xref: /freebsd/sys/net/bpf.c (revision 1603881667360c015f6685131f2f25474fa67a72)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1990, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * Copyright (c) 2019 Andrey V. Elsukov <ae@FreeBSD.org>
7  *
8  * This code is derived from the Stanford/CMU enet packet filter,
9  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
10  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
11  * Berkeley Laboratory.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *      @(#)bpf.c	8.4 (Berkeley) 1/9/95
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_bpf.h"
44 #include "opt_ddb.h"
45 #include "opt_netgraph.h"
46 
47 #include <sys/param.h>
48 #include <sys/conf.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
51 #include <sys/jail.h>
52 #include <sys/ktr.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/mutex.h>
57 #include <sys/time.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/signalvar.h>
61 #include <sys/filio.h>
62 #include <sys/sockio.h>
63 #include <sys/ttycom.h>
64 #include <sys/uio.h>
65 #include <sys/sysent.h>
66 #include <sys/systm.h>
67 
68 #include <sys/event.h>
69 #include <sys/file.h>
70 #include <sys/poll.h>
71 #include <sys/proc.h>
72 
73 #include <sys/socket.h>
74 
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #endif
78 
79 #include <net/if.h>
80 #include <net/if_var.h>
81 #include <net/if_dl.h>
82 #include <net/bpf.h>
83 #include <net/bpf_buffer.h>
84 #ifdef BPF_JITTER
85 #include <net/bpf_jitter.h>
86 #endif
87 #include <net/bpf_zerocopy.h>
88 #include <net/bpfdesc.h>
89 #include <net/route.h>
90 #include <net/vnet.h>
91 
92 #include <netinet/in.h>
93 #include <netinet/if_ether.h>
94 #include <sys/kernel.h>
95 #include <sys/sysctl.h>
96 
97 #include <net80211/ieee80211_freebsd.h>
98 
99 #include <security/mac/mac_framework.h>
100 
101 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
102 
103 static struct bpf_if_ext dead_bpf_if = {
104 	.bif_dlist = CK_LIST_HEAD_INITIALIZER()
105 };
106 
107 struct bpf_if {
108 #define	bif_next	bif_ext.bif_next
109 #define	bif_dlist	bif_ext.bif_dlist
110 	struct bpf_if_ext bif_ext;	/* public members */
111 	u_int		bif_dlt;	/* link layer type */
112 	u_int		bif_hdrlen;	/* length of link header */
113 	struct bpfd_list bif_wlist;	/* writer-only list */
114 	struct ifnet	*bif_ifp;	/* corresponding interface */
115 	struct bpf_if	**bif_bpf;	/* Pointer to pointer to us */
116 	volatile u_int	bif_refcnt;
117 	struct epoch_context epoch_ctx;
118 };
119 
120 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
121 
122 struct bpf_program_buffer {
123 	struct epoch_context	epoch_ctx;
124 #ifdef BPF_JITTER
125 	bpf_jit_filter		*func;
126 #endif
127 	void			*buffer[0];
128 };
129 
130 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
131 
132 #define PRINET  26			/* interruptible */
133 
134 #define	SIZEOF_BPF_HDR(type)	\
135     (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
136 
137 #ifdef COMPAT_FREEBSD32
138 #include <sys/mount.h>
139 #include <compat/freebsd32/freebsd32.h>
140 #define BPF_ALIGNMENT32 sizeof(int32_t)
141 #define	BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
142 
143 #ifndef BURN_BRIDGES
144 /*
145  * 32-bit version of structure prepended to each packet.  We use this header
146  * instead of the standard one for 32-bit streams.  We mark the a stream as
147  * 32-bit the first time we see a 32-bit compat ioctl request.
148  */
149 struct bpf_hdr32 {
150 	struct timeval32 bh_tstamp;	/* time stamp */
151 	uint32_t	bh_caplen;	/* length of captured portion */
152 	uint32_t	bh_datalen;	/* original length of packet */
153 	uint16_t	bh_hdrlen;	/* length of bpf header (this struct
154 					   plus alignment padding) */
155 };
156 #endif
157 
158 struct bpf_program32 {
159 	u_int bf_len;
160 	uint32_t bf_insns;
161 };
162 
163 struct bpf_dltlist32 {
164 	u_int	bfl_len;
165 	u_int	bfl_list;
166 };
167 
168 #define	BIOCSETF32	_IOW('B', 103, struct bpf_program32)
169 #define	BIOCSRTIMEOUT32	_IOW('B', 109, struct timeval32)
170 #define	BIOCGRTIMEOUT32	_IOR('B', 110, struct timeval32)
171 #define	BIOCGDLTLIST32	_IOWR('B', 121, struct bpf_dltlist32)
172 #define	BIOCSETWF32	_IOW('B', 123, struct bpf_program32)
173 #define	BIOCSETFNR32	_IOW('B', 130, struct bpf_program32)
174 #endif
175 
176 #define BPF_LOCK()	   sx_xlock(&bpf_sx)
177 #define BPF_UNLOCK()		sx_xunlock(&bpf_sx)
178 #define BPF_LOCK_ASSERT()	sx_assert(&bpf_sx, SA_XLOCKED)
179 /*
180  * bpf_iflist is a list of BPF interface structures, each corresponding to a
181  * specific DLT. The same network interface might have several BPF interface
182  * structures registered by different layers in the stack (i.e., 802.11
183  * frames, ethernet frames, etc).
184  */
185 CK_LIST_HEAD(bpf_iflist, bpf_if);
186 static struct bpf_iflist bpf_iflist;
187 static struct sx	bpf_sx;		/* bpf global lock */
188 static int		bpf_bpfd_cnt;
189 
190 static void	bpfif_ref(struct bpf_if *);
191 static void	bpfif_rele(struct bpf_if *);
192 
193 static void	bpfd_ref(struct bpf_d *);
194 static void	bpfd_rele(struct bpf_d *);
195 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
196 static void	bpf_detachd(struct bpf_d *);
197 static void	bpf_detachd_locked(struct bpf_d *, bool);
198 static void	bpfd_free(epoch_context_t);
199 static int	bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
200 		    struct sockaddr *, int *, struct bpf_d *);
201 static int	bpf_setif(struct bpf_d *, struct ifreq *);
202 static void	bpf_timed_out(void *);
203 static __inline void
204 		bpf_wakeup(struct bpf_d *);
205 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
206 		    void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
207 		    struct bintime *);
208 static void	reset_d(struct bpf_d *);
209 static int	bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
210 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
211 static int	bpf_setdlt(struct bpf_d *, u_int);
212 static void	filt_bpfdetach(struct knote *);
213 static int	filt_bpfread(struct knote *, long);
214 static void	bpf_drvinit(void *);
215 static int	bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
216 
217 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
218     "bpf sysctl");
219 int bpf_maxinsns = BPF_MAXINSNS;
220 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
221     &bpf_maxinsns, 0, "Maximum bpf program instructions");
222 static int bpf_zerocopy_enable = 0;
223 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
224     &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
225 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
226     bpf_stats_sysctl, "bpf statistics portal");
227 
228 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
229 #define	V_bpf_optimize_writers VNET(bpf_optimize_writers)
230 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RWTUN,
231     &VNET_NAME(bpf_optimize_writers), 0,
232     "Do not send packets until BPF program is set");
233 
234 static	d_open_t	bpfopen;
235 static	d_read_t	bpfread;
236 static	d_write_t	bpfwrite;
237 static	d_ioctl_t	bpfioctl;
238 static	d_poll_t	bpfpoll;
239 static	d_kqfilter_t	bpfkqfilter;
240 
241 static struct cdevsw bpf_cdevsw = {
242 	.d_version =	D_VERSION,
243 	.d_open =	bpfopen,
244 	.d_read =	bpfread,
245 	.d_write =	bpfwrite,
246 	.d_ioctl =	bpfioctl,
247 	.d_poll =	bpfpoll,
248 	.d_name =	"bpf",
249 	.d_kqfilter =	bpfkqfilter,
250 };
251 
252 static struct filterops bpfread_filtops = {
253 	.f_isfd = 1,
254 	.f_detach = filt_bpfdetach,
255 	.f_event = filt_bpfread,
256 };
257 
258 /*
259  * LOCKING MODEL USED BY BPF
260  *
261  * Locks:
262  * 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
263  * every bpf_iflist changes, serializes ioctl access to bpf descriptors.
264  * 2) Descriptor lock. Mutex, used to protect BPF buffers and various
265  * structure fields used by bpf_*tap* code.
266  *
267  * Lock order: global lock, then descriptor lock.
268  *
269  * There are several possible consumers:
270  *
271  * 1. The kernel registers interface pointer with bpfattach().
272  * Each call allocates new bpf_if structure, references ifnet pointer
273  * and links bpf_if into bpf_iflist chain. This is protected with global
274  * lock.
275  *
276  * 2. An userland application uses ioctl() call to bpf_d descriptor.
277  * All such call are serialized with global lock. BPF filters can be
278  * changed, but pointer to old filter will be freed using NET_EPOCH_CALL().
279  * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
280  * filter pointers, even if change will happen during bpf_tap execution.
281  * Destroying of bpf_d descriptor also is doing using NET_EPOCH_CALL().
282  *
283  * 3. An userland application can write packets into bpf_d descriptor.
284  * There we need to be sure, that ifnet won't disappear during bpfwrite().
285  *
286  * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
287  * bif_dlist is protected with net_epoch_preempt section. So, it should
288  * be safe to make access to bpf_d descriptor inside the section.
289  *
290  * 5. The kernel invokes bpfdetach() on interface destroying. All lists
291  * are modified with global lock held and actual free() is done using
292  * NET_EPOCH_CALL().
293  */
294 
295 static void
296 bpfif_free(epoch_context_t ctx)
297 {
298 	struct bpf_if *bp;
299 
300 	bp = __containerof(ctx, struct bpf_if, epoch_ctx);
301 	if_rele(bp->bif_ifp);
302 	free(bp, M_BPF);
303 }
304 
305 static void
306 bpfif_ref(struct bpf_if *bp)
307 {
308 
309 	refcount_acquire(&bp->bif_refcnt);
310 }
311 
312 static void
313 bpfif_rele(struct bpf_if *bp)
314 {
315 
316 	if (!refcount_release(&bp->bif_refcnt))
317 		return;
318 	NET_EPOCH_CALL(bpfif_free, &bp->epoch_ctx);
319 }
320 
321 static void
322 bpfd_ref(struct bpf_d *d)
323 {
324 
325 	refcount_acquire(&d->bd_refcnt);
326 }
327 
328 static void
329 bpfd_rele(struct bpf_d *d)
330 {
331 
332 	if (!refcount_release(&d->bd_refcnt))
333 		return;
334 	NET_EPOCH_CALL(bpfd_free, &d->epoch_ctx);
335 }
336 
337 static struct bpf_program_buffer*
338 bpf_program_buffer_alloc(size_t size, int flags)
339 {
340 
341 	return (malloc(sizeof(struct bpf_program_buffer) + size,
342 	    M_BPF, flags));
343 }
344 
345 static void
346 bpf_program_buffer_free(epoch_context_t ctx)
347 {
348 	struct bpf_program_buffer *ptr;
349 
350 	ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
351 #ifdef BPF_JITTER
352 	if (ptr->func != NULL)
353 		bpf_destroy_jit_filter(ptr->func);
354 #endif
355 	free(ptr, M_BPF);
356 }
357 
358 /*
359  * Wrapper functions for various buffering methods.  If the set of buffer
360  * modes expands, we will probably want to introduce a switch data structure
361  * similar to protosw, et.
362  */
363 static void
364 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
365     u_int len)
366 {
367 
368 	BPFD_LOCK_ASSERT(d);
369 
370 	switch (d->bd_bufmode) {
371 	case BPF_BUFMODE_BUFFER:
372 		return (bpf_buffer_append_bytes(d, buf, offset, src, len));
373 
374 	case BPF_BUFMODE_ZBUF:
375 		counter_u64_add(d->bd_zcopy, 1);
376 		return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
377 
378 	default:
379 		panic("bpf_buf_append_bytes");
380 	}
381 }
382 
383 static void
384 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
385     u_int len)
386 {
387 
388 	BPFD_LOCK_ASSERT(d);
389 
390 	switch (d->bd_bufmode) {
391 	case BPF_BUFMODE_BUFFER:
392 		return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
393 
394 	case BPF_BUFMODE_ZBUF:
395 		counter_u64_add(d->bd_zcopy, 1);
396 		return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
397 
398 	default:
399 		panic("bpf_buf_append_mbuf");
400 	}
401 }
402 
403 /*
404  * This function gets called when the free buffer is re-assigned.
