xref: /netbsd-src/sys/net/bpf.c (revision 867d70fc718005c0918b8b8b2f9d7f2d52d0a0db)
1 /*	$NetBSD: bpf.c,v 1.249 2022/11/30 06:02:37 ozaki-r Exp $	*/
2 
3 /*
4  * Copyright (c) 1990, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from the Stanford/CMU enet packet filter,
8  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
9  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
10  * Berkeley Laboratory.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)bpf.c	8.4 (Berkeley) 1/9/95
37  * static char rcsid[] =
38  * "Header: bpf.c,v 1.67 96/09/26 22:00:52 leres Exp ";
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: bpf.c,v 1.249 2022/11/30 06:02:37 ozaki-r Exp $");
43 
44 #if defined(_KERNEL_OPT)
45 #include "opt_bpf.h"
46 #include "sl.h"
47 #include "opt_net_mpsafe.h"
48 #endif
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/mbuf.h>
53 #include <sys/buf.h>
54 #include <sys/time.h>
55 #include <sys/proc.h>
56 #include <sys/ioctl.h>
57 #include <sys/conf.h>
58 #include <sys/vnode.h>
59 #include <sys/queue.h>
60 #include <sys/stat.h>
61 #include <sys/module.h>
62 #include <sys/atomic.h>
63 #include <sys/cpu.h>
64 
65 #include <sys/file.h>
66 #include <sys/filedesc.h>
67 #include <sys/tty.h>
68 #include <sys/uio.h>
69 
70 #include <sys/protosw.h>
71 #include <sys/socket.h>
72 #include <sys/errno.h>
73 #include <sys/kernel.h>
74 #include <sys/poll.h>
75 #include <sys/sysctl.h>
76 #include <sys/kauth.h>
77 #include <sys/syslog.h>
78 #include <sys/percpu.h>
79 #include <sys/pserialize.h>
80 #include <sys/lwp.h>
81 #include <sys/xcall.h>
82 
83 #include <net/if.h>
84 #include <net/slip.h>
85 
86 #include <net/bpf.h>
87 #include <net/bpfdesc.h>
88 #include <net/bpfjit.h>
89 
90 #include <net/if_arc.h>
91 #include <net/if_ether.h>
92 #include <net/if_types.h>
93 
94 #include <netinet/in.h>
95 #include <netinet/if_inarp.h>
96 
97 
98 #include <compat/sys/sockio.h>
99 
100 #ifndef BPF_BUFSIZE
101 /*
102  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
103  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
104  */
105 # define BPF_BUFSIZE 32768
106 #endif
107 
108 #define PRINET  26			/* interruptible */
109 
110 /*
111  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
112  * XXX the default values should be computed dynamically based
113  * on available memory size and available mbuf clusters.
114  */
115 static int bpf_bufsize = BPF_BUFSIZE;
116 static int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
117 static bool bpf_jit = false;
118 
119 struct bpfjit_ops bpfjit_module_ops = {
120 	.bj_generate_code = NULL,
121 	.bj_free_code = NULL
122 };
123 
124 /*
125  * Global BPF statistics returned by net.bpf.stats sysctl.
126  */
127 static struct percpu	*bpf_gstats_percpu; /* struct bpf_stat */
128 
129 #define BPF_STATINC(id)					\
130 	{						\
131 		struct bpf_stat *__stats =		\
132 		    percpu_getref(bpf_gstats_percpu);	\
133 		__stats->bs_##id++;			\
134 		percpu_putref(bpf_gstats_percpu);	\
135 	}
136 
137 /*
138  * Locking notes:
139  * - bpf_mtx (adaptive mutex) protects:
140  *   - Gobal lists: bpf_iflist and bpf_dlist
141  *   - struct bpf_if
142  *   - bpf_close
143  *   - bpf_psz (pserialize)
144  * - struct bpf_d has two mutexes:
145  *   - bd_buf_mtx (spin mutex) protects the buffers that can be accessed
146  *     on packet tapping
147  *   - bd_mtx (adaptive mutex) protects member variables other than the buffers
148  * - Locking order: bpf_mtx => bpf_d#bd_mtx => bpf_d#bd_buf_mtx
149  * - struct bpf_d obtained via fp->f_bpf in bpf_read and bpf_write is
150  *   never freed because struct bpf_d is only freed in bpf_close and
151  *   bpf_close never be called while executing bpf_read and bpf_write
152  * - A filter that is assigned to bpf_d can be replaced with another filter
153  *   while tapping packets, so it needs to be done atomically
154  * - struct bpf_d is iterated on bpf_dlist with psz
155  * - struct bpf_if is iterated on bpf_iflist with psz or psref
156  */
157 /*
158  * Use a mutex to avoid a race condition between gathering the stats/peers
159  * and opening/closing the device.
160  */
161 static kmutex_t bpf_mtx;
162 
163 static struct psref_class	*bpf_psref_class __read_mostly;
164 static pserialize_t		bpf_psz;
165 
166 static inline void
167 bpf_if_acquire(struct bpf_if *bp, struct psref *psref)
168 {
169 
170 	psref_acquire(psref, &bp->bif_psref, bpf_psref_class);
171 }
172 
173 static inline void
174 bpf_if_release(struct bpf_if *bp, struct psref *psref)
175 {
176 
177 	psref_release(psref, &bp->bif_psref, bpf_psref_class);
178 }
179 
180 /*
181  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
182  *  bpf_dtab holds the descriptors, indexed by minor device #
183  */
184 static struct pslist_head bpf_iflist;
185 static struct pslist_head bpf_dlist;
186 
187 /* Macros for bpf_d on bpf_dlist */
188 #define BPF_DLIST_WRITER_INSERT_HEAD(__d)				\
189 	PSLIST_WRITER_INSERT_HEAD(&bpf_dlist, (__d), bd_bpf_dlist_entry)
190 #define BPF_DLIST_READER_FOREACH(__d)					\
191 	PSLIST_READER_FOREACH((__d), &bpf_dlist, struct bpf_d,		\
192 	                      bd_bpf_dlist_entry)
193 #define BPF_DLIST_WRITER_FOREACH(__d)					\
194 	PSLIST_WRITER_FOREACH((__d), &bpf_dlist, struct bpf_d,		\
195 	                      bd_bpf_dlist_entry)
196 #define BPF_DLIST_ENTRY_INIT(__d)					\
197 	PSLIST_ENTRY_INIT((__d), bd_bpf_dlist_entry)
198 #define BPF_DLIST_WRITER_REMOVE(__d)					\
199 	PSLIST_WRITER_REMOVE((__d), bd_bpf_dlist_entry)
200 #define BPF_DLIST_ENTRY_DESTROY(__d)					\
201 	PSLIST_ENTRY_DESTROY((__d), bd_bpf_dlist_entry)
202 
203 /* Macros for bpf_if on bpf_iflist */
204 #define BPF_IFLIST_WRITER_INSERT_HEAD(__bp)				\
205 	PSLIST_WRITER_INSERT_HEAD(&bpf_iflist, (__bp), bif_iflist_entry)
206 #define BPF_IFLIST_READER_FOREACH(__bp)					\
207 	PSLIST_READER_FOREACH((__bp), &bpf_iflist, struct bpf_if,	\
208 	                      bif_iflist_entry)
209 #define BPF_IFLIST_WRITER_FOREACH(__bp)					\
210 	PSLIST_WRITER_FOREACH((__bp), &bpf_iflist, struct bpf_if,	\
211 	                      bif_iflist_entry)
212 #define BPF_IFLIST_WRITER_REMOVE(__bp)					\
213 	PSLIST_WRITER_REMOVE((__bp), bif_iflist_entry)
214 #define BPF_IFLIST_ENTRY_INIT(__bp)					\
215 	PSLIST_ENTRY_INIT((__bp), bif_iflist_entry)
216 #define BPF_IFLIST_ENTRY_DESTROY(__bp)					\
217 	PSLIST_ENTRY_DESTROY((__bp), bif_iflist_entry)
218 
219 /* Macros for bpf_d on bpf_if#bif_dlist_pslist */
220 #define BPFIF_DLIST_READER_FOREACH(__d, __bp)				\
221 	PSLIST_READER_FOREACH((__d), &(__bp)->bif_dlist_head, struct bpf_d, \
222 	                      bd_bif_dlist_entry)
223 #define BPFIF_DLIST_WRITER_INSERT_HEAD(__bp, __d)			\
224 	PSLIST_WRITER_INSERT_HEAD(&(__bp)->bif_dlist_head, (__d),	\
225 	                          bd_bif_dlist_entry)
226 #define BPFIF_DLIST_WRITER_REMOVE(__d)					\
227 	PSLIST_WRITER_REMOVE((__d), bd_bif_dlist_entry)
228 #define BPFIF_DLIST_ENTRY_INIT(__d)					\
229 	PSLIST_ENTRY_INIT((__d), bd_bif_dlist_entry)
230 #define	BPFIF_DLIST_READER_EMPTY(__bp)					\
231 	(PSLIST_READER_FIRST(&(__bp)->bif_dlist_head, struct bpf_d,	\
232 	                     bd_bif_dlist_entry) == NULL)
233 #define	BPFIF_DLIST_WRITER_EMPTY(__bp)					\
234 	(PSLIST_WRITER_FIRST(&(__bp)->bif_dlist_head, struct bpf_d,	\
235 	                     bd_bif_dlist_entry) == NULL)
236 #define BPFIF_DLIST_ENTRY_DESTROY(__d)					\
237 	PSLIST_ENTRY_DESTROY((__d), bd_bif_dlist_entry)
238 
239 static int	bpf_allocbufs(struct bpf_d *);
240 static u_int	bpf_xfilter(struct bpf_filter **, void *, u_int, u_int);
241 static void	bpf_deliver(struct bpf_if *,
242 		            void *(*cpfn)(void *, const void *, size_t),
243 		            void *, u_int, u_int, const u_int);
244 static void	bpf_freed(struct bpf_d *);
245 static void	bpf_free_filter(struct bpf_filter *);
246 static void	bpf_ifname(struct ifnet *, struct ifreq *);
247 static void	*bpf_mcpy(void *, const void *, size_t);
248 static int	bpf_movein(struct ifnet *, struct uio *, int, uint64_t,
249 			        struct mbuf **, struct sockaddr *,
250 				struct bpf_filter **);
251 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
252 static void	bpf_detachd(struct bpf_d *);
253 static int	bpf_setif(struct bpf_d *, struct ifreq *);
254 static int	bpf_setf(struct bpf_d *, struct bpf_program *, u_long);
255 static void	bpf_timed_out(void *);
256 static inline void
257 		bpf_wakeup(struct bpf_d *);
258 static int	bpf_hdrlen(struct bpf_d *);
259 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
260     void *(*)(void *, const void *, size_t), struct timespec *);
261 static void	reset_d(struct bpf_d *);
262 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
263 static int	bpf_setdlt(struct bpf_d *, u_int);
264 
265 static int	bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t,
266     int);
267 static int	bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t,
268     int);
269 static int	bpf_ioctl(struct file *, u_long, void *);
270 static int	bpf_poll(struct file *, int);
271 static int	bpf_stat(struct file *, struct stat *);
272 static int	bpf_close(struct file *);
273 static int	bpf_kqfilter(struct file *, struct knote *);
274 
275 static const struct fileops bpf_fileops = {
276 	.fo_name = "bpf",
277 	.fo_read = bpf_read,
278 	.fo_write = bpf_write,
279 	.fo_ioctl = bpf_ioctl,
280 	.fo_fcntl = fnullop_fcntl,
281 	.fo_poll = bpf_poll,
282 	.fo_stat = bpf_stat,
283 	.fo_close = bpf_close,
284 	.fo_kqfilter = bpf_kqfilter,
285 	.fo_restart = fnullop_restart,
286 };
287 
288 dev_type_open(bpfopen);
289 
290 const struct cdevsw bpf_cdevsw = {
291 	.d_open = bpfopen,
292 	.d_close = noclose,
293 	.d_read = noread,
294 	.d_write = nowrite,
295 	.d_ioctl = noioctl,
296 	.d_stop = nostop,
297 	.d_tty = notty,
298 	.d_poll = nopoll,
299 	.d_mmap = nommap,
300 	.d_kqfilter = nokqfilter,
301 	.d_discard = nodiscard,
302 	.d_flag = D_OTHER | D_MPSAFE
303 };
304 
305 bpfjit_func_t
306 bpf_jit_generate(bpf_ctx_t *bc, void *code, size_t size)
307 {
308 	struct bpfjit_ops *ops = &bpfjit_module_ops;
309 	bpfjit_func_t (*generate_code)(const bpf_ctx_t *,
310 	    const struct bpf_insn *, size_t);
311 
312 	generate_code = atomic_load_acquire(&ops->bj_generate_code);
313 	if (generate_code != NULL) {
314 		return generate_code(bc, code, size);
315 	}
316 	return NULL;
317 }
318 
319 void
320 bpf_jit_freecode(bpfjit_func_t jcode)
321 {
322 	KASSERT(bpfjit_module_ops.bj_free_code != NULL);
323 	bpfjit_module_ops.bj_free_code(jcode);
324 }
325 
326 static int
327 bpf_movein(struct ifnet *ifp, struct uio *uio, int linktype, uint64_t mtu, struct mbuf **mp,
328 	   struct sockaddr *sockp, struct bpf_filter **wfilter)
329 {
330 	struct mbuf *m, *m0, *n;
331 	int error;
332 	size_t len;
333 	size_t hlen;
334 	size_t align;
335 	u_int slen;
336 
337 	/*
338 	 * Build a sockaddr based on the data link layer type.
