xref: /freebsd/sys/net/bpf.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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_netgraph.h"
42 
43 #include <sys/types.h>
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/conf.h>
47 #include <sys/fcntl.h>
48 #include <sys/jail.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/time.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/signalvar.h>
55 #include <sys/filio.h>
56 #include <sys/sockio.h>
57 #include <sys/ttycom.h>
58 #include <sys/uio.h>
59 
60 #include <sys/event.h>
61 #include <sys/file.h>
62 #include <sys/poll.h>
63 #include <sys/proc.h>
64 
65 #include <sys/socket.h>
66 
67 #include <net/if.h>
68 #include <net/bpf.h>
69 #include <net/bpf_buffer.h>
70 #ifdef BPF_JITTER
71 #include <net/bpf_jitter.h>
72 #endif
73 #include <net/bpf_zerocopy.h>
74 #include <net/bpfdesc.h>
75 #include <net/vnet.h>
76 
77 #include <netinet/in.h>
78 #include <netinet/if_ether.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81 
82 #include <net80211/ieee80211_freebsd.h>
83 
84 #include <security/mac/mac_framework.h>
85 
86 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
87 
88 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
89 
90 #define PRINET  26			/* interruptible */
91 
92 /*
93  * bpf_iflist is a list of BPF interface structures, each corresponding to a
94  * specific DLT.  The same network interface might have several BPF interface
95  * structures registered by different layers in the stack (i.e., 802.11
96  * frames, ethernet frames, etc).
97  */
98 static LIST_HEAD(, bpf_if)	bpf_iflist;
99 static struct mtx	bpf_mtx;		/* bpf global lock */
100 static int		bpf_bpfd_cnt;
101 
102 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
103 static void	bpf_detachd(struct bpf_d *);
104 static void	bpf_freed(struct bpf_d *);
105 static int	bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
106 		    struct sockaddr *, int *, struct bpf_insn *);
107 static int	bpf_setif(struct bpf_d *, struct ifreq *);
108 static void	bpf_timed_out(void *);
109 static __inline void
110 		bpf_wakeup(struct bpf_d *);
111 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
112 		    void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
113 		    struct timeval *);
114 static void	reset_d(struct bpf_d *);
115 static int	 bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
116 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
117 static int	bpf_setdlt(struct bpf_d *, u_int);
118 static void	filt_bpfdetach(struct knote *);
119 static int	filt_bpfread(struct knote *, long);
120 static void	bpf_drvinit(void *);
121 static int	bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
122 
123 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
124 int bpf_maxinsns = BPF_MAXINSNS;
125 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
126     &bpf_maxinsns, 0, "Maximum bpf program instructions");
127 static int bpf_zerocopy_enable = 0;
128 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
129     &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
130 SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
131     bpf_stats_sysctl, "bpf statistics portal");
132 
133 static	d_open_t	bpfopen;
134 static	d_read_t	bpfread;
135 static	d_write_t	bpfwrite;
136 static	d_ioctl_t	bpfioctl;
137 static	d_poll_t	bpfpoll;
138 static	d_kqfilter_t	bpfkqfilter;
139 
140 static struct cdevsw bpf_cdevsw = {
141 	.d_version =	D_VERSION,
142 	.d_open =	bpfopen,
143 	.d_read =	bpfread,
144 	.d_write =	bpfwrite,
145 	.d_ioctl =	bpfioctl,
146 	.d_poll =	bpfpoll,
147 	.d_name =	"bpf",
148 	.d_kqfilter =	bpfkqfilter,
149 };
150 
151 static struct filterops bpfread_filtops =
152 	{ 1, NULL, filt_bpfdetach, filt_bpfread };
153 
154 /*
155  * Wrapper functions for various buffering methods.  If the set of buffer
156  * modes expands, we will probably want to introduce a switch data structure
157  * similar to protosw, et.
158  */
159 static void
160 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
161     u_int len)
162 {
163 
164 	BPFD_LOCK_ASSERT(d);
165 
166 	switch (d->bd_bufmode) {
167 	case BPF_BUFMODE_BUFFER:
168 		return (bpf_buffer_append_bytes(d, buf, offset, src, len));
169 
170 	case BPF_BUFMODE_ZBUF:
171 		d->bd_zcopy++;
172 		return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
173 
174 	default:
175 		panic("bpf_buf_append_bytes");
176 	}
177 }
178 
179 static void
180 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
181     u_int len)
182 {
183 
184 	BPFD_LOCK_ASSERT(d);
185 
186 	switch (d->bd_bufmode) {
187 	case BPF_BUFMODE_BUFFER:
188 		return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
189 
190 	case BPF_BUFMODE_ZBUF:
191 		d->bd_zcopy++;
192 		return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
193 
194 	default:
195 		panic("bpf_buf_append_mbuf");
196 	}
197 }
198 
199 /*
200  * This function gets called when the free buffer is re-assigned.
201  */
202 static void
203 bpf_buf_reclaimed(struct bpf_d *d)
204 {
205 
206 	BPFD_LOCK_ASSERT(d);
207 
208 	switch (d->bd_bufmode) {
209 	case BPF_BUFMODE_BUFFER:
210 		return;
211 
212 	case BPF_BUFMODE_ZBUF:
213 		bpf_zerocopy_buf_reclaimed(d);
214 		return;
215 
216 	default:
217 		panic("bpf_buf_reclaimed");
218 	}
219 }
220 
221 /*
222  * If the buffer mechanism has a way to decide that a held buffer can be made
223  * free, then it is exposed via the bpf_canfreebuf() interface.  (1) is
224  * returned if the buffer can be discarded, (0) is returned if it cannot.
225  */
226 static int
227 bpf_canfreebuf(struct bpf_d *d)
228 {
229 
230 	BPFD_LOCK_ASSERT(d);
231 
232 	switch (d->bd_bufmode) {
233 	case BPF_BUFMODE_ZBUF:
234 		return (bpf_zerocopy_canfreebuf(d));
235 	}
236 	return (0);
237 }
238 
239 /*
240  * Allow the buffer model to indicate that the current store buffer is
241  * immutable, regardless of the appearance of space.  Return (1) if the
242  * buffer is writable, and (0) if not.
243  */
244 static int
245 bpf_canwritebuf(struct bpf_d *d)
246 {
247 
248 	BPFD_LOCK_ASSERT(d);
249 
250 	switch (d->bd_bufmode) {
251 	case BPF_BUFMODE_ZBUF:
252 		return (bpf_zerocopy_canwritebuf(d));
253 	}
254 	return (1);
255 }
256 
257 /*
258  * Notify buffer model that an attempt to write to the store buffer has
259  * resulted in a dropped packet, in which case the buffer may be considered
260  * full.
261  */
262 static void
263 bpf_buffull(struct bpf_d *d)
264 {
265 
266 	BPFD_LOCK_ASSERT(d);
267 
268 	switch (d->bd_bufmode) {
269 	case BPF_BUFMODE_ZBUF:
270 		bpf_zerocopy_buffull(d);
271 		break;
272 	}
273 }
274 
275 /*
276  * Notify the buffer model that a buffer has moved into the hold position.
277  */
278 void
279 bpf_bufheld(struct bpf_d *d)
280 {
281 
282 	BPFD_LOCK_ASSERT(d);
283 
284 	switch (d->bd_bufmode) {
285 	case BPF_BUFMODE_ZBUF:
286 		bpf_zerocopy_bufheld(d);
287 		break;
288 	}
289 }
290 
291 static void
292 bpf_free(struct bpf_d *d)
293 {
294 
295 	switch (d->bd_bufmode) {
296 	case BPF_BUFMODE_BUFFER:
297 		return (bpf_buffer_free(d));
298 
299 	case BPF_BUFMODE_ZBUF:
300 		return (bpf_zerocopy_free(d));
301 
302 	default:
303 		panic("bpf_buf_free");
304 	}
305 }
306 
307 static int
308 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
309 {
310 
311 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
312 		return (EOPNOTSUPP);
313 	return (bpf_buffer_uiomove(d, buf, len, uio));
314 }
315 
316 static int
317 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
318 {
319 
320 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
321 		return (EOPNOTSUPP);
322 	return (bpf_buffer_ioctl_sblen(d, i));
323 }
324 
325 static int
326 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
327 {
328 
329 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
330 		return (EOPNOTSUPP);
331 	return (bpf_zerocopy_ioctl_getzmax(td, d, i));
332 }
333 
334 static int
335 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
336 {
337 
338 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
339 		return (EOPNOTSUPP);
340 	return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
341 }
342 
343 static int
344 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
345 {
346 
347 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
348 		return (EOPNOTSUPP);
349 	return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
350 }
351 
352 /*
353  * General BPF functions.
