xref: /freebsd/sys/net/bpf.c (revision ad30f8e79bd1007cc2476e491bd21b4f5e389e0a)
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/systm.h>
47 #include <sys/conf.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/time.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/signalvar.h>
56 #include <sys/filio.h>
57 #include <sys/sockio.h>
58 #include <sys/ttycom.h>
59 #include <sys/uio.h>
60 
61 #include <sys/event.h>
62 #include <sys/file.h>
63 #include <sys/poll.h>
64 #include <sys/proc.h>
65 
66 #include <sys/socket.h>
67 
68 #include <net/if.h>
69 #include <net/bpf.h>
70 #include <net/bpf_buffer.h>
71 #ifdef BPF_JITTER
72 #include <net/bpf_jitter.h>
73 #endif
74 #include <net/bpf_zerocopy.h>
75 #include <net/bpfdesc.h>
76 #include <net/vnet.h>
77 
78 #include <netinet/in.h>
79 #include <netinet/if_ether.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 
83 #include <net80211/ieee80211_freebsd.h>
84 
85 #include <security/mac/mac_framework.h>
86 
87 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
88 
89 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
90 
91 #define PRINET  26			/* interruptible */
92 
93 #define	SIZEOF_BPF_HDR(type)	\
94     (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
95 
96 #ifdef COMPAT_FREEBSD32
97 #include <sys/mount.h>
98 #include <compat/freebsd32/freebsd32.h>
99 #define BPF_ALIGNMENT32 sizeof(int32_t)
100 #define BPF_WORDALIGN32(x) (((x)+(BPF_ALIGNMENT32-1))&~(BPF_ALIGNMENT32-1))
101 
102 #ifndef BURN_BRIDGES
103 /*
104  * 32-bit version of structure prepended to each packet.  We use this header
105  * instead of the standard one for 32-bit streams.  We mark the a stream as
106  * 32-bit the first time we see a 32-bit compat ioctl request.
107  */
108 struct bpf_hdr32 {
109 	struct timeval32 bh_tstamp;	/* time stamp */
110 	uint32_t	bh_caplen;	/* length of captured portion */
111 	uint32_t	bh_datalen;	/* original length of packet */
112 	uint16_t	bh_hdrlen;	/* length of bpf header (this struct
113 					   plus alignment padding) */
114 };
115 #endif
116 
117 struct bpf_program32 {
118 	u_int bf_len;
119 	uint32_t bf_insns;
120 };
121 
122 struct bpf_dltlist32 {
123 	u_int	bfl_len;
124 	u_int	bfl_list;
125 };
126 
127 #define	BIOCSETF32	_IOW('B', 103, struct bpf_program32)
128 #define	BIOCSRTIMEOUT32	_IOW('B', 109, struct timeval32)
129 #define	BIOCGRTIMEOUT32	_IOR('B', 110, struct timeval32)
130 #define	BIOCGDLTLIST32	_IOWR('B', 121, struct bpf_dltlist32)
131 #define	BIOCSETWF32	_IOW('B', 123, struct bpf_program32)
132 #define	BIOCSETFNR32	_IOW('B', 130, struct bpf_program32)
133 #endif
134 
135 /*
136  * bpf_iflist is a list of BPF interface structures, each corresponding to a
137  * specific DLT.  The same network interface might have several BPF interface
138  * structures registered by different layers in the stack (i.e., 802.11
139  * frames, ethernet frames, etc).
140  */
141 static LIST_HEAD(, bpf_if)	bpf_iflist;
142 static struct mtx	bpf_mtx;		/* bpf global lock */
143 static int		bpf_bpfd_cnt;
144 
145 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
146 static void	bpf_detachd(struct bpf_d *);
147 static void	bpf_freed(struct bpf_d *);
148 static int	bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
149 		    struct sockaddr *, int *, struct bpf_insn *);
150 static int	bpf_setif(struct bpf_d *, struct ifreq *);
151 static void	bpf_timed_out(void *);
152 static __inline void
153 		bpf_wakeup(struct bpf_d *);
154 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
155 		    void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
156 		    struct bintime *);
157 static void	reset_d(struct bpf_d *);
158 static int	 bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
159 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
160 static int	bpf_setdlt(struct bpf_d *, u_int);
161 static void	filt_bpfdetach(struct knote *);
162 static int	filt_bpfread(struct knote *, long);
163 static void	bpf_drvinit(void *);
164 static int	bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
165 
166 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
167 int bpf_maxinsns = BPF_MAXINSNS;
168 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
169     &bpf_maxinsns, 0, "Maximum bpf program instructions");
170 static int bpf_zerocopy_enable = 0;
171 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
172     &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
173 SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
174     bpf_stats_sysctl, "bpf statistics portal");
175 
176 static	d_open_t	bpfopen;
177 static	d_read_t	bpfread;
178 static	d_write_t	bpfwrite;
179 static	d_ioctl_t	bpfioctl;
180 static	d_poll_t	bpfpoll;
181 static	d_kqfilter_t	bpfkqfilter;
182 
183 static struct cdevsw bpf_cdevsw = {
184 	.d_version =	D_VERSION,
185 	.d_open =	bpfopen,
186 	.d_read =	bpfread,
187 	.d_write =	bpfwrite,
188 	.d_ioctl =	bpfioctl,
189 	.d_poll =	bpfpoll,
190 	.d_name =	"bpf",
191 	.d_kqfilter =	bpfkqfilter,
192 };
193 
194 static struct filterops bpfread_filtops = {
195 	.f_isfd = 1,
196 	.f_detach = filt_bpfdetach,
197 	.f_event = filt_bpfread,
198 };
199 
200 /*
201  * Wrapper functions for various buffering methods.  If the set of buffer
202  * modes expands, we will probably want to introduce a switch data structure
203  * similar to protosw, et.
204  */
205 static void
206 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
207     u_int len)
208 {
209 
210 	BPFD_LOCK_ASSERT(d);
211 
212 	switch (d->bd_bufmode) {
213 	case BPF_BUFMODE_BUFFER:
214 		return (bpf_buffer_append_bytes(d, buf, offset, src, len));
215 
216 	case BPF_BUFMODE_ZBUF:
217 		d->bd_zcopy++;
218 		return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
219 
220 	default:
221 		panic("bpf_buf_append_bytes");
222 	}
223 }
224 
225 static void
226 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
227     u_int len)
228 {
229 
230 	BPFD_LOCK_ASSERT(d);
231 
232 	switch (d->bd_bufmode) {
233 	case BPF_BUFMODE_BUFFER:
234 		return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
235 
236 	case BPF_BUFMODE_ZBUF:
237 		d->bd_zcopy++;
238 		return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
239 
240 	default:
241 		panic("bpf_buf_append_mbuf");
242 	}
243 }
244 
245 /*
246  * This function gets called when the free buffer is re-assigned.
247  */
248 static void
249 bpf_buf_reclaimed(struct bpf_d *d)
250 {
251 
252 	BPFD_LOCK_ASSERT(d);
253 
254 	switch (d->bd_bufmode) {
255 	case BPF_BUFMODE_BUFFER:
256 		return;
257 
258 	case BPF_BUFMODE_ZBUF:
259 		bpf_zerocopy_buf_reclaimed(d);
260 		return;
261 
262 	default:
263 		panic("bpf_buf_reclaimed");
264 	}
265 }
266 
267 /*
268  * If the buffer mechanism has a way to decide that a held buffer can be made
269  * free, then it is exposed via the bpf_canfreebuf() interface.  (1) is
270  * returned if the buffer can be discarded, (0) is returned if it cannot.
271  */
272 static int
273 bpf_canfreebuf(struct bpf_d *d)
274 {
275 
276 	BPFD_LOCK_ASSERT(d);
277 
278 	switch (d->bd_bufmode) {
279 	case BPF_BUFMODE_ZBUF:
280 		return (bpf_zerocopy_canfreebuf(d));
281 	}
282 	return (0);
283 }
284 
285 /*
286  * Allow the buffer model to indicate that the current store buffer is
287  * immutable, regardless of the appearance of space.  Return (1) if the
288  * buffer is writable, and (0) if not.
289  */
290 static int
291 bpf_canwritebuf(struct bpf_d *d)
292 {
293 
294 	BPFD_LOCK_ASSERT(d);
295 
296 	switch (d->bd_bufmode) {
297 	case BPF_BUFMODE_ZBUF:
298 		return (bpf_zerocopy_canwritebuf(d));
299 	}
300 	return (1);
301 }
302 
303 /*
304  * Notify buffer model that an attempt to write to the store buffer has
305  * resulted in a dropped packet, in which case the buffer may be considered
306  * full.
307  */
308 static void
309 bpf_buffull(struct bpf_d *d)
310 {
311 
312 	BPFD_LOCK_ASSERT(d);
313 
314 	switch (d->bd_bufmode) {
315 	case BPF_BUFMODE_ZBUF:
316 		bpf_zerocopy_buffull(d);
317 		break;
318 	}
319 }
320 
321 /*
322  * Notify the buffer model that a buffer has moved into the hold position.
323  */
324 void
325 bpf_bufheld(struct bpf_d *d)
326 {
327 
328 	BPFD_LOCK_ASSERT(d);
329 
330 	switch (d->bd_bufmode) {
331 	case BPF_BUFMODE_ZBUF:
332 		bpf_zerocopy_bufheld(d);
333 		break;
334 	}
335 }
336 
337 static void
338 bpf_free(struct bpf_d *d)
339 {
340 
341 	switch (d->bd_bufmode) {
342 	case BPF_BUFMODE_BUFFER:
343 		return (bpf_buffer_free(d));
344 
345 	case BPF_BUFMODE_ZBUF:
346 		return (bpf_zerocopy_free(d));
347 
348 	default:
349 		panic("bpf_buf_free");
350 	}
351 }
352 
353 static int
354 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
355 {
356 
357 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
358 		return (EOPNOTSUPP);
359 	return (bpf_buffer_uiomove(d, buf, len, uio));
360 }
361 
362 static int
363 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
364 {
365 
366 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
367 		return (EOPNOTSUPP);
368 	return (bpf_buffer_ioctl_sblen(d, i));
369 }
370 
371 static int
372 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
373 {
374 
375 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
376 		return (EOPNOTSUPP);
377 	return (bpf_zerocopy_ioctl_getzmax(td, d, i));
378 }
379 
380 static int
381 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
382 {
383 
384 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
385 		return (EOPNOTSUPP);
386 	return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
387 }
388 
389 static int
390 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
391 {
392 
393 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
394 		return (EOPNOTSUPP);
395 	return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
396 }
397 
398 /*
399  * General BPF functions.
