xref: /freebsd/sys/net/bpf.c (revision aa64588d28258aef88cc33b8043112e8856948d0)
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 #ifdef COMPAT_FREEBSD32
94 #include <sys/mount.h>
95 #include <compat/freebsd32/freebsd32.h>
96 #define BPF_ALIGNMENT32 sizeof(int32_t)
97 #define BPF_WORDALIGN32(x) (((x)+(BPF_ALIGNMENT32-1))&~(BPF_ALIGNMENT32-1))
98 
99 /*
100  * 32-bit version of structure prepended to each packet.  We use this header
101  * instead of the standard one for 32-bit streams.  We mark the a stream as
102  * 32-bit the first time we see a 32-bit compat ioctl request.
103  */
104 struct bpf_hdr32 {
105 	struct timeval32 bh_tstamp;	/* time stamp */
106 	uint32_t	bh_caplen;	/* length of captured portion */
107 	uint32_t	bh_datalen;	/* original length of packet */
108 	uint16_t	bh_hdrlen;	/* length of bpf header (this struct
109 					   plus alignment padding) */
110 };
111 
112 struct bpf_program32 {
113 	u_int bf_len;
114 	uint32_t bf_insns;
115 };
116 
117 struct bpf_dltlist32 {
118 	u_int	bfl_len;
119 	u_int	bfl_list;
120 };
121 
122 #define	BIOCSETF32	_IOW('B', 103, struct bpf_program32)
123 #define	BIOCSRTIMEOUT32	_IOW('B',109, struct timeval32)
124 #define	BIOCGRTIMEOUT32	_IOR('B',110, struct timeval32)
125 #define	BIOCGDLTLIST32	_IOWR('B',121, struct bpf_dltlist32)
126 #define	BIOCSETWF32	_IOW('B',123, struct bpf_program32)
127 #define	BIOCSETFNR32	_IOW('B',130, struct bpf_program32)
128 #endif
129 
130 /*
131  * bpf_iflist is a list of BPF interface structures, each corresponding to a
132  * specific DLT.  The same network interface might have several BPF interface
133  * structures registered by different layers in the stack (i.e., 802.11
134  * frames, ethernet frames, etc).
135  */
136 static LIST_HEAD(, bpf_if)	bpf_iflist;
137 static struct mtx	bpf_mtx;		/* bpf global lock */
138 static int		bpf_bpfd_cnt;
139 
140 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
141 static void	bpf_detachd(struct bpf_d *);
142 static void	bpf_freed(struct bpf_d *);
143 static int	bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
144 		    struct sockaddr *, int *, struct bpf_insn *);
145 static int	bpf_setif(struct bpf_d *, struct ifreq *);
146 static void	bpf_timed_out(void *);
147 static __inline void
148 		bpf_wakeup(struct bpf_d *);
149 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
150 		    void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
151 		    struct timeval *);
152 static void	reset_d(struct bpf_d *);
153 static int	 bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
154 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
155 static int	bpf_setdlt(struct bpf_d *, u_int);
156 static void	filt_bpfdetach(struct knote *);
157 static int	filt_bpfread(struct knote *, long);
158 static void	bpf_drvinit(void *);
159 static int	bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
160 
161 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
162 int bpf_maxinsns = BPF_MAXINSNS;
163 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
164     &bpf_maxinsns, 0, "Maximum bpf program instructions");
165 static int bpf_zerocopy_enable = 0;
166 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
167     &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
168 SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
169     bpf_stats_sysctl, "bpf statistics portal");
170 
171 static	d_open_t	bpfopen;
172 static	d_read_t	bpfread;
173 static	d_write_t	bpfwrite;
174 static	d_ioctl_t	bpfioctl;
175 static	d_poll_t	bpfpoll;
176 static	d_kqfilter_t	bpfkqfilter;
177 
178 static struct cdevsw bpf_cdevsw = {
179 	.d_version =	D_VERSION,
180 	.d_open =	bpfopen,
181 	.d_read =	bpfread,
182 	.d_write =	bpfwrite,
183 	.d_ioctl =	bpfioctl,
184 	.d_poll =	bpfpoll,
185 	.d_name =	"bpf",
186 	.d_kqfilter =	bpfkqfilter,
187 };
188 
189 static struct filterops bpfread_filtops = {
190 	.f_isfd = 1,
191 	.f_detach = filt_bpfdetach,
192 	.f_event = filt_bpfread,
193 };
194 
195 /*
196  * Wrapper functions for various buffering methods.  If the set of buffer
197  * modes expands, we will probably want to introduce a switch data structure
198  * similar to protosw, et.
199  */
200 static void
201 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
202     u_int len)
203 {
204 
205 	BPFD_LOCK_ASSERT(d);
206 
207 	switch (d->bd_bufmode) {
208 	case BPF_BUFMODE_BUFFER:
209 		return (bpf_buffer_append_bytes(d, buf, offset, src, len));
210 
211 	case BPF_BUFMODE_ZBUF:
212 		d->bd_zcopy++;
213 		return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
214 
215 	default:
216 		panic("bpf_buf_append_bytes");
217 	}
218 }
219 
220 static void
221 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
222     u_int len)
223 {
224 
225 	BPFD_LOCK_ASSERT(d);
226 
227 	switch (d->bd_bufmode) {
228 	case BPF_BUFMODE_BUFFER:
229 		return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
230 
231 	case BPF_BUFMODE_ZBUF:
232 		d->bd_zcopy++;
233 		return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
234 
235 	default:
236 		panic("bpf_buf_append_mbuf");
237 	}
238 }
239 
240 /*
241  * This function gets called when the free buffer is re-assigned.
242  */
243 static void
244 bpf_buf_reclaimed(struct bpf_d *d)
245 {
246 
247 	BPFD_LOCK_ASSERT(d);
248 
249 	switch (d->bd_bufmode) {
250 	case BPF_BUFMODE_BUFFER:
251 		return;
252 
253 	case BPF_BUFMODE_ZBUF:
254 		bpf_zerocopy_buf_reclaimed(d);
255 		return;
256 
257 	default:
258 		panic("bpf_buf_reclaimed");
259 	}
260 }
261 
262 /*
263  * If the buffer mechanism has a way to decide that a held buffer can be made
264  * free, then it is exposed via the bpf_canfreebuf() interface.  (1) is
265  * returned if the buffer can be discarded, (0) is returned if it cannot.
266  */
267 static int
268 bpf_canfreebuf(struct bpf_d *d)
269 {
270 
271 	BPFD_LOCK_ASSERT(d);
272 
273 	switch (d->bd_bufmode) {
274 	case BPF_BUFMODE_ZBUF:
275 		return (bpf_zerocopy_canfreebuf(d));
276 	}
277 	return (0);
278 }
279 
280 /*
281  * Allow the buffer model to indicate that the current store buffer is
282  * immutable, regardless of the appearance of space.  Return (1) if the
283  * buffer is writable, and (0) if not.
284  */
285 static int
286 bpf_canwritebuf(struct bpf_d *d)
287 {
288 
289 	BPFD_LOCK_ASSERT(d);
290 
291 	switch (d->bd_bufmode) {
292 	case BPF_BUFMODE_ZBUF:
293 		return (bpf_zerocopy_canwritebuf(d));
294 	}
295 	return (1);
296 }
297 
298 /*
299  * Notify buffer model that an attempt to write to the store buffer has
300  * resulted in a dropped packet, in which case the buffer may be considered
301  * full.
302  */
303 static void
304 bpf_buffull(struct bpf_d *d)
305 {
306 
307 	BPFD_LOCK_ASSERT(d);
308 
309 	switch (d->bd_bufmode) {
310 	case BPF_BUFMODE_ZBUF:
311 		bpf_zerocopy_buffull(d);
312 		break;
313 	}
314 }
315 
316 /*
317  * Notify the buffer model that a buffer has moved into the hold position.
318  */
319 void
320 bpf_bufheld(struct bpf_d *d)
321 {
322 
323 	BPFD_LOCK_ASSERT(d);
324 
325 	switch (d->bd_bufmode) {
326 	case BPF_BUFMODE_ZBUF:
327 		bpf_zerocopy_bufheld(d);
328 		break;
329 	}
330 }
331 
332 static void
333 bpf_free(struct bpf_d *d)
334 {
335 
336 	switch (d->bd_bufmode) {
337 	case BPF_BUFMODE_BUFFER:
338 		return (bpf_buffer_free(d));
339 
340 	case BPF_BUFMODE_ZBUF:
341 		return (bpf_zerocopy_free(d));
342 
343 	default:
344 		panic("bpf_buf_free");
345 	}
346 }
347 
348 static int
349 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
350 {
351 
352 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
353 		return (EOPNOTSUPP);
354 	return (bpf_buffer_uiomove(d, buf, len, uio));
355 }
356 
357 static int
358 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
359 {
360 
361 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
362 		return (EOPNOTSUPP);
363 	return (bpf_buffer_ioctl_sblen(d, i));
364 }
365 
366 static int
367 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
368 {
369 
370 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
371 		return (EOPNOTSUPP);
372 	return (bpf_zerocopy_ioctl_getzmax(td, d, i));
373 }
374 
375 static int
376 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
377 {
378 
379 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
380 		return (EOPNOTSUPP);
381 	return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
382 }
383 
384 static int
385 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
386 {
387 
388 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
389 		return (EOPNOTSUPP);
390 	return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
391 }
392 
393 /*
394  * General BPF functions.