354  */
355 static int
356 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
357     struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter)
358 {
359 	const struct ieee80211_bpf_params *p;
360 	struct ether_header *eh;
361 	struct mbuf *m;
362 	int error;
363 	int len;
364 	int hlen;
365 	int slen;
366 
367 	/*
368 	 * Build a sockaddr based on the data link layer type.
369 	 * We do this at this level because the ethernet header
370 	 * is copied directly into the data field of the sockaddr.
371 	 * In the case of SLIP, there is no header and the packet
372 	 * is forwarded as is.
373 	 * Also, we are careful to leave room at the front of the mbuf
374 	 * for the link level header.
375 	 */
376 	switch (linktype) {
377 
378 	case DLT_SLIP:
379 		sockp->sa_family = AF_INET;
380 		hlen = 0;
381 		break;
382 
383 	case DLT_EN10MB:
384 		sockp->sa_family = AF_UNSPEC;
385 		/* XXX Would MAXLINKHDR be better? */
386 		hlen = ETHER_HDR_LEN;
387 		break;
388 
389 	case DLT_FDDI:
390 		sockp->sa_family = AF_IMPLINK;
391 		hlen = 0;
392 		break;
393 
394 	case DLT_RAW:
395 		sockp->sa_family = AF_UNSPEC;
396 		hlen = 0;
397 		break;
398 
399 	case DLT_NULL:
400 		/*
401 		 * null interface types require a 4 byte pseudo header which
402 		 * corresponds to the address family of the packet.
403 		 */
404 		sockp->sa_family = AF_UNSPEC;
405 		hlen = 4;
406 		break;
407 
408 	case DLT_ATM_RFC1483:
409 		/*
410 		 * en atm driver requires 4-byte atm pseudo header.
411 		 * though it isn't standard, vpi:vci needs to be
412 		 * specified anyway.
413 		 */
414 		sockp->sa_family = AF_UNSPEC;
415 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
416 		break;
417 
418 	case DLT_PPP:
419 		sockp->sa_family = AF_UNSPEC;
420 		hlen = 4;	/* This should match PPP_HDRLEN */
421 		break;
422 
423 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
424 		sockp->sa_family = AF_IEEE80211;
425 		hlen = 0;
426 		break;
427 
428 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
429 		sockp->sa_family = AF_IEEE80211;
430 		sockp->sa_len = 12;	/* XXX != 0 */
431 		hlen = sizeof(struct ieee80211_bpf_params);
432 		break;
433 
434 	default:
435 		return (EIO);
436 	}
437 
438 	len = uio->uio_resid;
439 
440 	if (len - hlen > ifp->if_mtu)
441 		return (EMSGSIZE);
442 
443 	if ((unsigned)len > MJUM16BYTES)
444 		return (EIO);
445 
446 	if (len <= MHLEN)
447 		MGETHDR(m, M_WAIT, MT_DATA);
448 	else if (len <= MCLBYTES)
449 		m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
450 	else
451 		m = m_getjcl(M_WAIT, MT_DATA, M_PKTHDR,
452 #if (MJUMPAGESIZE > MCLBYTES)
453 		    len <= MJUMPAGESIZE ? MJUMPAGESIZE :
454 #endif
455 		    (len <= MJUM9BYTES ? MJUM9BYTES : MJUM16BYTES));
456 	m->m_pkthdr.len = m->m_len = len;
457 	m->m_pkthdr.rcvif = NULL;
458 	*mp = m;
459 
460 	if (m->m_len < hlen) {
461 		error = EPERM;
462 		goto bad;
463 	}
464 
465 	error = uiomove(mtod(m, u_char *), len, uio);
466 	if (error)
467 		goto bad;
468 
469 	slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
470 	if (slen == 0) {
471 		error = EPERM;
472 		goto bad;
473 	}
474 
475 	/* Check for multicast destination */
476 	switch (linktype) {
477 	case DLT_EN10MB:
478 		eh = mtod(m, struct ether_header *);
479 		if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
480 			if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
481 			    ETHER_ADDR_LEN) == 0)
482 				m->m_flags |= M_BCAST;
483 			else
484 				m->m_flags |= M_MCAST;
485 		}
486 		break;
487 	}
488 
489 	/*
490 	 * Make room for link header, and copy it to sockaddr
491 	 */
492 	if (hlen != 0) {
493 		if (sockp->sa_family == AF_IEEE80211) {
494 			/*
495 			 * Collect true length from the parameter header
496 			 * NB: sockp is known to be zero'd so if we do a
497 			 *     short copy unspecified parameters will be
498 			 *     zero.
499 			 * NB: packet may not be aligned after stripping
500 			 *     bpf params
501 			 * XXX check ibp_vers
502 			 */
503 			p = mtod(m, const struct ieee80211_bpf_params *);
504 			hlen = p->ibp_len;
505 			if (hlen > sizeof(sockp->sa_data)) {
506 				error = EINVAL;
507 				goto bad;
508 			}
509 		}
510 		bcopy(m->m_data, sockp->sa_data, hlen);
511 	}
512 	*hdrlen = hlen;
513 
514 	return (0);
515 bad:
516 	m_freem(m);
517 	return (error);
518 }
519 
520 /*
521  * Attach file to the bpf interface, i.e. make d listen on bp.
522  */
523 static void
524 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
525 {
526 	/*
527 	 * Point d at bp, and add d to the interface's list of listeners.
528 	 * Finally, point the driver's bpf cookie at the interface so
529 	 * it will divert packets to bpf.
530 	 */
531 	BPFIF_LOCK(bp);
532 	d->bd_bif = bp;
533 	LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
534 
535 	bpf_bpfd_cnt++;
536 	BPFIF_UNLOCK(bp);
537 
538 	EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
539 }
540 
541 /*
542  * Detach a file from its interface.
543  */
544 static void
545 bpf_detachd(struct bpf_d *d)
546 {
547 	int error;
548 	struct bpf_if *bp;
549 	struct ifnet *ifp;
550 
551 	bp = d->bd_bif;
552 	BPFIF_LOCK(bp);
553 	BPFD_LOCK(d);
554 	ifp = d->bd_bif->bif_ifp;
555 
556 	/*
557 	 * Remove d from the interface's descriptor list.
558 	 */
559 	LIST_REMOVE(d, bd_next);
560 
561 	bpf_bpfd_cnt--;
562 	d->bd_bif = NULL;
563 	BPFD_UNLOCK(d);
564 	BPFIF_UNLOCK(bp);
565 
566 	EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
567 
568 	/*
569 	 * Check if this descriptor had requested promiscuous mode.
570 	 * If so, turn it off.
571 	 */
572 	if (d->bd_promisc) {
573 		d->bd_promisc = 0;
574 		CURVNET_SET(ifp->if_vnet);
575 		error = ifpromisc(ifp, 0);
576 		CURVNET_RESTORE();
577 		if (error != 0 && error != ENXIO) {
578 			/*
579 			 * ENXIO can happen if a pccard is unplugged
580 			 * Something is really wrong if we were able to put
581 			 * the driver into promiscuous mode, but can't
582 			 * take it out.
583 			 */
584 			if_printf(bp->bif_ifp,
585 				"bpf_detach: ifpromisc failed (%d)\n", error);
586 		}
587 	}
588 }
589 
590 /*
591  * Close the descriptor by detaching it from its interface,
592  * deallocating its buffers, and marking it free.
593  */
594 static void
595 bpf_dtor(void *data)
596 {
597 	struct bpf_d *d = data;
598 
599 	BPFD_LOCK(d);
600 	if (d->bd_state == BPF_WAITING)
601 		callout_stop(&d->bd_callout);
602 	d->bd_state = BPF_IDLE;
603 	BPFD_UNLOCK(d);
604 	funsetown(&d->bd_sigio);
605 	mtx_lock(&bpf_mtx);
606 	if (d->bd_bif)
607 		bpf_detachd(d);
608 	mtx_unlock(&bpf_mtx);
609 	selwakeuppri(&d->bd_sel, PRINET);
610 #ifdef MAC
611 	mac_bpfdesc_destroy(d);
612 #endif /* MAC */
613 	knlist_destroy(&d->bd_sel.si_note);
614 	bpf_freed(d);
615 	free(d, M_BPF);
616 }
617 
618 /*
619  * Open ethernet device.  Returns ENXIO for illegal minor device number,
620  * EBUSY if file is open by another process.
621  */
622 /* ARGSUSED */
623 static	int
624 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
625 {
626 	struct bpf_d *d;
627 	int error;
628 
629 	d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
630 	error = devfs_set_cdevpriv(d, bpf_dtor);
631 	if (error != 0) {
632 		free(d, M_BPF);
633 		return (error);
634 	}
635 
636 	/*
637 	 * For historical reasons, perform a one-time initialization call to
638 	 * the buffer routines, even though we're not yet committed to a
639 	 * particular buffer method.