400  */
401 static int
402 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
403     struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter)
404 {
405 	const struct ieee80211_bpf_params *p;
406 	struct ether_header *eh;
407 	struct mbuf *m;
408 	int error;
409 	int len;
410 	int hlen;
411 	int slen;
412 
413 	/*
414 	 * Build a sockaddr based on the data link layer type.
415 	 * We do this at this level because the ethernet header
416 	 * is copied directly into the data field of the sockaddr.
417 	 * In the case of SLIP, there is no header and the packet
418 	 * is forwarded as is.
419 	 * Also, we are careful to leave room at the front of the mbuf
420 	 * for the link level header.
421 	 */
422 	switch (linktype) {
423 
424 	case DLT_SLIP:
425 		sockp->sa_family = AF_INET;
426 		hlen = 0;
427 		break;
428 
429 	case DLT_EN10MB:
430 		sockp->sa_family = AF_UNSPEC;
431 		/* XXX Would MAXLINKHDR be better? */
432 		hlen = ETHER_HDR_LEN;
433 		break;
434 
435 	case DLT_FDDI:
436 		sockp->sa_family = AF_IMPLINK;
437 		hlen = 0;
438 		break;
439 
440 	case DLT_RAW:
441 		sockp->sa_family = AF_UNSPEC;
442 		hlen = 0;
443 		break;
444 
445 	case DLT_NULL:
446 		/*
447 		 * null interface types require a 4 byte pseudo header which
448 		 * corresponds to the address family of the packet.
449 		 */
450 		sockp->sa_family = AF_UNSPEC;
451 		hlen = 4;
452 		break;
453 
454 	case DLT_ATM_RFC1483:
455 		/*
456 		 * en atm driver requires 4-byte atm pseudo header.
457 		 * though it isn't standard, vpi:vci needs to be
458 		 * specified anyway.
459 		 */
460 		sockp->sa_family = AF_UNSPEC;
461 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
462 		break;
463 
464 	case DLT_PPP:
465 		sockp->sa_family = AF_UNSPEC;
466 		hlen = 4;	/* This should match PPP_HDRLEN */
467 		break;
468 
469 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
470 		sockp->sa_family = AF_IEEE80211;
471 		hlen = 0;
472 		break;
473 
474 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
475 		sockp->sa_family = AF_IEEE80211;
476 		sockp->sa_len = 12;	/* XXX != 0 */
477 		hlen = sizeof(struct ieee80211_bpf_params);
478 		break;
479 
480 	default:
481 		return (EIO);
482 	}
483 
484 	len = uio->uio_resid;
485 
486 	if (len - hlen > ifp->if_mtu)
487 		return (EMSGSIZE);
488 
489 	if ((unsigned)len > MJUM16BYTES)
490 		return (EIO);
491 
492 	if (len <= MHLEN)
493 		MGETHDR(m, M_WAIT, MT_DATA);
494 	else if (len <= MCLBYTES)
495 		m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
496 	else
497 		m = m_getjcl(M_WAIT, MT_DATA, M_PKTHDR,
498 #if (MJUMPAGESIZE > MCLBYTES)
499 		    len <= MJUMPAGESIZE ? MJUMPAGESIZE :
500 #endif
501 		    (len <= MJUM9BYTES ? MJUM9BYTES : MJUM16BYTES));
502 	m->m_pkthdr.len = m->m_len = len;
503 	m->m_pkthdr.rcvif = NULL;
504 	*mp = m;
505 
506 	if (m->m_len < hlen) {
507 		error = EPERM;
508 		goto bad;
509 	}
510 
511 	error = uiomove(mtod(m, u_char *), len, uio);
512 	if (error)
513 		goto bad;
514 
515 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
516 	if (slen == 0) {
517 		error = EPERM;
518 		goto bad;
519 	}
520 
521 	/* Check for multicast destination */
522 	switch (linktype) {
523 	case DLT_EN10MB:
524 		eh = mtod(m, struct ether_header *);
525 		if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
526 			if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
527 			    ETHER_ADDR_LEN) == 0)
528 				m->m_flags |= M_BCAST;
529 			else
530 				m->m_flags |= M_MCAST;
531 		}
532 		break;
533 	}
534 
535 	/*
536 	 * Make room for link header, and copy it to sockaddr
537 	 */
538 	if (hlen != 0) {
539 		if (sockp->sa_family == AF_IEEE80211) {
540 			/*
541 			 * Collect true length from the parameter header
542 			 * NB: sockp is known to be zero'd so if we do a
543 			 *     short copy unspecified parameters will be
544 			 *     zero.
545 			 * NB: packet may not be aligned after stripping
546 			 *     bpf params
547 			 * XXX check ibp_vers
548 			 */
549 			p = mtod(m, const struct ieee80211_bpf_params *);
550 			hlen = p->ibp_len;
551 			if (hlen > sizeof(sockp->sa_data)) {
552 				error = EINVAL;
553 				goto bad;
554 			}
555 		}
556 		bcopy(m->m_data, sockp->sa_data, hlen);
557 	}
558 	*hdrlen = hlen;
559 
560 	return (0);
561 bad:
562 	m_freem(m);
563 	return (error);
564 }
565 
566 /*
567  * Attach file to the bpf interface, i.e. make d listen on bp.
568  */
569 static void
570 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
571 {
572 	/*
573 	 * Point d at bp, and add d to the interface's list of listeners.
574 	 * Finally, point the driver's bpf cookie at the interface so
575 	 * it will divert packets to bpf.
576 	 */
577 	BPFIF_LOCK(bp);
578 	d->bd_bif = bp;
579 	LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
580 
581 	bpf_bpfd_cnt++;
582 	BPFIF_UNLOCK(bp);
583 
584 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
585 }
586 
587 /*
588  * Detach a file from its interface.
589  */
590 static void
591 bpf_detachd(struct bpf_d *d)
592 {
593 	int error;
594 	struct bpf_if *bp;
595 	struct ifnet *ifp;
596 
597 	bp = d->bd_bif;
598 	BPFIF_LOCK(bp);
599 	BPFD_LOCK(d);
600 	ifp = d->bd_bif->bif_ifp;
601 
602 	/*
603 	 * Remove d from the interface's descriptor list.
604 	 */
605 	LIST_REMOVE(d, bd_next);
606 
607 	bpf_bpfd_cnt--;
608 	d->bd_bif = NULL;
609 	BPFD_UNLOCK(d);
610 	BPFIF_UNLOCK(bp);
611 
612 	EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
613 
614 	/*
615 	 * Check if this descriptor had requested promiscuous mode.
616 	 * If so, turn it off.
617 	 */
618 	if (d->bd_promisc) {
619 		d->bd_promisc = 0;
620 		CURVNET_SET(ifp->if_vnet);
621 		error = ifpromisc(ifp, 0);
622 		CURVNET_RESTORE();
623 		if (error != 0 && error != ENXIO) {
624 			/*
625 			 * ENXIO can happen if a pccard is unplugged
626 			 * Something is really wrong if we were able to put
627 			 * the driver into promiscuous mode, but can't
628 			 * take it out.
629 			 */
630 			if_printf(bp->bif_ifp,
631 				"bpf_detach: ifpromisc failed (%d)\n", error);
632 		}
633 	}
634 }
635 
636 /*
637  * Close the descriptor by detaching it from its interface,
638  * deallocating its buffers, and marking it free.
639  */
640 static void
641 bpf_dtor(void *data)
642 {
643 	struct bpf_d *d = data;
644 
645 	BPFD_LOCK(d);
646 	if (d->bd_state == BPF_WAITING)
647 		callout_stop(&d->bd_callout);
648 	d->bd_state = BPF_IDLE;
649 	BPFD_UNLOCK(d);
650 	funsetown(&d->bd_sigio);
651 	mtx_lock(&bpf_mtx);
652 	if (d->bd_bif)
653 		bpf_detachd(d);
654 	mtx_unlock(&bpf_mtx);
655 	selwakeuppri(&d->bd_sel, PRINET);
656 #ifdef MAC
657 	mac_bpfdesc_destroy(d);
658 #endif /* MAC */
659 	knlist_destroy(&d->bd_sel.si_note);
660 	callout_drain(&d->bd_callout);
661 	bpf_freed(d);
662 	free(d, M_BPF);
663 }
664 
665 /*
666  * Open ethernet device.  Returns ENXIO for illegal minor device number,
667  * EBUSY if file is open by another process.
668  */
669 /* ARGSUSED */
670 static	int
671 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
672 {
673 	struct bpf_d *d;
674 	int error;
675 
676 	d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
677 	error = devfs_set_cdevpriv(d, bpf_dtor);
678 	if (error != 0) {
679 		free(d, M_BPF);
680 		return (error);
681 	}
682 
683 	/*
684 	 * For historical reasons, perform a one-time initialization call to
685 	 * the buffer routines, even though we're not yet committed to a
686 	 * particular buffer method.
687 	 */
688 	bpf_buffer_init(d);
689 	d->bd_bufmode = BPF_BUFMODE_BUFFER;
690 	d->bd_sig = SIGIO;
691 	d->bd_direction = BPF_D_INOUT;
692 	d->bd_pid = td->td_proc->p_pid;
693 #ifdef MAC
694 	mac_bpfdesc_init(d);
695 	mac_bpfdesc_create(td->td_ucred, d);
696 #endif
697 	mtx_init(&d->bd_mtx, devtoname(dev), "bpf cdev lock", MTX_DEF);
698 	callout_init_mtx(&d->bd_callout, &d->bd_mtx, 0);
699 	knlist_init_mtx(&d->bd_sel.si_note, &d->bd_mtx);
700 
701 	return (0);
702 }
703 
704 /*
705  *  bpfread - read next chunk of packets from buffers
706  */
707 static	int
708 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
709 {
710 	struct bpf_d *d;
711 	int error;
712 	int non_block;
713 	int timed_out;
714 
715 	error = devfs_get_cdevpriv((void **)&d);
716 	if (error != 0)
717 		return (error);
718 
719 	/*
720 	 * Restrict application to use a buffer the same size as
721 	 * as kernel buffers.
