xref: /freebsd/sys/net/bpf.c (revision 657729a89dd578d8cfc70d6616f5c65a48a8b33a)
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_vlan_var.h>
82 #include <net/if_dl.h>
83 #include <net/bpf.h>
84 #include <net/bpf_buffer.h>
85 #ifdef BPF_JITTER
86 #include <net/bpf_jitter.h>
87 #endif
88 #include <net/bpf_zerocopy.h>
89 #include <net/bpfdesc.h>
90 #include <net/route.h>
91 #include <net/vnet.h>
92 
93 #include <netinet/in.h>
94 #include <netinet/if_ether.h>
95 #include <sys/kernel.h>
96 #include <sys/sysctl.h>
97 
98 #include <net80211/ieee80211_freebsd.h>
99 
100 #include <security/mac/mac_framework.h>
101 
102 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
103 
104 static struct bpf_if_ext dead_bpf_if = {
105 	.bif_dlist = CK_LIST_HEAD_INITIALIZER()
106 };
107 
108 struct bpf_if {
109 #define	bif_next	bif_ext.bif_next
110 #define	bif_dlist	bif_ext.bif_dlist
111 	struct bpf_if_ext bif_ext;	/* public members */
112 	u_int		bif_dlt;	/* link layer type */
113 	u_int		bif_hdrlen;	/* length of link header */
114 	struct bpfd_list bif_wlist;	/* writer-only list */
115 	struct ifnet	*bif_ifp;	/* corresponding interface */
116 	struct bpf_if	**bif_bpf;	/* Pointer to pointer to us */
117 	volatile u_int	bif_refcnt;
118 	struct epoch_context epoch_ctx;
119 };
120 
121 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
122 
123 struct bpf_program_buffer {
124 	struct epoch_context	epoch_ctx;
125 #ifdef BPF_JITTER
126 	bpf_jit_filter		*func;
127 #endif
128 	void			*buffer[0];
129 };
130 
131 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
132 
133 #define PRINET  26			/* interruptible */
134 #define BPF_PRIO_MAX	7
135 
136 #define	SIZEOF_BPF_HDR(type)	\
137     (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
138 
139 #ifdef COMPAT_FREEBSD32
140 #include <sys/mount.h>
141 #include <compat/freebsd32/freebsd32.h>
142 #define BPF_ALIGNMENT32 sizeof(int32_t)
143 #define	BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
144 
145 #ifndef BURN_BRIDGES
146 /*
147  * 32-bit version of structure prepended to each packet.  We use this header
148  * instead of the standard one for 32-bit streams.  We mark the a stream as
149  * 32-bit the first time we see a 32-bit compat ioctl request.
150  */
151 struct bpf_hdr32 {
152 	struct timeval32 bh_tstamp;	/* time stamp */
153 	uint32_t	bh_caplen;	/* length of captured portion */
154 	uint32_t	bh_datalen;	/* original length of packet */
155 	uint16_t	bh_hdrlen;	/* length of bpf header (this struct
156 					   plus alignment padding) */
157 };
158 #endif
159 
160 struct bpf_program32 {
161 	u_int bf_len;
162 	uint32_t bf_insns;
163 };
164 
165 struct bpf_dltlist32 {
166 	u_int	bfl_len;
167 	u_int	bfl_list;
168 };
169 
170 #define	BIOCSETF32	_IOW('B', 103, struct bpf_program32)
171 #define	BIOCSRTIMEOUT32	_IOW('B', 109, struct timeval32)
172 #define	BIOCGRTIMEOUT32	_IOR('B', 110, struct timeval32)
173 #define	BIOCGDLTLIST32	_IOWR('B', 121, struct bpf_dltlist32)
174 #define	BIOCSETWF32	_IOW('B', 123, struct bpf_program32)
175 #define	BIOCSETFNR32	_IOW('B', 130, struct bpf_program32)
176 #endif
177 
178 #define BPF_LOCK()	   sx_xlock(&bpf_sx)
179 #define BPF_UNLOCK()		sx_xunlock(&bpf_sx)
180 #define BPF_LOCK_ASSERT()	sx_assert(&bpf_sx, SA_XLOCKED)
181 /*
182  * bpf_iflist is a list of BPF interface structures, each corresponding to a
183  * specific DLT. The same network interface might have several BPF interface
184  * structures registered by different layers in the stack (i.e., 802.11
185  * frames, ethernet frames, etc).
186  */
187 CK_LIST_HEAD(bpf_iflist, bpf_if);
188 static struct bpf_iflist bpf_iflist;
189 static struct sx	bpf_sx;		/* bpf global lock */
190 static int		bpf_bpfd_cnt;
191 
192 static void	bpfif_ref(struct bpf_if *);
193 static void	bpfif_rele(struct bpf_if *);
194 
195 static void	bpfd_ref(struct bpf_d *);
196 static void	bpfd_rele(struct bpf_d *);
197 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
198 static void	bpf_detachd(struct bpf_d *);
199 static void	bpf_detachd_locked(struct bpf_d *, bool);
200 static void	bpfd_free(epoch_context_t);
201 static int	bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
202 		    struct sockaddr *, int *, struct bpf_d *);
203 static int	bpf_setif(struct bpf_d *, struct ifreq *);
204 static void	bpf_timed_out(void *);
205 static __inline void
206 		bpf_wakeup(struct bpf_d *);
207 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
208 		    void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
209 		    struct bintime *);
210 static void	reset_d(struct bpf_d *);
211 static int	bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
212 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
213 static int	bpf_setdlt(struct bpf_d *, u_int);
214 static void	filt_bpfdetach(struct knote *);
215 static int	filt_bpfread(struct knote *, long);
216 static int	filt_bpfwrite(struct knote *, long);
217 static void	bpf_drvinit(void *);
218 static int	bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
219 
220 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
221     "bpf sysctl");
222 int bpf_maxinsns = BPF_MAXINSNS;
223 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
224     &bpf_maxinsns, 0, "Maximum bpf program instructions");
225 static int bpf_zerocopy_enable = 0;
226 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
227     &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
228 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
229     bpf_stats_sysctl, "bpf statistics portal");
230 
231 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
232 #define	V_bpf_optimize_writers VNET(bpf_optimize_writers)
233 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RWTUN,
234     &VNET_NAME(bpf_optimize_writers), 0,
235     "Do not send packets until BPF program is set");
236 
237 static	d_open_t	bpfopen;
238 static	d_read_t	bpfread;
239 static	d_write_t	bpfwrite;
240 static	d_ioctl_t	bpfioctl;
241 static	d_poll_t	bpfpoll;
242 static	d_kqfilter_t	bpfkqfilter;
243 
244 static struct cdevsw bpf_cdevsw = {
245 	.d_version =	D_VERSION,
246 	.d_open =	bpfopen,
247 	.d_read =	bpfread,
248 	.d_write =	bpfwrite,
249 	.d_ioctl =	bpfioctl,
250 	.d_poll =	bpfpoll,
251 	.d_name =	"bpf",
252 	.d_kqfilter =	bpfkqfilter,
253 };
254 
255 static struct filterops bpfread_filtops = {
256 	.f_isfd = 1,
257 	.f_detach = filt_bpfdetach,
258 	.f_event = filt_bpfread,
259 };
260 
261 static struct filterops bpfwrite_filtops = {
262 	.f_isfd = 1,
263 	.f_detach = filt_bpfdetach,
264 	.f_event = filt_bpfwrite,
265 };
266 
267 /*
268  * LOCKING MODEL USED BY BPF
269  *
270  * Locks:
271  * 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
272  * every bpf_iflist changes, serializes ioctl access to bpf descriptors.
273  * 2) Descriptor lock. Mutex, used to protect BPF buffers and various
274  * structure fields used by bpf_*tap* code.
275  *
276  * Lock order: global lock, then descriptor lock.
277  *
278  * There are several possible consumers:
279  *
280  * 1. The kernel registers interface pointer with bpfattach().
281  * Each call allocates new bpf_if structure, references ifnet pointer
282  * and links bpf_if into bpf_iflist chain. This is protected with global
283  * lock.
284  *
285  * 2. An userland application uses ioctl() call to bpf_d descriptor.
286  * All such call are serialized with global lock. BPF filters can be
287  * changed, but pointer to old filter will be freed using NET_EPOCH_CALL().
288  * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
289  * filter pointers, even if change will happen during bpf_tap execution.
290  * Destroying of bpf_d descriptor also is doing using NET_EPOCH_CALL().
291  *
292  * 3. An userland application can write packets into bpf_d descriptor.
293  * There we need to be sure, that ifnet won't disappear during bpfwrite().
294  *
295  * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
296  * bif_dlist is protected with net_epoch_preempt section. So, it should
297  * be safe to make access to bpf_d descriptor inside the section.
298  *
299  * 5. The kernel invokes bpfdetach() on interface destroying. All lists
300  * are modified with global lock held and actual free() is done using
301  * NET_EPOCH_CALL().
302  */
303 
304 static void
305 bpfif_free(epoch_context_t ctx)
306 {
307 	struct bpf_if *bp;
308 
309 	bp = __containerof(ctx, struct bpf_if, epoch_ctx);
310 	if_rele(bp->bif_ifp);
311 	free(bp, M_BPF);
312 }
313 
314 static void
315 bpfif_ref(struct bpf_if *bp)
316 {
317 
318 	refcount_acquire(&bp->bif_refcnt);
319 }
320 
321 static void
322 bpfif_rele(struct bpf_if *bp)
323 {
324 
325 	if (!refcount_release(&bp->bif_refcnt))
326 		return;
327 	NET_EPOCH_CALL(bpfif_free, &bp->epoch_ctx);
328 }
329 
330 static void
331 bpfd_ref(struct bpf_d *d)
332 {
333 
334 	refcount_acquire(&d->bd_refcnt);
335 }
336 
337 static void
338 bpfd_rele(struct bpf_d *d)
339 {
340 
341 	if (!refcount_release(&d->bd_refcnt))
342 		return;
343 	NET_EPOCH_CALL(bpfd_free, &d->epoch_ctx);
344 }
345 
346 static struct bpf_program_buffer*
347 bpf_program_buffer_alloc(size_t size, int flags)
348 {
349 
350 	return (malloc(sizeof(struct bpf_program_buffer) + size,
351 	    M_BPF, flags));
352 }
353 
354 static void
355 bpf_program_buffer_free(epoch_context_t ctx)
356 {
357 	struct bpf_program_buffer *ptr;
358 
359 	ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
360 #ifdef BPF_JITTER
361 	if (ptr->func != NULL)
362 		bpf_destroy_jit_filter(ptr->func);
363 #endif
364 	free(ptr, M_BPF);
365 }
366 
367 /*
368  * Wrapper functions for various buffering methods.  If the set of buffer
369  * modes expands, we will probably want to introduce a switch data structure
370  * similar to protosw, et.
371  */
372 static void
373 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
374     u_int len)
375 {
376 
377 	BPFD_LOCK_ASSERT(d);
378 
379 	switch (d->bd_bufmode) {
380 	case BPF_BUFMODE_BUFFER:
381 		return (bpf_buffer_append_bytes(d, buf, offset, src, len));
382 
383 	case BPF_BUFMODE_ZBUF:
384 		counter_u64_add(d->bd_zcopy, 1);
385 		return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
386 
387 	default:
388 		panic("bpf_buf_append_bytes");
389 	}
390 }
391 
392 static void
393 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
394     u_int len)
395 {
396 
397 	BPFD_LOCK_ASSERT(d);
398 
399 	switch (d->bd_bufmode) {
400 	case BPF_BUFMODE_BUFFER:
401 		return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
402 
403 	case BPF_BUFMODE_ZBUF:
404 		counter_u64_add(d->bd_zcopy, 1);
405 		return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
406 
407 	default:
408 		panic("bpf_buf_append_mbuf");
409 	}
410 }
411 
412 /*
413  * This function gets called when the free buffer is re-assigned.
414  */
415 static void
416 bpf_buf_reclaimed(struct bpf_d *d)
417 {
418 
419 	BPFD_LOCK_ASSERT(d);
420 
421 	switch (d->bd_bufmode) {
422 	case BPF_BUFMODE_BUFFER:
423 		return;
424 
425 	case BPF_BUFMODE_ZBUF:
426 		bpf_zerocopy_buf_reclaimed(d);
427 		return;
428 
429 	default:
430 		panic("bpf_buf_reclaimed");
431 	}
432 }
433 
434 /*
435  * If the buffer mechanism has a way to decide that a held buffer can be made
436  * free, then it is exposed via the bpf_canfreebuf() interface.  (1) is
437  * returned if the buffer can be discarded, (0) is returned if it cannot.