640 	 */
641 	bpf_buffer_init(d);
642 	d->bd_bufmode = BPF_BUFMODE_BUFFER;
643 	d->bd_sig = SIGIO;
644 	d->bd_direction = BPF_D_INOUT;
645 	d->bd_pid = td->td_proc->p_pid;
646 #ifdef MAC
647 	mac_bpfdesc_init(d);
648 	mac_bpfdesc_create(td->td_ucred, d);
649 #endif
650 	mtx_init(&d->bd_mtx, devtoname(dev), "bpf cdev lock", MTX_DEF);
651 	callout_init(&d->bd_callout, CALLOUT_MPSAFE);
652 	knlist_init_mtx(&d->bd_sel.si_note, &d->bd_mtx);
653 
654 	return (0);
655 }
656 
657 /*
658  *  bpfread - read next chunk of packets from buffers
659  */
660 static	int
661 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
662 {
663 	struct bpf_d *d;
664 	int timed_out;
665 	int error;
666 
667 	error = devfs_get_cdevpriv((void **)&d);
668 	if (error != 0)
669 		return (error);
670 
671 	/*
672 	 * Restrict application to use a buffer the same size as
673 	 * as kernel buffers.
674 	 */
675 	if (uio->uio_resid != d->bd_bufsize)
676 		return (EINVAL);
677 
678 	BPFD_LOCK(d);
679 	d->bd_pid = curthread->td_proc->p_pid;
680 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
681 		BPFD_UNLOCK(d);
682 		return (EOPNOTSUPP);
683 	}
684 	if (d->bd_state == BPF_WAITING)
685 		callout_stop(&d->bd_callout);
686 	timed_out = (d->bd_state == BPF_TIMED_OUT);
687 	d->bd_state = BPF_IDLE;
688 	/*
689 	 * If the hold buffer is empty, then do a timed sleep, which
690 	 * ends when the timeout expires or when enough packets
691 	 * have arrived to fill the store buffer.
692 	 */
693 	while (d->bd_hbuf == NULL) {
694 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
695 			/*
696 			 * A packet(s) either arrived since the previous
697 			 * read or arrived while we were asleep.
698 			 * Rotate the buffers and return what's here.
699 			 */
700 			ROTATE_BUFFERS(d);
701 			break;
702 		}
703 
704 		/*
705 		 * No data is available, check to see if the bpf device
706 		 * is still pointed at a real interface.  If not, return
707 		 * ENXIO so that the userland process knows to rebind
708 		 * it before using it again.
709 		 */
710 		if (d->bd_bif == NULL) {
711 			BPFD_UNLOCK(d);
712 			return (ENXIO);
713 		}
714 
715 		if (ioflag & O_NONBLOCK) {
716 			BPFD_UNLOCK(d);
717 			return (EWOULDBLOCK);
718 		}
719 		error = msleep(d, &d->bd_mtx, PRINET|PCATCH,
720 		     "bpf", d->bd_rtout);
721 		if (error == EINTR || error == ERESTART) {
722 			BPFD_UNLOCK(d);
723 			return (error);
724 		}
725 		if (error == EWOULDBLOCK) {
726 			/*
727 			 * On a timeout, return what's in the buffer,
728 			 * which may be nothing.  If there is something
729 			 * in the store buffer, we can rotate the buffers.
730 			 */
731 			if (d->bd_hbuf)
732 				/*
733 				 * We filled up the buffer in between
734 				 * getting the timeout and arriving
735 				 * here, so we don't need to rotate.
736 				 */
737 				break;
738 
739 			if (d->bd_slen == 0) {
740 				BPFD_UNLOCK(d);
741 				return (0);
742 			}
743 			ROTATE_BUFFERS(d);
744 			break;
745 		}
746 	}
747 	/*
748 	 * At this point, we know we have something in the hold slot.
749 	 */
750 	BPFD_UNLOCK(d);
751 
752 	/*
753 	 * Move data from hold buffer into user space.
754 	 * We know the entire buffer is transferred since
755 	 * we checked above that the read buffer is bpf_bufsize bytes.
756 	 *
757 	 * XXXRW: More synchronization needed here: what if a second thread
758 	 * issues a read on the same fd at the same time?  Don't want this
759 	 * getting invalidated.
760 	 */
761 	error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
762 
763 	BPFD_LOCK(d);
764 	d->bd_fbuf = d->bd_hbuf;
765 	d->bd_hbuf = NULL;
766 	d->bd_hlen = 0;
767 	bpf_buf_reclaimed(d);
768 	BPFD_UNLOCK(d);
769 
770 	return (error);
771 }
772 
773 /*
774  * If there are processes sleeping on this descriptor, wake them up.
775  */
776 static __inline void
777 bpf_wakeup(struct bpf_d *d)
778 {
779 
780 	BPFD_LOCK_ASSERT(d);
781 	if (d->bd_state == BPF_WAITING) {
782 		callout_stop(&d->bd_callout);
783 		d->bd_state = BPF_IDLE;
784 	}
785 	wakeup(d);
786 	if (d->bd_async && d->bd_sig && d->bd_sigio)
787 		pgsigio(&d->bd_sigio, d->bd_sig, 0);
788 
789 	selwakeuppri(&d->bd_sel, PRINET);
790 	KNOTE_LOCKED(&d->bd_sel.si_note, 0);
791 }
792 
793 static void
794 bpf_timed_out(void *arg)
795 {
796 	struct bpf_d *d = (struct bpf_d *)arg;
797 
798 	BPFD_LOCK(d);
799 	if (d->bd_state == BPF_WAITING) {
800 		d->bd_state = BPF_TIMED_OUT;
801 		if (d->bd_slen != 0)
802 			bpf_wakeup(d);
803 	}
804 	BPFD_UNLOCK(d);
805 }
806 
807 static int
808 bpf_ready(struct bpf_d *d)
809 {
810 
811 	BPFD_LOCK_ASSERT(d);
812 
813 	if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
814 		return (1);
815 	if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
816 	    d->bd_slen != 0)
817 		return (1);
818 	return (0);
819 }
820 
821 static int
822 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
823 {
824 	struct bpf_d *d;
825 	struct ifnet *ifp;
826 	struct mbuf *m, *mc;
827 	struct sockaddr dst;
828 	int error, hlen;
829 
830 	error = devfs_get_cdevpriv((void **)&d);
831 	if (error != 0)
832 		return (error);
833 
834 	d->bd_pid = curthread->td_proc->p_pid;
835 	d->bd_wcount++;
836 	if (d->bd_bif == NULL) {
837 		d->bd_wdcount++;
838 		return (ENXIO);
839 	}
840 
841 	ifp = d->bd_bif->bif_ifp;
842 
843 	if ((ifp->if_flags & IFF_UP) == 0) {
844 		d->bd_wdcount++;
845 		return (ENETDOWN);
846 	}
847 
848 	if (uio->uio_resid == 0) {
849 		d->bd_wdcount++;
850 		return (0);
851 	}
852 
853 	bzero(&dst, sizeof(dst));
854 	m = NULL;
855 	hlen = 0;
856 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
857 	    &m, &dst, &hlen, d->bd_wfilter);
858 	if (error) {
859 		d->bd_wdcount++;
860 		return (error);
861 	}
862 	d->bd_wfcount++;
863 	if (d->bd_hdrcmplt)
864 		dst.sa_family = pseudo_AF_HDRCMPLT;
865 
866 	if (d->bd_feedback) {
867 		mc = m_dup(m, M_DONTWAIT);
868 		if (mc != NULL)
869 			mc->m_pkthdr.rcvif = ifp;
870 		/* Set M_PROMISC for outgoing packets to be discarded. */
871 		if (d->bd_direction == BPF_D_INOUT)
872 			m->m_flags |= M_PROMISC;
873 	} else
874 		mc = NULL;
875 
876 	m->m_pkthdr.len -= hlen;
877 	m->m_len -= hlen;
878 	m->m_data += hlen;	/* XXX */
879 
880 	CURVNET_SET(ifp->if_vnet);
881 #ifdef MAC
882 	BPFD_LOCK(d);
883 	mac_bpfdesc_create_mbuf(d, m);
884 	if (mc != NULL)
885 		mac_bpfdesc_create_mbuf(d, mc);
886 	BPFD_UNLOCK(d);
887 #endif
888 
889 	error = (*ifp->if_output)(ifp, m, &dst, NULL);
890 	if (error)
891 		d->bd_wdcount++;
892 
893 	if (mc != NULL) {
894 		if (error == 0)
895 			(*ifp->if_input)(ifp, mc);
896 		else
897 			m_freem(mc);
898 	}
899 	CURVNET_RESTORE();
900 
901 	return (error);
902 }
903 
904 /*
905  * Reset a descriptor by flushing its packet buffer and clearing the receive
906  * and drop counts.  This is doable for kernel-only buffers, but with
907  * zero-copy buffers, we can't write to (or rotate) buffers that are
908  * currently owned by userspace.  It would be nice if we could encapsulate
909  * this logic in the buffer code rather than here.