722 	 */
723 	if (uio->uio_resid != d->bd_bufsize)
724 		return (EINVAL);
725 
726 	non_block = ((ioflag & O_NONBLOCK) != 0);
727 
728 	BPFD_LOCK(d);
729 	d->bd_pid = curthread->td_proc->p_pid;
730 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
731 		BPFD_UNLOCK(d);
732 		return (EOPNOTSUPP);
733 	}
734 	if (d->bd_state == BPF_WAITING)
735 		callout_stop(&d->bd_callout);
736 	timed_out = (d->bd_state == BPF_TIMED_OUT);
737 	d->bd_state = BPF_IDLE;
738 	/*
739 	 * If the hold buffer is empty, then do a timed sleep, which
740 	 * ends when the timeout expires or when enough packets
741 	 * have arrived to fill the store buffer.
742 	 */
743 	while (d->bd_hbuf == NULL) {
744 		if (d->bd_slen != 0) {
745 			/*
746 			 * A packet(s) either arrived since the previous
747 			 * read or arrived while we were asleep.
748 			 */
749 			if (d->bd_immediate || non_block || timed_out) {
750 				/*
751 				 * Rotate the buffers and return what's here
752 				 * if we are in immediate mode, non-blocking
753 				 * flag is set, or this descriptor timed out.
754 				 */
755 				ROTATE_BUFFERS(d);
756 				break;
757 			}
758 		}
759 
760 		/*
761 		 * No data is available, check to see if the bpf device
762 		 * is still pointed at a real interface.  If not, return
763 		 * ENXIO so that the userland process knows to rebind
764 		 * it before using it again.
765 		 */
766 		if (d->bd_bif == NULL) {
767 			BPFD_UNLOCK(d);
768 			return (ENXIO);
769 		}
770 
771 		if (non_block) {
772 			BPFD_UNLOCK(d);
773 			return (EWOULDBLOCK);
774 		}
775 		error = msleep(d, &d->bd_mtx, PRINET|PCATCH,
776 		     "bpf", d->bd_rtout);
777 		if (error == EINTR || error == ERESTART) {
778 			BPFD_UNLOCK(d);
779 			return (error);
780 		}
781 		if (error == EWOULDBLOCK) {
782 			/*
783 			 * On a timeout, return what's in the buffer,
784 			 * which may be nothing.  If there is something
785 			 * in the store buffer, we can rotate the buffers.
786 			 */
787 			if (d->bd_hbuf)
788 				/*
789 				 * We filled up the buffer in between
790 				 * getting the timeout and arriving
791 				 * here, so we don't need to rotate.
792 				 */
793 				break;
794 
795 			if (d->bd_slen == 0) {
796 				BPFD_UNLOCK(d);
797 				return (0);
798 			}
799 			ROTATE_BUFFERS(d);
800 			break;
801 		}
802 	}
803 	/*
804 	 * At this point, we know we have something in the hold slot.
805 	 */
806 	BPFD_UNLOCK(d);
807 
808 	/*
809 	 * Move data from hold buffer into user space.
810 	 * We know the entire buffer is transferred since
811 	 * we checked above that the read buffer is bpf_bufsize bytes.
812 	 *
813 	 * XXXRW: More synchronization needed here: what if a second thread
814 	 * issues a read on the same fd at the same time?  Don't want this
815 	 * getting invalidated.
816 	 */
817 	error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
818 
819 	BPFD_LOCK(d);
820 	d->bd_fbuf = d->bd_hbuf;
821 	d->bd_hbuf = NULL;
822 	d->bd_hlen = 0;
823 	bpf_buf_reclaimed(d);
824 	BPFD_UNLOCK(d);
825 
826 	return (error);
827 }
828 
829 /*
830  * If there are processes sleeping on this descriptor, wake them up.
831  */
832 static __inline void
833 bpf_wakeup(struct bpf_d *d)
834 {
835 
836 	BPFD_LOCK_ASSERT(d);
837 	if (d->bd_state == BPF_WAITING) {
838 		callout_stop(&d->bd_callout);
839 		d->bd_state = BPF_IDLE;
840 	}
841 	wakeup(d);
842 	if (d->bd_async && d->bd_sig && d->bd_sigio)
843 		pgsigio(&d->bd_sigio, d->bd_sig, 0);
844 
845 	selwakeuppri(&d->bd_sel, PRINET);
846 	KNOTE_LOCKED(&d->bd_sel.si_note, 0);
847 }
848 
849 static void
850 bpf_timed_out(void *arg)
851 {
852 	struct bpf_d *d = (struct bpf_d *)arg;
853 
854 	BPFD_LOCK_ASSERT(d);
855 
856 	if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
857 		return;
858 	if (d->bd_state == BPF_WAITING) {
859 		d->bd_state = BPF_TIMED_OUT;
860 		if (d->bd_slen != 0)
861 			bpf_wakeup(d);
862 	}
863 }
864 
865 static int
866 bpf_ready(struct bpf_d *d)
867 {
868 
869 	BPFD_LOCK_ASSERT(d);
870 
871 	if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
872 		return (1);
873 	if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
874 	    d->bd_slen != 0)
875 		return (1);
876 	return (0);
877 }
878 
879 static int
880 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
881 {
882 	struct bpf_d *d;
883 	struct ifnet *ifp;
884 	struct mbuf *m, *mc;
885 	struct sockaddr dst;
886 	int error, hlen;
887 
888 	error = devfs_get_cdevpriv((void **)&d);
889 	if (error != 0)
890 		return (error);
891 
892 	d->bd_pid = curthread->td_proc->p_pid;
893 	d->bd_wcount++;
894 	if (d->bd_bif == NULL) {
895 		d->bd_wdcount++;
896 		return (ENXIO);
897 	}
898 
899 	ifp = d->bd_bif->bif_ifp;
900 
901 	if ((ifp->if_flags & IFF_UP) == 0) {
902 		d->bd_wdcount++;
903 		return (ENETDOWN);
904 	}
905 
906 	if (uio->uio_resid == 0) {
907 		d->bd_wdcount++;
908 		return (0);
909 	}
910 
911 	bzero(&dst, sizeof(dst));
912 	m = NULL;
913 	hlen = 0;
914 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
915 	    &m, &dst, &hlen, d->bd_wfilter);
916 	if (error) {
917 		d->bd_wdcount++;
918 		return (error);
919 	}
920 	d->bd_wfcount++;
921 	if (d->bd_hdrcmplt)
922 		dst.sa_family = pseudo_AF_HDRCMPLT;
923 
924 	if (d->bd_feedback) {
925 		mc = m_dup(m, M_DONTWAIT);
926 		if (mc != NULL)
927 			mc->m_pkthdr.rcvif = ifp;
928 		/* Set M_PROMISC for outgoing packets to be discarded. */
929 		if (d->bd_direction == BPF_D_INOUT)
930 			m->m_flags |= M_PROMISC;
931 	} else
932 		mc = NULL;
933 
934 	m->m_pkthdr.len -= hlen;
935 	m->m_len -= hlen;
936 	m->m_data += hlen;	/* XXX */
937 
938 	CURVNET_SET(ifp->if_vnet);
939 #ifdef MAC
940 	BPFD_LOCK(d);
941 	mac_bpfdesc_create_mbuf(d, m);
942 	if (mc != NULL)
943 		mac_bpfdesc_create_mbuf(d, mc);
944 	BPFD_UNLOCK(d);
945 #endif
946 
947 	error = (*ifp->if_output)(ifp, m, &dst, NULL);
948 	if (error)
949 		d->bd_wdcount++;
950 
951 	if (mc != NULL) {
952 		if (error == 0)
953 			(*ifp->if_input)(ifp, mc);
954 		else
955 			m_freem(mc);
956 	}
957 	CURVNET_RESTORE();
958 
959 	return (error);
960 }
961 
962 /*
963  * Reset a descriptor by flushing its packet buffer and clearing the receive
964  * and drop counts.  This is doable for kernel-only buffers, but with
965  * zero-copy buffers, we can't write to (or rotate) buffers that are
966  * currently owned by userspace.  It would be nice if we could encapsulate
967  * this logic in the buffer code rather than here.
968  */
969 static void
970 reset_d(struct bpf_d *d)
971 {
972 
973 	mtx_assert(&d->bd_mtx, MA_OWNED);
974 
975 	if ((d->bd_hbuf != NULL) &&
976 	    (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
977 		/* Free the hold buffer. */
978 		d->bd_fbuf = d->bd_hbuf;
979 		d->bd_hbuf = NULL;
980 		d->bd_hlen = 0;
981 		bpf_buf_reclaimed(d);
982 	}
983 	if (bpf_canwritebuf(d))
984 		d->bd_slen = 0;
985 	d->bd_rcount = 0;
986 	d->bd_dcount = 0;
987 	d->bd_fcount = 0;
988 	d->bd_wcount = 0;
989 	d->bd_wfcount = 0;
990 	d->bd_wdcount = 0;
991 	d->bd_zcopy = 0;
992 }
993 
994 /*
995  *  FIONREAD		Check for read packet available.
996  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
997  *  BIOCGBLEN		Get buffer len [for read()].
998  *  BIOCSETF		Set read filter.
999  *  BIOCSETFNR		Set read filter without resetting descriptor.
1000  *  BIOCSETWF		Set write filter.
1001  *  BIOCFLUSH		Flush read packet buffer.
1002  *  BIOCPROMISC		Put interface into promiscuous mode.
1003  *  BIOCGDLT		Get link layer type.
1004  *  BIOCGETIF		Get interface name.
1005  *  BIOCSETIF		Set interface.
1006  *  BIOCSRTIMEOUT	Set read timeout.
1007  *  BIOCGRTIMEOUT	Get read timeout.
1008  *  BIOCGSTATS		Get packet stats.
1009  *  BIOCIMMEDIATE	Set immediate mode.
1010  *  BIOCVERSION		Get filter language version.
