
2395 			 */
2396 			bpf_buffull(d);
2397 			++d->bd_dcount;
2398 			return;
2399 		}
2400 		while (d->bd_hbuf_in_use)
2401 			mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2402 			    PRINET, "bd_hbuf", 0);
2403 		ROTATE_BUFFERS(d);
2404 		do_wakeup = 1;
2405 		curlen = 0;
2406 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2407 		/*
2408 		 * Immediate mode is set, or the read timeout has already
2409 		 * expired during a select call.  A packet arrived, so the
2410 		 * reader should be woken up.
2411 		 */
2412 		do_wakeup = 1;
2413 	caplen = totlen - hdrlen;
2414 	tstype = d->bd_tstamp;
2415 	do_timestamp = tstype != BPF_T_NONE;
2416 #ifndef BURN_BRIDGES
2417 	if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2418 		struct bpf_ts ts;
2419 		if (do_timestamp)
2420 			bpf_bintime2ts(bt, &ts, tstype);
2421 #ifdef COMPAT_FREEBSD32
2422 		if (d->bd_compat32) {
2423 			bzero(&hdr32_old, sizeof(hdr32_old));
2424 			if (do_timestamp) {
2425 				hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2426 				hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2427 			}
2428 			hdr32_old.bh_datalen = pktlen;
2429 			hdr32_old.bh_hdrlen = hdrlen;
2430 			hdr32_old.bh_caplen = caplen;
2431 			bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2432 			    sizeof(hdr32_old));
2433 			goto copy;
2434 		}
2435 #endif
2436 		bzero(&hdr_old, sizeof(hdr_old));
2437 		if (do_timestamp) {
2438 			hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2439 			hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2440 		}
2441 		hdr_old.bh_datalen = pktlen;
2442 		hdr_old.bh_hdrlen = hdrlen;
2443 		hdr_old.bh_caplen = caplen;
2444 		bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2445 		    sizeof(hdr_old));
2446 		goto copy;
2447 	}
2448 #endif
2449 
2450 	/*
2451 	 * Append the bpf header.  Note we append the actual header size, but
2452 	 * move forward the length of the header plus padding.
2453 	 */
2454 	bzero(&hdr, sizeof(hdr));
2455 	if (do_timestamp)
2456 		bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2457 	hdr.bh_datalen = pktlen;
2458 	hdr.bh_hdrlen = hdrlen;
2459 	hdr.bh_caplen = caplen;
2460 	bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2461 
2462 	/*
2463 	 * Copy the packet data into the store buffer and update its length.
2464 	 */
2465 #ifndef BURN_BRIDGES
2466 copy:
2467 #endif
2468 	(*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2469 	d->bd_slen = curlen + totlen;
2470 
2471 	if (do_wakeup)
2472 		bpf_wakeup(d);
2473 }
2474 
2475 /*
2476  * Free buffers currently in use by a descriptor.
2477  * Called on close.
2478  */
2479 static void
2480 bpf_freed(struct bpf_d *d)
2481 {
2482 
2483 	/*
2484 	 * We don't need to lock out interrupts since this descriptor has
2485 	 * been detached from its interface and it yet hasn't been marked
2486 	 * free.
2487 	 */
2488 	bpf_free(d);
2489 	if (d->bd_rfilter != NULL) {
2490 		free((caddr_t)d->bd_rfilter, M_BPF);
2491 #ifdef BPF_JITTER
2492 		if (d->bd_bfilter != NULL)
2493 			bpf_destroy_jit_filter(d->bd_bfilter);
2494 #endif
2495 	}
2496 	if (d->bd_wfilter != NULL)
2497 		free((caddr_t)d->bd_wfilter, M_BPF);
2498 	mtx_destroy(&d->bd_lock);
2499 }
2500 
2501 /*
2502  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
2503  * fixed size of the link header (variable length headers not yet supported).
2504  */
2505 void
2506 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2507 {
2508 
2509 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2510 }
2511 
2512 /*
2513  * Attach an interface to bpf.  ifp is a pointer to the structure
2514  * defining the interface to be attached, dlt is the link layer type,
2515  * and hdrlen is the fixed size of the link header (variable length
2516  * headers are not yet supporrted).