405  */
406 static void
407 bpf_buf_reclaimed(struct bpf_d *d)
408 {
409 
410 	BPFD_LOCK_ASSERT(d);
411 
412 	switch (d->bd_bufmode) {
413 	case BPF_BUFMODE_BUFFER:
414 		return;
415 
416 	case BPF_BUFMODE_ZBUF:
417 		bpf_zerocopy_buf_reclaimed(d);
418 		return;
419 
420 	default:
421 		panic("bpf_buf_reclaimed");
422 	}
423 }
424 
425 /*
426  * If the buffer mechanism has a way to decide that a held buffer can be made
427  * free, then it is exposed via the bpf_canfreebuf() interface.  (1) is
428  * returned if the buffer can be discarded, (0) is returned if it cannot.
429  */
430 static int
431 bpf_canfreebuf(struct bpf_d *d)
432 {
433 
434 	BPFD_LOCK_ASSERT(d);
435 
436 	switch (d->bd_bufmode) {
437 	case BPF_BUFMODE_ZBUF:
438 		return (bpf_zerocopy_canfreebuf(d));
439 	}
440 	return (0);
441 }
442 
443 /*
444  * Allow the buffer model to indicate that the current store buffer is
445  * immutable, regardless of the appearance of space.  Return (1) if the
446  * buffer is writable, and (0) if not.
447  */
448 static int
449 bpf_canwritebuf(struct bpf_d *d)
450 {
451 	BPFD_LOCK_ASSERT(d);
452 
453 	switch (d->bd_bufmode) {
454 	case BPF_BUFMODE_ZBUF:
455 		return (bpf_zerocopy_canwritebuf(d));
456 	}
457 	return (1);
458 }
459 
460 /*
461  * Notify buffer model that an attempt to write to the store buffer has
462  * resulted in a dropped packet, in which case the buffer may be considered
463  * full.
464  */
465 static void
466 bpf_buffull(struct bpf_d *d)
467 {
468 
469 	BPFD_LOCK_ASSERT(d);
470 
471 	switch (d->bd_bufmode) {
472 	case BPF_BUFMODE_ZBUF:
473 		bpf_zerocopy_buffull(d);
474 		break;
475 	}
476 }
477 
478 /*
479  * Notify the buffer model that a buffer has moved into the hold position.
480  */
481 void
482 bpf_bufheld(struct bpf_d *d)
483 {
484 
485 	BPFD_LOCK_ASSERT(d);
486 
487 	switch (d->bd_bufmode) {
488 	case BPF_BUFMODE_ZBUF:
489 		bpf_zerocopy_bufheld(d);
490 		break;
491 	}
492 }
493 
494 static void
495 bpf_free(struct bpf_d *d)
496 {
497 
498 	switch (d->bd_bufmode) {
499 	case BPF_BUFMODE_BUFFER:
500 		return (bpf_buffer_free(d));
501 
502 	case BPF_BUFMODE_ZBUF:
503 		return (bpf_zerocopy_free(d));
504 
505 	default:
506 		panic("bpf_buf_free");
507 	}
508 }
509 
510 static int
511 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
512 {
513 
514 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
515 		return (EOPNOTSUPP);
516 	return (bpf_buffer_uiomove(d, buf, len, uio));
517 }
518 
519 static int
520 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
521 {
522 
523 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
524 		return (EOPNOTSUPP);
525 	return (bpf_buffer_ioctl_sblen(d, i));
526 }
527 
528 static int
529 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
530 {
531 
532 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
533 		return (EOPNOTSUPP);
534 	return (bpf_zerocopy_ioctl_getzmax(td, d, i));
535 }
536 
537 static int
538 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
539 {
540 
541 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
542 		return (EOPNOTSUPP);
543 	return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
544 }
545 
546 static int
547 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
548 {
549 
550 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
551 		return (EOPNOTSUPP);
552 	return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
553 }
554 
555 /*
556  * General BPF functions.
557  */
558 static int
559 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
560     struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
561 {
562 	const struct ieee80211_bpf_params *p;
563 	struct ether_header *eh;
564 	struct mbuf *m;
565 	int error;
566 	int len;
567 	int hlen;
568 	int slen;
569 
570 	/*
571 	 * Build a sockaddr based on the data link layer type.
572 	 * We do this at this level because the ethernet header
573 	 * is copied directly into the data field of the sockaddr.
574 	 * In the case of SLIP, there is no header and the packet
575 	 * is forwarded as is.
576 	 * Also, we are careful to leave room at the front of the mbuf
577 	 * for the link level header.
578 	 */
579 	switch (linktype) {
580 	case DLT_SLIP:
581 		sockp->sa_family = AF_INET;
582 		hlen = 0;
583 		break;
584 
585 	case DLT_EN10MB:
586 		sockp->sa_family = AF_UNSPEC;
587 		/* XXX Would MAXLINKHDR be better? */
588 		hlen = ETHER_HDR_LEN;
589 		break;
590 
591 	case DLT_FDDI:
592 		sockp->sa_family = AF_IMPLINK;
593 		hlen = 0;
594 		break;
595 
596 	case DLT_RAW:
597 		sockp->sa_family = AF_UNSPEC;
598 		hlen = 0;
599 		break;
600 
601 	case DLT_NULL:
602 		/*
603 		 * null interface types require a 4 byte pseudo header which
604 		 * corresponds to the address family of the packet.
605 		 */
606 		sockp->sa_family = AF_UNSPEC;
607 		hlen = 4;
608 		break;
609 
610 	case DLT_ATM_RFC1483:
611 		/*
612 		 * en atm driver requires 4-byte atm pseudo header.
613 		 * though it isn't standard, vpi:vci needs to be
614 		 * specified anyway.
615 		 */
616 		sockp->sa_family = AF_UNSPEC;
617 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
618 		break;
619 
620 	case DLT_PPP:
621 		sockp->sa_family = AF_UNSPEC;
622 		hlen = 4;	/* This should match PPP_HDRLEN */
623 		break;
624 
625 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
626 		sockp->sa_family = AF_IEEE80211;
627 		hlen = 0;
628 		break;
629 
630 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
631 		sockp->sa_family = AF_IEEE80211;
632 		sockp->sa_len = 12;	/* XXX != 0 */
633 		hlen = sizeof(struct ieee80211_bpf_params);
634 		break;
635 
636 	default:
637 		return (EIO);
638 	}
639 
640 	len = uio->uio_resid;
641 	if (len < hlen || len - hlen > ifp->if_mtu)
642 		return (EMSGSIZE);
643 
644 	m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
645 	if (m == NULL)
646 		return (EIO);
647 	m->m_pkthdr.len = m->m_len = len;
648 	*mp = m;
649 
650 	error = uiomove(mtod(m, u_char *), len, uio);
651 	if (error)
652 		goto bad;
653 
654 	slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
655 	if (slen == 0) {
656 		error = EPERM;
657 		goto bad;
658 	}
659 
660 	/* Check for multicast destination */
661 	switch (linktype) {
662 	case DLT_EN10MB:
663 		eh = mtod(m, struct ether_header *);
664 		if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
665 			if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
666 			    ETHER_ADDR_LEN) == 0)
667 				m->m_flags |= M_BCAST;
668 			else
669 				m->m_flags |= M_MCAST;
670 		}
671 		if (d->bd_hdrcmplt == 0) {
672 			memcpy(eh->ether_shost, IF_LLADDR(ifp),
673 			    sizeof(eh->ether_shost));
674 		}
675 		break;
676 	}
677 
678 	/*
679 	 * Make room for link header, and copy it to sockaddr
680 	 */
681 	if (hlen != 0) {
682 		if (sockp->sa_family == AF_IEEE80211) {
683 			/*
684 			 * Collect true length from the parameter header
685 			 * NB: sockp is known to be zero'd so if we do a
686 			 *     short copy unspecified parameters will be
687 			 *     zero.
688 			 * NB: packet may not be aligned after stripping
689 			 *     bpf params
690 			 * XXX check ibp_vers
691 			 */
692 			p = mtod(m, const struct ieee80211_bpf_params *);
693 			hlen = p->ibp_len;
694 			if (hlen > sizeof(sockp->sa_data)) {
695 				error = EINVAL;
696 				goto bad;
697 			}
698 		}
699 		bcopy(mtod(m, const void *), sockp->sa_data, hlen);
700 	}
701 	*hdrlen = hlen;
702 
703 	return (0);
704 bad:
705 	m_freem(m);
706 	return (error);
707 }
708 
709 /*
710  * Attach descriptor to the bpf interface, i.e. make d listen on bp,
711  * then reset its buffers and counters with reset_d().
712  */
713 static void
714 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
715 {
716 	int op_w;
717 
718 	BPF_LOCK_ASSERT();
719 
720 	/*
721 	 * Save sysctl value to protect from sysctl change
722 	 * between reads
723 	 */
724 	op_w = V_bpf_optimize_writers || d->bd_writer;
725 
726 	if (d->bd_bif != NULL)
727 		bpf_detachd_locked(d, false);
728 	/*
729 	 * Point d at bp, and add d to the interface's list.
730 	 * Since there are many applications using BPF for
731 	 * sending raw packets only (dhcpd, cdpd are good examples)
732 	 * we can delay adding d to the list of active listeners until
733 	 * some filter is configured.
734 	 */
735 
736 	BPFD_LOCK(d);
737 	/*
738 	 * Hold reference to bpif while descriptor uses this interface.
739 	 */
740 	bpfif_ref(bp);
741 	d->bd_bif = bp;
742 	if (op_w != 0) {
743 		/* Add to writers-only list */
744 		CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
745 		/*
746 		 * We decrement bd_writer on every filter set operation.
747 		 * First BIOCSETF is done by pcap_open_live() to set up
748 		 * snap length. After that appliation usually sets its own
749 		 * filter.
750 		 */
751 		d->bd_writer = 2;
752 	} else
753 		CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
754 
755 	reset_d(d);
756 	BPFD_UNLOCK(d);
757 	bpf_bpfd_cnt++;
758 
759 	CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
760 	    __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
761 
762 	if (op_w == 0)
763 		EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
764 }
765 
766 /*
767  * Check if we need to upgrade our descriptor @d from write-only mode.
768  */
769 static int
770 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
771     int flen)
772 {
773 	int is_snap, need_upgrade;
774 
775 	/*
776 	 * Check if we've already upgraded or new filter is empty.
777 	 */
778 	if (d->bd_writer == 0 || fcode == NULL)
779 		return (0);
780 
781 	need_upgrade = 0;
782 
783 	/*
784 	 * Check if cmd looks like snaplen setting from
785 	 * pcap_bpf.c:pcap_open_live().
786 	 * Note we're not checking .k value here:
787 	 * while pcap_open_live() definitely sets to non-zero value,
788 	 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
789 	 * do not consider upgrading immediately
790 	 */
791 	if (cmd == BIOCSETF && flen == 1 &&
792 	    fcode[0].code == (BPF_RET | BPF_K))
793 		is_snap = 1;
794 	else
795 		is_snap = 0;
796 
797 	if (is_snap == 0) {
798 		/*
799 		 * We're setting first filter and it doesn't look like
800 		 * setting snaplen.  We're probably using bpf directly.
801 		 * Upgrade immediately.
802 		 */
803 		need_upgrade = 1;
804 	} else {
805 		/*
806 		 * Do not require upgrade by first BIOCSETF
807 		 * (used to set snaplen) by pcap_open_live().
808 		 */
809 
810 		if (--d->bd_writer == 0) {
811 			/*
812 			 * First snaplen filter has already
813 			 * been set. This is probably catch-all
814 			 * filter
815 			 */
816 			need_upgrade = 1;
817 		}
818 	}
819 
820 	CTR5(KTR_NET,
821 	    "%s: filter function set by pid %d, "
822 	    "bd_writer counter %d, snap %d upgrade %d",
823 	    __func__, d->bd_pid, d->bd_writer,
824 	    is_snap, need_upgrade);
825 
826 	return (need_upgrade);
827 }
828 
829 /*
830  * Detach a file from its interface.