1011  *  BIOCGHDRCMPLT	Get "header already complete" flag
1012  *  BIOCSHDRCMPLT	Set "header already complete" flag
1013  *  BIOCGDIRECTION	Get packet direction flag
1014  *  BIOCSDIRECTION	Set packet direction flag
1015  *  BIOCGTSTAMP		Get time stamp format and resolution.
1016  *  BIOCSTSTAMP		Set time stamp format and resolution.
1017  *  BIOCLOCK		Set "locked" flag
1018  *  BIOCFEEDBACK	Set packet feedback mode.
1019  *  BIOCSETZBUF		Set current zero-copy buffer locations.
1020  *  BIOCGETZMAX		Get maximum zero-copy buffer size.
1021  *  BIOCROTZBUF		Force rotation of zero-copy buffer
1022  *  BIOCSETBUFMODE	Set buffer mode.
1023  *  BIOCGETBUFMODE	Get current buffer mode.
1024  */
1025 /* ARGSUSED */
1026 static	int
1027 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1028     struct thread *td)
1029 {
1030 	struct bpf_d *d;
1031 	int error;
1032 
1033 	error = devfs_get_cdevpriv((void **)&d);
1034 	if (error != 0)
1035 		return (error);
1036 
1037 	/*
1038 	 * Refresh PID associated with this descriptor.
1039 	 */
1040 	BPFD_LOCK(d);
1041 	d->bd_pid = td->td_proc->p_pid;
1042 	if (d->bd_state == BPF_WAITING)
1043 		callout_stop(&d->bd_callout);
1044 	d->bd_state = BPF_IDLE;
1045 	BPFD_UNLOCK(d);
1046 
1047 	if (d->bd_locked == 1) {
1048 		switch (cmd) {
1049 		case BIOCGBLEN:
1050 		case BIOCFLUSH:
1051 		case BIOCGDLT:
1052 		case BIOCGDLTLIST:
1053 #ifdef COMPAT_FREEBSD32
1054 		case BIOCGDLTLIST32:
1055 #endif
1056 		case BIOCGETIF:
1057 		case BIOCGRTIMEOUT:
1058 #ifdef COMPAT_FREEBSD32
1059 		case BIOCGRTIMEOUT32:
1060 #endif
1061 		case BIOCGSTATS:
1062 		case BIOCVERSION:
1063 		case BIOCGRSIG:
1064 		case BIOCGHDRCMPLT:
1065 		case BIOCSTSTAMP:
1066 		case BIOCFEEDBACK:
1067 		case FIONREAD:
1068 		case BIOCLOCK:
1069 		case BIOCSRTIMEOUT:
1070 #ifdef COMPAT_FREEBSD32
1071 		case BIOCSRTIMEOUT32:
1072 #endif
1073 		case BIOCIMMEDIATE:
1074 		case TIOCGPGRP:
1075 		case BIOCROTZBUF:
1076 			break;
1077 		default:
1078 			return (EPERM);
1079 		}
1080 	}
1081 #ifdef COMPAT_FREEBSD32
1082 	/*
1083 	 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1084 	 * that it will get 32-bit packet headers.
1085 	 */
1086 	switch (cmd) {
1087 	case BIOCSETF32:
1088 	case BIOCSETFNR32:
1089 	case BIOCSETWF32:
1090 	case BIOCGDLTLIST32:
1091 	case BIOCGRTIMEOUT32:
1092 	case BIOCSRTIMEOUT32:
1093 		d->bd_compat32 = 1;
1094 	}
1095 #endif
1096 
1097 	CURVNET_SET(TD_TO_VNET(td));
1098 	switch (cmd) {
1099 
1100 	default:
1101 		error = EINVAL;
1102 		break;
1103 
1104 	/*
1105 	 * Check for read packet available.
1106 	 */
1107 	case FIONREAD:
1108 		{
1109 			int n;
1110 
1111 			BPFD_LOCK(d);
1112 			n = d->bd_slen;
1113 			if (d->bd_hbuf)
1114 				n += d->bd_hlen;
1115 			BPFD_UNLOCK(d);
1116 
1117 			*(int *)addr = n;
1118 			break;
1119 		}
1120 
1121 	case SIOCGIFADDR:
1122 		{
1123 			struct ifnet *ifp;
1124 
1125 			if (d->bd_bif == NULL)
1126 				error = EINVAL;
1127 			else {
1128 				ifp = d->bd_bif->bif_ifp;
1129 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
1130 			}
1131 			break;
1132 		}
1133 
1134 	/*
1135 	 * Get buffer len [for read()].
1136 	 */
1137 	case BIOCGBLEN:
1138 		*(u_int *)addr = d->bd_bufsize;
1139 		break;
1140 
1141 	/*
1142 	 * Set buffer length.
1143 	 */
1144 	case BIOCSBLEN:
1145 		error = bpf_ioctl_sblen(d, (u_int *)addr);
1146 		break;
1147 
1148 	/*
1149 	 * Set link layer read filter.
1150 	 */
1151 	case BIOCSETF:
1152 	case BIOCSETFNR:
1153 	case BIOCSETWF:
1154 #ifdef COMPAT_FREEBSD32
1155 	case BIOCSETF32:
1156 	case BIOCSETFNR32:
1157 	case BIOCSETWF32:
1158 #endif
1159 		error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1160 		break;
1161 
1162 	/*
1163 	 * Flush read packet buffer.
1164 	 */
1165 	case BIOCFLUSH:
1166 		BPFD_LOCK(d);
1167 		reset_d(d);
1168 		BPFD_UNLOCK(d);
1169 		break;
1170 
1171 	/*
1172 	 * Put interface into promiscuous mode.
1173 	 */
1174 	case BIOCPROMISC:
1175 		if (d->bd_bif == NULL) {
1176 			/*
1177 			 * No interface attached yet.
1178 			 */
1179 			error = EINVAL;
1180 			break;
1181 		}
1182 		if (d->bd_promisc == 0) {
1183 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1184 			if (error == 0)
1185 				d->bd_promisc = 1;
1186 		}
1187 		break;
1188 
1189 	/*
1190 	 * Get current data link type.
1191 	 */
1192 	case BIOCGDLT:
1193 		if (d->bd_bif == NULL)
1194 			error = EINVAL;
1195 		else
1196 			*(u_int *)addr = d->bd_bif->bif_dlt;
1197 		break;
1198 
1199 	/*
1200 	 * Get a list of supported data link types.
1201 	 */
1202 #ifdef COMPAT_FREEBSD32
1203 	case BIOCGDLTLIST32:
1204 		{
1205 			struct bpf_dltlist32 *list32;
1206 			struct bpf_dltlist dltlist;
1207 
1208 			list32 = (struct bpf_dltlist32 *)addr;
1209 			dltlist.bfl_len = list32->bfl_len;
1210 			dltlist.bfl_list = PTRIN(list32->bfl_list);
1211 			if (d->bd_bif == NULL)
1212 				error = EINVAL;
1213 			else {
1214 				error = bpf_getdltlist(d, &dltlist);
1215 				if (error == 0)
1216 					list32->bfl_len = dltlist.bfl_len;
1217 			}
1218 			break;
1219 		}
1220 #endif
1221 
1222 	case BIOCGDLTLIST:
1223 		if (d->bd_bif == NULL)
1224 			error = EINVAL;
1225 		else
1226 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1227 		break;
1228 
1229 	/*
1230 	 * Set data link type.
1231 	 */
1232 	case BIOCSDLT:
1233 		if (d->bd_bif == NULL)
1234 			error = EINVAL;
1235 		else
1236 			error = bpf_setdlt(d, *(u_int *)addr);
1237 		break;
1238 
1239 	/*
1240 	 * Get interface name.
1241 	 */
1242 	case BIOCGETIF:
1243 		if (d->bd_bif == NULL)
1244 			error = EINVAL;
1245 		else {
1246 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
1247 			struct ifreq *const ifr = (struct ifreq *)addr;
1248 
1249 			strlcpy(ifr->ifr_name, ifp->if_xname,
1250 			    sizeof(ifr->ifr_name));
1251 		}
1252 		break;
1253 
1254 	/*
1255 	 * Set interface.
1256 	 */
1257 	case BIOCSETIF:
1258 		error = bpf_setif(d, (struct ifreq *)addr);
1259 		break;
1260 
1261 	/*
1262 	 * Set read timeout.
1263 	 */
1264 	case BIOCSRTIMEOUT:
1265 #ifdef COMPAT_FREEBSD32
1266 	case BIOCSRTIMEOUT32:
1267 #endif
1268 		{
1269 			struct timeval *tv = (struct timeval *)addr;
1270 #ifdef COMPAT_FREEBSD32
1271 			struct timeval32 *tv32;
1272 			struct timeval tv64;
1273 
1274 			if (cmd == BIOCSRTIMEOUT32) {
1275 				tv32 = (struct timeval32 *)addr;
1276 				tv = &tv64;
1277 				tv->tv_sec = tv32->tv_sec;
1278 				tv->tv_usec = tv32->tv_usec;
1279 			} else
1280 #endif
1281 				tv = (struct timeval *)addr;
1282 
1283 			/*
1284 			 * Subtract 1 tick from tvtohz() since this isn't
1285 			 * a one-shot timer.
1286 			 */
1287 			if ((error = itimerfix(tv)) == 0)
1288 				d->bd_rtout = tvtohz(tv) - 1;
1289 			break;
1290 		}
1291 
1292 	/*
1293 	 * Get read timeout.
1294 	 */
1295 	case BIOCGRTIMEOUT:
1296 #ifdef COMPAT_FREEBSD32
1297 	case BIOCGRTIMEOUT32:
1298 #endif
1299 		{
1300 			struct timeval *tv;
1301 #ifdef COMPAT_FREEBSD32
1302 			struct timeval32 *tv32;
1303 			struct timeval tv64;
1304 
1305 			if (cmd == BIOCGRTIMEOUT32)
1306 				tv = &tv64;
1307 			else
1308 #endif
1309 				tv = (struct timeval *)addr;
1310 
1311 			tv->tv_sec = d->bd_rtout / hz;
1312 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1313 #ifdef COMPAT_FREEBSD32
1314 			if (cmd == BIOCGRTIMEOUT32) {
1315 				tv32 = (struct timeval32 *)addr;
1316 				tv32->tv_sec = tv->tv_sec;
1317 				tv32->tv_usec = tv->tv_usec;
1318 			}
1319 #endif
1320 
1321 			break;
1322 		}
1323 
1324 	/*
1325 	 * Get packet stats.
