xref: /netbsd-src/sys/net/bpf.c (revision d909946ca08dceb44d7d0f22ec9488679695d976)
1 /*	$NetBSD: bpf.c,v 1.203 2016/07/19 02:47:45 pgoyette 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 1.203 2016/07/19 02:47:45 pgoyette Exp $");
43 
44 #if defined(_KERNEL_OPT)
45 #include "opt_bpf.h"
46 #include "sl.h"
47 #include "strip.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 
64 #include <sys/file.h>
65 #include <sys/filedesc.h>
66 #include <sys/tty.h>
67 #include <sys/uio.h>
68 
69 #include <sys/protosw.h>
70 #include <sys/socket.h>
71 #include <sys/errno.h>
72 #include <sys/kernel.h>
73 #include <sys/poll.h>
74 #include <sys/sysctl.h>
75 #include <sys/kauth.h>
76 
77 #include <net/if.h>
78 #include <net/slip.h>
79 
80 #include <net/bpf.h>
81 #include <net/bpfdesc.h>
82 #include <net/bpfjit.h>
83 
84 #include <net/if_arc.h>
85 #include <net/if_ether.h>
86 
87 #include <netinet/in.h>
88 #include <netinet/if_inarp.h>
89 
90 
91 #include <compat/sys/sockio.h>
92 
93 #ifndef BPF_BUFSIZE
94 /*
95  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
96  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
97  */
98 # define BPF_BUFSIZE 32768
99 #endif
100 
101 #define PRINET  26			/* interruptible */
102 
103 /*
104  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
105  * XXX the default values should be computed dynamically based
106  * on available memory size and available mbuf clusters.
107  */
108 int bpf_bufsize = BPF_BUFSIZE;
109 int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
110 bool bpf_jit = false;
111 
112 struct bpfjit_ops bpfjit_module_ops = {
113 	.bj_generate_code = NULL,
114 	.bj_free_code = NULL
115 };
116 
117 /*
118  * Global BPF statistics returned by net.bpf.stats sysctl.
119  */
120 struct bpf_stat	bpf_gstats;
121 
122 /*
123  * Use a mutex to avoid a race condition between gathering the stats/peers
124  * and opening/closing the device.
125  */
126 static kmutex_t bpf_mtx;
127 
128 /*
129  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
130  *  bpf_dtab holds the descriptors, indexed by minor device #
131  */
132 struct bpf_if	*bpf_iflist;
133 LIST_HEAD(, bpf_d) bpf_list;
134 
135 static int	bpf_allocbufs(struct bpf_d *);
136 static void	bpf_deliver(struct bpf_if *,
137 		            void *(*cpfn)(void *, const void *, size_t),
138 		            void *, u_int, u_int, const bool);
139 static void	bpf_freed(struct bpf_d *);
140 static void	bpf_ifname(struct ifnet *, struct ifreq *);
141 static void	*bpf_mcpy(void *, const void *, size_t);
142 static int	bpf_movein(struct uio *, int, uint64_t,
143 			        struct mbuf **, struct sockaddr *);
144 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
145 static void	bpf_detachd(struct bpf_d *);
146 static int	bpf_setif(struct bpf_d *, struct ifreq *);
147 static void	bpf_timed_out(void *);
148 static inline void
149 		bpf_wakeup(struct bpf_d *);
150 static int	bpf_hdrlen(struct bpf_d *);
151 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
152     void *(*)(void *, const void *, size_t), struct timespec *);
153 static void	reset_d(struct bpf_d *);
154 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
155 static int	bpf_setdlt(struct bpf_d *, u_int);
156 
157 static int	bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t,
158     int);
159 static int	bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t,
160     int);
161 static int	bpf_ioctl(struct file *, u_long, void *);
162 static int	bpf_poll(struct file *, int);
163 static int	bpf_stat(struct file *, struct stat *);
164 static int	bpf_close(struct file *);
165 static int	bpf_kqfilter(struct file *, struct knote *);
166 static void	bpf_softintr(void *);
167 
168 static const struct fileops bpf_fileops = {
169 	.fo_read = bpf_read,
170 	.fo_write = bpf_write,
171 	.fo_ioctl = bpf_ioctl,
172 	.fo_fcntl = fnullop_fcntl,
173 	.fo_poll = bpf_poll,
174 	.fo_stat = bpf_stat,
175 	.fo_close = bpf_close,
176 	.fo_kqfilter = bpf_kqfilter,
177 	.fo_restart = fnullop_restart,
178 };
179 
180 dev_type_open(bpfopen);
181 
182 const struct cdevsw bpf_cdevsw = {
183 	.d_open = bpfopen,
184 	.d_close = noclose,
185 	.d_read = noread,
186 	.d_write = nowrite,
187 	.d_ioctl = noioctl,
188 	.d_stop = nostop,
189 	.d_tty = notty,
190 	.d_poll = nopoll,
191 	.d_mmap = nommap,
192 	.d_kqfilter = nokqfilter,
193 	.d_discard = nodiscard,
194 	.d_flag = D_OTHER
195 };
196 
197 bpfjit_func_t
198 bpf_jit_generate(bpf_ctx_t *bc, void *code, size_t size)
199 {
200 
201 	membar_consumer();
202 	if (bpfjit_module_ops.bj_generate_code != NULL) {
203 		return bpfjit_module_ops.bj_generate_code(bc, code, size);
204 	}
205 	return NULL;
206 }
207 
208 void
209 bpf_jit_freecode(bpfjit_func_t jcode)
210 {
211 	KASSERT(bpfjit_module_ops.bj_free_code != NULL);
212 	bpfjit_module_ops.bj_free_code(jcode);
213 }
214 
215 static int
216 bpf_movein(struct uio *uio, int linktype, uint64_t mtu, struct mbuf **mp,
217 	   struct sockaddr *sockp)
218 {
219 	struct mbuf *m;
220 	int error;
221 	size_t len;
222 	size_t hlen;
223 	size_t align;
224 
225 	/*
226 	 * Build a sockaddr based on the data link layer type.
227 	 * We do this at this level because the ethernet header
228 	 * is copied directly into the data field of the sockaddr.
229 	 * In the case of SLIP, there is no header and the packet
230 	 * is forwarded as is.
231 	 * Also, we are careful to leave room at the front of the mbuf
232 	 * for the link level header.
233 	 */
234 	switch (linktype) {
235 
236 	case DLT_SLIP:
237 		sockp->sa_family = AF_INET;
238 		hlen = 0;
239 		align = 0;
240 		break;
241 
242 	case DLT_PPP:
243 		sockp->sa_family = AF_UNSPEC;
244 		hlen = 0;
245 		align = 0;
246 		break;
247 
248 	case DLT_EN10MB:
249 		sockp->sa_family = AF_UNSPEC;
250 		/* XXX Would MAXLINKHDR be better? */
251  		/* 6(dst)+6(src)+2(type) */
252 		hlen = sizeof(struct ether_header);
253 		align = 2;
254 		break;
255 
256 	case DLT_ARCNET:
257 		sockp->sa_family = AF_UNSPEC;
258 		hlen = ARC_HDRLEN;
259 		align = 5;
260 		break;
261 
262 	case DLT_FDDI:
263 		sockp->sa_family = AF_LINK;
264 		/* XXX 4(FORMAC)+6(dst)+6(src) */
265 		hlen = 16;
266 		align = 0;
267 		break;
268 
269 	case DLT_ECONET:
270 		sockp->sa_family = AF_UNSPEC;
271 		hlen = 6;
272 		align = 2;
273 		break;
274 
275 	case DLT_NULL:
276 		sockp->sa_family = AF_UNSPEC;
277 		hlen = 0;
278 		align = 0;
279 		break;
280 
281 	default:
282 		return (EIO);
283 	}
284 
285 	len = uio->uio_resid;
286 	/*
287 	 * If there aren't enough bytes for a link level header or the
288 	 * packet length exceeds the interface mtu, return an error.
289 	 */
290 	if (len - hlen > mtu)
291 		return (EMSGSIZE);
292 
293 	/*
294 	 * XXX Avoid complicated buffer chaining ---
295 	 * bail if it won't fit in a single mbuf.