339 	 * We do this at this level because the ethernet header
340 	 * is copied directly into the data field of the sockaddr.
341 	 * In the case of SLIP, there is no header and the packet
342 	 * is forwarded as is.
343 	 * Also, we are careful to leave room at the front of the mbuf
344 	 * for the link level header.
345 	 */
346 	switch (linktype) {
347 
348 	case DLT_SLIP:
349 		sockp->sa_family = AF_INET;
350 		hlen = 0;
351 		align = 0;
352 		break;
353 
354 	case DLT_PPP:
355 		sockp->sa_family = AF_UNSPEC;
356 		hlen = 0;
357 		align = 0;
358 		break;
359 
360 	case DLT_EN10MB:
361 		sockp->sa_family = AF_UNSPEC;
362 		/* XXX Would MAXLINKHDR be better? */
363  		/* 6(dst)+6(src)+2(type) */
364 		hlen = sizeof(struct ether_header);
365 		align = 2;
366 		break;
367 
368 	case DLT_ARCNET:
369 		sockp->sa_family = AF_UNSPEC;
370 		hlen = ARC_HDRLEN;
371 		align = 5;
372 		break;
373 
374 	case DLT_FDDI:
375 		sockp->sa_family = AF_LINK;
376 		/* XXX 4(FORMAC)+6(dst)+6(src) */
377 		hlen = 16;
378 		align = 0;
379 		break;
380 
381 	case DLT_ECONET:
382 		sockp->sa_family = AF_UNSPEC;
383 		hlen = 6;
384 		align = 2;
385 		break;
386 
387 	case DLT_NULL:
388 		sockp->sa_family = AF_UNSPEC;
389 		if (ifp->if_type == IFT_LOOP) {
390 			/* Set here to apply the following validations */
391 			hlen = sizeof(uint32_t);
392 		} else
393 			hlen = 0;
394 		align = 0;
395 		break;
396 
397 	default:
398 		return (EIO);
399 	}
400 
401 	len = uio->uio_resid;
402 	/*
403 	 * If there aren't enough bytes for a link level header or the
404 	 * packet length exceeds the interface mtu, return an error.
405 	 */
406 	if (len - hlen > mtu)
407 		return (EMSGSIZE);
408 
409 	m0 = m = m_gethdr(M_WAIT, MT_DATA);
410 	m_reset_rcvif(m);
411 	m->m_pkthdr.len = (int)(len - hlen);
412 	if (len + align > MHLEN) {
413 		m_clget(m, M_WAIT);
414 		if ((m->m_flags & M_EXT) == 0) {
415 			error = ENOBUFS;
416 			goto bad;
417 		}
418 	}
419 
420 	/* Insure the data is properly aligned */
421 	if (align > 0)
422 		m->m_data += align;
423 
424 	for (;;) {
425 		len = M_TRAILINGSPACE(m);
426 		if (len > uio->uio_resid)
427 			len = uio->uio_resid;
428 		error = uiomove(mtod(m, void *), len, uio);
429 		if (error)
430 			goto bad;
431 		m->m_len = len;
432 
433 		if (uio->uio_resid == 0)
434 			break;
435 
436 		n = m_get(M_WAIT, MT_DATA);
437 		m_clget(n, M_WAIT);	/* if fails, there is no problem */
438 		m->m_next = n;
439 		m = n;
440 	}
441 
442 	slen = bpf_xfilter(wfilter, mtod(m, u_char *), len, len);
443 	if (slen == 0) {
444 		error = EPERM;
445 		goto bad;
446 	}
447 
448 	if (hlen != 0) {
449 		if (linktype == DLT_NULL && ifp->if_type == IFT_LOOP) {
450 			uint32_t af;
451 			/* the link header indicates the address family */
452 			memcpy(&af, mtod(m0, void *), sizeof(af));
453 			sockp->sa_family = af;
454 		} else {
455 			/* move link level header in the top of mbuf to sa_data */
456 			memcpy(sockp->sa_data, mtod(m0, void *), hlen);
457 		}
458 		m0->m_data += hlen;
459 		m0->m_len -= hlen;
460 	}
461 
462 	*mp = m0;
463 	return (0);
464 
465 bad:
466 	m_freem(m0);
467 	return (error);
468 }
469 
470 /*
471  * Attach file to the bpf interface, i.e. make d listen on bp.
472  */
473 static void
474 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
475 {
476 	struct bpf_event_tracker *t;
477 
478 	KASSERT(mutex_owned(&bpf_mtx));
479 	KASSERT(mutex_owned(d->bd_mtx));
480 	/*
481 	 * Point d at bp, and add d to the interface's list of listeners.
482 	 * Finally, point the driver's bpf cookie at the interface so
483 	 * it will divert packets to bpf.
484 	 */
485 	d->bd_bif = bp;
486 	BPFIF_DLIST_WRITER_INSERT_HEAD(bp, d);
487 
488 	*bp->bif_driverp = bp;
489 
490 	SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) {
491 		t->bet_notify(bp, bp->bif_ifp, bp->bif_dlt,
492 		    BPF_TRACK_EVENT_ATTACH);
493 	}
494 }
495 
496 /*
497  * Detach a file from its interface.
498  */
499 static void
500 bpf_detachd(struct bpf_d *d)
501 {
502 	struct bpf_if *bp;
503 	struct bpf_event_tracker *t;
504 
505 	KASSERT(mutex_owned(&bpf_mtx));
506 	KASSERT(mutex_owned(d->bd_mtx));
507 
508 	bp = d->bd_bif;
509 	/*
510 	 * Check if this descriptor had requested promiscuous mode.
511 	 * If so, turn it off.
512 	 */
513 	if (d->bd_promisc) {
514 		int error __diagused;
515 
516 		d->bd_promisc = 0;
517 		/*
518 		 * Take device out of promiscuous mode.  Since we were
519 		 * able to enter promiscuous mode, we should be able
520 		 * to turn it off.  But we can get an error if
521 		 * the interface was configured down, so only panic
522 		 * if we don't get an unexpected error.
523 		 */
524 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
525   		error = ifpromisc(bp->bif_ifp, 0);
526 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
527 #ifdef DIAGNOSTIC
528 		if (error)
529 			printf("%s: ifpromisc failed: %d", __func__, error);
530 #endif
531 	}
532 
533 	/* Remove d from the interface's descriptor list. */
534 	BPFIF_DLIST_WRITER_REMOVE(d);
535 
536 	pserialize_perform(bpf_psz);
537 
538 	if (BPFIF_DLIST_WRITER_EMPTY(bp)) {
539 		/*
540 		 * Let the driver know that there are no more listeners.
541 		 */
542 		*d->bd_bif->bif_driverp = NULL;
543 	}
544 
545 	d->bd_bif = NULL;
546 
547 	SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) {
548 		t->bet_notify(bp, bp->bif_ifp, bp->bif_dlt,
549 		    BPF_TRACK_EVENT_DETACH);
550 	}
551 }
552 
553 static void
554 bpf_init(void)
555 {
556 
557 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
558 	bpf_psz = pserialize_create();
559 	bpf_psref_class = psref_class_create("bpf", IPL_SOFTNET);
560 
561 	PSLIST_INIT(&bpf_iflist);
562 	PSLIST_INIT(&bpf_dlist);
563 
564 	bpf_gstats_percpu = percpu_alloc(sizeof(struct bpf_stat));
565 
566 	return;
567 }
568 
569 /*
570  * bpfilterattach() is called at boot time.  We don't need to do anything
571  * here, since any initialization will happen as part of module init code.
572  */
573 /* ARGSUSED */
574 void
575 bpfilterattach(int n)
576 {
577 
578 }
579 
580 /*
581  * Open ethernet device. Clones.
582  */
583 /* ARGSUSED */
584 int
585 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
586 {
587 	struct bpf_d *d;
588 	struct file *fp;
589 	int error, fd;
590 
591 	/* falloc() will fill in the descriptor for us. */
592 	if ((error = fd_allocfile(&fp, &fd)) != 0)
593 		return error;
594 
595 	d = kmem_zalloc(sizeof(*d), KM_SLEEP);
596 	d->bd_bufsize = bpf_bufsize;
597 	d->bd_direction = BPF_D_INOUT;
598 	d->bd_feedback = 0;
599 	d->bd_pid = l->l_proc->p_pid;
600 #ifdef _LP64
601 	if (curproc->p_flag & PK_32)
602 		d->bd_compat32 = 1;
603 #endif
604 	getnanotime(&d->bd_btime);
605 	d->bd_atime = d->bd_mtime = d->bd_btime;
606 	callout_init(&d->bd_callout, CALLOUT_MPSAFE);
607 	selinit(&d->bd_sel);
608 	d->bd_jitcode = NULL;
609 	d->bd_rfilter = NULL;
610 	d->bd_wfilter = NULL;
611 	d->bd_locked = 0;
612 	BPF_DLIST_ENTRY_INIT(d);
613 	BPFIF_DLIST_ENTRY_INIT(d);
614 	d->bd_mtx = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
615 	d->bd_buf_mtx = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
616 	cv_init(&d->bd_cv, "bpf");
617 
618 	mutex_enter(&bpf_mtx);
619 	BPF_DLIST_WRITER_INSERT_HEAD(d);
620 	mutex_exit(&bpf_mtx);
621 
622 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
623 }
624 
625 /*
626  * Close the descriptor by detaching it from its interface,
627  * deallocating its buffers, and marking it free.
628  */
629 /* ARGSUSED */
630 static int
631 bpf_close(struct file *fp)
632 {
633 	struct bpf_d *d;
634 
635 	mutex_enter(&bpf_mtx);
636 
637 	if ((d = fp->f_bpf) == NULL) {
638 		mutex_exit(&bpf_mtx);
639 		return 0;
640 	}
641 
642 	/*
643 	 * Refresh the PID associated with this bpf file.