1326 	 */
1327 	case BIOCGSTATS:
1328 		{
1329 			struct bpf_stat *bs = (struct bpf_stat *)addr;
1330 
1331 			/* XXXCSJP overflow */
1332 			bs->bs_recv = d->bd_rcount;
1333 			bs->bs_drop = d->bd_dcount;
1334 			break;
1335 		}
1336 
1337 	/*
1338 	 * Set immediate mode.
1339 	 */
1340 	case BIOCIMMEDIATE:
1341 		d->bd_immediate = *(u_int *)addr;
1342 		break;
1343 
1344 	case BIOCVERSION:
1345 		{
1346 			struct bpf_version *bv = (struct bpf_version *)addr;
1347 
1348 			bv->bv_major = BPF_MAJOR_VERSION;
1349 			bv->bv_minor = BPF_MINOR_VERSION;
1350 			break;
1351 		}
1352 
1353 	/*
1354 	 * Get "header already complete" flag
1355 	 */
1356 	case BIOCGHDRCMPLT:
1357 		*(u_int *)addr = d->bd_hdrcmplt;
1358 		break;
1359 
1360 	/*
1361 	 * Set "header already complete" flag
1362 	 */
1363 	case BIOCSHDRCMPLT:
1364 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1365 		break;
1366 
1367 	/*
1368 	 * Get packet direction flag
1369 	 */
1370 	case BIOCGDIRECTION:
1371 		*(u_int *)addr = d->bd_direction;
1372 		break;
1373 
1374 	/*
1375 	 * Set packet direction flag
1376 	 */
1377 	case BIOCSDIRECTION:
1378 		{
1379 			u_int	direction;
1380 
1381 			direction = *(u_int *)addr;
1382 			switch (direction) {
1383 			case BPF_D_IN:
1384 			case BPF_D_INOUT:
1385 			case BPF_D_OUT:
1386 				d->bd_direction = direction;
1387 				break;
1388 			default:
1389 				error = EINVAL;
1390 			}
1391 		}
1392 		break;
1393 
1394 	/*
1395 	 * Get packet timestamp format and resolution.
1396 	 */
1397 	case BIOCGTSTAMP:
1398 		*(u_int *)addr = d->bd_tstamp;
1399 		break;
1400 
1401 	/*
1402 	 * Set packet timestamp format and resolution.
1403 	 */
1404 	case BIOCSTSTAMP:
1405 		{
1406 			u_int	func;
1407 
1408 			func = *(u_int *)addr;
1409 			if (BPF_T_VALID(func))
1410 				d->bd_tstamp = func;
1411 			else
1412 				error = EINVAL;
1413 		}
1414 		break;
1415 
1416 	case BIOCFEEDBACK:
1417 		d->bd_feedback = *(u_int *)addr;
1418 		break;
1419 
1420 	case BIOCLOCK:
1421 		d->bd_locked = 1;
1422 		break;
1423 
1424 	case FIONBIO:		/* Non-blocking I/O */
1425 		break;
1426 
1427 	case FIOASYNC:		/* Send signal on receive packets */
1428 		d->bd_async = *(int *)addr;
1429 		break;
1430 
1431 	case FIOSETOWN:
1432 		error = fsetown(*(int *)addr, &d->bd_sigio);
1433 		break;
1434 
1435 	case FIOGETOWN:
1436 		*(int *)addr = fgetown(&d->bd_sigio);
1437 		break;
1438 
1439 	/* This is deprecated, FIOSETOWN should be used instead. */
1440 	case TIOCSPGRP:
1441 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
1442 		break;
1443 
1444 	/* This is deprecated, FIOGETOWN should be used instead. */
1445 	case TIOCGPGRP:
1446 		*(int *)addr = -fgetown(&d->bd_sigio);
1447 		break;
1448 
1449 	case BIOCSRSIG:		/* Set receive signal */
1450 		{
1451 			u_int sig;
1452 
1453 			sig = *(u_int *)addr;
1454 
1455 			if (sig >= NSIG)
1456 				error = EINVAL;
1457 			else
1458 				d->bd_sig = sig;
1459 			break;
1460 		}
1461 	case BIOCGRSIG:
1462 		*(u_int *)addr = d->bd_sig;
1463 		break;
1464 
1465 	case BIOCGETBUFMODE:
1466 		*(u_int *)addr = d->bd_bufmode;
1467 		break;
1468 
1469 	case BIOCSETBUFMODE:
1470 		/*
1471 		 * Allow the buffering mode to be changed as long as we
1472 		 * haven't yet committed to a particular mode.  Our
1473 		 * definition of commitment, for now, is whether or not a
1474 		 * buffer has been allocated or an interface attached, since
1475 		 * that's the point where things get tricky.
1476 		 */
1477 		switch (*(u_int *)addr) {
1478 		case BPF_BUFMODE_BUFFER:
1479 			break;
1480 
1481 		case BPF_BUFMODE_ZBUF:
1482 			if (bpf_zerocopy_enable)
1483 				break;
1484 			/* FALLSTHROUGH */
1485 
1486 		default:
1487 			CURVNET_RESTORE();
1488 			return (EINVAL);
1489 		}
1490 
1491 		BPFD_LOCK(d);
1492 		if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1493 		    d->bd_fbuf != NULL || d->bd_bif != NULL) {
1494 			BPFD_UNLOCK(d);
1495 			CURVNET_RESTORE();
1496 			return (EBUSY);
1497 		}
1498 		d->bd_bufmode = *(u_int *)addr;
1499 		BPFD_UNLOCK(d);
1500 		break;
1501 
1502 	case BIOCGETZMAX:
1503 		error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1504 		break;
1505 
1506 	case BIOCSETZBUF:
1507 		error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1508 		break;
1509 
1510 	case BIOCROTZBUF:
1511 		error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1512 		break;
1513 	}
1514 	CURVNET_RESTORE();
1515 	return (error);
1516 }
1517 
1518 /*
1519  * Set d's packet filter program to fp.  If this file already has a filter,
1520  * free it and replace it.  Returns EINVAL for bogus requests.
1521  */
1522 static int
1523 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1524 {
1525 	struct bpf_insn *fcode, *old;
1526 	u_int wfilter, flen, size;
1527 #ifdef BPF_JITTER
1528 	bpf_jit_filter *ofunc;
1529 #endif
1530 #ifdef COMPAT_FREEBSD32
1531 	struct bpf_program32 *fp32;
1532 	struct bpf_program fp_swab;
1533 
1534 	if (cmd == BIOCSETWF32 || cmd == BIOCSETF32 || cmd == BIOCSETFNR32) {
1535 		fp32 = (struct bpf_program32 *)fp;
1536 		fp_swab.bf_len = fp32->bf_len;
1537 		fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1538 		fp = &fp_swab;
1539 		if (cmd == BIOCSETWF32)
1540 			cmd = BIOCSETWF;
1541 	}
1542 #endif
1543 	if (cmd == BIOCSETWF) {
1544 		old = d->bd_wfilter;
1545 		wfilter = 1;
1546 #ifdef BPF_JITTER
1547 		ofunc = NULL;
1548 #endif
1549 	} else {
1550 		wfilter = 0;
1551 		old = d->bd_rfilter;
1552 #ifdef BPF_JITTER
1553 		ofunc = d->bd_bfilter;
1554 #endif
1555 	}
1556 	if (fp->bf_insns == NULL) {
1557 		if (fp->bf_len != 0)
1558 			return (EINVAL);
1559 		BPFD_LOCK(d);
1560 		if (wfilter)
1561 			d->bd_wfilter = NULL;
1562 		else {
1563 			d->bd_rfilter = NULL;
1564 #ifdef BPF_JITTER
1565 			d->bd_bfilter = NULL;
1566 #endif
1567 			if (cmd == BIOCSETF)
1568 				reset_d(d);
1569 		}
1570 		BPFD_UNLOCK(d);
1571 		if (old != NULL)
1572 			free((caddr_t)old, M_BPF);
1573 #ifdef BPF_JITTER
1574 		if (ofunc != NULL)
1575 			bpf_destroy_jit_filter(ofunc);
1576 #endif
1577 		return (0);
1578 	}
1579 	flen = fp->bf_len;
1580 	if (flen > bpf_maxinsns)
1581 		return (EINVAL);
1582 
1583 	size = flen * sizeof(*fp->bf_insns);
1584 	fcode = (struct bpf_insn *)malloc(size, M_BPF, M_WAITOK);
1585 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
1586 	    bpf_validate(fcode, (int)flen)) {
1587 		BPFD_LOCK(d);
1588 		if (wfilter)
1589 			d->bd_wfilter = fcode;
1590 		else {
1591 			d->bd_rfilter = fcode;
1592 #ifdef BPF_JITTER
1593 			d->bd_bfilter = bpf_jitter(fcode, flen);
1594 #endif
1595 			if (cmd == BIOCSETF)
1596 				reset_d(d);
1597 		}
1598 		BPFD_UNLOCK(d);
1599 		if (old != NULL)
1600 			free((caddr_t)old, M_BPF);
1601 #ifdef BPF_JITTER
1602 		if (ofunc != NULL)
1603 			bpf_destroy_jit_filter(ofunc);
1604 #endif
1605 
1606 		return (0);
1607 	}
1608 	free((caddr_t)fcode, M_BPF);
1609 	return (EINVAL);
1610 }
1611 
1612 /*
1613  * Detach a file from its current interface (if attached at all) and attach
1614  * to the interface indicated by the name stored in ifr.
1615  * Return an errno or 0.
1616  */
1617 static int
1618 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1619 {
1620 	struct bpf_if *bp;
1621 	struct ifnet *theywant;
1622 
1623 	theywant = ifunit(ifr->ifr_name);
1624 	if (theywant == NULL || theywant->if_bpf == NULL)
1625 		return (ENXIO);
1626 
1627 	bp = theywant->if_bpf;
1628 
1629 	/*
1630 	 * Behavior here depends on the buffering model.  If we're using
1631 	 * kernel memory buffers, then we can allocate them here.  If we're
1632 	 * using zero-copy, then the user process must have registered
1633 	 * buffers by the time we get here.  If not, return an error.