438  */
439 static int
440 bpf_canfreebuf(struct bpf_d *d)
441 {
442 
443 	BPFD_LOCK_ASSERT(d);
444 
445 	switch (d->bd_bufmode) {
446 	case BPF_BUFMODE_ZBUF:
447 		return (bpf_zerocopy_canfreebuf(d));
448 	}
449 	return (0);
450 }
451 
452 /*
453  * Allow the buffer model to indicate that the current store buffer is
454  * immutable, regardless of the appearance of space.  Return (1) if the
455  * buffer is writable, and (0) if not.
456  */
457 static int
458 bpf_canwritebuf(struct bpf_d *d)
459 {
460 	BPFD_LOCK_ASSERT(d);
461 
462 	switch (d->bd_bufmode) {
463 	case BPF_BUFMODE_ZBUF:
464 		return (bpf_zerocopy_canwritebuf(d));
465 	}
466 	return (1);
467 }
468 
469 /*
470  * Notify buffer model that an attempt to write to the store buffer has
471  * resulted in a dropped packet, in which case the buffer may be considered
472  * full.
473  */
474 static void
475 bpf_buffull(struct bpf_d *d)
476 {
477 
478 	BPFD_LOCK_ASSERT(d);
479 
480 	switch (d->bd_bufmode) {
481 	case BPF_BUFMODE_ZBUF:
482 		bpf_zerocopy_buffull(d);
483 		break;
484 	}
485 }
486 
487 /*
488  * Notify the buffer model that a buffer has moved into the hold position.
489  */
490 void
491 bpf_bufheld(struct bpf_d *d)
492 {
493 
494 	BPFD_LOCK_ASSERT(d);
495 
496 	switch (d->bd_bufmode) {
497 	case BPF_BUFMODE_ZBUF:
498 		bpf_zerocopy_bufheld(d);
499 		break;
500 	}
501 }
502 
503 static void
504 bpf_free(struct bpf_d *d)
505 {
506 
507 	switch (d->bd_bufmode) {
508 	case BPF_BUFMODE_BUFFER:
509 		return (bpf_buffer_free(d));
510 
511 	case BPF_BUFMODE_ZBUF:
512 		return (bpf_zerocopy_free(d));
513 
514 	default:
515 		panic("bpf_buf_free");
516 	}
517 }
518 
519 static int
520 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
521 {
522 
523 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
524 		return (EOPNOTSUPP);
525 	return (bpf_buffer_uiomove(d, buf, len, uio));
526 }
527 
528 static int
529 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
530 {
531 
532 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
533 		return (EOPNOTSUPP);
534 	return (bpf_buffer_ioctl_sblen(d, i));
535 }
536 
537 static int
538 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
539 {
540 
541 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
542 		return (EOPNOTSUPP);
543 	return (bpf_zerocopy_ioctl_getzmax(td, d, i));
544 }
545 
546 static int
547 bpf_ioctl_rotzbuf(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_rotzbuf(td, d, bz));
553 }
554 
555 static int
556 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
557 {
558 
559 	if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
560 		return (EOPNOTSUPP);
561 	return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
562 }
563 
564 /*
565  * General BPF functions.
566  */
567 static int
568 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
569     struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
570 {
571 	const struct ieee80211_bpf_params *p;
572 	struct ether_header *eh;
573 	struct mbuf *m;
574 	int error;
575 	int len;
576 	int hlen;
577 	int slen;
578 
579 	/*
580 	 * Build a sockaddr based on the data link layer type.
581 	 * We do this at this level because the ethernet header
582 	 * is copied directly into the data field of the sockaddr.
583 	 * In the case of SLIP, there is no header and the packet
584 	 * is forwarded as is.
585 	 * Also, we are careful to leave room at the front of the mbuf
586 	 * for the link level header.
587 	 */
588 	switch (linktype) {
589 	case DLT_SLIP:
590 		sockp->sa_family = AF_INET;
591 		hlen = 0;
592 		break;
593 
594 	case DLT_EN10MB:
595 		sockp->sa_family = AF_UNSPEC;
596 		/* XXX Would MAXLINKHDR be better? */
597 		hlen = ETHER_HDR_LEN;
598 		break;
599 
600 	case DLT_FDDI:
601 		sockp->sa_family = AF_IMPLINK;
602 		hlen = 0;
603 		break;
604 
605 	case DLT_RAW:
606 		sockp->sa_family = AF_UNSPEC;
607 		hlen = 0;
608 		break;
609 
610 	case DLT_NULL:
611 		/*
612 		 * null interface types require a 4 byte pseudo header which
613 		 * corresponds to the address family of the packet.
614 		 */
615 		sockp->sa_family = AF_UNSPEC;
616 		hlen = 4;
617 		break;
618 
619 	case DLT_ATM_RFC1483:
620 		/*
621 		 * en atm driver requires 4-byte atm pseudo header.
622 		 * though it isn't standard, vpi:vci needs to be
623 		 * specified anyway.
624 		 */
625 		sockp->sa_family = AF_UNSPEC;
626 		hlen = 12;	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
627 		break;
628 
629 	case DLT_PPP:
630 		sockp->sa_family = AF_UNSPEC;
631 		hlen = 4;	/* This should match PPP_HDRLEN */
632 		break;
633 
634 	case DLT_IEEE802_11:		/* IEEE 802.11 wireless */
635 		sockp->sa_family = AF_IEEE80211;
636 		hlen = 0;
637 		break;
638 
639 	case DLT_IEEE802_11_RADIO:	/* IEEE 802.11 wireless w/ phy params */
640 		sockp->sa_family = AF_IEEE80211;
641 		sockp->sa_len = 12;	/* XXX != 0 */
642 		hlen = sizeof(struct ieee80211_bpf_params);
643 		break;
644 
645 	default:
646 		return (EIO);
647 	}
648 
649 	len = uio->uio_resid;
650 	if (len < hlen || len - hlen > ifp->if_mtu)
651 		return (EMSGSIZE);
652 
653 	/* Allocate a mbuf for our write, since m_get2 fails if len >= to MJUMPAGESIZE, use m_getjcl for bigger buffers */
654 	m = m_get3(len, M_WAITOK, MT_DATA, M_PKTHDR);
655 	if (m == NULL)
656 		return (EIO);
657 	m->m_pkthdr.len = m->m_len = len;
658 	*mp = m;
659 
660 	error = uiomove(mtod(m, u_char *), len, uio);
661 	if (error)
662 		goto bad;
663 
664 	slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
665 	if (slen == 0) {
666 		error = EPERM;
667 		goto bad;
668 	}
669 
670 	/* Check for multicast destination */
671 	switch (linktype) {
672 	case DLT_EN10MB:
673 		eh = mtod(m, struct ether_header *);
674 		if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
675 			if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
676 			    ETHER_ADDR_LEN) == 0)
677 				m->m_flags |= M_BCAST;
678 			else
679 				m->m_flags |= M_MCAST;
680 		}
681 		if (d->bd_hdrcmplt == 0) {
682 			memcpy(eh->ether_shost, IF_LLADDR(ifp),
683 			    sizeof(eh->ether_shost));
684 		}
685 		break;
686 	}
687 
688 	/*
689 	 * Make room for link header, and copy it to sockaddr
690 	 */
691 	if (hlen != 0) {
692 		if (sockp->sa_family == AF_IEEE80211) {
693 			/*
694 			 * Collect true length from the parameter header
695 			 * NB: sockp is known to be zero'd so if we do a
696 			 *     short copy unspecified parameters will be
697 			 *     zero.
698 			 * NB: packet may not be aligned after stripping
699 			 *     bpf params
700 			 * XXX check ibp_vers
701 			 */
702 			p = mtod(m, const struct ieee80211_bpf_params *);
703 			hlen = p->ibp_len;
704 			if (hlen > sizeof(sockp->sa_data)) {
705 				error = EINVAL;
706 				goto bad;
707 			}
708 		}
709 		bcopy(mtod(m, const void *), sockp->sa_data, hlen);
710 	}
711 	*hdrlen = hlen;
712 
713 	return (0);
714 bad:
715 	m_freem(m);
716 	return (error);
717 }
718 
719 /*
720  * Attach descriptor to the bpf interface, i.e. make d listen on bp,
721  * then reset its buffers and counters with reset_d().
722  */
723 static void
724 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
725 {
726 	int op_w;
727 
728 	BPF_LOCK_ASSERT();
729 
730 	/*
731 	 * Save sysctl value to protect from sysctl change
732 	 * between reads
733 	 */
734 	op_w = V_bpf_optimize_writers || d->bd_writer;
735 
736 	if (d->bd_bif != NULL)
737 		bpf_detachd_locked(d, false);
738 	/*
739 	 * Point d at bp, and add d to the interface's list.
740 	 * Since there are many applications using BPF for
741 	 * sending raw packets only (dhcpd, cdpd are good examples)
742 	 * we can delay adding d to the list of active listeners until
743 	 * some filter is configured.
744 	 */
745 
746 	BPFD_LOCK(d);
747 	/*
748 	 * Hold reference to bpif while descriptor uses this interface.
749 	 */
750 	bpfif_ref(bp);
751 	d->bd_bif = bp;
752 	if (op_w != 0) {
753 		/* Add to writers-only list */
754 		CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
755 		/*
756 		 * We decrement bd_writer on every filter set operation.
757 		 * First BIOCSETF is done by pcap_open_live() to set up
758 		 * snap length. After that appliation usually sets its own
759 		 * filter.
760 		 */
761 		d->bd_writer = 2;
762 	} else
763 		CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
764 
765 	reset_d(d);
766 
767 	/* Trigger EVFILT_WRITE events. */
768 	bpf_wakeup(d);
769 
770 	BPFD_UNLOCK(d);
771 	bpf_bpfd_cnt++;
772 
773 	CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
774 	    __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
775 
776 	if (op_w == 0)
777 		EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
778 }
779 
780 /*
781  * Check if we need to upgrade our descriptor @d from write-only mode.
782  */
783 static int
784 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
785     int flen)
786 {
787 	int is_snap, need_upgrade;
788 
789 	/*
790 	 * Check if we've already upgraded or new filter is empty.
791 	 */
792 	if (d->bd_writer == 0 || fcode == NULL)
793 		return (0);
794 
795 	need_upgrade = 0;
796 
797 	/*
798 	 * Check if cmd looks like snaplen setting from
799 	 * pcap_bpf.c:pcap_open_live().
800 	 * Note we're not checking .k value here:
801 	 * while pcap_open_live() definitely sets to non-zero value,
802 	 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
803 	 * do not consider upgrading immediately
804 	 */
805 	if (cmd == BIOCSETF && flen == 1 &&
806 	    fcode[0].code == (BPF_RET | BPF_K))
807 		is_snap = 1;
808 	else
809 		is_snap = 0;
810 
811 	if (is_snap == 0) {
812 		/*
813 		 * We're setting first filter and it doesn't look like
814 		 * setting snaplen.  We're probably using bpf directly.
815 		 * Upgrade immediately.
816 		 */
817 		need_upgrade = 1;
818 	} else {
819 		/*
820 		 * Do not require upgrade by first BIOCSETF
821 		 * (used to set snaplen) by pcap_open_live().
822 		 */
823 
824 		if (--d->bd_writer == 0) {
825 			/*
826 			 * First snaplen filter has already
827 			 * been set. This is probably catch-all
828 			 * filter
829 			 */
830 			need_upgrade = 1;
831 		}
832 	}
833 
834 	CTR5(KTR_NET,
835 	    "%s: filter function set by pid %d, "
836 	    "bd_writer counter %d, snap %d upgrade %d",
837 	    __func__, d->bd_pid, d->bd_writer,
838 	    is_snap, need_upgrade);
839 
840 	return (need_upgrade);
841 }
842 
843 /*
844  * Detach a file from its interface.
845  */
846 static void
847 bpf_detachd(struct bpf_d *d)
848 {
849 	BPF_LOCK();
850 	bpf_detachd_locked(d, false);
851 	BPF_UNLOCK();
852 }
853 
854 static void
855 bpf_detachd_locked(struct bpf_d *d, bool detached_ifp)
856 {
857 	struct bpf_if *bp;
858 	struct ifnet *ifp;
859 	int error;
860 
861 	BPF_LOCK_ASSERT();
862 	CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
863 
864 	/* Check if descriptor is attached */
865 	if ((bp = d->bd_bif) == NULL)
866 		return;
867 
868 	BPFD_LOCK(d);
869 	/* Remove d from the interface's descriptor list. */
870 	CK_LIST_REMOVE(d, bd_next);
871 	/* Save bd_writer value */
872 	error = d->bd_writer;
873 	ifp = bp->bif_ifp;
874 	d->bd_bif = NULL;
875 	if (detached_ifp) {
876 		/*
877 		 * Notify descriptor as it's detached, so that any
878 		 * sleepers wake up and get ENXIO.