644 	 */
645 	d->bd_pid = curproc->p_pid;
646 
647 	mutex_enter(d->bd_mtx);
648 	if (d->bd_state == BPF_WAITING)
649 		callout_halt(&d->bd_callout, d->bd_mtx);
650 	d->bd_state = BPF_IDLE;
651 	if (d->bd_bif)
652 		bpf_detachd(d);
653 	mutex_exit(d->bd_mtx);
654 
655 	BPF_DLIST_WRITER_REMOVE(d);
656 
657 	pserialize_perform(bpf_psz);
658 	mutex_exit(&bpf_mtx);
659 
660 	BPFIF_DLIST_ENTRY_DESTROY(d);
661 	BPF_DLIST_ENTRY_DESTROY(d);
662 	fp->f_bpf = NULL;
663 	bpf_freed(d);
664 	callout_destroy(&d->bd_callout);
665 	seldestroy(&d->bd_sel);
666 	mutex_obj_free(d->bd_mtx);
667 	mutex_obj_free(d->bd_buf_mtx);
668 	cv_destroy(&d->bd_cv);
669 
670 	kmem_free(d, sizeof(*d));
671 
672 	return (0);
673 }
674 
675 /*
676  * Rotate the packet buffers in descriptor d.  Move the store buffer
677  * into the hold slot, and the free buffer into the store slot.
678  * Zero the length of the new store buffer.
679  */
680 #define ROTATE_BUFFERS(d) \
681 	(d)->bd_hbuf = (d)->bd_sbuf; \
682 	(d)->bd_hlen = (d)->bd_slen; \
683 	(d)->bd_sbuf = (d)->bd_fbuf; \
684 	(d)->bd_slen = 0; \
685 	(d)->bd_fbuf = NULL;
686 /*
687  *  bpfread - read next chunk of packets from buffers
688  */
689 static int
690 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
691     kauth_cred_t cred, int flags)
692 {
693 	struct bpf_d *d = fp->f_bpf;
694 	int timed_out;
695 	int error;
696 
697 	/*
698 	 * Refresh the PID associated with this bpf file.
699 	 */
700 	d->bd_pid = curproc->p_pid;
701 
702 	getnanotime(&d->bd_atime);
703 	/*
704 	 * Restrict application to use a buffer the same size as
705 	 * the kernel buffers.
706 	 */
707 	if (uio->uio_resid != d->bd_bufsize)
708 		return (EINVAL);
709 
710 	mutex_enter(d->bd_mtx);
711 	if (d->bd_state == BPF_WAITING)
712 		callout_halt(&d->bd_callout, d->bd_mtx);
713 	timed_out = (d->bd_state == BPF_TIMED_OUT);
714 	d->bd_state = BPF_IDLE;
715 	mutex_exit(d->bd_mtx);
716 	/*
717 	 * If the hold buffer is empty, then do a timed sleep, which
718 	 * ends when the timeout expires or when enough packets
719 	 * have arrived to fill the store buffer.
720 	 */
721 	mutex_enter(d->bd_buf_mtx);
722 	while (d->bd_hbuf == NULL) {
723 		if (fp->f_flag & FNONBLOCK) {
724 			if (d->bd_slen == 0) {
725 				error = EWOULDBLOCK;
726 				goto out;
727 			}
728 			ROTATE_BUFFERS(d);
729 			break;
730 		}
731 
732 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
733 			/*
734 			 * A packet(s) either arrived since the previous
735 			 * read or arrived while we were asleep.
736 			 * Rotate the buffers and return what's here.
737 			 */
738 			ROTATE_BUFFERS(d);
739 			break;
740 		}
741 
742 		error = cv_timedwait_sig(&d->bd_cv, d->bd_buf_mtx, d->bd_rtout);
743 
744 		if (error == EINTR || error == ERESTART)
745 			goto out;
746 
747 		if (error == EWOULDBLOCK) {
748 			/*
749 			 * On a timeout, return what's in the buffer,
750 			 * which may be nothing.  If there is something
751 			 * in the store buffer, we can rotate the buffers.
752 			 */
753 			if (d->bd_hbuf)
754 				/*
755 				 * We filled up the buffer in between
756 				 * getting the timeout and arriving
757 				 * here, so we don't need to rotate.
758 				 */
759 				break;
760 
761 			if (d->bd_slen == 0) {
762 				error = 0;
763 				goto out;
764 			}
765 			ROTATE_BUFFERS(d);
766 			break;
767 		}
768 		if (error != 0)
769 			goto out;
770 	}
771 	/*
772 	 * At this point, we know we have something in the hold slot.
773 	 */
774 	mutex_exit(d->bd_buf_mtx);
775 
776 	/*
777 	 * Move data from hold buffer into user space.
778 	 * We know the entire buffer is transferred since
779 	 * we checked above that the read buffer is bpf_bufsize bytes.
780 	 */
781 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
782 
783 	mutex_enter(d->bd_buf_mtx);
784 	d->bd_fbuf = d->bd_hbuf;
785 	d->bd_hbuf = NULL;
786 	d->bd_hlen = 0;
787 out:
788 	mutex_exit(d->bd_buf_mtx);
789 	return (error);
790 }
791 
792 
793 /*
794  * If there are processes sleeping on this descriptor, wake them up.
795  */
796 static inline void
797 bpf_wakeup(struct bpf_d *d)
798 {
799 
800 	mutex_enter(d->bd_buf_mtx);
801 	cv_broadcast(&d->bd_cv);
802 	mutex_exit(d->bd_buf_mtx);
803 
804 	if (d->bd_async)
805 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
806 	selnotify(&d->bd_sel, 0, 0);
807 }
808 
809 static void
810 bpf_timed_out(void *arg)
811 {
812 	struct bpf_d *d = arg;
813 
814 	mutex_enter(d->bd_mtx);
815 	if (d->bd_state == BPF_WAITING) {
816 		d->bd_state = BPF_TIMED_OUT;
817 		if (d->bd_slen != 0)
818 			bpf_wakeup(d);
819 	}
820 	mutex_exit(d->bd_mtx);
821 }
822 
823 
824 static int
825 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
826     kauth_cred_t cred, int flags)
827 {
828 	struct bpf_d *d = fp->f_bpf;
829 	struct bpf_if *bp;
830 	struct ifnet *ifp;
831 	struct mbuf *m, *mc;
832 	int error;
833 	static struct sockaddr_storage dst;
834 	struct psref psref;
835 	int bound;
836 
837 	/*
838 	 * Refresh the PID associated with this bpf file.
839 	 */
840 	d->bd_pid = curproc->p_pid;
841 
842 	m = NULL;	/* XXX gcc */
843 
844 	bound = curlwp_bind();
845 	mutex_enter(d->bd_mtx);
846 	bp = d->bd_bif;
847 	if (bp == NULL) {
848 		mutex_exit(d->bd_mtx);
849 		error = ENXIO;
850 		goto out_bindx;
851 	}
852 	bpf_if_acquire(bp, &psref);
853 	mutex_exit(d->bd_mtx);
854 
855 	getnanotime(&d->bd_mtime);
856 
857 	ifp = bp->bif_ifp;
858 	if (if_is_deactivated(ifp)) {
859 		error = ENXIO;
860 		goto out;
861 	}
862 
863 	if (uio->uio_resid == 0) {
864 		error = 0;
865 		goto out;
866 	}
867 
868 	error = bpf_movein(ifp, uio, (int)bp->bif_dlt, ifp->if_mtu, &m,
869 		(struct sockaddr *) &dst, &d->bd_wfilter);
870 	if (error)
871 		goto out;
872 
873 	if (m->m_pkthdr.len > ifp->if_mtu) {
874 		m_freem(m);
875 		error = EMSGSIZE;
876 		goto out;
877 	}
878 
879 	if (d->bd_hdrcmplt)
880 		dst.ss_family = pseudo_AF_HDRCMPLT;
881 
882 	if (d->bd_feedback) {
883 		mc = m_dup(m, 0, M_COPYALL, M_NOWAIT);
884 		if (mc != NULL)
885 			m_set_rcvif(mc, ifp);
886 		/* Set M_PROMISC for outgoing packets to be discarded. */
887 		if (1 /*d->bd_direction == BPF_D_INOUT*/)
888 			m->m_flags |= M_PROMISC;
889 	} else
890 		mc = NULL;
891 
892 	error = if_output_lock(ifp, ifp, m, (struct sockaddr *) &dst, NULL);
893 
894 	if (mc != NULL) {
895 		if (error == 0) {
896 			int s = splsoftnet();
897 			KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
898 			ifp->_if_input(ifp, mc);
899 			KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
900 			splx(s);
901 		} else
902 			m_freem(mc);
903 	}
904 	/*
905 	 * The driver frees the mbuf.
906 	 */
907 out:
908 	bpf_if_release(bp, &psref);
909 out_bindx:
910 	curlwp_bindx(bound);
911 	return error;
912 }
913 
914 /*
915  * Reset a descriptor by flushing its packet buffer and clearing the
916  * receive and drop counts.
917  */
918 static void
919 reset_d(struct bpf_d *d)
920 {
921 
922 	KASSERT(mutex_owned(d->bd_mtx));
923 
924 	mutex_enter(d->bd_buf_mtx);
925 	if (d->bd_hbuf) {
926 		/* Free the hold buffer. */
927 		d->bd_fbuf = d->bd_hbuf;
928 		d->bd_hbuf = NULL;
929 	}
930 	d->bd_slen = 0;
931 	d->bd_hlen = 0;
932 	d->bd_rcount = 0;
933 	d->bd_dcount = 0;
934 	d->bd_ccount = 0;
935 	mutex_exit(d->bd_buf_mtx);
936 }
937 
938 /*
939  *  FIONREAD		Check for read packet available.
940  *  BIOCGBLEN		Get buffer len [for read()].
941  *  BIOCSETF		Set ethernet read filter.
942  *  BIOCFLUSH		Flush read packet buffer.
943  *  BIOCPROMISC		Put interface into promiscuous mode.
944  *  BIOCGDLT		Get link layer type.
945  *  BIOCGETIF		Get interface name.
946  *  BIOCSETIF		Set interface.
947  *  BIOCSRTIMEOUT	Set read timeout.
948  *  BIOCGRTIMEOUT	Get read timeout.
949  *  BIOCGSTATS		Get packet stats.
950  *  BIOCIMMEDIATE	Set immediate mode.
951  *  BIOCVERSION		Get filter language version.
952  *  BIOCGHDRCMPLT	Get "header already complete" flag.
953  *  BIOCSHDRCMPLT	Set "header already complete" flag.
954  *  BIOCSFEEDBACK	Set packet feedback mode.
955  *  BIOCGFEEDBACK	Get packet feedback mode.
956  *  BIOCGDIRECTION	Get packet direction flag
957  *  BIOCSDIRECTION	Set packet direction flag
958  */
959 /* ARGSUSED */
960 static int
961 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
962 {
963 	struct bpf_d *d = fp->f_bpf;
964 	int error = 0;
965 
966 	/*
967 	 * Refresh the PID associated with this bpf file.
968 	 */
969 	d->bd_pid = curproc->p_pid;
970 #ifdef _LP64
971 	if (curproc->p_flag & PK_32)
972 		d->bd_compat32 = 1;
973 	else
974 		d->bd_compat32 = 0;
975 #endif
976 
977 	mutex_enter(d->bd_mtx);
978 	if (d->bd_state == BPF_WAITING)
979 		callout_halt(&d->bd_callout, d->bd_mtx);
980 	d->bd_state = BPF_IDLE;
981 	mutex_exit(d->bd_mtx);
982 
983 	if (d->bd_locked) {
984 		switch (cmd) {
985 		case BIOCGBLEN:		/* FALLTHROUGH */
986 		case BIOCFLUSH:		/* FALLTHROUGH */
987 		case BIOCGDLT:		/* FALLTHROUGH */
988 		case BIOCGDLTLIST:	/* FALLTHROUGH */
989 		case BIOCGETIF:		/* FALLTHROUGH */
990 		case BIOCGRTIMEOUT:	/* FALLTHROUGH */
991 		case BIOCGSTATS:	/* FALLTHROUGH */
992 		case BIOCVERSION:	/* FALLTHROUGH */
993 		case BIOCGHDRCMPLT:	/* FALLTHROUGH */
994 		case FIONREAD:		/* FALLTHROUGH */
995 		case BIOCLOCK:		/* FALLTHROUGH */
996 		case BIOCSRTIMEOUT:	/* FALLTHROUGH */
997 		case BIOCIMMEDIATE:	/* FALLTHROUGH */
998 		case TIOCGPGRP:
999 			break;
1000 		default:
1001 			return EPERM;
1002 		}
1003 	}
1004 
1005 	switch (cmd) {
1006 
1007 	default:
1008 		error = EINVAL;
1009 		break;
1010 
1011 	/*
1012 	 * Check for read packet available.