2517  */
2518 void
2519 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2520 {
2521 	struct bpf_if *bp;
2522 
2523 	bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2524 	if (bp == NULL)
2525 		panic("bpfattach");
2526 
2527 	LIST_INIT(&bp->bif_dlist);
2528 	LIST_INIT(&bp->bif_wlist);
2529 	bp->bif_ifp = ifp;
2530 	bp->bif_dlt = dlt;
2531 	rw_init(&bp->bif_lock, "bpf interface lock");
2532 	KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2533 	*driverp = bp;
2534 
2535 	BPF_LOCK();
2536 	LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2537 	BPF_UNLOCK();
2538 
2539 	bp->bif_hdrlen = hdrlen;
2540 
2541 	if (bootverbose)
2542 		if_printf(ifp, "bpf attached\n");
2543 }
2544 
2545 /*
2546  * Detach bpf from an interface. This involves detaching each descriptor
2547  * associated with the interface. Notify each descriptor as it's detached
2548  * so that any sleepers wake up and get ENXIO.
2549  */
2550 void
2551 bpfdetach(struct ifnet *ifp)
2552 {
2553 	struct bpf_if	*bp, *bp_temp;
2554 	struct bpf_d	*d;
2555 	int ndetached;
2556 
2557 	ndetached = 0;
2558 
2559 	BPF_LOCK();
2560 	/* Find all bpf_if struct's which reference ifp and detach them. */
2561 	LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2562 		if (ifp != bp->bif_ifp)
2563 			continue;
2564 
2565 		LIST_REMOVE(bp, bif_next);
2566 		/* Add to to-be-freed list */
2567 		LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next);
2568 
2569 		ndetached++;
2570 		/*
2571 		 * Delay freeing bp till interface is detached
2572 		 * and all routes through this interface are removed.
2573 		 * Mark bp as detached to restrict new consumers.
2574 		 */
2575 		BPFIF_WLOCK(bp);
2576 		bp->flags |= BPFIF_FLAG_DYING;
2577 		BPFIF_WUNLOCK(bp);
2578 
2579 		CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p",
2580 		    __func__, bp->bif_dlt, bp, ifp);
2581 
2582 		/* Free common descriptors */
2583 		while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2584 			bpf_detachd_locked(d);
2585 			BPFD_LOCK(d);
2586 			bpf_wakeup(d);
2587 			BPFD_UNLOCK(d);
2588 		}
2589 
2590 		/* Free writer-only descriptors */
2591 		while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2592 			bpf_detachd_locked(d);
2593 			BPFD_LOCK(d);
2594 			bpf_wakeup(d);
2595 			BPFD_UNLOCK(d);
2596 		}
2597 	}
2598 	BPF_UNLOCK();
2599 
2600 #ifdef INVARIANTS
2601 	if (ndetached == 0)
2602 		printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2603 #endif
2604 }
2605 
2606 /*
2607  * Interface departure handler.
2608  * Note departure event does not guarantee interface is going down.
2609  * Interface renaming is currently done via departure/arrival event set.
2610  *
2611  * Departure handled is called after all routes pointing to
2612  * given interface are removed and interface is in down state
2613  * restricting any packets to be sent/received. We assume it is now safe
2614  * to free data allocated by BPF.
2615  */
2616 static void
2617 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2618 {
2619 	struct bpf_if *bp, *bp_temp;
2620 	int nmatched = 0;
2621 
2622 	BPF_LOCK();
2623 	/*
2624 	 * Find matching entries in free list.
2625 	 * Nothing should be found if bpfdetach() was not called.
2626 	 */
2627 	LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) {
2628 		if (ifp != bp->bif_ifp)
2629 			continue;
2630 
2631 		CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p",
2632 		    __func__, bp, ifp);
2633 
2634 		LIST_REMOVE(bp, bif_next);
2635 
2636 		rw_destroy(&bp->bif_lock);
2637 		free(bp, M_BPF);
2638 
2639 		nmatched++;
2640 	}
2641 	BPF_UNLOCK();
2642 
2643 	/*
2644 	 * Note that we cannot zero other pointers to
2645 	 * custom DLTs possibly used by given interface.
2646 	 */
2647 	if (nmatched != 0)
2648 		ifp->if_bpf = NULL;
2649 }
2650 
2651 /*
2652  * Get a list of available data link type of the interface.