879 		 */
880 		bpf_wakeup(d);
881 	}
882 	BPFD_UNLOCK(d);
883 	bpf_bpfd_cnt--;
884 
885 	/* Call event handler iff d is attached */
886 	if (error == 0)
887 		EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
888 
889 	/*
890 	 * Check if this descriptor had requested promiscuous mode.
891 	 * If so and ifnet is not detached, turn it off.
892 	 */
893 	if (d->bd_promisc && !detached_ifp) {
894 		d->bd_promisc = 0;
895 		CURVNET_SET(ifp->if_vnet);
896 		error = ifpromisc(ifp, 0);
897 		CURVNET_RESTORE();
898 		if (error != 0 && error != ENXIO) {
899 			/*
900 			 * ENXIO can happen if a pccard is unplugged
901 			 * Something is really wrong if we were able to put
902 			 * the driver into promiscuous mode, but can't
903 			 * take it out.
904 			 */
905 			if_printf(bp->bif_ifp,
906 				"bpf_detach: ifpromisc failed (%d)\n", error);
907 		}
908 	}
909 	bpfif_rele(bp);
910 }
911 
912 /*
913  * Close the descriptor by detaching it from its interface,
914  * deallocating its buffers, and marking it free.
915  */
916 static void
917 bpf_dtor(void *data)
918 {
919 	struct bpf_d *d = data;
920 
921 	BPFD_LOCK(d);
922 	if (d->bd_state == BPF_WAITING)
923 		callout_stop(&d->bd_callout);
924 	d->bd_state = BPF_IDLE;
925 	BPFD_UNLOCK(d);
926 	funsetown(&d->bd_sigio);
927 	bpf_detachd(d);
928 #ifdef MAC
929 	mac_bpfdesc_destroy(d);
930 #endif /* MAC */
931 	seldrain(&d->bd_sel);
932 	knlist_destroy(&d->bd_sel.si_note);
933 	callout_drain(&d->bd_callout);
934 	bpfd_rele(d);
935 }
936 
937 /*
938  * Open ethernet device.  Returns ENXIO for illegal minor device number,
939  * EBUSY if file is open by another process.
940  */
941 /* ARGSUSED */
942 static	int
943 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
944 {
945 	struct bpf_d *d;
946 	int error;
947 
948 	d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
949 	error = devfs_set_cdevpriv(d, bpf_dtor);
950 	if (error != 0) {
951 		free(d, M_BPF);
952 		return (error);
953 	}
954 
955 	/* Setup counters */
956 	d->bd_rcount = counter_u64_alloc(M_WAITOK);
957 	d->bd_dcount = counter_u64_alloc(M_WAITOK);
958 	d->bd_fcount = counter_u64_alloc(M_WAITOK);
959 	d->bd_wcount = counter_u64_alloc(M_WAITOK);
960 	d->bd_wfcount = counter_u64_alloc(M_WAITOK);
961 	d->bd_wdcount = counter_u64_alloc(M_WAITOK);
962 	d->bd_zcopy = counter_u64_alloc(M_WAITOK);
963 
964 	/*
965 	 * For historical reasons, perform a one-time initialization call to
966 	 * the buffer routines, even though we're not yet committed to a
967 	 * particular buffer method.
968 	 */
969 	bpf_buffer_init(d);
970 	if ((flags & FREAD) == 0)
971 		d->bd_writer = 2;
972 	d->bd_hbuf_in_use = 0;
973 	d->bd_bufmode = BPF_BUFMODE_BUFFER;
974 	d->bd_sig = SIGIO;
975 	d->bd_direction = BPF_D_INOUT;
976 	refcount_init(&d->bd_refcnt, 1);
977 	BPF_PID_REFRESH(d, td);
978 #ifdef MAC
979 	mac_bpfdesc_init(d);
980 	mac_bpfdesc_create(td->td_ucred, d);
981 #endif
982 	mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
983 	callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
984 	knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
985 
986 	/* Disable VLAN pcp tagging. */
987 	d->bd_pcp = 0;
988 
989 	return (0);
990 }
991 
992 /*
993  *  bpfread - read next chunk of packets from buffers
994  */
995 static	int
996 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
997 {
998 	struct bpf_d *d;
999 	int error;
1000 	int non_block;
1001 	int timed_out;
1002 
1003 	error = devfs_get_cdevpriv((void **)&d);
1004 	if (error != 0)
1005 		return (error);
1006 
1007 	/*
1008 	 * Restrict application to use a buffer the same size as
1009 	 * as kernel buffers.
1010 	 */
1011 	if (uio->uio_resid != d->bd_bufsize)
1012 		return (EINVAL);
1013 
1014 	non_block = ((ioflag & O_NONBLOCK) != 0);
1015 
1016 	BPFD_LOCK(d);
1017 	BPF_PID_REFRESH_CUR(d);
1018 	if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1019 		BPFD_UNLOCK(d);
1020 		return (EOPNOTSUPP);
1021 	}
1022 	if (d->bd_state == BPF_WAITING)
1023 		callout_stop(&d->bd_callout);
1024 	timed_out = (d->bd_state == BPF_TIMED_OUT);
1025 	d->bd_state = BPF_IDLE;
1026 	while (d->bd_hbuf_in_use) {
1027 		error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1028 		    PRINET|PCATCH, "bd_hbuf", 0);
1029 		if (error != 0) {
1030 			BPFD_UNLOCK(d);
1031 			return (error);
1032 		}
1033 	}
1034 	/*
1035 	 * If the hold buffer is empty, then do a timed sleep, which
1036 	 * ends when the timeout expires or when enough packets
1037 	 * have arrived to fill the store buffer.
1038 	 */
1039 	while (d->bd_hbuf == NULL) {
1040 		if (d->bd_slen != 0) {
1041 			/*
1042 			 * A packet(s) either arrived since the previous
1043 			 * read or arrived while we were asleep.
1044 			 */
1045 			if (d->bd_immediate || non_block || timed_out) {
1046 				/*
1047 				 * Rotate the buffers and return what's here
1048 				 * if we are in immediate mode, non-blocking
1049 				 * flag is set, or this descriptor timed out.
1050 				 */
1051 				ROTATE_BUFFERS(d);
1052 				break;
1053 			}
1054 		}
1055 
1056 		/*
1057 		 * No data is available, check to see if the bpf device
1058 		 * is still pointed at a real interface.  If not, return
1059 		 * ENXIO so that the userland process knows to rebind
1060 		 * it before using it again.
1061 		 */
1062 		if (d->bd_bif == NULL) {
1063 			BPFD_UNLOCK(d);
1064 			return (ENXIO);
1065 		}
1066 
1067 		if (non_block) {
1068 			BPFD_UNLOCK(d);
1069 			return (EWOULDBLOCK);
1070 		}
1071 		error = msleep(d, &d->bd_lock, PRINET|PCATCH,
1072 		     "bpf", d->bd_rtout);
1073 		if (error == EINTR || error == ERESTART) {
1074 			BPFD_UNLOCK(d);
1075 			return (error);
1076 		}
1077 		if (error == EWOULDBLOCK) {
1078 			/*
1079 			 * On a timeout, return what's in the buffer,
1080 			 * which may be nothing.  If there is something
1081 			 * in the store buffer, we can rotate the buffers.
1082 			 */
1083 			if (d->bd_hbuf)
1084 				/*
1085 				 * We filled up the buffer in between
1086 				 * getting the timeout and arriving
1087 				 * here, so we don't need to rotate.
1088 				 */
1089 				break;
1090 
1091 			if (d->bd_slen == 0) {
1092 				BPFD_UNLOCK(d);
1093 				return (0);
1094 			}
1095 			ROTATE_BUFFERS(d);
1096 			break;
1097 		}
1098 	}
1099 	/*
1100 	 * At this point, we know we have something in the hold slot.
1101 	 */
1102 	d->bd_hbuf_in_use = 1;
1103 	BPFD_UNLOCK(d);
1104 
1105 	/*
1106 	 * Move data from hold buffer into user space.
1107 	 * We know the entire buffer is transferred since
1108 	 * we checked above that the read buffer is bpf_bufsize bytes.
1109   	 *
1110 	 * We do not have to worry about simultaneous reads because
1111 	 * we waited for sole access to the hold buffer above.
1112 	 */
1113 	error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1114 
1115 	BPFD_LOCK(d);
1116 	KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1117 	d->bd_fbuf = d->bd_hbuf;
1118 	d->bd_hbuf = NULL;
1119 	d->bd_hlen = 0;
1120 	bpf_buf_reclaimed(d);
1121 	d->bd_hbuf_in_use = 0;
1122 	wakeup(&d->bd_hbuf_in_use);
1123 	BPFD_UNLOCK(d);
1124 
1125 	return (error);
1126 }
1127 
1128 /*
1129  * If there are processes sleeping on this descriptor, wake them up.
1130  */
1131 static __inline void
1132 bpf_wakeup(struct bpf_d *d)
1133 {
1134 
1135 	BPFD_LOCK_ASSERT(d);
1136 	if (d->bd_state == BPF_WAITING) {
1137 		callout_stop(&d->bd_callout);
1138 		d->bd_state = BPF_IDLE;
1139 	}
1140 	wakeup(d);
1141 	if (d->bd_async && d->bd_sig && d->bd_sigio)
1142 		pgsigio(&d->bd_sigio, d->bd_sig, 0);
1143 
1144 	selwakeuppri(&d->bd_sel, PRINET);
1145 	KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1146 }
1147 
1148 static void
1149 bpf_timed_out(void *arg)
1150 {
1151 	struct bpf_d *d = (struct bpf_d *)arg;
1152 
1153 	BPFD_LOCK_ASSERT(d);
1154 
1155 	if (callout_pending(&d->bd_callout) ||
1156 	    !callout_active(&d->bd_callout))
1157 		return;
1158 	if (d->bd_state == BPF_WAITING) {
1159 		d->bd_state = BPF_TIMED_OUT;
1160 		if (d->bd_slen != 0)
1161 			bpf_wakeup(d);
1162 	}
1163 }
1164 
1165 static int
1166 bpf_ready(struct bpf_d *d)
1167 {
1168 
1169 	BPFD_LOCK_ASSERT(d);
1170 
1171 	if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1172 		return (1);
1173 	if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1174 	    d->bd_slen != 0)
1175 		return (1);
1176 	return (0);
1177 }
1178 
1179 static int
1180 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1181 {
1182 	struct route ro;
1183 	struct sockaddr dst;
1184 	struct epoch_tracker et;
1185 	struct bpf_if *bp;
1186 	struct bpf_d *d;
1187 	struct ifnet *ifp;
1188 	struct mbuf *m, *mc;
1189 	int error, hlen;
1190 
1191 	error = devfs_get_cdevpriv((void **)&d);
1192 	if (error != 0)
1193 		return (error);
1194 
1195 	NET_EPOCH_ENTER(et);
1196 	BPFD_LOCK(d);
1197 	BPF_PID_REFRESH_CUR(d);
1198 	counter_u64_add(d->bd_wcount, 1);
1199 	if ((bp = d->bd_bif) == NULL) {
1200 		error = ENXIO;
1201 		goto out_locked;
1202 	}
1203 
1204 	ifp = bp->bif_ifp;
1205 	if ((ifp->if_flags & IFF_UP) == 0) {
1206 		error = ENETDOWN;
1207 		goto out_locked;
1208 	}
1209 
1210 	if (uio->uio_resid == 0)
1211 		goto out_locked;
1212 
1213 	bzero(&dst, sizeof(dst));
1214 	m = NULL;
1215 	hlen = 0;
1216 
1217 	/*
1218 	 * Take extra reference, unlock d and exit from epoch section,
1219 	 * since bpf_movein() can sleep.
1220 	 */
1221 	bpfd_ref(d);
1222 	NET_EPOCH_EXIT(et);
1223 	BPFD_UNLOCK(d);
1224 
1225 	error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1226 	    &m, &dst, &hlen, d);
1227 
1228 	if (error != 0) {
1229 		counter_u64_add(d->bd_wdcount, 1);
1230 		bpfd_rele(d);
1231 		return (error);
1232 	}
1233 
1234 	BPFD_LOCK(d);
1235 	/*
1236 	 * Check that descriptor is still attached to the interface.
1237 	 * This can happen on bpfdetach(). To avoid access to detached
1238 	 * ifnet, free mbuf and return ENXIO.