910  */
911 static void
912 reset_d(struct bpf_d *d)
913 {
914 
915 	mtx_assert(&d->bd_mtx, MA_OWNED);
916 
917 	if ((d->bd_hbuf != NULL) &&
918 	    (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
919 		/* Free the hold buffer. */
920 		d->bd_fbuf = d->bd_hbuf;
921 		d->bd_hbuf = NULL;
922 		d->bd_hlen = 0;
923 		bpf_buf_reclaimed(d);
924 	}
925 	if (bpf_canwritebuf(d))
926 		d->bd_slen = 0;
927 	d->bd_rcount = 0;
928 	d->bd_dcount = 0;
929 	d->bd_fcount = 0;
930 	d->bd_wcount = 0;
931 	d->bd_wfcount = 0;
932 	d->bd_wdcount = 0;
933 	d->bd_zcopy = 0;
934 }
935 
936 /*
937  *  FIONREAD		Check for read packet available.
938  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
939  *  BIOCGBLEN		Get buffer len [for read()].
940  *  BIOCSETF		Set read filter.
941  *  BIOCSETFNR		Set read filter without resetting descriptor.
942  *  BIOCSETWF		Set write filter.
943  *  BIOCFLUSH		Flush read packet buffer.
944  *  BIOCPROMISC		Put interface into promiscuous mode.
945  *  BIOCGDLT		Get link layer type.
946  *  BIOCGETIF		Get interface name.
947  *  BIOCSETIF		Set interface.
948  *  BIOCSRTIMEOUT	Set read timeout.
949  *  BIOCGRTIMEOUT	Get read timeout.
950  *  BIOCGSTATS		Get packet stats.
951  *  BIOCIMMEDIATE	Set immediate mode.
952  *  BIOCVERSION		Get filter language version.
953  *  BIOCGHDRCMPLT	Get "header already complete" flag
954  *  BIOCSHDRCMPLT	Set "header already complete" flag
955  *  BIOCGDIRECTION	Get packet direction flag
956  *  BIOCSDIRECTION	Set packet direction flag
957  *  BIOCLOCK		Set "locked" flag
958  *  BIOCFEEDBACK	Set packet feedback mode.
959  *  BIOCSETZBUF		Set current zero-copy buffer locations.
960  *  BIOCGETZMAX		Get maximum zero-copy buffer size.
961  *  BIOCROTZBUF		Force rotation of zero-copy buffer
962  *  BIOCSETBUFMODE	Set buffer mode.
963  *  BIOCGETBUFMODE	Get current buffer mode.
964  */
965 /* ARGSUSED */
966 static	int
967 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
968     struct thread *td)
969 {
970 	struct bpf_d *d;
971 	int error;
972 
973 	error = devfs_get_cdevpriv((void **)&d);
974 	if (error != 0)
975 		return (error);
976 
977 	/*
978 	 * Refresh PID associated with this descriptor.
979 	 */
980 	BPFD_LOCK(d);
981 	d->bd_pid = td->td_proc->p_pid;
982 	if (d->bd_state == BPF_WAITING)
983 		callout_stop(&d->bd_callout);
984 	d->bd_state = BPF_IDLE;
985 	BPFD_UNLOCK(d);
986 
987 	if (d->bd_locked == 1) {
988 		switch (cmd) {
989 		case BIOCGBLEN:
990 		case BIOCFLUSH:
991 		case BIOCGDLT:
992 		case BIOCGDLTLIST:
993 		case BIOCGETIF:
994 		case BIOCGRTIMEOUT:
995 		case BIOCGSTATS:
996 		case BIOCVERSION:
997 		case BIOCGRSIG:
998 		case BIOCGHDRCMPLT:
999 		case BIOCFEEDBACK:
1000 		case FIONREAD:
1001 		case BIOCLOCK:
1002 		case BIOCSRTIMEOUT:
1003 		case BIOCIMMEDIATE:
1004 		case TIOCGPGRP:
1005 		case BIOCROTZBUF:
1006 			break;
1007 		default:
1008 			return (EPERM);
1009 		}
1010 	}
1011 	CURVNET_SET(TD_TO_VNET(td));
1012 	switch (cmd) {
1013 
1014 	default:
1015 		error = EINVAL;
1016 		break;
1017 
1018 	/*
1019 	 * Check for read packet available.
1020 	 */
1021 	case FIONREAD:
1022 		{
1023 			int n;
1024 
1025 			BPFD_LOCK(d);
1026 			n = d->bd_slen;
1027 			if (d->bd_hbuf)
1028 				n += d->bd_hlen;
1029 			BPFD_UNLOCK(d);
1030 
1031 			*(int *)addr = n;
1032 			break;
1033 		}
1034 
1035 	case SIOCGIFADDR:
1036 		{
1037 			struct ifnet *ifp;
1038 
1039 			if (d->bd_bif == NULL)
1040 				error = EINVAL;
1041 			else {
1042 				ifp = d->bd_bif->bif_ifp;
1043 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
1044 			}
1045 			break;
1046 		}
1047 
1048 	/*
1049 	 * Get buffer len [for read()].
1050 	 */
1051 	case BIOCGBLEN:
1052 		*(u_int *)addr = d->bd_bufsize;
1053 		break;
1054 
1055 	/*
1056 	 * Set buffer length.
1057 	 */
1058 	case BIOCSBLEN:
1059 		error = bpf_ioctl_sblen(d, (u_int *)addr);
1060 		break;
1061 
1062 	/*
1063 	 * Set link layer read filter.
1064 	 */
1065 	case BIOCSETF:
1066 	case BIOCSETFNR:
1067 	case BIOCSETWF:
1068 		error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1069 		break;
1070 
1071 	/*
1072 	 * Flush read packet buffer.
1073 	 */
1074 	case BIOCFLUSH:
1075 		BPFD_LOCK(d);
1076 		reset_d(d);
1077 		BPFD_UNLOCK(d);
1078 		break;
1079 
1080 	/*
1081 	 * Put interface into promiscuous mode.
1082 	 */
1083 	case BIOCPROMISC:
1084 		if (d->bd_bif == NULL) {
1085 			/*
1086 			 * No interface attached yet.
1087 			 */
1088 			error = EINVAL;
1089 			break;
1090 		}
1091 		if (d->bd_promisc == 0) {
1092 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1093 			if (error == 0)
1094 				d->bd_promisc = 1;
1095 		}
1096 		break;
1097 
1098 	/*
1099 	 * Get current data link type.
1100 	 */
1101 	case BIOCGDLT:
1102 		if (d->bd_bif == NULL)
1103 			error = EINVAL;
1104 		else
1105 			*(u_int *)addr = d->bd_bif->bif_dlt;
1106 		break;
1107 
1108 	/*
1109 	 * Get a list of supported data link types.
1110 	 */
1111 	case BIOCGDLTLIST:
1112 		if (d->bd_bif == NULL)
1113 			error = EINVAL;
1114 		else
1115 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1116 		break;
1117 
1118 	/*
1119 	 * Set data link type.
1120 	 */
1121 	case BIOCSDLT:
1122 		if (d->bd_bif == NULL)
1123 			error = EINVAL;
1124 		else
1125 			error = bpf_setdlt(d, *(u_int *)addr);
1126 		break;
1127 
1128 	/*
1129 	 * Get interface name.
1130 	 */
1131 	case BIOCGETIF:
1132 		if (d->bd_bif == NULL)
1133 			error = EINVAL;
1134 		else {
1135 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
1136 			struct ifreq *const ifr = (struct ifreq *)addr;
1137 
1138 			strlcpy(ifr->ifr_name, ifp->if_xname,
1139 			    sizeof(ifr->ifr_name));
1140 		}
1141 		break;
1142 
1143 	/*
1144 	 * Set interface.
1145 	 */
1146 	case BIOCSETIF:
1147 		error = bpf_setif(d, (struct ifreq *)addr);
1148 		break;
1149 
1150 	/*
1151 	 * Set read timeout.
1152 	 */
1153 	case BIOCSRTIMEOUT:
1154 		{
1155 			struct timeval *tv = (struct timeval *)addr;
1156 
1157 			/*
1158 			 * Subtract 1 tick from tvtohz() since this isn't
1159 			 * a one-shot timer.