1013 	 */
1014 	case FIONREAD:
1015 		{
1016 			int n;
1017 
1018 			mutex_enter(d->bd_buf_mtx);
1019 			n = d->bd_slen;
1020 			if (d->bd_hbuf)
1021 				n += d->bd_hlen;
1022 			mutex_exit(d->bd_buf_mtx);
1023 
1024 			*(int *)addr = n;
1025 			break;
1026 		}
1027 
1028 	/*
1029 	 * Get buffer len [for read()].
1030 	 */
1031 	case BIOCGBLEN:
1032 		*(u_int *)addr = d->bd_bufsize;
1033 		break;
1034 
1035 	/*
1036 	 * Set buffer length.
1037 	 */
1038 	case BIOCSBLEN:
1039 		/*
1040 		 * Forbid to change the buffer length if buffers are already
1041 		 * allocated.
1042 		 */
1043 		mutex_enter(d->bd_mtx);
1044 		mutex_enter(d->bd_buf_mtx);
1045 		if (d->bd_bif != NULL || d->bd_sbuf != NULL)
1046 			error = EINVAL;
1047 		else {
1048 			u_int size = *(u_int *)addr;
1049 
1050 			if (size > bpf_maxbufsize)
1051 				*(u_int *)addr = size = bpf_maxbufsize;
1052 			else if (size < BPF_MINBUFSIZE)
1053 				*(u_int *)addr = size = BPF_MINBUFSIZE;
1054 			d->bd_bufsize = size;
1055 		}
1056 		mutex_exit(d->bd_buf_mtx);
1057 		mutex_exit(d->bd_mtx);
1058 		break;
1059 
1060 	/*
1061 	 * Set link layer read filter.
1062 	 */
1063 	case BIOCSETF:		/* FALLTHROUGH */
1064 	case BIOCSETWF:
1065 		error = bpf_setf(d, addr, cmd);
1066 		break;
1067 
1068 	case BIOCLOCK:
1069 		d->bd_locked = 1;
1070 		break;
1071 
1072 	/*
1073 	 * Flush read packet buffer.
1074 	 */
1075 	case BIOCFLUSH:
1076 		mutex_enter(d->bd_mtx);
1077 		reset_d(d);
1078 		mutex_exit(d->bd_mtx);
1079 		break;
1080 
1081 	/*
1082 	 * Put interface into promiscuous mode.
1083 	 */
1084 	case BIOCPROMISC:
1085 		mutex_enter(d->bd_mtx);
1086 		if (d->bd_bif == NULL) {
1087 			mutex_exit(d->bd_mtx);
1088 			/*
1089 			 * No interface attached yet.
1090 			 */
1091 			error = EINVAL;
1092 			break;
1093 		}
1094 		if (d->bd_promisc == 0) {
1095 			KERNEL_LOCK_UNLESS_NET_MPSAFE();
1096 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1097 			KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
1098 			if (error == 0)
1099 				d->bd_promisc = 1;
1100 		}
1101 		mutex_exit(d->bd_mtx);
1102 		break;
1103 
1104 	/*
1105 	 * Get device parameters.
1106 	 */
1107 	case BIOCGDLT:
1108 		mutex_enter(d->bd_mtx);
1109 		if (d->bd_bif == NULL)
1110 			error = EINVAL;
1111 		else
1112 			*(u_int *)addr = d->bd_bif->bif_dlt;
1113 		mutex_exit(d->bd_mtx);
1114 		break;
1115 
1116 	/*
1117 	 * Get a list of supported device parameters.
1118 	 */
1119 	case BIOCGDLTLIST:
1120 		mutex_enter(d->bd_mtx);
1121 		if (d->bd_bif == NULL)
1122 			error = EINVAL;
1123 		else
1124 			error = bpf_getdltlist(d, addr);
1125 		mutex_exit(d->bd_mtx);
1126 		break;
1127 
1128 	/*
1129 	 * Set device parameters.
1130 	 */
1131 	case BIOCSDLT:
1132 		mutex_enter(&bpf_mtx);
1133 		mutex_enter(d->bd_mtx);
1134 		if (d->bd_bif == NULL)
1135 			error = EINVAL;
1136 		else
1137 			error = bpf_setdlt(d, *(u_int *)addr);
1138 		mutex_exit(d->bd_mtx);
1139 		mutex_exit(&bpf_mtx);
1140 		break;
1141 
1142 	/*
1143 	 * Set interface name.
1144 	 */
1145 #ifdef OBIOCGETIF
1146 	case OBIOCGETIF:
1147 #endif
1148 	case BIOCGETIF:
1149 		mutex_enter(d->bd_mtx);
1150 		if (d->bd_bif == NULL)
1151 			error = EINVAL;
1152 		else
1153 			bpf_ifname(d->bd_bif->bif_ifp, addr);
1154 		mutex_exit(d->bd_mtx);
1155 		break;
1156 
1157 	/*
1158 	 * Set interface.
1159 	 */
1160 #ifdef OBIOCSETIF
1161 	case OBIOCSETIF:
1162 #endif
1163 	case BIOCSETIF:
1164 		mutex_enter(&bpf_mtx);
1165 		error = bpf_setif(d, addr);
1166 		mutex_exit(&bpf_mtx);
1167 		break;
1168 
1169 	/*
1170 	 * Set read timeout.
1171 	 */
1172 	case BIOCSRTIMEOUT:
1173 		{
1174 			struct timeval *tv = addr;
1175 
1176 			/* Compute number of ticks. */
1177 			if (tv->tv_sec < 0 ||
1178 			    tv->tv_usec < 0 || tv->tv_usec >= 1000000) {
1179 				error = EINVAL;
1180 				break;
1181 			} else if (tv->tv_sec > INT_MAX/hz - 1) {
1182 				d->bd_rtout = INT_MAX;
1183 			} else {
1184 				d->bd_rtout = tv->tv_sec * hz
1185 				    + tv->tv_usec / tick;
1186 			}
1187 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
1188 				d->bd_rtout = 1;
1189 			break;
1190 		}
1191 
1192 #ifdef BIOCGORTIMEOUT
1193 	/*
1194 	 * Get read timeout.
1195 	 */
1196 	case BIOCGORTIMEOUT:
1197 		{
1198 			struct timeval50 *tv = addr;
1199 
1200 			tv->tv_sec = d->bd_rtout / hz;
1201 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1202 			break;
1203 		}
1204 #endif
1205 
1206 #ifdef BIOCSORTIMEOUT
1207 	/*
1208 	 * Set read timeout.
1209 	 */
1210 	case BIOCSORTIMEOUT:
1211 		{
1212 			struct timeval50 *tv = addr;
1213 
1214 			/* Compute number of ticks. */
1215 			if (tv->tv_sec < 0 ||
1216 			    tv->tv_usec < 0 || tv->tv_usec >= 1000000) {
1217 				error = EINVAL;
1218 				break;
1219 			} else if (tv->tv_sec > INT_MAX/hz - 1) {
1220 				d->bd_rtout = INT_MAX;
1221 			} else {
1222 				d->bd_rtout = tv->tv_sec * hz
1223 				    + tv->tv_usec / tick;
1224 			}
1225 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
1226 				d->bd_rtout = 1;
1227 			break;
1228 		}
1229 #endif
1230 
1231 	/*
1232 	 * Get read timeout.
1233 	 */
1234 	case BIOCGRTIMEOUT:
1235 		{
1236 			struct timeval *tv = addr;
1237 
1238 			tv->tv_sec = d->bd_rtout / hz;
1239 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1240 			break;
1241 		}
1242 	/*
1243 	 * Get packet stats.
1244 	 */
1245 	case BIOCGSTATS:
1246 		{
1247 			struct bpf_stat *bs = addr;
1248 
1249 			bs->bs_recv = d->bd_rcount;
1250 			bs->bs_drop = d->bd_dcount;
1251 			bs->bs_capt = d->bd_ccount;
1252 			break;
1253 		}
1254 
1255 	case BIOCGSTATSOLD:
1256 		{
1257 			struct bpf_stat_old *bs = addr;
1258 
1259 			bs->bs_recv = d->bd_rcount;
1260 			bs->bs_drop = d->bd_dcount;
1261 			break;
1262 		}
1263 
1264 	/*
1265 	 * Set immediate mode.
1266 	 */
1267 	case BIOCIMMEDIATE:
1268 		d->bd_immediate = *(u_int *)addr;
1269 		break;
1270 
1271 	case BIOCVERSION:
1272 		{
1273 			struct bpf_version *bv = addr;
1274 
1275 			bv->bv_major = BPF_MAJOR_VERSION;
1276 			bv->bv_minor = BPF_MINOR_VERSION;
1277 			break;
1278 		}
1279 
1280 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
1281 		*(u_int *)addr = d->bd_hdrcmplt;
1282 		break;
1283 
1284 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
1285 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1286 		break;
1287 
1288 	/*
1289 	 * Get packet direction flag
1290 	 */
1291 	case BIOCGDIRECTION:
1292 		*(u_int *)addr = d->bd_direction;
1293 		break;
1294 
1295 	/*
1296 	 * Set packet direction flag
1297 	 */
1298 	case BIOCSDIRECTION:
1299 		{
1300 			u_int	direction;
1301 
1302 			direction = *(u_int *)addr;
1303 			switch (direction) {
1304 			case BPF_D_IN:
1305 			case BPF_D_INOUT:
1306 			case BPF_D_OUT:
1307 				d->bd_direction = direction;
1308 				break;
1309 			default:
1310 				error = EINVAL;
1311 			}
1312 		}
1313 		break;
1314 
1315 	/*
1316 	 * Set "feed packets from bpf back to input" mode
1317 	 */
1318 	case BIOCSFEEDBACK:
1319 		d->bd_feedback = *(u_int *)addr;
1320 		break;
1321 
1322 	/*
1323 	 * Get "feed packets from bpf back to input" mode
1324 	 */
1325 	case BIOCGFEEDBACK:
1326 		*(u_int *)addr = d->bd_feedback;
1327 		break;
1328 
1329 	case FIONBIO:		/* Non-blocking I/O */
1330 		/*
1331 		 * No need to do anything special as we use IO_NDELAY in
1332 		 * bpfread() as an indication of whether or not to block
1333 		 * the read.
1334 		 */
1335 		break;
1336 
1337 	case FIOASYNC:		/* Send signal on receive packets */
1338 		mutex_enter(d->bd_mtx);
1339 		d->bd_async = *(int *)addr;
1340 		mutex_exit(d->bd_mtx);
1341 		break;
1342 
1343 	case TIOCSPGRP:		/* Process or group to send signals to */
1344 	case FIOSETOWN:
1345 		error = fsetown(&d->bd_pgid, cmd, addr);
1346 		break;
1347 
1348 	case TIOCGPGRP:
1349 	case FIOGETOWN:
1350 		error = fgetown(d->bd_pgid, cmd, addr);
1351 		break;
1352 	}
1353 	return (error);
1354 }
1355 
1356 /*
1357  * Set d's packet filter program to fp.  If this file already has a filter,
1358  * free it and replace it.  Returns EINVAL for bogus requests.