296 	 * (Take into account possible alignment bytes)
297 	 */
298 	if (len + align > MCLBYTES)
299 		return (EIO);
300 
301 	m = m_gethdr(M_WAIT, MT_DATA);
302 	m_reset_rcvif(m);
303 	m->m_pkthdr.len = (int)(len - hlen);
304 	if (len + align > MHLEN) {
305 		m_clget(m, M_WAIT);
306 		if ((m->m_flags & M_EXT) == 0) {
307 			error = ENOBUFS;
308 			goto bad;
309 		}
310 	}
311 
312 	/* Insure the data is properly aligned */
313 	if (align > 0) {
314 		m->m_data += align;
315 		m->m_len -= (int)align;
316 	}
317 
318 	error = uiomove(mtod(m, void *), len, uio);
319 	if (error)
320 		goto bad;
321 	if (hlen != 0) {
322 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
323 		m->m_data += hlen; /* XXX */
324 		len -= hlen;
325 	}
326 	m->m_len = (int)len;
327 	*mp = m;
328 	return (0);
329 
330 bad:
331 	m_freem(m);
332 	return (error);
333 }
334 
335 /*
336  * Attach file to the bpf interface, i.e. make d listen on bp.
337  * Must be called at splnet.
338  */
339 static void
340 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
341 {
342 	KASSERT(mutex_owned(&bpf_mtx));
343 	/*
344 	 * Point d at bp, and add d to the interface's list of listeners.
345 	 * Finally, point the driver's bpf cookie at the interface so
346 	 * it will divert packets to bpf.
347 	 */
348 	d->bd_bif = bp;
349 	d->bd_next = bp->bif_dlist;
350 	bp->bif_dlist = d;
351 
352 	*bp->bif_driverp = bp;
353 }
354 
355 /*
356  * Detach a file from its interface.
357  */
358 static void
359 bpf_detachd(struct bpf_d *d)
360 {
361 	struct bpf_d **p;
362 	struct bpf_if *bp;
363 
364 	KASSERT(mutex_owned(&bpf_mtx));
365 
366 	bp = d->bd_bif;
367 	/*
368 	 * Check if this descriptor had requested promiscuous mode.
369 	 * If so, turn it off.
370 	 */
371 	if (d->bd_promisc) {
372 		int error __diagused;
373 
374 		d->bd_promisc = 0;
375 		/*
376 		 * Take device out of promiscuous mode.  Since we were
377 		 * able to enter promiscuous mode, we should be able
378 		 * to turn it off.  But we can get an error if
379 		 * the interface was configured down, so only panic
380 		 * if we don't get an unexpected error.
381 		 */
382   		error = ifpromisc(bp->bif_ifp, 0);
383 #ifdef DIAGNOSTIC
384 		if (error)
385 			printf("%s: ifpromisc failed: %d", __func__, error);
386 #endif
387 	}
388 	/* Remove d from the interface's descriptor list. */
389 	p = &bp->bif_dlist;
390 	while (*p != d) {
391 		p = &(*p)->bd_next;
392 		if (*p == NULL)
393 			panic("%s: descriptor not in list", __func__);
394 	}
395 	*p = (*p)->bd_next;
396 	if (bp->bif_dlist == NULL)
397 		/*
398 		 * Let the driver know that there are no more listeners.
399 		 */
400 		*d->bd_bif->bif_driverp = NULL;
401 	d->bd_bif = NULL;
402 }
403 
404 static void
405 bpf_init(void)
406 {
407 
408 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
409 
410 	LIST_INIT(&bpf_list);
411 
412 	bpf_gstats.bs_recv = 0;
413 	bpf_gstats.bs_drop = 0;
414 	bpf_gstats.bs_capt = 0;
415 
416 	return;
417 }
418 
419 /*
420  * bpfilterattach() is called at boot time.  We don't need to do anything
421  * here, since any initialization will happen as part of module init code.
422  */
423 /* ARGSUSED */
424 void
425 bpfilterattach(int n)
426 {
427 
428 }
429 
430 /*
431  * Open ethernet device. Clones.
432  */
433 /* ARGSUSED */
434 int
435 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
436 {
437 	struct bpf_d *d;
438 	struct file *fp;
439 	int error, fd;
440 
441 	/* falloc() will fill in the descriptor for us. */
442 	if ((error = fd_allocfile(&fp, &fd)) != 0)
443 		return error;
444 
445 	d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK|M_ZERO);
446 	d->bd_bufsize = bpf_bufsize;
447 	d->bd_seesent = 1;
448 	d->bd_feedback = 0;
449 	d->bd_pid = l->l_proc->p_pid;
450 #ifdef _LP64
451 	if (curproc->p_flag & PK_32)
452 		d->bd_compat32 = 1;
453 #endif
454 	getnanotime(&d->bd_btime);
455 	d->bd_atime = d->bd_mtime = d->bd_btime;
456 	callout_init(&d->bd_callout, 0);
457 	selinit(&d->bd_sel);
458 	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
459 	d->bd_jitcode = NULL;
460 
461 	mutex_enter(&bpf_mtx);
462 	LIST_INSERT_HEAD(&bpf_list, d, bd_list);
463 	mutex_exit(&bpf_mtx);
464 
465 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
466 }
467 
468 /*
469  * Close the descriptor by detaching it from its interface,
470  * deallocating its buffers, and marking it free.
471  */
472 /* ARGSUSED */
473 static int
474 bpf_close(struct file *fp)
475 {
476 	struct bpf_d *d;
477 	int s;
478 
479 	KERNEL_LOCK(1, NULL);
480 	mutex_enter(&bpf_mtx);
481 
482 	if ((d = fp->f_bpf) == NULL) {
483 		mutex_exit(&bpf_mtx);
484 		KERNEL_UNLOCK_ONE(NULL);
485 		return 0;
486 	}
487 
488 	/*
489 	 * Refresh the PID associated with this bpf file.
490 	 */
491 	d->bd_pid = curproc->p_pid;
492 
493 	s = splnet();
494 	if (d->bd_state == BPF_WAITING)
495 		callout_stop(&d->bd_callout);
496 	d->bd_state = BPF_IDLE;
497 	if (d->bd_bif)
498 		bpf_detachd(d);
499 	splx(s);
500 	bpf_freed(d);
501 	LIST_REMOVE(d, bd_list);
502 	fp->f_bpf = NULL;
503 
504 	mutex_exit(&bpf_mtx);
505 	KERNEL_UNLOCK_ONE(NULL);
506 
507 	callout_destroy(&d->bd_callout);
508 	seldestroy(&d->bd_sel);
509 	softint_disestablish(d->bd_sih);
510 	free(d, M_DEVBUF);
511 
512 	return (0);
513 }
514 
515 /*
516  * Rotate the packet buffers in descriptor d.  Move the store buffer
517  * into the hold slot, and the free buffer into the store slot.
518  * Zero the length of the new store buffer.
519  */
520 #define ROTATE_BUFFERS(d) \
521 	(d)->bd_hbuf = (d)->bd_sbuf; \
522 	(d)->bd_hlen = (d)->bd_slen; \
523 	(d)->bd_sbuf = (d)->bd_fbuf; \
524 	(d)->bd_slen = 0; \
525 	(d)->bd_fbuf = NULL;
526 /*
527  *  bpfread - read next chunk of packets from buffers
528  */
529 static int
530 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
531     kauth_cred_t cred, int flags)
532 {
533 	struct bpf_d *d = fp->f_bpf;
534 	int timed_out;
535 	int error;
536 	int s;
537 
538 	getnanotime(&d->bd_atime);
539 	/*
540 	 * Restrict application to use a buffer the same size as
541 	 * the kernel buffers.
542 	 */
543 	if (uio->uio_resid != d->bd_bufsize)
544 		return (EINVAL);
545 
546 	KERNEL_LOCK(1, NULL);
547 	s = splnet();
548 	if (d->bd_state == BPF_WAITING)
549 		callout_stop(&d->bd_callout);
550 	timed_out = (d->bd_state == BPF_TIMED_OUT);
551 	d->bd_state = BPF_IDLE;
552 	/*
553 	 * If the hold buffer is empty, then do a timed sleep, which
554 	 * ends when the timeout expires or when enough packets
555 	 * have arrived to fill the store buffer.
556 	 */
557 	while (d->bd_hbuf == NULL) {
558 		if (fp->f_flag & FNONBLOCK) {
559 			if (d->bd_slen == 0) {
560 				splx(s);
561 				KERNEL_UNLOCK_ONE(NULL);
562 				return (EWOULDBLOCK);
563 			}
564 			ROTATE_BUFFERS(d);
565 			break;
566 		}
567 
568 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
569 			/*
570 			 * A packet(s) either arrived since the previous
571 			 * read or arrived while we were asleep.