2653  */
2654 static int
2655 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2656 {
2657 	int n, error;
2658 	struct ifnet *ifp;
2659 	struct bpf_if *bp;
2660 
2661 	BPF_LOCK_ASSERT();
2662 
2663 	ifp = d->bd_bif->bif_ifp;
2664 	n = 0;
2665 	error = 0;
2666 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2667 		if (bp->bif_ifp != ifp)
2668 			continue;
2669 		if (bfl->bfl_list != NULL) {
2670 			if (n >= bfl->bfl_len)
2671 				return (ENOMEM);
2672 			error = copyout(&bp->bif_dlt,
2673 			    bfl->bfl_list + n, sizeof(u_int));
2674 		}
2675 		n++;
2676 	}
2677 	bfl->bfl_len = n;
2678 	return (error);
2679 }
2680 
2681 /*
2682  * Set the data link type of a BPF instance.
2683  */
2684 static int
2685 bpf_setdlt(struct bpf_d *d, u_int dlt)
2686 {
2687 	int error, opromisc;
2688 	struct ifnet *ifp;
2689 	struct bpf_if *bp;
2690 
2691 	BPF_LOCK_ASSERT();
2692 
2693 	if (d->bd_bif->bif_dlt == dlt)
2694 		return (0);
2695 	ifp = d->bd_bif->bif_ifp;
2696 
2697 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2698 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2699 			break;
2700 	}
2701 
2702 	if (bp != NULL) {
2703 		opromisc = d->bd_promisc;
2704 		bpf_attachd(d, bp);
2705 		BPFD_LOCK(d);
2706 		reset_d(d);
2707 		BPFD_UNLOCK(d);
2708 		if (opromisc) {
2709 			error = ifpromisc(bp->bif_ifp, 1);
2710 			if (error)
2711 				if_printf(bp->bif_ifp,
2712 					"bpf_setdlt: ifpromisc failed (%d)\n",
2713 					error);
2714 			else
2715 				d->bd_promisc = 1;
2716 		}
2717 	}
2718 	return (bp == NULL ? EINVAL : 0);
2719 }
2720 
2721 static void
2722 bpf_drvinit(void *unused)
2723 {
2724 	struct cdev *dev;
2725 
2726 	mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2727 	LIST_INIT(&bpf_iflist);
2728 	LIST_INIT(&bpf_freelist);
2729 
2730 	dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2731 	/* For compatibility */
2732 	make_dev_alias(dev, "bpf0");
2733 
2734 	/* Register interface departure handler */
2735 	bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2736 		    ifnet_departure_event, bpf_ifdetach, NULL,
2737 		    EVENTHANDLER_PRI_ANY);
2738 }
2739 
2740 /*
2741  * Zero out the various packet counters associated with all of the bpf
2742  * descriptors.  At some point, we will probably want to get a bit more
2743  * granular and allow the user to specify descriptors to be zeroed.
2744  */
2745 static void
2746 bpf_zero_counters(void)
2747 {
2748 	struct bpf_if *bp;
2749 	struct bpf_d *bd;
2750 
2751 	BPF_LOCK();
2752 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2753 		BPFIF_RLOCK(bp);
2754 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2755 			BPFD_LOCK(bd);
2756 			bd->bd_rcount = 0;
2757 			bd->bd_dcount = 0;
2758 			bd->bd_fcount = 0;
2759 			bd->bd_wcount = 0;
2760 			bd->bd_wfcount = 0;
2761 			bd->bd_zcopy = 0;
2762 			BPFD_UNLOCK(bd);
2763 		}
2764 		BPFIF_RUNLOCK(bp);
2765 	}
2766 	BPF_UNLOCK();
2767 }
2768 
2769 /*
2770  * Fill filter statistics
2771  */
2772 static void
2773 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2774 {
2775 
2776 	bzero(d, sizeof(*d));
2777 	BPFD_LOCK_ASSERT(bd);
2778 	d->bd_structsize = sizeof(*d);
2779 	/* XXX: reading should be protected by global lock */
2780 	d->bd_immediate = bd->bd_immediate;
2781 	d->bd_promisc = bd->bd_promisc;
2782 	d->bd_hdrcmplt = bd->bd_hdrcmplt;
2783 	d->bd_direction = bd->bd_direction;
2784 	d->bd_feedback = bd->bd_feedback;
2785 	d->bd_async = bd->bd_async;
2786 	d->bd_rcount = bd->bd_rcount;
2787 	d->bd_dcount = bd->bd_dcount;
2788 	d->bd_fcount = bd->bd_fcount;
2789 	d->bd_sig = bd->bd_sig;
2790 	d->bd_slen = bd->bd_slen;
2791 	d->bd_hlen = bd->bd_hlen;
2792 	d->bd_bufsize = bd->bd_bufsize;
2793 	d->bd_pid = bd->bd_pid;
2794 	strlcpy(d->bd_ifname,
2795 	    bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2796 	d->bd_locked = bd->bd_locked;
2797 	d->bd_wcount = bd->bd_wcount;
2798 	d->bd_wdcount = bd->bd_wdcount;
2799 	d->bd_wfcount = bd->bd_wfcount;
2800 	d->bd_zcopy = bd->bd_zcopy;
2801 	d->bd_bufmode = bd->bd_bufmode;
2802 }
2803 
2804 /*
2805  * Handle `netstat -B' stats request
2806  */
2807 static int
2808 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2809 {
2810 	static const struct xbpf_d zerostats;
2811 	struct xbpf_d *xbdbuf, *xbd, tempstats;
2812 	int index, error;
2813 	struct bpf_if *bp;
2814 	struct bpf_d *bd;
2815 
2816 	/*
2817 	 * XXX This is not technically correct. It is possible for non
2818 	 * privileged users to open bpf devices. It would make sense
2819 	 * if the users who opened the devices were able to retrieve
2820 	 * the statistics for them, too.
