xref: /netbsd-src/sys/net/bpf.c (revision 4391d5e9d4f291db41e3b3ba26a01b5e51364aae)
1 /*	$NetBSD: bpf.c,v 1.179 2013/11/16 01:13:52 rmind 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 1.179 2013/11/16 01:13:52 rmind 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/once.h>
63 #include <sys/atomic.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 
78 #include <net/if.h>
79 #include <net/slip.h>
80 
81 #include <net/bpf.h>
82 #include <net/bpfdesc.h>
83 #include <net/bpfjit.h>
84 
85 #include <net/if_arc.h>
86 #include <net/if_ether.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/if_inarp.h>
90 
91 
92 #include <compat/sys/sockio.h>
93 
94 #ifndef BPF_BUFSIZE
95 /*
96  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
97  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
98  */
99 # define BPF_BUFSIZE 32768
100 #endif
101 
102 #define PRINET  26			/* interruptible */
103 
104 /*
105  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
106  * XXX the default values should be computed dynamically based
107  * on available memory size and available mbuf clusters.
108  */
109 int bpf_bufsize = BPF_BUFSIZE;
110 int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
111 bool bpf_jit = false;
112 
113 struct bpfjit_ops bpfjit_module_ops = {
114 	.bj_generate_code = NULL,
115 	.bj_free_code = NULL
116 };
117 
118 /*
119  * Global BPF statistics returned by net.bpf.stats sysctl.
120  */
121 struct bpf_stat	bpf_gstats;
122 
123 /*
124  * Use a mutex to avoid a race condition between gathering the stats/peers
125  * and opening/closing the device.
126  */
127 static kmutex_t bpf_mtx;
128 
129 /*
130  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
131  *  bpf_dtab holds the descriptors, indexed by minor device #
132  */
133 struct bpf_if	*bpf_iflist;
134 LIST_HEAD(, bpf_d) bpf_list;
135 
136 static int	bpf_allocbufs(struct bpf_d *);
137 static void	bpf_deliver(struct bpf_if *,
138 		            void *(*cpfn)(void *, const void *, size_t),
139 		            void *, u_int, u_int, const bool);
140 static void	bpf_freed(struct bpf_d *);
141 static void	bpf_ifname(struct ifnet *, struct ifreq *);
142 static void	*bpf_mcpy(void *, const void *, size_t);
143 static int	bpf_movein(struct uio *, int, uint64_t,
144 			        struct mbuf **, struct sockaddr *);
145 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
146 static void	bpf_detachd(struct bpf_d *);
147 static int	bpf_setif(struct bpf_d *, struct ifreq *);
148 static void	bpf_timed_out(void *);
149 static inline void
150 		bpf_wakeup(struct bpf_d *);
151 static int	bpf_hdrlen(struct bpf_d *);
152 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
153     void *(*)(void *, const void *, size_t), struct timespec *);
154 static void	reset_d(struct bpf_d *);
155 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
156 static int	bpf_setdlt(struct bpf_d *, u_int);
157 
158 static int	bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t,
159     int);
160 static int	bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t,
161     int);
162 static int	bpf_ioctl(struct file *, u_long, void *);
163 static int	bpf_poll(struct file *, int);
164 static int	bpf_stat(struct file *, struct stat *);
165 static int	bpf_close(struct file *);
166 static int	bpf_kqfilter(struct file *, struct knote *);
167 static void	bpf_softintr(void *);
168 
169 static const struct fileops bpf_fileops = {
170 	.fo_read = bpf_read,
171 	.fo_write = bpf_write,
172 	.fo_ioctl = bpf_ioctl,
173 	.fo_fcntl = fnullop_fcntl,
174 	.fo_poll = bpf_poll,
175 	.fo_stat = bpf_stat,
176 	.fo_close = bpf_close,
177 	.fo_kqfilter = bpf_kqfilter,
178 	.fo_restart = fnullop_restart,
179 };
180 
181 dev_type_open(bpfopen);
182 
183 const struct cdevsw bpf_cdevsw = {
184 	bpfopen, noclose, noread, nowrite, noioctl,
185 	nostop, notty, nopoll, nommap, nokqfilter, D_OTHER
186 };
187 
188 bpfjit_func_t
189 bpf_jit_generate(bpf_ctx_t *bc, void *code, size_t size)
190 {
191 	membar_consumer();
192 	if (bpfjit_module_ops.bj_generate_code != NULL) {
193 		return bpfjit_module_ops.bj_generate_code(bc, code, size);
194 	}
195 	return NULL;
196 }
197 
198 void
199 bpf_jit_freecode(bpfjit_func_t jcode)
200 {
201 	KASSERT(bpfjit_module_ops.bj_free_code != NULL);
202 	bpfjit_module_ops.bj_free_code(jcode);
203 }
204 
205 static int
206 bpf_movein(struct uio *uio, int linktype, uint64_t mtu, struct mbuf **mp,
207 	   struct sockaddr *sockp)
208 {
209 	struct mbuf *m;
210 	int error;
211 	size_t len;
212 	size_t hlen;
213 	size_t align;
214 
215 	/*
216 	 * Build a sockaddr based on the data link layer type.
217 	 * We do this at this level because the ethernet header
218 	 * is copied directly into the data field of the sockaddr.
219 	 * In the case of SLIP, there is no header and the packet
220 	 * is forwarded as is.
221 	 * Also, we are careful to leave room at the front of the mbuf
222 	 * for the link level header.
223 	 */
224 	switch (linktype) {
225 
226 	case DLT_SLIP:
227 		sockp->sa_family = AF_INET;
228 		hlen = 0;
229 		align = 0;
230 		break;
231 
232 	case DLT_PPP:
233 		sockp->sa_family = AF_UNSPEC;
234 		hlen = 0;
235 		align = 0;
236 		break;
237 
238 	case DLT_EN10MB:
239 		sockp->sa_family = AF_UNSPEC;
240 		/* XXX Would MAXLINKHDR be better? */
241  		/* 6(dst)+6(src)+2(type) */
242 		hlen = sizeof(struct ether_header);
243 		align = 2;
244 		break;
245 
246 	case DLT_ARCNET:
247 		sockp->sa_family = AF_UNSPEC;
248 		hlen = ARC_HDRLEN;
249 		align = 5;
250 		break;
251 
252 	case DLT_FDDI:
253 		sockp->sa_family = AF_LINK;
254 		/* XXX 4(FORMAC)+6(dst)+6(src) */
255 		hlen = 16;
256 		align = 0;
257 		break;
258 
259 	case DLT_ECONET:
260 		sockp->sa_family = AF_UNSPEC;
261 		hlen = 6;
262 		align = 2;
263 		break;
264 
265 	case DLT_NULL:
266 		sockp->sa_family = AF_UNSPEC;
267 		hlen = 0;
268 		align = 0;
269 		break;
270 
271 	default:
272 		return (EIO);
273 	}
274 
275 	len = uio->uio_resid;
276 	/*
277 	 * If there aren't enough bytes for a link level header or the
278 	 * packet length exceeds the interface mtu, return an error.
279 	 */
280 	if (len - hlen > mtu)
281 		return (EMSGSIZE);
282 
283 	/*
284 	 * XXX Avoid complicated buffer chaining ---
285 	 * bail if it won't fit in a single mbuf.
286 	 * (Take into account possible alignment bytes)
287 	 */
288 	if (len + align > MCLBYTES)
289 		return (EIO);
290 
291 	m = m_gethdr(M_WAIT, MT_DATA);
292 	m->m_pkthdr.rcvif = 0;
293 	m->m_pkthdr.len = (int)(len - hlen);
294 	if (len + align > MHLEN) {
295 		m_clget(m, M_WAIT);
296 		if ((m->m_flags & M_EXT) == 0) {
297 			error = ENOBUFS;
298 			goto bad;
299 		}
300 	}
301 
302 	/* Insure the data is properly aligned */
303 	if (align > 0) {
304 		m->m_data += align;
305 		m->m_len -= (int)align;
306 	}
307 
308 	error = uiomove(mtod(m, void *), len, uio);
309 	if (error)
310 		goto bad;
311 	if (hlen != 0) {
312 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
313 		m->m_data += hlen; /* XXX */
314 		len -= hlen;
315 	}
316 	m->m_len = (int)len;
317 	*mp = m;
318 	return (0);
319 
320 bad:
321 	m_freem(m);
322 	return (error);
323 }
324 
325 /*
326  * Attach file to the bpf interface, i.e. make d listen on bp.