572 			 * Rotate the buffers and return what's here.
573 			 */
574 			ROTATE_BUFFERS(d);
575 			break;
576 		}
577 		error = tsleep(d, PRINET|PCATCH, "bpf",
578 				d->bd_rtout);
579 		if (error == EINTR || error == ERESTART) {
580 			splx(s);
581 			KERNEL_UNLOCK_ONE(NULL);
582 			return (error);
583 		}
584 		if (error == EWOULDBLOCK) {
585 			/*
586 			 * On a timeout, return what's in the buffer,
587 			 * which may be nothing.  If there is something
588 			 * in the store buffer, we can rotate the buffers.
589 			 */
590 			if (d->bd_hbuf)
591 				/*
592 				 * We filled up the buffer in between
593 				 * getting the timeout and arriving
594 				 * here, so we don't need to rotate.
595 				 */
596 				break;
597 
598 			if (d->bd_slen == 0) {
599 				splx(s);
600 				KERNEL_UNLOCK_ONE(NULL);
601 				return (0);
602 			}
603 			ROTATE_BUFFERS(d);
604 			break;
605 		}
606 		if (error != 0)
607 			goto done;
608 	}
609 	/*
610 	 * At this point, we know we have something in the hold slot.
611 	 */
612 	splx(s);
613 
614 	/*
615 	 * Move data from hold buffer into user space.
616 	 * We know the entire buffer is transferred since
617 	 * we checked above that the read buffer is bpf_bufsize bytes.
618 	 */
619 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
620 
621 	s = splnet();
622 	d->bd_fbuf = d->bd_hbuf;
623 	d->bd_hbuf = NULL;
624 	d->bd_hlen = 0;
625 done:
626 	splx(s);
627 	KERNEL_UNLOCK_ONE(NULL);
628 	return (error);
629 }
630 
631 
632 /*
633  * If there are processes sleeping on this descriptor, wake them up.
634  */
635 static inline void
636 bpf_wakeup(struct bpf_d *d)
637 {
638 	wakeup(d);
639 	if (d->bd_async)
640 		softint_schedule(d->bd_sih);
641 	selnotify(&d->bd_sel, 0, 0);
642 }
643 
644 static void
645 bpf_softintr(void *cookie)
646 {
647 	struct bpf_d *d;
648 
649 	d = cookie;
650 	if (d->bd_async)
651 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
652 }
653 
654 static void
655 bpf_timed_out(void *arg)
656 {
657 	struct bpf_d *d = arg;
658 	int s;
659 
660 	s = splnet();
661 	if (d->bd_state == BPF_WAITING) {
662 		d->bd_state = BPF_TIMED_OUT;
663 		if (d->bd_slen != 0)
664 			bpf_wakeup(d);
665 	}
666 	splx(s);
667 }
668 
669 
670 static int
671 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
672     kauth_cred_t cred, int flags)
673 {
674 	struct bpf_d *d = fp->f_bpf;
675 	struct ifnet *ifp;
676 	struct mbuf *m, *mc;
677 	int error, s;
678 	static struct sockaddr_storage dst;
679 
680 	m = NULL;	/* XXX gcc */
681 
682 	KERNEL_LOCK(1, NULL);
683 
684 	if (d->bd_bif == NULL) {
685 		KERNEL_UNLOCK_ONE(NULL);
686 		return (ENXIO);
687 	}
688 	getnanotime(&d->bd_mtime);
689 
690 	ifp = d->bd_bif->bif_ifp;
691 
692 	if (uio->uio_resid == 0) {
693 		KERNEL_UNLOCK_ONE(NULL);
694 		return (0);
695 	}
696 
697 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
698 		(struct sockaddr *) &dst);
699 	if (error) {
700 		KERNEL_UNLOCK_ONE(NULL);
701 		return (error);
702 	}
703 
704 	if (m->m_pkthdr.len > ifp->if_mtu) {
705 		KERNEL_UNLOCK_ONE(NULL);
706 		m_freem(m);
707 		return (EMSGSIZE);
708 	}
709 
710 	if (d->bd_hdrcmplt)
711 		dst.ss_family = pseudo_AF_HDRCMPLT;
712 
713 	if (d->bd_feedback) {
714 		mc = m_dup(m, 0, M_COPYALL, M_NOWAIT);
715 		if (mc != NULL)
716 			m_set_rcvif(mc, ifp);
717 		/* Set M_PROMISC for outgoing packets to be discarded. */
718 		if (1 /*d->bd_direction == BPF_D_INOUT*/)
719 			m->m_flags |= M_PROMISC;
720 	} else
721 		mc = NULL;
722 
723 	s = splsoftnet();
724 	error = if_output_lock(ifp, ifp, m, (struct sockaddr *) &dst, NULL);
725 
726 	if (mc != NULL) {
727 		if (error == 0)
728 			ifp->_if_input(ifp, mc);
729 		else
730 			m_freem(mc);
731 	}
732 	splx(s);
733 	KERNEL_UNLOCK_ONE(NULL);
734 	/*
735 	 * The driver frees the mbuf.
736 	 */
737 	return (error);
738 }
739 
740 /*
741  * Reset a descriptor by flushing its packet buffer and clearing the
742  * receive and drop counts.  Should be called at splnet.
743  */
744 static void
745 reset_d(struct bpf_d *d)
746 {
747 	if (d->bd_hbuf) {
748 		/* Free the hold buffer. */
749 		d->bd_fbuf = d->bd_hbuf;
750 		d->bd_hbuf = NULL;
751 	}
752 	d->bd_slen = 0;
753 	d->bd_hlen = 0;
754 	d->bd_rcount = 0;
755 	d->bd_dcount = 0;
756 	d->bd_ccount = 0;
757 }
758 
759 /*
760  *  FIONREAD		Check for read packet available.
761  *  BIOCGBLEN		Get buffer len [for read()].
762  *  BIOCSETF		Set ethernet read filter.
763  *  BIOCFLUSH		Flush read packet buffer.
764  *  BIOCPROMISC		Put interface into promiscuous mode.
765  *  BIOCGDLT		Get link layer type.
766  *  BIOCGETIF		Get interface name.
767  *  BIOCSETIF		Set interface.
768  *  BIOCSRTIMEOUT	Set read timeout.
769  *  BIOCGRTIMEOUT	Get read timeout.
770  *  BIOCGSTATS		Get packet stats.
771  *  BIOCIMMEDIATE	Set immediate mode.
772  *  BIOCVERSION		Get filter language version.
773  *  BIOCGHDRCMPLT	Get "header already complete" flag.
774  *  BIOCSHDRCMPLT	Set "header already complete" flag.
775  *  BIOCSFEEDBACK	Set packet feedback mode.
776  *  BIOCGFEEDBACK	Get packet feedback mode.
777  *  BIOCGSEESENT  	Get "see sent packets" mode.
778  *  BIOCSSEESENT  	Set "see sent packets" mode.
779  */
780 /* ARGSUSED */
781 static int
782 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
783 {
784 	struct bpf_d *d = fp->f_bpf;
785 	int s, error = 0;
786 
787 	/*
788 	 * Refresh the PID associated with this bpf file.
789 	 */
790 	KERNEL_LOCK(1, NULL);
791 	d->bd_pid = curproc->p_pid;
792 #ifdef _LP64
793 	if (curproc->p_flag & PK_32)
794 		d->bd_compat32 = 1;
795 	else
796 		d->bd_compat32 = 0;
797 #endif
798 
799 	s = splnet();
800 	if (d->bd_state == BPF_WAITING)
801 		callout_stop(&d->bd_callout);
802 	d->bd_state = BPF_IDLE;
803 	splx(s);
804 
805 	switch (cmd) {
806 
807 	default:
808 		error = EINVAL;
809 		break;
810 
811 	/*
812 	 * Check for read packet available.
813 	 */
814 	case FIONREAD:
815 		{
816 			int n;
817 
818 			s = splnet();
819 			n = d->bd_slen;
820 			if (d->bd_hbuf)
821 				n += d->bd_hlen;
822 			splx(s);
823 
824 			*(int *)addr = n;
825 			break;
826 		}
827 
828 	/*
829 	 * Get buffer len [for read()].
830 	 */
831 	case BIOCGBLEN:
832 		*(u_int *)addr = d->bd_bufsize;
833 		break;
834 
835 	/*
836 	 * Set buffer length.