1359  */
1360 static int
1361 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1362 {
1363 	struct bpf_insn *fcode;
1364 	bpfjit_func_t jcode;
1365 	size_t flen, size = 0;
1366 	struct bpf_filter *oldf, *newf, **storef;
1367 
1368 	jcode = NULL;
1369 	flen = fp->bf_len;
1370 
1371 	if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) {
1372 		return EINVAL;
1373 	}
1374 
1375 	if (flen) {
1376 		/*
1377 		 * Allocate the buffer, copy the byte-code from
1378 		 * userspace and validate it.
1379 		 */
1380 		size = flen * sizeof(*fp->bf_insns);
1381 		fcode = kmem_alloc(size, KM_SLEEP);
1382 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1383 		    !bpf_validate(fcode, (int)flen)) {
1384 			kmem_free(fcode, size);
1385 			return EINVAL;
1386 		}
1387 		if (bpf_jit)
1388 			jcode = bpf_jit_generate(NULL, fcode, flen);
1389 	} else {
1390 		fcode = NULL;
1391 	}
1392 
1393 	newf = kmem_alloc(sizeof(*newf), KM_SLEEP);
1394 	newf->bf_insn = fcode;
1395 	newf->bf_size = size;
1396 	newf->bf_jitcode = jcode;
1397 	if (cmd == BIOCSETF)
1398 		d->bd_jitcode = jcode; /* XXX just for kvm(3) users */
1399 
1400 	/* Need to hold bpf_mtx for pserialize_perform */
1401 	mutex_enter(&bpf_mtx);
1402 	mutex_enter(d->bd_mtx);
1403 	if (cmd == BIOCSETWF) {
1404 		oldf = d->bd_wfilter;
1405 		storef = &d->bd_wfilter;
1406 	} else {
1407 		oldf = d->bd_rfilter;
1408 		storef = &d->bd_rfilter;
1409 	}
1410 	atomic_store_release(storef, newf);
1411 	reset_d(d);
1412 	pserialize_perform(bpf_psz);
1413 	mutex_exit(d->bd_mtx);
1414 	mutex_exit(&bpf_mtx);
1415 
1416 	if (oldf != NULL)
1417 		bpf_free_filter(oldf);
1418 
1419 	return 0;
1420 }
1421 
1422 /*
1423  * Detach a file from its current interface (if attached at all) and attach
1424  * to the interface indicated by the name stored in ifr.
1425  * Return an errno or 0.
1426  */
1427 static int
1428 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1429 {
1430 	struct bpf_if *bp;
1431 	char *cp;
1432 	int unit_seen, i, error;
1433 
1434 	KASSERT(mutex_owned(&bpf_mtx));
1435 	/*
1436 	 * Make sure the provided name has a unit number, and default
1437 	 * it to '0' if not specified.
1438 	 * XXX This is ugly ... do this differently?
1439 	 */
1440 	unit_seen = 0;
1441 	cp = ifr->ifr_name;
1442 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1443 	while (*cp++)
1444 		if (*cp >= '0' && *cp <= '9')
1445 			unit_seen = 1;
1446 	if (!unit_seen) {
1447 		/* Make sure to leave room for the '\0'. */
1448 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1449 			if ((ifr->ifr_name[i] >= 'a' &&
1450 			     ifr->ifr_name[i] <= 'z') ||
1451 			    (ifr->ifr_name[i] >= 'A' &&
1452 			     ifr->ifr_name[i] <= 'Z'))
1453 				continue;
1454 			ifr->ifr_name[i] = '0';
1455 		}
1456 	}
1457 
1458 	/*
1459 	 * Look through attached interfaces for the named one.
1460 	 */
1461 	BPF_IFLIST_WRITER_FOREACH(bp) {
1462 		struct ifnet *ifp = bp->bif_ifp;
1463 
1464 		if (ifp == NULL ||
1465 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1466 			continue;
1467 		/* skip additional entry */
1468 		if (bp->bif_driverp != &ifp->if_bpf)
1469 			continue;
1470 		/*
1471 		 * We found the requested interface.
1472 		 * Allocate the packet buffers if we need to.
1473 		 * If we're already attached to requested interface,
1474 		 * just flush the buffer.
1475 		 */
1476 		/*
1477 		 * bpf_allocbufs is called only here. bpf_mtx ensures that
1478 		 * no race condition happen on d->bd_sbuf.
1479 		 */
1480 		if (d->bd_sbuf == NULL) {
1481 			error = bpf_allocbufs(d);
1482 			if (error != 0)
1483 				return (error);
1484 		}
1485 		mutex_enter(d->bd_mtx);
1486 		if (bp != d->bd_bif) {
1487 			if (d->bd_bif) {
1488 				/*
1489 				 * Detach if attached to something else.
1490 				 */
1491 				bpf_detachd(d);
1492 				BPFIF_DLIST_ENTRY_INIT(d);
1493 			}
1494 
1495 			bpf_attachd(d, bp);
1496 		}
1497 		reset_d(d);
1498 		mutex_exit(d->bd_mtx);
1499 		return (0);
1500 	}
1501 	/* Not found. */
1502 	return (ENXIO);
1503 }
1504 
1505 /*
1506  * Copy the interface name to the ifreq.
1507  */
1508 static void
1509 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1510 {
1511 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1512 }
1513 
1514 static int
1515 bpf_stat(struct file *fp, struct stat *st)
1516 {
1517 	struct bpf_d *d = fp->f_bpf;
1518 
1519 	(void)memset(st, 0, sizeof(*st));
1520 	mutex_enter(d->bd_mtx);
1521 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1522 	st->st_atimespec = d->bd_atime;
1523 	st->st_mtimespec = d->bd_mtime;
1524 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1525 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1526 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1527 	st->st_mode = S_IFCHR;
1528 	mutex_exit(d->bd_mtx);
1529 	return 0;
1530 }
1531 
1532 /*
1533  * Support for poll() system call
1534  *
1535  * Return true iff the specific operation will not block indefinitely - with
1536  * the assumption that it is safe to positively acknowledge a request for the
1537  * ability to write to the BPF device.
1538  * Otherwise, return false but make a note that a selnotify() must be done.
1539  */
1540 static int
1541 bpf_poll(struct file *fp, int events)
1542 {
1543 	struct bpf_d *d = fp->f_bpf;
1544 	int revents;
1545 
1546 	/*
1547 	 * Refresh the PID associated with this bpf file.
1548 	 */
1549 	mutex_enter(&bpf_mtx);
1550 	d->bd_pid = curproc->p_pid;
1551 
1552 	revents = events & (POLLOUT | POLLWRNORM);
1553 	if (events & (POLLIN | POLLRDNORM)) {
1554 		/*
1555 		 * An imitation of the FIONREAD ioctl code.
1556 		 */
1557 		mutex_enter(d->bd_mtx);
1558 		if (d->bd_hlen != 0 ||
1559 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1560 		     d->bd_slen != 0)) {
1561 			revents |= events & (POLLIN | POLLRDNORM);
1562 		} else {
1563 			selrecord(curlwp, &d->bd_sel);
1564 			/* Start the read timeout if necessary */
1565 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1566 				callout_reset(&d->bd_callout, d->bd_rtout,
1567 					      bpf_timed_out, d);
1568 				d->bd_state = BPF_WAITING;
1569 			}
1570 		}
1571 		mutex_exit(d->bd_mtx);
1572 	}
1573 
1574 	mutex_exit(&bpf_mtx);
1575 	return (revents);
1576 }
1577 
1578 static void
1579 filt_bpfrdetach(struct knote *kn)
1580 {
1581 	struct bpf_d *d = kn->kn_hook;
1582 
1583 	mutex_enter(d->bd_buf_mtx);
1584 	selremove_knote(&d->bd_sel, kn);
1585 	mutex_exit(d->bd_buf_mtx);
1586 }
1587 
1588 static int
1589 filt_bpfread(struct knote *kn, long hint)
1590 {
1591 	struct bpf_d *d = kn->kn_hook;
1592 	int rv;
1593 
1594 	/*
1595 	 * Refresh the PID associated with this bpf file.
1596 	 */
1597 	d->bd_pid = curproc->p_pid;
1598 
1599 	mutex_enter(d->bd_buf_mtx);
1600 	kn->kn_data = d->bd_hlen;
1601 	if (d->bd_immediate)
1602 		kn->kn_data += d->bd_slen;
1603 	rv = (kn->kn_data > 0);
1604 	mutex_exit(d->bd_buf_mtx);
1605 	return rv;
1606 }
1607 
1608 static const struct filterops bpfread_filtops = {
1609 	.f_flags = FILTEROP_ISFD,
1610 	.f_attach = NULL,
1611 	.f_detach = filt_bpfrdetach,
1612 	.f_event = filt_bpfread,
1613 };
1614 
1615 static int
1616 bpf_kqfilter(struct file *fp, struct knote *kn)
1617 {
1618 	struct bpf_d *d = fp->f_bpf;
1619 
1620 	switch (kn->kn_filter) {
1621 	case EVFILT_READ:
1622 		kn->kn_fop = &bpfread_filtops;
1623 		break;
1624 
1625 	default:
1626 		return (EINVAL);
1627 	}
1628 
1629 	kn->kn_hook = d;
1630 
1631 	mutex_enter(d->bd_buf_mtx);
1632 	selrecord_knote(&d->bd_sel, kn);
1633 	mutex_exit(d->bd_buf_mtx);
1634 
1635 	return (0);
1636 }
1637 
1638 /*
1639  * Copy data from an mbuf chain into a buffer.  This code is derived
1640  * from m_copydata in sys/uipc_mbuf.c.
1641  */
1642 static void *
1643 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1644 {
1645 	const struct mbuf *m;
1646 	u_int count;
1647 	u_char *dst;
1648 
1649 	m = src_arg;
1650 	dst = dst_arg;
1651 	while (len > 0) {
1652 		if (m == NULL)
1653 			panic("bpf_mcpy");
1654 		count = uimin(m->m_len, len);
1655 		memcpy(dst, mtod(m, const void *), count);
1656 		m = m->m_next;
1657 		dst += count;
1658 		len -= count;
1659 	}
1660 	return dst_arg;
1661 }
1662 
1663 static inline u_int
1664 bpf_xfilter(struct bpf_filter **filter, void *pkt, u_int pktlen, u_int buflen)
1665 {
1666 	struct bpf_filter *filt;
1667 	uint32_t mem[BPF_MEMWORDS];
1668 	bpf_args_t args = {
1669 		.pkt = (const uint8_t *)pkt,
1670 		.wirelen = pktlen,
1671 		.buflen = buflen,
1672 		.mem = mem,
1673 		.arg = NULL
1674 	};
1675 	u_int slen;
1676 
1677 	filt = atomic_load_consume(filter);
1678 	if (filt == NULL) /* No filter means accept all. */
1679 		return (u_int)-1;
1680 
1681 	if (filt->bf_jitcode != NULL)
1682 		slen = filt->bf_jitcode(NULL, &args);
1683 	else
1684 		slen = bpf_filter_ext(NULL, filt->bf_insn, &args);
1685 	return slen;
1686 }
1687 
1688 /*
1689  * Dispatch a packet to all the listeners on interface bp.
1690  *
1691  * pkt       pointer to the packet, either a data buffer or an mbuf chain
1692  * buflen    buffer length, if pkt is a data buffer
1693  * cpfn      a function that can copy pkt into the listener's buffer
1694  * pktlen    length of the packet
1695  * direction BPF_D_IN or BPF_D_OUT
1696  */
1697 static inline void
1698 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1699     void *pkt, u_int pktlen, u_int buflen, const u_int direction)
1700 {
1701 	bool gottime = false;
1702 	struct timespec ts;
1703 	struct bpf_d *d;
1704 	int s;
1705 	u_int slen;
1706 
1707 	KASSERT(!cpu_intr_p());
1708 
1709 	/*
1710 	 * Note that the IPL does not have to be raised at this point.
1711 	 * The only problem that could arise here is that if two different
1712 	 * interfaces shared any data.  This is not the case.