395  */
396 static int
397 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
398     struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter)
399 {
400 	const struct ieee80211_bpf_params *p;
401 	struct ether_header *eh;
402 	struct mbuf *m;
403 	int error;
404 	int len;
405 	int hlen;
406 	int slen;
407 
408 	/*
409 	 * Build a sockaddr based on the data link layer type.
410 	 * We do this at this level because the ethernet header
411 	 * is copied directly into the data field of the sockaddr.
412 	 * In the case of SLIP, there is no header and the packet
413 	 * is forwarded as is.
414 	 * Also, we are careful to leave room at the front of the mbuf
415 	 * for the link level header.
416 	 */
417 	switch (linktype) {
418 
419 	case DLT_SLIP:
420 		sockp->sa_family = AF_INET;
421 		hlen = 0;
422 		break;
423 
424 	case DLT_EN10MB:
425 		sockp->sa_family = AF_UNSPEC;
426 		/* XXX Would MAXLINKHDR be better? */
427 		hlen = ETHER_HDR_LEN;
428 		break;
429 
430 	case DLT_FDDI:
431 		sockp->sa_family = AF_IMPLINK;
432 		hlen = 0;
433 		break;
434 
435 	case DLT_RAW:
436 		sockp->sa_family = AF_UNSPEC;
437 		hlen = 0;
438 		break;
439 
440 	case DLT_NULL:
441 		/*
442 		 * null interface types require a 4 byte pseudo header which
443 		 * corresponds to the address family of the packet.
444 		 */
445 		sockp->sa_family = AF_UNSPEC;
446 		hlen = 4;
447 		break;
448 
449 	case DLT_ATM_RFC1483:
450 		/*
451 		 * en atm driver requires 4-byte atm pseudo header.
452 		 * though it isn't standard, vpi:vci needs to be
453 		 * specified anyway.
454 		 */
455 		sockp->sa_family = AF_UNSPEC;
456 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
457 		break;
458 
459 	case DLT_PPP:
460 		sockp->sa_family = AF_UNSPEC;
461 		hlen = 4;	/* This should match PPP_HDRLEN */
462 		break;
463 
464 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
465 		sockp->sa_family = AF_IEEE80211;
466 		hlen = 0;
467 		break;
468 
469 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
470 		sockp->sa_family = AF_IEEE80211;
471 		sockp->sa_len = 12;	/* XXX != 0 */
472 		hlen = sizeof(struct ieee80211_bpf_params);
473 		break;
474 
475 	default:
476 		return (EIO);
477 	}
478 
479 	len = uio->uio_resid;
480 
481 	if (len - hlen > ifp->if_mtu)
482 		return (EMSGSIZE);
483 
484 	if ((unsigned)len > MJUM16BYTES)
485 		return (EIO);
486 
487 	if (len <= MHLEN)
488 		MGETHDR(m, M_WAIT, MT_DATA);
489 	else if (len <= MCLBYTES)
490 		m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
491 	else
492 		m = m_getjcl(M_WAIT, MT_DATA, M_PKTHDR,
493 #if (MJUMPAGESIZE > MCLBYTES)
494 		    len <= MJUMPAGESIZE ? MJUMPAGESIZE :
495 #endif
496 		    (len <= MJUM9BYTES ? MJUM9BYTES : MJUM16BYTES));
497 	m->m_pkthdr.len = m->m_len = len;
498 	m->m_pkthdr.rcvif = NULL;
499 	*mp = m;
500 
501 	if (m->m_len < hlen) {
502 		error = EPERM;
503 		goto bad;
504 	}
505 
506 	error = uiomove(mtod(m, u_char *), len, uio);
507 	if (error)
508 		goto bad;
509 
510 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
511 	if (slen == 0) {
512 		error = EPERM;
513 		goto bad;
514 	}
515 
516 	/* Check for multicast destination */
517 	switch (linktype) {
518 	case DLT_EN10MB:
519 		eh = mtod(m, struct ether_header *);
520 		if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
521 			if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
522 			    ETHER_ADDR_LEN) == 0)
523 				m->m_flags |= M_BCAST;
524 			else
525 				m->m_flags |= M_MCAST;
526 		}
527 		break;
528 	}
529 
530 	/*
531 	 * Make room for link header, and copy it to sockaddr
532 	 */
533 	if (hlen != 0) {
534 		if (sockp->sa_family == AF_IEEE80211) {
535 			/*
536 			 * Collect true length from the parameter header
537 			 * NB: sockp is known to be zero'd so if we do a
538 			 *     short copy unspecified parameters will be
539 			 *     zero.
540 			 * NB: packet may not be aligned after stripping
541 			 *     bpf params
542 			 * XXX check ibp_vers
543 			 */
544 			p = mtod(m, const struct ieee80211_bpf_params *);
545 			hlen = p->ibp_len;
546 			if (hlen > sizeof(sockp->sa_data)) {
547 				error = EINVAL;
548 				goto bad;
549 			}
550 		}
551 		bcopy(m->m_data, sockp->sa_data, hlen);
552 	}
553 	*hdrlen = hlen;
554 
555 	return (0);
556 bad:
557 	m_freem(m);
558 	return (error);
559 }
560 
561 /*
562  * Attach file to the bpf interface, i.e. make d listen on bp.
563  */
564 static void
565 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
566 {
567 	/*
568 	 * Point d at bp, and add d to the interface's list of listeners.
569 	 * Finally, point the driver's bpf cookie at the interface so
570 	 * it will divert packets to bpf.
571 	 */
572 	BPFIF_LOCK(bp);
573 	d->bd_bif = bp;
574 	LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
575 
576 	bpf_bpfd_cnt++;
577 	BPFIF_UNLOCK(bp);
578 
579 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
580 }
581 
582 /*
583  * Detach a file from its interface.
584  */
585 static void
586 bpf_detachd(struct bpf_d *d)
587 {
588 	int error;
589 	struct bpf_if *bp;
590 	struct ifnet *ifp;
591 
592 	bp = d->bd_bif;
593 	BPFIF_LOCK(bp);
594 	BPFD_LOCK(d);
595 	ifp = d->bd_bif->bif_ifp;
596 
597 	/*
598 	 * Remove d from the interface's descriptor list.
599 	 */
600 	LIST_REMOVE(d, bd_next);
601 
602 	bpf_bpfd_cnt--;
603 	d->bd_bif = NULL;
604 	BPFD_UNLOCK(d);
605 	BPFIF_UNLOCK(bp);
606 
607 	EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
608 
609 	/*
610 	 * Check if this descriptor had requested promiscuous mode.
611 	 * If so, turn it off.
612 	 */
613 	if (d->bd_promisc) {
614 		d->bd_promisc = 0;
615 		CURVNET_SET(ifp->if_vnet);
616 		error = ifpromisc(ifp, 0);
617 		CURVNET_RESTORE();
618 		if (error != 0 && error != ENXIO) {
619 			/*
620 			 * ENXIO can happen if a pccard is unplugged
621 			 * Something is really wrong if we were able to put
622 			 * the driver into promiscuous mode, but can't
623 			 * take it out.
624 			 */
625 			if_printf(bp->bif_ifp,
626 				"bpf_detach: ifpromisc failed (%d)\n", error);
627 		}
628 	}
629 }
630 
631 /*
632  * Close the descriptor by detaching it from its interface,
633  * deallocating its buffers, and marking it free.
634  */
635 static void
636 bpf_dtor(void *data)
637 {
638 	struct bpf_d *d = data;
639 
640 	BPFD_LOCK(d);
641 	if (d->bd_state == BPF_WAITING)
642 		callout_stop(&d->bd_callout);
643 	d->bd_state = BPF_IDLE;
644 	BPFD_UNLOCK(d);
645 	funsetown(&d->bd_sigio);
646 	mtx_lock(&bpf_mtx);
647 	if (d->bd_bif)
648 		bpf_detachd(d);
649 	mtx_unlock(&bpf_mtx);
650 	selwakeuppri(&d->bd_sel, PRINET);
651 #ifdef MAC
652 	mac_bpfdesc_destroy(d);
653 #endif /* MAC */
654 	knlist_destroy(&d->bd_sel.si_note);
655 	callout_drain(&d->bd_callout);
656 	bpf_freed(d);
657 	free(d, M_BPF);
658 }
659 
660 /*
661  * Open ethernet device.  Returns ENXIO for illegal minor device number,
662  * EBUSY if file is open by another process.
663  */
664 /* ARGSUSED */
665 static	int
666 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
667 {
668 	struct bpf_d *d;
669 	int error;
670 
671 	d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
672 	error = devfs_set_cdevpriv(d, bpf_dtor);
673 	if (error != 0) {
674 		free(d, M_BPF);
675 		return (error);
676 	}
677 
678 	/*
679 	 * For historical reasons, perform a one-time initialization call to
680 	 * the buffer routines, even though we're not yet committed to a
681 	 * particular buffer method.