831  */
832 static void
833 bpf_detachd(struct bpf_d *d)
834 {
835 	BPF_LOCK();
836 	bpf_detachd_locked(d, false);
837 	BPF_UNLOCK();
838 }
839 
840 static void
841 bpf_detachd_locked(struct bpf_d *d, bool detached_ifp)
842 {
843 	struct bpf_if *bp;
844 	struct ifnet *ifp;
845 	int error;
846 
847 	BPF_LOCK_ASSERT();
848 	CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
849 
850 	/* Check if descriptor is attached */
851 	if ((bp = d->bd_bif) == NULL)
852 		return;
853 
854 	BPFD_LOCK(d);
855 	/* Remove d from the interface's descriptor list. */
856 	CK_LIST_REMOVE(d, bd_next);
857 	/* Save bd_writer value */
858 	error = d->bd_writer;
859 	ifp = bp->bif_ifp;
860 	d->bd_bif = NULL;
861 	if (detached_ifp) {
862 		/*
863 		 * Notify descriptor as it's detached, so that any
864 		 * sleepers wake up and get ENXIO.
865 		 */
866 		bpf_wakeup(d);
867 	}
868 	BPFD_UNLOCK(d);
869 	bpf_bpfd_cnt--;
870 
871 	/* Call event handler iff d is attached */
872 	if (error == 0)
873 		EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
874 
875 	/*
876 	 * Check if this descriptor had requested promiscuous mode.
877 	 * If so and ifnet is not detached, turn it off.
878 	 */
879 	if (d->bd_promisc && !detached_ifp) {
880 		d->bd_promisc = 0;
881 		CURVNET_SET(ifp->if_vnet);
882 		error = ifpromisc(ifp, 0);
883 		CURVNET_RESTORE();
884 		if (error != 0 && error != ENXIO) {
885 			/*
886 			 * ENXIO can happen if a pccard is unplugged
887 			 * Something is really wrong if we were able to put
888 			 * the driver into promiscuous mode, but can't
889 			 * take it out.
890 			 */
891 			if_printf(bp->bif_ifp,
892 				"bpf_detach: ifpromisc failed (%d)\n", error);
893 		}
894 	}
895 	bpfif_rele(bp);
896 }
897 
898 /*
899  * Close the descriptor by detaching it from its interface,
900  * deallocating its buffers, and marking it free.
901  */
902 static void
903 bpf_dtor(void *data)
904 {
905 	struct bpf_d *d = data;
906 
907 	BPFD_LOCK(d);
908 	if (d->bd_state == BPF_WAITING)
909 		callout_stop(&d->bd_callout);
910 	d->bd_state = BPF_IDLE;
911 	BPFD_UNLOCK(d);
912 	funsetown(&d->bd_sigio);
913 	bpf_detachd(d);
914 #ifdef MAC
915 	mac_bpfdesc_destroy(d);
916 #endif /* MAC */
917 	seldrain(&d->bd_sel);
918 	knlist_destroy(&d->bd_sel.si_note);
919 	callout_drain(&d->bd_callout);
920 	bpfd_rele(d);
921 }
922 
923 /*
924  * Open ethernet device.  Returns ENXIO for illegal minor device number,
925  * EBUSY if file is open by another process.
926  */
927 /* ARGSUSED */
928 static	int
929 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
930 {
931 	struct bpf_d *d;
932 	int error;
933 
934 	d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
935 	error = devfs_set_cdevpriv(d, bpf_dtor);
936 	if (error != 0) {
937 		free(d, M_BPF);
938 		return (error);
939 	}
940 
941 	/* Setup counters */
942 	d->bd_rcount = counter_u64_alloc(M_WAITOK);
943 	d->bd_dcount = counter_u64_alloc(M_WAITOK);
944 	d->bd_fcount = counter_u64_alloc(M_WAITOK);
945 	d->bd_wcount = counter_u64_alloc(M_WAITOK);
946 	d->bd_wfcount = counter_u64_alloc(M_WAITOK);
947 	d->bd_wdcount = counter_u64_alloc(M_WAITOK);
948 	d->bd_zcopy = counter_u64_alloc(M_WAITOK);
949 
950 	/*
951 	 * For historical reasons, perform a one-time initialization call to
952 	 * the buffer routines, even though we're not yet committed to a
953 	 * particular buffer method.
954 	 */
955 	bpf_buffer_init(d);
956 	if ((flags & FREAD) == 0)
957 		d->bd_writer = 2;
958 	d->bd_hbuf_in_use = 0;
959 	d->bd_bufmode = BPF_BUFMODE_BUFFER;
960 	d->bd_sig = SIGIO;
961 	d->bd_direction = BPF_D_INOUT;
962 	d->bd_refcnt = 1;
963 	BPF_PID_REFRESH(d, td);
964 #ifdef MAC
965 	mac_bpfdesc_init(d);
966 	mac_bpfdesc_create(td->td_ucred, d);
967 #endif
968 	mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
969 	callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
970 	knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
971 
972 	return (0);
973 }
974 
975 /*
976  *  bpfread - read next chunk of packets from buffers
977  */
978 static	int
979 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
980 {
981 	struct bpf_d *d;
982 	int error;
983 	int non_block;
984 	int timed_out;
985 
986 	error = devfs_get_cdevpriv((void **)&d);
987 	if (error != 0)
988 		return (error);
989 
990 	/*
991 	 * Restrict application to use a buffer the same size as
992 	 * as kernel buffers.
993 	 */
994 	if (uio->uio_resid != d->bd_bufsize)
995 		return (EINVAL);
996 
997 	non_block = ((ioflag & O_NONBLOCK) != 0);
998 
999 	BPFD_LOCK(d);
1000 	BPF_PID_REFRESH_CUR(d);
1001 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1002 		BPFD_UNLOCK(d);
1003 		return (EOPNOTSUPP);
1004 	}
1005 	if (d->bd_state == BPF_WAITING)
1006 		callout_stop(&d->bd_callout);
1007 	timed_out = (d->bd_state == BPF_TIMED_OUT);
1008 	d->bd_state = BPF_IDLE;
1009 	while (d->bd_hbuf_in_use) {
1010 		error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1011 		    PRINET|PCATCH, "bd_hbuf", 0);
1012 		if (error != 0) {
1013 			BPFD_UNLOCK(d);
1014 			return (error);
1015 		}
1016 	}
1017 	/*
1018 	 * If the hold buffer is empty, then do a timed sleep, which
1019 	 * ends when the timeout expires or when enough packets
1020 	 * have arrived to fill the store buffer.
1021 	 */
1022 	while (d->bd_hbuf == NULL) {
1023 		if (d->bd_slen != 0) {
1024 			/*
1025 			 * A packet(s) either arrived since the previous
1026 			 * read or arrived while we were asleep.
1027 			 */
1028 			if (d->bd_immediate || non_block || timed_out) {
1029 				/*
1030 				 * Rotate the buffers and return what's here
1031 				 * if we are in immediate mode, non-blocking
1032 				 * flag is set, or this descriptor timed out.
1033 				 */
1034 				ROTATE_BUFFERS(d);
1035 				break;
1036 			}
1037 		}
1038 
1039 		/*
1040 		 * No data is available, check to see if the bpf device
1041 		 * is still pointed at a real interface.  If not, return
1042 		 * ENXIO so that the userland process knows to rebind
1043 		 * it before using it again.
1044 		 */
1045 		if (d->bd_bif == NULL) {
1046 			BPFD_UNLOCK(d);
1047 			return (ENXIO);
1048 		}
1049 
1050 		if (non_block) {
1051 			BPFD_UNLOCK(d);
1052 			return (EWOULDBLOCK);
1053 		}
1054 		error = msleep(d, &d->bd_lock, PRINET|PCATCH,
1055 		     "bpf", d->bd_rtout);
1056 		if (error == EINTR || error == ERESTART) {
1057 			BPFD_UNLOCK(d);
1058 			return (error);
1059 		}
1060 		if (error == EWOULDBLOCK) {
1061 			/*
1062 			 * On a timeout, return what's in the buffer,
1063 			 * which may be nothing.  If there is something
1064 			 * in the store buffer, we can rotate the buffers.
1065 			 */
1066 			if (d->bd_hbuf)
1067 				/*
1068 				 * We filled up the buffer in between
1069 				 * getting the timeout and arriving
1070 				 * here, so we don't need to rotate.
1071 				 */
1072 				break;
1073 
1074 			if (d->bd_slen == 0) {
1075 				BPFD_UNLOCK(d);
1076 				return (0);
1077 			}
1078 			ROTATE_BUFFERS(d);
1079 			break;
1080 		}
1081 	}
1082 	/*
1083 	 * At this point, we know we have something in the hold slot.
1084 	 */
1085 	d->bd_hbuf_in_use = 1;
1086 	BPFD_UNLOCK(d);
1087 
1088 	/*
1089 	 * Move data from hold buffer into user space.
1090 	 * We know the entire buffer is transferred since
1091 	 * we checked above that the read buffer is bpf_bufsize bytes.
1092   	 *
1093 	 * We do not have to worry about simultaneous reads because
1094 	 * we waited for sole access to the hold buffer above.
1095 	 */
1096 	error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1097 
1098 	BPFD_LOCK(d);
1099 	KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1100 	d->bd_fbuf = d->bd_hbuf;
1101 	d->bd_hbuf = NULL;
1102 	d->bd_hlen = 0;
1103 	bpf_buf_reclaimed(d);
1104 	d->bd_hbuf_in_use = 0;
1105 	wakeup(&d->bd_hbuf_in_use);
1106 	BPFD_UNLOCK(d);
1107 
1108 	return (error);
1109 }
1110 
1111 /*
1112  * If there are processes sleeping on this descriptor, wake them up.
1113  */
1114 static __inline void
1115 bpf_wakeup(struct bpf_d *d)
1116 {
1117 
1118 	BPFD_LOCK_ASSERT(d);
1119 	if (d->bd_state == BPF_WAITING) {
1120 		callout_stop(&d->bd_callout);
1121 		d->bd_state = BPF_IDLE;
1122 	}
1123 	wakeup(d);
1124 	if (d->bd_async && d->bd_sig && d->bd_sigio)
1125 		pgsigio(&d->bd_sigio, d->bd_sig, 0);
1126 
1127 	selwakeuppri(&d->bd_sel, PRINET);
1128 	KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1129 }
1130 
1131 static void
1132 bpf_timed_out(void *arg)
1133 {
1134 	struct bpf_d *d = (struct bpf_d *)arg;
1135 
1136 	BPFD_LOCK_ASSERT(d);
1137 
1138 	if (callout_pending(&d->bd_callout) ||
1139 	    !callout_active(&d->bd_callout))
1140 		return;
1141 	if (d->bd_state == BPF_WAITING) {
1142 		d->bd_state = BPF_TIMED_OUT;
1143 		if (d->bd_slen != 0)
1144 			bpf_wakeup(d);
1145 	}
1146 }
1147 
1148 static int
1149 bpf_ready(struct bpf_d *d)
1150 {
1151 
1152 	BPFD_LOCK_ASSERT(d);
1153 
1154 	if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1155 		return (1);
1156 	if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1157 	    d->bd_slen != 0)
1158 		return (1);
1159 	return (0);
1160 }
1161 
1162 static int
1163 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1164 {
1165 	struct route ro;
1166 	struct sockaddr dst;
1167 	struct epoch_tracker et;
1168 	struct bpf_if *bp;
1169 	struct bpf_d *d;
1170 	struct ifnet *ifp;
1171 	struct mbuf *m, *mc;
1172 	int error, hlen;
1173 
1174 	error = devfs_get_cdevpriv((void **)&d);
1175 	if (error != 0)
1176 		return (error);
1177 
1178 	NET_EPOCH_ENTER(et);
1179 	BPFD_LOCK(d);
1180 	BPF_PID_REFRESH_CUR(d);
1181 	counter_u64_add(d->bd_wcount, 1);
1182 	if ((bp = d->bd_bif) == NULL) {
1183 		error = ENXIO;
1184 		goto out_locked;
1185 	}
1186 
1187 	ifp = bp->bif_ifp;
1188 	if ((ifp->if_flags & IFF_UP) == 0) {
1189 		error = ENETDOWN;
1190 		goto out_locked;
1191 	}
1192 
1193 	if (uio->uio_resid == 0)
1194 		goto out_locked;
1195 
1196 	bzero(&dst, sizeof(dst));
1197 	m = NULL;
1198 	hlen = 0;
1199 
1200 	/*
1201 	 * Take extra reference, unlock d and exit from epoch section,
1202 	 * since bpf_movein() can sleep.
1203 	 */
1204 	bpfd_ref(d);
1205 	NET_EPOCH_EXIT(et);
1206 	BPFD_UNLOCK(d);
1207 
1208 	error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1209 	    &m, &dst, &hlen, d);
1210 
1211 	if (error != 0) {
1212 		counter_u64_add(d->bd_wdcount, 1);
1213 		bpfd_rele(d);
1214 		return (error);
1215 	}
1216 
1217 	BPFD_LOCK(d);
1218 	/*
1219 	 * Check that descriptor is still attached to the interface.