1239 	 */
1240 	if (d->bd_bif == NULL) {
1241 		counter_u64_add(d->bd_wdcount, 1);
1242 		BPFD_UNLOCK(d);
1243 		bpfd_rele(d);
1244 		m_freem(m);
1245 		return (ENXIO);
1246 	}
1247 	counter_u64_add(d->bd_wfcount, 1);
1248 	if (d->bd_hdrcmplt)
1249 		dst.sa_family = pseudo_AF_HDRCMPLT;
1250 
1251 	if (d->bd_feedback) {
1252 		mc = m_dup(m, M_NOWAIT);
1253 		if (mc != NULL)
1254 			mc->m_pkthdr.rcvif = ifp;
1255 		/* Set M_PROMISC for outgoing packets to be discarded. */
1256 		if (d->bd_direction == BPF_D_INOUT)
1257 			m->m_flags |= M_PROMISC;
1258 	} else
1259 		mc = NULL;
1260 
1261 	m->m_pkthdr.len -= hlen;
1262 	m->m_len -= hlen;
1263 	m->m_data += hlen;	/* XXX */
1264 
1265 	CURVNET_SET(ifp->if_vnet);
1266 #ifdef MAC
1267 	mac_bpfdesc_create_mbuf(d, m);
1268 	if (mc != NULL)
1269 		mac_bpfdesc_create_mbuf(d, mc);
1270 #endif
1271 
1272 	bzero(&ro, sizeof(ro));
1273 	if (hlen != 0) {
1274 		ro.ro_prepend = (u_char *)&dst.sa_data;
1275 		ro.ro_plen = hlen;
1276 		ro.ro_flags = RT_HAS_HEADER;
1277 	}
1278 
1279 	if (d->bd_pcp != 0)
1280 		vlan_set_pcp(m, d->bd_pcp);
1281 
1282 	/* Avoid possible recursion on BPFD_LOCK(). */
1283 	NET_EPOCH_ENTER(et);
1284 	BPFD_UNLOCK(d);
1285 	error = (*ifp->if_output)(ifp, m, &dst, &ro);
1286 	if (error)
1287 		counter_u64_add(d->bd_wdcount, 1);
1288 
1289 	if (mc != NULL) {
1290 		if (error == 0)
1291 			(*ifp->if_input)(ifp, mc);
1292 		else
1293 			m_freem(mc);
1294 	}
1295 	NET_EPOCH_EXIT(et);
1296 	CURVNET_RESTORE();
1297 	bpfd_rele(d);
1298 	return (error);
1299 
1300 out_locked:
1301 	counter_u64_add(d->bd_wdcount, 1);
1302 	NET_EPOCH_EXIT(et);
1303 	BPFD_UNLOCK(d);
1304 	return (error);
1305 }
1306 
1307 /*
1308  * Reset a descriptor by flushing its packet buffer and clearing the receive
1309  * and drop counts.  This is doable for kernel-only buffers, but with
1310  * zero-copy buffers, we can't write to (or rotate) buffers that are
1311  * currently owned by userspace.  It would be nice if we could encapsulate
1312  * this logic in the buffer code rather than here.
1313  */
1314 static void
1315 reset_d(struct bpf_d *d)
1316 {
1317 
1318 	BPFD_LOCK_ASSERT(d);
1319 
1320 	while (d->bd_hbuf_in_use)
1321 		mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1322 		    "bd_hbuf", 0);
1323 	if ((d->bd_hbuf != NULL) &&
1324 	    (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1325 		/* Free the hold buffer. */
1326 		d->bd_fbuf = d->bd_hbuf;
1327 		d->bd_hbuf = NULL;
1328 		d->bd_hlen = 0;
1329 		bpf_buf_reclaimed(d);
1330 	}
1331 	if (bpf_canwritebuf(d))
1332 		d->bd_slen = 0;
1333 	counter_u64_zero(d->bd_rcount);
1334 	counter_u64_zero(d->bd_dcount);
1335 	counter_u64_zero(d->bd_fcount);
1336 	counter_u64_zero(d->bd_wcount);
1337 	counter_u64_zero(d->bd_wfcount);
1338 	counter_u64_zero(d->bd_wdcount);
1339 	counter_u64_zero(d->bd_zcopy);
1340 }
1341 
1342 /*
1343  *  FIONREAD		Check for read packet available.
1344  *  BIOCGBLEN		Get buffer len [for read()].
1345  *  BIOCSETF		Set read filter.
1346  *  BIOCSETFNR		Set read filter without resetting descriptor.
1347  *  BIOCSETWF		Set write filter.
1348  *  BIOCFLUSH		Flush read packet buffer.
1349  *  BIOCPROMISC		Put interface into promiscuous mode.
1350  *  BIOCGDLT		Get link layer type.
1351  *  BIOCGETIF		Get interface name.
1352  *  BIOCSETIF		Set interface.
1353  *  BIOCSRTIMEOUT	Set read timeout.
1354  *  BIOCGRTIMEOUT	Get read timeout.
1355  *  BIOCGSTATS		Get packet stats.
1356  *  BIOCIMMEDIATE	Set immediate mode.
1357  *  BIOCVERSION		Get filter language version.
1358  *  BIOCGHDRCMPLT	Get "header already complete" flag
1359  *  BIOCSHDRCMPLT	Set "header already complete" flag
1360  *  BIOCGDIRECTION	Get packet direction flag
1361  *  BIOCSDIRECTION	Set packet direction flag
1362  *  BIOCGTSTAMP		Get time stamp format and resolution.
1363  *  BIOCSTSTAMP		Set time stamp format and resolution.
1364  *  BIOCLOCK		Set "locked" flag
1365  *  BIOCFEEDBACK	Set packet feedback mode.
1366  *  BIOCSETZBUF		Set current zero-copy buffer locations.
1367  *  BIOCGETZMAX		Get maximum zero-copy buffer size.
1368  *  BIOCROTZBUF		Force rotation of zero-copy buffer
1369  *  BIOCSETBUFMODE	Set buffer mode.
1370  *  BIOCGETBUFMODE	Get current buffer mode.
1371  *  BIOCSETVLANPCP	Set VLAN PCP tag.
1372  */
1373 /* ARGSUSED */
1374 static	int
1375 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1376     struct thread *td)
1377 {
1378 	struct bpf_d *d;
1379 	int error;
1380 
1381 	error = devfs_get_cdevpriv((void **)&d);
1382 	if (error != 0)
1383 		return (error);
1384 
1385 	/*
1386 	 * Refresh PID associated with this descriptor.
1387 	 */
1388 	BPFD_LOCK(d);
1389 	BPF_PID_REFRESH(d, td);
1390 	if (d->bd_state == BPF_WAITING)
1391 		callout_stop(&d->bd_callout);
1392 	d->bd_state = BPF_IDLE;
1393 	BPFD_UNLOCK(d);
1394 
1395 	if (d->bd_locked == 1) {
1396 		switch (cmd) {
1397 		case BIOCGBLEN:
1398 		case BIOCFLUSH:
1399 		case BIOCGDLT:
1400 		case BIOCGDLTLIST:
1401 #ifdef COMPAT_FREEBSD32
1402 		case BIOCGDLTLIST32:
1403 #endif
1404 		case BIOCGETIF:
1405 		case BIOCGRTIMEOUT:
1406 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1407 		case BIOCGRTIMEOUT32:
1408 #endif
1409 		case BIOCGSTATS:
1410 		case BIOCVERSION:
1411 		case BIOCGRSIG:
1412 		case BIOCGHDRCMPLT:
1413 		case BIOCSTSTAMP:
1414 		case BIOCFEEDBACK:
1415 		case FIONREAD:
1416 		case BIOCLOCK:
1417 		case BIOCSRTIMEOUT:
1418 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1419 		case BIOCSRTIMEOUT32:
1420 #endif
1421 		case BIOCIMMEDIATE:
1422 		case TIOCGPGRP:
1423 		case BIOCROTZBUF:
1424 			break;
1425 		default:
1426 			return (EPERM);
1427 		}
1428 	}
1429 #ifdef COMPAT_FREEBSD32
1430 	/*
1431 	 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1432 	 * that it will get 32-bit packet headers.
1433 	 */
1434 	switch (cmd) {
1435 	case BIOCSETF32:
1436 	case BIOCSETFNR32:
1437 	case BIOCSETWF32:
1438 	case BIOCGDLTLIST32:
1439 	case BIOCGRTIMEOUT32:
1440 	case BIOCSRTIMEOUT32:
1441 		if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1442 			BPFD_LOCK(d);
1443 			d->bd_compat32 = 1;
1444 			BPFD_UNLOCK(d);
1445 		}
1446 	}
1447 #endif
1448 
1449 	CURVNET_SET(TD_TO_VNET(td));
1450 	switch (cmd) {
1451 	default:
1452 		error = EINVAL;
1453 		break;
1454 
1455 	/*
1456 	 * Check for read packet available.
1457 	 */
1458 	case FIONREAD:
1459 		{
1460 			int n;
1461 
1462 			BPFD_LOCK(d);
1463 			n = d->bd_slen;
1464 			while (d->bd_hbuf_in_use)
1465 				mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1466 				    PRINET, "bd_hbuf", 0);
1467 			if (d->bd_hbuf)
1468 				n += d->bd_hlen;
1469 			BPFD_UNLOCK(d);
1470 
1471 			*(int *)addr = n;
1472 			break;
1473 		}
1474 
1475 	/*
1476 	 * Get buffer len [for read()].
1477 	 */
1478 	case BIOCGBLEN:
1479 		BPFD_LOCK(d);
1480 		*(u_int *)addr = d->bd_bufsize;
1481 		BPFD_UNLOCK(d);
1482 		break;
1483 
1484 	/*
1485 	 * Set buffer length.
1486 	 */
1487 	case BIOCSBLEN:
1488 		error = bpf_ioctl_sblen(d, (u_int *)addr);
1489 		break;
1490 
1491 	/*
1492 	 * Set link layer read filter.
1493 	 */
1494 	case BIOCSETF:
1495 	case BIOCSETFNR:
1496 	case BIOCSETWF:
1497 #ifdef COMPAT_FREEBSD32
1498 	case BIOCSETF32:
1499 	case BIOCSETFNR32:
1500 	case BIOCSETWF32:
1501 #endif
1502 		error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1503 		break;
1504 
1505 	/*
1506 	 * Flush read packet buffer.
1507 	 */
1508 	case BIOCFLUSH:
1509 		BPFD_LOCK(d);
1510 		reset_d(d);
1511 		BPFD_UNLOCK(d);
1512 		break;
1513 
1514 	/*
1515 	 * Put interface into promiscuous mode.
1516 	 */
1517 	case BIOCPROMISC:
1518 		BPF_LOCK();
1519 		if (d->bd_bif == NULL) {
1520 			/*
1521 			 * No interface attached yet.
1522 			 */
1523 			error = EINVAL;
1524 		} else if (d->bd_promisc == 0) {
1525 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1526 			if (error == 0)
1527 				d->bd_promisc = 1;
1528 		}
1529 		BPF_UNLOCK();
1530 		break;
1531 
1532 	/*
1533 	 * Get current data link type.
1534 	 */
1535 	case BIOCGDLT:
1536 		BPF_LOCK();
1537 		if (d->bd_bif == NULL)
1538 			error = EINVAL;
1539 		else
1540 			*(u_int *)addr = d->bd_bif->bif_dlt;
1541 		BPF_UNLOCK();
1542 		break;
1543 
1544 	/*
1545 	 * Get a list of supported data link types.
1546 	 */
1547 #ifdef COMPAT_FREEBSD32
1548 	case BIOCGDLTLIST32:
1549 		{
1550 			struct bpf_dltlist32 *list32;
1551 			struct bpf_dltlist dltlist;
1552 
1553 			list32 = (struct bpf_dltlist32 *)addr;
1554 			dltlist.bfl_len = list32->bfl_len;
1555 			dltlist.bfl_list = PTRIN(list32->bfl_list);
1556 			BPF_LOCK();
1557 			if (d->bd_bif == NULL)
1558 				error = EINVAL;
1559 			else {
1560 				error = bpf_getdltlist(d, &dltlist);
1561 				if (error == 0)
1562 					list32->bfl_len = dltlist.bfl_len;
1563 			}
1564 			BPF_UNLOCK();
1565 			break;
1566 		}
1567 #endif
1568 
1569 	case BIOCGDLTLIST:
1570 		BPF_LOCK();
1571 		if (d->bd_bif == NULL)
1572 			error = EINVAL;
1573 		else
1574 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1575 		BPF_UNLOCK();
1576 		break;
1577 
1578 	/*
1579 	 * Set data link type.
1580 	 */
1581 	case BIOCSDLT:
1582 		BPF_LOCK();
1583 		if (d->bd_bif == NULL)
1584 			error = EINVAL;
1585 		else
1586 			error = bpf_setdlt(d, *(u_int *)addr);
1587 		BPF_UNLOCK();
1588 		break;
1589 
1590 	/*
1591 	 * Get interface name.
1592 	 */
1593 	case BIOCGETIF:
1594 		BPF_LOCK();
1595 		if (d->bd_bif == NULL)
1596 			error = EINVAL;
1597 		else {
1598 			struct ifnet *const ifp = d->bd_bif->bif_ifp;
1599 			struct ifreq *const ifr = (struct ifreq *)addr;
1600 
1601 			strlcpy(ifr->ifr_name, ifp->if_xname,
1602 			    sizeof(ifr->ifr_name));
1603 		}
1604 		BPF_UNLOCK();
1605 		break;
1606 
1607 	/*
1608 	 * Set interface.