1713 	 */
1714 	s = pserialize_read_enter();
1715 	BPFIF_DLIST_READER_FOREACH(d, bp) {
1716 		if (direction == BPF_D_IN) {
1717 			if (d->bd_direction == BPF_D_OUT)
1718 				continue;
1719 		} else { /* BPF_D_OUT */
1720 			if (d->bd_direction == BPF_D_IN)
1721 				continue;
1722 		}
1723 
1724 		atomic_inc_ulong(&d->bd_rcount);
1725 		BPF_STATINC(recv);
1726 
1727 		slen = bpf_xfilter(&d->bd_rfilter, pkt, pktlen, buflen);
1728 		if (slen == 0)
1729 			continue;
1730 
1731 		if (!gottime) {
1732 			gottime = true;
1733 			nanotime(&ts);
1734 		}
1735 		/* Assume catchpacket doesn't sleep */
1736 		catchpacket(d, pkt, pktlen, slen, cpfn, &ts);
1737 	}
1738 	pserialize_read_exit(s);
1739 }
1740 
1741 /*
1742  * Incoming linkage from device drivers, when the head of the packet is in
1743  * a buffer, and the tail is in an mbuf chain.
1744  */
1745 static void
1746 _bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m,
1747 	u_int direction)
1748 {
1749 	u_int pktlen;
1750 	struct mbuf mb;
1751 
1752 	/* Skip outgoing duplicate packets. */
1753 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) {
1754 		m->m_flags &= ~M_PROMISC;
1755 		return;
1756 	}
1757 
1758 	pktlen = m_length(m) + dlen;
1759 
1760 	/*
1761 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1762 	 * Note that we cut corners here; we only setup what's
1763 	 * absolutely needed--this mbuf should never go anywhere else.
1764 	 */
1765 	(void)memset(&mb, 0, sizeof(mb));
1766 	mb.m_type = MT_DATA;
1767 	mb.m_next = m;
1768 	mb.m_data = data;
1769 	mb.m_len = dlen;
1770 
1771 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, direction);
1772 }
1773 
1774 /*
1775  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1776  */
1777 static void
1778 _bpf_mtap(struct bpf_if *bp, struct mbuf *m, u_int direction)
1779 {
1780 	void *(*cpfn)(void *, const void *, size_t);
1781 	u_int pktlen, buflen;
1782 	void *marg;
1783 
1784 	/* Skip outgoing duplicate packets. */
1785 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) {
1786 		m->m_flags &= ~M_PROMISC;
1787 		return;
1788 	}
1789 
1790 	pktlen = m_length(m);
1791 
1792 	/* Skip zero-sized packets. */
1793 	if (__predict_false(pktlen == 0)) {
1794 		return;
1795 	}
1796 
1797 	if (pktlen == m->m_len) {
1798 		cpfn = (void *)memcpy;
1799 		marg = mtod(m, void *);
1800 		buflen = pktlen;
1801 		KASSERT(buflen != 0);
1802 	} else {
1803 		cpfn = bpf_mcpy;
1804 		marg = m;
1805 		buflen = 0;
1806 	}
1807 
1808 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, direction);
1809 }
1810 
1811 /*
1812  * We need to prepend the address family as
1813  * a four byte field.  Cons up a dummy header
1814  * to pacify bpf.  This is safe because bpf
1815  * will only read from the mbuf (i.e., it won't
1816  * try to free it or keep a pointer a to it).
1817  */
1818 static void
1819 _bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m, u_int direction)
1820 {
1821 	struct mbuf m0;
1822 
1823 	m0.m_type = MT_DATA;
1824 	m0.m_flags = 0;
1825 	m0.m_next = m;
1826 	m0.m_nextpkt = NULL;
1827 	m0.m_owner = NULL;
1828 	m0.m_len = 4;
1829 	m0.m_data = (char *)&af;
1830 
1831 	_bpf_mtap(bp, &m0, direction);
1832 }
1833 
1834 /*
1835  * Put the SLIP pseudo-"link header" in place.
1836  * Note this M_PREPEND() should never fail,
1837  * swince we know we always have enough space
1838  * in the input buffer.
1839  */
1840 static void
1841 _bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m)
1842 {
1843 	u_char *hp;
1844 
1845 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1846 	if (*m == NULL)
1847 		return;
1848 
1849 	hp = mtod(*m, u_char *);
1850 	hp[SLX_DIR] = SLIPDIR_IN;
1851 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1852 
1853 	_bpf_mtap(bp, *m, BPF_D_IN);
1854 
1855 	m_adj(*m, SLIP_HDRLEN);
1856 }
1857 
1858 /*
1859  * Put the SLIP pseudo-"link header" in
1860  * place.  The compressed header is now
1861  * at the beginning of the mbuf.
1862  */
1863 static void
1864 _bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m)
1865 {
1866 	struct mbuf m0;
1867 	u_char *hp;
1868 
1869 	m0.m_type = MT_DATA;
1870 	m0.m_flags = 0;
1871 	m0.m_next = m;
1872 	m0.m_nextpkt = NULL;
1873 	m0.m_owner = NULL;
1874 	m0.m_data = m0.m_dat;
1875 	m0.m_len = SLIP_HDRLEN;
1876 
1877 	hp = mtod(&m0, u_char *);
1878 
1879 	hp[SLX_DIR] = SLIPDIR_OUT;
1880 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1881 
1882 	_bpf_mtap(bp, &m0, BPF_D_OUT);
1883 	m_freem(m);
1884 }
1885 
1886 static struct mbuf *
1887 bpf_mbuf_enqueue(struct bpf_if *bp, struct mbuf *m)
1888 {
1889 	struct mbuf *dup;
1890 
1891 	dup = m_dup(m, 0, M_COPYALL, M_NOWAIT);
1892 	if (dup == NULL)
1893 		return NULL;
1894 
1895 	if (bp->bif_mbuf_tail != NULL) {
1896 		bp->bif_mbuf_tail->m_nextpkt = dup;
1897 	} else {
1898 		bp->bif_mbuf_head = dup;
1899 	}
1900 	bp->bif_mbuf_tail = dup;
1901 #ifdef BPF_MTAP_SOFTINT_DEBUG
1902 	log(LOG_DEBUG, "%s: enqueued mbuf=%p to %s\n",
1903 	    __func__, dup, bp->bif_ifp->if_xname);
1904 #endif
1905 
1906 	return dup;
1907 }
1908 
1909 static struct mbuf *
1910 bpf_mbuf_dequeue(struct bpf_if *bp)
1911 {
1912 	struct mbuf *m;
1913 	int s;
1914 
1915 	/* XXX NOMPSAFE: assumed running on one CPU */
1916 	s = splnet();
1917 	m = bp->bif_mbuf_head;
1918 	if (m != NULL) {
1919 		bp->bif_mbuf_head = m->m_nextpkt;
1920 		m->m_nextpkt = NULL;
1921 
1922 		if (bp->bif_mbuf_head == NULL)
1923 			bp->bif_mbuf_tail = NULL;
1924 #ifdef BPF_MTAP_SOFTINT_DEBUG
1925 		log(LOG_DEBUG, "%s: dequeued mbuf=%p from %s\n",
1926 		    __func__, m, bp->bif_ifp->if_xname);
1927 #endif
1928 	}
1929 	splx(s);
1930 
1931 	return m;
1932 }
1933 
1934 static void
1935 bpf_mtap_si(void *arg)
1936 {
1937 	struct bpf_if *bp = arg;
1938 	struct mbuf *m;
1939 
1940 	while ((m = bpf_mbuf_dequeue(bp)) != NULL) {
1941 #ifdef BPF_MTAP_SOFTINT_DEBUG
1942 		log(LOG_DEBUG, "%s: tapping mbuf=%p on %s\n",
1943 		    __func__, m, bp->bif_ifp->if_xname);
1944 #endif
1945 		bpf_ops->bpf_mtap(bp, m, BPF_D_IN);
1946 		m_freem(m);
1947 	}
1948 }
1949 
1950 static void
1951 _bpf_mtap_softint(struct ifnet *ifp, struct mbuf *m)
1952 {
1953 	struct bpf_if *bp = ifp->if_bpf;
1954 	struct mbuf *dup;
1955 
1956 	KASSERT(cpu_intr_p());
1957 
1958 	/* To avoid extra invocations of the softint */
1959 	if (BPFIF_DLIST_READER_EMPTY(bp))
1960 		return;
1961 	KASSERT(bp->bif_si != NULL);
1962 
1963 	dup = bpf_mbuf_enqueue(bp, m);
1964 	if (dup != NULL)
1965 		softint_schedule(bp->bif_si);
1966 }
1967 
1968 static int
1969 bpf_hdrlen(struct bpf_d *d)
1970 {
1971 	int hdrlen = d->bd_bif->bif_hdrlen;
1972 	/*
1973 	 * Compute the length of the bpf header.  This is not necessarily
1974 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1975 	 * that the network layer header begins on a longword boundary (for
1976 	 * performance reasons and to alleviate alignment restrictions).
1977 	 */
1978 #ifdef _LP64
1979 	if (d->bd_compat32)
1980 		return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen);
1981 	else
1982 #endif
1983 		return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen);
1984 }
1985 
1986 /*
1987  * Move the packet data from interface memory (pkt) into the
1988  * store buffer. Call the wakeup functions if it's time to wakeup
1989  * a listener (buffer full), "cpfn" is the routine called to do the
1990  * actual data transfer. memcpy is passed in to copy contiguous chunks,
1991  * while bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1992  * pkt is really an mbuf.
1993  */
1994 static void
1995 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1996     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1997 {
1998 	char *h;
1999 	int totlen, curlen, caplen;
2000 	int hdrlen = bpf_hdrlen(d);
2001 	int do_wakeup = 0;
2002 
2003 	atomic_inc_ulong(&d->bd_ccount);
2004 	BPF_STATINC(capt);
2005 	/*
2006 	 * Figure out how many bytes to move.  If the packet is
2007 	 * greater or equal to the snapshot length, transfer that
2008 	 * much.  Otherwise, transfer the whole packet (unless
2009 	 * we hit the buffer size limit).
2010 	 */
2011 	totlen = hdrlen + uimin(snaplen, pktlen);
2012 	if (totlen > d->bd_bufsize)
2013 		totlen = d->bd_bufsize;
2014 	/*
2015 	 * If we adjusted totlen to fit the bufsize, it could be that
2016 	 * totlen is smaller than hdrlen because of the link layer header.
2017 	 */
2018 	caplen = totlen - hdrlen;
2019 	if (caplen < 0)
2020 		caplen = 0;
2021 
2022 	mutex_enter(d->bd_buf_mtx);
2023 	/*
2024 	 * Round up the end of the previous packet to the next longword.
2025 	 */
2026 #ifdef _LP64
2027 	if (d->bd_compat32)
2028 		curlen = BPF_WORDALIGN32(d->bd_slen);
2029 	else
2030 #endif
2031 		curlen = BPF_WORDALIGN(d->bd_slen);
2032 	if (curlen + totlen > d->bd_bufsize) {
2033 		/*
2034 		 * This packet will overflow the storage buffer.
2035 		 * Rotate the buffers if we can, then wakeup any
2036 		 * pending reads.
2037 		 */
2038 		if (d->bd_fbuf == NULL) {
2039 			mutex_exit(d->bd_buf_mtx);
2040 			/*
2041 			 * We haven't completed the previous read yet,
2042 			 * so drop the packet.
2043 			 */
2044 			atomic_inc_ulong(&d->bd_dcount);
2045 			BPF_STATINC(drop);
2046 			return;
2047 		}
2048 		ROTATE_BUFFERS(d);
2049 		do_wakeup = 1;
2050 		curlen = 0;
2051 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
2052 		/*
2053 		 * Immediate mode is set, or the read timeout has
2054 		 * already expired during a select call.  A packet
2055 		 * arrived, so the reader should be woken up.
2056 		 */
2057 		do_wakeup = 1;
2058 	}
2059 
2060 	/*
2061 	 * Append the bpf header.