837 	 */
838 	case BIOCSBLEN:
839 		if (d->bd_bif != NULL)
840 			error = EINVAL;
841 		else {
842 			u_int size = *(u_int *)addr;
843 
844 			if (size > bpf_maxbufsize)
845 				*(u_int *)addr = size = bpf_maxbufsize;
846 			else if (size < BPF_MINBUFSIZE)
847 				*(u_int *)addr = size = BPF_MINBUFSIZE;
848 			d->bd_bufsize = size;
849 		}
850 		break;
851 
852 	/*
853 	 * Set link layer read filter.
854 	 */
855 	case BIOCSETF:
856 		error = bpf_setf(d, addr);
857 		break;
858 
859 	/*
860 	 * Flush read packet buffer.
861 	 */
862 	case BIOCFLUSH:
863 		s = splnet();
864 		reset_d(d);
865 		splx(s);
866 		break;
867 
868 	/*
869 	 * Put interface into promiscuous mode.
870 	 */
871 	case BIOCPROMISC:
872 		if (d->bd_bif == NULL) {
873 			/*
874 			 * No interface attached yet.
875 			 */
876 			error = EINVAL;
877 			break;
878 		}
879 		s = splnet();
880 		if (d->bd_promisc == 0) {
881 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
882 			if (error == 0)
883 				d->bd_promisc = 1;
884 		}
885 		splx(s);
886 		break;
887 
888 	/*
889 	 * Get device parameters.
890 	 */
891 	case BIOCGDLT:
892 		if (d->bd_bif == NULL)
893 			error = EINVAL;
894 		else
895 			*(u_int *)addr = d->bd_bif->bif_dlt;
896 		break;
897 
898 	/*
899 	 * Get a list of supported device parameters.
900 	 */
901 	case BIOCGDLTLIST:
902 		if (d->bd_bif == NULL)
903 			error = EINVAL;
904 		else
905 			error = bpf_getdltlist(d, addr);
906 		break;
907 
908 	/*
909 	 * Set device parameters.
910 	 */
911 	case BIOCSDLT:
912 		mutex_enter(&bpf_mtx);
913 		if (d->bd_bif == NULL)
914 			error = EINVAL;
915 		else
916 			error = bpf_setdlt(d, *(u_int *)addr);
917 		mutex_exit(&bpf_mtx);
918 		break;
919 
920 	/*
921 	 * Set interface name.
922 	 */
923 #ifdef OBIOCGETIF
924 	case OBIOCGETIF:
925 #endif
926 	case BIOCGETIF:
927 		if (d->bd_bif == NULL)
928 			error = EINVAL;
929 		else
930 			bpf_ifname(d->bd_bif->bif_ifp, addr);
931 		break;
932 
933 	/*
934 	 * Set interface.
935 	 */
936 #ifdef OBIOCSETIF
937 	case OBIOCSETIF:
938 #endif
939 	case BIOCSETIF:
940 		mutex_enter(&bpf_mtx);
941 		error = bpf_setif(d, addr);
942 		mutex_exit(&bpf_mtx);
943 		break;
944 
945 	/*
946 	 * Set read timeout.
947 	 */
948 	case BIOCSRTIMEOUT:
949 		{
950 			struct timeval *tv = addr;
951 
952 			/* Compute number of ticks. */
953 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
954 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
955 				d->bd_rtout = 1;
956 			break;
957 		}
958 
959 #ifdef BIOCGORTIMEOUT
960 	/*
961 	 * Get read timeout.
962 	 */
963 	case BIOCGORTIMEOUT:
964 		{
965 			struct timeval50 *tv = addr;
966 
967 			tv->tv_sec = d->bd_rtout / hz;
968 			tv->tv_usec = (d->bd_rtout % hz) * tick;
969 			break;
970 		}
971 #endif
972 
973 #ifdef BIOCSORTIMEOUT
974 	/*
975 	 * Set read timeout.
976 	 */
977 	case BIOCSORTIMEOUT:
978 		{
979 			struct timeval50 *tv = addr;
980 
981 			/* Compute number of ticks. */
982 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
983 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
984 				d->bd_rtout = 1;
985 			break;
986 		}
987 #endif
988 
989 	/*
990 	 * Get read timeout.
991 	 */
992 	case BIOCGRTIMEOUT:
993 		{
994 			struct timeval *tv = addr;
995 
996 			tv->tv_sec = d->bd_rtout / hz;
997 			tv->tv_usec = (d->bd_rtout % hz) * tick;
998 			break;
999 		}
1000 	/*
1001 	 * Get packet stats.
1002 	 */
1003 	case BIOCGSTATS:
1004 		{
1005 			struct bpf_stat *bs = addr;
1006 
1007 			bs->bs_recv = d->bd_rcount;
1008 			bs->bs_drop = d->bd_dcount;
1009 			bs->bs_capt = d->bd_ccount;
1010 			break;
1011 		}
1012 
1013 	case BIOCGSTATSOLD:
1014 		{
1015 			struct bpf_stat_old *bs = addr;
1016 
1017 			bs->bs_recv = d->bd_rcount;
1018 			bs->bs_drop = d->bd_dcount;
1019 			break;
1020 		}
1021 
1022 	/*
1023 	 * Set immediate mode.
1024 	 */
1025 	case BIOCIMMEDIATE:
1026 		d->bd_immediate = *(u_int *)addr;
1027 		break;
1028 
1029 	case BIOCVERSION:
1030 		{
1031 			struct bpf_version *bv = addr;
1032 
1033 			bv->bv_major = BPF_MAJOR_VERSION;
1034 			bv->bv_minor = BPF_MINOR_VERSION;
1035 			break;
1036 		}
1037 
1038 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
1039 		*(u_int *)addr = d->bd_hdrcmplt;
1040 		break;
1041 
1042 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
1043 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1044 		break;
1045 
1046 	/*
1047 	 * Get "see sent packets" flag
1048 	 */
1049 	case BIOCGSEESENT:
1050 		*(u_int *)addr = d->bd_seesent;
1051 		break;
1052 
1053 	/*
1054 	 * Set "see sent" packets flag
1055 	 */
1056 	case BIOCSSEESENT:
1057 		d->bd_seesent = *(u_int *)addr;
1058 		break;
1059 
1060 	/*
1061 	 * Set "feed packets from bpf back to input" mode
1062 	 */
1063 	case BIOCSFEEDBACK:
1064 		d->bd_feedback = *(u_int *)addr;
1065 		break;
1066 
1067 	/*
1068 	 * Get "feed packets from bpf back to input" mode
1069 	 */
1070 	case BIOCGFEEDBACK:
1071 		*(u_int *)addr = d->bd_feedback;
1072 		break;
1073 
1074 	case FIONBIO:		/* Non-blocking I/O */
1075 		/*
1076 		 * No need to do anything special as we use IO_NDELAY in
1077 		 * bpfread() as an indication of whether or not to block
1078 		 * the read.
1079 		 */
1080 		break;
1081 
1082 	case FIOASYNC:		/* Send signal on receive packets */
1083 		d->bd_async = *(int *)addr;
1084 		break;
1085 
1086 	case TIOCSPGRP:		/* Process or group to send signals to */
1087 	case FIOSETOWN:
1088 		error = fsetown(&d->bd_pgid, cmd, addr);
1089 		break;
1090 
1091 	case TIOCGPGRP:
1092 	case FIOGETOWN:
1093 		error = fgetown(d->bd_pgid, cmd, addr);
1094 		break;
1095 	}
1096 	KERNEL_UNLOCK_ONE(NULL);
1097 	return (error);
1098 }
1099 
1100 /*
1101  * Set d's packet filter program to fp.  If this file already has a filter,
1102  * free it and replace it.  Returns EINVAL for bogus requests.
1103  */
1104 int
1105 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
1106 {
1107 	struct bpf_insn *fcode, *old;
1108 	bpfjit_func_t jcode, oldj;
1109 	size_t flen, size;
1110 	int s;
1111 
1112 	jcode = NULL;
1113 	flen = fp->bf_len;
1114 
1115 	if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) {
1116 		return EINVAL;
1117 	}
1118 
1119 	if (flen) {
1120 		/*
1121 		 * Allocate the buffer, copy the byte-code from
1122 		 * userspace and validate it.
