xref: /netbsd-src/sys/net/bpf.c (revision ba65fde2d7fefa7d39838fa5fa855e62bd606b5e)
1 /*	$NetBSD: bpf.c,v 1.173 2012/10/27 22:36:14 alnsn 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 1.173 2012/10/27 22:36:14 alnsn 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 static int
189 bpf_movein(struct uio *uio, int linktype, uint64_t mtu, struct mbuf **mp,
190 	   struct sockaddr *sockp)
191 {
192 	struct mbuf *m;
193 	int error;
194 	size_t len;
195 	size_t hlen;
196 	size_t align;
197 
198 	/*
199 	 * Build a sockaddr based on the data link layer type.
200 	 * We do this at this level because the ethernet header
201 	 * is copied directly into the data field of the sockaddr.
202 	 * In the case of SLIP, there is no header and the packet
203 	 * is forwarded as is.
204 	 * Also, we are careful to leave room at the front of the mbuf
205 	 * for the link level header.
206 	 */
207 	switch (linktype) {
208 
209 	case DLT_SLIP:
210 		sockp->sa_family = AF_INET;
211 		hlen = 0;
212 		align = 0;
213 		break;
214 
215 	case DLT_PPP:
216 		sockp->sa_family = AF_UNSPEC;
217 		hlen = 0;
218 		align = 0;
219 		break;
220 
221 	case DLT_EN10MB:
222 		sockp->sa_family = AF_UNSPEC;
223 		/* XXX Would MAXLINKHDR be better? */
224  		/* 6(dst)+6(src)+2(type) */
225 		hlen = sizeof(struct ether_header);
226 		align = 2;
227 		break;
228 
229 	case DLT_ARCNET:
230 		sockp->sa_family = AF_UNSPEC;
231 		hlen = ARC_HDRLEN;
232 		align = 5;
233 		break;
234 
235 	case DLT_FDDI:
236 		sockp->sa_family = AF_LINK;
237 		/* XXX 4(FORMAC)+6(dst)+6(src) */
238 		hlen = 16;
239 		align = 0;
240 		break;
241 
242 	case DLT_ECONET:
243 		sockp->sa_family = AF_UNSPEC;
244 		hlen = 6;
245 		align = 2;
246 		break;
247 
248 	case DLT_NULL:
249 		sockp->sa_family = AF_UNSPEC;
250 		hlen = 0;
251 		align = 0;
252 		break;
253 
254 	default:
255 		return (EIO);
256 	}
257 
258 	len = uio->uio_resid;
259 	/*
260 	 * If there aren't enough bytes for a link level header or the
261 	 * packet length exceeds the interface mtu, return an error.
262 	 */
263 	if (len - hlen > mtu)
264 		return (EMSGSIZE);
265 
266 	/*
267 	 * XXX Avoid complicated buffer chaining ---
268 	 * bail if it won't fit in a single mbuf.
269 	 * (Take into account possible alignment bytes)
270 	 */
271 	if (len + align > MCLBYTES)
272 		return (EIO);
273 
274 	m = m_gethdr(M_WAIT, MT_DATA);
275 	m->m_pkthdr.rcvif = 0;
276 	m->m_pkthdr.len = (int)(len - hlen);
277 	if (len + align > MHLEN) {
278 		m_clget(m, M_WAIT);
279 		if ((m->m_flags & M_EXT) == 0) {
280 			error = ENOBUFS;
281 			goto bad;
282 		}
283 	}
284 
285 	/* Insure the data is properly aligned */
286 	if (align > 0) {
287 		m->m_data += align;
288 		m->m_len -= (int)align;
289 	}
290 
291 	error = uiomove(mtod(m, void *), len, uio);
292 	if (error)
293 		goto bad;
294 	if (hlen != 0) {
295 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
296 		m->m_data += hlen; /* XXX */
297 		len -= hlen;
298 	}
299 	m->m_len = (int)len;
300 	*mp = m;
301 	return (0);
302 
303 bad:
304 	m_freem(m);
305 	return (error);
306 }
307 
308 /*
309  * Attach file to the bpf interface, i.e. make d listen on bp.
310  * Must be called at splnet.