2062 	 */
2063 	h = (char *)d->bd_sbuf + curlen;
2064 #ifdef _LP64
2065 	if (d->bd_compat32) {
2066 		struct bpf_hdr32 *hp32;
2067 
2068 		hp32 = (struct bpf_hdr32 *)h;
2069 		hp32->bh_tstamp.tv_sec = ts->tv_sec;
2070 		hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
2071 		hp32->bh_datalen = pktlen;
2072 		hp32->bh_hdrlen = hdrlen;
2073 		hp32->bh_caplen = caplen;
2074 	} else
2075 #endif
2076 	{
2077 		struct bpf_hdr *hp;
2078 
2079 		hp = (struct bpf_hdr *)h;
2080 		hp->bh_tstamp.tv_sec = ts->tv_sec;
2081 		hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
2082 		hp->bh_datalen = pktlen;
2083 		hp->bh_hdrlen = hdrlen;
2084 		hp->bh_caplen = caplen;
2085 	}
2086 
2087 	/*
2088 	 * Copy the packet data into the store buffer and update its length.
2089 	 */
2090 	(*cpfn)(h + hdrlen, pkt, caplen);
2091 	d->bd_slen = curlen + totlen;
2092 	mutex_exit(d->bd_buf_mtx);
2093 
2094 	/*
2095 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
2096 	 * will cause filt_bpfread() to be called with it adjusted.
2097 	 */
2098 	if (do_wakeup)
2099 		bpf_wakeup(d);
2100 }
2101 
2102 /*
2103  * Initialize all nonzero fields of a descriptor.
2104  */
2105 static int
2106 bpf_allocbufs(struct bpf_d *d)
2107 {
2108 
2109 	d->bd_fbuf = kmem_zalloc(d->bd_bufsize, KM_NOSLEEP);
2110 	if (!d->bd_fbuf)
2111 		return (ENOBUFS);
2112 	d->bd_sbuf = kmem_zalloc(d->bd_bufsize, KM_NOSLEEP);
2113 	if (!d->bd_sbuf) {
2114 		kmem_free(d->bd_fbuf, d->bd_bufsize);
2115 		return (ENOBUFS);
2116 	}
2117 	d->bd_slen = 0;
2118 	d->bd_hlen = 0;
2119 	return (0);
2120 }
2121 
2122 static void
2123 bpf_free_filter(struct bpf_filter *filter)
2124 {
2125 
2126 	KASSERT(filter != NULL);
2127 
2128 	if (filter->bf_insn != NULL)
2129 		kmem_free(filter->bf_insn, filter->bf_size);
2130 	if (filter->bf_jitcode != NULL)
2131 		bpf_jit_freecode(filter->bf_jitcode);
2132 	kmem_free(filter, sizeof(*filter));
2133 }
2134 
2135 /*
2136  * Free buffers currently in use by a descriptor.
2137  * Called on close.
2138  */
2139 static void
2140 bpf_freed(struct bpf_d *d)
2141 {
2142 	/*
2143 	 * We don't need to lock out interrupts since this descriptor has
2144 	 * been detached from its interface and it yet hasn't been marked
2145 	 * free.
2146 	 */
2147 	if (d->bd_sbuf != NULL) {
2148 		kmem_free(d->bd_sbuf, d->bd_bufsize);
2149 		if (d->bd_hbuf != NULL)
2150 			kmem_free(d->bd_hbuf, d->bd_bufsize);
2151 		if (d->bd_fbuf != NULL)
2152 			kmem_free(d->bd_fbuf, d->bd_bufsize);
2153 	}
2154 	if (d->bd_rfilter != NULL) {
2155 		bpf_free_filter(d->bd_rfilter);
2156 		d->bd_rfilter = NULL;
2157 	}
2158 	if (d->bd_wfilter != NULL) {
2159 		bpf_free_filter(d->bd_wfilter);
2160 		d->bd_wfilter = NULL;
2161 	}
2162 	d->bd_jitcode = NULL;
2163 }
2164 
2165 /*
2166  * Attach an interface to bpf.  dlt is the link layer type;
2167  * hdrlen is the fixed size of the link header for the specified dlt
2168  * (variable length headers not yet supported).
2169  */
2170 static void
2171 _bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2172 {
2173 	struct bpf_if *bp;
2174 
2175 	bp = kmem_alloc(sizeof(*bp), KM_SLEEP);
2176 
2177 	mutex_enter(&bpf_mtx);
2178 	bp->bif_driverp = driverp;
2179 	bp->bif_ifp = ifp;
2180 	bp->bif_dlt = dlt;
2181 	bp->bif_si = NULL;
2182 	BPF_IFLIST_ENTRY_INIT(bp);
2183 	PSLIST_INIT(&bp->bif_dlist_head);
2184 	psref_target_init(&bp->bif_psref, bpf_psref_class);
2185 	SLIST_INIT(&bp->bif_trackers);
2186 
2187 	BPF_IFLIST_WRITER_INSERT_HEAD(bp);
2188 
2189 	*bp->bif_driverp = NULL;
2190 
2191 	bp->bif_hdrlen = hdrlen;
2192 	mutex_exit(&bpf_mtx);
2193 #if 0
2194 	printf("bpf: %s attached with dlt %x\n", ifp->if_xname, dlt);
2195 #endif
2196 }
2197 
2198 static void
2199 _bpf_mtap_softint_init(struct ifnet *ifp)
2200 {
2201 	struct bpf_if *bp;
2202 
2203 	mutex_enter(&bpf_mtx);
2204 	BPF_IFLIST_WRITER_FOREACH(bp) {
2205 		if (bp->bif_ifp != ifp)
2206 			continue;
2207 
2208 		bp->bif_mbuf_head = NULL;
2209 		bp->bif_mbuf_tail = NULL;
2210 		bp->bif_si = softint_establish(SOFTINT_NET, bpf_mtap_si, bp);
2211 		if (bp->bif_si == NULL)
2212 			panic("%s: softint_establish() failed", __func__);
2213 		break;
2214 	}
2215 	mutex_exit(&bpf_mtx);
2216 
2217 	if (bp == NULL)
2218 		panic("%s: no bpf_if found for %s", __func__, ifp->if_xname);
2219 }
2220 
2221 /*
2222  * Remove an interface from bpf.
2223  */
2224 static void
2225 _bpfdetach(struct ifnet *ifp)
2226 {
2227 	struct bpf_if *bp;
2228 	struct bpf_d *d;
2229 	int s;
2230 
2231 	mutex_enter(&bpf_mtx);
2232 	/* Nuke the vnodes for any open instances */
2233   again_d:
2234 	BPF_DLIST_WRITER_FOREACH(d) {
2235 		mutex_enter(d->bd_mtx);
2236 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
2237 			/*
2238 			 * Detach the descriptor from an interface now.
2239 			 * It will be free'ed later by close routine.
2240 			 */
2241 			bpf_detachd(d);
2242 			mutex_exit(d->bd_mtx);
2243 			goto again_d;
2244 		}
2245 		mutex_exit(d->bd_mtx);
2246 	}
2247 
2248   again:
2249 	BPF_IFLIST_WRITER_FOREACH(bp) {
2250 		if (bp->bif_ifp == ifp) {
2251 			BPF_IFLIST_WRITER_REMOVE(bp);
2252 
2253 			pserialize_perform(bpf_psz);
2254 			psref_target_destroy(&bp->bif_psref, bpf_psref_class);
2255 
2256 			while (!SLIST_EMPTY(&bp->bif_trackers)) {
2257 				struct bpf_event_tracker *t =
2258 				    SLIST_FIRST(&bp->bif_trackers);
2259 				SLIST_REMOVE_HEAD(&bp->bif_trackers,
2260 				    bet_entries);
2261 				kmem_free(t, sizeof(*t));
2262 			}
2263 
2264 			BPF_IFLIST_ENTRY_DESTROY(bp);
2265 			if (bp->bif_si != NULL) {
2266 				/* XXX NOMPSAFE: assumed running on one CPU */
2267 				s = splnet();
2268 				while (bp->bif_mbuf_head != NULL) {
2269 					struct mbuf *m = bp->bif_mbuf_head;
2270 					bp->bif_mbuf_head = m->m_nextpkt;
2271 					m_freem(m);
2272 				}
2273 				splx(s);
2274 				softint_disestablish(bp->bif_si);
2275 			}
2276 			kmem_free(bp, sizeof(*bp));
2277 			goto again;
2278 		}
2279 	}
2280 	mutex_exit(&bpf_mtx);
2281 }
2282 
2283 /*
2284  * Change the data link type of a interface.
2285  */
2286 static void
2287 _bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2288 {
2289 	struct bpf_if *bp;
2290 
2291 	mutex_enter(&bpf_mtx);
2292 	BPF_IFLIST_WRITER_FOREACH(bp) {
2293 		if (bp->bif_driverp == &ifp->if_bpf)
2294 			break;
2295 	}
2296 	if (bp == NULL)
2297 		panic("bpf_change_type");
2298 
2299 	bp->bif_dlt = dlt;
2300 
2301 	bp->bif_hdrlen = hdrlen;
2302 	mutex_exit(&bpf_mtx);
2303 }
2304 
2305 /*
2306  * Get a list of available data link type of the interface.
2307  */
2308 static int
2309 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2310 {
2311 	int n, error;
2312 	struct ifnet *ifp;
2313 	struct bpf_if *bp;
2314 	int s, bound;
2315 
2316 	KASSERT(mutex_owned(d->bd_mtx));
2317 
2318 	ifp = d->bd_bif->bif_ifp;
2319 	n = 0;
2320 	error = 0;
2321 
2322 	bound = curlwp_bind();
2323 	s = pserialize_read_enter();
2324 	BPF_IFLIST_READER_FOREACH(bp) {
2325 		if (bp->bif_ifp != ifp)
2326 			continue;
2327 		if (bfl->bfl_list != NULL) {
2328 			struct psref psref;
2329 
2330 			if (n >= bfl->bfl_len) {
2331 				pserialize_read_exit(s);
2332 				return ENOMEM;
2333 			}
2334 
2335 			bpf_if_acquire(bp, &psref);
2336 			pserialize_read_exit(s);
2337 
2338 			error = copyout(&bp->bif_dlt,
2339 			    bfl->bfl_list + n, sizeof(u_int));
2340 
2341 			s = pserialize_read_enter();
2342 			bpf_if_release(bp, &psref);
2343 		}
2344 		n++;
2345 	}
2346 	pserialize_read_exit(s);
2347 	curlwp_bindx(bound);
2348 
2349 	bfl->bfl_len = n;
2350 	return error;
2351 }
2352 
2353 /*
2354  * Set the data link type of a BPF instance.