682 	 */
683 	bpf_buffer_init(d);
684 	d->bd_bufmode = BPF_BUFMODE_BUFFER;
685 	d->bd_sig = SIGIO;
686 	d->bd_direction = BPF_D_INOUT;
687 	d->bd_pid = td->td_proc->p_pid;
688 #ifdef MAC
689 	mac_bpfdesc_init(d);
690 	mac_bpfdesc_create(td->td_ucred, d);
691 #endif
692 	mtx_init(&d->bd_mtx, devtoname(dev), "bpf cdev lock", MTX_DEF);
693 	callout_init_mtx(&d->bd_callout, &d->bd_mtx, 0);
694 	knlist_init_mtx(&d->bd_sel.si_note, &d->bd_mtx);
695 
696 	return (0);
697 }
698 
699 /*
700  *  bpfread - read next chunk of packets from buffers
701  */
702 static	int
703 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
704 {
705 	struct bpf_d *d;
706 	int error;
707 	int non_block;
708 	int timed_out;
709 
710 	error = devfs_get_cdevpriv((void **)&d);
711 	if (error != 0)
712 		return (error);
713 
714 	/*
715 	 * Restrict application to use a buffer the same size as
716 	 * as kernel buffers.
717 	 */
718 	if (uio->uio_resid != d->bd_bufsize)
719 		return (EINVAL);
720 
721 	non_block = ((ioflag & O_NONBLOCK) != 0);
722 
723 	BPFD_LOCK(d);
724 	d->bd_pid = curthread->td_proc->p_pid;
725 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
726 		BPFD_UNLOCK(d);
727 		return (EOPNOTSUPP);
728 	}
729 	if (d->bd_state == BPF_WAITING)
730 		callout_stop(&d->bd_callout);
731 	timed_out = (d->bd_state == BPF_TIMED_OUT);
732 	d->bd_state = BPF_IDLE;
733 	/*
734 	 * If the hold buffer is empty, then do a timed sleep, which
735 	 * ends when the timeout expires or when enough packets
736 	 * have arrived to fill the store buffer.
737 	 */
738 	while (d->bd_hbuf == NULL) {
739 		if (d->bd_slen != 0) {
740 			/*
741 			 * A packet(s) either arrived since the previous
742 			 * read or arrived while we were asleep.
743 			 */
744 			if (d->bd_immediate || non_block || timed_out) {
745 				/*
746 				 * Rotate the buffers and return what's here
747 				 * if we are in immediate mode, non-blocking
748 				 * flag is set, or this descriptor timed out.
749 				 */
750 				ROTATE_BUFFERS(d);
751 				break;
752 			}
753 		}
754 
755 		/*
756 		 * No data is available, check to see if the bpf device
757 		 * is still pointed at a real interface.  If not, return
758 		 * ENXIO so that the userland process knows to rebind
759 		 * it before using it again.
760 		 */
761 		if (d->bd_bif == NULL) {
762 			BPFD_UNLOCK(d);
763 			return (ENXIO);
764 		}
765 
766 		if (non_block) {
767 			BPFD_UNLOCK(d);
768 			return (EWOULDBLOCK);
769 		}
770 		error = msleep(d, &d->bd_mtx, PRINET|PCATCH,
771 		     "bpf", d->bd_rtout);
772 		if (error == EINTR || error == ERESTART) {
773 			BPFD_UNLOCK(d);
774 			return (error);
775 		}
776 		if (error == EWOULDBLOCK) {
777 			/*
778 			 * On a timeout, return what's in the buffer,
779 			 * which may be nothing.  If there is something
780 			 * in the store buffer, we can rotate the buffers.
781 			 */
782 			if (d->bd_hbuf)
783 				/*
784 				 * We filled up the buffer in between
785 				 * getting the timeout and arriving
786 				 * here, so we don't need to rotate.
787 				 */
788 				break;
789 
790 			if (d->bd_slen == 0) {
791 				BPFD_UNLOCK(d);
792 				return (0);
793 			}
794 			ROTATE_BUFFERS(d);
795 			break;
796 		}
797 	}
798 	/*
799 	 * At this point, we know we have something in the hold slot.
800 	 */
801 	BPFD_UNLOCK(d);
802 
803 	/*
804 	 * Move data from hold buffer into user space.
805 	 * We know the entire buffer is transferred since
806 	 * we checked above that the read buffer is bpf_bufsize bytes.
807 	 *
808 	 * XXXRW: More synchronization needed here: what if a second thread
809 	 * issues a read on the same fd at the same time?  Don't want this
810 	 * getting invalidated.
811 	 */
812 	error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
813 
814 	BPFD_LOCK(d);
815 	d->bd_fbuf = d->bd_hbuf;
816 	d->bd_hbuf = NULL;
817 	d->bd_hlen = 0;
818 	bpf_buf_reclaimed(d);
819 	BPFD_UNLOCK(d);
820 
821 	return (error);
822 }
823 
824 /*
825  * If there are processes sleeping on this descriptor, wake them up.
826  */
827 static __inline void
828 bpf_wakeup(struct bpf_d *d)
829 {
830 
831 	BPFD_LOCK_ASSERT(d);
832 	if (d->bd_state == BPF_WAITING) {
833 		callout_stop(&d->bd_callout);
834 		d->bd_state = BPF_IDLE;
835 	}
836 	wakeup(d);
837 	if (d->bd_async && d->bd_sig && d->bd_sigio)
838 		pgsigio(&d->bd_sigio, d->bd_sig, 0);
839 
840 	selwakeuppri(&d->bd_sel, PRINET);
841 	KNOTE_LOCKED(&d->bd_sel.si_note, 0);
842 }
843 
844 static void
845 bpf_timed_out(void *arg)
846 {
847 	struct bpf_d *d = (struct bpf_d *)arg;
848 
849 	BPFD_LOCK_ASSERT(d);
850 
851 	if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
852 		return;
853 	if (d->bd_state == BPF_WAITING) {
854 		d->bd_state = BPF_TIMED_OUT;
855 		if (d->bd_slen != 0)
856 			bpf_wakeup(d);
857 	}
858 }
859 
860 static int
861 bpf_ready(struct bpf_d *d)
862 {
863 
864 	BPFD_LOCK_ASSERT(d);
865 
866 	if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
867 		return (1);
868 	if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
869 	    d->bd_slen != 0)
870 		return (1);
871 	return (0);
872 }
873 
874 static int
875 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
876 {
877 	struct bpf_d *d;
878 	struct ifnet *ifp;
879 	struct mbuf *m, *mc;
880 	struct sockaddr dst;
881 	int error, hlen;
882 
883 	error = devfs_get_cdevpriv((void **)&d);
884 	if (error != 0)
885 		return (error);
886 
887 	d->bd_pid = curthread->td_proc->p_pid;
888 	d->bd_wcount++;
889 	if (d->bd_bif == NULL) {
890 		d->bd_wdcount++;
891 		return (ENXIO);
892 	}
893 
894 	ifp = d->bd_bif->bif_ifp;
895 
896 	if ((ifp->if_flags & IFF_UP) == 0) {
897 		d->bd_wdcount++;
898 		return (ENETDOWN);
899 	}
900 
901 	if (uio->uio_resid == 0) {
902 		d->bd_wdcount++;
903 		return (0);
904 	}
905 
906 	bzero(&dst, sizeof(dst));
907 	m = NULL;
908 	hlen = 0;
909 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
910 	    &m, &dst, &hlen, d->bd_wfilter);
911 	if (error) {
912 		d->bd_wdcount++;
913 		return (error);
914 	}
915 	d->bd_wfcount++;
916 	if (d->bd_hdrcmplt)
917 		dst.sa_family = pseudo_AF_HDRCMPLT;
918 
919 	if (d->bd_feedback) {
920 		mc = m_dup(m, M_DONTWAIT);
921 		if (mc != NULL)
922 			mc->m_pkthdr.rcvif = ifp;
923 		/* Set M_PROMISC for outgoing packets to be discarded. */
924 		if (d->bd_direction == BPF_D_INOUT)
925 			m->m_flags |= M_PROMISC;
926 	} else
927 		mc = NULL;
928 
929 	m->m_pkthdr.len -= hlen;
930 	m->m_len -= hlen;
931 	m->m_data += hlen;	/* XXX */
932 
933 	CURVNET_SET(ifp->if_vnet);
934 #ifdef MAC
935 	BPFD_LOCK(d);
936 	mac_bpfdesc_create_mbuf(d, m);
937 	if (mc != NULL)
938 		mac_bpfdesc_create_mbuf(d, mc);
939 	BPFD_UNLOCK(d);
940 #endif
941 
942 	error = (*ifp->if_output)(ifp, m, &dst, NULL);
943 	if (error)
944 		d->bd_wdcount++;
945 
946 	if (mc != NULL) {
947 		if (error == 0)
948 			(*ifp->if_input)(ifp, mc);
949 		else
950 			m_freem(mc);
951 	}
952 	CURVNET_RESTORE();
953 
954 	return (error);
955 }
956 
957 /*
958  * Reset a descriptor by flushing its packet buffer and clearing the receive
959  * and drop counts.  This is doable for kernel-only buffers, but with
960  * zero-copy buffers, we can't write to (or rotate) buffers that are
961  * currently owned by userspace.  It would be nice if we could encapsulate
962  * this logic in the buffer code rather than here.
963  */
964 static void
965 reset_d(struct bpf_d *d)
966 {
967 
968 	mtx_assert(&d->bd_mtx, MA_OWNED);
969 
970 	if ((d->bd_hbuf != NULL) &&
971 	    (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
972 		/* Free the hold buffer. */
973 		d->bd_fbuf = d->bd_hbuf;
974 		d->bd_hbuf = NULL;
975 		d->bd_hlen = 0;
976 		bpf_buf_reclaimed(d);
977 	}
978 	if (bpf_canwritebuf(d))
979 		d->bd_slen = 0;
980 	d->bd_rcount = 0;
981 	d->bd_dcount = 0;
982 	d->bd_fcount = 0;
983 	d->bd_wcount = 0;
984 	d->bd_wfcount = 0;
985 	d->bd_wdcount = 0;
986 	d->bd_zcopy = 0;
987 }
988 
989 /*
990  *  FIONREAD		Check for read packet available.