1609 	 */
1610 	case BIOCSETIF:
1611 		{
1612 			int alloc_buf, size;
1613 
1614 			/*
1615 			 * Behavior here depends on the buffering model.  If
1616 			 * we're using kernel memory buffers, then we can
1617 			 * allocate them here.  If we're using zero-copy,
1618 			 * then the user process must have registered buffers
1619 			 * by the time we get here.
1620 			 */
1621 			alloc_buf = 0;
1622 			BPFD_LOCK(d);
1623 			if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1624 			    d->bd_sbuf == NULL)
1625 				alloc_buf = 1;
1626 			BPFD_UNLOCK(d);
1627 			if (alloc_buf) {
1628 				size = d->bd_bufsize;
1629 				error = bpf_buffer_ioctl_sblen(d, &size);
1630 				if (error != 0)
1631 					break;
1632 			}
1633 			BPF_LOCK();
1634 			error = bpf_setif(d, (struct ifreq *)addr);
1635 			BPF_UNLOCK();
1636 			break;
1637 		}
1638 
1639 	/*
1640 	 * Set read timeout.
1641 	 */
1642 	case BIOCSRTIMEOUT:
1643 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1644 	case BIOCSRTIMEOUT32:
1645 #endif
1646 		{
1647 			struct timeval *tv = (struct timeval *)addr;
1648 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1649 			struct timeval32 *tv32;
1650 			struct timeval tv64;
1651 
1652 			if (cmd == BIOCSRTIMEOUT32) {
1653 				tv32 = (struct timeval32 *)addr;
1654 				tv = &tv64;
1655 				tv->tv_sec = tv32->tv_sec;
1656 				tv->tv_usec = tv32->tv_usec;
1657 			} else
1658 #endif
1659 				tv = (struct timeval *)addr;
1660 
1661 			/*
1662 			 * Subtract 1 tick from tvtohz() since this isn't
1663 			 * a one-shot timer.
1664 			 */
1665 			if ((error = itimerfix(tv)) == 0)
1666 				d->bd_rtout = tvtohz(tv) - 1;
1667 			break;
1668 		}
1669 
1670 	/*
1671 	 * Get read timeout.
1672 	 */
1673 	case BIOCGRTIMEOUT:
1674 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1675 	case BIOCGRTIMEOUT32:
1676 #endif
1677 		{
1678 			struct timeval *tv;
1679 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1680 			struct timeval32 *tv32;
1681 			struct timeval tv64;
1682 
1683 			if (cmd == BIOCGRTIMEOUT32)
1684 				tv = &tv64;
1685 			else
1686 #endif
1687 				tv = (struct timeval *)addr;
1688 
1689 			tv->tv_sec = d->bd_rtout / hz;
1690 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1691 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1692 			if (cmd == BIOCGRTIMEOUT32) {
1693 				tv32 = (struct timeval32 *)addr;
1694 				tv32->tv_sec = tv->tv_sec;
1695 				tv32->tv_usec = tv->tv_usec;
1696 			}
1697 #endif
1698 
1699 			break;
1700 		}
1701 
1702 	/*
1703 	 * Get packet stats.
1704 	 */
1705 	case BIOCGSTATS:
1706 		{
1707 			struct bpf_stat *bs = (struct bpf_stat *)addr;
1708 
1709 			/* XXXCSJP overflow */
1710 			bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1711 			bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1712 			break;
1713 		}
1714 
1715 	/*
1716 	 * Set immediate mode.
1717 	 */
1718 	case BIOCIMMEDIATE:
1719 		BPFD_LOCK(d);
1720 		d->bd_immediate = *(u_int *)addr;
1721 		BPFD_UNLOCK(d);
1722 		break;
1723 
1724 	case BIOCVERSION:
1725 		{
1726 			struct bpf_version *bv = (struct bpf_version *)addr;
1727 
1728 			bv->bv_major = BPF_MAJOR_VERSION;
1729 			bv->bv_minor = BPF_MINOR_VERSION;
1730 			break;
1731 		}
1732 
1733 	/*
1734 	 * Get "header already complete" flag
1735 	 */
1736 	case BIOCGHDRCMPLT:
1737 		BPFD_LOCK(d);
1738 		*(u_int *)addr = d->bd_hdrcmplt;
1739 		BPFD_UNLOCK(d);
1740 		break;
1741 
1742 	/*
1743 	 * Set "header already complete" flag
1744 	 */
1745 	case BIOCSHDRCMPLT:
1746 		BPFD_LOCK(d);
1747 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1748 		BPFD_UNLOCK(d);
1749 		break;
1750 
1751 	/*
1752 	 * Get packet direction flag
1753 	 */
1754 	case BIOCGDIRECTION:
1755 		BPFD_LOCK(d);
1756 		*(u_int *)addr = d->bd_direction;
1757 		BPFD_UNLOCK(d);
1758 		break;
1759 
1760 	/*
1761 	 * Set packet direction flag
1762 	 */
1763 	case BIOCSDIRECTION:
1764 		{
1765 			u_int	direction;
1766 
1767 			direction = *(u_int *)addr;
1768 			switch (direction) {
1769 			case BPF_D_IN:
1770 			case BPF_D_INOUT:
1771 			case BPF_D_OUT:
1772 				BPFD_LOCK(d);
1773 				d->bd_direction = direction;
1774 				BPFD_UNLOCK(d);
1775 				break;
1776 			default:
1777 				error = EINVAL;
1778 			}
1779 		}
1780 		break;
1781 
1782 	/*
1783 	 * Get packet timestamp format and resolution.
1784 	 */
1785 	case BIOCGTSTAMP:
1786 		BPFD_LOCK(d);
1787 		*(u_int *)addr = d->bd_tstamp;
1788 		BPFD_UNLOCK(d);
1789 		break;
1790 
1791 	/*
1792 	 * Set packet timestamp format and resolution.
1793 	 */
1794 	case BIOCSTSTAMP:
1795 		{
1796 			u_int	func;
1797 
1798 			func = *(u_int *)addr;
1799 			if (BPF_T_VALID(func))
1800 				d->bd_tstamp = func;
1801 			else
1802 				error = EINVAL;
1803 		}
1804 		break;
1805 
1806 	case BIOCFEEDBACK:
1807 		BPFD_LOCK(d);
1808 		d->bd_feedback = *(u_int *)addr;
1809 		BPFD_UNLOCK(d);
1810 		break;
1811 
1812 	case BIOCLOCK:
1813 		BPFD_LOCK(d);
1814 		d->bd_locked = 1;
1815 		BPFD_UNLOCK(d);
1816 		break;
1817 
1818 	case FIONBIO:		/* Non-blocking I/O */
1819 		break;
1820 
1821 	case FIOASYNC:		/* Send signal on receive packets */
1822 		BPFD_LOCK(d);
1823 		d->bd_async = *(int *)addr;
1824 		BPFD_UNLOCK(d);
1825 		break;
1826 
1827 	case FIOSETOWN:
1828 		/*
1829 		 * XXX: Add some sort of locking here?
1830 		 * fsetown() can sleep.
1831 		 */
1832 		error = fsetown(*(int *)addr, &d->bd_sigio);
1833 		break;
1834 
1835 	case FIOGETOWN:
1836 		BPFD_LOCK(d);
1837 		*(int *)addr = fgetown(&d->bd_sigio);
1838 		BPFD_UNLOCK(d);
1839 		break;
1840 
1841 	/* This is deprecated, FIOSETOWN should be used instead. */
1842 	case TIOCSPGRP:
1843 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
1844 		break;
1845 
1846 	/* This is deprecated, FIOGETOWN should be used instead. */
1847 	case TIOCGPGRP:
1848 		*(int *)addr = -fgetown(&d->bd_sigio);
1849 		break;
1850 
1851 	case BIOCSRSIG:		/* Set receive signal */
1852 		{
1853 			u_int sig;
1854 
1855 			sig = *(u_int *)addr;
1856 
1857 			if (sig >= NSIG)
1858 				error = EINVAL;
1859 			else {
1860 				BPFD_LOCK(d);
1861 				d->bd_sig = sig;
1862 				BPFD_UNLOCK(d);
1863 			}
1864 			break;
1865 		}
1866 	case BIOCGRSIG:
1867 		BPFD_LOCK(d);
1868 		*(u_int *)addr = d->bd_sig;
1869 		BPFD_UNLOCK(d);
1870 		break;
1871 
1872 	case BIOCGETBUFMODE:
1873 		BPFD_LOCK(d);
1874 		*(u_int *)addr = d->bd_bufmode;
1875 		BPFD_UNLOCK(d);
1876 		break;
1877 
1878 	case BIOCSETBUFMODE:
1879 		/*
1880 		 * Allow the buffering mode to be changed as long as we
1881 		 * haven't yet committed to a particular mode.  Our
1882 		 * definition of commitment, for now, is whether or not a
1883 		 * buffer has been allocated or an interface attached, since
1884 		 * that's the point where things get tricky.
1885 		 */
1886 		switch (*(u_int *)addr) {
1887 		case BPF_BUFMODE_BUFFER:
1888 			break;
1889 
1890 		case BPF_BUFMODE_ZBUF:
1891 			if (bpf_zerocopy_enable)
1892 				break;
1893 			/* FALLSTHROUGH */
1894 
1895 		default:
1896 			CURVNET_RESTORE();
1897 			return (EINVAL);
1898 		}
1899 
1900 		BPFD_LOCK(d);
1901 		if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1902 		    d->bd_fbuf != NULL || d->bd_bif != NULL) {
1903 			BPFD_UNLOCK(d);
1904 			CURVNET_RESTORE();
1905 			return (EBUSY);
1906 		}
1907 		d->bd_bufmode = *(u_int *)addr;
1908 		BPFD_UNLOCK(d);
1909 		break;
1910 
1911 	case BIOCGETZMAX:
1912 		error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1913 		break;
1914 
1915 	case BIOCSETZBUF:
1916 		error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1917 		break;
1918 
1919 	case BIOCROTZBUF:
1920 		error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1921 		break;
1922 
1923 	case BIOCSETVLANPCP:
1924 		{
1925 			u_int pcp;
1926 
1927 			pcp = *(u_int *)addr;
1928 			if (pcp > BPF_PRIO_MAX || pcp < 0) {
1929 				error = EINVAL;
1930 				break;
1931 			}
1932 			d->bd_pcp = pcp;
1933 			break;
1934 		}
1935 	}
1936 	CURVNET_RESTORE();
1937 	return (error);
1938 }
1939 
1940 /*
1941  * Set d's packet filter program to fp. If this file already has a filter,
1942  * free it and replace it. Returns EINVAL for bogus requests.
1943  *
1944  * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1945  * calls.
1946  */
1947 static int
1948 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1949 {
1950 #ifdef COMPAT_FREEBSD32
1951 	struct bpf_program fp_swab;
1952 	struct bpf_program32 *fp32;
1953 #endif
1954 	struct bpf_program_buffer *fcode;
1955 	struct bpf_insn *filter;
1956 #ifdef BPF_JITTER
1957 	bpf_jit_filter *jfunc;
1958 #endif
1959 	size_t size;
1960 	u_int flen;
1961 	bool track_event;
1962 
1963 #ifdef COMPAT_FREEBSD32
1964 	switch (cmd) {
1965 	case BIOCSETF32:
1966 	case BIOCSETWF32:
1967 	case BIOCSETFNR32:
1968 		fp32 = (struct bpf_program32 *)fp;
1969 		fp_swab.bf_len = fp32->bf_len;
1970 		fp_swab.bf_insns =
1971 		    (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1972 		fp = &fp_swab;
1973 		switch (cmd) {
1974 		case BIOCSETF32:
1975 			cmd = BIOCSETF;
1976 			break;
1977 		case BIOCSETWF32:
1978 			cmd = BIOCSETWF;
1979 			break;
1980 		}
1981 		break;
1982 	}
1983 #endif
1984 
1985 	filter = NULL;
1986 #ifdef BPF_JITTER
1987 	jfunc = NULL;
1988 #endif
1989 	/*
1990 	 * Check new filter validness before acquiring any locks.
1991 	 * Allocate memory for new filter, if needed.
1992 	 */
1993 	flen = fp->bf_len;
1994 	if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1995 		return (EINVAL);
1996 	size = flen * sizeof(*fp->bf_insns);
1997 	if (size > 0) {
1998 		/* We're setting up new filter. Copy and check actual data. */
1999 		fcode = bpf_program_buffer_alloc(size, M_WAITOK);
2000 		filter = (struct bpf_insn *)fcode->buffer;
2001 		if (copyin(fp->bf_insns, filter, size) != 0 ||
2002 		    !bpf_validate(filter, flen)) {
2003 			free(fcode, M_BPF);
2004 			return (EINVAL);
2005 		}
2006 #ifdef BPF_JITTER
2007 		if (cmd != BIOCSETWF) {
2008 			/*
2009 			 * Filter is copied inside fcode and is
2010 			 * perfectly valid.