1220 	 * This can happen on bpfdetach(). To avoid access to detached
1221 	 * ifnet, free mbuf and return ENXIO.
1222 	 */
1223 	if (d->bd_bif == NULL) {
1224 		counter_u64_add(d->bd_wdcount, 1);
1225 		BPFD_UNLOCK(d);
1226 		bpfd_rele(d);
1227 		m_freem(m);
1228 		return (ENXIO);
1229 	}
1230 	counter_u64_add(d->bd_wfcount, 1);
1231 	if (d->bd_hdrcmplt)
1232 		dst.sa_family = pseudo_AF_HDRCMPLT;
1233 
1234 	if (d->bd_feedback) {
1235 		mc = m_dup(m, M_NOWAIT);
1236 		if (mc != NULL)
1237 			mc->m_pkthdr.rcvif = ifp;
1238 		/* Set M_PROMISC for outgoing packets to be discarded. */
1239 		if (d->bd_direction == BPF_D_INOUT)
1240 			m->m_flags |= M_PROMISC;
1241 	} else
1242 		mc = NULL;
1243 
1244 	m->m_pkthdr.len -= hlen;
1245 	m->m_len -= hlen;
1246 	m->m_data += hlen;	/* XXX */
1247 
1248 	CURVNET_SET(ifp->if_vnet);
1249 #ifdef MAC
1250 	mac_bpfdesc_create_mbuf(d, m);
1251 	if (mc != NULL)
1252 		mac_bpfdesc_create_mbuf(d, mc);
1253 #endif
1254 
1255 	bzero(&ro, sizeof(ro));
1256 	if (hlen != 0) {
1257 		ro.ro_prepend = (u_char *)&dst.sa_data;
1258 		ro.ro_plen = hlen;
1259 		ro.ro_flags = RT_HAS_HEADER;
1260 	}
1261 
1262 	/* Avoid possible recursion on BPFD_LOCK(). */
1263 	NET_EPOCH_ENTER(et);
1264 	BPFD_UNLOCK(d);
1265 	error = (*ifp->if_output)(ifp, m, &dst, &ro);
1266 	if (error)
1267 		counter_u64_add(d->bd_wdcount, 1);
1268 
1269 	if (mc != NULL) {
1270 		if (error == 0)
1271 			(*ifp->if_input)(ifp, mc);
1272 		else
1273 			m_freem(mc);
1274 	}
1275 	NET_EPOCH_EXIT(et);
1276 	CURVNET_RESTORE();
1277 	bpfd_rele(d);
1278 	return (error);
1279 
1280 out_locked:
1281 	counter_u64_add(d->bd_wdcount, 1);
1282 	NET_EPOCH_EXIT(et);
1283 	BPFD_UNLOCK(d);
1284 	return (error);
1285 }
1286 
1287 /*
1288  * Reset a descriptor by flushing its packet buffer and clearing the receive
1289  * and drop counts.  This is doable for kernel-only buffers, but with
1290  * zero-copy buffers, we can't write to (or rotate) buffers that are
1291  * currently owned by userspace.  It would be nice if we could encapsulate
1292  * this logic in the buffer code rather than here.
1293  */
1294 static void
1295 reset_d(struct bpf_d *d)
1296 {
1297 
1298 	BPFD_LOCK_ASSERT(d);
1299 
1300 	while (d->bd_hbuf_in_use)
1301 		mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1302 		    "bd_hbuf", 0);
1303 	if ((d->bd_hbuf != NULL) &&
1304 	    (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1305 		/* Free the hold buffer. */
1306 		d->bd_fbuf = d->bd_hbuf;
1307 		d->bd_hbuf = NULL;
1308 		d->bd_hlen = 0;
1309 		bpf_buf_reclaimed(d);
1310 	}
1311 	if (bpf_canwritebuf(d))
1312 		d->bd_slen = 0;
1313 	counter_u64_zero(d->bd_rcount);
1314 	counter_u64_zero(d->bd_dcount);
1315 	counter_u64_zero(d->bd_fcount);
1316 	counter_u64_zero(d->bd_wcount);
1317 	counter_u64_zero(d->bd_wfcount);
1318 	counter_u64_zero(d->bd_wdcount);
1319 	counter_u64_zero(d->bd_zcopy);
1320 }
1321 
1322 /*
1323  *  FIONREAD		Check for read packet available.
1324  *  BIOCGBLEN		Get buffer len [for read()].
1325  *  BIOCSETF		Set read filter.
1326  *  BIOCSETFNR		Set read filter without resetting descriptor.
1327  *  BIOCSETWF		Set write filter.
1328  *  BIOCFLUSH		Flush read packet buffer.
1329  *  BIOCPROMISC		Put interface into promiscuous mode.
1330  *  BIOCGDLT		Get link layer type.
1331  *  BIOCGETIF		Get interface name.
1332  *  BIOCSETIF		Set interface.
1333  *  BIOCSRTIMEOUT	Set read timeout.
1334  *  BIOCGRTIMEOUT	Get read timeout.
1335  *  BIOCGSTATS		Get packet stats.
1336  *  BIOCIMMEDIATE	Set immediate mode.
1337  *  BIOCVERSION		Get filter language version.
1338  *  BIOCGHDRCMPLT	Get "header already complete" flag
1339  *  BIOCSHDRCMPLT	Set "header already complete" flag
1340  *  BIOCGDIRECTION	Get packet direction flag
1341  *  BIOCSDIRECTION	Set packet direction flag
1342  *  BIOCGTSTAMP		Get time stamp format and resolution.
1343  *  BIOCSTSTAMP		Set time stamp format and resolution.
1344  *  BIOCLOCK		Set "locked" flag
1345  *  BIOCFEEDBACK	Set packet feedback mode.
1346  *  BIOCSETZBUF		Set current zero-copy buffer locations.
1347  *  BIOCGETZMAX		Get maximum zero-copy buffer size.
1348  *  BIOCROTZBUF		Force rotation of zero-copy buffer
1349  *  BIOCSETBUFMODE	Set buffer mode.
1350  *  BIOCGETBUFMODE	Get current buffer mode.
1351  */
1352 /* ARGSUSED */
1353 static	int
1354 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1355     struct thread *td)
1356 {
1357 	struct bpf_d *d;
1358 	int error;
1359 
1360 	error = devfs_get_cdevpriv((void **)&d);
1361 	if (error != 0)
1362 		return (error);
1363 
1364 	/*
1365 	 * Refresh PID associated with this descriptor.
1366 	 */
1367 	BPFD_LOCK(d);
1368 	BPF_PID_REFRESH(d, td);
1369 	if (d->bd_state == BPF_WAITING)
1370 		callout_stop(&d->bd_callout);
1371 	d->bd_state = BPF_IDLE;
1372 	BPFD_UNLOCK(d);
1373 
1374 	if (d->bd_locked == 1) {
1375 		switch (cmd) {
1376 		case BIOCGBLEN:
1377 		case BIOCFLUSH:
1378 		case BIOCGDLT:
1379 		case BIOCGDLTLIST:
1380 #ifdef COMPAT_FREEBSD32
1381 		case BIOCGDLTLIST32:
1382 #endif
1383 		case BIOCGETIF:
1384 		case BIOCGRTIMEOUT:
1385 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1386 		case BIOCGRTIMEOUT32:
1387 #endif
1388 		case BIOCGSTATS:
1389 		case BIOCVERSION:
1390 		case BIOCGRSIG:
1391 		case BIOCGHDRCMPLT:
1392 		case BIOCSTSTAMP:
1393 		case BIOCFEEDBACK:
1394 		case FIONREAD:
1395 		case BIOCLOCK:
1396 		case BIOCSRTIMEOUT:
1397 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1398 		case BIOCSRTIMEOUT32:
1399 #endif
1400 		case BIOCIMMEDIATE:
1401 		case TIOCGPGRP:
1402 		case BIOCROTZBUF:
1403 			break;
1404 		default:
1405 			return (EPERM);
1406 		}
1407 	}
1408 #ifdef COMPAT_FREEBSD32
1409 	/*
1410 	 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1411 	 * that it will get 32-bit packet headers.
1412 	 */
1413 	switch (cmd) {
1414 	case BIOCSETF32:
1415 	case BIOCSETFNR32:
1416 	case BIOCSETWF32:
1417 	case BIOCGDLTLIST32:
1418 	case BIOCGRTIMEOUT32:
1419 	case BIOCSRTIMEOUT32:
1420 		if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1421 			BPFD_LOCK(d);
1422 			d->bd_compat32 = 1;
1423 			BPFD_UNLOCK(d);
1424 		}
1425 	}
1426 #endif
1427 
1428 	CURVNET_SET(TD_TO_VNET(td));
1429 	switch (cmd) {
1430 	default:
1431 		error = EINVAL;
1432 		break;
1433 
1434 	/*
1435 	 * Check for read packet available.
1436 	 */
1437 	case FIONREAD:
1438 		{
1439 			int n;
1440 
1441 			BPFD_LOCK(d);
1442 			n = d->bd_slen;
1443 			while (d->bd_hbuf_in_use)
1444 				mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1445 				    PRINET, "bd_hbuf", 0);
1446 			if (d->bd_hbuf)
1447 				n += d->bd_hlen;
1448 			BPFD_UNLOCK(d);
1449 
1450 			*(int *)addr = n;
1451 			break;
1452 		}
1453 
1454 	/*
1455 	 * Get buffer len [for read()].
1456 	 */
1457 	case BIOCGBLEN:
1458 		BPFD_LOCK(d);
1459 		*(u_int *)addr = d->bd_bufsize;
1460 		BPFD_UNLOCK(d);
1461 		break;
1462 
1463 	/*
1464 	 * Set buffer length.
1465 	 */
1466 	case BIOCSBLEN:
1467 		error = bpf_ioctl_sblen(d, (u_int *)addr);
1468 		break;
1469 
1470 	/*
1471 	 * Set link layer read filter.
1472 	 */
1473 	case BIOCSETF:
1474 	case BIOCSETFNR:
1475 	case BIOCSETWF:
1476 #ifdef COMPAT_FREEBSD32
1477 	case BIOCSETF32:
1478 	case BIOCSETFNR32:
1479 	case BIOCSETWF32:
1480 #endif
1481 		error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1482 		break;
1483 
1484 	/*
1485 	 * Flush read packet buffer.
1486 	 */
1487 	case BIOCFLUSH:
1488 		BPFD_LOCK(d);
1489 		reset_d(d);
1490 		BPFD_UNLOCK(d);
1491 		break;
1492 
1493 	/*
1494 	 * Put interface into promiscuous mode.
1495 	 */
1496 	case BIOCPROMISC:
1497 		if (d->bd_bif == NULL) {
1498 			/*
1499 			 * No interface attached yet.
1500 			 */
1501 			error = EINVAL;
1502 			break;
1503 		}
1504 		if (d->bd_promisc == 0) {
1505 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1506 			if (error == 0)
1507 				d->bd_promisc = 1;
1508 		}
1509 		break;
1510 
1511 	/*
1512 	 * Get current data link type.
1513 	 */
1514 	case BIOCGDLT:
1515 		BPF_LOCK();
1516 		if (d->bd_bif == NULL)
1517 			error = EINVAL;
1518 		else
1519 			*(u_int *)addr = d->bd_bif->bif_dlt;
1520 		BPF_UNLOCK();
1521 		break;
1522 
1523 	/*
1524 	 * Get a list of supported data link types.
1525 	 */
1526 #ifdef COMPAT_FREEBSD32
1527 	case BIOCGDLTLIST32:
1528 		{
1529 			struct bpf_dltlist32 *list32;
1530 			struct bpf_dltlist dltlist;
1531 
1532 			list32 = (struct bpf_dltlist32 *)addr;
1533 			dltlist.bfl_len = list32->bfl_len;
1534 			dltlist.bfl_list = PTRIN(list32->bfl_list);
1535 			BPF_LOCK();
1536 			if (d->bd_bif == NULL)
1537 				error = EINVAL;
1538 			else {
1539 				error = bpf_getdltlist(d, &dltlist);
1540 				if (error == 0)
1541 					list32->bfl_len = dltlist.bfl_len;
1542 			}
1543 			BPF_UNLOCK();
1544 			break;
1545 		}
1546 #endif
1547 
1548 	case BIOCGDLTLIST:
1549 		BPF_LOCK();
1550 		if (d->bd_bif == NULL)
1551 			error = EINVAL;
1552 		else
1553 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1554 		BPF_UNLOCK();
1555 		break;
1556 
1557 	/*
1558 	 * Set data link type.