991  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
992  *  BIOCGBLEN		Get buffer len [for read()].
993  *  BIOCSETF		Set read filter.
994  *  BIOCSETFNR		Set read filter without resetting descriptor.
995  *  BIOCSETWF		Set write filter.
996  *  BIOCFLUSH		Flush read packet buffer.
997  *  BIOCPROMISC		Put interface into promiscuous mode.
998  *  BIOCGDLT		Get link layer type.
999  *  BIOCGETIF		Get interface name.
1000  *  BIOCSETIF		Set interface.
1001  *  BIOCSRTIMEOUT	Set read timeout.
1002  *  BIOCGRTIMEOUT	Get read timeout.
1003  *  BIOCGSTATS		Get packet stats.
1004  *  BIOCIMMEDIATE	Set immediate mode.
1005  *  BIOCVERSION		Get filter language version.
1006  *  BIOCGHDRCMPLT	Get "header already complete" flag
1007  *  BIOCSHDRCMPLT	Set "header already complete" flag
1008  *  BIOCGDIRECTION	Get packet direction flag
1009  *  BIOCSDIRECTION	Set packet direction flag
1010  *  BIOCLOCK		Set "locked" flag
1011  *  BIOCFEEDBACK	Set packet feedback mode.
1012  *  BIOCSETZBUF		Set current zero-copy buffer locations.
1013  *  BIOCGETZMAX		Get maximum zero-copy buffer size.
1014  *  BIOCROTZBUF		Force rotation of zero-copy buffer
1015  *  BIOCSETBUFMODE	Set buffer mode.
1016  *  BIOCGETBUFMODE	Get current buffer mode.
1017  */
1018 /* ARGSUSED */
1019 static	int
1020 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1021     struct thread *td)
1022 {
1023 	struct bpf_d *d;
1024 	int error;
1025 
1026 	error = devfs_get_cdevpriv((void **)&d);
1027 	if (error != 0)
1028 		return (error);
1029 
1030 	/*
1031 	 * Refresh PID associated with this descriptor.
1032 	 */
1033 	BPFD_LOCK(d);
1034 	d->bd_pid = td->td_proc->p_pid;
1035 	if (d->bd_state == BPF_WAITING)
1036 		callout_stop(&d->bd_callout);
1037 	d->bd_state = BPF_IDLE;
1038 	BPFD_UNLOCK(d);
1039 
1040 	if (d->bd_locked == 1) {
1041 		switch (cmd) {
1042 		case BIOCGBLEN:
1043 		case BIOCFLUSH:
1044 		case BIOCGDLT:
1045 		case BIOCGDLTLIST:
1046 #ifdef COMPAT_FREEBSD32
1047 		case BIOCGDLTLIST32:
1048 #endif
1049 		case BIOCGETIF:
1050 		case BIOCGRTIMEOUT:
1051 #ifdef COMPAT_FREEBSD32
1052 		case BIOCGRTIMEOUT32:
1053 #endif
1054 		case BIOCGSTATS:
1055 		case BIOCVERSION:
1056 		case BIOCGRSIG:
1057 		case BIOCGHDRCMPLT:
1058 		case BIOCFEEDBACK:
1059 		case FIONREAD:
1060 		case BIOCLOCK:
1061 		case BIOCSRTIMEOUT:
1062 #ifdef COMPAT_FREEBSD32
1063 		case BIOCSRTIMEOUT32:
1064 #endif
1065 		case BIOCIMMEDIATE:
1066 		case TIOCGPGRP:
1067 		case BIOCROTZBUF:
1068 			break;
1069 		default:
1070 			return (EPERM);
1071 		}
1072 	}
1073 #ifdef COMPAT_FREEBSD32
1074 	/*
1075 	 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1076 	 * that it will get 32-bit packet headers.
1077 	 */
1078 	switch (cmd) {
1079 	case BIOCSETF32:
1080 	case BIOCSETFNR32:
1081 	case BIOCSETWF32:
1082 	case BIOCGDLTLIST32:
1083 	case BIOCGRTIMEOUT32:
1084 	case BIOCSRTIMEOUT32:
1085 		d->bd_compat32 = 1;
1086 	}
1087 #endif
1088 
1089 	CURVNET_SET(TD_TO_VNET(td));
1090 	switch (cmd) {
1091 
1092 	default:
1093 		error = EINVAL;
1094 		break;
1095 
1096 	/*
1097 	 * Check for read packet available.
1098 	 */
1099 	case FIONREAD:
1100 		{
1101 			int n;
1102 
1103 			BPFD_LOCK(d);
1104 			n = d->bd_slen;
1105 			if (d->bd_hbuf)
1106 				n += d->bd_hlen;
1107 			BPFD_UNLOCK(d);
1108 
1109 			*(int *)addr = n;
1110 			break;
1111 		}
1112 
1113 	case SIOCGIFADDR:
1114 		{
1115 			struct ifnet *ifp;
1116 
1117 			if (d->bd_bif == NULL)
1118 				error = EINVAL;
1119 			else {
1120 				ifp = d->bd_bif->bif_ifp;
1121 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
1122 			}
1123 			break;
1124 		}
1125 
1126 	/*
1127 	 * Get buffer len [for read()].
1128 	 */
1129 	case BIOCGBLEN:
1130 		*(u_int *)addr = d->bd_bufsize;
1131 		break;
1132 
1133 	/*
1134 	 * Set buffer length.
1135 	 */
1136 	case BIOCSBLEN:
1137 		error = bpf_ioctl_sblen(d, (u_int *)addr);
1138 		break;
1139 
1140 	/*
1141 	 * Set link layer read filter.
1142 	 */
1143 	case BIOCSETF:
1144 	case BIOCSETFNR:
1145 	case BIOCSETWF:
1146 #ifdef COMPAT_FREEBSD32
1147 	case BIOCSETF32:
1148 	case BIOCSETFNR32:
1149 	case BIOCSETWF32:
1150 #endif
1151 		error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1152 		break;
1153 
1154 	/*
1155 	 * Flush read packet buffer.
1156 	 */
1157 	case BIOCFLUSH:
1158 		BPFD_LOCK(d);
1159 		reset_d(d);
1160 		BPFD_UNLOCK(d);
1161 		break;
1162 
1163 	/*
1164 	 * Put interface into promiscuous mode.
1165 	 */
1166 	case BIOCPROMISC:
1167 		if (d->bd_bif == NULL) {
1168 			/*
1169 			 * No interface attached yet.
1170 			 */
1171 			error = EINVAL;
1172 			break;
1173 		}
1174 		if (d->bd_promisc == 0) {
1175 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1176 			if (error == 0)
1177 				d->bd_promisc = 1;
1178 		}
1179 		break;
1180 
1181 	/*
1182 	 * Get current data link type.
1183 	 */
1184 	case BIOCGDLT:
1185 		if (d->bd_bif == NULL)
1186 			error = EINVAL;
1187 		else
1188 			*(u_int *)addr = d->bd_bif->bif_dlt;
1189 		break;
1190 
1191 	/*
1192 	 * Get a list of supported data link types.
1193 	 */
1194 #ifdef COMPAT_FREEBSD32
1195 	case BIOCGDLTLIST32:
1196 		{
1197 			struct bpf_dltlist32 *list32;
1198 			struct bpf_dltlist dltlist;
1199 
1200 			list32 = (struct bpf_dltlist32 *)addr;
1201 			dltlist.bfl_len = list32->bfl_len;
1202 			dltlist.bfl_list = PTRIN(list32->bfl_list);
1203 			if (d->bd_bif == NULL)
1204 				error = EINVAL;
1205 			else {
1206 				error = bpf_getdltlist(d, &dltlist);
1207 				if (error == 0)
1208 					list32->bfl_len = dltlist.bfl_len;
1209 			}
1210 			break;
1211 		}
1212 #endif
1213 
1214 	case BIOCGDLTLIST:
1215 		if (d->bd_bif == NULL)
1216 			error = EINVAL;
1217 		else
1218 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1219 		break;
1220 
1221 	/*
1222 	 * Set data link type.
1223 	 */
1224 	case BIOCSDLT:
1225 		if (d->bd_bif == NULL)
1226 			error = EINVAL;
1227 		else
1228 			error = bpf_setdlt(d, *(u_int *)addr);
1229 		break;
1230 
1231 	/*
1232 	 * Get interface name.
1233 	 */
1234 	case BIOCGETIF:
1235 		if (d->bd_bif == NULL)
1236 			error = EINVAL;
1237 		else {
1238 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
1239 			struct ifreq *const ifr = (struct ifreq *)addr;
1240 
1241 			strlcpy(ifr->ifr_name, ifp->if_xname,
1242 			    sizeof(ifr->ifr_name));
1243 		}
1244 		break;
1245 
1246 	/*
1247 	 * Set interface.
1248 	 */
1249 	case BIOCSETIF:
1250 		error = bpf_setif(d, (struct ifreq *)addr);
1251 		break;
1252 
1253 	/*
1254 	 * Set read timeout.