327  * Must be called at splnet.
328  */
329 static void
330 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
331 {
332 	/*
333 	 * Point d at bp, and add d to the interface's list of listeners.
334 	 * Finally, point the driver's bpf cookie at the interface so
335 	 * it will divert packets to bpf.
336 	 */
337 	d->bd_bif = bp;
338 	d->bd_next = bp->bif_dlist;
339 	bp->bif_dlist = d;
340 
341 	*bp->bif_driverp = bp;
342 }
343 
344 /*
345  * Detach a file from its interface.
346  */
347 static void
348 bpf_detachd(struct bpf_d *d)
349 {
350 	struct bpf_d **p;
351 	struct bpf_if *bp;
352 
353 	bp = d->bd_bif;
354 	/*
355 	 * Check if this descriptor had requested promiscuous mode.
356 	 * If so, turn it off.
357 	 */
358 	if (d->bd_promisc) {
359 		int error;
360 
361 		d->bd_promisc = 0;
362 		/*
363 		 * Take device out of promiscuous mode.  Since we were
364 		 * able to enter promiscuous mode, we should be able
365 		 * to turn it off.  But we can get an error if
366 		 * the interface was configured down, so only panic
367 		 * if we don't get an unexpected error.
368 		 */
369   		error = ifpromisc(bp->bif_ifp, 0);
370 		if (error && error != EINVAL)
371 			panic("%s: ifpromisc failed: %d", __func__, error);
372 	}
373 	/* Remove d from the interface's descriptor list. */
374 	p = &bp->bif_dlist;
375 	while (*p != d) {
376 		p = &(*p)->bd_next;
377 		if (*p == 0)
378 			panic("%s: descriptor not in list", __func__);
379 	}
380 	*p = (*p)->bd_next;
381 	if (bp->bif_dlist == 0)
382 		/*
383 		 * Let the driver know that there are no more listeners.
384 		 */
385 		*d->bd_bif->bif_driverp = 0;
386 	d->bd_bif = 0;
387 }
388 
389 static int
390 doinit(void)
391 {
392 
393 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
394 
395 	LIST_INIT(&bpf_list);
396 
397 	bpf_gstats.bs_recv = 0;
398 	bpf_gstats.bs_drop = 0;
399 	bpf_gstats.bs_capt = 0;
400 
401 	return 0;
402 }
403 
404 /*
405  * bpfilterattach() is called at boot time.
406  */
407 /* ARGSUSED */
408 void
409 bpfilterattach(int n)
410 {
411 	static ONCE_DECL(control);
412 
413 	RUN_ONCE(&control, doinit);
414 }
415 
416 /*
417  * Open ethernet device. Clones.
418  */
419 /* ARGSUSED */
420 int
421 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
422 {
423 	struct bpf_d *d;
424 	struct file *fp;
425 	int error, fd;
426 
427 	/* falloc() will use the descriptor for us. */
428 	if ((error = fd_allocfile(&fp, &fd)) != 0)
429 		return error;
430 
431 	d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK|M_ZERO);
432 	d->bd_bufsize = bpf_bufsize;
433 	d->bd_seesent = 1;
434 	d->bd_feedback = 0;
435 	d->bd_pid = l->l_proc->p_pid;
436 #ifdef _LP64
437 	if (curproc->p_flag & PK_32)
438 		d->bd_compat32 = 1;
439 #endif
440 	getnanotime(&d->bd_btime);
441 	d->bd_atime = d->bd_mtime = d->bd_btime;
442 	callout_init(&d->bd_callout, 0);
443 	selinit(&d->bd_sel);
444 	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
445 	d->bd_jitcode = NULL;
446 
447 	mutex_enter(&bpf_mtx);
448 	LIST_INSERT_HEAD(&bpf_list, d, bd_list);
449 	mutex_exit(&bpf_mtx);
450 
451 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
452 }
453 
454 /*
455  * Close the descriptor by detaching it from its interface,
456  * deallocating its buffers, and marking it free.
457  */
458 /* ARGSUSED */
459 static int
460 bpf_close(struct file *fp)
461 {
462 	struct bpf_d *d = fp->f_data;
463 	int s;
464 
465 	KERNEL_LOCK(1, NULL);
466 
467 	/*
468 	 * Refresh the PID associated with this bpf file.
469 	 */
470 	d->bd_pid = curproc->p_pid;
471 
472 	s = splnet();
473 	if (d->bd_state == BPF_WAITING)
474 		callout_stop(&d->bd_callout);
475 	d->bd_state = BPF_IDLE;
476 	if (d->bd_bif)
477 		bpf_detachd(d);
478 	splx(s);
479 	bpf_freed(d);
480 	mutex_enter(&bpf_mtx);
481 	LIST_REMOVE(d, bd_list);
482 	mutex_exit(&bpf_mtx);
483 	callout_destroy(&d->bd_callout);
484 	seldestroy(&d->bd_sel);
485 	softint_disestablish(d->bd_sih);
486 	free(d, M_DEVBUF);
487 	fp->f_data = NULL;
488 
489 	KERNEL_UNLOCK_ONE(NULL);
490 
491 	return (0);
492 }
493 
494 /*
495  * Rotate the packet buffers in descriptor d.  Move the store buffer
496  * into the hold slot, and the free buffer into the store slot.
497  * Zero the length of the new store buffer.
498  */
499 #define ROTATE_BUFFERS(d) \
500 	(d)->bd_hbuf = (d)->bd_sbuf; \
501 	(d)->bd_hlen = (d)->bd_slen; \
502 	(d)->bd_sbuf = (d)->bd_fbuf; \
503 	(d)->bd_slen = 0; \
504 	(d)->bd_fbuf = 0;
505 /*
506  *  bpfread - read next chunk of packets from buffers
507  */
508 static int
509 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
510     kauth_cred_t cred, int flags)
511 {
512 	struct bpf_d *d = fp->f_data;
513 	int timed_out;
514 	int error;
515 	int s;
516 
517 	getnanotime(&d->bd_atime);
518 	/*
519 	 * Restrict application to use a buffer the same size as
520 	 * the kernel buffers.
521 	 */
522 	if (uio->uio_resid != d->bd_bufsize)
523 		return (EINVAL);
524 
525 	KERNEL_LOCK(1, NULL);
526 	s = splnet();
527 	if (d->bd_state == BPF_WAITING)
528 		callout_stop(&d->bd_callout);
529 	timed_out = (d->bd_state == BPF_TIMED_OUT);
530 	d->bd_state = BPF_IDLE;
531 	/*
532 	 * If the hold buffer is empty, then do a timed sleep, which
533 	 * ends when the timeout expires or when enough packets
534 	 * have arrived to fill the store buffer.
535 	 */
536 	while (d->bd_hbuf == 0) {
537 		if (fp->f_flag & FNONBLOCK) {
538 			if (d->bd_slen == 0) {
539 				splx(s);
540 				KERNEL_UNLOCK_ONE(NULL);
541 				return (EWOULDBLOCK);
542 			}
543 			ROTATE_BUFFERS(d);
544 			break;
545 		}
546 
547 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
548 			/*
549 			 * A packet(s) either arrived since the previous
550 			 * read or arrived while we were asleep.
551 			 * Rotate the buffers and return what's here.
552 			 */
553 			ROTATE_BUFFERS(d);
554 			break;
555 		}
556 		error = tsleep(d, PRINET|PCATCH, "bpf",
557 				d->bd_rtout);
558 		if (error == EINTR || error == ERESTART) {
559 			splx(s);
560 			KERNEL_UNLOCK_ONE(NULL);
561 			return (error);
562 		}
563 		if (error == EWOULDBLOCK) {
564 			/*
565 			 * On a timeout, return what's in the buffer,
566 			 * which may be nothing.  If there is something
567 			 * in the store buffer, we can rotate the buffers.
568 			 */
569 			if (d->bd_hbuf)
570 				/*
571 				 * We filled up the buffer in between
572 				 * getting the timeout and arriving
573 				 * here, so we don't need to rotate.