1160 			 */
1161 			if ((error = itimerfix(tv)) == 0)
1162 				d->bd_rtout = tvtohz(tv) - 1;
1163 			break;
1164 		}
1165 
1166 	/*
1167 	 * Get read timeout.
1168 	 */
1169 	case BIOCGRTIMEOUT:
1170 		{
1171 			struct timeval *tv = (struct timeval *)addr;
1172 
1173 			tv->tv_sec = d->bd_rtout / hz;
1174 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1175 			break;
1176 		}
1177 
1178 	/*
1179 	 * Get packet stats.
1180 	 */
1181 	case BIOCGSTATS:
1182 		{
1183 			struct bpf_stat *bs = (struct bpf_stat *)addr;
1184 
1185 			/* XXXCSJP overflow */
1186 			bs->bs_recv = d->bd_rcount;
1187 			bs->bs_drop = d->bd_dcount;
1188 			break;
1189 		}
1190 
1191 	/*
1192 	 * Set immediate mode.
1193 	 */
1194 	case BIOCIMMEDIATE:
1195 		d->bd_immediate = *(u_int *)addr;
1196 		break;
1197 
1198 	case BIOCVERSION:
1199 		{
1200 			struct bpf_version *bv = (struct bpf_version *)addr;
1201 
1202 			bv->bv_major = BPF_MAJOR_VERSION;
1203 			bv->bv_minor = BPF_MINOR_VERSION;
1204 			break;
1205 		}
1206 
1207 	/*
1208 	 * Get "header already complete" flag
1209 	 */
1210 	case BIOCGHDRCMPLT:
1211 		*(u_int *)addr = d->bd_hdrcmplt;
1212 		break;
1213 
1214 	/*
1215 	 * Set "header already complete" flag
1216 	 */
1217 	case BIOCSHDRCMPLT:
1218 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1219 		break;
1220 
1221 	/*
1222 	 * Get packet direction flag
1223 	 */
1224 	case BIOCGDIRECTION:
1225 		*(u_int *)addr = d->bd_direction;
1226 		break;
1227 
1228 	/*
1229 	 * Set packet direction flag
1230 	 */
1231 	case BIOCSDIRECTION:
1232 		{
1233 			u_int	direction;
1234 
1235 			direction = *(u_int *)addr;
1236 			switch (direction) {
1237 			case BPF_D_IN:
1238 			case BPF_D_INOUT:
1239 			case BPF_D_OUT:
1240 				d->bd_direction = direction;
1241 				break;
1242 			default:
1243 				error = EINVAL;
1244 			}
1245 		}
1246 		break;
1247 
1248 	case BIOCFEEDBACK:
1249 		d->bd_feedback = *(u_int *)addr;
1250 		break;
1251 
1252 	case BIOCLOCK:
1253 		d->bd_locked = 1;
1254 		break;
1255 
1256 	case FIONBIO:		/* Non-blocking I/O */
1257 		break;
1258 
1259 	case FIOASYNC:		/* Send signal on receive packets */
1260 		d->bd_async = *(int *)addr;
1261 		break;
1262 
1263 	case FIOSETOWN:
1264 		error = fsetown(*(int *)addr, &d->bd_sigio);
1265 		break;
1266 
1267 	case FIOGETOWN:
1268 		*(int *)addr = fgetown(&d->bd_sigio);
1269 		break;
1270 
1271 	/* This is deprecated, FIOSETOWN should be used instead. */
1272 	case TIOCSPGRP:
1273 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
1274 		break;
1275 
1276 	/* This is deprecated, FIOGETOWN should be used instead. */
1277 	case TIOCGPGRP:
1278 		*(int *)addr = -fgetown(&d->bd_sigio);
1279 		break;
1280 
1281 	case BIOCSRSIG:		/* Set receive signal */
1282 		{
1283 			u_int sig;
1284 
1285 			sig = *(u_int *)addr;
1286 
1287 			if (sig >= NSIG)
1288 				error = EINVAL;
1289 			else
1290 				d->bd_sig = sig;
1291 			break;
1292 		}
1293 	case BIOCGRSIG:
1294 		*(u_int *)addr = d->bd_sig;
1295 		break;
1296 
1297 	case BIOCGETBUFMODE:
1298 		*(u_int *)addr = d->bd_bufmode;
1299 		break;
1300 
1301 	case BIOCSETBUFMODE:
1302 		/*
1303 		 * Allow the buffering mode to be changed as long as we
1304 		 * haven't yet committed to a particular mode.  Our
1305 		 * definition of commitment, for now, is whether or not a
1306 		 * buffer has been allocated or an interface attached, since
1307 		 * that's the point where things get tricky.
1308 		 */
1309 		switch (*(u_int *)addr) {
1310 		case BPF_BUFMODE_BUFFER:
1311 			break;
1312 
1313 		case BPF_BUFMODE_ZBUF:
1314 			if (bpf_zerocopy_enable)
1315 				break;
1316 			/* FALLSTHROUGH */
1317 
1318 		default:
1319 			return (EINVAL);
1320 		}
1321 
1322 		BPFD_LOCK(d);
1323 		if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1324 		    d->bd_fbuf != NULL || d->bd_bif != NULL) {
1325 			BPFD_UNLOCK(d);
1326 			return (EBUSY);
1327 		}
1328 		d->bd_bufmode = *(u_int *)addr;
1329 		BPFD_UNLOCK(d);
1330 		break;
1331 
1332 	case BIOCGETZMAX:
1333 		return (bpf_ioctl_getzmax(td, d, (size_t *)addr));
1334 
1335 	case BIOCSETZBUF:
1336 		return (bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr));
1337 
1338 	case BIOCROTZBUF:
1339 		return (bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr));
1340 	}
1341 	CURVNET_RESTORE();
1342 	return (error);
1343 }
1344 
1345 /*
1346  * Set d's packet filter program to fp.  If this file already has a filter,
1347  * free it and replace it.  Returns EINVAL for bogus requests.
1348  */
1349 static int
1350 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1351 {
1352 	struct bpf_insn *fcode, *old;
1353 	u_int wfilter, flen, size;
1354 #ifdef BPF_JITTER
1355 	bpf_jit_filter *ofunc;
1356 #endif
1357 
1358 	if (cmd == BIOCSETWF) {
1359 		old = d->bd_wfilter;
1360 		wfilter = 1;
1361 #ifdef BPF_JITTER
1362 		ofunc = NULL;
1363 #endif
1364 	} else {
1365 		wfilter = 0;
1366 		old = d->bd_rfilter;
1367 #ifdef BPF_JITTER
1368 		ofunc = d->bd_bfilter;
1369 #endif
1370 	}
1371 	if (fp->bf_insns == NULL) {
1372 		if (fp->bf_len != 0)
1373 			return (EINVAL);
1374 		BPFD_LOCK(d);
1375 		if (wfilter)
1376 			d->bd_wfilter = NULL;
1377 		else {
1378 			d->bd_rfilter = NULL;
1379 #ifdef BPF_JITTER
1380 			d->bd_bfilter = NULL;
1381 #endif
1382 			if (cmd == BIOCSETF)
1383 				reset_d(d);
1384 		}
1385 		BPFD_UNLOCK(d);
1386 		if (old != NULL)
1387 			free((caddr_t)old, M_BPF);
1388 #ifdef BPF_JITTER
1389 		if (ofunc != NULL)
1390 			bpf_destroy_jit_filter(ofunc);
1391 #endif
1392 		return (0);
1393 	}
1394 	flen = fp->bf_len;
1395 	if (flen > bpf_maxinsns)
1396 		return (EINVAL);
1397 
1398 	size = flen * sizeof(*fp->bf_insns);
1399 	fcode = (struct bpf_insn *)malloc(size, M_BPF, M_WAITOK);
1400 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
1401 	    bpf_validate(fcode, (int)flen)) {
1402 		BPFD_LOCK(d);
1403 		if (wfilter)
1404 			d->bd_wfilter = fcode;
1405 		else {
1406 			d->bd_rfilter = fcode;
1407 #ifdef BPF_JITTER
1408 			d->bd_bfilter = bpf_jitter(fcode, flen);
1409 #endif
1410 			if (cmd == BIOCSETF)
1411 				reset_d(d);
1412 		}
1413 		BPFD_UNLOCK(d);
1414 		if (old != NULL)
1415 			free((caddr_t)old, M_BPF);
1416 #ifdef BPF_JITTER
1417 		if (ofunc != NULL)
1418 			bpf_destroy_jit_filter(ofunc);
1419 #endif
1420 
1421 		return (0);
1422 	}
1423 	free((caddr_t)fcode, M_BPF);
1424 	return (EINVAL);
1425 }
1426 
1427 /*
1428  * Detach a file from its current interface (if attached at all) and attach
1429  * to the interface indicated by the name stored in ifr.