1634 	 *
1635 	 * XXXRW: There are locking issues here with multi-threaded use: what
1636 	 * if two threads try to set the interface at once?
1637 	 */
1638 	switch (d->bd_bufmode) {
1639 	case BPF_BUFMODE_BUFFER:
1640 		if (d->bd_sbuf == NULL)
1641 			bpf_buffer_alloc(d);
1642 		KASSERT(d->bd_sbuf != NULL, ("bpf_setif: bd_sbuf NULL"));
1643 		break;
1644 
1645 	case BPF_BUFMODE_ZBUF:
1646 		if (d->bd_sbuf == NULL)
1647 			return (EINVAL);
1648 		break;
1649 
1650 	default:
1651 		panic("bpf_setif: bufmode %d", d->bd_bufmode);
1652 	}
1653 	if (bp != d->bd_bif) {
1654 		if (d->bd_bif)
1655 			/*
1656 			 * Detach if attached to something else.
1657 			 */
1658 			bpf_detachd(d);
1659 
1660 		bpf_attachd(d, bp);
1661 	}
1662 	BPFD_LOCK(d);
1663 	reset_d(d);
1664 	BPFD_UNLOCK(d);
1665 	return (0);
1666 }
1667 
1668 /*
1669  * Support for select() and poll() system calls
1670  *
1671  * Return true iff the specific operation will not block indefinitely.
1672  * Otherwise, return false but make a note that a selwakeup() must be done.
1673  */
1674 static int
1675 bpfpoll(struct cdev *dev, int events, struct thread *td)
1676 {
1677 	struct bpf_d *d;
1678 	int revents;
1679 
1680 	if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1681 		return (events &
1682 		    (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1683 
1684 	/*
1685 	 * Refresh PID associated with this descriptor.
1686 	 */
1687 	revents = events & (POLLOUT | POLLWRNORM);
1688 	BPFD_LOCK(d);
1689 	d->bd_pid = td->td_proc->p_pid;
1690 	if (events & (POLLIN | POLLRDNORM)) {
1691 		if (bpf_ready(d))
1692 			revents |= events & (POLLIN | POLLRDNORM);
1693 		else {
1694 			selrecord(td, &d->bd_sel);
1695 			/* Start the read timeout if necessary. */
1696 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1697 				callout_reset(&d->bd_callout, d->bd_rtout,
1698 				    bpf_timed_out, d);
1699 				d->bd_state = BPF_WAITING;
1700 			}
1701 		}
1702 	}
1703 	BPFD_UNLOCK(d);
1704 	return (revents);
1705 }
1706 
1707 /*
1708  * Support for kevent() system call.  Register EVFILT_READ filters and
1709  * reject all others.
1710  */
1711 int
1712 bpfkqfilter(struct cdev *dev, struct knote *kn)
1713 {
1714 	struct bpf_d *d;
1715 
1716 	if (devfs_get_cdevpriv((void **)&d) != 0 ||
1717 	    kn->kn_filter != EVFILT_READ)
1718 		return (1);
1719 
1720 	/*
1721 	 * Refresh PID associated with this descriptor.
1722 	 */
1723 	BPFD_LOCK(d);
1724 	d->bd_pid = curthread->td_proc->p_pid;
1725 	kn->kn_fop = &bpfread_filtops;
1726 	kn->kn_hook = d;
1727 	knlist_add(&d->bd_sel.si_note, kn, 1);
1728 	BPFD_UNLOCK(d);
1729 
1730 	return (0);
1731 }
1732 
1733 static void
1734 filt_bpfdetach(struct knote *kn)
1735 {
1736 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1737 
1738 	knlist_remove(&d->bd_sel.si_note, kn, 0);
1739 }
1740 
1741 static int
1742 filt_bpfread(struct knote *kn, long hint)
1743 {
1744 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1745 	int ready;
1746 
1747 	BPFD_LOCK_ASSERT(d);
1748 	ready = bpf_ready(d);
1749 	if (ready) {
1750 		kn->kn_data = d->bd_slen;
1751 		if (d->bd_hbuf)
1752 			kn->kn_data += d->bd_hlen;
1753 	} else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1754 		callout_reset(&d->bd_callout, d->bd_rtout,
1755 		    bpf_timed_out, d);
1756 		d->bd_state = BPF_WAITING;
1757 	}
1758 
1759 	return (ready);
1760 }
1761 
1762 #define	BPF_TSTAMP_NONE		0
1763 #define	BPF_TSTAMP_FAST		1
1764 #define	BPF_TSTAMP_NORMAL	2
1765 #define	BPF_TSTAMP_EXTERN	3
1766 
1767 static int
1768 bpf_ts_quality(int tstype)
1769 {
1770 
1771 	if (tstype == BPF_T_NONE)
1772 		return (BPF_TSTAMP_NONE);
1773 	if ((tstype & BPF_T_FAST) != 0)
1774 		return (BPF_TSTAMP_FAST);
1775 
1776 	return (BPF_TSTAMP_NORMAL);
1777 }
1778 
1779 static int
1780 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
1781 {
1782 	struct m_tag *tag;
1783 	int quality;
1784 
1785 	quality = bpf_ts_quality(tstype);
1786 	if (quality == BPF_TSTAMP_NONE)
1787 		return (quality);
1788 
1789 	if (m != NULL) {
1790 		tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
1791 		if (tag != NULL) {
1792 			*bt = *(struct bintime *)(tag + 1);
1793 			return (BPF_TSTAMP_EXTERN);
1794 		}
1795 	}
1796 	if (quality == BPF_TSTAMP_NORMAL)
1797 		binuptime(bt);
1798 	else
1799 		getbinuptime(bt);
1800 
1801 	return (quality);
1802 }
1803 
1804 /*
1805  * Incoming linkage from device drivers.  Process the packet pkt, of length
1806  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1807  * by each process' filter, and if accepted, stashed into the corresponding
1808  * buffer.
1809  */
1810 void
1811 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1812 {
1813 	struct bintime bt;
1814 	struct bpf_d *d;
1815 #ifdef BPF_JITTER
1816 	bpf_jit_filter *bf;
1817 #endif
1818 	u_int slen;
1819 	int gottime;
1820 
1821 	gottime = BPF_TSTAMP_NONE;
1822 	BPFIF_LOCK(bp);
1823 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1824 		BPFD_LOCK(d);
1825 		++d->bd_rcount;
1826 		/*
1827 		 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
1828 		 * way for the caller to indiciate to us whether this packet
1829 		 * is inbound or outbound.  In the bpf_mtap() routines, we use
1830 		 * the interface pointers on the mbuf to figure it out.
1831 		 */
1832 #ifdef BPF_JITTER
1833 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
1834 		if (bf != NULL)
1835 			slen = (*(bf->func))(pkt, pktlen, pktlen);
1836 		else
1837 #endif
1838 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1839 		if (slen != 0) {
1840 			d->bd_fcount++;
1841 			if (gottime < bpf_ts_quality(d->bd_tstamp))
1842 				gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
1843 #ifdef MAC
1844 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1845 #endif
1846 				catchpacket(d, pkt, pktlen, slen,
1847 				    bpf_append_bytes, &bt);
1848 		}
1849 		BPFD_UNLOCK(d);
1850 	}
1851 	BPFIF_UNLOCK(bp);
1852 }
1853 
1854 #define	BPF_CHECK_DIRECTION(d, r, i)				\
1855 	    (((d)->bd_direction == BPF_D_IN && (r) != (i)) ||	\
1856 	    ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
1857 
1858 /*
1859  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1860  */
1861 void
1862 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1863 {
1864 	struct bintime bt;
1865 	struct bpf_d *d;
1866 #ifdef BPF_JITTER
1867 	bpf_jit_filter *bf;
1868 #endif
1869 	u_int pktlen, slen;
1870 	int gottime;
1871 
1872 	/* Skip outgoing duplicate packets. */
1873 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1874 		m->m_flags &= ~M_PROMISC;
1875 		return;
1876 	}
1877 
1878 	pktlen = m_length(m, NULL);
1879 
1880 	gottime = BPF_TSTAMP_NONE;
1881 	BPFIF_LOCK(bp);
1882 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1883 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
1884 			continue;
1885 		BPFD_LOCK(d);
1886 		++d->bd_rcount;
1887 #ifdef BPF_JITTER
1888 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
1889 		/* XXX We cannot handle multiple mbufs. */
1890 		if (bf != NULL && m->m_next == NULL)
1891 			slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
1892 		else
1893 #endif
1894 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1895 		if (slen != 0) {
1896 			d->bd_fcount++;
1897 			if (gottime < bpf_ts_quality(d->bd_tstamp))
1898 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
1899 #ifdef MAC
1900 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1901 #endif
1902 				catchpacket(d, (u_char *)m, pktlen, slen,
1903 				    bpf_append_mbuf, &bt);
1904 		}
1905 		BPFD_UNLOCK(d);
1906 	}
1907 	BPFIF_UNLOCK(bp);
1908 }
1909 
1910 /*
1911  * Incoming linkage from device drivers, when packet is in
1912  * an mbuf chain and to be prepended by a contiguous header.
1913  */
1914 void
1915 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1916 {
1917 	struct bintime bt;
1918 	struct mbuf mb;
1919 	struct bpf_d *d;
1920 	u_int pktlen, slen;
1921 	int gottime;
1922 
1923 	/* Skip outgoing duplicate packets. */
1924 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1925 		m->m_flags &= ~M_PROMISC;
1926 		return;
1927 	}
1928 
1929 	pktlen = m_length(m, NULL);
1930 	/*
1931 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1932 	 * Note that we cut corners here; we only setup what's
1933 	 * absolutely needed--this mbuf should never go anywhere else.