574 				 */
575 				break;
576 
577 			if (d->bd_slen == 0) {
578 				splx(s);
579 				KERNEL_UNLOCK_ONE(NULL);
580 				return (0);
581 			}
582 			ROTATE_BUFFERS(d);
583 			break;
584 		}
585 		if (error != 0)
586 			goto done;
587 	}
588 	/*
589 	 * At this point, we know we have something in the hold slot.
590 	 */
591 	splx(s);
592 
593 	/*
594 	 * Move data from hold buffer into user space.
595 	 * We know the entire buffer is transferred since
596 	 * we checked above that the read buffer is bpf_bufsize bytes.
597 	 */
598 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
599 
600 	s = splnet();
601 	d->bd_fbuf = d->bd_hbuf;
602 	d->bd_hbuf = 0;
603 	d->bd_hlen = 0;
604 done:
605 	splx(s);
606 	KERNEL_UNLOCK_ONE(NULL);
607 	return (error);
608 }
609 
610 
611 /*
612  * If there are processes sleeping on this descriptor, wake them up.
613  */
614 static inline void
615 bpf_wakeup(struct bpf_d *d)
616 {
617 	wakeup(d);
618 	if (d->bd_async)
619 		softint_schedule(d->bd_sih);
620 	selnotify(&d->bd_sel, 0, 0);
621 }
622 
623 static void
624 bpf_softintr(void *cookie)
625 {
626 	struct bpf_d *d;
627 
628 	d = cookie;
629 	if (d->bd_async)
630 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
631 }
632 
633 static void
634 bpf_timed_out(void *arg)
635 {
636 	struct bpf_d *d = arg;
637 	int s;
638 
639 	s = splnet();
640 	if (d->bd_state == BPF_WAITING) {
641 		d->bd_state = BPF_TIMED_OUT;
642 		if (d->bd_slen != 0)
643 			bpf_wakeup(d);
644 	}
645 	splx(s);
646 }
647 
648 
649 static int
650 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
651     kauth_cred_t cred, int flags)
652 {
653 	struct bpf_d *d = fp->f_data;
654 	struct ifnet *ifp;
655 	struct mbuf *m, *mc;
656 	int error, s;
657 	static struct sockaddr_storage dst;
658 
659 	m = NULL;	/* XXX gcc */
660 
661 	KERNEL_LOCK(1, NULL);
662 
663 	if (d->bd_bif == 0) {
664 		KERNEL_UNLOCK_ONE(NULL);
665 		return (ENXIO);
666 	}
667 	getnanotime(&d->bd_mtime);
668 
669 	ifp = d->bd_bif->bif_ifp;
670 
671 	if (uio->uio_resid == 0) {
672 		KERNEL_UNLOCK_ONE(NULL);
673 		return (0);
674 	}
675 
676 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
677 		(struct sockaddr *) &dst);
678 	if (error) {
679 		KERNEL_UNLOCK_ONE(NULL);
680 		return (error);
681 	}
682 
683 	if (m->m_pkthdr.len > ifp->if_mtu) {
684 		KERNEL_UNLOCK_ONE(NULL);
685 		m_freem(m);
686 		return (EMSGSIZE);
687 	}
688 
689 	if (d->bd_hdrcmplt)
690 		dst.ss_family = pseudo_AF_HDRCMPLT;
691 
692 	if (d->bd_feedback) {
693 		mc = m_dup(m, 0, M_COPYALL, M_NOWAIT);
694 		if (mc != NULL)
695 			mc->m_pkthdr.rcvif = ifp;
696 		/* Set M_PROMISC for outgoing packets to be discarded. */
697 		if (1 /*d->bd_direction == BPF_D_INOUT*/)
698 			m->m_flags |= M_PROMISC;
699 	} else
700 		mc = NULL;
701 
702 	s = splsoftnet();
703 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *) &dst, NULL);
704 
705 	if (mc != NULL) {
706 		if (error == 0)
707 			(*ifp->if_input)(ifp, mc);
708 		m_freem(mc);
709 	}
710 	splx(s);
711 	KERNEL_UNLOCK_ONE(NULL);
712 	/*
713 	 * The driver frees the mbuf.
714 	 */
715 	return (error);
716 }
717 
718 /*
719  * Reset a descriptor by flushing its packet buffer and clearing the
720  * receive and drop counts.  Should be called at splnet.
721  */
722 static void
723 reset_d(struct bpf_d *d)
724 {
725 	if (d->bd_hbuf) {
726 		/* Free the hold buffer. */
727 		d->bd_fbuf = d->bd_hbuf;
728 		d->bd_hbuf = 0;
729 	}
730 	d->bd_slen = 0;
731 	d->bd_hlen = 0;
732 	d->bd_rcount = 0;
733 	d->bd_dcount = 0;
734 	d->bd_ccount = 0;
735 }
736 
737 /*
738  *  FIONREAD		Check for read packet available.
739  *  BIOCGBLEN		Get buffer len [for read()].
740  *  BIOCSETF		Set ethernet read filter.
741  *  BIOCFLUSH		Flush read packet buffer.
742  *  BIOCPROMISC		Put interface into promiscuous mode.
743  *  BIOCGDLT		Get link layer type.
744  *  BIOCGETIF		Get interface name.
745  *  BIOCSETIF		Set interface.
746  *  BIOCSRTIMEOUT	Set read timeout.
747  *  BIOCGRTIMEOUT	Get read timeout.
748  *  BIOCGSTATS		Get packet stats.
749  *  BIOCIMMEDIATE	Set immediate mode.
750  *  BIOCVERSION		Get filter language version.
751  *  BIOCGHDRCMPLT	Get "header already complete" flag.
752  *  BIOCSHDRCMPLT	Set "header already complete" flag.
753  *  BIOCSFEEDBACK	Set packet feedback mode.
754  *  BIOCGFEEDBACK	Get packet feedback mode.
755  *  BIOCGSEESENT  	Get "see sent packets" mode.
756  *  BIOCSSEESENT  	Set "see sent packets" mode.
757  */
758 /* ARGSUSED */
759 static int
760 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
761 {
762 	struct bpf_d *d = fp->f_data;
763 	int s, error = 0;
764 
765 	/*
766 	 * Refresh the PID associated with this bpf file.
767 	 */
768 	KERNEL_LOCK(1, NULL);
769 	d->bd_pid = curproc->p_pid;
770 #ifdef _LP64
771 	if (curproc->p_flag & PK_32)
772 		d->bd_compat32 = 1;
773 	else
774 		d->bd_compat32 = 0;
775 #endif
776 
777 	s = splnet();
778 	if (d->bd_state == BPF_WAITING)
779 		callout_stop(&d->bd_callout);
780 	d->bd_state = BPF_IDLE;
781 	splx(s);
782 
783 	switch (cmd) {
784 
785 	default:
786 		error = EINVAL;
787 		break;
788 
789 	/*
790 	 * Check for read packet available.
791 	 */
792 	case FIONREAD:
793 		{
794 			int n;
795 
796 			s = splnet();
797 			n = d->bd_slen;
798 			if (d->bd_hbuf)
799 				n += d->bd_hlen;
800 			splx(s);
801 
802 			*(int *)addr = n;
803 			break;
804 		}
805 
806 	/*
807 	 * Get buffer len [for read()].
808 	 */
809 	case BIOCGBLEN:
810 		*(u_int *)addr = d->bd_bufsize;
811 		break;
812 
813 	/*
814 	 * Set buffer length.
815 	 */
816 	case BIOCSBLEN:
817 		if (d->bd_bif != 0)
818 			error = EINVAL;
819 		else {
820 			u_int size = *(u_int *)addr;
821 
822 			if (size > bpf_maxbufsize)
823 				*(u_int *)addr = size = bpf_maxbufsize;
824 			else if (size < BPF_MINBUFSIZE)
825 				*(u_int *)addr = size = BPF_MINBUFSIZE;
826 			d->bd_bufsize = size;
827 		}
828 		break;
829 
830 	/*
831 	 * Set link layer read filter.
832 	 */
833 	case BIOCSETF:
834 		error = bpf_setf(d, addr);
835 		break;
836 
837 	/*
838 	 * Flush read packet buffer.