1255 	 */
1256 	case BIOCSRTIMEOUT:
1257 #ifdef COMPAT_FREEBSD32
1258 	case BIOCSRTIMEOUT32:
1259 #endif
1260 		{
1261 			struct timeval *tv = (struct timeval *)addr;
1262 #ifdef COMPAT_FREEBSD32
1263 			struct timeval32 *tv32;
1264 			struct timeval tv64;
1265 
1266 			if (cmd == BIOCSRTIMEOUT32) {
1267 				tv32 = (struct timeval32 *)addr;
1268 				tv = &tv64;
1269 				tv->tv_sec = tv32->tv_sec;
1270 				tv->tv_usec = tv32->tv_usec;
1271 			} else
1272 #endif
1273 				tv = (struct timeval *)addr;
1274 
1275 			/*
1276 			 * Subtract 1 tick from tvtohz() since this isn't
1277 			 * a one-shot timer.
1278 			 */
1279 			if ((error = itimerfix(tv)) == 0)
1280 				d->bd_rtout = tvtohz(tv) - 1;
1281 			break;
1282 		}
1283 
1284 	/*
1285 	 * Get read timeout.
1286 	 */
1287 	case BIOCGRTIMEOUT:
1288 #ifdef COMPAT_FREEBSD32
1289 	case BIOCGRTIMEOUT32:
1290 #endif
1291 		{
1292 			struct timeval *tv;
1293 #ifdef COMPAT_FREEBSD32
1294 			struct timeval32 *tv32;
1295 			struct timeval tv64;
1296 
1297 			if (cmd == BIOCGRTIMEOUT32)
1298 				tv = &tv64;
1299 			else
1300 #endif
1301 				tv = (struct timeval *)addr;
1302 
1303 			tv->tv_sec = d->bd_rtout / hz;
1304 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1305 #ifdef COMPAT_FREEBSD32
1306 			if (cmd == BIOCGRTIMEOUT32) {
1307 				tv32 = (struct timeval32 *)addr;
1308 				tv32->tv_sec = tv->tv_sec;
1309 				tv32->tv_usec = tv->tv_usec;
1310 			}
1311 #endif
1312 
1313 			break;
1314 		}
1315 
1316 	/*
1317 	 * Get packet stats.
1318 	 */
1319 	case BIOCGSTATS:
1320 		{
1321 			struct bpf_stat *bs = (struct bpf_stat *)addr;
1322 
1323 			/* XXXCSJP overflow */
1324 			bs->bs_recv = d->bd_rcount;
1325 			bs->bs_drop = d->bd_dcount;
1326 			break;
1327 		}
1328 
1329 	/*
1330 	 * Set immediate mode.
1331 	 */
1332 	case BIOCIMMEDIATE:
1333 		d->bd_immediate = *(u_int *)addr;
1334 		break;
1335 
1336 	case BIOCVERSION:
1337 		{
1338 			struct bpf_version *bv = (struct bpf_version *)addr;
1339 
1340 			bv->bv_major = BPF_MAJOR_VERSION;
1341 			bv->bv_minor = BPF_MINOR_VERSION;
1342 			break;
1343 		}
1344 
1345 	/*
1346 	 * Get "header already complete" flag
1347 	 */
1348 	case BIOCGHDRCMPLT:
1349 		*(u_int *)addr = d->bd_hdrcmplt;
1350 		break;
1351 
1352 	/*
1353 	 * Set "header already complete" flag
1354 	 */
1355 	case BIOCSHDRCMPLT:
1356 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1357 		break;
1358 
1359 	/*
1360 	 * Get packet direction flag
1361 	 */
1362 	case BIOCGDIRECTION:
1363 		*(u_int *)addr = d->bd_direction;
1364 		break;
1365 
1366 	/*
1367 	 * Set packet direction flag
1368 	 */
1369 	case BIOCSDIRECTION:
1370 		{
1371 			u_int	direction;
1372 
1373 			direction = *(u_int *)addr;
1374 			switch (direction) {
1375 			case BPF_D_IN:
1376 			case BPF_D_INOUT:
1377 			case BPF_D_OUT:
1378 				d->bd_direction = direction;
1379 				break;
1380 			default:
1381 				error = EINVAL;
1382 			}
1383 		}
1384 		break;
1385 
1386 	case BIOCFEEDBACK:
1387 		d->bd_feedback = *(u_int *)addr;
1388 		break;
1389 
1390 	case BIOCLOCK:
1391 		d->bd_locked = 1;
1392 		break;
1393 
1394 	case FIONBIO:		/* Non-blocking I/O */
1395 		break;
1396 
1397 	case FIOASYNC:		/* Send signal on receive packets */
1398 		d->bd_async = *(int *)addr;
1399 		break;
1400 
1401 	case FIOSETOWN:
1402 		error = fsetown(*(int *)addr, &d->bd_sigio);
1403 		break;
1404 
1405 	case FIOGETOWN:
1406 		*(int *)addr = fgetown(&d->bd_sigio);
1407 		break;
1408 
1409 	/* This is deprecated, FIOSETOWN should be used instead. */
1410 	case TIOCSPGRP:
1411 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
1412 		break;
1413 
1414 	/* This is deprecated, FIOGETOWN should be used instead. */
1415 	case TIOCGPGRP:
1416 		*(int *)addr = -fgetown(&d->bd_sigio);
1417 		break;
1418 
1419 	case BIOCSRSIG:		/* Set receive signal */
1420 		{
1421 			u_int sig;
1422 
1423 			sig = *(u_int *)addr;
1424 
1425 			if (sig >= NSIG)
1426 				error = EINVAL;
1427 			else
1428 				d->bd_sig = sig;
1429 			break;
1430 		}
1431 	case BIOCGRSIG:
1432 		*(u_int *)addr = d->bd_sig;
1433 		break;
1434 
1435 	case BIOCGETBUFMODE:
1436 		*(u_int *)addr = d->bd_bufmode;
1437 		break;
1438 
1439 	case BIOCSETBUFMODE:
1440 		/*
1441 		 * Allow the buffering mode to be changed as long as we
1442 		 * haven't yet committed to a particular mode.  Our
1443 		 * definition of commitment, for now, is whether or not a
1444 		 * buffer has been allocated or an interface attached, since
1445 		 * that's the point where things get tricky.
1446 		 */
1447 		switch (*(u_int *)addr) {
1448 		case BPF_BUFMODE_BUFFER:
1449 			break;
1450 
1451 		case BPF_BUFMODE_ZBUF:
1452 			if (bpf_zerocopy_enable)
1453 				break;
1454 			/* FALLSTHROUGH */
1455 
1456 		default:
1457 			CURVNET_RESTORE();
1458 			return (EINVAL);
1459 		}
1460 
1461 		BPFD_LOCK(d);
1462 		if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1463 		    d->bd_fbuf != NULL || d->bd_bif != NULL) {
1464 			BPFD_UNLOCK(d);
1465 			CURVNET_RESTORE();
1466 			return (EBUSY);
1467 		}
1468 		d->bd_bufmode = *(u_int *)addr;
1469 		BPFD_UNLOCK(d);
1470 		break;
1471 
1472 	case BIOCGETZMAX:
1473 		error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1474 		break;
1475 
1476 	case BIOCSETZBUF:
1477 		error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1478 		break;
1479 
1480 	case BIOCROTZBUF:
1481 		error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1482 		break;
1483 	}
1484 	CURVNET_RESTORE();
1485 	return (error);
1486 }
1487 
1488 /*
1489  * Set d's packet filter program to fp.  If this file already has a filter,
1490  * free it and replace it.  Returns EINVAL for bogus requests.
1491  */
1492 static int
1493 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1494 {
1495 	struct bpf_insn *fcode, *old;
1496 	u_int wfilter, flen, size;
1497 #ifdef BPF_JITTER
1498 	bpf_jit_filter *ofunc;
1499 #endif
1500 #ifdef COMPAT_FREEBSD32
1501 	struct bpf_program32 *fp32;
1502 	struct bpf_program fp_swab;
1503 
1504 	if (cmd == BIOCSETWF32 || cmd == BIOCSETF32 || cmd == BIOCSETFNR32) {
1505 		fp32 = (struct bpf_program32 *)fp;
1506 		fp_swab.bf_len = fp32->bf_len;
1507 		fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1508 		fp = &fp_swab;
1509 		if (cmd == BIOCSETWF32)
1510 			cmd = BIOCSETWF;
1511 	}
1512 #endif
1513 	if (cmd == BIOCSETWF) {
1514 		old = d->bd_wfilter;
1515 		wfilter = 1;
1516 #ifdef BPF_JITTER
1517 		ofunc = NULL;
1518 #endif
1519 	} else {
1520 		wfilter = 0;
1521 		old = d->bd_rfilter;
1522 #ifdef BPF_JITTER
1523 		ofunc = d->bd_bfilter;
1524 #endif
1525 	}
1526 	if (fp->bf_insns == NULL) {
1527 		if (fp->bf_len != 0)
1528 			return (EINVAL);
1529 		BPFD_LOCK(d);
1530 		if (wfilter)
1531 			d->bd_wfilter = NULL;
1532 		else {
1533 			d->bd_rfilter = NULL;
1534 #ifdef BPF_JITTER
1535 			d->bd_bfilter = NULL;
1536 #endif
1537 			if (cmd == BIOCSETF)
1538 				reset_d(d);
1539 		}
1540 		BPFD_UNLOCK(d);
1541 		if (old != NULL)
1542 			free((caddr_t)old, M_BPF);
1543 #ifdef BPF_JITTER
1544 		if (ofunc != NULL)
1545 			bpf_destroy_jit_filter(ofunc);
1546 #endif
1547 		return (0);
1548 	}
1549 	flen = fp->bf_len;
1550 	if (flen > bpf_maxinsns)
1551 		return (EINVAL);
1552 
1553 	size = flen * sizeof(*fp->bf_insns);
1554 	fcode = (struct bpf_insn *)malloc(size, M_BPF, M_WAITOK);
1555 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
1556 	    bpf_validate(fcode, (int)flen)) {
1557 		BPFD_LOCK(d);
1558 		if (wfilter)
1559 			d->bd_wfilter = fcode;
1560 		else {
1561 			d->bd_rfilter = fcode;
1562 #ifdef BPF_JITTER
1563 			d->bd_bfilter = bpf_jitter(fcode, flen);
1564 #endif
1565 			if (cmd == BIOCSETF)
1566 				reset_d(d);
1567 		}
1568 		BPFD_UNLOCK(d);
1569 		if (old != NULL)
1570 			free((caddr_t)old, M_BPF);
1571 #ifdef BPF_JITTER
1572 		if (ofunc != NULL)
1573 			bpf_destroy_jit_filter(ofunc);
1574 #endif
1575 
1576 		return (0);
1577 	}
1578 	free((caddr_t)fcode, M_BPF);
1579 	return (EINVAL);
1580 }
1581 
1582 /*
1583  * Detach a file from its current interface (if attached at all) and attach
1584  * to the interface indicated by the name stored in ifr.