311  */
312 static void
313 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
314 {
315 	/*
316 	 * Point d at bp, and add d to the interface's list of listeners.
317 	 * Finally, point the driver's bpf cookie at the interface so
318 	 * it will divert packets to bpf.
319 	 */
320 	d->bd_bif = bp;
321 	d->bd_next = bp->bif_dlist;
322 	bp->bif_dlist = d;
323 
324 	*bp->bif_driverp = bp;
325 }
326 
327 /*
328  * Detach a file from its interface.
329  */
330 static void
331 bpf_detachd(struct bpf_d *d)
332 {
333 	struct bpf_d **p;
334 	struct bpf_if *bp;
335 
336 	bp = d->bd_bif;
337 	/*
338 	 * Check if this descriptor had requested promiscuous mode.
339 	 * If so, turn it off.
340 	 */
341 	if (d->bd_promisc) {
342 		int error;
343 
344 		d->bd_promisc = 0;
345 		/*
346 		 * Take device out of promiscuous mode.  Since we were
347 		 * able to enter promiscuous mode, we should be able
348 		 * to turn it off.  But we can get an error if
349 		 * the interface was configured down, so only panic
350 		 * if we don't get an unexpected error.
351 		 */
352   		error = ifpromisc(bp->bif_ifp, 0);
353 		if (error && error != EINVAL)
354 			panic("%s: ifpromisc failed: %d", __func__, error);
355 	}
356 	/* Remove d from the interface's descriptor list. */
357 	p = &bp->bif_dlist;
358 	while (*p != d) {
359 		p = &(*p)->bd_next;
360 		if (*p == 0)
361 			panic("%s: descriptor not in list", __func__);
362 	}
363 	*p = (*p)->bd_next;
364 	if (bp->bif_dlist == 0)
365 		/*
366 		 * Let the driver know that there are no more listeners.
367 		 */
368 		*d->bd_bif->bif_driverp = 0;
369 	d->bd_bif = 0;
370 }
371 
372 static int
373 doinit(void)
374 {
375 
376 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
377 
378 	LIST_INIT(&bpf_list);
379 
380 	bpf_gstats.bs_recv = 0;
381 	bpf_gstats.bs_drop = 0;
382 	bpf_gstats.bs_capt = 0;
383 
384 	return 0;
385 }
386 
387 /*
388  * bpfilterattach() is called at boot time.
389  */
390 /* ARGSUSED */
391 void
392 bpfilterattach(int n)
393 {
394 	static ONCE_DECL(control);
395 
396 	RUN_ONCE(&control, doinit);
397 }
398 
399 /*
400  * Open ethernet device. Clones.
401  */
402 /* ARGSUSED */
403 int
404 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
405 {
406 	struct bpf_d *d;
407 	struct file *fp;
408 	int error, fd;
409 
410 	/* falloc() will use the descriptor for us. */
411 	if ((error = fd_allocfile(&fp, &fd)) != 0)
412 		return error;
413 
414 	d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK|M_ZERO);
415 	d->bd_bufsize = bpf_bufsize;
416 	d->bd_seesent = 1;
417 	d->bd_feedback = 0;
418 	d->bd_pid = l->l_proc->p_pid;
419 #ifdef _LP64
420 	if (curproc->p_flag & PK_32)
421 		d->bd_compat32 = 1;
422 #endif
423 	getnanotime(&d->bd_btime);
424 	d->bd_atime = d->bd_mtime = d->bd_btime;
425 	callout_init(&d->bd_callout, 0);
426 	selinit(&d->bd_sel);
427 	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
428 	d->bd_jitcode = NULL;
429 
430 	mutex_enter(&bpf_mtx);
431 	LIST_INSERT_HEAD(&bpf_list, d, bd_list);
432 	mutex_exit(&bpf_mtx);
433 
434 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
435 }
436 
437 /*
438  * Close the descriptor by detaching it from its interface,
439  * deallocating its buffers, and marking it free.
440  */
441 /* ARGSUSED */
442 static int
443 bpf_close(struct file *fp)
444 {
445 	struct bpf_d *d = fp->f_data;
446 	int s;
447 
448 	KERNEL_LOCK(1, NULL);
449 
450 	/*
451 	 * Refresh the PID associated with this bpf file.