839 	 */
840 	case BIOCFLUSH:
841 		s = splnet();
842 		reset_d(d);
843 		splx(s);
844 		break;
845 
846 	/*
847 	 * Put interface into promiscuous mode.
848 	 */
849 	case BIOCPROMISC:
850 		if (d->bd_bif == 0) {
851 			/*
852 			 * No interface attached yet.
853 			 */
854 			error = EINVAL;
855 			break;
856 		}
857 		s = splnet();
858 		if (d->bd_promisc == 0) {
859 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
860 			if (error == 0)
861 				d->bd_promisc = 1;
862 		}
863 		splx(s);
864 		break;
865 
866 	/*
867 	 * Get device parameters.
868 	 */
869 	case BIOCGDLT:
870 		if (d->bd_bif == 0)
871 			error = EINVAL;
872 		else
873 			*(u_int *)addr = d->bd_bif->bif_dlt;
874 		break;
875 
876 	/*
877 	 * Get a list of supported device parameters.
878 	 */
879 	case BIOCGDLTLIST:
880 		if (d->bd_bif == 0)
881 			error = EINVAL;
882 		else
883 			error = bpf_getdltlist(d, addr);
884 		break;
885 
886 	/*
887 	 * Set device parameters.
888 	 */
889 	case BIOCSDLT:
890 		if (d->bd_bif == 0)
891 			error = EINVAL;
892 		else
893 			error = bpf_setdlt(d, *(u_int *)addr);
894 		break;
895 
896 	/*
897 	 * Set interface name.
898 	 */
899 #ifdef OBIOCGETIF
900 	case OBIOCGETIF:
901 #endif
902 	case BIOCGETIF:
903 		if (d->bd_bif == 0)
904 			error = EINVAL;
905 		else
906 			bpf_ifname(d->bd_bif->bif_ifp, addr);
907 		break;
908 
909 	/*
910 	 * Set interface.
911 	 */
912 #ifdef OBIOCSETIF
913 	case OBIOCSETIF:
914 #endif
915 	case BIOCSETIF:
916 		error = bpf_setif(d, addr);
917 		break;
918 
919 	/*
920 	 * Set read timeout.
921 	 */
922 	case BIOCSRTIMEOUT:
923 		{
924 			struct timeval *tv = addr;
925 
926 			/* Compute number of ticks. */
927 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
928 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
929 				d->bd_rtout = 1;
930 			break;
931 		}
932 
933 #ifdef BIOCGORTIMEOUT
934 	/*
935 	 * Get read timeout.
936 	 */
937 	case BIOCGORTIMEOUT:
938 		{
939 			struct timeval50 *tv = addr;
940 
941 			tv->tv_sec = d->bd_rtout / hz;
942 			tv->tv_usec = (d->bd_rtout % hz) * tick;
943 			break;
944 		}
945 #endif
946 
947 #ifdef BIOCSORTIMEOUT
948 	/*
949 	 * Set read timeout.
950 	 */
951 	case BIOCSORTIMEOUT:
952 		{
953 			struct timeval50 *tv = addr;
954 
955 			/* Compute number of ticks. */
956 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
957 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
958 				d->bd_rtout = 1;
959 			break;
960 		}
961 #endif
962 
963 	/*
964 	 * Get read timeout.
965 	 */
966 	case BIOCGRTIMEOUT:
967 		{
968 			struct timeval *tv = addr;
969 
970 			tv->tv_sec = d->bd_rtout / hz;
971 			tv->tv_usec = (d->bd_rtout % hz) * tick;
972 			break;
973 		}
974 	/*
975 	 * Get packet stats.
976 	 */
977 	case BIOCGSTATS:
978 		{
979 			struct bpf_stat *bs = addr;
980 
981 			bs->bs_recv = d->bd_rcount;
982 			bs->bs_drop = d->bd_dcount;
983 			bs->bs_capt = d->bd_ccount;
984 			break;
985 		}
986 
987 	case BIOCGSTATSOLD:
988 		{
989 			struct bpf_stat_old *bs = addr;
990 
991 			bs->bs_recv = d->bd_rcount;
992 			bs->bs_drop = d->bd_dcount;
993 			break;
994 		}
995 
996 	/*
997 	 * Set immediate mode.
998 	 */
999 	case BIOCIMMEDIATE:
1000 		d->bd_immediate = *(u_int *)addr;
1001 		break;
1002 
1003 	case BIOCVERSION:
1004 		{
1005 			struct bpf_version *bv = addr;
1006 
1007 			bv->bv_major = BPF_MAJOR_VERSION;
1008 			bv->bv_minor = BPF_MINOR_VERSION;
1009 			break;
1010 		}
1011 
1012 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
1013 		*(u_int *)addr = d->bd_hdrcmplt;
1014 		break;
1015 
1016 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
1017 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1018 		break;
1019 
1020 	/*
1021 	 * Get "see sent packets" flag
1022 	 */
1023 	case BIOCGSEESENT:
1024 		*(u_int *)addr = d->bd_seesent;
1025 		break;
1026 
1027 	/*
1028 	 * Set "see sent" packets flag
1029 	 */
1030 	case BIOCSSEESENT:
1031 		d->bd_seesent = *(u_int *)addr;
1032 		break;
1033 
1034 	/*
1035 	 * Set "feed packets from bpf back to input" mode
1036 	 */
1037 	case BIOCSFEEDBACK:
1038 		d->bd_feedback = *(u_int *)addr;
1039 		break;
1040 
1041 	/*
1042 	 * Get "feed packets from bpf back to input" mode
1043 	 */
1044 	case BIOCGFEEDBACK:
1045 		*(u_int *)addr = d->bd_feedback;
1046 		break;
1047 
1048 	case FIONBIO:		/* Non-blocking I/O */
1049 		/*
1050 		 * No need to do anything special as we use IO_NDELAY in
1051 		 * bpfread() as an indication of whether or not to block
1052 		 * the read.
1053 		 */
1054 		break;
1055 
1056 	case FIOASYNC:		/* Send signal on receive packets */
1057 		d->bd_async = *(int *)addr;
1058 		break;
1059 
1060 	case TIOCSPGRP:		/* Process or group to send signals to */
1061 	case FIOSETOWN:
1062 		error = fsetown(&d->bd_pgid, cmd, addr);
1063 		break;
1064 
1065 	case TIOCGPGRP:
1066 	case FIOGETOWN:
1067 		error = fgetown(d->bd_pgid, cmd, addr);
1068 		break;
1069 	}
1070 	KERNEL_UNLOCK_ONE(NULL);
1071 	return (error);
1072 }
1073 
1074 /*
1075  * Set d's packet filter program to fp.  If this file already has a filter,
1076  * free it and replace it.  Returns EINVAL for bogus requests.
1077  */
1078 int
1079 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
1080 {
1081 	struct bpf_insn *fcode, *old;
1082 	bpfjit_func_t jcode, oldj;
1083 	size_t flen, size;
1084 	int s;
1085 
1086 	jcode = NULL;
1087 	flen = fp->bf_len;
1088 
1089 	if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) {
1090 		return EINVAL;
1091 	}
1092 
1093 	if (flen) {
1094 		/*
1095 		 * Allocate the buffer, copy the byte-code from
1096 		 * userspace and validate it.
1097 		 */
1098 		size = flen * sizeof(*fp->bf_insns);
1099 		fcode = malloc(size, M_DEVBUF, M_WAITOK);
1100 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1101 		    !bpf_validate(fcode, (int)flen)) {
1102 			free(fcode, M_DEVBUF);
1103 			return EINVAL;
1104 		}
1105 		membar_consumer();
1106 		if (bpf_jit) {
1107 			bpf_ctx_t *bc = bpf_default_ctx();
1108 			jcode = bpf_jit_generate(bc, fcode, flen);
1109 		}
1110 	} else {
1111 		fcode = NULL;
1112 	}
1113 
1114 	s = splnet();
1115 	old = d->bd_filter;
1116 	d->bd_filter = fcode;
1117 	oldj = d->bd_jitcode;
1118 	d->bd_jitcode = jcode;
1119 	reset_d(d);
1120 	splx(s);
1121 
1122 	if (old) {
1123 		free(old, M_DEVBUF);
1124 	}
1125 	if (oldj) {
1126 		bpf_jit_freecode(oldj);
1127 	}
1128 
1129 	return 0;
1130 }
1131 
1132 /*
1133  * Detach a file from its current interface (if attached at all) and attach
1134  * to the interface indicated by the name stored in ifr.