2821 	 */
2822 	error = priv_check(req->td, PRIV_NET_BPF);
2823 	if (error)
2824 		return (error);
2825 	/*
2826 	 * Check to see if the user is requesting that the counters be
2827 	 * zeroed out.  Explicitly check that the supplied data is zeroed,
2828 	 * as we aren't allowing the user to set the counters currently.
2829 	 */
2830 	if (req->newptr != NULL) {
2831 		if (req->newlen != sizeof(tempstats))
2832 			return (EINVAL);
2833 		memset(&tempstats, 0, sizeof(tempstats));
2834 		error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2835 		if (error)
2836 			return (error);
2837 		if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2838 			return (EINVAL);
2839 		bpf_zero_counters();
2840 		return (0);
2841 	}
2842 	if (req->oldptr == NULL)
2843 		return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2844 	if (bpf_bpfd_cnt == 0)
2845 		return (SYSCTL_OUT(req, 0, 0));
2846 	xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2847 	BPF_LOCK();
2848 	if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2849 		BPF_UNLOCK();
2850 		free(xbdbuf, M_BPF);
2851 		return (ENOMEM);
2852 	}
2853 	index = 0;
2854 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2855 		BPFIF_RLOCK(bp);
2856 		/* Send writers-only first */
2857 		LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2858 			xbd = &xbdbuf[index++];
2859 			BPFD_LOCK(bd);
2860 			bpfstats_fill_xbpf(xbd, bd);
2861 			BPFD_UNLOCK(bd);
2862 		}
2863 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2864 			xbd = &xbdbuf[index++];
2865 			BPFD_LOCK(bd);
2866 			bpfstats_fill_xbpf(xbd, bd);
2867 			BPFD_UNLOCK(bd);
2868 		}
2869 		BPFIF_RUNLOCK(bp);
2870 	}
2871 	BPF_UNLOCK();
2872 	error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2873 	free(xbdbuf, M_BPF);
2874 	return (error);
2875 }
2876 
2877 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2878 
2879 #else /* !DEV_BPF && !NETGRAPH_BPF */
2880 /*
2881  * NOP stubs to allow bpf-using drivers to load and function.
2882  *
2883  * A 'better' implementation would allow the core bpf functionality
2884  * to be loaded at runtime.
2885  */
2886 static struct bpf_if bp_null;
2887 
2888 void
2889 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2890 {
2891 }
2892 
2893 void
2894 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2895 {
2896 }
2897 
2898 void
2899 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2900 {
2901 }
2902 
2903 void
2904 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2905 {
2906 
2907 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2908 }
2909 
2910 void
2911 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2912 {
2913 
2914 	*driverp = &bp_null;
2915 }
2916 
2917 void
2918 bpfdetach(struct ifnet *ifp)
2919 {
2920 }
2921 
2922 u_int
2923 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2924 {
2925 	return -1;	/* "no filter" behaviour */
2926 }
2927 
2928 int
2929 bpf_validate(const struct bpf_insn *f, int len)
2930 {
2931 	return 0;		/* false */
2932 }
2933 
2934 #endif /* !DEV_BPF && !NETGRAPH_BPF */
2935