1559 	 */
1560 	case BIOCSDLT:
1561 		BPF_LOCK();
1562 		if (d->bd_bif == NULL)
1563 			error = EINVAL;
1564 		else
1565 			error = bpf_setdlt(d, *(u_int *)addr);
1566 		BPF_UNLOCK();
1567 		break;
1568 
1569 	/*
1570 	 * Get interface name.
1571 	 */
1572 	case BIOCGETIF:
1573 		BPF_LOCK();
1574 		if (d->bd_bif == NULL)
1575 			error = EINVAL;
1576 		else {
1577 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
1578 			struct ifreq *const ifr = (struct ifreq *)addr;
1579 
1580 			strlcpy(ifr->ifr_name, ifp->if_xname,
1581 			    sizeof(ifr->ifr_name));
1582 		}
1583 		BPF_UNLOCK();
1584 		break;
1585 
1586 	/*
1587 	 * Set interface.
1588 	 */
1589 	case BIOCSETIF:
1590 		{
1591 			int alloc_buf, size;
1592 
1593 			/*
1594 			 * Behavior here depends on the buffering model.  If
1595 			 * we're using kernel memory buffers, then we can
1596 			 * allocate them here.  If we're using zero-copy,
1597 			 * then the user process must have registered buffers
1598 			 * by the time we get here.
1599 			 */
1600 			alloc_buf = 0;
1601 			BPFD_LOCK(d);
1602 			if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1603 			    d->bd_sbuf == NULL)
1604 				alloc_buf = 1;
1605 			BPFD_UNLOCK(d);
1606 			if (alloc_buf) {
1607 				size = d->bd_bufsize;
1608 				error = bpf_buffer_ioctl_sblen(d, &size);
1609 				if (error != 0)
1610 					break;
1611 			}
1612 			BPF_LOCK();
1613 			error = bpf_setif(d, (struct ifreq *)addr);
1614 			BPF_UNLOCK();
1615 			break;
1616 		}
1617 
1618 	/*
1619 	 * Set read timeout.
1620 	 */
1621 	case BIOCSRTIMEOUT:
1622 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1623 	case BIOCSRTIMEOUT32:
1624 #endif
1625 		{
1626 			struct timeval *tv = (struct timeval *)addr;
1627 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1628 			struct timeval32 *tv32;
1629 			struct timeval tv64;
1630 
1631 			if (cmd == BIOCSRTIMEOUT32) {
1632 				tv32 = (struct timeval32 *)addr;
1633 				tv = &tv64;
1634 				tv->tv_sec = tv32->tv_sec;
1635 				tv->tv_usec = tv32->tv_usec;
1636 			} else
1637 #endif
1638 				tv = (struct timeval *)addr;
1639 
1640 			/*
1641 			 * Subtract 1 tick from tvtohz() since this isn't
1642 			 * a one-shot timer.
1643 			 */
1644 			if ((error = itimerfix(tv)) == 0)
1645 				d->bd_rtout = tvtohz(tv) - 1;
1646 			break;
1647 		}
1648 
1649 	/*
1650 	 * Get read timeout.
1651 	 */
1652 	case BIOCGRTIMEOUT:
1653 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1654 	case BIOCGRTIMEOUT32:
1655 #endif
1656 		{
1657 			struct timeval *tv;
1658 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1659 			struct timeval32 *tv32;
1660 			struct timeval tv64;
1661 
1662 			if (cmd == BIOCGRTIMEOUT32)
1663 				tv = &tv64;
1664 			else
1665 #endif
1666 				tv = (struct timeval *)addr;
1667 
1668 			tv->tv_sec = d->bd_rtout / hz;
1669 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1670 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1671 			if (cmd == BIOCGRTIMEOUT32) {
1672 				tv32 = (struct timeval32 *)addr;
1673 				tv32->tv_sec = tv->tv_sec;
1674 				tv32->tv_usec = tv->tv_usec;
1675 			}
1676 #endif
1677 
1678 			break;
1679 		}
1680 
1681 	/*
1682 	 * Get packet stats.
1683 	 */
1684 	case BIOCGSTATS:
1685 		{
1686 			struct bpf_stat *bs = (struct bpf_stat *)addr;
1687 
1688 			/* XXXCSJP overflow */
1689 			bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1690 			bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1691 			break;
1692 		}
1693 
1694 	/*
1695 	 * Set immediate mode.
1696 	 */
1697 	case BIOCIMMEDIATE:
1698 		BPFD_LOCK(d);
1699 		d->bd_immediate = *(u_int *)addr;
1700 		BPFD_UNLOCK(d);
1701 		break;
1702 
1703 	case BIOCVERSION:
1704 		{
1705 			struct bpf_version *bv = (struct bpf_version *)addr;
1706 
1707 			bv->bv_major = BPF_MAJOR_VERSION;
1708 			bv->bv_minor = BPF_MINOR_VERSION;
1709 			break;
1710 		}
1711 
1712 	/*
1713 	 * Get "header already complete" flag
1714 	 */
1715 	case BIOCGHDRCMPLT:
1716 		BPFD_LOCK(d);
1717 		*(u_int *)addr = d->bd_hdrcmplt;
1718 		BPFD_UNLOCK(d);
1719 		break;
1720 
1721 	/*
1722 	 * Set "header already complete" flag
1723 	 */
1724 	case BIOCSHDRCMPLT:
1725 		BPFD_LOCK(d);
1726 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1727 		BPFD_UNLOCK(d);
1728 		break;
1729 
1730 	/*
1731 	 * Get packet direction flag
1732 	 */
1733 	case BIOCGDIRECTION:
1734 		BPFD_LOCK(d);
1735 		*(u_int *)addr = d->bd_direction;
1736 		BPFD_UNLOCK(d);
1737 		break;
1738 
1739 	/*
1740 	 * Set packet direction flag
1741 	 */
1742 	case BIOCSDIRECTION:
1743 		{
1744 			u_int	direction;
1745 
1746 			direction = *(u_int *)addr;
1747 			switch (direction) {
1748 			case BPF_D_IN:
1749 			case BPF_D_INOUT:
1750 			case BPF_D_OUT:
1751 				BPFD_LOCK(d);
1752 				d->bd_direction = direction;
1753 				BPFD_UNLOCK(d);
1754 				break;
1755 			default:
1756 				error = EINVAL;
1757 			}
1758 		}
1759 		break;
1760 
1761 	/*
1762 	 * Get packet timestamp format and resolution.
1763 	 */
1764 	case BIOCGTSTAMP:
1765 		BPFD_LOCK(d);
1766 		*(u_int *)addr = d->bd_tstamp;
1767 		BPFD_UNLOCK(d);
1768 		break;
1769 
1770 	/*
1771 	 * Set packet timestamp format and resolution.
1772 	 */
1773 	case BIOCSTSTAMP:
1774 		{
1775 			u_int	func;
1776 
1777 			func = *(u_int *)addr;
1778 			if (BPF_T_VALID(func))
1779 				d->bd_tstamp = func;
1780 			else
1781 				error = EINVAL;
1782 		}
1783 		break;
1784 
1785 	case BIOCFEEDBACK:
1786 		BPFD_LOCK(d);
1787 		d->bd_feedback = *(u_int *)addr;
1788 		BPFD_UNLOCK(d);
1789 		break;
1790 
1791 	case BIOCLOCK:
1792 		BPFD_LOCK(d);
1793 		d->bd_locked = 1;
1794 		BPFD_UNLOCK(d);
1795 		break;
1796 
1797 	case FIONBIO:		/* Non-blocking I/O */
1798 		break;
1799 
1800 	case FIOASYNC:		/* Send signal on receive packets */
1801 		BPFD_LOCK(d);
1802 		d->bd_async = *(int *)addr;
1803 		BPFD_UNLOCK(d);
1804 		break;
1805 
1806 	case FIOSETOWN:
1807 		/*
1808 		 * XXX: Add some sort of locking here?
1809 		 * fsetown() can sleep.
1810 		 */
1811 		error = fsetown(*(int *)addr, &d->bd_sigio);
1812 		break;
1813 
1814 	case FIOGETOWN:
1815 		BPFD_LOCK(d);
1816 		*(int *)addr = fgetown(&d->bd_sigio);
1817 		BPFD_UNLOCK(d);
1818 		break;
1819 
1820 	/* This is deprecated, FIOSETOWN should be used instead. */
1821 	case TIOCSPGRP:
1822 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
1823 		break;
1824 
1825 	/* This is deprecated, FIOGETOWN should be used instead. */
1826 	case TIOCGPGRP:
1827 		*(int *)addr = -fgetown(&d->bd_sigio);
1828 		break;
1829 
1830 	case BIOCSRSIG:		/* Set receive signal */
1831 		{
1832 			u_int sig;
1833 
1834 			sig = *(u_int *)addr;
1835 
1836 			if (sig >= NSIG)
1837 				error = EINVAL;
1838 			else {
1839 				BPFD_LOCK(d);
1840 				d->bd_sig = sig;
1841 				BPFD_UNLOCK(d);
1842 			}
1843 			break;
1844 		}
1845 	case BIOCGRSIG:
1846 		BPFD_LOCK(d);
1847 		*(u_int *)addr = d->bd_sig;
1848 		BPFD_UNLOCK(d);
1849 		break;
1850 
1851 	case BIOCGETBUFMODE:
1852 		BPFD_LOCK(d);
1853 		*(u_int *)addr = d->bd_bufmode;
1854 		BPFD_UNLOCK(d);
1855 		break;
1856 
1857 	case BIOCSETBUFMODE:
1858 		/*
1859 		 * Allow the buffering mode to be changed as long as we
1860 		 * haven't yet committed to a particular mode.  Our
1861 		 * definition of commitment, for now, is whether or not a
1862 		 * buffer has been allocated or an interface attached, since
1863 		 * that's the point where things get tricky.
1864 		 */
1865 		switch (*(u_int *)addr) {
1866 		case BPF_BUFMODE_BUFFER:
1867 			break;
1868 
1869 		case BPF_BUFMODE_ZBUF:
1870 			if (bpf_zerocopy_enable)
1871 				break;
1872 			/* FALLSTHROUGH */
1873 
1874 		default:
1875 			CURVNET_RESTORE();
1876 			return (EINVAL);
1877 		}
1878 
1879 		BPFD_LOCK(d);
1880 		if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1881 		    d->bd_fbuf != NULL || d->bd_bif != NULL) {
1882 			BPFD_UNLOCK(d);
1883 			CURVNET_RESTORE();
1884 			return (EBUSY);
1885 		}
1886 		d->bd_bufmode = *(u_int *)addr;
1887 		BPFD_UNLOCK(d);
1888 		break;
1889 
1890 	case BIOCGETZMAX:
1891 		error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1892 		break;
1893 
1894 	case BIOCSETZBUF:
1895 		error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1896 		break;
1897 
1898 	case BIOCROTZBUF:
1899 		error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1900 		break;
1901 	}
1902 	CURVNET_RESTORE();
1903 	return (error);
1904 }
1905 
1906 /*
1907  * Set d's packet filter program to fp. If this file already has a filter,
1908  * free it and replace it. Returns EINVAL for bogus requests.
1909  *
1910  * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1911  * calls.
1912  */
1913 static int
1914 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1915 {
1916 #ifdef COMPAT_FREEBSD32
1917 	struct bpf_program fp_swab;
1918 	struct bpf_program32 *fp32;
1919 #endif
1920 	struct bpf_program_buffer *fcode;
1921 	struct bpf_insn *filter;
1922 #ifdef BPF_JITTER
1923 	bpf_jit_filter *jfunc;
1924 #endif
1925 	size_t size;
1926 	u_int flen;
1927 	bool track_event;
1928 
1929 #ifdef COMPAT_FREEBSD32
1930 	switch (cmd) {
1931 	case BIOCSETF32:
1932 	case BIOCSETWF32:
1933 	case BIOCSETFNR32:
1934 		fp32 = (struct bpf_program32 *)fp;
1935 		fp_swab.bf_len = fp32->bf_len;
1936 		fp_swab.bf_insns =
1937 		    (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1938 		fp = &fp_swab;
1939 		switch (cmd) {
1940 		case BIOCSETF32:
1941 			cmd = BIOCSETF;
1942 			break;
1943 		case BIOCSETWF32:
1944 			cmd = BIOCSETWF;
1945 			break;
1946 		}
1947 		break;
1948 	}
1949 #endif
1950 
1951 	filter = NULL;
1952 #ifdef BPF_JITTER
1953 	jfunc = NULL;
1954 #endif
1955 	/*
1956 	 * Check new filter validness before acquiring any locks.