1135  * Return an errno or 0.
1136  */
1137 static int
1138 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1139 {
1140 	struct bpf_if *bp;
1141 	char *cp;
1142 	int unit_seen, i, s, error;
1143 
1144 	/*
1145 	 * Make sure the provided name has a unit number, and default
1146 	 * it to '0' if not specified.
1147 	 * XXX This is ugly ... do this differently?
1148 	 */
1149 	unit_seen = 0;
1150 	cp = ifr->ifr_name;
1151 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1152 	while (*cp++)
1153 		if (*cp >= '0' && *cp <= '9')
1154 			unit_seen = 1;
1155 	if (!unit_seen) {
1156 		/* Make sure to leave room for the '\0'. */
1157 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1158 			if ((ifr->ifr_name[i] >= 'a' &&
1159 			     ifr->ifr_name[i] <= 'z') ||
1160 			    (ifr->ifr_name[i] >= 'A' &&
1161 			     ifr->ifr_name[i] <= 'Z'))
1162 				continue;
1163 			ifr->ifr_name[i] = '0';
1164 		}
1165 	}
1166 
1167 	/*
1168 	 * Look through attached interfaces for the named one.
1169 	 */
1170 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
1171 		struct ifnet *ifp = bp->bif_ifp;
1172 
1173 		if (ifp == 0 ||
1174 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1175 			continue;
1176 		/* skip additional entry */
1177 		if (bp->bif_driverp != &ifp->if_bpf)
1178 			continue;
1179 		/*
1180 		 * We found the requested interface.
1181 		 * Allocate the packet buffers if we need to.
1182 		 * If we're already attached to requested interface,
1183 		 * just flush the buffer.
1184 		 */
1185 		if (d->bd_sbuf == 0) {
1186 			error = bpf_allocbufs(d);
1187 			if (error != 0)
1188 				return (error);
1189 		}
1190 		s = splnet();
1191 		if (bp != d->bd_bif) {
1192 			if (d->bd_bif)
1193 				/*
1194 				 * Detach if attached to something else.
1195 				 */
1196 				bpf_detachd(d);
1197 
1198 			bpf_attachd(d, bp);
1199 		}
1200 		reset_d(d);
1201 		splx(s);
1202 		return (0);
1203 	}
1204 	/* Not found. */
1205 	return (ENXIO);
1206 }
1207 
1208 /*
1209  * Copy the interface name to the ifreq.
1210  */
1211 static void
1212 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1213 {
1214 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1215 }
1216 
1217 static int
1218 bpf_stat(struct file *fp, struct stat *st)
1219 {
1220 	struct bpf_d *d = fp->f_data;
1221 
1222 	(void)memset(st, 0, sizeof(*st));
1223 	KERNEL_LOCK(1, NULL);
1224 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1225 	st->st_atimespec = d->bd_atime;
1226 	st->st_mtimespec = d->bd_mtime;
1227 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1228 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1229 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1230 	st->st_mode = S_IFCHR;
1231 	KERNEL_UNLOCK_ONE(NULL);
1232 	return 0;
1233 }
1234 
1235 /*
1236  * Support for poll() system call
1237  *
1238  * Return true iff the specific operation will not block indefinitely - with
1239  * the assumption that it is safe to positively acknowledge a request for the
1240  * ability to write to the BPF device.
1241  * Otherwise, return false but make a note that a selnotify() must be done.
1242  */
1243 static int
1244 bpf_poll(struct file *fp, int events)
1245 {
1246 	struct bpf_d *d = fp->f_data;
1247 	int s = splnet();
1248 	int revents;
1249 
1250 	/*
1251 	 * Refresh the PID associated with this bpf file.
1252 	 */
1253 	KERNEL_LOCK(1, NULL);
1254 	d->bd_pid = curproc->p_pid;
1255 
1256 	revents = events & (POLLOUT | POLLWRNORM);
1257 	if (events & (POLLIN | POLLRDNORM)) {
1258 		/*
1259 		 * An imitation of the FIONREAD ioctl code.
1260 		 */
1261 		if (d->bd_hlen != 0 ||
1262 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1263 		     d->bd_slen != 0)) {
1264 			revents |= events & (POLLIN | POLLRDNORM);
1265 		} else {
1266 			selrecord(curlwp, &d->bd_sel);
1267 			/* Start the read timeout if necessary */
1268 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1269 				callout_reset(&d->bd_callout, d->bd_rtout,
1270 					      bpf_timed_out, d);
1271 				d->bd_state = BPF_WAITING;
1272 			}
1273 		}
1274 	}
1275 
1276 	KERNEL_UNLOCK_ONE(NULL);
1277 	splx(s);
1278 	return (revents);
1279 }
1280 
1281 static void
1282 filt_bpfrdetach(struct knote *kn)
1283 {
1284 	struct bpf_d *d = kn->kn_hook;
1285 	int s;
1286 
1287 	KERNEL_LOCK(1, NULL);
1288 	s = splnet();
1289 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1290 	splx(s);
1291 	KERNEL_UNLOCK_ONE(NULL);
1292 }
1293 
1294 static int
1295 filt_bpfread(struct knote *kn, long hint)
1296 {
1297 	struct bpf_d *d = kn->kn_hook;
1298 	int rv;
1299 
1300 	KERNEL_LOCK(1, NULL);
1301 	kn->kn_data = d->bd_hlen;
1302 	if (d->bd_immediate)
1303 		kn->kn_data += d->bd_slen;
1304 	rv = (kn->kn_data > 0);
1305 	KERNEL_UNLOCK_ONE(NULL);
1306 	return rv;
1307 }
1308 
1309 static const struct filterops bpfread_filtops =
1310 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1311 
1312 static int
1313 bpf_kqfilter(struct file *fp, struct knote *kn)
1314 {
1315 	struct bpf_d *d = fp->f_data;
1316 	struct klist *klist;
1317 	int s;
1318 
1319 	KERNEL_LOCK(1, NULL);
1320 
1321 	switch (kn->kn_filter) {
1322 	case EVFILT_READ:
1323 		klist = &d->bd_sel.sel_klist;
1324 		kn->kn_fop = &bpfread_filtops;
1325 		break;
1326 
1327 	default:
1328 		KERNEL_UNLOCK_ONE(NULL);
1329 		return (EINVAL);
1330 	}
1331 
1332 	kn->kn_hook = d;
1333 
1334 	s = splnet();
1335 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1336 	splx(s);
1337 	KERNEL_UNLOCK_ONE(NULL);
1338 
1339 	return (0);
1340 }
1341 
1342 /*
1343  * Copy data from an mbuf chain into a buffer.  This code is derived
1344  * from m_copydata in sys/uipc_mbuf.c.
1345  */
1346 static void *
1347 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1348 {
1349 	const struct mbuf *m;
1350 	u_int count;
1351 	u_char *dst;
1352 
1353 	m = src_arg;
1354 	dst = dst_arg;
1355 	while (len > 0) {
1356 		if (m == NULL)
1357 			panic("bpf_mcpy");
1358 		count = min(m->m_len, len);
1359 		memcpy(dst, mtod(m, const void *), count);
1360 		m = m->m_next;
1361 		dst += count;
1362 		len -= count;
1363 	}
1364 	return dst_arg;
1365 }
1366 
1367 /*
1368  * Dispatch a packet to all the listeners on interface bp.
1369  *
1370  * pkt     pointer to the packet, either a data buffer or an mbuf chain
1371  * buflen  buffer length, if pkt is a data buffer
1372  * cpfn    a function that can copy pkt into the listener's buffer
1373  * pktlen  length of the packet
1374  * rcv     true if packet came in
1375  */
1376 static inline void
1377 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1378     void *pkt, u_int pktlen, u_int buflen, const bool rcv)
1379 {
1380 	bpf_ctx_t *bc = bpf_default_ctx();
1381 	bpf_args_t args = {
1382 		.pkt = pkt,
1383 		.wirelen = pktlen,
1384 		.buflen = buflen,
1385 		.arg = NULL
1386 	};
1387 	struct bpf_d *d;
1388 	struct timespec ts;
1389 	bool gottime = false;
1390 
1391 	/*
1392 	 * Note that the IPL does not have to be raised at this point.