452 	 */
453 	d->bd_pid = curproc->p_pid;
454 
455 	s = splnet();
456 	if (d->bd_state == BPF_WAITING)
457 		callout_stop(&d->bd_callout);
458 	d->bd_state = BPF_IDLE;
459 	if (d->bd_bif)
460 		bpf_detachd(d);
461 	splx(s);
462 	bpf_freed(d);
463 	mutex_enter(&bpf_mtx);
464 	LIST_REMOVE(d, bd_list);
465 	mutex_exit(&bpf_mtx);
466 	callout_destroy(&d->bd_callout);
467 	seldestroy(&d->bd_sel);
468 	softint_disestablish(d->bd_sih);
469 	free(d, M_DEVBUF);
470 	fp->f_data = NULL;
471 
472 	KERNEL_UNLOCK_ONE(NULL);
473 
474 	return (0);
475 }
476 
477 /*
478  * Rotate the packet buffers in descriptor d.  Move the store buffer
479  * into the hold slot, and the free buffer into the store slot.
480  * Zero the length of the new store buffer.
481  */
482 #define ROTATE_BUFFERS(d) \
483 	(d)->bd_hbuf = (d)->bd_sbuf; \
484 	(d)->bd_hlen = (d)->bd_slen; \
485 	(d)->bd_sbuf = (d)->bd_fbuf; \
486 	(d)->bd_slen = 0; \
487 	(d)->bd_fbuf = 0;
488 /*
489  *  bpfread - read next chunk of packets from buffers
490  */
491 static int
492 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
493     kauth_cred_t cred, int flags)
494 {
495 	struct bpf_d *d = fp->f_data;
496 	int timed_out;
497 	int error;
498 	int s;
499 
500 	getnanotime(&d->bd_atime);
501 	/*
502 	 * Restrict application to use a buffer the same size as
503 	 * the kernel buffers.
504 	 */
505 	if (uio->uio_resid != d->bd_bufsize)
506 		return (EINVAL);
507 
508 	KERNEL_LOCK(1, NULL);
509 	s = splnet();
510 	if (d->bd_state == BPF_WAITING)
511 		callout_stop(&d->bd_callout);
512 	timed_out = (d->bd_state == BPF_TIMED_OUT);
513 	d->bd_state = BPF_IDLE;
514 	/*
515 	 * If the hold buffer is empty, then do a timed sleep, which
516 	 * ends when the timeout expires or when enough packets
517 	 * have arrived to fill the store buffer.
518 	 */
519 	while (d->bd_hbuf == 0) {
520 		if (fp->f_flag & FNONBLOCK) {
521 			if (d->bd_slen == 0) {
522 				splx(s);
523 				KERNEL_UNLOCK_ONE(NULL);
524 				return (EWOULDBLOCK);
525 			}
526 			ROTATE_BUFFERS(d);
527 			break;
528 		}
529 
530 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
531 			/*
532 			 * A packet(s) either arrived since the previous
533 			 * read or arrived while we were asleep.
534 			 * Rotate the buffers and return what's here.
535 			 */
536 			ROTATE_BUFFERS(d);
537 			break;
538 		}
539 		error = tsleep(d, PRINET|PCATCH, "bpf",
540 				d->bd_rtout);
541 		if (error == EINTR || error == ERESTART) {
542 			splx(s);
543 			KERNEL_UNLOCK_ONE(NULL);
544 			return (error);
545 		}
546 		if (error == EWOULDBLOCK) {
547 			/*
548 			 * On a timeout, return what's in the buffer,
549 			 * which may be nothing.  If there is something
550 			 * in the store buffer, we can rotate the buffers.
551 			 */
552 			if (d->bd_hbuf)
553 				/*
554 				 * We filled up the buffer in between
555 				 * getting the timeout and arriving
556 				 * here, so we don't need to rotate.