1934 	 */
1935 	mb.m_next = m;
1936 	mb.m_data = data;
1937 	mb.m_len = dlen;
1938 	pktlen += dlen;
1939 
1940 	gottime = BPF_TSTAMP_NONE;
1941 	BPFIF_LOCK(bp);
1942 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1943 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
1944 			continue;
1945 		BPFD_LOCK(d);
1946 		++d->bd_rcount;
1947 		slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
1948 		if (slen != 0) {
1949 			d->bd_fcount++;
1950 			if (gottime < bpf_ts_quality(d->bd_tstamp))
1951 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
1952 #ifdef MAC
1953 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1954 #endif
1955 				catchpacket(d, (u_char *)&mb, pktlen, slen,
1956 				    bpf_append_mbuf, &bt);
1957 		}
1958 		BPFD_UNLOCK(d);
1959 	}
1960 	BPFIF_UNLOCK(bp);
1961 }
1962 
1963 #undef	BPF_CHECK_DIRECTION
1964 
1965 #undef	BPF_TSTAMP_NONE
1966 #undef	BPF_TSTAMP_FAST
1967 #undef	BPF_TSTAMP_NORMAL
1968 #undef	BPF_TSTAMP_EXTERN
1969 
1970 static int
1971 bpf_hdrlen(struct bpf_d *d)
1972 {
1973 	int hdrlen;
1974 
1975 	hdrlen = d->bd_bif->bif_hdrlen;
1976 #ifndef BURN_BRIDGES
1977 	if (d->bd_tstamp == BPF_T_NONE ||
1978 	    BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
1979 #ifdef COMPAT_FREEBSD32
1980 		if (d->bd_compat32)
1981 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
1982 		else
1983 #endif
1984 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
1985 	else
1986 #endif
1987 		hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
1988 #ifdef COMPAT_FREEBSD32
1989 	if (d->bd_compat32)
1990 		hdrlen = BPF_WORDALIGN32(hdrlen);
1991 	else
1992 #endif
1993 		hdrlen = BPF_WORDALIGN(hdrlen);
1994 
1995 	return (hdrlen - d->bd_bif->bif_hdrlen);
1996 }
1997 
1998 static void
1999 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2000 {
2001 	struct bintime bt2;
2002 	struct timeval tsm;
2003 	struct timespec tsn;
2004 
2005 	if ((tstype & BPF_T_MONOTONIC) == 0) {
2006 		bt2 = *bt;
2007 		bintime_add(&bt2, &boottimebin);
2008 		bt = &bt2;
2009 	}
2010 	switch (BPF_T_FORMAT(tstype)) {
2011 	case BPF_T_MICROTIME:
2012 		bintime2timeval(bt, &tsm);
2013 		ts->bt_sec = tsm.tv_sec;
2014 		ts->bt_frac = tsm.tv_usec;
2015 		break;
2016 	case BPF_T_NANOTIME:
2017 		bintime2timespec(bt, &tsn);
2018 		ts->bt_sec = tsn.tv_sec;
2019 		ts->bt_frac = tsn.tv_nsec;
2020 		break;
2021 	case BPF_T_BINTIME:
2022 		ts->bt_sec = bt->sec;
2023 		ts->bt_frac = bt->frac;
2024 		break;
2025 	}
2026 }
2027 
2028 /*
2029  * Move the packet data from interface memory (pkt) into the
2030  * store buffer.  "cpfn" is the routine called to do the actual data
2031  * transfer.  bcopy is passed in to copy contiguous chunks, while
2032  * bpf_append_mbuf is passed in to copy mbuf chains.  In the latter case,
2033  * pkt is really an mbuf.
2034  */
2035 static void
2036 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2037     void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2038     struct bintime *bt)
2039 {
2040 	struct bpf_xhdr hdr;
2041 #ifndef BURN_BRIDGES
2042 	struct bpf_hdr hdr_old;
2043 #ifdef COMPAT_FREEBSD32
2044 	struct bpf_hdr32 hdr32_old;
2045 #endif
2046 #endif
2047 	int caplen, curlen, hdrlen, totlen;
2048 	int do_wakeup = 0;
2049 	int do_timestamp;
2050 	int tstype;
2051 
2052 	BPFD_LOCK_ASSERT(d);
2053 
2054 	/*
2055 	 * Detect whether user space has released a buffer back to us, and if
2056 	 * so, move it from being a hold buffer to a free buffer.  This may
2057 	 * not be the best place to do it (for example, we might only want to
2058 	 * run this check if we need the space), but for now it's a reliable
2059 	 * spot to do it.
2060 	 */
2061 	if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2062 		d->bd_fbuf = d->bd_hbuf;
2063 		d->bd_hbuf = NULL;
2064 		d->bd_hlen = 0;
2065 		bpf_buf_reclaimed(d);
2066 	}
2067 
2068 	/*
2069 	 * Figure out how many bytes to move.  If the packet is
2070 	 * greater or equal to the snapshot length, transfer that
2071 	 * much.  Otherwise, transfer the whole packet (unless
2072 	 * we hit the buffer size limit).
2073 	 */
2074 	hdrlen = bpf_hdrlen(d);
2075 	totlen = hdrlen + min(snaplen, pktlen);
2076 	if (totlen > d->bd_bufsize)
2077 		totlen = d->bd_bufsize;
2078 
2079 	/*
2080 	 * Round up the end of the previous packet to the next longword.
2081 	 *
2082 	 * Drop the packet if there's no room and no hope of room
2083 	 * If the packet would overflow the storage buffer or the storage
2084 	 * buffer is considered immutable by the buffer model, try to rotate
2085 	 * the buffer and wakeup pending processes.
2086 	 */
2087 #ifdef COMPAT_FREEBSD32
2088 	if (d->bd_compat32)
2089 		curlen = BPF_WORDALIGN32(d->bd_slen);
2090 	else
2091 #endif
2092 		curlen = BPF_WORDALIGN(d->bd_slen);
2093 	if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2094 		if (d->bd_fbuf == NULL) {
2095 			/*
2096 			 * There's no room in the store buffer, and no
2097 			 * prospect of room, so drop the packet.  Notify the
2098 			 * buffer model.
2099 			 */
2100 			bpf_buffull(d);
2101 			++d->bd_dcount;
2102 			return;
2103 		}
2104 		ROTATE_BUFFERS(d);
2105 		do_wakeup = 1;
2106 		curlen = 0;
2107 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2108 		/*
2109 		 * Immediate mode is set, or the read timeout has already
2110 		 * expired during a select call.  A packet arrived, so the
2111 		 * reader should be woken up.
2112 		 */
2113 		do_wakeup = 1;
2114 	caplen = totlen - hdrlen;
2115 	tstype = d->bd_tstamp;
2116 	do_timestamp = tstype != BPF_T_NONE;
2117 #ifndef BURN_BRIDGES
2118 	if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2119 		struct bpf_ts ts;
2120 		if (do_timestamp)
2121 			bpf_bintime2ts(bt, &ts, tstype);
2122 #ifdef COMPAT_FREEBSD32
2123 		if (d->bd_compat32) {
2124 			bzero(&hdr32_old, sizeof(hdr32_old));
2125 			if (do_timestamp) {
2126 				hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2127 				hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2128 			}
2129 			hdr32_old.bh_datalen = pktlen;
2130 			hdr32_old.bh_hdrlen = hdrlen;
2131 			hdr32_old.bh_caplen = caplen;
2132 			bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2133 			    sizeof(hdr32_old));
2134 			goto copy;
2135 		}
2136 #endif
2137 		bzero(&hdr_old, sizeof(hdr_old));
2138 		if (do_timestamp) {
2139 			hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2140 			hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2141 		}
2142 		hdr_old.bh_datalen = pktlen;
2143 		hdr_old.bh_hdrlen = hdrlen;
2144 		hdr_old.bh_caplen = caplen;
2145 		bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2146 		    sizeof(hdr_old));
2147 		goto copy;
2148 	}
2149 #endif
2150 
2151 	/*
2152 	 * Append the bpf header.  Note we append the actual header size, but
2153 	 * move forward the length of the header plus padding.
2154 	 */
2155 	bzero(&hdr, sizeof(hdr));
2156 	if (do_timestamp)
2157 		bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2158 	hdr.bh_datalen = pktlen;
2159 	hdr.bh_hdrlen = hdrlen;
2160 	hdr.bh_caplen = caplen;
2161 	bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2162 
2163 	/*
2164 	 * Copy the packet data into the store buffer and update its length.
2165 	 */
2166 #ifndef BURN_BRIDGES
2167 copy:
2168 #endif
2169 	(*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2170 	d->bd_slen = curlen + totlen;
2171 
2172 	if (do_wakeup)
2173 		bpf_wakeup(d);
2174 }
2175 
2176 /*
2177  * Free buffers currently in use by a descriptor.
2178  * Called on close.
2179  */
2180 static void
2181 bpf_freed(struct bpf_d *d)
2182 {
2183 
2184 	/*
2185 	 * We don't need to lock out interrupts since this descriptor has
2186 	 * been detached from its interface and it yet hasn't been marked
2187 	 * free.
2188 	 */
2189 	bpf_free(d);
2190 	if (d->bd_rfilter != NULL) {
2191 		free((caddr_t)d->bd_rfilter, M_BPF);
2192 #ifdef BPF_JITTER
2193 		if (d->bd_bfilter != NULL)
2194 			bpf_destroy_jit_filter(d->bd_bfilter);
2195 #endif
2196 	}
2197 	if (d->bd_wfilter != NULL)
2198 		free((caddr_t)d->bd_wfilter, M_BPF);
2199 	mtx_destroy(&d->bd_mtx);
2200 }
2201 
2202 /*
2203  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
2204  * fixed size of the link header (variable length headers not yet supported).
2205  */
2206 void
2207 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2208 {
2209 
2210 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2211 }
2212 
2213 /*
2214  * Attach an interface to bpf.  ifp is a pointer to the structure
2215  * defining the interface to be attached, dlt is the link layer type,
2216  * and hdrlen is the fixed size of the link header (variable length
2217  * headers are not yet supporrted).