2011 			 */
2012 			jfunc = bpf_jitter(filter, flen);
2013 		}
2014 #endif
2015 	}
2016 
2017 	track_event = false;
2018 	fcode = NULL;
2019 
2020 	BPF_LOCK();
2021 	BPFD_LOCK(d);
2022 	/* Set up new filter. */
2023 	if (cmd == BIOCSETWF) {
2024 		if (d->bd_wfilter != NULL) {
2025 			fcode = __containerof((void *)d->bd_wfilter,
2026 			    struct bpf_program_buffer, buffer);
2027 #ifdef BPF_JITTER
2028 			fcode->func = NULL;
2029 #endif
2030 		}
2031 		d->bd_wfilter = filter;
2032 	} else {
2033 		if (d->bd_rfilter != NULL) {
2034 			fcode = __containerof((void *)d->bd_rfilter,
2035 			    struct bpf_program_buffer, buffer);
2036 #ifdef BPF_JITTER
2037 			fcode->func = d->bd_bfilter;
2038 #endif
2039 		}
2040 		d->bd_rfilter = filter;
2041 #ifdef BPF_JITTER
2042 		d->bd_bfilter = jfunc;
2043 #endif
2044 		if (cmd == BIOCSETF)
2045 			reset_d(d);
2046 
2047 		if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2048 			/*
2049 			 * Filter can be set several times without
2050 			 * specifying interface. In this case just mark d
2051 			 * as reader.
2052 			 */
2053 			d->bd_writer = 0;
2054 			if (d->bd_bif != NULL) {
2055 				/*
2056 				 * Remove descriptor from writers-only list
2057 				 * and add it to active readers list.
2058 				 */
2059 				CK_LIST_REMOVE(d, bd_next);
2060 				CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2061 				    d, bd_next);
2062 				CTR2(KTR_NET,
2063 				    "%s: upgrade required by pid %d",
2064 				    __func__, d->bd_pid);
2065 				track_event = true;
2066 			}
2067 		}
2068 	}
2069 	BPFD_UNLOCK(d);
2070 
2071 	if (fcode != NULL)
2072 		NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2073 
2074 	if (track_event)
2075 		EVENTHANDLER_INVOKE(bpf_track,
2076 		    d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2077 
2078 	BPF_UNLOCK();
2079 	return (0);
2080 }
2081 
2082 /*
2083  * Detach a file from its current interface (if attached at all) and attach
2084  * to the interface indicated by the name stored in ifr.
2085  * Return an errno or 0.
2086  */
2087 static int
2088 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2089 {
2090 	struct bpf_if *bp;
2091 	struct ifnet *theywant;
2092 
2093 	BPF_LOCK_ASSERT();
2094 
2095 	theywant = ifunit(ifr->ifr_name);
2096 	if (theywant == NULL || theywant->if_bpf == NULL)
2097 		return (ENXIO);
2098 
2099 	bp = theywant->if_bpf;
2100 	/*
2101 	 * At this point, we expect the buffer is already allocated.  If not,
2102 	 * return an error.
2103 	 */
2104 	switch (d->bd_bufmode) {
2105 	case BPF_BUFMODE_BUFFER:
2106 	case BPF_BUFMODE_ZBUF:
2107 		if (d->bd_sbuf == NULL)
2108 			return (EINVAL);
2109 		break;
2110 
2111 	default:
2112 		panic("bpf_setif: bufmode %d", d->bd_bufmode);
2113 	}
2114 	if (bp != d->bd_bif)
2115 		bpf_attachd(d, bp);
2116 	else {
2117 		BPFD_LOCK(d);
2118 		reset_d(d);
2119 		BPFD_UNLOCK(d);
2120 	}
2121 	return (0);
2122 }
2123 
2124 /*
2125  * Support for select() and poll() system calls
2126  *
2127  * Return true iff the specific operation will not block indefinitely.
2128  * Otherwise, return false but make a note that a selwakeup() must be done.
2129  */
2130 static int
2131 bpfpoll(struct cdev *dev, int events, struct thread *td)
2132 {
2133 	struct bpf_d *d;
2134 	int revents;
2135 
2136 	if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2137 		return (events &
2138 		    (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
2139 
2140 	/*
2141 	 * Refresh PID associated with this descriptor.
2142 	 */
2143 	revents = events & (POLLOUT | POLLWRNORM);
2144 	BPFD_LOCK(d);
2145 	BPF_PID_REFRESH(d, td);
2146 	if (events & (POLLIN | POLLRDNORM)) {
2147 		if (bpf_ready(d))
2148 			revents |= events & (POLLIN | POLLRDNORM);
2149 		else {
2150 			selrecord(td, &d->bd_sel);
2151 			/* Start the read timeout if necessary. */
2152 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2153 				callout_reset(&d->bd_callout, d->bd_rtout,
2154 				    bpf_timed_out, d);
2155 				d->bd_state = BPF_WAITING;
2156 			}
2157 		}
2158 	}
2159 	BPFD_UNLOCK(d);
2160 	return (revents);
2161 }
2162 
2163 /*
2164  * Support for kevent() system call.  Register EVFILT_READ filters and
2165  * reject all others.
2166  */
2167 int
2168 bpfkqfilter(struct cdev *dev, struct knote *kn)
2169 {
2170 	struct bpf_d *d;
2171 
2172 	if (devfs_get_cdevpriv((void **)&d) != 0)
2173 		return (1);
2174 
2175 	switch (kn->kn_filter) {
2176 	case EVFILT_READ:
2177 		kn->kn_fop = &bpfread_filtops;
2178 		break;
2179 
2180 	case EVFILT_WRITE:
2181 		kn->kn_fop = &bpfwrite_filtops;
2182 		break;
2183 
2184 	default:
2185 		return (1);
2186 	}
2187 
2188 	/*
2189 	 * Refresh PID associated with this descriptor.
2190 	 */
2191 	BPFD_LOCK(d);
2192 	BPF_PID_REFRESH_CUR(d);
2193 	kn->kn_hook = d;
2194 	knlist_add(&d->bd_sel.si_note, kn, 1);
2195 	BPFD_UNLOCK(d);
2196 
2197 	return (0);
2198 }
2199 
2200 static void
2201 filt_bpfdetach(struct knote *kn)
2202 {
2203 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2204 
2205 	knlist_remove(&d->bd_sel.si_note, kn, 0);
2206 }
2207 
2208 static int
2209 filt_bpfread(struct knote *kn, long hint)
2210 {
2211 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2212 	int ready;
2213 
2214 	BPFD_LOCK_ASSERT(d);
2215 	ready = bpf_ready(d);
2216 	if (ready) {
2217 		kn->kn_data = d->bd_slen;
2218 		/*
2219 		 * Ignore the hold buffer if it is being copied to user space.
2220 		 */
2221 		if (!d->bd_hbuf_in_use && d->bd_hbuf)
2222 			kn->kn_data += d->bd_hlen;
2223 	} else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2224 		callout_reset(&d->bd_callout, d->bd_rtout,
2225 		    bpf_timed_out, d);
2226 		d->bd_state = BPF_WAITING;
2227 	}
2228 
2229 	return (ready);
2230 }
2231 
2232 static int
2233 filt_bpfwrite(struct knote *kn, long hint)
2234 {
2235 	struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2236 
2237 	BPFD_LOCK_ASSERT(d);
2238 
2239 	if (d->bd_bif == NULL) {
2240 		kn->kn_data = 0;
2241 		return (0);
2242 	} else {
2243 		kn->kn_data = d->bd_bif->bif_ifp->if_mtu;
2244 		return (1);
2245 	}
2246 }
2247 
2248 #define	BPF_TSTAMP_NONE		0
2249 #define	BPF_TSTAMP_FAST		1
2250 #define	BPF_TSTAMP_NORMAL	2
2251 #define	BPF_TSTAMP_EXTERN	3
2252 
2253 static int
2254 bpf_ts_quality(int tstype)
2255 {
2256 
2257 	if (tstype == BPF_T_NONE)
2258 		return (BPF_TSTAMP_NONE);
2259 	if ((tstype & BPF_T_FAST) != 0)
2260 		return (BPF_TSTAMP_FAST);
2261 
2262 	return (BPF_TSTAMP_NORMAL);
2263 }
2264 
2265 static int
2266 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2267 {
2268 	struct timespec ts;
2269 	struct m_tag *tag;
2270 	int quality;
2271 
2272 	quality = bpf_ts_quality(tstype);
2273 	if (quality == BPF_TSTAMP_NONE)
2274 		return (quality);
2275 
2276 	if (m != NULL) {
2277 		if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) {
2278 			mbuf_tstmp2timespec(m, &ts);
2279 			timespec2bintime(&ts, bt);
2280 			return (BPF_TSTAMP_EXTERN);
2281 		}
2282 		tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2283 		if (tag != NULL) {
2284 			*bt = *(struct bintime *)(tag + 1);
2285 			return (BPF_TSTAMP_EXTERN);
2286 		}
2287 	}
2288 	if (quality == BPF_TSTAMP_NORMAL)
2289 		binuptime(bt);
2290 	else
2291 		getbinuptime(bt);
2292 
2293 	return (quality);
2294 }
2295 
2296 /*
2297  * Incoming linkage from device drivers.  Process the packet pkt, of length
2298  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
2299  * by each process' filter, and if accepted, stashed into the corresponding
2300  * buffer.
2301  */
2302 void
2303 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2304 {
2305 	struct epoch_tracker et;
2306 	struct bintime bt;
2307 	struct bpf_d *d;
2308 #ifdef BPF_JITTER
2309 	bpf_jit_filter *bf;
2310 #endif
2311 	u_int slen;
2312 	int gottime;
2313 
2314 	gottime = BPF_TSTAMP_NONE;
2315 	NET_EPOCH_ENTER(et);
2316 	CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2317 		counter_u64_add(d->bd_rcount, 1);
2318 		/*
2319 		 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2320 		 * is no way for the caller to indiciate to us whether this
2321 		 * packet is inbound or outbound. In the bpf_mtap() routines,
2322 		 * we use the interface pointers on the mbuf to figure it out.
2323 		 */
2324 #ifdef BPF_JITTER
2325 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2326 		if (bf != NULL)
2327 			slen = (*(bf->func))(pkt, pktlen, pktlen);
2328 		else
2329 #endif
2330 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2331 		if (slen != 0) {
2332 			/*
2333 			 * Filter matches. Let's to acquire write lock.
2334 			 */
2335 			BPFD_LOCK(d);
2336 			counter_u64_add(d->bd_fcount, 1);
2337 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2338 				gottime = bpf_gettime(&bt, d->bd_tstamp,
2339 				    NULL);
2340 #ifdef MAC
2341 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2342 #endif
2343 				catchpacket(d, pkt, pktlen, slen,
2344 				    bpf_append_bytes, &bt);
2345 			BPFD_UNLOCK(d);
2346 		}
2347 	}
2348 	NET_EPOCH_EXIT(et);
2349 }
2350 
2351 #define	BPF_CHECK_DIRECTION(d, r, i)				\
2352 	    (((d)->bd_direction == BPF_D_IN && (r) != (i)) ||	\
2353 	    ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2354 
2355 /*
2356  * Incoming linkage from device drivers, when packet is in an mbuf chain.
2357  * Locking model is explained in bpf_tap().
2358  */
2359 void
2360 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2361 {
2362 	struct epoch_tracker et;
2363 	struct bintime bt;
2364 	struct bpf_d *d;
2365 #ifdef BPF_JITTER
2366 	bpf_jit_filter *bf;
2367 #endif
2368 	u_int pktlen, slen;
2369 	int gottime;
2370 
2371 	/* Skip outgoing duplicate packets. */
2372 	if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2373 		m->m_flags &= ~M_PROMISC;
2374 		return;
2375 	}
2376 
2377 	pktlen = m_length(m, NULL);
2378 	gottime = BPF_TSTAMP_NONE;
2379 
2380 	NET_EPOCH_ENTER(et);
2381 	CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2382 		if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2383 			continue;
2384 		counter_u64_add(d->bd_rcount, 1);
2385 #ifdef BPF_JITTER
2386 		bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2387 		/* XXX We cannot handle multiple mbufs. */
2388 		if (bf != NULL && m->m_next == NULL)
2389 			slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2390 			    pktlen);
2391 		else
2392 #endif
2393 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2394 		if (slen != 0) {
2395 			BPFD_LOCK(d);
2396 
2397 			counter_u64_add(d->bd_fcount, 1);
2398 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2399 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2400 #ifdef MAC
2401 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2402 #endif
2403 				catchpacket(d, (u_char *)m, pktlen, slen,
2404 				    bpf_append_mbuf, &bt);
2405 			BPFD_UNLOCK(d);
2406 		}
2407 	}
2408 	NET_EPOCH_EXIT(et);
2409 }
2410 
2411 /*
2412  * Incoming linkage from device drivers, when packet is in
2413  * an mbuf chain and to be prepended by a contiguous header.