1585  * Return an errno or 0.
1586  */
1587 static int
1588 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1589 {
1590 	struct bpf_if *bp;
1591 	struct ifnet *theywant;
1592 
1593 	theywant = ifunit(ifr->ifr_name);
1594 	if (theywant == NULL || theywant->if_bpf == NULL)
1595 		return (ENXIO);
1596 
1597 	bp = theywant->if_bpf;
1598 
1599 	/*
1600 	 * Behavior here depends on the buffering model.  If we're using
1601 	 * kernel memory buffers, then we can allocate them here.  If we're
1602 	 * using zero-copy, then the user process must have registered
1603 	 * buffers by the time we get here.  If not, return an error.
1604 	 *
1605 	 * XXXRW: There are locking issues here with multi-threaded use: what
1606 	 * if two threads try to set the interface at once?
1607 	 */
1608 	switch (d->bd_bufmode) {
1609 	case BPF_BUFMODE_BUFFER:
1610 		if (d->bd_sbuf == NULL)
1611 			bpf_buffer_alloc(d);
1612 		KASSERT(d->bd_sbuf != NULL, ("bpf_setif: bd_sbuf NULL"));
1613 		break;
1614 
1615 	case BPF_BUFMODE_ZBUF:
1616 		if (d->bd_sbuf == NULL)
1617 			return (EINVAL);
1618 		break;
1619 
1620 	default:
1621 		panic("bpf_setif: bufmode %d", d->bd_bufmode);
1622 	}
1623 	if (bp != d->bd_bif) {
1624 		if (d->bd_bif)
1625 			/*
1626 			 * Detach if attached to something else.
1627 			 */
1628 			bpf_detachd(d);
1629 
1630 		bpf_attachd(d, bp);
1631 	}
1632 	BPFD_LOCK(d);
1633 	reset_d(d);
1634 	BPFD_UNLOCK(d);
1635 	return (0);
1636 }
1637 
1638 /*
1639  * Support for select() and poll() system calls
1640  *
1641  * Return true iff the specific operation will not block indefinitely.
1642  * Otherwise, return false but make a note that a selwakeup() must be done.
1643  */
1644 static int
1645 bpfpoll(struct cdev *dev, int events, struct thread *td)
1646 {
1647 	struct bpf_d *d;
1648 	int revents;
1649 
1650 	if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1651 		return (events &
1652 		    (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1653 
1654 	/*
1655 	 * Refresh PID associated with this descriptor.
1656 	 */
1657 	revents = events & (POLLOUT | POLLWRNORM);
1658 	BPFD_LOCK(d);
1659 	d->bd_pid = td->td_proc->p_pid;
1660 	if (events & (POLLIN | POLLRDNORM)) {
1661 		if (bpf_ready(d))
1662 			revents |= events & (POLLIN | POLLRDNORM);
1663 		else {
1664 			selrecord(td, &d->bd_sel);
1665 			/* Start the read timeout if necessary. */
1666 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1667 				callout_reset(&d->bd_callout, d->bd_rtout,
1668 				    bpf_timed_out, d);
1669 				d->bd_state = BPF_WAITING;
1670 			}
1671 		}
1672 	}
1673 	BPFD_UNLOCK(d);
1674 	return (revents);
1675 }
1676 
1677 /*
1678  * Support for kevent() system call.  Register EVFILT_READ filters and
1679  * reject all others.
1680  */
1681 int
1682 bpfkqfilter(struct cdev *dev, struct knote *kn)
1683 {
1684 	struct bpf_d *d;
1685 
1686 	if (devfs_get_cdevpriv((void **)&d) != 0 ||
1687 	    kn->kn_filter != EVFILT_READ)
1688 		return (1);
1689 
1690 	/*
1691 	 * Refresh PID associated with this descriptor.
1692 	 */
1693 	BPFD_LOCK(d);
1694 	d->bd_pid = curthread->td_proc->p_pid;
1695 	kn->kn_fop = &bpfread_filtops;
1696 	kn->kn_hook = d;
1697 	knlist_add(&d->bd_sel.si_note, kn, 1);
1698 	BPFD_UNLOCK(d);
1699 
1700 	return (0);
1701 }
1702 
1703 static void
1704 filt_bpfdetach(struct knote *kn)
1705 {
1706 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1707 
1708 	knlist_remove(&d->bd_sel.si_note, kn, 0);
1709 }
1710 
1711 static int
1712 filt_bpfread(struct knote *kn, long hint)
1713 {
1714 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1715 	int ready;
1716 
1717 	BPFD_LOCK_ASSERT(d);
1718 	ready = bpf_ready(d);
1719 	if (ready) {
1720 		kn->kn_data = d->bd_slen;
1721 		if (d->bd_hbuf)
1722 			kn->kn_data += d->bd_hlen;
1723 	} else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1724 		callout_reset(&d->bd_callout, d->bd_rtout,
1725 		    bpf_timed_out, d);
1726 		d->bd_state = BPF_WAITING;
1727 	}
1728 
1729 	return (ready);
1730 }
1731 
1732 /*
1733  * Incoming linkage from device drivers.  Process the packet pkt, of length
1734  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1735  * by each process' filter, and if accepted, stashed into the corresponding
1736  * buffer.
1737  */
1738 void
1739 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1740 {
1741 	struct bpf_d *d;
1742 #ifdef BPF_JITTER
1743 	bpf_jit_filter *bf;
1744 #endif
1745 	u_int slen;
1746 	int gottime;
1747 	struct timeval tv;
1748 
1749 	gottime = 0;
1750 	BPFIF_LOCK(bp);
1751 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1752 		BPFD_LOCK(d);
1753 		++d->bd_rcount;
1754 		/*
1755 		 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
1756 		 * way for the caller to indiciate to us whether this packet
1757 		 * is inbound or outbound.  In the bpf_mtap() routines, we use
1758 		 * the interface pointers on the mbuf to figure it out.
1759 		 */
1760 #ifdef BPF_JITTER
1761 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
1762 		if (bf != NULL)
1763 			slen = (*(bf->func))(pkt, pktlen, pktlen);
1764 		else
1765 #endif
1766 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1767 		if (slen != 0) {
1768 			d->bd_fcount++;
1769 			if (!gottime) {
1770 				microtime(&tv);
1771 				gottime = 1;
1772 			}
1773 #ifdef MAC
1774 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1775 #endif
1776 				catchpacket(d, pkt, pktlen, slen,
1777 				    bpf_append_bytes, &tv);
1778 		}
1779 		BPFD_UNLOCK(d);
1780 	}
1781 	BPFIF_UNLOCK(bp);
1782 }
1783 
1784 #define	BPF_CHECK_DIRECTION(d, r, i)				\
1785 	    (((d)->bd_direction == BPF_D_IN && (r) != (i)) ||	\
1786 	    ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
1787 
1788 /*
1789  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1790  */
1791 void
1792 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1793 {
1794 	struct bpf_d *d;
1795 #ifdef BPF_JITTER
1796 	bpf_jit_filter *bf;
1797 #endif
1798 	u_int pktlen, slen;
1799 	int gottime;
1800 	struct timeval tv;
1801 
1802 	/* Skip outgoing duplicate packets. */
1803 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1804 		m->m_flags &= ~M_PROMISC;
1805 		return;
1806 	}
1807 
1808 	gottime = 0;
1809 
1810 	pktlen = m_length(m, NULL);
1811 
1812 	BPFIF_LOCK(bp);
1813 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1814 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
1815 			continue;
1816 		BPFD_LOCK(d);
1817 		++d->bd_rcount;
1818 #ifdef BPF_JITTER
1819 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
1820 		/* XXX We cannot handle multiple mbufs. */
1821 		if (bf != NULL && m->m_next == NULL)
1822 			slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
1823 		else
1824 #endif
1825 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1826 		if (slen != 0) {
1827 			d->bd_fcount++;
1828 			if (!gottime) {
1829 				microtime(&tv);
1830 				gottime = 1;
1831 			}
1832 #ifdef MAC
1833 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1834 #endif
1835 				catchpacket(d, (u_char *)m, pktlen, slen,
1836 				    bpf_append_mbuf, &tv);
1837 		}
1838 		BPFD_UNLOCK(d);
1839 	}
1840 	BPFIF_UNLOCK(bp);
1841 }
1842 
1843 /*
1844  * Incoming linkage from device drivers, when packet is in
1845  * an mbuf chain and to be prepended by a contiguous header.