1123 		 */
1124 		size = flen * sizeof(*fp->bf_insns);
1125 		fcode = malloc(size, M_DEVBUF, M_WAITOK);
1126 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1127 		    !bpf_validate(fcode, (int)flen)) {
1128 			free(fcode, M_DEVBUF);
1129 			return EINVAL;
1130 		}
1131 		membar_consumer();
1132 		if (bpf_jit)
1133 			jcode = bpf_jit_generate(NULL, fcode, flen);
1134 	} else {
1135 		fcode = NULL;
1136 	}
1137 
1138 	s = splnet();
1139 	old = d->bd_filter;
1140 	d->bd_filter = fcode;
1141 	oldj = d->bd_jitcode;
1142 	d->bd_jitcode = jcode;
1143 	reset_d(d);
1144 	splx(s);
1145 
1146 	if (old) {
1147 		free(old, M_DEVBUF);
1148 	}
1149 	if (oldj) {
1150 		bpf_jit_freecode(oldj);
1151 	}
1152 
1153 	return 0;
1154 }
1155 
1156 /*
1157  * Detach a file from its current interface (if attached at all) and attach
1158  * to the interface indicated by the name stored in ifr.
1159  * Return an errno or 0.
1160  */
1161 static int
1162 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1163 {
1164 	struct bpf_if *bp;
1165 	char *cp;
1166 	int unit_seen, i, s, error;
1167 
1168 	KASSERT(mutex_owned(&bpf_mtx));
1169 	/*
1170 	 * Make sure the provided name has a unit number, and default
1171 	 * it to '0' if not specified.
1172 	 * XXX This is ugly ... do this differently?
1173 	 */
1174 	unit_seen = 0;
1175 	cp = ifr->ifr_name;
1176 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1177 	while (*cp++)
1178 		if (*cp >= '0' && *cp <= '9')
1179 			unit_seen = 1;
1180 	if (!unit_seen) {
1181 		/* Make sure to leave room for the '\0'. */
1182 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1183 			if ((ifr->ifr_name[i] >= 'a' &&
1184 			     ifr->ifr_name[i] <= 'z') ||
1185 			    (ifr->ifr_name[i] >= 'A' &&
1186 			     ifr->ifr_name[i] <= 'Z'))
1187 				continue;
1188 			ifr->ifr_name[i] = '0';
1189 		}
1190 	}
1191 
1192 	/*
1193 	 * Look through attached interfaces for the named one.
1194 	 */
1195 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1196 		struct ifnet *ifp = bp->bif_ifp;
1197 
1198 		if (ifp == NULL ||
1199 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1200 			continue;
1201 		/* skip additional entry */
1202 		if (bp->bif_driverp != &ifp->if_bpf)
1203 			continue;
1204 		/*
1205 		 * We found the requested interface.
1206 		 * Allocate the packet buffers if we need to.
1207 		 * If we're already attached to requested interface,
1208 		 * just flush the buffer.
1209 		 */
1210 		if (d->bd_sbuf == NULL) {
1211 			error = bpf_allocbufs(d);
1212 			if (error != 0)
1213 				return (error);
1214 		}
1215 		s = splnet();
1216 		if (bp != d->bd_bif) {
1217 			if (d->bd_bif)
1218 				/*
1219 				 * Detach if attached to something else.
1220 				 */
1221 				bpf_detachd(d);
1222 
1223 			bpf_attachd(d, bp);
1224 		}
1225 		reset_d(d);
1226 		splx(s);
1227 		return (0);
1228 	}
1229 	/* Not found. */
1230 	return (ENXIO);
1231 }
1232 
1233 /*
1234  * Copy the interface name to the ifreq.
1235  */
1236 static void
1237 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1238 {
1239 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1240 }
1241 
1242 static int
1243 bpf_stat(struct file *fp, struct stat *st)
1244 {
1245 	struct bpf_d *d = fp->f_bpf;
1246 
1247 	(void)memset(st, 0, sizeof(*st));
1248 	KERNEL_LOCK(1, NULL);
1249 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1250 	st->st_atimespec = d->bd_atime;
1251 	st->st_mtimespec = d->bd_mtime;
1252 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1253 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1254 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1255 	st->st_mode = S_IFCHR;
1256 	KERNEL_UNLOCK_ONE(NULL);
1257 	return 0;
1258 }
1259 
1260 /*
1261  * Support for poll() system call
1262  *
1263  * Return true iff the specific operation will not block indefinitely - with
1264  * the assumption that it is safe to positively acknowledge a request for the
1265  * ability to write to the BPF device.
1266  * Otherwise, return false but make a note that a selnotify() must be done.
1267  */
1268 static int
1269 bpf_poll(struct file *fp, int events)
1270 {
1271 	struct bpf_d *d = fp->f_bpf;
1272 	int s = splnet();
1273 	int revents;
1274 
1275 	/*
1276 	 * Refresh the PID associated with this bpf file.
1277 	 */
1278 	KERNEL_LOCK(1, NULL);
1279 	d->bd_pid = curproc->p_pid;
1280 
1281 	revents = events & (POLLOUT | POLLWRNORM);
1282 	if (events & (POLLIN | POLLRDNORM)) {
1283 		/*
1284 		 * An imitation of the FIONREAD ioctl code.
1285 		 */
1286 		if (d->bd_hlen != 0 ||
1287 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1288 		     d->bd_slen != 0)) {
1289 			revents |= events & (POLLIN | POLLRDNORM);
1290 		} else {
1291 			selrecord(curlwp, &d->bd_sel);
1292 			/* Start the read timeout if necessary */
1293 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1294 				callout_reset(&d->bd_callout, d->bd_rtout,
1295 					      bpf_timed_out, d);
1296 				d->bd_state = BPF_WAITING;
1297 			}
1298 		}
1299 	}
1300 
1301 	KERNEL_UNLOCK_ONE(NULL);
1302 	splx(s);
1303 	return (revents);
1304 }
1305 
1306 static void
1307 filt_bpfrdetach(struct knote *kn)
1308 {
1309 	struct bpf_d *d = kn->kn_hook;
1310 	int s;
1311 
1312 	KERNEL_LOCK(1, NULL);
1313 	s = splnet();
1314 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1315 	splx(s);
1316 	KERNEL_UNLOCK_ONE(NULL);
1317 }
1318 
1319 static int
1320 filt_bpfread(struct knote *kn, long hint)
1321 {
1322 	struct bpf_d *d = kn->kn_hook;
1323 	int rv;
1324 
1325 	KERNEL_LOCK(1, NULL);
1326 	kn->kn_data = d->bd_hlen;
1327 	if (d->bd_immediate)
1328 		kn->kn_data += d->bd_slen;
1329 	rv = (kn->kn_data > 0);
1330 	KERNEL_UNLOCK_ONE(NULL);
1331 	return rv;
1332 }
1333 
1334 static const struct filterops bpfread_filtops =
1335 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1336 
1337 static int
1338 bpf_kqfilter(struct file *fp, struct knote *kn)
1339 {
1340 	struct bpf_d *d = fp->f_bpf;
1341 	struct klist *klist;
1342 	int s;
1343 
1344 	KERNEL_LOCK(1, NULL);
1345 
1346 	switch (kn->kn_filter) {
1347 	case EVFILT_READ:
1348 		klist = &d->bd_sel.sel_klist;
1349 		kn->kn_fop = &bpfread_filtops;
1350 		break;
1351 
1352 	default:
1353 		KERNEL_UNLOCK_ONE(NULL);
1354 		return (EINVAL);
1355 	}
1356 
1357 	kn->kn_hook = d;
1358 
1359 	s = splnet();
1360 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1361 	splx(s);
1362 	KERNEL_UNLOCK_ONE(NULL);
1363 
1364 	return (0);
1365 }
1366 
1367 /*
1368  * Copy data from an mbuf chain into a buffer.  This code is derived
1369  * from m_copydata in sys/uipc_mbuf.c.
1370  */
1371 static void *
1372 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1373 {
1374 	const struct mbuf *m;
1375 	u_int count;
1376 	u_char *dst;
1377 
1378 	m = src_arg;
1379 	dst = dst_arg;
1380 	while (len > 0) {
1381 		if (m == NULL)
1382 			panic("bpf_mcpy");
1383 		count = min(m->m_len, len);
1384 		memcpy(dst, mtod(m, const void *), count);
1385 		m = m->m_next;
1386 		dst += count;
1387 		len -= count;
1388 	}
1389 	return dst_arg;
1390 }
1391 
1392 /*
1393  * Dispatch a packet to all the listeners on interface bp.