2355  */
2356 static int
2357 bpf_setdlt(struct bpf_d *d, u_int dlt)
2358 {
2359 	int error, opromisc;
2360 	struct ifnet *ifp;
2361 	struct bpf_if *bp;
2362 
2363 	KASSERT(mutex_owned(&bpf_mtx));
2364 	KASSERT(mutex_owned(d->bd_mtx));
2365 
2366 	if (d->bd_bif->bif_dlt == dlt)
2367 		return 0;
2368 	ifp = d->bd_bif->bif_ifp;
2369 	BPF_IFLIST_WRITER_FOREACH(bp) {
2370 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2371 			break;
2372 	}
2373 	if (bp == NULL)
2374 		return EINVAL;
2375 	opromisc = d->bd_promisc;
2376 	bpf_detachd(d);
2377 	BPFIF_DLIST_ENTRY_INIT(d);
2378 	bpf_attachd(d, bp);
2379 	reset_d(d);
2380 	if (opromisc) {
2381 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
2382 		error = ifpromisc(bp->bif_ifp, 1);
2383 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2384 		if (error)
2385 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
2386 			    bp->bif_ifp->if_xname, error);
2387 		else
2388 			d->bd_promisc = 1;
2389 	}
2390 	return 0;
2391 }
2392 
2393 static int
2394 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
2395 {
2396 	int newsize, error;
2397 	struct sysctlnode node;
2398 
2399 	node = *rnode;
2400 	node.sysctl_data = &newsize;
2401 	newsize = bpf_maxbufsize;
2402 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2403 	if (error || newp == NULL)
2404 		return (error);
2405 
2406 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
2407 		return (EINVAL);
2408 
2409 	bpf_maxbufsize = newsize;
2410 
2411 	return (0);
2412 }
2413 
2414 #if defined(MODULAR) || defined(BPFJIT)
2415 static int
2416 sysctl_net_bpf_jit(SYSCTLFN_ARGS)
2417 {
2418 	bool newval;
2419 	int error;
2420 	struct sysctlnode node;
2421 
2422 	node = *rnode;
2423 	node.sysctl_data = &newval;
2424 	newval = bpf_jit;
2425 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2426 	if (error != 0 || newp == NULL)
2427 		return error;
2428 
2429 	bpf_jit = newval;
2430 	if (newval && bpfjit_module_ops.bj_generate_code == NULL) {
2431 		printf("JIT compilation is postponed "
2432 		    "until after bpfjit module is loaded\n");
2433 	}
2434 
2435 	return 0;
2436 }
2437 #endif
2438 
2439 static int
2440 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
2441 {
2442 	int    error, elem_count;
2443 	struct bpf_d	 *dp;
2444 	struct bpf_d_ext  dpe;
2445 	size_t len, needed, elem_size, out_size;
2446 	char   *sp;
2447 
2448 	if (namelen == 1 && name[0] == CTL_QUERY)
2449 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
2450 
2451 	if (namelen != 2)
2452 		return (EINVAL);
2453 
2454 	/* BPF peers is privileged information. */
2455 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
2456 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
2457 	if (error)
2458 		return (EPERM);
2459 
2460 	len = (oldp != NULL) ? *oldlenp : 0;
2461 	sp = oldp;
2462 	elem_size = name[0];
2463 	elem_count = name[1];
2464 	out_size = MIN(sizeof(dpe), elem_size);
2465 	needed = 0;
2466 
2467 	if (elem_size < 1 || elem_count < 0)
2468 		return (EINVAL);
2469 
2470 	mutex_enter(&bpf_mtx);
2471 	BPF_DLIST_WRITER_FOREACH(dp) {
2472 		if (len >= elem_size && elem_count > 0) {
2473 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
2474 			BPF_EXT(bufsize);
2475 			BPF_EXT(promisc);
2476 			BPF_EXT(state);
2477 			BPF_EXT(immediate);
2478 			BPF_EXT(hdrcmplt);
2479 			BPF_EXT(direction);
2480 			BPF_EXT(pid);
2481 			BPF_EXT(rcount);
2482 			BPF_EXT(dcount);
2483 			BPF_EXT(ccount);
2484 #undef BPF_EXT
2485 			mutex_enter(dp->bd_mtx);
2486 			if (dp->bd_bif)
2487 				(void)strlcpy(dpe.bde_ifname,
2488 				    dp->bd_bif->bif_ifp->if_xname,
2489 				    IFNAMSIZ - 1);
2490 			else
2491 				dpe.bde_ifname[0] = '\0';
2492 			dpe.bde_locked = dp->bd_locked;
2493 			mutex_exit(dp->bd_mtx);
2494 
2495 			error = copyout(&dpe, sp, out_size);
2496 			if (error)
2497 				break;
2498 			sp += elem_size;
2499 			len -= elem_size;
2500 		}
2501 		needed += elem_size;
2502 		if (elem_count > 0 && elem_count != INT_MAX)
2503 			elem_count--;
2504 	}
2505 	mutex_exit(&bpf_mtx);
2506 
2507 	*oldlenp = needed;
2508 
2509 	return (error);
2510 }
2511 
2512 static void
2513 bpf_stats(void *p, void *arg, struct cpu_info *ci __unused)
2514 {
2515 	struct bpf_stat *const stats = p;
2516 	struct bpf_stat *sum = arg;
2517 
2518 	int s = splnet();
2519 
2520 	sum->bs_recv += stats->bs_recv;
2521 	sum->bs_drop += stats->bs_drop;
2522 	sum->bs_capt += stats->bs_capt;
2523 
2524 	splx(s);
2525 }
2526 
2527 static int
2528 bpf_sysctl_gstats_handler(SYSCTLFN_ARGS)
2529 {
2530 	struct sysctlnode node;
2531 	int error;
2532 	struct bpf_stat sum;
2533 
2534 	memset(&sum, 0, sizeof(sum));
2535 	node = *rnode;
2536 
2537 	percpu_foreach_xcall(bpf_gstats_percpu, XC_HIGHPRI_IPL(IPL_SOFTNET),
2538 	    bpf_stats, &sum);
2539 
2540 	node.sysctl_data = &sum;
2541 	node.sysctl_size = sizeof(sum);
2542 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2543 	if (error != 0 || newp == NULL)
2544 		return error;
2545 
2546 	return 0;
2547 }
2548 
2549 SYSCTL_SETUP(sysctl_net_bpf_setup, "bpf sysctls")
2550 {
2551 	const struct sysctlnode *node;
2552 
2553 	node = NULL;
2554 	sysctl_createv(clog, 0, NULL, &node,
2555 		       CTLFLAG_PERMANENT,
2556 		       CTLTYPE_NODE, "bpf",
2557 		       SYSCTL_DESCR("BPF options"),
2558 		       NULL, 0, NULL, 0,
2559 		       CTL_NET, CTL_CREATE, CTL_EOL);
2560 	if (node != NULL) {
2561 #if defined(MODULAR) || defined(BPFJIT)
2562 		sysctl_createv(clog, 0, NULL, NULL,
2563 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2564 			CTLTYPE_BOOL, "jit",
2565 			SYSCTL_DESCR("Toggle Just-In-Time compilation"),
2566 			sysctl_net_bpf_jit, 0, &bpf_jit, 0,
2567 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2568 #endif
2569 		sysctl_createv(clog, 0, NULL, NULL,
2570 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2571 			CTLTYPE_INT, "maxbufsize",
2572 			SYSCTL_DESCR("Maximum size for data capture buffer"),
2573 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
2574 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2575 		sysctl_createv(clog, 0, NULL, NULL,
2576 			CTLFLAG_PERMANENT,
2577 			CTLTYPE_STRUCT, "stats",
2578 			SYSCTL_DESCR("BPF stats"),
2579 			bpf_sysctl_gstats_handler, 0, NULL, 0,
2580 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2581 		sysctl_createv(clog, 0, NULL, NULL,
2582 			CTLFLAG_PERMANENT,
2583 			CTLTYPE_STRUCT, "peers",
2584 			SYSCTL_DESCR("BPF peers"),
2585 			sysctl_net_bpf_peers, 0, NULL, 0,
2586 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2587 	}
2588 
2589 }
2590 
2591 static int
2592 _bpf_register_track_event(struct bpf_if **driverp,
2593 	    void (*_fun)(struct bpf_if *, struct ifnet *, int, int))
2594 {
2595 	struct bpf_if *bp;
2596 	struct bpf_event_tracker *t;
2597 	int ret = ENOENT;
2598 
2599 	t = kmem_zalloc(sizeof(*t), KM_SLEEP);
2600 	if (!t)
2601 		return ENOMEM;
2602 	t->bet_notify = _fun;
2603 
2604 	mutex_enter(&bpf_mtx);
2605 	BPF_IFLIST_WRITER_FOREACH(bp) {
2606 		if (bp->bif_driverp != driverp)
2607 			continue;
2608 		SLIST_INSERT_HEAD(&bp->bif_trackers, t, bet_entries);
2609 		ret = 0;
2610 		break;
2611 	}
2612 	mutex_exit(&bpf_mtx);
2613 
2614 	return ret;
2615 }
2616 
2617 static int
2618 _bpf_deregister_track_event(struct bpf_if **driverp,
2619 	    void (*_fun)(struct bpf_if *, struct ifnet *, int, int))
2620 {
2621 	struct bpf_if *bp;
2622 	struct bpf_event_tracker *t = NULL;
2623 	int ret = ENOENT;
2624 
2625 	mutex_enter(&bpf_mtx);
2626 	BPF_IFLIST_WRITER_FOREACH(bp) {
2627 		if (bp->bif_driverp != driverp)
2628 			continue;
2629 		SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) {
2630 			if (t->bet_notify == _fun) {
2631 				ret = 0;
2632 				break;
2633 			}
2634 		}
2635 		if (ret == 0)
2636 			break;
2637 	}
2638 	if (ret == 0 && t && t->bet_notify == _fun) {
2639 		SLIST_REMOVE(&bp->bif_trackers, t, bpf_event_tracker,
2640 		    bet_entries);
2641 	}
2642 	mutex_exit(&bpf_mtx);
2643 	if (ret == 0)
2644 		kmem_free(t, sizeof(*t));
2645 	return ret;
2646 }
2647 
2648 struct bpf_ops bpf_ops_kernel = {
2649 	.bpf_attach =		_bpfattach,
2650 	.bpf_detach =		_bpfdetach,
2651 	.bpf_change_type =	_bpf_change_type,
2652 	.bpf_register_track_event = _bpf_register_track_event,
2653 	.bpf_deregister_track_event = _bpf_deregister_track_event,
2654 
2655 	.bpf_mtap =		_bpf_mtap,
2656 	.bpf_mtap2 =		_bpf_mtap2,
2657 	.bpf_mtap_af =		_bpf_mtap_af,
2658 	.bpf_mtap_sl_in =	_bpf_mtap_sl_in,
2659 	.bpf_mtap_sl_out =	_bpf_mtap_sl_out,
2660 
2661 	.bpf_mtap_softint =		_bpf_mtap_softint,
2662 	.bpf_mtap_softint_init =	_bpf_mtap_softint_init,
2663 };
2664 
2665 MODULE(MODULE_CLASS_DRIVER, bpf, "bpf_filter");
2666 
2667 static int
2668 bpf_modcmd(modcmd_t cmd, void *arg)
2669 {
2670 #ifdef _MODULE
2671 	devmajor_t bmajor, cmajor;
2672 #endif
2673 	int error = 0;
2674 
2675 	switch (cmd) {
2676 	case MODULE_CMD_INIT:
2677 		bpf_init();
2678 #ifdef _MODULE
2679 		bmajor = cmajor = NODEVMAJOR;
2680 		error = devsw_attach("bpf", NULL, &bmajor,
2681 		    &bpf_cdevsw, &cmajor);
2682 		if (error)
2683 			break;
2684 #endif
2685 
2686 		bpf_ops_handover_enter(&bpf_ops_kernel);
2687 		atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel);
2688 		bpf_ops_handover_exit();
2689 		break;
2690 
2691 	case MODULE_CMD_FINI:
2692 		/*
2693 		 * While there is no reference counting for bpf callers,
2694 		 * unload could at least in theory be done similarly to
2695 		 * system call disestablishment.  This should even be
2696 		 * a little simpler:
2697 		 *
2698 		 * 1) replace op vector with stubs
2699 		 * 2) post update to all cpus with xc
2700 		 * 3) check that nobody is in bpf anymore
2701 		 *    (it's doubtful we'd want something like l_sysent,
2702 		 *     but we could do something like *signed* percpu
2703 		 *     counters.  if the sum is 0, we're good).
2704 		 * 4) if fail, unroll changes
2705 		 *
2706 		 * NOTE: change won't be atomic to the outside.  some
2707 		 * packets may be not captured even if unload is
2708 		 * not successful.  I think packet capture not working
2709 		 * is a perfectly logical consequence of trying to
2710 		 * disable packet capture.
2711 		 */
2712 		error = EOPNOTSUPP;
2713 		break;
2714 
2715 	default:
2716 		error = ENOTTY;
2717 		break;
2718 	}
2719 
2720 	return error;
2721 }
2722