2218  */
2219 void
2220 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2221 {
2222 	struct bpf_if *bp;
2223 
2224 	bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2225 	if (bp == NULL)
2226 		panic("bpfattach");
2227 
2228 	LIST_INIT(&bp->bif_dlist);
2229 	bp->bif_ifp = ifp;
2230 	bp->bif_dlt = dlt;
2231 	mtx_init(&bp->bif_mtx, "bpf interface lock", NULL, MTX_DEF);
2232 	KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2233 	*driverp = bp;
2234 
2235 	mtx_lock(&bpf_mtx);
2236 	LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2237 	mtx_unlock(&bpf_mtx);
2238 
2239 	bp->bif_hdrlen = hdrlen;
2240 
2241 	if (bootverbose)
2242 		if_printf(ifp, "bpf attached\n");
2243 }
2244 
2245 /*
2246  * Detach bpf from an interface.  This involves detaching each descriptor
2247  * associated with the interface, and leaving bd_bif NULL.  Notify each
2248  * descriptor as it's detached so that any sleepers wake up and get
2249  * ENXIO.
2250  */
2251 void
2252 bpfdetach(struct ifnet *ifp)
2253 {
2254 	struct bpf_if	*bp;
2255 	struct bpf_d	*d;
2256 
2257 	/* Locate BPF interface information */
2258 	mtx_lock(&bpf_mtx);
2259 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2260 		if (ifp == bp->bif_ifp)
2261 			break;
2262 	}
2263 
2264 	/* Interface wasn't attached */
2265 	if ((bp == NULL) || (bp->bif_ifp == NULL)) {
2266 		mtx_unlock(&bpf_mtx);
2267 		printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2268 		return;
2269 	}
2270 
2271 	LIST_REMOVE(bp, bif_next);
2272 	mtx_unlock(&bpf_mtx);
2273 
2274 	while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2275 		bpf_detachd(d);
2276 		BPFD_LOCK(d);
2277 		bpf_wakeup(d);
2278 		BPFD_UNLOCK(d);
2279 	}
2280 
2281 	mtx_destroy(&bp->bif_mtx);
2282 	free(bp, M_BPF);
2283 }
2284 
2285 /*
2286  * Get a list of available data link type of the interface.
2287  */
2288 static int
2289 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2290 {
2291 	int n, error;
2292 	struct ifnet *ifp;
2293 	struct bpf_if *bp;
2294 
2295 	ifp = d->bd_bif->bif_ifp;
2296 	n = 0;
2297 	error = 0;
2298 	mtx_lock(&bpf_mtx);
2299 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2300 		if (bp->bif_ifp != ifp)
2301 			continue;
2302 		if (bfl->bfl_list != NULL) {
2303 			if (n >= bfl->bfl_len) {
2304 				mtx_unlock(&bpf_mtx);
2305 				return (ENOMEM);
2306 			}
2307 			error = copyout(&bp->bif_dlt,
2308 			    bfl->bfl_list + n, sizeof(u_int));
2309 		}
2310 		n++;
2311 	}
2312 	mtx_unlock(&bpf_mtx);
2313 	bfl->bfl_len = n;
2314 	return (error);
2315 }
2316 
2317 /*
2318  * Set the data link type of a BPF instance.
2319  */
2320 static int
2321 bpf_setdlt(struct bpf_d *d, u_int dlt)
2322 {
2323 	int error, opromisc;
2324 	struct ifnet *ifp;
2325 	struct bpf_if *bp;
2326 
2327 	if (d->bd_bif->bif_dlt == dlt)
2328 		return (0);
2329 	ifp = d->bd_bif->bif_ifp;
2330 	mtx_lock(&bpf_mtx);
2331 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2332 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2333 			break;
2334 	}
2335 	mtx_unlock(&bpf_mtx);
2336 	if (bp != NULL) {
2337 		opromisc = d->bd_promisc;
2338 		bpf_detachd(d);
2339 		bpf_attachd(d, bp);
2340 		BPFD_LOCK(d);
2341 		reset_d(d);
2342 		BPFD_UNLOCK(d);
2343 		if (opromisc) {
2344 			error = ifpromisc(bp->bif_ifp, 1);
2345 			if (error)
2346 				if_printf(bp->bif_ifp,
2347 					"bpf_setdlt: ifpromisc failed (%d)\n",
2348 					error);
2349 			else
2350 				d->bd_promisc = 1;
2351 		}
2352 	}
2353 	return (bp == NULL ? EINVAL : 0);
2354 }
2355 
2356 static void
2357 bpf_drvinit(void *unused)
2358 {
2359 	struct cdev *dev;
2360 
2361 	mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2362 	LIST_INIT(&bpf_iflist);
2363 
2364 	dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2365 	/* For compatibility */
2366 	make_dev_alias(dev, "bpf0");
2367 }
2368 
2369 /*
2370  * Zero out the various packet counters associated with all of the bpf
2371  * descriptors.  At some point, we will probably want to get a bit more
2372  * granular and allow the user to specify descriptors to be zeroed.
2373  */
2374 static void
2375 bpf_zero_counters(void)
2376 {
2377 	struct bpf_if *bp;
2378 	struct bpf_d *bd;
2379 
2380 	mtx_lock(&bpf_mtx);
2381 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2382 		BPFIF_LOCK(bp);
2383 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2384 			BPFD_LOCK(bd);
2385 			bd->bd_rcount = 0;
2386 			bd->bd_dcount = 0;
2387 			bd->bd_fcount = 0;
2388 			bd->bd_wcount = 0;
2389 			bd->bd_wfcount = 0;
2390 			bd->bd_zcopy = 0;
2391 			BPFD_UNLOCK(bd);
2392 		}
2393 		BPFIF_UNLOCK(bp);
2394 	}
2395 	mtx_unlock(&bpf_mtx);
2396 }
2397 
2398 static void
2399 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2400 {
2401 
2402 	bzero(d, sizeof(*d));
2403 	BPFD_LOCK_ASSERT(bd);
2404 	d->bd_structsize = sizeof(*d);
2405 	d->bd_immediate = bd->bd_immediate;
2406 	d->bd_promisc = bd->bd_promisc;
2407 	d->bd_hdrcmplt = bd->bd_hdrcmplt;
2408 	d->bd_direction = bd->bd_direction;
2409 	d->bd_feedback = bd->bd_feedback;
2410 	d->bd_async = bd->bd_async;
2411 	d->bd_rcount = bd->bd_rcount;
2412 	d->bd_dcount = bd->bd_dcount;
2413 	d->bd_fcount = bd->bd_fcount;
2414 	d->bd_sig = bd->bd_sig;
2415 	d->bd_slen = bd->bd_slen;
2416 	d->bd_hlen = bd->bd_hlen;
2417 	d->bd_bufsize = bd->bd_bufsize;
2418 	d->bd_pid = bd->bd_pid;
2419 	strlcpy(d->bd_ifname,
2420 	    bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2421 	d->bd_locked = bd->bd_locked;
2422 	d->bd_wcount = bd->bd_wcount;
2423 	d->bd_wdcount = bd->bd_wdcount;
2424 	d->bd_wfcount = bd->bd_wfcount;
2425 	d->bd_zcopy = bd->bd_zcopy;
2426 	d->bd_bufmode = bd->bd_bufmode;
2427 }
2428 
2429 static int
2430 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2431 {
2432 	struct xbpf_d *xbdbuf, *xbd, zerostats;
2433 	int index, error;
2434 	struct bpf_if *bp;
2435 	struct bpf_d *bd;
2436 
2437 	/*
2438 	 * XXX This is not technically correct. It is possible for non
2439 	 * privileged users to open bpf devices. It would make sense
2440 	 * if the users who opened the devices were able to retrieve
2441 	 * the statistics for them, too.
2442 	 */
2443 	error = priv_check(req->td, PRIV_NET_BPF);
2444 	if (error)
2445 		return (error);
2446 	/*
2447 	 * Check to see if the user is requesting that the counters be
2448 	 * zeroed out.  Explicitly check that the supplied data is zeroed,
2449 	 * as we aren't allowing the user to set the counters currently.
2450 	 */
2451 	if (req->newptr != NULL) {
2452 		if (req->newlen != sizeof(zerostats))
2453 			return (EINVAL);
2454 		bzero(&zerostats, sizeof(zerostats));
2455 		xbd = req->newptr;
2456 		if (bcmp(xbd, &zerostats, sizeof(*xbd)) != 0)
2457 			return (EINVAL);
2458 		bpf_zero_counters();
2459 		return (0);
2460 	}
2461 	if (req->oldptr == NULL)
2462 		return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2463 	if (bpf_bpfd_cnt == 0)
2464 		return (SYSCTL_OUT(req, 0, 0));
2465 	xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2466 	mtx_lock(&bpf_mtx);
2467 	if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2468 		mtx_unlock(&bpf_mtx);
2469 		free(xbdbuf, M_BPF);
2470 		return (ENOMEM);
2471 	}
2472 	index = 0;
2473 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2474 		BPFIF_LOCK(bp);
2475 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2476 			xbd = &xbdbuf[index++];
2477 			BPFD_LOCK(bd);
2478 			bpfstats_fill_xbpf(xbd, bd);
2479 			BPFD_UNLOCK(bd);
2480 		}
2481 		BPFIF_UNLOCK(bp);
2482 	}
2483 	mtx_unlock(&bpf_mtx);
2484 	error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2485 	free(xbdbuf, M_BPF);
2486 	return (error);
2487 }
2488 
2489 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2490 
2491 #else /* !DEV_BPF && !NETGRAPH_BPF */
2492 /*
2493  * NOP stubs to allow bpf-using drivers to load and function.
2494  *
2495  * A 'better' implementation would allow the core bpf functionality
2496  * to be loaded at runtime.
2497  */
2498 static struct bpf_if bp_null;
2499 
2500 void
2501 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2502 {
2503 }
2504 
2505 void
2506 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2507 {
2508 }
2509 
2510 void
2511 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2512 {
2513 }
2514 
2515 void
2516 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2517 {
2518 
2519 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2520 }
2521 
2522 void
2523 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2524 {
2525 
2526 	*driverp = &bp_null;
2527 }
2528 
2529 void
2530 bpfdetach(struct ifnet *ifp)
2531 {
2532 }
2533 
2534 u_int
2535 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2536 {
2537 	return -1;	/* "no filter" behaviour */
2538 }
2539 
2540 int
2541 bpf_validate(const struct bpf_insn *f, int len)
2542 {
2543 	return 0;		/* false */
2544 }
2545 
2546 #endif /* !DEV_BPF && !NETGRAPH_BPF */
2547