1957 	 * Allocate memory for new filter, if needed.
1958 	 */
1959 	flen = fp->bf_len;
1960 	if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1961 		return (EINVAL);
1962 	size = flen * sizeof(*fp->bf_insns);
1963 	if (size > 0) {
1964 		/* We're setting up new filter. Copy and check actual data. */
1965 		fcode = bpf_program_buffer_alloc(size, M_WAITOK);
1966 		filter = (struct bpf_insn *)fcode->buffer;
1967 		if (copyin(fp->bf_insns, filter, size) != 0 ||
1968 		    !bpf_validate(filter, flen)) {
1969 			free(fcode, M_BPF);
1970 			return (EINVAL);
1971 		}
1972 #ifdef BPF_JITTER
1973 		if (cmd != BIOCSETWF) {
1974 			/*
1975 			 * Filter is copied inside fcode and is
1976 			 * perfectly valid.
1977 			 */
1978 			jfunc = bpf_jitter(filter, flen);
1979 		}
1980 #endif
1981 	}
1982 
1983 	track_event = false;
1984 	fcode = NULL;
1985 
1986 	BPF_LOCK();
1987 	BPFD_LOCK(d);
1988 	/* Set up new filter. */
1989 	if (cmd == BIOCSETWF) {
1990 		if (d->bd_wfilter != NULL) {
1991 			fcode = __containerof((void *)d->bd_wfilter,
1992 			    struct bpf_program_buffer, buffer);
1993 #ifdef BPF_JITTER
1994 			fcode->func = NULL;
1995 #endif
1996 		}
1997 		d->bd_wfilter = filter;
1998 	} else {
1999 		if (d->bd_rfilter != NULL) {
2000 			fcode = __containerof((void *)d->bd_rfilter,
2001 			    struct bpf_program_buffer, buffer);
2002 #ifdef BPF_JITTER
2003 			fcode->func = d->bd_bfilter;
2004 #endif
2005 		}
2006 		d->bd_rfilter = filter;
2007 #ifdef BPF_JITTER
2008 		d->bd_bfilter = jfunc;
2009 #endif
2010 		if (cmd == BIOCSETF)
2011 			reset_d(d);
2012 
2013 		if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2014 			/*
2015 			 * Filter can be set several times without
2016 			 * specifying interface. In this case just mark d
2017 			 * as reader.
2018 			 */
2019 			d->bd_writer = 0;
2020 			if (d->bd_bif != NULL) {
2021 				/*
2022 				 * Remove descriptor from writers-only list
2023 				 * and add it to active readers list.
2024 				 */
2025 				CK_LIST_REMOVE(d, bd_next);
2026 				CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2027 				    d, bd_next);
2028 				CTR2(KTR_NET,
2029 				    "%s: upgrade required by pid %d",
2030 				    __func__, d->bd_pid);
2031 				track_event = true;
2032 			}
2033 		}
2034 	}
2035 	BPFD_UNLOCK(d);
2036 
2037 	if (fcode != NULL)
2038 		NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2039 
2040 	if (track_event)
2041 		EVENTHANDLER_INVOKE(bpf_track,
2042 		    d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2043 
2044 	BPF_UNLOCK();
2045 	return (0);
2046 }
2047 
2048 /*
2049  * Detach a file from its current interface (if attached at all) and attach
2050  * to the interface indicated by the name stored in ifr.
2051  * Return an errno or 0.
2052  */
2053 static int
2054 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2055 {
2056 	struct bpf_if *bp;
2057 	struct ifnet *theywant;
2058 
2059 	BPF_LOCK_ASSERT();
2060 
2061 	theywant = ifunit(ifr->ifr_name);
2062 	if (theywant == NULL || theywant->if_bpf == NULL)
2063 		return (ENXIO);
2064 
2065 	bp = theywant->if_bpf;
2066 	/*
2067 	 * At this point, we expect the buffer is already allocated.  If not,
2068 	 * return an error.
2069 	 */
2070 	switch (d->bd_bufmode) {
2071 	case BPF_BUFMODE_BUFFER:
2072 	case BPF_BUFMODE_ZBUF:
2073 		if (d->bd_sbuf == NULL)
2074 			return (EINVAL);
2075 		break;
2076 
2077 	default:
2078 		panic("bpf_setif: bufmode %d", d->bd_bufmode);
2079 	}
2080 	if (bp != d->bd_bif)
2081 		bpf_attachd(d, bp);
2082 	else {
2083 		BPFD_LOCK(d);
2084 		reset_d(d);
2085 		BPFD_UNLOCK(d);
2086 	}
2087 	return (0);
2088 }
2089 
2090 /*
2091  * Support for select() and poll() system calls
2092  *
2093  * Return true iff the specific operation will not block indefinitely.
2094  * Otherwise, return false but make a note that a selwakeup() must be done.
2095  */
2096 static int
2097 bpfpoll(struct cdev *dev, int events, struct thread *td)
2098 {
2099 	struct bpf_d *d;
2100 	int revents;
2101 
2102 	if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2103 		return (events &
2104 		    (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
2105 
2106 	/*
2107 	 * Refresh PID associated with this descriptor.
2108 	 */
2109 	revents = events & (POLLOUT | POLLWRNORM);
2110 	BPFD_LOCK(d);
2111 	BPF_PID_REFRESH(d, td);
2112 	if (events & (POLLIN | POLLRDNORM)) {
2113 		if (bpf_ready(d))
2114 			revents |= events & (POLLIN | POLLRDNORM);
2115 		else {
2116 			selrecord(td, &d->bd_sel);
2117 			/* Start the read timeout if necessary. */
2118 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2119 				callout_reset(&d->bd_callout, d->bd_rtout,
2120 				    bpf_timed_out, d);
2121 				d->bd_state = BPF_WAITING;
2122 			}
2123 		}
2124 	}
2125 	BPFD_UNLOCK(d);
2126 	return (revents);
2127 }
2128 
2129 /*
2130  * Support for kevent() system call.  Register EVFILT_READ filters and
2131  * reject all others.
2132  */
2133 int
2134 bpfkqfilter(struct cdev *dev, struct knote *kn)
2135 {
2136 	struct bpf_d *d;
2137 
2138 	if (devfs_get_cdevpriv((void **)&d) != 0 ||
2139 	    kn->kn_filter != EVFILT_READ)
2140 		return (1);
2141 
2142 	/*
2143 	 * Refresh PID associated with this descriptor.
2144 	 */
2145 	BPFD_LOCK(d);
2146 	BPF_PID_REFRESH_CUR(d);
2147 	kn->kn_fop = &bpfread_filtops;
2148 	kn->kn_hook = d;
2149 	knlist_add(&d->bd_sel.si_note, kn, 1);
2150 	BPFD_UNLOCK(d);
2151 
2152 	return (0);
2153 }
2154 
2155 static void
2156 filt_bpfdetach(struct knote *kn)
2157 {
2158 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2159 
2160 	knlist_remove(&d->bd_sel.si_note, kn, 0);
2161 }
2162 
2163 static int
2164 filt_bpfread(struct knote *kn, long hint)
2165 {
2166 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2167 	int ready;
2168 
2169 	BPFD_LOCK_ASSERT(d);
2170 	ready = bpf_ready(d);
2171 	if (ready) {
2172 		kn->kn_data = d->bd_slen;
2173 		/*
2174 		 * Ignore the hold buffer if it is being copied to user space.
2175 		 */
2176 		if (!d->bd_hbuf_in_use && d->bd_hbuf)
2177 			kn->kn_data += d->bd_hlen;
2178 	} else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2179 		callout_reset(&d->bd_callout, d->bd_rtout,
2180 		    bpf_timed_out, d);
2181 		d->bd_state = BPF_WAITING;
2182 	}
2183 
2184 	return (ready);
2185 }
2186 
2187 #define	BPF_TSTAMP_NONE		0
2188 #define	BPF_TSTAMP_FAST		1
2189 #define	BPF_TSTAMP_NORMAL	2
2190 #define	BPF_TSTAMP_EXTERN	3
2191 
2192 static int
2193 bpf_ts_quality(int tstype)
2194 {
2195 
2196 	if (tstype == BPF_T_NONE)
2197 		return (BPF_TSTAMP_NONE);
2198 	if ((tstype & BPF_T_FAST) != 0)
2199 		return (BPF_TSTAMP_FAST);
2200 
2201 	return (BPF_TSTAMP_NORMAL);
2202 }
2203 
2204 static int
2205 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2206 {
2207 	struct m_tag *tag;
2208 	int quality;
2209 
2210 	quality = bpf_ts_quality(tstype);
2211 	if (quality == BPF_TSTAMP_NONE)
2212 		return (quality);
2213 
2214 	if (m != NULL) {
2215 		tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2216 		if (tag != NULL) {
2217 			*bt = *(struct bintime *)(tag + 1);
2218 			return (BPF_TSTAMP_EXTERN);
2219 		}
2220 	}
2221 	if (quality == BPF_TSTAMP_NORMAL)
2222 		binuptime(bt);
2223 	else
2224 		getbinuptime(bt);
2225 
2226 	return (quality);
2227 }
2228 
2229 /*
2230  * Incoming linkage from device drivers.  Process the packet pkt, of length
2231  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
2232  * by each process' filter, and if accepted, stashed into the corresponding
2233  * buffer.
2234  */
2235 void
2236 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2237 {
2238 	struct epoch_tracker et;
2239 	struct bintime bt;
2240 	struct bpf_d *d;
2241 #ifdef BPF_JITTER
2242 	bpf_jit_filter *bf;
2243 #endif
2244 	u_int slen;
2245 	int gottime;
2246 
2247 	gottime = BPF_TSTAMP_NONE;
2248 	NET_EPOCH_ENTER(et);
2249 	CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2250 		counter_u64_add(d->bd_rcount, 1);
2251 		/*
2252 		 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2253 		 * is no way for the caller to indiciate to us whether this
2254 		 * packet is inbound or outbound. In the bpf_mtap() routines,
2255 		 * we use the interface pointers on the mbuf to figure it out.
2256 		 */
2257 #ifdef BPF_JITTER
2258 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2259 		if (bf != NULL)
2260 			slen = (*(bf->func))(pkt, pktlen, pktlen);
2261 		else
2262 #endif
2263 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2264 		if (slen != 0) {
2265 			/*
2266 			 * Filter matches. Let's to acquire write lock.
2267 			 */
2268 			BPFD_LOCK(d);
2269 			counter_u64_add(d->bd_fcount, 1);
2270 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2271 				gottime = bpf_gettime(&bt, d->bd_tstamp,
2272 				    NULL);
2273 #ifdef MAC
2274 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2275 #endif
2276 				catchpacket(d, pkt, pktlen, slen,
2277 				    bpf_append_bytes, &bt);
2278 			BPFD_UNLOCK(d);
2279 		}
2280 	}
2281 	NET_EPOCH_EXIT(et);
2282 }
2283 
2284 #define	BPF_CHECK_DIRECTION(d, r, i)				\
2285 	    (((d)->bd_direction == BPF_D_IN && (r) != (i)) ||	\
2286 	    ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2287 
2288 /*
2289  * Incoming linkage from device drivers, when packet is in an mbuf chain.
2290  * Locking model is explained in bpf_tap().
2291  */
2292 void
2293 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2294 {
2295 	struct epoch_tracker et;
2296 	struct bintime bt;
2297 	struct bpf_d *d;
2298 #ifdef BPF_JITTER
2299 	bpf_jit_filter *bf;
2300 #endif
2301 	u_int pktlen, slen;
2302 	int gottime;
2303 
2304 	/* Skip outgoing duplicate packets. */
2305 	if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2306 		m->m_flags &= ~M_PROMISC;
2307 		return;
2308 	}
2309 
2310 	pktlen = m_length(m, NULL);
2311 	gottime = BPF_TSTAMP_NONE;
2312 
2313 	NET_EPOCH_ENTER(et);
2314 	CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2315 		if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2316 			continue;
2317 		counter_u64_add(d->bd_rcount, 1);
2318 #ifdef BPF_JITTER
2319 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2320 		/* XXX We cannot handle multiple mbufs. */
2321 		if (bf != NULL && m->m_next == NULL)
2322 			slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2323 			    pktlen);
2324 		else
2325 #endif
2326 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2327 		if (slen != 0) {
2328 			BPFD_LOCK(d);
2329 
2330 			counter_u64_add(d->bd_fcount, 1);
2331 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2332 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2333 #ifdef MAC
2334 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2335 #endif
2336 				catchpacket(d, (u_char *)m, pktlen, slen,
2337 				    bpf_append_mbuf, &bt);
2338 			BPFD_UNLOCK(d);
2339 		}
2340 	}
2341 	NET_EPOCH_EXIT(et);
2342 }
2343 
2344 /*
2345  * Incoming linkage from device drivers, when packet is in
2346  * an mbuf chain and to be prepended by a contiguous header.