557 				 */
558 				break;
559 
560 			if (d->bd_slen == 0) {
561 				splx(s);
562 				KERNEL_UNLOCK_ONE(NULL);
563 				return (0);
564 			}
565 			ROTATE_BUFFERS(d);
566 			break;
567 		}
568 		if (error != 0)
569 			goto done;
570 	}
571 	/*
572 	 * At this point, we know we have something in the hold slot.
573 	 */
574 	splx(s);
575 
576 	/*
577 	 * Move data from hold buffer into user space.
578 	 * We know the entire buffer is transferred since
579 	 * we checked above that the read buffer is bpf_bufsize bytes.
580 	 */
581 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
582 
583 	s = splnet();
584 	d->bd_fbuf = d->bd_hbuf;
585 	d->bd_hbuf = 0;
586 	d->bd_hlen = 0;
587 done:
588 	splx(s);
589 	KERNEL_UNLOCK_ONE(NULL);
590 	return (error);
591 }
592 
593 
594 /*
595  * If there are processes sleeping on this descriptor, wake them up.
596  */
597 static inline void
598 bpf_wakeup(struct bpf_d *d)
599 {
600 	wakeup(d);
601 	if (d->bd_async)
602 		softint_schedule(d->bd_sih);
603 	selnotify(&d->bd_sel, 0, 0);
604 }
605 
606 static void
607 bpf_softintr(void *cookie)
608 {
609 	struct bpf_d *d;
610 
611 	d = cookie;
612 	if (d->bd_async)
613 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
614 }
615 
616 static void
617 bpf_timed_out(void *arg)
618 {
619 	struct bpf_d *d = arg;
620 	int s;
621 
622 	s = splnet();
623 	if (d->bd_state == BPF_WAITING) {
624 		d->bd_state = BPF_TIMED_OUT;
625 		if (d->bd_slen != 0)
626 			bpf_wakeup(d);
627 	}
628 	splx(s);
629 }
630 
631 
632 static int
633 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
634     kauth_cred_t cred, int flags)
635 {
636 	struct bpf_d *d = fp->f_data;
637 	struct ifnet *ifp;
638 	struct mbuf *m, *mc;
639 	int error, s;
640 	static struct sockaddr_storage dst;
641 
642 	m = NULL;	/* XXX gcc */
643 
644 	KERNEL_LOCK(1, NULL);
645 
646 	if (d->bd_bif == 0) {
647 		KERNEL_UNLOCK_ONE(NULL);
648 		return (ENXIO);
649 	}
650 	getnanotime(&d->bd_mtime);
651 
652 	ifp = d->bd_bif->bif_ifp;
653 
654 	if (uio->uio_resid == 0) {
655 		KERNEL_UNLOCK_ONE(NULL);
656 		return (0);
657 	}
658 
659 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
660 		(struct sockaddr *) &dst);
661 	if (error) {
662 		KERNEL_UNLOCK_ONE(NULL);
663 		return (error);
664 	}
665 
666 	if (m->m_pkthdr.len > ifp->if_mtu) {
667 		KERNEL_UNLOCK_ONE(NULL);
668 		m_freem(m);
669 		return (EMSGSIZE);
670 	}
671 
672 	if (d->bd_hdrcmplt)
673 		dst.ss_family = pseudo_AF_HDRCMPLT;
674 
675 	if (d->bd_feedback) {
676 		mc = m_dup(m, 0, M_COPYALL, M_NOWAIT);
677 		if (mc != NULL)
678 			mc->m_pkthdr.rcvif = ifp;
679 		/* Set M_PROMISC for outgoing packets to be discarded. */
680 		if (1 /*d->bd_direction == BPF_D_INOUT*/)
681 			m->m_flags |= M_PROMISC;
682 	} else
683 		mc = NULL;
684 
685 	s = splsoftnet();
686 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *) &dst, NULL);
687 
688 	if (mc != NULL) {
689 		if (error == 0)
690 			(*ifp->if_input)(ifp, mc);
691 		m_freem(mc);
692 	}
693 	splx(s);
694 	KERNEL_UNLOCK_ONE(NULL);
695 	/*
696 	 * The driver frees the mbuf.
697 	 */
698 	return (error);
699 }
700 
701 /*
702  * Reset a descriptor by flushing its packet buffer and clearing the
703  * receive and drop counts.  Should be called at splnet.