1393 	 * The only problem that could arise here is that if two different
1394 	 * interfaces shared any data.  This is not the case.
1395 	 */
1396 	for (d = bp->bif_dlist; d != NULL; d = d->bd_next) {
1397 		u_int slen;
1398 
1399 		if (!d->bd_seesent && !rcv) {
1400 			continue;
1401 		}
1402 		d->bd_rcount++;
1403 		bpf_gstats.bs_recv++;
1404 
1405 		if (d->bd_jitcode)
1406 			slen = d->bd_jitcode(pkt, pktlen, buflen);
1407 		else
1408 			slen = bpf_filter_ext(bc, d->bd_filter, &args);
1409 
1410 		if (!slen) {
1411 			continue;
1412 		}
1413 		if (!gottime) {
1414 			gottime = true;
1415 			nanotime(&ts);
1416 		}
1417 		catchpacket(d, pkt, pktlen, slen, cpfn, &ts);
1418 	}
1419 }
1420 
1421 /*
1422  * Incoming linkage from device drivers.  Process the packet pkt, of length
1423  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1424  * by each process' filter, and if accepted, stashed into the corresponding
1425  * buffer.
1426  */
1427 static void
1428 _bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1429 {
1430 
1431 	bpf_deliver(bp, memcpy, pkt, pktlen, pktlen, true);
1432 }
1433 
1434 /*
1435  * Incoming linkage from device drivers, when the head of the packet is in
1436  * a buffer, and the tail is in an mbuf chain.
1437  */
1438 static void
1439 _bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1440 {
1441 	u_int pktlen;
1442 	struct mbuf mb;
1443 
1444 	/* Skip outgoing duplicate packets. */
1445 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1446 		m->m_flags &= ~M_PROMISC;
1447 		return;
1448 	}
1449 
1450 	pktlen = m_length(m) + dlen;
1451 
1452 	/*
1453 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1454 	 * Note that we cut corners here; we only setup what's
1455 	 * absolutely needed--this mbuf should never go anywhere else.
1456 	 */
1457 	(void)memset(&mb, 0, sizeof(mb));
1458 	mb.m_next = m;
1459 	mb.m_data = data;
1460 	mb.m_len = dlen;
1461 
1462 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif != NULL);
1463 }
1464 
1465 /*
1466  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1467  */
1468 static void
1469 _bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1470 {
1471 	void *(*cpfn)(void *, const void *, size_t);
1472 	u_int pktlen, buflen;
1473 	void *marg;
1474 
1475 	/* Skip outgoing duplicate packets. */
1476 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1477 		m->m_flags &= ~M_PROMISC;
1478 		return;
1479 	}
1480 
1481 	pktlen = m_length(m);
1482 
1483 	if (pktlen == m->m_len) {
1484 		cpfn = (void *)memcpy;
1485 		marg = mtod(m, void *);
1486 		buflen = pktlen;
1487 	} else {
1488 		cpfn = bpf_mcpy;
1489 		marg = m;
1490 		buflen = 0;
1491 	}
1492 
1493 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif != NULL);
1494 }
1495 
1496 /*
1497  * We need to prepend the address family as
1498  * a four byte field.  Cons up a dummy header
1499  * to pacify bpf.  This is safe because bpf
1500  * will only read from the mbuf (i.e., it won't
1501  * try to free it or keep a pointer a to it).
1502  */
1503 static void
1504 _bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m)
1505 {
1506 	struct mbuf m0;
1507 
1508 	m0.m_flags = 0;
1509 	m0.m_next = m;
1510 	m0.m_len = 4;
1511 	m0.m_data = (char *)&af;
1512 
1513 	_bpf_mtap(bp, &m0);
1514 }
1515 
1516 /*
1517  * Put the SLIP pseudo-"link header" in place.
1518  * Note this M_PREPEND() should never fail,
1519  * swince we know we always have enough space
1520  * in the input buffer.
1521  */
1522 static void
1523 _bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m)
1524 {
1525 	int s;
1526 	u_char *hp;
1527 
1528 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1529 	if (*m == NULL)
1530 		return;
1531 
1532 	hp = mtod(*m, u_char *);
1533 	hp[SLX_DIR] = SLIPDIR_IN;
1534 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1535 
1536 	s = splnet();
1537 	_bpf_mtap(bp, *m);
1538 	splx(s);
1539 
1540 	m_adj(*m, SLIP_HDRLEN);
1541 }
1542 
1543 /*
1544  * Put the SLIP pseudo-"link header" in
1545  * place.  The compressed header is now
1546  * at the beginning of the mbuf.
1547  */
1548 static void
1549 _bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m)
1550 {
1551 	struct mbuf m0;
1552 	u_char *hp;
1553 	int s;
1554 
1555 	m0.m_flags = 0;
1556 	m0.m_next = m;
1557 	m0.m_data = m0.m_dat;
1558 	m0.m_len = SLIP_HDRLEN;
1559 
1560 	hp = mtod(&m0, u_char *);
1561 
1562 	hp[SLX_DIR] = SLIPDIR_OUT;
1563 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1564 
1565 	s = splnet();
1566 	_bpf_mtap(bp, &m0);
1567 	splx(s);
1568 	m_freem(m);
1569 }
1570 
1571 static int
1572 bpf_hdrlen(struct bpf_d *d)
1573 {
1574 	int hdrlen = d->bd_bif->bif_hdrlen;
1575 	/*
1576 	 * Compute the length of the bpf header.  This is not necessarily
1577 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1578 	 * that the network layer header begins on a longword boundary (for
1579 	 * performance reasons and to alleviate alignment restrictions).
1580 	 */
1581 #ifdef _LP64
1582 	if (d->bd_compat32)
1583 		return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen);
1584 	else
1585 #endif
1586 		return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen);
1587 }
1588 
1589 /*
1590  * Move the packet data from interface memory (pkt) into the
1591  * store buffer. Call the wakeup functions if it's time to wakeup
1592  * a listener (buffer full), "cpfn" is the routine called to do the
1593  * actual data transfer. memcpy is passed in to copy contiguous chunks,
1594  * while bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1595  * pkt is really an mbuf.
1596  */
1597 static void
1598 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1599     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1600 {
1601 	char *h;
1602 	int totlen, curlen, caplen;
1603 	int hdrlen = bpf_hdrlen(d);
1604 	int do_wakeup = 0;
1605 
1606 	++d->bd_ccount;
1607 	++bpf_gstats.bs_capt;
1608 	/*
1609 	 * Figure out how many bytes to move.  If the packet is
1610 	 * greater or equal to the snapshot length, transfer that
1611 	 * much.  Otherwise, transfer the whole packet (unless
1612 	 * we hit the buffer size limit).
1613 	 */
1614 	totlen = hdrlen + min(snaplen, pktlen);
1615 	if (totlen > d->bd_bufsize)
1616 		totlen = d->bd_bufsize;
1617 	/*
1618 	 * If we adjusted totlen to fit the bufsize, it could be that
1619 	 * totlen is smaller than hdrlen because of the link layer header.
1620 	 */
1621 	caplen = totlen - hdrlen;
1622 	if (caplen < 0)
1623 		caplen = 0;
1624 
1625 	/*
1626 	 * Round up the end of the previous packet to the next longword.
1627 	 */
1628 #ifdef _LP64
1629 	if (d->bd_compat32)
1630 		curlen = BPF_WORDALIGN32(d->bd_slen);
1631 	else
1632 #endif
1633 		curlen = BPF_WORDALIGN(d->bd_slen);
1634 	if (curlen + totlen > d->bd_bufsize) {
1635 		/*
1636 		 * This packet will overflow the storage buffer.
1637 		 * Rotate the buffers if we can, then wakeup any
1638 		 * pending reads.
1639 		 */
1640 		if (d->bd_fbuf == 0) {
1641 			/*
1642 			 * We haven't completed the previous read yet,
1643 			 * so drop the packet.