2414  */
2415 void
2416 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2417 {
2418 	struct epoch_tracker et;
2419 	struct bintime bt;
2420 	struct mbuf mb;
2421 	struct bpf_d *d;
2422 	u_int pktlen, slen;
2423 	int gottime;
2424 
2425 	/* Skip outgoing duplicate packets. */
2426 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2427 		m->m_flags &= ~M_PROMISC;
2428 		return;
2429 	}
2430 
2431 	pktlen = m_length(m, NULL);
2432 	/*
2433 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
2434 	 * Note that we cut corners here; we only setup what's
2435 	 * absolutely needed--this mbuf should never go anywhere else.
2436 	 */
2437 	mb.m_flags = 0;
2438 	mb.m_next = m;
2439 	mb.m_data = data;
2440 	mb.m_len = dlen;
2441 	pktlen += dlen;
2442 
2443 	gottime = BPF_TSTAMP_NONE;
2444 
2445 	NET_EPOCH_ENTER(et);
2446 	CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2447 		if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2448 			continue;
2449 		counter_u64_add(d->bd_rcount, 1);
2450 		slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2451 		if (slen != 0) {
2452 			BPFD_LOCK(d);
2453 
2454 			counter_u64_add(d->bd_fcount, 1);
2455 			if (gottime < bpf_ts_quality(d->bd_tstamp))
2456 				gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2457 #ifdef MAC
2458 			if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2459 #endif
2460 				catchpacket(d, (u_char *)&mb, pktlen, slen,
2461 				    bpf_append_mbuf, &bt);
2462 			BPFD_UNLOCK(d);
2463 		}
2464 	}
2465 	NET_EPOCH_EXIT(et);
2466 }
2467 
2468 #undef	BPF_CHECK_DIRECTION
2469 #undef	BPF_TSTAMP_NONE
2470 #undef	BPF_TSTAMP_FAST
2471 #undef	BPF_TSTAMP_NORMAL
2472 #undef	BPF_TSTAMP_EXTERN
2473 
2474 static int
2475 bpf_hdrlen(struct bpf_d *d)
2476 {
2477 	int hdrlen;
2478 
2479 	hdrlen = d->bd_bif->bif_hdrlen;
2480 #ifndef BURN_BRIDGES
2481 	if (d->bd_tstamp == BPF_T_NONE ||
2482 	    BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2483 #ifdef COMPAT_FREEBSD32
2484 		if (d->bd_compat32)
2485 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2486 		else
2487 #endif
2488 			hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2489 	else
2490 #endif
2491 		hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2492 #ifdef COMPAT_FREEBSD32
2493 	if (d->bd_compat32)
2494 		hdrlen = BPF_WORDALIGN32(hdrlen);
2495 	else
2496 #endif
2497 		hdrlen = BPF_WORDALIGN(hdrlen);
2498 
2499 	return (hdrlen - d->bd_bif->bif_hdrlen);
2500 }
2501 
2502 static void
2503 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2504 {
2505 	struct bintime bt2, boottimebin;
2506 	struct timeval tsm;
2507 	struct timespec tsn;
2508 
2509 	if ((tstype & BPF_T_MONOTONIC) == 0) {
2510 		bt2 = *bt;
2511 		getboottimebin(&boottimebin);
2512 		bintime_add(&bt2, &boottimebin);
2513 		bt = &bt2;
2514 	}
2515 	switch (BPF_T_FORMAT(tstype)) {
2516 	case BPF_T_MICROTIME:
2517 		bintime2timeval(bt, &tsm);
2518 		ts->bt_sec = tsm.tv_sec;
2519 		ts->bt_frac = tsm.tv_usec;
2520 		break;
2521 	case BPF_T_NANOTIME:
2522 		bintime2timespec(bt, &tsn);
2523 		ts->bt_sec = tsn.tv_sec;
2524 		ts->bt_frac = tsn.tv_nsec;
2525 		break;
2526 	case BPF_T_BINTIME:
2527 		ts->bt_sec = bt->sec;
2528 		ts->bt_frac = bt->frac;
2529 		break;
2530 	}
2531 }
2532 
2533 /*
2534  * Move the packet data from interface memory (pkt) into the
2535  * store buffer.  "cpfn" is the routine called to do the actual data
2536  * transfer.  bcopy is passed in to copy contiguous chunks, while
2537  * bpf_append_mbuf is passed in to copy mbuf chains.  In the latter case,
2538  * pkt is really an mbuf.
2539  */
2540 static void
2541 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2542     void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2543     struct bintime *bt)
2544 {
2545 	static char zeroes[BPF_ALIGNMENT];
2546 	struct bpf_xhdr hdr;
2547 #ifndef BURN_BRIDGES
2548 	struct bpf_hdr hdr_old;
2549 #ifdef COMPAT_FREEBSD32
2550 	struct bpf_hdr32 hdr32_old;
2551 #endif
2552 #endif
2553 	int caplen, curlen, hdrlen, pad, totlen;
2554 	int do_wakeup = 0;
2555 	int do_timestamp;
2556 	int tstype;
2557 
2558 	BPFD_LOCK_ASSERT(d);
2559 	if (d->bd_bif == NULL) {
2560 		/* Descriptor was detached in concurrent thread */
2561 		counter_u64_add(d->bd_dcount, 1);
2562 		return;
2563 	}
2564 
2565 	/*
2566 	 * Detect whether user space has released a buffer back to us, and if
2567 	 * so, move it from being a hold buffer to a free buffer.  This may
2568 	 * not be the best place to do it (for example, we might only want to
2569 	 * run this check if we need the space), but for now it's a reliable
2570 	 * spot to do it.
2571 	 */
2572 	if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2573 		d->bd_fbuf = d->bd_hbuf;
2574 		d->bd_hbuf = NULL;
2575 		d->bd_hlen = 0;
2576 		bpf_buf_reclaimed(d);
2577 	}
2578 
2579 	/*
2580 	 * Figure out how many bytes to move.  If the packet is
2581 	 * greater or equal to the snapshot length, transfer that
2582 	 * much.  Otherwise, transfer the whole packet (unless
2583 	 * we hit the buffer size limit).
2584 	 */
2585 	hdrlen = bpf_hdrlen(d);
2586 	totlen = hdrlen + min(snaplen, pktlen);
2587 	if (totlen > d->bd_bufsize)
2588 		totlen = d->bd_bufsize;
2589 
2590 	/*
2591 	 * Round up the end of the previous packet to the next longword.
2592 	 *
2593 	 * Drop the packet if there's no room and no hope of room
2594 	 * If the packet would overflow the storage buffer or the storage
2595 	 * buffer is considered immutable by the buffer model, try to rotate
2596 	 * the buffer and wakeup pending processes.
2597 	 */
2598 #ifdef COMPAT_FREEBSD32
2599 	if (d->bd_compat32)
2600 		curlen = BPF_WORDALIGN32(d->bd_slen);
2601 	else
2602 #endif
2603 		curlen = BPF_WORDALIGN(d->bd_slen);
2604 	if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2605 		if (d->bd_fbuf == NULL) {
2606 			/*
2607 			 * There's no room in the store buffer, and no
2608 			 * prospect of room, so drop the packet.  Notify the
2609 			 * buffer model.
2610 			 */
2611 			bpf_buffull(d);
2612 			counter_u64_add(d->bd_dcount, 1);
2613 			return;
2614 		}
2615 		KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2616 		ROTATE_BUFFERS(d);
2617 		do_wakeup = 1;
2618 		curlen = 0;
2619 	} else {
2620 		if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
2621 			/*
2622 			 * Immediate mode is set, or the read timeout has
2623 			 * already expired during a select call.  A packet
2624 			 * arrived, so the reader should be woken up.
2625 			 */
2626 			do_wakeup = 1;
2627 		}
2628 		pad = curlen - d->bd_slen;
2629 		KASSERT(pad >= 0 && pad <= sizeof(zeroes),
2630 		    ("%s: invalid pad byte count %d", __func__, pad));
2631 		if (pad > 0) {
2632 			/* Zero pad bytes. */
2633 			bpf_append_bytes(d, d->bd_sbuf, d->bd_slen, zeroes,
2634 			    pad);
2635 		}
2636 	}
2637 
2638 	caplen = totlen - hdrlen;
2639 	tstype = d->bd_tstamp;
2640 	do_timestamp = tstype != BPF_T_NONE;
2641 #ifndef BURN_BRIDGES
2642 	if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2643 		struct bpf_ts ts;
2644 		if (do_timestamp)
2645 			bpf_bintime2ts(bt, &ts, tstype);
2646 #ifdef COMPAT_FREEBSD32
2647 		if (d->bd_compat32) {
2648 			bzero(&hdr32_old, sizeof(hdr32_old));
2649 			if (do_timestamp) {
2650 				hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2651 				hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2652 			}
2653 			hdr32_old.bh_datalen = pktlen;
2654 			hdr32_old.bh_hdrlen = hdrlen;
2655 			hdr32_old.bh_caplen = caplen;
2656 			bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2657 			    sizeof(hdr32_old));
2658 			goto copy;
2659 		}
2660 #endif
2661 		bzero(&hdr_old, sizeof(hdr_old));
2662 		if (do_timestamp) {
2663 			hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2664 			hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2665 		}
2666 		hdr_old.bh_datalen = pktlen;
2667 		hdr_old.bh_hdrlen = hdrlen;
2668 		hdr_old.bh_caplen = caplen;
2669 		bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2670 		    sizeof(hdr_old));
2671 		goto copy;
2672 	}
2673 #endif
2674 
2675 	/*
2676 	 * Append the bpf header.  Note we append the actual header size, but
2677 	 * move forward the length of the header plus padding.
2678 	 */
2679 	bzero(&hdr, sizeof(hdr));
2680 	if (do_timestamp)
2681 		bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2682 	hdr.bh_datalen = pktlen;
2683 	hdr.bh_hdrlen = hdrlen;
2684 	hdr.bh_caplen = caplen;
2685 	bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2686 
2687 	/*
2688 	 * Copy the packet data into the store buffer and update its length.
2689 	 */
2690 #ifndef BURN_BRIDGES
2691 copy:
2692 #endif
2693 	(*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2694 	d->bd_slen = curlen + totlen;
2695 
2696 	if (do_wakeup)
2697 		bpf_wakeup(d);
2698 }
2699 
2700 /*
2701  * Free buffers currently in use by a descriptor.
2702  * Called on close.
2703  */
2704 static void
2705 bpfd_free(epoch_context_t ctx)
2706 {
2707 	struct bpf_d *d;
2708 	struct bpf_program_buffer *p;
2709 
2710 	/*
2711 	 * We don't need to lock out interrupts since this descriptor has
2712 	 * been detached from its interface and it yet hasn't been marked
2713 	 * free.
2714 	 */
2715 	d = __containerof(ctx, struct bpf_d, epoch_ctx);
2716 	bpf_free(d);
2717 	if (d->bd_rfilter != NULL) {
2718 		p = __containerof((void *)d->bd_rfilter,
2719 		    struct bpf_program_buffer, buffer);
2720 #ifdef BPF_JITTER
2721 		p->func = d->bd_bfilter;
2722 #endif
2723 		bpf_program_buffer_free(&p->epoch_ctx);
2724 	}
2725 	if (d->bd_wfilter != NULL) {
2726 		p = __containerof((void *)d->bd_wfilter,
2727 		    struct bpf_program_buffer, buffer);
2728 #ifdef BPF_JITTER
2729 		p->func = NULL;
2730 #endif
2731 		bpf_program_buffer_free(&p->epoch_ctx);
2732 	}
2733 
2734 	mtx_destroy(&d->bd_lock);
2735 	counter_u64_free(d->bd_rcount);
2736 	counter_u64_free(d->bd_dcount);
2737 	counter_u64_free(d->bd_fcount);
2738 	counter_u64_free(d->bd_wcount);
2739 	counter_u64_free(d->bd_wfcount);
2740 	counter_u64_free(d->bd_wdcount);
2741 	counter_u64_free(d->bd_zcopy);
2742 	free(d, M_BPF);
2743 }
2744 
2745 /*
2746  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
2747  * fixed size of the link header (variable length headers not yet supported).
2748  */
2749 void
2750 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2751 {
2752 
2753 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2754 }
2755 
2756 /*
2757  * Attach an interface to bpf.  ifp is a pointer to the structure
2758  * defining the interface to be attached, dlt is the link layer type,
2759  * and hdrlen is the fixed size of the link header (variable length
2760  * headers are not yet supporrted).