704  */
705 static void
706 reset_d(struct bpf_d *d)
707 {
708 	if (d->bd_hbuf) {
709 		/* Free the hold buffer. */
710 		d->bd_fbuf = d->bd_hbuf;
711 		d->bd_hbuf = 0;
712 	}
713 	d->bd_slen = 0;
714 	d->bd_hlen = 0;
715 	d->bd_rcount = 0;
716 	d->bd_dcount = 0;
717 	d->bd_ccount = 0;
718 }
719 
720 /*
721  *  FIONREAD		Check for read packet available.
722  *  BIOCGBLEN		Get buffer len [for read()].
723  *  BIOCSETF		Set ethernet read filter.
724  *  BIOCFLUSH		Flush read packet buffer.
725  *  BIOCPROMISC		Put interface into promiscuous mode.
726  *  BIOCGDLT		Get link layer type.
727  *  BIOCGETIF		Get interface name.
728  *  BIOCSETIF		Set interface.
729  *  BIOCSRTIMEOUT	Set read timeout.
730  *  BIOCGRTIMEOUT	Get read timeout.
731  *  BIOCGSTATS		Get packet stats.
732  *  BIOCIMMEDIATE	Set immediate mode.
733  *  BIOCVERSION		Get filter language version.
734  *  BIOCGHDRCMPLT	Get "header already complete" flag.
735  *  BIOCSHDRCMPLT	Set "header already complete" flag.
736  *  BIOCSFEEDBACK	Set packet feedback mode.
737  *  BIOCGFEEDBACK	Get packet feedback mode.
738  *  BIOCGSEESENT  	Get "see sent packets" mode.
739  *  BIOCSSEESENT  	Set "see sent packets" mode.
740  */
741 /* ARGSUSED */
742 static int
743 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
744 {
745 	struct bpf_d *d = fp->f_data;
746 	int s, error = 0;
747 
748 	/*
749 	 * Refresh the PID associated with this bpf file.
750 	 */
751 	KERNEL_LOCK(1, NULL);
752 	d->bd_pid = curproc->p_pid;
753 #ifdef _LP64
754 	if (curproc->p_flag & PK_32)
755 		d->bd_compat32 = 1;
756 	else
757 		d->bd_compat32 = 0;
758 #endif
759 
760 	s = splnet();
761 	if (d->bd_state == BPF_WAITING)
762 		callout_stop(&d->bd_callout);
763 	d->bd_state = BPF_IDLE;
764 	splx(s);
765 
766 	switch (cmd) {
767 
768 	default:
769 		error = EINVAL;
770 		break;
771 
772 	/*
773 	 * Check for read packet available.
774 	 */
775 	case FIONREAD:
776 		{
777 			int n;
778 
779 			s = splnet();
780 			n = d->bd_slen;
781 			if (d->bd_hbuf)
782 				n += d->bd_hlen;
783 			splx(s);
784 
785 			*(int *)addr = n;
786 			break;
787 		}
788 
789 	/*
790 	 * Get buffer len [for read()].
791 	 */
792 	case BIOCGBLEN:
793 		*(u_int *)addr = d->bd_bufsize;
794 		break;
795 
796 	/*
797 	 * Set buffer length.
798 	 */
799 	case BIOCSBLEN:
800 		if (d->bd_bif != 0)
801 			error = EINVAL;
802 		else {
803 			u_int size = *(u_int *)addr;
804 
805 			if (size > bpf_maxbufsize)
806 				*(u_int *)addr = size = bpf_maxbufsize;
807 			else if (size < BPF_MINBUFSIZE)
808 				*(u_int *)addr = size = BPF_MINBUFSIZE;
809 			d->bd_bufsize = size;
810 		}
811 		break;
812 
813 	/*
814 	 * Set link layer read filter.
815 	 */
816 	case BIOCSETF:
817 		error = bpf_setf(d, addr);
818 		break;
819 
820 	/*
821 	 * Flush read packet buffer.
822 	 */
823 	case BIOCFLUSH:
824 		s = splnet();
825 		reset_d(d);
826 		splx(s);
827 		break;
828 
829 	/*
830 	 * Put interface into promiscuous mode.