1644 			 */
1645 			++d->bd_dcount;
1646 			++bpf_gstats.bs_drop;
1647 			return;
1648 		}
1649 		ROTATE_BUFFERS(d);
1650 		do_wakeup = 1;
1651 		curlen = 0;
1652 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1653 		/*
1654 		 * Immediate mode is set, or the read timeout has
1655 		 * already expired during a select call.  A packet
1656 		 * arrived, so the reader should be woken up.
1657 		 */
1658 		do_wakeup = 1;
1659 	}
1660 
1661 	/*
1662 	 * Append the bpf header.
1663 	 */
1664 	h = (char *)d->bd_sbuf + curlen;
1665 #ifdef _LP64
1666 	if (d->bd_compat32) {
1667 		struct bpf_hdr32 *hp32;
1668 
1669 		hp32 = (struct bpf_hdr32 *)h;
1670 		hp32->bh_tstamp.tv_sec = ts->tv_sec;
1671 		hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1672 		hp32->bh_datalen = pktlen;
1673 		hp32->bh_hdrlen = hdrlen;
1674 		hp32->bh_caplen = caplen;
1675 	} else
1676 #endif
1677 	{
1678 		struct bpf_hdr *hp;
1679 
1680 		hp = (struct bpf_hdr *)h;
1681 		hp->bh_tstamp.tv_sec = ts->tv_sec;
1682 		hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1683 		hp->bh_datalen = pktlen;
1684 		hp->bh_hdrlen = hdrlen;
1685 		hp->bh_caplen = caplen;
1686 	}
1687 
1688 	/*
1689 	 * Copy the packet data into the store buffer and update its length.
1690 	 */
1691 	(*cpfn)(h + hdrlen, pkt, caplen);
1692 	d->bd_slen = curlen + totlen;
1693 
1694 	/*
1695 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1696 	 * will cause filt_bpfread() to be called with it adjusted.
1697 	 */
1698 	if (do_wakeup)
1699 		bpf_wakeup(d);
1700 }
1701 
1702 /*
1703  * Initialize all nonzero fields of a descriptor.
1704  */
1705 static int
1706 bpf_allocbufs(struct bpf_d *d)
1707 {
1708 
1709 	d->bd_fbuf = malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK | M_CANFAIL);
1710 	if (!d->bd_fbuf)
1711 		return (ENOBUFS);
1712 	d->bd_sbuf = malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK | M_CANFAIL);
1713 	if (!d->bd_sbuf) {
1714 		free(d->bd_fbuf, M_DEVBUF);
1715 		return (ENOBUFS);
1716 	}
1717 	d->bd_slen = 0;
1718 	d->bd_hlen = 0;
1719 	return (0);
1720 }
1721 
1722 /*
1723  * Free buffers currently in use by a descriptor.
1724  * Called on close.
1725  */
1726 static void
1727 bpf_freed(struct bpf_d *d)
1728 {
1729 	/*
1730 	 * We don't need to lock out interrupts since this descriptor has
1731 	 * been detached from its interface and it yet hasn't been marked
1732 	 * free.
1733 	 */
1734 	if (d->bd_sbuf != NULL) {
1735 		free(d->bd_sbuf, M_DEVBUF);
1736 		if (d->bd_hbuf != NULL)
1737 			free(d->bd_hbuf, M_DEVBUF);
1738 		if (d->bd_fbuf != NULL)
1739 			free(d->bd_fbuf, M_DEVBUF);
1740 	}
1741 	if (d->bd_filter)
1742 		free(d->bd_filter, M_DEVBUF);
1743 
1744 	if (d->bd_jitcode != NULL) {
1745 		bpf_jit_freecode(d->bd_jitcode);
1746 	}
1747 }
1748 
1749 /*
1750  * Attach an interface to bpf.  dlt is the link layer type;
1751  * hdrlen is the fixed size of the link header for the specified dlt
1752  * (variable length headers not yet supported).
1753  */
1754 static void
1755 _bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1756 {
1757 	struct bpf_if *bp;
1758 	bp = malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1759 	if (bp == 0)
1760 		panic("bpfattach");
1761 
1762 	bp->bif_dlist = 0;
1763 	bp->bif_driverp = driverp;
1764 	bp->bif_ifp = ifp;
1765 	bp->bif_dlt = dlt;
1766 
1767 	bp->bif_next = bpf_iflist;
1768 	bpf_iflist = bp;
1769 
1770 	*bp->bif_driverp = 0;
1771 
1772 	bp->bif_hdrlen = hdrlen;
1773 #if 0
1774 	printf("bpf: %s attached\n", ifp->if_xname);
1775 #endif
1776 }
1777 
1778 /*
1779  * Remove an interface from bpf.
1780  */
1781 static void
1782 _bpfdetach(struct ifnet *ifp)
1783 {
1784 	struct bpf_if *bp, **pbp;
1785 	struct bpf_d *d;
1786 	int s;
1787 
1788 	/* Nuke the vnodes for any open instances */
1789 	LIST_FOREACH(d, &bpf_list, bd_list) {
1790 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
1791 			/*
1792 			 * Detach the descriptor from an interface now.
1793 			 * It will be free'ed later by close routine.
1794 			 */
1795 			s = splnet();
1796 			d->bd_promisc = 0;	/* we can't touch device. */
1797 			bpf_detachd(d);
1798 			splx(s);
1799 		}
1800 	}
1801 
1802   again:
1803 	for (bp = bpf_iflist, pbp = &bpf_iflist;
1804 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
1805 		if (bp->bif_ifp == ifp) {
1806 			*pbp = bp->bif_next;
1807 			free(bp, M_DEVBUF);
1808 			goto again;
1809 		}
1810 	}
1811 }
1812 
1813 /*
1814  * Change the data link type of a interface.
1815  */
1816 static void
1817 _bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1818 {
1819 	struct bpf_if *bp;
1820 
1821 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1822 		if (bp->bif_driverp == &ifp->if_bpf)
1823 			break;
1824 	}
1825 	if (bp == NULL)
1826 		panic("bpf_change_type");
1827 
1828 	bp->bif_dlt = dlt;
1829 
1830 	bp->bif_hdrlen = hdrlen;
1831 }
1832 
1833 /*
1834  * Get a list of available data link type of the interface.
1835  */
1836 static int
1837 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1838 {
1839 	int n, error;
1840 	struct ifnet *ifp;
1841 	struct bpf_if *bp;
1842 
1843 	ifp = d->bd_bif->bif_ifp;
1844 	n = 0;
1845 	error = 0;
1846 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1847 		if (bp->bif_ifp != ifp)
1848 			continue;
1849 		if (bfl->bfl_list != NULL) {
1850 			if (n >= bfl->bfl_len)
1851 				return ENOMEM;
1852 			error = copyout(&bp->bif_dlt,
1853 			    bfl->bfl_list + n, sizeof(u_int));
1854 		}
1855 		n++;
1856 	}
1857 	bfl->bfl_len = n;
1858 	return error;
1859 }
1860 
1861 /*
1862  * Set the data link type of a BPF instance.
1863  */
1864 static int
1865 bpf_setdlt(struct bpf_d *d, u_int dlt)
1866 {
1867 	int s, error, opromisc;
1868 	struct ifnet *ifp;
1869 	struct bpf_if *bp;
1870 
1871 	if (d->bd_bif->bif_dlt == dlt)
1872 		return 0;
1873 	ifp = d->bd_bif->bif_ifp;
1874 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1875 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1876 			break;
1877 	}
1878 	if (bp == NULL)
1879 		return EINVAL;
1880 	s = splnet();
1881 	opromisc = d->bd_promisc;
1882 	bpf_detachd(d);
1883 	bpf_attachd(d, bp);
1884 	reset_d(d);
1885 	if (opromisc) {
1886 		error = ifpromisc(bp->bif_ifp, 1);
1887 		if (error)
1888 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
1889 			    bp->bif_ifp->if_xname, error);
1890 		else
1891 			d->bd_promisc = 1;
1892 	}
1893 	splx(s);
1894 	return 0;
1895 }
1896 
1897 static int
1898 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
1899 {
1900 	int newsize, error;
1901 	struct sysctlnode node;
1902 
1903 	node = *rnode;
1904 	node.sysctl_data = &newsize;
1905 	newsize = bpf_maxbufsize;
1906 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1907 	if (error || newp == NULL)
1908 		return (error);
1909 
1910 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
1911 		return (EINVAL);
1912 
1913 	bpf_maxbufsize = newsize;
1914 
1915 	return (0);
1916 }
1917 
1918 static int
1919 sysctl_net_bpf_jit(SYSCTLFN_ARGS)
1920 {
1921 	bool newval;
1922 	int error;
1923 	struct sysctlnode node;
1924 
1925 	node = *rnode;
1926 	node.sysctl_data = &newval;
1927 	newval = bpf_jit;
1928 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1929 	if (error != 0 || newp == NULL)
1930 		return error;
1931 
1932 	bpf_jit = newval;
1933 
1934 	/*
1935 	 * Do a full sync to publish new bpf_jit value and
1936 	 * update bpfjit_module_ops.bj_generate_code variable.