1430  * Return an errno or 0.
1431  */
1432 static int
1433 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1434 {
1435 	struct bpf_if *bp;
1436 	struct ifnet *theywant;
1437 
1438 	theywant = ifunit(ifr->ifr_name);
1439 	if (theywant == NULL || theywant->if_bpf == NULL)
1440 		return (ENXIO);
1441 
1442 	bp = theywant->if_bpf;
1443 
1444 	/*
1445 	 * Behavior here depends on the buffering model.  If we're using
1446 	 * kernel memory buffers, then we can allocate them here.  If we're
1447 	 * using zero-copy, then the user process must have registered
1448 	 * buffers by the time we get here.  If not, return an error.
1449 	 *
1450 	 * XXXRW: There are locking issues here with multi-threaded use: what
1451 	 * if two threads try to set the interface at once?
1452 	 */
1453 	switch (d->bd_bufmode) {
1454 	case BPF_BUFMODE_BUFFER:
1455 		if (d->bd_sbuf == NULL)
1456 			bpf_buffer_alloc(d);
1457 		KASSERT(d->bd_sbuf != NULL, ("bpf_setif: bd_sbuf NULL"));
1458 		break;
1459 
1460 	case BPF_BUFMODE_ZBUF:
1461 		if (d->bd_sbuf == NULL)
1462 			return (EINVAL);
1463 		break;
1464 
1465 	default:
1466 		panic("bpf_setif: bufmode %d", d->bd_bufmode);
1467 	}
1468 	if (bp != d->bd_bif) {
1469 		if (d->bd_bif)
1470 			/*
1471 			 * Detach if attached to something else.
1472 			 */
1473 			bpf_detachd(d);
1474 
1475 		bpf_attachd(d, bp);
1476 	}
1477 	BPFD_LOCK(d);
1478 	reset_d(d);
1479 	BPFD_UNLOCK(d);
1480 	return (0);
1481 }
1482 
1483 /*
1484  * Support for select() and poll() system calls
1485  *
1486  * Return true iff the specific operation will not block indefinitely.
1487  * Otherwise, return false but make a note that a selwakeup() must be done.
1488  */
1489 static int
1490 bpfpoll(struct cdev *dev, int events, struct thread *td)
1491 {
1492 	struct bpf_d *d;
1493 	int revents;
1494 
1495 	if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1496 		return (events &
1497 		    (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1498 
1499 	/*
1500 	 * Refresh PID associated with this descriptor.
1501 	 */
1502 	revents = events & (POLLOUT | POLLWRNORM);
1503 	BPFD_LOCK(d);
1504 	d->bd_pid = td->td_proc->p_pid;
1505 	if (events & (POLLIN | POLLRDNORM)) {
1506 		if (bpf_ready(d))
1507 			revents |= events & (POLLIN | POLLRDNORM);
1508 		else {
1509 			selrecord(td, &d->bd_sel);
1510 			/* Start the read timeout if necessary. */
1511 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1512 				callout_reset(&d->bd_callout, d->bd_rtout,
1513 				    bpf_timed_out, d);
1514 				d->bd_state = BPF_WAITING;
1515 			}
1516 		}
1517 	}
1518 	BPFD_UNLOCK(d);
1519 	return (revents);
1520 }
1521 
1522 /*
1523  * Support for kevent() system call.  Register EVFILT_READ filters and
1524  * reject all others.
1525  */
1526 int
1527 bpfkqfilter(struct cdev *dev, struct knote *kn)
1528 {
1529 	struct bpf_d *d;
1530 
1531 	if (devfs_get_cdevpriv((void **)&d) != 0 ||
1532 	    kn->kn_filter != EVFILT_READ)
1533 		return (1);
1534 
1535 	/*
1536 	 * Refresh PID associated with this descriptor.
1537 	 */
1538 	BPFD_LOCK(d);
1539 	d->bd_pid = curthread->td_proc->p_pid;
1540 	kn->kn_fop = &bpfread_filtops;
1541 	kn->kn_hook = d;
1542 	knlist_add(&d->bd_sel.si_note, kn, 1);
1543 	BPFD_UNLOCK(d);
1544 
1545 	return (0);
1546 }
1547 
1548 static void
1549 filt_bpfdetach(struct knote *kn)
1550 {
1551 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1552 
1553 	knlist_remove(&d->bd_sel.si_note, kn, 0);
1554 }
1555 
1556 static int
1557 filt_bpfread(struct knote *kn, long hint)
1558 {
1559 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1560 	int ready;
1561 
1562 	BPFD_LOCK_ASSERT(d);
1563 	ready = bpf_ready(d);
1564 	if (ready) {
1565 		kn->kn_data = d->bd_slen;
1566 		if (d->bd_hbuf)
1567 			kn->kn_data += d->bd_hlen;
1568 	}
1569 	else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1570 		callout_reset(&d->bd_callout, d->bd_rtout,
1571 		    bpf_timed_out, d);
1572 		d->bd_state = BPF_WAITING;
1573 	}
1574 
1575 	return (ready);
1576 }
1577 
1578 /*
1579  * Incoming linkage from device drivers.  Process the packet pkt, of length
1580  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1581  * by each process' filter, and if accepted, stashed into the corresponding
1582  * buffer.
1583  */
1584 void
1585 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1586 {
1587 	struct bpf_d *d;
1588 #ifdef BPF_JITTER
1589 	bpf_jit_filter *bf;
1590 #endif
1591 	u_int slen;
1592 	int gottime;
1593 	struct timeval tv;
1594 
1595 	gottime = 0;
1596 	BPFIF_LOCK(bp);
1597 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1598 		BPFD_LOCK(d);
1599 		++d->bd_rcount;
1600 		/*
1601 		 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
1602 		 * way for the caller to indiciate to us whether this packet
1603 		 * is inbound or outbound.  In the bpf_mtap() routines, we use
1604 		 * the interface pointers on the mbuf to figure it out.
1605 		 */
1606 #ifdef BPF_JITTER
1607 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
1608 		if (bf != NULL)
1609 			slen = (*(bf->func))(pkt, pktlen, pktlen);
1610 		else
1611 #endif
1612 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1613 		if (slen != 0) {
1614 			d->bd_fcount++;
1615 			if (!gottime) {
1616 				microtime(&tv);
1617 				gottime = 1;
1618 			}
1619 #ifdef MAC
1620 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1621 #endif
1622 				catchpacket(d, pkt, pktlen, slen,
1623 				    bpf_append_bytes, &tv);
1624 		}
1625 		BPFD_UNLOCK(d);
1626 	}
1627 	BPFIF_UNLOCK(bp);
1628 }
1629 
1630 #define	BPF_CHECK_DIRECTION(d, r, i)				\
1631 	    (((d)->bd_direction == BPF_D_IN && (r) != (i)) ||	\
1632 	    ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
1633 
1634 /*
1635  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1636  */
1637 void
1638 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1639 {
1640 	struct bpf_d *d;
1641 #ifdef BPF_JITTER
1642 	bpf_jit_filter *bf;
1643 #endif
1644 	u_int pktlen, slen;
1645 	int gottime;
1646 	struct timeval tv;
1647 
1648 	/* Skip outgoing duplicate packets. */
1649 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1650 		m->m_flags &= ~M_PROMISC;
1651 		return;
1652 	}
1653 
1654 	gottime = 0;
1655 
1656 	pktlen = m_length(m, NULL);
1657 
1658 	BPFIF_LOCK(bp);
1659 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1660 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
1661 			continue;
1662 		BPFD_LOCK(d);
1663 		++d->bd_rcount;
1664 #ifdef BPF_JITTER
1665 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
1666 		/* XXX We cannot handle multiple mbufs. */
1667 		if (bf != NULL && m->m_next == NULL)
1668 			slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
1669 		else
1670 #endif
1671 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1672 		if (slen != 0) {
1673 			d->bd_fcount++;
1674 			if (!gottime) {
1675 				microtime(&tv);
1676 				gottime = 1;
1677 			}
1678 #ifdef MAC
1679 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1680 #endif
1681 				catchpacket(d, (u_char *)m, pktlen, slen,
1682 				    bpf_append_mbuf, &tv);
1683 		}
1684 		BPFD_UNLOCK(d);
1685 	}
1686 	BPFIF_UNLOCK(bp);
1687 }
1688 
1689 /*
1690  * Incoming linkage from device drivers, when packet is in
1691  * an mbuf chain and to be prepended by a contiguous header.