2347  */
2348 void
2349 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2350 {
2351 	struct epoch_tracker et;
2352 	struct bintime bt;
2353 	struct mbuf mb;
2354 	struct bpf_d *d;
2355 	u_int pktlen, slen;
2356 	int gottime;
2357 
2358 	/* Skip outgoing duplicate packets. */
2359 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2360 		m->m_flags &= ~M_PROMISC;
2361 		return;
2362 	}
2363 
2364 	pktlen = m_length(m, NULL);
2365 	/*
2366 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
2367 	 * Note that we cut corners here; we only setup what's
2368 	 * absolutely needed--this mbuf should never go anywhere else.
2369 	 */
2370 	mb.m_flags = 0;
2371 	mb.m_next = m;
2372 	mb.m_data = data;
2373 	mb.m_len = dlen;
2374 	pktlen += dlen;
2375 
2376 	gottime = BPF_TSTAMP_NONE;
2377 
2378 	NET_EPOCH_ENTER(et);
2379 	CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2380 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2381 			continue;
2382 		counter_u64_add(d->bd_rcount, 1);
2383 		slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2384 		if (slen != 0) {
2385 			BPFD_LOCK(d);
2386 
2387 			counter_u64_add(d->bd_fcount, 1);
2388 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2389 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2390 #ifdef MAC
2391 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2392 #endif
2393 				catchpacket(d, (u_char *)&mb, pktlen, slen,
2394 				    bpf_append_mbuf, &bt);
2395 			BPFD_UNLOCK(d);
2396 		}
2397 	}
2398 	NET_EPOCH_EXIT(et);
2399 }
2400 
2401 #undef	BPF_CHECK_DIRECTION
2402 #undef	BPF_TSTAMP_NONE
2403 #undef	BPF_TSTAMP_FAST
2404 #undef	BPF_TSTAMP_NORMAL
2405 #undef	BPF_TSTAMP_EXTERN
2406 
2407 static int
2408 bpf_hdrlen(struct bpf_d *d)
2409 {
2410 	int hdrlen;
2411 
2412 	hdrlen = d->bd_bif->bif_hdrlen;
2413 #ifndef BURN_BRIDGES
2414 	if (d->bd_tstamp == BPF_T_NONE ||
2415 	    BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2416 #ifdef COMPAT_FREEBSD32
2417 		if (d->bd_compat32)
2418 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2419 		else
2420 #endif
2421 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2422 	else
2423 #endif
2424 		hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2425 #ifdef COMPAT_FREEBSD32
2426 	if (d->bd_compat32)
2427 		hdrlen = BPF_WORDALIGN32(hdrlen);
2428 	else
2429 #endif
2430 		hdrlen = BPF_WORDALIGN(hdrlen);
2431 
2432 	return (hdrlen - d->bd_bif->bif_hdrlen);
2433 }
2434 
2435 static void
2436 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2437 {
2438 	struct bintime bt2, boottimebin;
2439 	struct timeval tsm;
2440 	struct timespec tsn;
2441 
2442 	if ((tstype & BPF_T_MONOTONIC) == 0) {
2443 		bt2 = *bt;
2444 		getboottimebin(&boottimebin);
2445 		bintime_add(&bt2, &boottimebin);
2446 		bt = &bt2;
2447 	}
2448 	switch (BPF_T_FORMAT(tstype)) {
2449 	case BPF_T_MICROTIME:
2450 		bintime2timeval(bt, &tsm);
2451 		ts->bt_sec = tsm.tv_sec;
2452 		ts->bt_frac = tsm.tv_usec;
2453 		break;
2454 	case BPF_T_NANOTIME:
2455 		bintime2timespec(bt, &tsn);
2456 		ts->bt_sec = tsn.tv_sec;
2457 		ts->bt_frac = tsn.tv_nsec;
2458 		break;
2459 	case BPF_T_BINTIME:
2460 		ts->bt_sec = bt->sec;
2461 		ts->bt_frac = bt->frac;
2462 		break;
2463 	}
2464 }
2465 
2466 /*
2467  * Move the packet data from interface memory (pkt) into the
2468  * store buffer.  "cpfn" is the routine called to do the actual data
2469  * transfer.  bcopy is passed in to copy contiguous chunks, while
2470  * bpf_append_mbuf is passed in to copy mbuf chains.  In the latter case,
2471  * pkt is really an mbuf.
2472  */
2473 static void
2474 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2475     void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2476     struct bintime *bt)
2477 {
2478 	struct bpf_xhdr hdr;
2479 #ifndef BURN_BRIDGES
2480 	struct bpf_hdr hdr_old;
2481 #ifdef COMPAT_FREEBSD32
2482 	struct bpf_hdr32 hdr32_old;
2483 #endif
2484 #endif
2485 	int caplen, curlen, hdrlen, totlen;
2486 	int do_wakeup = 0;
2487 	int do_timestamp;
2488 	int tstype;
2489 
2490 	BPFD_LOCK_ASSERT(d);
2491 	if (d->bd_bif == NULL) {
2492 		/* Descriptor was detached in concurrent thread */
2493 		counter_u64_add(d->bd_dcount, 1);
2494 		return;
2495 	}
2496 
2497 	/*
2498 	 * Detect whether user space has released a buffer back to us, and if
2499 	 * so, move it from being a hold buffer to a free buffer.  This may
2500 	 * not be the best place to do it (for example, we might only want to
2501 	 * run this check if we need the space), but for now it's a reliable
2502 	 * spot to do it.
2503 	 */
2504 	if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2505 		d->bd_fbuf = d->bd_hbuf;
2506 		d->bd_hbuf = NULL;
2507 		d->bd_hlen = 0;
2508 		bpf_buf_reclaimed(d);
2509 	}
2510 
2511 	/*
2512 	 * Figure out how many bytes to move.  If the packet is
2513 	 * greater or equal to the snapshot length, transfer that
2514 	 * much.  Otherwise, transfer the whole packet (unless
2515 	 * we hit the buffer size limit).
2516 	 */
2517 	hdrlen = bpf_hdrlen(d);
2518 	totlen = hdrlen + min(snaplen, pktlen);
2519 	if (totlen > d->bd_bufsize)
2520 		totlen = d->bd_bufsize;
2521 
2522 	/*
2523 	 * Round up the end of the previous packet to the next longword.
2524 	 *
2525 	 * Drop the packet if there's no room and no hope of room
2526 	 * If the packet would overflow the storage buffer or the storage
2527 	 * buffer is considered immutable by the buffer model, try to rotate
2528 	 * the buffer and wakeup pending processes.
2529 	 */
2530 #ifdef COMPAT_FREEBSD32
2531 	if (d->bd_compat32)
2532 		curlen = BPF_WORDALIGN32(d->bd_slen);
2533 	else
2534 #endif
2535 		curlen = BPF_WORDALIGN(d->bd_slen);
2536 	if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2537 		if (d->bd_fbuf == NULL) {
2538 			/*
2539 			 * There's no room in the store buffer, and no
2540 			 * prospect of room, so drop the packet.  Notify the
2541 			 * buffer model.
2542 			 */
2543 			bpf_buffull(d);
2544 			counter_u64_add(d->bd_dcount, 1);
2545 			return;
2546 		}
2547 		KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2548 		ROTATE_BUFFERS(d);
2549 		do_wakeup = 1;
2550 		curlen = 0;
2551 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2552 		/*
2553 		 * Immediate mode is set, or the read timeout has already
2554 		 * expired during a select call.  A packet arrived, so the
2555 		 * reader should be woken up.
2556 		 */
2557 		do_wakeup = 1;
2558 	caplen = totlen - hdrlen;
2559 	tstype = d->bd_tstamp;
2560 	do_timestamp = tstype != BPF_T_NONE;
2561 #ifndef BURN_BRIDGES
2562 	if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2563 		struct bpf_ts ts;
2564 		if (do_timestamp)
2565 			bpf_bintime2ts(bt, &ts, tstype);
2566 #ifdef COMPAT_FREEBSD32
2567 		if (d->bd_compat32) {
2568 			bzero(&hdr32_old, sizeof(hdr32_old));
2569 			if (do_timestamp) {
2570 				hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2571 				hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2572 			}
2573 			hdr32_old.bh_datalen = pktlen;
2574 			hdr32_old.bh_hdrlen = hdrlen;
2575 			hdr32_old.bh_caplen = caplen;
2576 			bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2577 			    sizeof(hdr32_old));
2578 			goto copy;
2579 		}
2580 #endif
2581 		bzero(&hdr_old, sizeof(hdr_old));
2582 		if (do_timestamp) {
2583 			hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2584 			hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2585 		}
2586 		hdr_old.bh_datalen = pktlen;
2587 		hdr_old.bh_hdrlen = hdrlen;
2588 		hdr_old.bh_caplen = caplen;
2589 		bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2590 		    sizeof(hdr_old));
2591 		goto copy;
2592 	}
2593 #endif
2594 
2595 	/*
2596 	 * Append the bpf header.  Note we append the actual header size, but
2597 	 * move forward the length of the header plus padding.
2598 	 */
2599 	bzero(&hdr, sizeof(hdr));
2600 	if (do_timestamp)
2601 		bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2602 	hdr.bh_datalen = pktlen;
2603 	hdr.bh_hdrlen = hdrlen;
2604 	hdr.bh_caplen = caplen;
2605 	bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2606 
2607 	/*
2608 	 * Copy the packet data into the store buffer and update its length.
2609 	 */
2610 #ifndef BURN_BRIDGES
2611 copy:
2612 #endif
2613 	(*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2614 	d->bd_slen = curlen + totlen;
2615 
2616 	if (do_wakeup)
2617 		bpf_wakeup(d);
2618 }
2619 
2620 /*
2621  * Free buffers currently in use by a descriptor.
2622  * Called on close.
2623  */
2624 static void
2625 bpfd_free(epoch_context_t ctx)
2626 {
2627 	struct bpf_d *d;
2628 	struct bpf_program_buffer *p;
2629 
2630 	/*
2631 	 * We don't need to lock out interrupts since this descriptor has
2632 	 * been detached from its interface and it yet hasn't been marked
2633 	 * free.
2634 	 */
2635 	d = __containerof(ctx, struct bpf_d, epoch_ctx);
2636 	bpf_free(d);
2637 	if (d->bd_rfilter != NULL) {
2638 		p = __containerof((void *)d->bd_rfilter,
2639 		    struct bpf_program_buffer, buffer);
2640 #ifdef BPF_JITTER
2641 		p->func = d->bd_bfilter;
2642 #endif
2643 		bpf_program_buffer_free(&p->epoch_ctx);
2644 	}
2645 	if (d->bd_wfilter != NULL) {
2646 		p = __containerof((void *)d->bd_wfilter,
2647 		    struct bpf_program_buffer, buffer);
2648 #ifdef BPF_JITTER
2649 		p->func = NULL;
2650 #endif
2651 		bpf_program_buffer_free(&p->epoch_ctx);
2652 	}
2653 
2654 	mtx_destroy(&d->bd_lock);
2655 	counter_u64_free(d->bd_rcount);
2656 	counter_u64_free(d->bd_dcount);
2657 	counter_u64_free(d->bd_fcount);
2658 	counter_u64_free(d->bd_wcount);
2659 	counter_u64_free(d->bd_wfcount);
2660 	counter_u64_free(d->bd_wdcount);
2661 	counter_u64_free(d->bd_zcopy);
2662 	free(d, M_BPF);
2663 }
2664 
2665 /*
2666  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
2667  * fixed size of the link header (variable length headers not yet supported).
2668  */
2669 void
2670 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2671 {
2672 
2673 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2674 }
2675 
2676 /*
2677  * Attach an interface to bpf.  ifp is a pointer to the structure
2678  * defining the interface to be attached, dlt is the link layer type,
2679  * and hdrlen is the fixed size of the link header (variable length
2680  * headers are not yet supporrted).