1937 	 */
1938 	membar_sync();
1939 
1940 	if (newval && bpfjit_module_ops.bj_generate_code == NULL) {
1941 		printf("WARNING jit activation is postponed "
1942 		    "until after bpfjit module is loaded\n");
1943 	}
1944 
1945 	return 0;
1946 }
1947 
1948 static int
1949 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
1950 {
1951 	int    error, elem_count;
1952 	struct bpf_d	 *dp;
1953 	struct bpf_d_ext  dpe;
1954 	size_t len, needed, elem_size, out_size;
1955 	char   *sp;
1956 
1957 	if (namelen == 1 && name[0] == CTL_QUERY)
1958 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
1959 
1960 	if (namelen != 2)
1961 		return (EINVAL);
1962 
1963 	/* BPF peers is privileged information. */
1964 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
1965 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
1966 	if (error)
1967 		return (EPERM);
1968 
1969 	len = (oldp != NULL) ? *oldlenp : 0;
1970 	sp = oldp;
1971 	elem_size = name[0];
1972 	elem_count = name[1];
1973 	out_size = MIN(sizeof(dpe), elem_size);
1974 	needed = 0;
1975 
1976 	if (elem_size < 1 || elem_count < 0)
1977 		return (EINVAL);
1978 
1979 	mutex_enter(&bpf_mtx);
1980 	LIST_FOREACH(dp, &bpf_list, bd_list) {
1981 		if (len >= elem_size && elem_count > 0) {
1982 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
1983 			BPF_EXT(bufsize);
1984 			BPF_EXT(promisc);
1985 			BPF_EXT(state);
1986 			BPF_EXT(immediate);
1987 			BPF_EXT(hdrcmplt);
1988 			BPF_EXT(seesent);
1989 			BPF_EXT(pid);
1990 			BPF_EXT(rcount);
1991 			BPF_EXT(dcount);
1992 			BPF_EXT(ccount);
1993 #undef BPF_EXT
1994 			if (dp->bd_bif)
1995 				(void)strlcpy(dpe.bde_ifname,
1996 				    dp->bd_bif->bif_ifp->if_xname,
1997 				    IFNAMSIZ - 1);
1998 			else
1999 				dpe.bde_ifname[0] = '\0';
2000 
2001 			error = copyout(&dpe, sp, out_size);
2002 			if (error)
2003 				break;
2004 			sp += elem_size;
2005 			len -= elem_size;
2006 		}
2007 		needed += elem_size;
2008 		if (elem_count > 0 && elem_count != INT_MAX)
2009 			elem_count--;
2010 	}
2011 	mutex_exit(&bpf_mtx);
2012 
2013 	*oldlenp = needed;
2014 
2015 	return (error);
2016 }
2017 
2018 static struct sysctllog *bpf_sysctllog;
2019 static void
2020 sysctl_net_bpf_setup(void)
2021 {
2022 	const struct sysctlnode *node;
2023 
2024 	sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2025 		       CTLFLAG_PERMANENT,
2026 		       CTLTYPE_NODE, "net", NULL,
2027 		       NULL, 0, NULL, 0,
2028 		       CTL_NET, CTL_EOL);
2029 
2030 	node = NULL;
2031 	sysctl_createv(&bpf_sysctllog, 0, NULL, &node,
2032 		       CTLFLAG_PERMANENT,
2033 		       CTLTYPE_NODE, "bpf",
2034 		       SYSCTL_DESCR("BPF options"),
2035 		       NULL, 0, NULL, 0,
2036 		       CTL_NET, CTL_CREATE, CTL_EOL);
2037 	if (node != NULL) {
2038 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2039 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2040 			CTLTYPE_BOOL, "jit",
2041 			SYSCTL_DESCR("Toggle Just-In-Time compilation"),
2042 			sysctl_net_bpf_jit, 0, &bpf_jit, 0,
2043 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2044 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2045 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2046 			CTLTYPE_INT, "maxbufsize",
2047 			SYSCTL_DESCR("Maximum size for data capture buffer"),
2048 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
2049 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2050 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2051 			CTLFLAG_PERMANENT,
2052 			CTLTYPE_STRUCT, "stats",
2053 			SYSCTL_DESCR("BPF stats"),
2054 			NULL, 0, &bpf_gstats, sizeof(bpf_gstats),
2055 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2056 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2057 			CTLFLAG_PERMANENT,
2058 			CTLTYPE_STRUCT, "peers",
2059 			SYSCTL_DESCR("BPF peers"),
2060 			sysctl_net_bpf_peers, 0, NULL, 0,
2061 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2062 	}
2063 
2064 }
2065 
2066 struct bpf_ops bpf_ops_kernel = {
2067 	.bpf_attach =		_bpfattach,
2068 	.bpf_detach =		_bpfdetach,
2069 	.bpf_change_type =	_bpf_change_type,
2070 
2071 	.bpf_tap =		_bpf_tap,
2072 	.bpf_mtap =		_bpf_mtap,
2073 	.bpf_mtap2 =		_bpf_mtap2,
2074 	.bpf_mtap_af =		_bpf_mtap_af,
2075 	.bpf_mtap_sl_in =	_bpf_mtap_sl_in,
2076 	.bpf_mtap_sl_out =	_bpf_mtap_sl_out,
2077 };
2078 
2079 MODULE(MODULE_CLASS_DRIVER, bpf, NULL);
2080 
2081 static int
2082 bpf_modcmd(modcmd_t cmd, void *arg)
2083 {
2084 	devmajor_t bmajor, cmajor;
2085 	int error;
2086 
2087 	bmajor = cmajor = NODEVMAJOR;
2088 
2089 	switch (cmd) {
2090 	case MODULE_CMD_INIT:
2091 		bpfilterattach(0);
2092 		error = devsw_attach("bpf", NULL, &bmajor,
2093 		    &bpf_cdevsw, &cmajor);
2094 		if (error == EEXIST)
2095 			error = 0; /* maybe built-in ... improve eventually */
2096 		if (error)
2097 			break;
2098 
2099 		bpf_ops_handover_enter(&bpf_ops_kernel);
2100 		atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel);
2101 		bpf_ops_handover_exit();
2102 		sysctl_net_bpf_setup();
2103 		break;
2104 
2105 	case MODULE_CMD_FINI:
2106 		/*
2107 		 * While there is no reference counting for bpf callers,
2108 		 * unload could at least in theory be done similarly to
2109 		 * system call disestablishment.  This should even be
2110 		 * a little simpler:
2111 		 *
2112 		 * 1) replace op vector with stubs
2113 		 * 2) post update to all cpus with xc
2114 		 * 3) check that nobody is in bpf anymore
2115 		 *    (it's doubtful we'd want something like l_sysent,
2116 		 *     but we could do something like *signed* percpu
2117 		 *     counters.  if the sum is 0, we're good).
2118 		 * 4) if fail, unroll changes
2119 		 *
2120 		 * NOTE: change won't be atomic to the outside.  some
2121 		 * packets may be not captured even if unload is
2122 		 * not succesful.  I think packet capture not working
2123 		 * is a perfectly logical consequence of trying to
2124 		 * disable packet capture.
2125 		 */
2126 		error = EOPNOTSUPP;
2127 		/* insert sysctl teardown */
2128 		break;
2129 
2130 	default:
2131 		error = ENOTTY;
2132 		break;
2133 	}
2134 
2135 	return error;
2136 }
2137