2761  */
2762 void
2763 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2764     struct bpf_if **driverp)
2765 {
2766 	struct bpf_if *bp;
2767 
2768 	KASSERT(*driverp == NULL,
2769 	    ("bpfattach2: driverp already initialized"));
2770 
2771 	bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2772 
2773 	CK_LIST_INIT(&bp->bif_dlist);
2774 	CK_LIST_INIT(&bp->bif_wlist);
2775 	bp->bif_ifp = ifp;
2776 	bp->bif_dlt = dlt;
2777 	bp->bif_hdrlen = hdrlen;
2778 	bp->bif_bpf = driverp;
2779 	refcount_init(&bp->bif_refcnt, 1);
2780 	*driverp = bp;
2781 	/*
2782 	 * Reference ifnet pointer, so it won't freed until
2783 	 * we release it.
2784 	 */
2785 	if_ref(ifp);
2786 	BPF_LOCK();
2787 	CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2788 	BPF_UNLOCK();
2789 
2790 	if (bootverbose && IS_DEFAULT_VNET(curvnet))
2791 		if_printf(ifp, "bpf attached\n");
2792 }
2793 
2794 #ifdef VIMAGE
2795 /*
2796  * When moving interfaces between vnet instances we need a way to
2797  * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2798  * after the vmove.  We unfortunately have no device driver infrastructure
2799  * to query the interface for these values after creation/attach, thus
2800  * add this as a workaround.
2801  */
2802 int
2803 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2804 {
2805 
2806 	if (bp == NULL)
2807 		return (ENXIO);
2808 	if (bif_dlt == NULL && bif_hdrlen == NULL)
2809 		return (0);
2810 
2811 	if (bif_dlt != NULL)
2812 		*bif_dlt = bp->bif_dlt;
2813 	if (bif_hdrlen != NULL)
2814 		*bif_hdrlen = bp->bif_hdrlen;
2815 
2816 	return (0);
2817 }
2818 #endif
2819 
2820 /*
2821  * Detach bpf from an interface. This involves detaching each descriptor
2822  * associated with the interface. Notify each descriptor as it's detached
2823  * so that any sleepers wake up and get ENXIO.
2824  */
2825 void
2826 bpfdetach(struct ifnet *ifp)
2827 {
2828 	struct bpf_if *bp, *bp_temp;
2829 	struct bpf_d *d;
2830 
2831 	BPF_LOCK();
2832 	/* Find all bpf_if struct's which reference ifp and detach them. */
2833 	CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2834 		if (ifp != bp->bif_ifp)
2835 			continue;
2836 
2837 		CK_LIST_REMOVE(bp, bif_next);
2838 		*bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
2839 
2840 		CTR4(KTR_NET,
2841 		    "%s: sheduling free for encap %d (%p) for if %p",
2842 		    __func__, bp->bif_dlt, bp, ifp);
2843 
2844 		/* Detach common descriptors */
2845 		while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2846 			bpf_detachd_locked(d, true);
2847 		}
2848 
2849 		/* Detach writer-only descriptors */
2850 		while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2851 			bpf_detachd_locked(d, true);
2852 		}
2853 		bpfif_rele(bp);
2854 	}
2855 	BPF_UNLOCK();
2856 }
2857 
2858 /*
2859  * Get a list of available data link type of the interface.
2860  */
2861 static int
2862 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2863 {
2864 	struct ifnet *ifp;
2865 	struct bpf_if *bp;
2866 	u_int *lst;
2867 	int error, n, n1;
2868 
2869 	BPF_LOCK_ASSERT();
2870 
2871 	ifp = d->bd_bif->bif_ifp;
2872 	n1 = 0;
2873 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2874 		if (bp->bif_ifp == ifp)
2875 			n1++;
2876 	}
2877 	if (bfl->bfl_list == NULL) {
2878 		bfl->bfl_len = n1;
2879 		return (0);
2880 	}
2881 	if (n1 > bfl->bfl_len)
2882 		return (ENOMEM);
2883 
2884 	lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2885 	n = 0;
2886 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2887 		if (bp->bif_ifp != ifp)
2888 			continue;
2889 		lst[n++] = bp->bif_dlt;
2890 	}
2891 	error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2892 	free(lst, M_TEMP);
2893 	bfl->bfl_len = n;
2894 	return (error);
2895 }
2896 
2897 /*
2898  * Set the data link type of a BPF instance.
2899  */
2900 static int
2901 bpf_setdlt(struct bpf_d *d, u_int dlt)
2902 {
2903 	int error, opromisc;
2904 	struct ifnet *ifp;
2905 	struct bpf_if *bp;
2906 
2907 	BPF_LOCK_ASSERT();
2908 	MPASS(d->bd_bif != NULL);
2909 
2910 	/*
2911 	 * It is safe to check bd_bif without BPFD_LOCK, it can not be
2912 	 * changed while we hold global lock.
2913 	 */
2914 	if (d->bd_bif->bif_dlt == dlt)
2915 		return (0);
2916 
2917 	ifp = d->bd_bif->bif_ifp;
2918 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2919 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2920 			break;
2921 	}
2922 	if (bp == NULL)
2923 		return (EINVAL);
2924 
2925 	opromisc = d->bd_promisc;
2926 	bpf_attachd(d, bp);
2927 	if (opromisc) {
2928 		error = ifpromisc(bp->bif_ifp, 1);
2929 		if (error)
2930 			if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
2931 			    __func__, error);
2932 		else
2933 			d->bd_promisc = 1;
2934 	}
2935 	return (0);
2936 }
2937 
2938 static void
2939 bpf_drvinit(void *unused)
2940 {
2941 	struct cdev *dev;
2942 
2943 	sx_init(&bpf_sx, "bpf global lock");
2944 	CK_LIST_INIT(&bpf_iflist);
2945 
2946 	dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2947 	/* For compatibility */
2948 	make_dev_alias(dev, "bpf0");
2949 }
2950 
2951 /*
2952  * Zero out the various packet counters associated with all of the bpf
2953  * descriptors.  At some point, we will probably want to get a bit more
2954  * granular and allow the user to specify descriptors to be zeroed.
2955  */
2956 static void
2957 bpf_zero_counters(void)
2958 {
2959 	struct bpf_if *bp;
2960 	struct bpf_d *bd;
2961 
2962 	BPF_LOCK();
2963 	/*
2964 	 * We are protected by global lock here, interfaces and
2965 	 * descriptors can not be deleted while we hold it.
2966 	 */
2967 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2968 		CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2969 			counter_u64_zero(bd->bd_rcount);
2970 			counter_u64_zero(bd->bd_dcount);
2971 			counter_u64_zero(bd->bd_fcount);
2972 			counter_u64_zero(bd->bd_wcount);
2973 			counter_u64_zero(bd->bd_wfcount);
2974 			counter_u64_zero(bd->bd_zcopy);
2975 		}
2976 	}
2977 	BPF_UNLOCK();
2978 }
2979 
2980 /*
2981  * Fill filter statistics
2982  */
2983 static void
2984 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2985 {
2986 
2987 	BPF_LOCK_ASSERT();
2988 	bzero(d, sizeof(*d));
2989 	d->bd_structsize = sizeof(*d);
2990 	d->bd_immediate = bd->bd_immediate;
2991 	d->bd_promisc = bd->bd_promisc;
2992 	d->bd_hdrcmplt = bd->bd_hdrcmplt;
2993 	d->bd_direction = bd->bd_direction;
2994 	d->bd_feedback = bd->bd_feedback;
2995 	d->bd_async = bd->bd_async;
2996 	d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
2997 	d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
2998 	d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
2999 	d->bd_sig = bd->bd_sig;
3000 	d->bd_slen = bd->bd_slen;
3001 	d->bd_hlen = bd->bd_hlen;
3002 	d->bd_bufsize = bd->bd_bufsize;
3003 	d->bd_pid = bd->bd_pid;
3004 	strlcpy(d->bd_ifname,
3005 	    bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
3006 	d->bd_locked = bd->bd_locked;
3007 	d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
3008 	d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
3009 	d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
3010 	d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
3011 	d->bd_bufmode = bd->bd_bufmode;
3012 }
3013 
3014 /*
3015  * Handle `netstat -B' stats request
3016  */
3017 static int
3018 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
3019 {
3020 	static const struct xbpf_d zerostats;
3021 	struct xbpf_d *xbdbuf, *xbd, tempstats;
3022 	int index, error;
3023 	struct bpf_if *bp;
3024 	struct bpf_d *bd;
3025 
3026 	/*
3027 	 * XXX This is not technically correct. It is possible for non
3028 	 * privileged users to open bpf devices. It would make sense
3029 	 * if the users who opened the devices were able to retrieve
3030 	 * the statistics for them, too.
3031 	 */
3032 	error = priv_check(req->td, PRIV_NET_BPF);
3033 	if (error)
3034 		return (error);
3035 	/*
3036 	 * Check to see if the user is requesting that the counters be
3037 	 * zeroed out.  Explicitly check that the supplied data is zeroed,
3038 	 * as we aren't allowing the user to set the counters currently.
3039 	 */
3040 	if (req->newptr != NULL) {
3041 		if (req->newlen != sizeof(tempstats))
3042 			return (EINVAL);
3043 		memset(&tempstats, 0, sizeof(tempstats));
3044 		error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
3045 		if (error)
3046 			return (error);
3047 		if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
3048 			return (EINVAL);
3049 		bpf_zero_counters();
3050 		return (0);
3051 	}
3052 	if (req->oldptr == NULL)
3053 		return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
3054 	if (bpf_bpfd_cnt == 0)
3055 		return (SYSCTL_OUT(req, 0, 0));
3056 	xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
3057 	BPF_LOCK();
3058 	if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
3059 		BPF_UNLOCK();
3060 		free(xbdbuf, M_BPF);
3061 		return (ENOMEM);
3062 	}
3063 	index = 0;
3064 	CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
3065 		/* Send writers-only first */
3066 		CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
3067 			xbd = &xbdbuf[index++];
3068 			bpfstats_fill_xbpf(xbd, bd);
3069 		}
3070 		CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
3071 			xbd = &xbdbuf[index++];
3072 			bpfstats_fill_xbpf(xbd, bd);
3073 		}
3074 	}
3075 	BPF_UNLOCK();
3076 	error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
3077 	free(xbdbuf, M_BPF);
3078 	return (error);
3079 }
3080 
3081 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
3082 
3083 #else /* !DEV_BPF && !NETGRAPH_BPF */
3084 
3085 /*
3086  * NOP stubs to allow bpf-using drivers to load and function.
3087  *
3088  * A 'better' implementation would allow the core bpf functionality
3089  * to be loaded at runtime.
3090  */
3091 
3092 void
3093 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3094 {
3095 }
3096 
3097 void
3098 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3099 {
3100 }
3101 
3102 void
3103 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3104 {
3105 }
3106 
3107 void
3108 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3109 {
3110 
3111 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3112 }
3113 
3114 void
3115 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3116 {
3117 
3118 	*driverp = (struct bpf_if *)&dead_bpf_if;
3119 }
3120 
3121 void
3122 bpfdetach(struct ifnet *ifp)
3123 {
3124 }
3125 
3126 u_int
3127 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3128 {
3129 	return -1;	/* "no filter" behaviour */
3130 }
3131 
3132 int
3133 bpf_validate(const struct bpf_insn *f, int len)
3134 {
3135 	return 0;		/* false */
3136 }
3137 
3138 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3139 
3140 #ifdef DDB
3141 static void
3142 bpf_show_bpf_if(struct bpf_if *bpf_if)
3143 {
3144 
3145 	if (bpf_if == NULL)
3146 		return;
3147 	db_printf("%p:\n", bpf_if);
3148 #define	BPF_DB_PRINTF(f, e)	db_printf("   %s = " f "\n", #e, bpf_if->e);
3149 #define	BPF_DB_PRINTF_RAW(f, e)	db_printf("   %s = " f "\n", #e, e);
3150 	/* bif_ext.bif_next */
3151 	/* bif_ext.bif_dlist */
3152 	BPF_DB_PRINTF("%#x", bif_dlt);
3153 	BPF_DB_PRINTF("%u", bif_hdrlen);
3154 	/* bif_wlist */
3155 	BPF_DB_PRINTF("%p", bif_ifp);
3156 	BPF_DB_PRINTF("%p", bif_bpf);
3157 	BPF_DB_PRINTF_RAW("%u", refcount_load(&bpf_if->bif_refcnt));
3158 }
3159 
3160 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3161 {
3162 
3163 	if (!have_addr) {
3164 		db_printf("usage: show bpf_if <struct bpf_if *>\n");
3165 		return;
3166 	}
3167 
3168 	bpf_show_bpf_if((struct bpf_if *)addr);
3169 }
3170 #endif
3171