1394  *
1395  * pkt     pointer to the packet, either a data buffer or an mbuf chain
1396  * buflen  buffer length, if pkt is a data buffer
1397  * cpfn    a function that can copy pkt into the listener's buffer
1398  * pktlen  length of the packet
1399  * rcv     true if packet came in
1400  */
1401 static inline void
1402 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1403     void *pkt, u_int pktlen, u_int buflen, const bool rcv)
1404 {
1405 	uint32_t mem[BPF_MEMWORDS];
1406 	bpf_args_t args = {
1407 		.pkt = (const uint8_t *)pkt,
1408 		.wirelen = pktlen,
1409 		.buflen = buflen,
1410 		.mem = mem,
1411 		.arg = NULL
1412 	};
1413 	bool gottime = false;
1414 	struct timespec ts;
1415 
1416 	/*
1417 	 * Note that the IPL does not have to be raised at this point.
1418 	 * The only problem that could arise here is that if two different
1419 	 * interfaces shared any data.  This is not the case.
1420 	 */
1421 	for (struct bpf_d *d = bp->bif_dlist; d != NULL; d = d->bd_next) {
1422 		u_int slen;
1423 
1424 		if (!d->bd_seesent && !rcv) {
1425 			continue;
1426 		}
1427 		d->bd_rcount++;
1428 		bpf_gstats.bs_recv++;
1429 
1430 		if (d->bd_jitcode)
1431 			slen = d->bd_jitcode(NULL, &args);
1432 		else
1433 			slen = bpf_filter_ext(NULL, d->bd_filter, &args);
1434 
1435 		if (!slen) {
1436 			continue;
1437 		}
1438 		if (!gottime) {
1439 			gottime = true;
1440 			nanotime(&ts);
1441 		}
1442 		catchpacket(d, pkt, pktlen, slen, cpfn, &ts);
1443 	}
1444 }
1445 
1446 /*
1447  * Incoming linkage from device drivers.  Process the packet pkt, of length
1448  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1449  * by each process' filter, and if accepted, stashed into the corresponding
1450  * buffer.
1451  */
1452 static void
1453 _bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1454 {
1455 
1456 	bpf_deliver(bp, memcpy, pkt, pktlen, pktlen, true);
1457 }
1458 
1459 /*
1460  * Incoming linkage from device drivers, when the head of the packet is in
1461  * a buffer, and the tail is in an mbuf chain.
1462  */
1463 static void
1464 _bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1465 {
1466 	u_int pktlen;
1467 	struct mbuf mb;
1468 
1469 	/* Skip outgoing duplicate packets. */
1470 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) {
1471 		m->m_flags &= ~M_PROMISC;
1472 		return;
1473 	}
1474 
1475 	pktlen = m_length(m) + dlen;
1476 
1477 	/*
1478 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1479 	 * Note that we cut corners here; we only setup what's
1480 	 * absolutely needed--this mbuf should never go anywhere else.
1481 	 */
1482 	(void)memset(&mb, 0, sizeof(mb));
1483 	mb.m_next = m;
1484 	mb.m_data = data;
1485 	mb.m_len = dlen;
1486 
1487 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif_index != 0);
1488 }
1489 
1490 /*
1491  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1492  */
1493 static void
1494 _bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1495 {
1496 	void *(*cpfn)(void *, const void *, size_t);
1497 	u_int pktlen, buflen;
1498 	void *marg;
1499 
1500 	/* Skip outgoing duplicate packets. */
1501 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) {
1502 		m->m_flags &= ~M_PROMISC;
1503 		return;
1504 	}
1505 
1506 	pktlen = m_length(m);
1507 
1508 	if (pktlen == m->m_len) {
1509 		cpfn = (void *)memcpy;
1510 		marg = mtod(m, void *);
1511 		buflen = pktlen;
1512 	} else {
1513 		cpfn = bpf_mcpy;
1514 		marg = m;
1515 		buflen = 0;
1516 	}
1517 
1518 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif_index != 0);
1519 }
1520 
1521 /*
1522  * We need to prepend the address family as
1523  * a four byte field.  Cons up a dummy header
1524  * to pacify bpf.  This is safe because bpf
1525  * will only read from the mbuf (i.e., it won't
1526  * try to free it or keep a pointer a to it).
1527  */
1528 static void
1529 _bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m)
1530 {
1531 	struct mbuf m0;
1532 
1533 	m0.m_flags = 0;
1534 	m0.m_next = m;
1535 	m0.m_len = 4;
1536 	m0.m_data = (char *)&af;
1537 
1538 	_bpf_mtap(bp, &m0);
1539 }
1540 
1541 /*
1542  * Put the SLIP pseudo-"link header" in place.
1543  * Note this M_PREPEND() should never fail,
1544  * swince we know we always have enough space
1545  * in the input buffer.
1546  */
1547 static void
1548 _bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m)
1549 {
1550 	int s;
1551 	u_char *hp;
1552 
1553 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1554 	if (*m == NULL)
1555 		return;
1556 
1557 	hp = mtod(*m, u_char *);
1558 	hp[SLX_DIR] = SLIPDIR_IN;
1559 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1560 
1561 	s = splnet();
1562 	_bpf_mtap(bp, *m);
1563 	splx(s);
1564 
1565 	m_adj(*m, SLIP_HDRLEN);
1566 }
1567 
1568 /*
1569  * Put the SLIP pseudo-"link header" in
1570  * place.  The compressed header is now
1571  * at the beginning of the mbuf.
1572  */
1573 static void
1574 _bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m)
1575 {
1576 	struct mbuf m0;
1577 	u_char *hp;
1578 	int s;
1579 
1580 	m0.m_flags = 0;
1581 	m0.m_next = m;
1582 	m0.m_data = m0.m_dat;
1583 	m0.m_len = SLIP_HDRLEN;
1584 
1585 	hp = mtod(&m0, u_char *);
1586 
1587 	hp[SLX_DIR] = SLIPDIR_OUT;
1588 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1589 
1590 	s = splnet();
1591 	_bpf_mtap(bp, &m0);
1592 	splx(s);
1593 	m_freem(m);
1594 }
1595 
1596 static int
1597 bpf_hdrlen(struct bpf_d *d)
1598 {
1599 	int hdrlen = d->bd_bif->bif_hdrlen;
1600 	/*
1601 	 * Compute the length of the bpf header.  This is not necessarily
1602 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1603 	 * that the network layer header begins on a longword boundary (for
1604 	 * performance reasons and to alleviate alignment restrictions).
1605 	 */
1606 #ifdef _LP64
1607 	if (d->bd_compat32)
1608 		return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen);
1609 	else
1610 #endif
1611 		return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen);
1612 }
1613 
1614 /*
1615  * Move the packet data from interface memory (pkt) into the
1616  * store buffer. Call the wakeup functions if it's time to wakeup
1617  * a listener (buffer full), "cpfn" is the routine called to do the
1618  * actual data transfer. memcpy is passed in to copy contiguous chunks,
1619  * while bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1620  * pkt is really an mbuf.
1621  */
1622 static void
1623 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1624     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1625 {
1626 	char *h;
1627 	int totlen, curlen, caplen;
1628 	int hdrlen = bpf_hdrlen(d);
1629 	int do_wakeup = 0;
1630 
1631 	++d->bd_ccount;
1632 	++bpf_gstats.bs_capt;
1633 	/*
1634 	 * Figure out how many bytes to move.  If the packet is
1635 	 * greater or equal to the snapshot length, transfer that
1636 	 * much.  Otherwise, transfer the whole packet (unless
1637 	 * we hit the buffer size limit).
1638 	 */
1639 	totlen = hdrlen + min(snaplen, pktlen);
1640 	if (totlen > d->bd_bufsize)
1641 		totlen = d->bd_bufsize;
1642 	/*
1643 	 * If we adjusted totlen to fit the bufsize, it could be that
1644 	 * totlen is smaller than hdrlen because of the link layer header.
1645 	 */
1646 	caplen = totlen - hdrlen;
1647 	if (caplen < 0)
1648 		caplen = 0;
1649 
1650 	/*
1651 	 * Round up the end of the previous packet to the next longword.
1652 	 */
1653 #ifdef _LP64
1654 	if (d->bd_compat32)
1655 		curlen = BPF_WORDALIGN32(d->bd_slen);
1656 	else
1657 #endif
1658 		curlen = BPF_WORDALIGN(d->bd_slen);
1659 	if (curlen + totlen > d->bd_bufsize) {
1660 		/*
1661 		 * This packet will overflow the storage buffer.