1846  */
1847 void
1848 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1849 {
1850 	struct mbuf mb;
1851 	struct bpf_d *d;
1852 	u_int pktlen, slen;
1853 	int gottime;
1854 	struct timeval tv;
1855 
1856 	/* Skip outgoing duplicate packets. */
1857 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1858 		m->m_flags &= ~M_PROMISC;
1859 		return;
1860 	}
1861 
1862 	gottime = 0;
1863 
1864 	pktlen = m_length(m, NULL);
1865 	/*
1866 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1867 	 * Note that we cut corners here; we only setup what's
1868 	 * absolutely needed--this mbuf should never go anywhere else.
1869 	 */
1870 	mb.m_next = m;
1871 	mb.m_data = data;
1872 	mb.m_len = dlen;
1873 	pktlen += dlen;
1874 
1875 	BPFIF_LOCK(bp);
1876 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1877 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
1878 			continue;
1879 		BPFD_LOCK(d);
1880 		++d->bd_rcount;
1881 		slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
1882 		if (slen != 0) {
1883 			d->bd_fcount++;
1884 			if (!gottime) {
1885 				microtime(&tv);
1886 				gottime = 1;
1887 			}
1888 #ifdef MAC
1889 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1890 #endif
1891 				catchpacket(d, (u_char *)&mb, pktlen, slen,
1892 				    bpf_append_mbuf, &tv);
1893 		}
1894 		BPFD_UNLOCK(d);
1895 	}
1896 	BPFIF_UNLOCK(bp);
1897 }
1898 
1899 #undef	BPF_CHECK_DIRECTION
1900 
1901 /*
1902  * Move the packet data from interface memory (pkt) into the
1903  * store buffer.  "cpfn" is the routine called to do the actual data
1904  * transfer.  bcopy is passed in to copy contiguous chunks, while
1905  * bpf_append_mbuf is passed in to copy mbuf chains.  In the latter case,
1906  * pkt is really an mbuf.
1907  */
1908 static void
1909 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1910     void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
1911     struct timeval *tv)
1912 {
1913 	struct bpf_hdr hdr;
1914 #ifdef COMPAT_FREEBSD32
1915 	struct bpf_hdr32 hdr32;
1916 #endif
1917 	int totlen, curlen;
1918 	int hdrlen = d->bd_bif->bif_hdrlen;
1919 	int do_wakeup = 0;
1920 
1921 	BPFD_LOCK_ASSERT(d);
1922 
1923 	/*
1924 	 * Detect whether user space has released a buffer back to us, and if
1925 	 * so, move it from being a hold buffer to a free buffer.  This may
1926 	 * not be the best place to do it (for example, we might only want to
1927 	 * run this check if we need the space), but for now it's a reliable
1928 	 * spot to do it.
1929 	 */
1930 	if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
1931 		d->bd_fbuf = d->bd_hbuf;
1932 		d->bd_hbuf = NULL;
1933 		d->bd_hlen = 0;
1934 		bpf_buf_reclaimed(d);
1935 	}
1936 
1937 	/*
1938 	 * Figure out how many bytes to move.  If the packet is
1939 	 * greater or equal to the snapshot length, transfer that
1940 	 * much.  Otherwise, transfer the whole packet (unless
1941 	 * we hit the buffer size limit).
1942 	 */
1943 	totlen = hdrlen + min(snaplen, pktlen);
1944 	if (totlen > d->bd_bufsize)
1945 		totlen = d->bd_bufsize;
1946 
1947 	/*
1948 	 * Round up the end of the previous packet to the next longword.
1949 	 *
1950 	 * Drop the packet if there's no room and no hope of room
1951 	 * If the packet would overflow the storage buffer or the storage
1952 	 * buffer is considered immutable by the buffer model, try to rotate
1953 	 * the buffer and wakeup pending processes.
1954 	 */
1955 #ifdef COMPAT_FREEBSD32
1956 	if (d->bd_compat32)
1957 		curlen = BPF_WORDALIGN32(d->bd_slen);
1958 	else
1959 #endif
1960 		curlen = BPF_WORDALIGN(d->bd_slen);
1961 	if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
1962 		if (d->bd_fbuf == NULL) {
1963 			/*
1964 			 * There's no room in the store buffer, and no
1965 			 * prospect of room, so drop the packet.  Notify the
1966 			 * buffer model.
1967 			 */
1968 			bpf_buffull(d);
1969 			++d->bd_dcount;
1970 			return;
1971 		}
1972 		ROTATE_BUFFERS(d);
1973 		do_wakeup = 1;
1974 		curlen = 0;
1975 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
1976 		/*
1977 		 * Immediate mode is set, or the read timeout has already
1978 		 * expired during a select call.  A packet arrived, so the
1979 		 * reader should be woken up.
1980 		 */
1981 		do_wakeup = 1;
1982 #ifdef COMPAT_FREEBSD32
1983 	/*
1984 	 * If this is a 32-bit stream, then stick a 32-bit header at the
1985 	 * front and copy the data into the buffer.
1986 	 */
1987 	if (d->bd_compat32) {
1988 		bzero(&hdr32, sizeof(hdr32));
1989 		hdr32.bh_tstamp.tv_sec = tv->tv_sec;
1990 		hdr32.bh_tstamp.tv_usec = tv->tv_usec;
1991 		hdr32.bh_datalen = pktlen;
1992 		hdr32.bh_hdrlen = hdrlen;
1993 		hdr.bh_caplen = hdr32.bh_caplen = totlen - hdrlen;
1994 		bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32, sizeof(hdr32));
1995 		goto copy;
1996 	}
1997 #endif
1998 
1999 	/*
2000 	 * Append the bpf header.  Note we append the actual header size, but
2001 	 * move forward the length of the header plus padding.
2002 	 */
2003 	bzero(&hdr, sizeof(hdr));
2004 	hdr.bh_tstamp = *tv;
2005 	hdr.bh_datalen = pktlen;
2006 	hdr.bh_hdrlen = hdrlen;
2007 	hdr.bh_caplen = totlen - hdrlen;
2008 	bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2009 
2010 	/*
2011 	 * Copy the packet data into the store buffer and update its length.
2012 	 */
2013 #ifdef COMPAT_FREEBSD32
2014  copy:
2015 #endif
2016 	(*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, hdr.bh_caplen);
2017 	d->bd_slen = curlen + totlen;
2018 
2019 	if (do_wakeup)
2020 		bpf_wakeup(d);
2021 }
2022 
2023 /*
2024  * Free buffers currently in use by a descriptor.
2025  * Called on close.
2026  */
2027 static void
2028 bpf_freed(struct bpf_d *d)
2029 {
2030 
2031 	/*
2032 	 * We don't need to lock out interrupts since this descriptor has
2033 	 * been detached from its interface and it yet hasn't been marked
2034 	 * free.
2035 	 */
2036 	bpf_free(d);
2037 	if (d->bd_rfilter != NULL) {
2038 		free((caddr_t)d->bd_rfilter, M_BPF);
2039 #ifdef BPF_JITTER
2040 		if (d->bd_bfilter != NULL)
2041 			bpf_destroy_jit_filter(d->bd_bfilter);
2042 #endif
2043 	}
2044 	if (d->bd_wfilter != NULL)
2045 		free((caddr_t)d->bd_wfilter, M_BPF);
2046 	mtx_destroy(&d->bd_mtx);
2047 }
2048 
2049 /*
2050  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
2051  * fixed size of the link header (variable length headers not yet supported).
2052  */
2053 void
2054 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2055 {
2056 
2057 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2058 }
2059 
2060 /*
2061  * Attach an interface to bpf.  ifp is a pointer to the structure
2062  * defining the interface to be attached, dlt is the link layer type,
2063  * and hdrlen is the fixed size of the link header (variable length
2064  * headers are not yet supporrted).
2065  */
2066 void
2067 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2068 {
2069 	struct bpf_if *bp;
2070 
2071 	bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2072 	if (bp == NULL)
2073 		panic("bpfattach");
2074 
2075 	LIST_INIT(&bp->bif_dlist);
2076 	bp->bif_ifp = ifp;
2077 	bp->bif_dlt = dlt;
2078 	mtx_init(&bp->bif_mtx, "bpf interface lock", NULL, MTX_DEF);
2079 	KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2080 	*driverp = bp;
2081 
2082 	mtx_lock(&bpf_mtx);
2083 	LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2084 	mtx_unlock(&bpf_mtx);
2085 
2086 	/*
2087 	 * Compute the length of the bpf header.  This is not necessarily
2088 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
2089 	 * that the network layer header begins on a longword boundary (for
2090 	 * performance reasons and to alleviate alignment restrictions).
2091 	 */
2092 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
2093 
2094 	if (bootverbose)
2095 		if_printf(ifp, "bpf attached\n");
2096 }
2097 
2098 /*
2099  * Detach bpf from an interface.  This involves detaching each descriptor
2100  * associated with the interface, and leaving bd_bif NULL.  Notify each
2101  * descriptor as it's detached so that any sleepers wake up and get
2102  * ENXIO.