1692  */
1693 void
1694 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1695 {
1696 	struct mbuf mb;
1697 	struct bpf_d *d;
1698 	u_int pktlen, slen;
1699 	int gottime;
1700 	struct timeval tv;
1701 
1702 	/* Skip outgoing duplicate packets. */
1703 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1704 		m->m_flags &= ~M_PROMISC;
1705 		return;
1706 	}
1707 
1708 	gottime = 0;
1709 
1710 	pktlen = m_length(m, NULL);
1711 	/*
1712 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1713 	 * Note that we cut corners here; we only setup what's
1714 	 * absolutely needed--this mbuf should never go anywhere else.
1715 	 */
1716 	mb.m_next = m;
1717 	mb.m_data = data;
1718 	mb.m_len = dlen;
1719 	pktlen += dlen;
1720 
1721 	BPFIF_LOCK(bp);
1722 	LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
1723 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
1724 			continue;
1725 		BPFD_LOCK(d);
1726 		++d->bd_rcount;
1727 		slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
1728 		if (slen != 0) {
1729 			d->bd_fcount++;
1730 			if (!gottime) {
1731 				microtime(&tv);
1732 				gottime = 1;
1733 			}
1734 #ifdef MAC
1735 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
1736 #endif
1737 				catchpacket(d, (u_char *)&mb, pktlen, slen,
1738 				    bpf_append_mbuf, &tv);
1739 		}
1740 		BPFD_UNLOCK(d);
1741 	}
1742 	BPFIF_UNLOCK(bp);
1743 }
1744 
1745 #undef	BPF_CHECK_DIRECTION
1746 
1747 /*
1748  * Move the packet data from interface memory (pkt) into the
1749  * store buffer.  "cpfn" is the routine called to do the actual data
1750  * transfer.  bcopy is passed in to copy contiguous chunks, while
1751  * bpf_append_mbuf is passed in to copy mbuf chains.  In the latter case,
1752  * pkt is really an mbuf.
1753  */
1754 static void
1755 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1756     void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
1757     struct timeval *tv)
1758 {
1759 	struct bpf_hdr hdr;
1760 	int totlen, curlen;
1761 	int hdrlen = d->bd_bif->bif_hdrlen;
1762 	int do_wakeup = 0;
1763 
1764 	BPFD_LOCK_ASSERT(d);
1765 
1766 	/*
1767 	 * Detect whether user space has released a buffer back to us, and if
1768 	 * so, move it from being a hold buffer to a free buffer.  This may
1769 	 * not be the best place to do it (for example, we might only want to
1770 	 * run this check if we need the space), but for now it's a reliable
1771 	 * spot to do it.
1772 	 */
1773 	if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
1774 		d->bd_fbuf = d->bd_hbuf;
1775 		d->bd_hbuf = NULL;
1776 		d->bd_hlen = 0;
1777 		bpf_buf_reclaimed(d);
1778 	}
1779 
1780 	/*
1781 	 * Figure out how many bytes to move.  If the packet is
1782 	 * greater or equal to the snapshot length, transfer that
1783 	 * much.  Otherwise, transfer the whole packet (unless
1784 	 * we hit the buffer size limit).
1785 	 */
1786 	totlen = hdrlen + min(snaplen, pktlen);
1787 	if (totlen > d->bd_bufsize)
1788 		totlen = d->bd_bufsize;
1789 
1790 	/*
1791 	 * Round up the end of the previous packet to the next longword.
1792 	 *
1793 	 * Drop the packet if there's no room and no hope of room
1794 	 * If the packet would overflow the storage buffer or the storage
1795 	 * buffer is considered immutable by the buffer model, try to rotate
1796 	 * the buffer and wakeup pending processes.
1797 	 */
1798 	curlen = BPF_WORDALIGN(d->bd_slen);
1799 	if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
1800 		if (d->bd_fbuf == NULL) {
1801 			/*
1802 			 * There's no room in the store buffer, and no
1803 			 * prospect of room, so drop the packet.  Notify the
1804 			 * buffer model.
1805 			 */
1806 			bpf_buffull(d);
1807 			++d->bd_dcount;
1808 			return;
1809 		}
1810 		ROTATE_BUFFERS(d);
1811 		do_wakeup = 1;
1812 		curlen = 0;
1813 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
1814 		/*
1815 		 * Immediate mode is set, or the read timeout has already
1816 		 * expired during a select call.  A packet arrived, so the
1817 		 * reader should be woken up.
1818 		 */
1819 		do_wakeup = 1;
1820 
1821 	/*
1822 	 * Append the bpf header.  Note we append the actual header size, but
1823 	 * move forward the length of the header plus padding.
1824 	 */
1825 	bzero(&hdr, sizeof(hdr));
1826 	hdr.bh_tstamp = *tv;
1827 	hdr.bh_datalen = pktlen;
1828 	hdr.bh_hdrlen = hdrlen;
1829 	hdr.bh_caplen = totlen - hdrlen;
1830 	bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
1831 
1832 	/*
1833 	 * Copy the packet data into the store buffer and update its length.
1834 	 */
1835 	(*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, hdr.bh_caplen);
1836 	d->bd_slen = curlen + totlen;
1837 
1838 	if (do_wakeup)
1839 		bpf_wakeup(d);
1840 }
1841 
1842 /*
1843  * Free buffers currently in use by a descriptor.
1844  * Called on close.
1845  */
1846 static void
1847 bpf_freed(struct bpf_d *d)
1848 {
1849 
1850 	/*
1851 	 * We don't need to lock out interrupts since this descriptor has
1852 	 * been detached from its interface and it yet hasn't been marked
1853 	 * free.
1854 	 */
1855 	bpf_free(d);
1856 	if (d->bd_rfilter) {
1857 		free((caddr_t)d->bd_rfilter, M_BPF);
1858 #ifdef BPF_JITTER
1859 		bpf_destroy_jit_filter(d->bd_bfilter);
1860 #endif
1861 	}
1862 	if (d->bd_wfilter)
1863 		free((caddr_t)d->bd_wfilter, M_BPF);
1864 	mtx_destroy(&d->bd_mtx);
1865 }
1866 
1867 /*
1868  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
1869  * fixed size of the link header (variable length headers not yet supported).
1870  */
1871 void
1872 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1873 {
1874 
1875 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
1876 }
1877 
1878 /*
1879  * Attach an interface to bpf.  ifp is a pointer to the structure
1880  * defining the interface to be attached, dlt is the link layer type,
1881  * and hdrlen is the fixed size of the link header (variable length
1882  * headers are not yet supporrted).
1883  */
1884 void
1885 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1886 {
1887 	struct bpf_if *bp;
1888 
1889 	bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
1890 	if (bp == NULL)
1891 		panic("bpfattach");
1892 
1893 	LIST_INIT(&bp->bif_dlist);
1894 	bp->bif_ifp = ifp;
1895 	bp->bif_dlt = dlt;
1896 	mtx_init(&bp->bif_mtx, "bpf interface lock", NULL, MTX_DEF);
1897 	KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
1898 	*driverp = bp;
1899 
1900 	mtx_lock(&bpf_mtx);
1901 	LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
1902 	mtx_unlock(&bpf_mtx);
1903 
1904 	/*
1905 	 * Compute the length of the bpf header.  This is not necessarily
1906 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1907 	 * that the network layer header begins on a longword boundary (for
1908 	 * performance reasons and to alleviate alignment restrictions).
1909 	 */
1910 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1911 
1912 	if (bootverbose)
1913 		if_printf(ifp, "bpf attached\n");
1914 }
1915 
1916 /*
1917  * Detach bpf from an interface.  This involves detaching each descriptor
1918  * associated with the interface, and leaving bd_bif NULL.  Notify each
1919  * descriptor as it's detached so that any sleepers wake up and get
1920  * ENXIO.
1921  */
1922 void
1923 bpfdetach(struct ifnet *ifp)
1924 {
1925 	struct bpf_if	*bp;
1926 	struct bpf_d	*d;
1927 
1928 	/* Locate BPF interface information */
1929 	mtx_lock(&bpf_mtx);
1930 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
1931 		if (ifp == bp->bif_ifp)
1932 			break;
1933 	}
1934 
1935 	/* Interface wasn't attached */
1936 	if ((bp == NULL) || (bp->bif_ifp == NULL)) {
1937 		mtx_unlock(&bpf_mtx);
1938 		printf("bpfdetach: %s was not attached\n", ifp->if_xname);
1939 		return;
1940 	}
1941 
1942 	LIST_REMOVE(bp, bif_next);
1943 	mtx_unlock(&bpf_mtx);
1944 
1945 	while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
1946 		bpf_detachd(d);
1947 		BPFD_LOCK(d);
1948 		bpf_wakeup(d);
1949 		BPFD_UNLOCK(d);
1950 	}
1951 
1952 	mtx_destroy(&bp->bif_mtx);
1953 	free(bp, M_BPF);
1954 }
1955 
1956 /*
1957  * Get a list of available data link type of the interface.