1662 		 * Rotate the buffers if we can, then wakeup any
1663 		 * pending reads.
1664 		 */
1665 		if (d->bd_fbuf == NULL) {
1666 			/*
1667 			 * We haven't completed the previous read yet,
1668 			 * so drop the packet.
1669 			 */
1670 			++d->bd_dcount;
1671 			++bpf_gstats.bs_drop;
1672 			return;
1673 		}
1674 		ROTATE_BUFFERS(d);
1675 		do_wakeup = 1;
1676 		curlen = 0;
1677 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1678 		/*
1679 		 * Immediate mode is set, or the read timeout has
1680 		 * already expired during a select call.  A packet
1681 		 * arrived, so the reader should be woken up.
1682 		 */
1683 		do_wakeup = 1;
1684 	}
1685 
1686 	/*
1687 	 * Append the bpf header.
1688 	 */
1689 	h = (char *)d->bd_sbuf + curlen;
1690 #ifdef _LP64
1691 	if (d->bd_compat32) {
1692 		struct bpf_hdr32 *hp32;
1693 
1694 		hp32 = (struct bpf_hdr32 *)h;
1695 		hp32->bh_tstamp.tv_sec = ts->tv_sec;
1696 		hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1697 		hp32->bh_datalen = pktlen;
1698 		hp32->bh_hdrlen = hdrlen;
1699 		hp32->bh_caplen = caplen;
1700 	} else
1701 #endif
1702 	{
1703 		struct bpf_hdr *hp;
1704 
1705 		hp = (struct bpf_hdr *)h;
1706 		hp->bh_tstamp.tv_sec = ts->tv_sec;
1707 		hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1708 		hp->bh_datalen = pktlen;
1709 		hp->bh_hdrlen = hdrlen;
1710 		hp->bh_caplen = caplen;
1711 	}
1712 
1713 	/*
1714 	 * Copy the packet data into the store buffer and update its length.
1715 	 */
1716 	(*cpfn)(h + hdrlen, pkt, caplen);
1717 	d->bd_slen = curlen + totlen;
1718 
1719 	/*
1720 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1721 	 * will cause filt_bpfread() to be called with it adjusted.
1722 	 */
1723 	if (do_wakeup)
1724 		bpf_wakeup(d);
1725 }
1726 
1727 /*
1728  * Initialize all nonzero fields of a descriptor.
1729  */
1730 static int
1731 bpf_allocbufs(struct bpf_d *d)
1732 {
1733 
1734 	d->bd_fbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1735 	if (!d->bd_fbuf)
1736 		return (ENOBUFS);
1737 	d->bd_sbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1738 	if (!d->bd_sbuf) {
1739 		free(d->bd_fbuf, M_DEVBUF);
1740 		return (ENOBUFS);
1741 	}
1742 	d->bd_slen = 0;
1743 	d->bd_hlen = 0;
1744 	return (0);
1745 }
1746 
1747 /*
1748  * Free buffers currently in use by a descriptor.
1749  * Called on close.
1750  */
1751 static void
1752 bpf_freed(struct bpf_d *d)
1753 {
1754 	/*
1755 	 * We don't need to lock out interrupts since this descriptor has
1756 	 * been detached from its interface and it yet hasn't been marked
1757 	 * free.
1758 	 */
1759 	if (d->bd_sbuf != NULL) {
1760 		free(d->bd_sbuf, M_DEVBUF);
1761 		if (d->bd_hbuf != NULL)
1762 			free(d->bd_hbuf, M_DEVBUF);
1763 		if (d->bd_fbuf != NULL)
1764 			free(d->bd_fbuf, M_DEVBUF);
1765 	}
1766 	if (d->bd_filter)
1767 		free(d->bd_filter, M_DEVBUF);
1768 
1769 	if (d->bd_jitcode != NULL) {
1770 		bpf_jit_freecode(d->bd_jitcode);
1771 	}
1772 }
1773 
1774 /*
1775  * Attach an interface to bpf.  dlt is the link layer type;
1776  * hdrlen is the fixed size of the link header for the specified dlt
1777  * (variable length headers not yet supported).
1778  */
1779 static void
1780 _bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1781 {
1782 	struct bpf_if *bp;
1783 	bp = malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1784 	if (bp == NULL)
1785 		panic("bpfattach");
1786 
1787 	mutex_enter(&bpf_mtx);
1788 	bp->bif_dlist = NULL;
1789 	bp->bif_driverp = driverp;
1790 	bp->bif_ifp = ifp;
1791 	bp->bif_dlt = dlt;
1792 
1793 	bp->bif_next = bpf_iflist;
1794 	bpf_iflist = bp;
1795 
1796 	*bp->bif_driverp = NULL;
1797 
1798 	bp->bif_hdrlen = hdrlen;
1799 	mutex_exit(&bpf_mtx);
1800 #if 0
1801 	printf("bpf: %s attached\n", ifp->if_xname);
1802 #endif
1803 }
1804 
1805 /*
1806  * Remove an interface from bpf.
1807  */
1808 static void
1809 _bpfdetach(struct ifnet *ifp)
1810 {
1811 	struct bpf_if *bp, **pbp;
1812 	struct bpf_d *d;
1813 	int s;
1814 
1815 	mutex_enter(&bpf_mtx);
1816 	/* Nuke the vnodes for any open instances */
1817 	LIST_FOREACH(d, &bpf_list, bd_list) {
1818 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
1819 			/*
1820 			 * Detach the descriptor from an interface now.
1821 			 * It will be free'ed later by close routine.
1822 			 */
1823 			s = splnet();
1824 			d->bd_promisc = 0;	/* we can't touch device. */
1825 			bpf_detachd(d);
1826 			splx(s);
1827 		}
1828 	}
1829 
1830   again:
1831 	for (bp = bpf_iflist, pbp = &bpf_iflist;
1832 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
1833 		if (bp->bif_ifp == ifp) {
1834 			*pbp = bp->bif_next;
1835 			free(bp, M_DEVBUF);
1836 			goto again;
1837 		}
1838 	}
1839 	mutex_exit(&bpf_mtx);
1840 }
1841 
1842 /*
1843  * Change the data link type of a interface.
1844  */
1845 static void
1846 _bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1847 {
1848 	struct bpf_if *bp;
1849 
1850 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1851 		if (bp->bif_driverp == &ifp->if_bpf)
1852 			break;
1853 	}
1854 	if (bp == NULL)
1855 		panic("bpf_change_type");
1856 
1857 	bp->bif_dlt = dlt;
1858 
1859 	bp->bif_hdrlen = hdrlen;
1860 }
1861 
1862 /*
1863  * Get a list of available data link type of the interface.
1864  */
1865 static int
1866 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1867 {
1868 	int n, error;
1869 	struct ifnet *ifp;
1870 	struct bpf_if *bp;
1871 
1872 	ifp = d->bd_bif->bif_ifp;
1873 	n = 0;
1874 	error = 0;
1875 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1876 		if (bp->bif_ifp != ifp)
1877 			continue;
1878 		if (bfl->bfl_list != NULL) {
1879 			if (n >= bfl->bfl_len)
1880 				return ENOMEM;
1881 			error = copyout(&bp->bif_dlt,
1882 			    bfl->bfl_list + n, sizeof(u_int));
1883 		}
1884 		n++;
1885 	}
1886 	bfl->bfl_len = n;
1887 	return error;
1888 }
1889 
1890 /*
1891  * Set the data link type of a BPF instance.