2103  */
2104 void
2105 bpfdetach(struct ifnet *ifp)
2106 {
2107 	struct bpf_if	*bp;
2108 	struct bpf_d	*d;
2109 
2110 	/* Locate BPF interface information */
2111 	mtx_lock(&bpf_mtx);
2112 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2113 		if (ifp == bp->bif_ifp)
2114 			break;
2115 	}
2116 
2117 	/* Interface wasn't attached */
2118 	if ((bp == NULL) || (bp->bif_ifp == NULL)) {
2119 		mtx_unlock(&bpf_mtx);
2120 		printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2121 		return;
2122 	}
2123 
2124 	LIST_REMOVE(bp, bif_next);
2125 	mtx_unlock(&bpf_mtx);
2126 
2127 	while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2128 		bpf_detachd(d);
2129 		BPFD_LOCK(d);
2130 		bpf_wakeup(d);
2131 		BPFD_UNLOCK(d);
2132 	}
2133 
2134 	mtx_destroy(&bp->bif_mtx);
2135 	free(bp, M_BPF);
2136 }
2137 
2138 /*
2139  * Get a list of available data link type of the interface.
2140  */
2141 static int
2142 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2143 {
2144 	int n, error;
2145 	struct ifnet *ifp;
2146 	struct bpf_if *bp;
2147 
2148 	ifp = d->bd_bif->bif_ifp;
2149 	n = 0;
2150 	error = 0;
2151 	mtx_lock(&bpf_mtx);
2152 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2153 		if (bp->bif_ifp != ifp)
2154 			continue;
2155 		if (bfl->bfl_list != NULL) {
2156 			if (n >= bfl->bfl_len) {
2157 				mtx_unlock(&bpf_mtx);
2158 				return (ENOMEM);
2159 			}
2160 			error = copyout(&bp->bif_dlt,
2161 			    bfl->bfl_list + n, sizeof(u_int));
2162 		}
2163 		n++;
2164 	}
2165 	mtx_unlock(&bpf_mtx);
2166 	bfl->bfl_len = n;
2167 	return (error);
2168 }
2169 
2170 /*
2171  * Set the data link type of a BPF instance.
2172  */
2173 static int
2174 bpf_setdlt(struct bpf_d *d, u_int dlt)
2175 {
2176 	int error, opromisc;
2177 	struct ifnet *ifp;
2178 	struct bpf_if *bp;
2179 
2180 	if (d->bd_bif->bif_dlt == dlt)
2181 		return (0);
2182 	ifp = d->bd_bif->bif_ifp;
2183 	mtx_lock(&bpf_mtx);
2184 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2185 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2186 			break;
2187 	}
2188 	mtx_unlock(&bpf_mtx);
2189 	if (bp != NULL) {
2190 		opromisc = d->bd_promisc;
2191 		bpf_detachd(d);
2192 		bpf_attachd(d, bp);
2193 		BPFD_LOCK(d);
2194 		reset_d(d);
2195 		BPFD_UNLOCK(d);
2196 		if (opromisc) {
2197 			error = ifpromisc(bp->bif_ifp, 1);
2198 			if (error)
2199 				if_printf(bp->bif_ifp,
2200 					"bpf_setdlt: ifpromisc failed (%d)\n",
2201 					error);
2202 			else
2203 				d->bd_promisc = 1;
2204 		}
2205 	}
2206 	return (bp == NULL ? EINVAL : 0);
2207 }
2208 
2209 static void
2210 bpf_drvinit(void *unused)
2211 {
2212 	struct cdev *dev;
2213 
2214 	mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2215 	LIST_INIT(&bpf_iflist);
2216 
2217 	dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2218 	/* For compatibility */
2219 	make_dev_alias(dev, "bpf0");
2220 }
2221 
2222 /*
2223  * Zero out the various packet counters associated with all of the bpf
2224  * descriptors.  At some point, we will probably want to get a bit more
2225  * granular and allow the user to specify descriptors to be zeroed.
2226  */
2227 static void
2228 bpf_zero_counters(void)
2229 {
2230 	struct bpf_if *bp;
2231 	struct bpf_d *bd;
2232 
2233 	mtx_lock(&bpf_mtx);
2234 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2235 		BPFIF_LOCK(bp);
2236 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2237 			BPFD_LOCK(bd);
2238 			bd->bd_rcount = 0;
2239 			bd->bd_dcount = 0;
2240 			bd->bd_fcount = 0;
2241 			bd->bd_wcount = 0;
2242 			bd->bd_wfcount = 0;
2243 			bd->bd_zcopy = 0;
2244 			BPFD_UNLOCK(bd);
2245 		}
2246 		BPFIF_UNLOCK(bp);
2247 	}
2248 	mtx_unlock(&bpf_mtx);
2249 }
2250 
2251 static void
2252 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2253 {
2254 
2255 	bzero(d, sizeof(*d));
2256 	BPFD_LOCK_ASSERT(bd);
2257 	d->bd_structsize = sizeof(*d);
2258 	d->bd_immediate = bd->bd_immediate;
2259 	d->bd_promisc = bd->bd_promisc;
2260 	d->bd_hdrcmplt = bd->bd_hdrcmplt;
2261 	d->bd_direction = bd->bd_direction;
2262 	d->bd_feedback = bd->bd_feedback;
2263 	d->bd_async = bd->bd_async;
2264 	d->bd_rcount = bd->bd_rcount;
2265 	d->bd_dcount = bd->bd_dcount;
2266 	d->bd_fcount = bd->bd_fcount;
2267 	d->bd_sig = bd->bd_sig;
2268 	d->bd_slen = bd->bd_slen;
2269 	d->bd_hlen = bd->bd_hlen;
2270 	d->bd_bufsize = bd->bd_bufsize;
2271 	d->bd_pid = bd->bd_pid;
2272 	strlcpy(d->bd_ifname,
2273 	    bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2274 	d->bd_locked = bd->bd_locked;
2275 	d->bd_wcount = bd->bd_wcount;
2276 	d->bd_wdcount = bd->bd_wdcount;
2277 	d->bd_wfcount = bd->bd_wfcount;
2278 	d->bd_zcopy = bd->bd_zcopy;
2279 	d->bd_bufmode = bd->bd_bufmode;
2280 }
2281 
2282 static int
2283 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2284 {
2285 	struct xbpf_d *xbdbuf, *xbd, zerostats;
2286 	int index, error;
2287 	struct bpf_if *bp;
2288 	struct bpf_d *bd;
2289 
2290 	/*
2291 	 * XXX This is not technically correct. It is possible for non
2292 	 * privileged users to open bpf devices. It would make sense
2293 	 * if the users who opened the devices were able to retrieve
2294 	 * the statistics for them, too.
2295 	 */
2296 	error = priv_check(req->td, PRIV_NET_BPF);
2297 	if (error)
2298 		return (error);
2299 	/*
2300 	 * Check to see if the user is requesting that the counters be
2301 	 * zeroed out.  Explicitly check that the supplied data is zeroed,
2302 	 * as we aren't allowing the user to set the counters currently.
2303 	 */
2304 	if (req->newptr != NULL) {
2305 		if (req->newlen != sizeof(zerostats))
2306 			return (EINVAL);
2307 		bzero(&zerostats, sizeof(zerostats));
2308 		xbd = req->newptr;
2309 		if (bcmp(xbd, &zerostats, sizeof(*xbd)) != 0)
2310 			return (EINVAL);
2311 		bpf_zero_counters();
2312 		return (0);
2313 	}
2314 	if (req->oldptr == NULL)
2315 		return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2316 	if (bpf_bpfd_cnt == 0)
2317 		return (SYSCTL_OUT(req, 0, 0));
2318 	xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2319 	mtx_lock(&bpf_mtx);
2320 	if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2321 		mtx_unlock(&bpf_mtx);
2322 		free(xbdbuf, M_BPF);
2323 		return (ENOMEM);
2324 	}
2325 	index = 0;
2326 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2327 		BPFIF_LOCK(bp);
2328 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2329 			xbd = &xbdbuf[index++];
2330 			BPFD_LOCK(bd);
2331 			bpfstats_fill_xbpf(xbd, bd);
2332 			BPFD_UNLOCK(bd);
2333 		}
2334 		BPFIF_UNLOCK(bp);
2335 	}
2336 	mtx_unlock(&bpf_mtx);
2337 	error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2338 	free(xbdbuf, M_BPF);
2339 	return (error);
2340 }
2341 
2342 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2343 
2344 #else /* !DEV_BPF && !NETGRAPH_BPF */
2345 /*
2346  * NOP stubs to allow bpf-using drivers to load and function.
2347  *
2348  * A 'better' implementation would allow the core bpf functionality
2349  * to be loaded at runtime.
2350  */
2351 static struct bpf_if bp_null;
2352 
2353 void
2354 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2355 {
2356 }
2357 
2358 void
2359 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2360 {
2361 }
2362 
2363 void
2364 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2365 {
2366 }
2367 
2368 void
2369 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2370 {
2371 
2372 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2373 }
2374 
2375 void
2376 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2377 {
2378 
2379 	*driverp = &bp_null;
2380 }
2381 
2382 void
2383 bpfdetach(struct ifnet *ifp)
2384 {
2385 }
2386 
2387 u_int
2388 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2389 {
2390 	return -1;	/* "no filter" behaviour */
2391 }
2392 
2393 int
2394 bpf_validate(const struct bpf_insn *f, int len)
2395 {
2396 	return 0;		/* false */
2397 }
2398 
2399 #endif /* !DEV_BPF && !NETGRAPH_BPF */
2400