1958  */
1959 static int
1960 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1961 {
1962 	int n, error;
1963 	struct ifnet *ifp;
1964 	struct bpf_if *bp;
1965 
1966 	ifp = d->bd_bif->bif_ifp;
1967 	n = 0;
1968 	error = 0;
1969 	mtx_lock(&bpf_mtx);
1970 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
1971 		if (bp->bif_ifp != ifp)
1972 			continue;
1973 		if (bfl->bfl_list != NULL) {
1974 			if (n >= bfl->bfl_len) {
1975 				mtx_unlock(&bpf_mtx);
1976 				return (ENOMEM);
1977 			}
1978 			error = copyout(&bp->bif_dlt,
1979 			    bfl->bfl_list + n, sizeof(u_int));
1980 		}
1981 		n++;
1982 	}
1983 	mtx_unlock(&bpf_mtx);
1984 	bfl->bfl_len = n;
1985 	return (error);
1986 }
1987 
1988 /*
1989  * Set the data link type of a BPF instance.
1990  */
1991 static int
1992 bpf_setdlt(struct bpf_d *d, u_int dlt)
1993 {
1994 	int error, opromisc;
1995 	struct ifnet *ifp;
1996 	struct bpf_if *bp;
1997 
1998 	if (d->bd_bif->bif_dlt == dlt)
1999 		return (0);
2000 	ifp = d->bd_bif->bif_ifp;
2001 	mtx_lock(&bpf_mtx);
2002 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2003 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2004 			break;
2005 	}
2006 	mtx_unlock(&bpf_mtx);
2007 	if (bp != NULL) {
2008 		opromisc = d->bd_promisc;
2009 		bpf_detachd(d);
2010 		bpf_attachd(d, bp);
2011 		BPFD_LOCK(d);
2012 		reset_d(d);
2013 		BPFD_UNLOCK(d);
2014 		if (opromisc) {
2015 			error = ifpromisc(bp->bif_ifp, 1);
2016 			if (error)
2017 				if_printf(bp->bif_ifp,
2018 					"bpf_setdlt: ifpromisc failed (%d)\n",
2019 					error);
2020 			else
2021 				d->bd_promisc = 1;
2022 		}
2023 	}
2024 	return (bp == NULL ? EINVAL : 0);
2025 }
2026 
2027 static void
2028 bpf_drvinit(void *unused)
2029 {
2030 	struct cdev *dev;
2031 
2032 	mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2033 	LIST_INIT(&bpf_iflist);
2034 
2035 	dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2036 	/* For compatibility */
2037 	make_dev_alias(dev, "bpf0");
2038 
2039 }
2040 
2041 /*
2042  * Zero out the various packet counters associated with all of the bpf
2043  * descriptors.  At some point, we will probably want to get a bit more
2044  * granular and allow the user to specify descriptors to be zeroed.
2045  */
2046 static void
2047 bpf_zero_counters(void)
2048 {
2049 	struct bpf_if *bp;
2050 	struct bpf_d *bd;
2051 
2052 	mtx_lock(&bpf_mtx);
2053 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2054 		BPFIF_LOCK(bp);
2055 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2056 			BPFD_LOCK(bd);
2057 			bd->bd_rcount = 0;
2058 			bd->bd_dcount = 0;
2059 			bd->bd_fcount = 0;
2060 			bd->bd_wcount = 0;
2061 			bd->bd_wfcount = 0;
2062 			bd->bd_zcopy = 0;
2063 			BPFD_UNLOCK(bd);
2064 		}
2065 		BPFIF_UNLOCK(bp);
2066 	}
2067 	mtx_unlock(&bpf_mtx);
2068 }
2069 
2070 static void
2071 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2072 {
2073 
2074 	bzero(d, sizeof(*d));
2075 	BPFD_LOCK_ASSERT(bd);
2076 	d->bd_structsize = sizeof(*d);
2077 	d->bd_immediate = bd->bd_immediate;
2078 	d->bd_promisc = bd->bd_promisc;
2079 	d->bd_hdrcmplt = bd->bd_hdrcmplt;
2080 	d->bd_direction = bd->bd_direction;
2081 	d->bd_feedback = bd->bd_feedback;
2082 	d->bd_async = bd->bd_async;
2083 	d->bd_rcount = bd->bd_rcount;
2084 	d->bd_dcount = bd->bd_dcount;
2085 	d->bd_fcount = bd->bd_fcount;
2086 	d->bd_sig = bd->bd_sig;
2087 	d->bd_slen = bd->bd_slen;
2088 	d->bd_hlen = bd->bd_hlen;
2089 	d->bd_bufsize = bd->bd_bufsize;
2090 	d->bd_pid = bd->bd_pid;
2091 	strlcpy(d->bd_ifname,
2092 	    bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2093 	d->bd_locked = bd->bd_locked;
2094 	d->bd_wcount = bd->bd_wcount;
2095 	d->bd_wdcount = bd->bd_wdcount;
2096 	d->bd_wfcount = bd->bd_wfcount;
2097 	d->bd_zcopy = bd->bd_zcopy;
2098 	d->bd_bufmode = bd->bd_bufmode;
2099 }
2100 
2101 static int
2102 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2103 {
2104 	struct xbpf_d *xbdbuf, *xbd, zerostats;
2105 	int index, error;
2106 	struct bpf_if *bp;
2107 	struct bpf_d *bd;
2108 
2109 	/*
2110 	 * XXX This is not technically correct. It is possible for non
2111 	 * privileged users to open bpf devices. It would make sense
2112 	 * if the users who opened the devices were able to retrieve
2113 	 * the statistics for them, too.
2114 	 */
2115 	error = priv_check(req->td, PRIV_NET_BPF);
2116 	if (error)
2117 		return (error);
2118 	/*
2119 	 * Check to see if the user is requesting that the counters be
2120 	 * zeroed out.  Explicitly check that the supplied data is zeroed,
2121 	 * as we aren't allowing the user to set the counters currently.
2122 	 */
2123 	if (req->newptr != NULL) {
2124 		if (req->newlen != sizeof(zerostats))
2125 			return (EINVAL);
2126 		bzero(&zerostats, sizeof(zerostats));
2127 		xbd = req->newptr;
2128 		if (bcmp(xbd, &zerostats, sizeof(*xbd)) != 0)
2129 			return (EINVAL);
2130 		bpf_zero_counters();
2131 		return (0);
2132 	}
2133 	if (req->oldptr == NULL)
2134 		return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2135 	if (bpf_bpfd_cnt == 0)
2136 		return (SYSCTL_OUT(req, 0, 0));
2137 	xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2138 	mtx_lock(&bpf_mtx);
2139 	if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2140 		mtx_unlock(&bpf_mtx);
2141 		free(xbdbuf, M_BPF);
2142 		return (ENOMEM);
2143 	}
2144 	index = 0;
2145 	LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2146 		BPFIF_LOCK(bp);
2147 		LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2148 			xbd = &xbdbuf[index++];
2149 			BPFD_LOCK(bd);
2150 			bpfstats_fill_xbpf(xbd, bd);
2151 			BPFD_UNLOCK(bd);
2152 		}
2153 		BPFIF_UNLOCK(bp);
2154 	}
2155 	mtx_unlock(&bpf_mtx);
2156 	error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2157 	free(xbdbuf, M_BPF);
2158 	return (error);
2159 }
2160 
2161 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2162 
2163 #else /* !DEV_BPF && !NETGRAPH_BPF */
2164 /*
2165  * NOP stubs to allow bpf-using drivers to load and function.
2166  *
2167  * A 'better' implementation would allow the core bpf functionality
2168  * to be loaded at runtime.
2169  */
2170 static struct bpf_if bp_null;
2171 
2172 void
2173 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2174 {
2175 }
2176 
2177 void
2178 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2179 {
2180 }
2181 
2182 void
2183 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2184 {
2185 }
2186 
2187 void
2188 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2189 {
2190 
2191 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2192 }
2193 
2194 void
2195 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2196 {
2197 
2198 	*driverp = &bp_null;
2199 }
2200 
2201 void
2202 bpfdetach(struct ifnet *ifp)
2203 {
2204 }
2205 
2206 u_int
2207 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2208 {
2209 	return -1;	/* "no filter" behaviour */
2210 }
2211 
2212 int
2213 bpf_validate(const struct bpf_insn *f, int len)
2214 {
2215 	return 0;		/* false */
2216 }
2217 
2218 #endif /* !DEV_BPF && !NETGRAPH_BPF */
2219