1892  */
1893 static int
1894 bpf_setdlt(struct bpf_d *d, u_int dlt)
1895 {
1896 	int s, error, opromisc;
1897 	struct ifnet *ifp;
1898 	struct bpf_if *bp;
1899 
1900 	KASSERT(mutex_owned(&bpf_mtx));
1901 
1902 	if (d->bd_bif->bif_dlt == dlt)
1903 		return 0;
1904 	ifp = d->bd_bif->bif_ifp;
1905 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1906 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1907 			break;
1908 	}
1909 	if (bp == NULL)
1910 		return EINVAL;
1911 	s = splnet();
1912 	opromisc = d->bd_promisc;
1913 	bpf_detachd(d);
1914 	bpf_attachd(d, bp);
1915 	reset_d(d);
1916 	if (opromisc) {
1917 		error = ifpromisc(bp->bif_ifp, 1);
1918 		if (error)
1919 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
1920 			    bp->bif_ifp->if_xname, error);
1921 		else
1922 			d->bd_promisc = 1;
1923 	}
1924 	splx(s);
1925 	return 0;
1926 }
1927 
1928 static int
1929 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
1930 {
1931 	int newsize, error;
1932 	struct sysctlnode node;
1933 
1934 	node = *rnode;
1935 	node.sysctl_data = &newsize;
1936 	newsize = bpf_maxbufsize;
1937 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1938 	if (error || newp == NULL)
1939 		return (error);
1940 
1941 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
1942 		return (EINVAL);
1943 
1944 	bpf_maxbufsize = newsize;
1945 
1946 	return (0);
1947 }
1948 
1949 #if defined(MODULAR) || defined(BPFJIT)
1950 static int
1951 sysctl_net_bpf_jit(SYSCTLFN_ARGS)
1952 {
1953 	bool newval;
1954 	int error;
1955 	struct sysctlnode node;
1956 
1957 	node = *rnode;
1958 	node.sysctl_data = &newval;
1959 	newval = bpf_jit;
1960 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1961 	if (error != 0 || newp == NULL)
1962 		return error;
1963 
1964 	bpf_jit = newval;
1965 
1966 	/*
1967 	 * Do a full sync to publish new bpf_jit value and
1968 	 * update bpfjit_module_ops.bj_generate_code variable.
1969 	 */
1970 	membar_sync();
1971 
1972 	if (newval && bpfjit_module_ops.bj_generate_code == NULL) {
1973 		printf("JIT compilation is postponed "
1974 		    "until after bpfjit module is loaded\n");
1975 	}
1976 
1977 	return 0;
1978 }
1979 #endif
1980 
1981 static int
1982 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
1983 {
1984 	int    error, elem_count;
1985 	struct bpf_d	 *dp;
1986 	struct bpf_d_ext  dpe;
1987 	size_t len, needed, elem_size, out_size;
1988 	char   *sp;
1989 
1990 	if (namelen == 1 && name[0] == CTL_QUERY)
1991 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
1992 
1993 	if (namelen != 2)
1994 		return (EINVAL);
1995 
1996 	/* BPF peers is privileged information. */
1997 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
1998 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
1999 	if (error)
2000 		return (EPERM);
2001 
2002 	len = (oldp != NULL) ? *oldlenp : 0;
2003 	sp = oldp;
2004 	elem_size = name[0];
2005 	elem_count = name[1];
2006 	out_size = MIN(sizeof(dpe), elem_size);
2007 	needed = 0;
2008 
2009 	if (elem_size < 1 || elem_count < 0)
2010 		return (EINVAL);
2011 
2012 	mutex_enter(&bpf_mtx);
2013 	LIST_FOREACH(dp, &bpf_list, bd_list) {
2014 		if (len >= elem_size && elem_count > 0) {
2015 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
2016 			BPF_EXT(bufsize);
2017 			BPF_EXT(promisc);
2018 			BPF_EXT(state);
2019 			BPF_EXT(immediate);
2020 			BPF_EXT(hdrcmplt);
2021 			BPF_EXT(seesent);
2022 			BPF_EXT(pid);
2023 			BPF_EXT(rcount);
2024 			BPF_EXT(dcount);
2025 			BPF_EXT(ccount);
2026 #undef BPF_EXT
2027 			if (dp->bd_bif)
2028 				(void)strlcpy(dpe.bde_ifname,
2029 				    dp->bd_bif->bif_ifp->if_xname,
2030 				    IFNAMSIZ - 1);
2031 			else
2032 				dpe.bde_ifname[0] = '\0';
2033 
2034 			error = copyout(&dpe, sp, out_size);
2035 			if (error)
2036 				break;
2037 			sp += elem_size;
2038 			len -= elem_size;
2039 		}
2040 		needed += elem_size;
2041 		if (elem_count > 0 && elem_count != INT_MAX)
2042 			elem_count--;
2043 	}
2044 	mutex_exit(&bpf_mtx);
2045 
2046 	*oldlenp = needed;
2047 
2048 	return (error);
2049 }
2050 
2051 static struct sysctllog *bpf_sysctllog;
2052 static void
2053 sysctl_net_bpf_setup(void)
2054 {
2055 	const struct sysctlnode *node;
2056 
2057 	node = NULL;
2058 	sysctl_createv(&bpf_sysctllog, 0, NULL, &node,
2059 		       CTLFLAG_PERMANENT,
2060 		       CTLTYPE_NODE, "bpf",
2061 		       SYSCTL_DESCR("BPF options"),
2062 		       NULL, 0, NULL, 0,
2063 		       CTL_NET, CTL_CREATE, CTL_EOL);
2064 	if (node != NULL) {
2065 #if defined(MODULAR) || defined(BPFJIT)
2066 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2067 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2068 			CTLTYPE_BOOL, "jit",
2069 			SYSCTL_DESCR("Toggle Just-In-Time compilation"),
2070 			sysctl_net_bpf_jit, 0, &bpf_jit, 0,
2071 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2072 #endif
2073 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2074 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2075 			CTLTYPE_INT, "maxbufsize",
2076 			SYSCTL_DESCR("Maximum size for data capture buffer"),
2077 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
2078 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2079 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2080 			CTLFLAG_PERMANENT,
2081 			CTLTYPE_STRUCT, "stats",
2082 			SYSCTL_DESCR("BPF stats"),
2083 			NULL, 0, &bpf_gstats, sizeof(bpf_gstats),
2084 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2085 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2086 			CTLFLAG_PERMANENT,
2087 			CTLTYPE_STRUCT, "peers",
2088 			SYSCTL_DESCR("BPF peers"),
2089 			sysctl_net_bpf_peers, 0, NULL, 0,
2090 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2091 	}
2092 
2093 }
2094 
2095 struct bpf_ops bpf_ops_kernel = {
2096 	.bpf_attach =		_bpfattach,
2097 	.bpf_detach =		_bpfdetach,
2098 	.bpf_change_type =	_bpf_change_type,
2099 
2100 	.bpf_tap =		_bpf_tap,
2101 	.bpf_mtap =		_bpf_mtap,
2102 	.bpf_mtap2 =		_bpf_mtap2,
2103 	.bpf_mtap_af =		_bpf_mtap_af,
2104 	.bpf_mtap_sl_in =	_bpf_mtap_sl_in,
2105 	.bpf_mtap_sl_out =	_bpf_mtap_sl_out,
2106 };
2107 
2108 MODULE(MODULE_CLASS_DRIVER, bpf, "bpf_filter");
2109 
2110 static int
2111 bpf_modcmd(modcmd_t cmd, void *arg)
2112 {
2113 #ifdef _MODULE
2114 	devmajor_t bmajor, cmajor;
2115 #endif
2116 	int error = 0;
2117 
2118 	switch (cmd) {
2119 	case MODULE_CMD_INIT:
2120 		bpf_init();
2121 #ifdef _MODULE
2122 		bmajor = cmajor = NODEVMAJOR;
2123 		error = devsw_attach("bpf", NULL, &bmajor,
2124 		    &bpf_cdevsw, &cmajor);
2125 		if (error)
2126 			break;
2127 #endif
2128 
2129 		bpf_ops_handover_enter(&bpf_ops_kernel);
2130 		atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel);
2131 		bpf_ops_handover_exit();
2132 		sysctl_net_bpf_setup();
2133 		break;
2134 
2135 	case MODULE_CMD_FINI:
2136 		/*
2137 		 * While there is no reference counting for bpf callers,
2138 		 * unload could at least in theory be done similarly to
2139 		 * system call disestablishment.  This should even be
2140 		 * a little simpler:
2141 		 *
2142 		 * 1) replace op vector with stubs
2143 		 * 2) post update to all cpus with xc
2144 		 * 3) check that nobody is in bpf anymore
2145 		 *    (it's doubtful we'd want something like l_sysent,
2146 		 *     but we could do something like *signed* percpu
2147 		 *     counters.  if the sum is 0, we're good).
2148 		 * 4) if fail, unroll changes
2149 		 *
2150 		 * NOTE: change won't be atomic to the outside.  some
2151 		 * packets may be not captured even if unload is
2152 		 * not succesful.  I think packet capture not working
2153 		 * is a perfectly logical consequence of trying to
2154 		 * disable packet capture.
2155 		 */
2156 		error = EOPNOTSUPP;
2157 		/* insert sysctl teardown */
2158 		break;
2159 
2160 	default:
2161 		error = ENOTTY;
2162 		break;
2163 	}
2164 
2165 	return error;
2166 }
2167