2681  */
2682 void
2683 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2684     struct bpf_if **driverp)
2685 {
2686 	struct bpf_if *bp;
2687 
2688 	KASSERT(*driverp == NULL,
2689 	    ("bpfattach2: driverp already initialized"));
2690 
2691 	bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2692 
2693 	CK_LIST_INIT(&bp->bif_dlist);
2694 	CK_LIST_INIT(&bp->bif_wlist);
2695 	bp->bif_ifp = ifp;
2696 	bp->bif_dlt = dlt;
2697 	bp->bif_hdrlen = hdrlen;
2698 	bp->bif_bpf = driverp;
2699 	bp->bif_refcnt = 1;
2700 	*driverp = bp;
2701 	/*
2702 	 * Reference ifnet pointer, so it won't freed until
2703 	 * we release it.
2704 	 */
2705 	if_ref(ifp);
2706 	BPF_LOCK();
2707 	CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2708 	BPF_UNLOCK();
2709 
2710 	if (bootverbose && IS_DEFAULT_VNET(curvnet))
2711 		if_printf(ifp, "bpf attached\n");
2712 }
2713 
2714 #ifdef VIMAGE
2715 /*
2716  * When moving interfaces between vnet instances we need a way to
2717  * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2718  * after the vmove.  We unfortunately have no device driver infrastructure
2719  * to query the interface for these values after creation/attach, thus
2720  * add this as a workaround.
2721  */
2722 int
2723 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2724 {
2725 
2726 	if (bp == NULL)
2727 		return (ENXIO);
2728 	if (bif_dlt == NULL && bif_hdrlen == NULL)
2729 		return (0);
2730 
2731 	if (bif_dlt != NULL)
2732 		*bif_dlt = bp->bif_dlt;
2733 	if (bif_hdrlen != NULL)
2734 		*bif_hdrlen = bp->bif_hdrlen;
2735 
2736 	return (0);
2737 }
2738 #endif
2739 
2740 /*
2741  * Detach bpf from an interface. This involves detaching each descriptor
2742  * associated with the interface. Notify each descriptor as it's detached
2743  * so that any sleepers wake up and get ENXIO.
2744  */
2745 void
2746 bpfdetach(struct ifnet *ifp)
2747 {
2748 	struct bpf_if *bp, *bp_temp;
2749 	struct bpf_d *d;
2750 
2751 	BPF_LOCK();
2752 	/* Find all bpf_if struct's which reference ifp and detach them. */
2753 	CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2754 		if (ifp != bp->bif_ifp)
2755 			continue;
2756 
2757 		CK_LIST_REMOVE(bp, bif_next);
2758 		*bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
2759 
2760 		CTR4(KTR_NET,
2761 		    "%s: sheduling free for encap %d (%p) for if %p",
2762 		    __func__, bp->bif_dlt, bp, ifp);
2763 
2764 		/* Detach common descriptors */
2765 		while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2766 			bpf_detachd_locked(d, true);
2767 		}
2768 
2769 		/* Detach writer-only descriptors */
2770 		while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2771 			bpf_detachd_locked(d, true);
2772 		}
2773 		bpfif_rele(bp);
2774 	}
2775 	BPF_UNLOCK();
2776 }
2777 
2778 /*
2779  * Get a list of available data link type of the interface.
2780  */
2781 static int
2782 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2783 {
2784 	struct ifnet *ifp;
2785 	struct bpf_if *bp;
2786 	u_int *lst;
2787 	int error, n, n1;
2788 
2789 	BPF_LOCK_ASSERT();
2790 
2791 	ifp = d->bd_bif->bif_ifp;
2792 	n1 = 0;
2793 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2794 		if (bp->bif_ifp == ifp)
2795 			n1++;
2796 	}
2797 	if (bfl->bfl_list == NULL) {
2798 		bfl->bfl_len = n1;
2799 		return (0);
2800 	}
2801 	if (n1 > bfl->bfl_len)
2802 		return (ENOMEM);
2803 
2804 	lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2805 	n = 0;
2806 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2807 		if (bp->bif_ifp != ifp)
2808 			continue;
2809 		lst[n++] = bp->bif_dlt;
2810 	}
2811 	error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2812 	free(lst, M_TEMP);
2813 	bfl->bfl_len = n;
2814 	return (error);
2815 }
2816 
2817 /*
2818  * Set the data link type of a BPF instance.
2819  */
2820 static int
2821 bpf_setdlt(struct bpf_d *d, u_int dlt)
2822 {
2823 	int error, opromisc;
2824 	struct ifnet *ifp;
2825 	struct bpf_if *bp;
2826 
2827 	BPF_LOCK_ASSERT();
2828 	MPASS(d->bd_bif != NULL);
2829 
2830 	/*
2831 	 * It is safe to check bd_bif without BPFD_LOCK, it can not be
2832 	 * changed while we hold global lock.
2833 	 */
2834 	if (d->bd_bif->bif_dlt == dlt)
2835 		return (0);
2836 
2837 	ifp = d->bd_bif->bif_ifp;
2838 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2839 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2840 			break;
2841 	}
2842 	if (bp == NULL)
2843 		return (EINVAL);
2844 
2845 	opromisc = d->bd_promisc;
2846 	bpf_attachd(d, bp);
2847 	if (opromisc) {
2848 		error = ifpromisc(bp->bif_ifp, 1);
2849 		if (error)
2850 			if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
2851 			    __func__, error);
2852 		else
2853 			d->bd_promisc = 1;
2854 	}
2855 	return (0);
2856 }
2857 
2858 static void
2859 bpf_drvinit(void *unused)
2860 {
2861 	struct cdev *dev;
2862 
2863 	sx_init(&bpf_sx, "bpf global lock");
2864 	CK_LIST_INIT(&bpf_iflist);
2865 
2866 	dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2867 	/* For compatibility */
2868 	make_dev_alias(dev, "bpf0");
2869 }
2870 
2871 /*
2872  * Zero out the various packet counters associated with all of the bpf
2873  * descriptors.  At some point, we will probably want to get a bit more
2874  * granular and allow the user to specify descriptors to be zeroed.
2875  */
2876 static void
2877 bpf_zero_counters(void)
2878 {
2879 	struct bpf_if *bp;
2880 	struct bpf_d *bd;
2881 
2882 	BPF_LOCK();
2883 	/*
2884 	 * We are protected by global lock here, interfaces and
2885 	 * descriptors can not be deleted while we hold it.
2886 	 */
2887 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2888 		CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2889 			counter_u64_zero(bd->bd_rcount);
2890 			counter_u64_zero(bd->bd_dcount);
2891 			counter_u64_zero(bd->bd_fcount);
2892 			counter_u64_zero(bd->bd_wcount);
2893 			counter_u64_zero(bd->bd_wfcount);
2894 			counter_u64_zero(bd->bd_zcopy);
2895 		}
2896 	}
2897 	BPF_UNLOCK();
2898 }
2899 
2900 /*
2901  * Fill filter statistics
2902  */
2903 static void
2904 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2905 {
2906 
2907 	BPF_LOCK_ASSERT();
2908 	bzero(d, sizeof(*d));
2909 	d->bd_structsize = sizeof(*d);
2910 	d->bd_immediate = bd->bd_immediate;
2911 	d->bd_promisc = bd->bd_promisc;
2912 	d->bd_hdrcmplt = bd->bd_hdrcmplt;
2913 	d->bd_direction = bd->bd_direction;
2914 	d->bd_feedback = bd->bd_feedback;
2915 	d->bd_async = bd->bd_async;
2916 	d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
2917 	d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
2918 	d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
2919 	d->bd_sig = bd->bd_sig;
2920 	d->bd_slen = bd->bd_slen;
2921 	d->bd_hlen = bd->bd_hlen;
2922 	d->bd_bufsize = bd->bd_bufsize;
2923 	d->bd_pid = bd->bd_pid;
2924 	strlcpy(d->bd_ifname,
2925 	    bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2926 	d->bd_locked = bd->bd_locked;
2927 	d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
2928 	d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
2929 	d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
2930 	d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
2931 	d->bd_bufmode = bd->bd_bufmode;
2932 }
2933 
2934 /*
2935  * Handle `netstat -B' stats request
2936  */
2937 static int
2938 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2939 {
2940 	static const struct xbpf_d zerostats;
2941 	struct xbpf_d *xbdbuf, *xbd, tempstats;
2942 	int index, error;
2943 	struct bpf_if *bp;
2944 	struct bpf_d *bd;
2945 
2946 	/*
2947 	 * XXX This is not technically correct. It is possible for non
2948 	 * privileged users to open bpf devices. It would make sense
2949 	 * if the users who opened the devices were able to retrieve
2950 	 * the statistics for them, too.
2951 	 */
2952 	error = priv_check(req->td, PRIV_NET_BPF);
2953 	if (error)
2954 		return (error);
2955 	/*
2956 	 * Check to see if the user is requesting that the counters be
2957 	 * zeroed out.  Explicitly check that the supplied data is zeroed,
2958 	 * as we aren't allowing the user to set the counters currently.
2959 	 */
2960 	if (req->newptr != NULL) {
2961 		if (req->newlen != sizeof(tempstats))
2962 			return (EINVAL);
2963 		memset(&tempstats, 0, sizeof(tempstats));
2964 		error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2965 		if (error)
2966 			return (error);
2967 		if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2968 			return (EINVAL);
2969 		bpf_zero_counters();
2970 		return (0);
2971 	}
2972 	if (req->oldptr == NULL)
2973 		return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2974 	if (bpf_bpfd_cnt == 0)
2975 		return (SYSCTL_OUT(req, 0, 0));
2976 	xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2977 	BPF_LOCK();
2978 	if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2979 		BPF_UNLOCK();
2980 		free(xbdbuf, M_BPF);
2981 		return (ENOMEM);
2982 	}
2983 	index = 0;
2984 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2985 		/* Send writers-only first */
2986 		CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2987 			xbd = &xbdbuf[index++];
2988 			bpfstats_fill_xbpf(xbd, bd);
2989 		}
2990 		CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2991 			xbd = &xbdbuf[index++];
2992 			bpfstats_fill_xbpf(xbd, bd);
2993 		}
2994 	}
2995 	BPF_UNLOCK();
2996 	error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2997 	free(xbdbuf, M_BPF);
2998 	return (error);
2999 }
3000 
3001 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
3002 
3003 #else /* !DEV_BPF && !NETGRAPH_BPF */
3004 
3005 /*
3006  * NOP stubs to allow bpf-using drivers to load and function.
3007  *
3008  * A 'better' implementation would allow the core bpf functionality
3009  * to be loaded at runtime.
3010  */
3011 
3012 void
3013 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3014 {
3015 }
3016 
3017 void
3018 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3019 {
3020 }
3021 
3022 void
3023 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3024 {
3025 }
3026 
3027 void
3028 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3029 {
3030 
3031 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3032 }
3033 
3034 void
3035 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3036 {
3037 
3038 	*driverp = (struct bpf_if *)&dead_bpf_if;
3039 }
3040 
3041 void
3042 bpfdetach(struct ifnet *ifp)
3043 {
3044 }
3045 
3046 u_int
3047 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3048 {
3049 	return -1;	/* "no filter" behaviour */
3050 }
3051 
3052 int
3053 bpf_validate(const struct bpf_insn *f, int len)
3054 {
3055 	return 0;		/* false */
3056 }
3057 
3058 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3059 
3060 #ifdef DDB
3061 static void
3062 bpf_show_bpf_if(struct bpf_if *bpf_if)
3063 {
3064 
3065 	if (bpf_if == NULL)
3066 		return;
3067 	db_printf("%p:\n", bpf_if);
3068 #define	BPF_DB_PRINTF(f, e)	db_printf("   %s = " f "\n", #e, bpf_if->e);
3069 	/* bif_ext.bif_next */
3070 	/* bif_ext.bif_dlist */
3071 	BPF_DB_PRINTF("%#x", bif_dlt);
3072 	BPF_DB_PRINTF("%u", bif_hdrlen);
3073 	/* bif_wlist */
3074 	BPF_DB_PRINTF("%p", bif_ifp);
3075 	BPF_DB_PRINTF("%p", bif_bpf);
3076 	BPF_DB_PRINTF("%u", bif_refcnt);
3077 }
3078 
3079 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3080 {
3081 
3082 	if (!have_addr) {
3083 		db_printf("usage: show bpf_if <struct bpf_if *>\n");
3084 		return;
3085 	}
3086 
3087 	bpf_show_bpf_if((struct bpf_if *)addr);
3088 }
3089 #endif
3090