831 	 */
832 	case BIOCPROMISC:
833 		if (d->bd_bif == 0) {
834 			/*
835 			 * No interface attached yet.
836 			 */
837 			error = EINVAL;
838 			break;
839 		}
840 		s = splnet();
841 		if (d->bd_promisc == 0) {
842 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
843 			if (error == 0)
844 				d->bd_promisc = 1;
845 		}
846 		splx(s);
847 		break;
848 
849 	/*
850 	 * Get device parameters.
851 	 */
852 	case BIOCGDLT:
853 		if (d->bd_bif == 0)
854 			error = EINVAL;
855 		else
856 			*(u_int *)addr = d->bd_bif->bif_dlt;
857 		break;
858 
859 	/*
860 	 * Get a list of supported device parameters.
861 	 */
862 	case BIOCGDLTLIST:
863 		if (d->bd_bif == 0)
864 			error = EINVAL;
865 		else
866 			error = bpf_getdltlist(d, addr);
867 		break;
868 
869 	/*
870 	 * Set device parameters.
871 	 */
872 	case BIOCSDLT:
873 		if (d->bd_bif == 0)
874 			error = EINVAL;
875 		else
876 			error = bpf_setdlt(d, *(u_int *)addr);
877 		break;
878 
879 	/*
880 	 * Set interface name.
881 	 */
882 #ifdef OBIOCGETIF
883 	case OBIOCGETIF:
884 #endif
885 	case BIOCGETIF:
886 		if (d->bd_bif == 0)
887 			error = EINVAL;
888 		else
889 			bpf_ifname(d->bd_bif->bif_ifp, addr);
890 		break;
891 
892 	/*
893 	 * Set interface.
894 	 */
895 #ifdef OBIOCSETIF
896 	case OBIOCSETIF:
897 #endif
898 	case BIOCSETIF:
899 		error = bpf_setif(d, addr);
900 		break;
901 
902 	/*
903 	 * Set read timeout.
904 	 */
905 	case BIOCSRTIMEOUT:
906 		{
907 			struct timeval *tv = addr;
908 
909 			/* Compute number of ticks. */
910 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
911 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
912 				d->bd_rtout = 1;
913 			break;
914 		}
915 
916 #ifdef BIOCGORTIMEOUT
917 	/*
918 	 * Get read timeout.
919 	 */
920 	case BIOCGORTIMEOUT:
921 		{
922 			struct timeval50 *tv = addr;
923 
924 			tv->tv_sec = d->bd_rtout / hz;
925 			tv->tv_usec = (d->bd_rtout % hz) * tick;
926 			break;
927 		}
928 #endif
929 
930 #ifdef BIOCSORTIMEOUT
931 	/*
932 	 * Set read timeout.
933 	 */
934 	case BIOCSORTIMEOUT:
935 		{
936 			struct timeval50 *tv = addr;
937 
938 			/* Compute number of ticks. */
939 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
940 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
941 				d->bd_rtout = 1;
942 			break;
943 		}
944 #endif
945 
946 	/*
947 	 * Get read timeout.
948 	 */
949 	case BIOCGRTIMEOUT:
950 		{
951 			struct timeval *tv = addr;
952 
953 			tv->tv_sec = d->bd_rtout / hz;
954 			tv->tv_usec = (d->bd_rtout % hz) * tick;
955 			break;
956 		}
957 	/*
958 	 * Get packet stats.
959 	 */
960 	case BIOCGSTATS:
961 		{
962 			struct bpf_stat *bs = addr;
963 
964 			bs->bs_recv = d->bd_rcount;
965 			bs->bs_drop = d->bd_dcount;
966 			bs->bs_capt = d->bd_ccount;
967 			break;
968 		}
969 
970 	case BIOCGSTATSOLD:
971 		{
972 			struct bpf_stat_old *bs = addr;
973 
974 			bs->bs_recv = d->bd_rcount;
975 			bs->bs_drop = d->bd_dcount;
976 			break;
977 		}
978 
979 	/*
980 	 * Set immediate mode.
981 	 */
982 	case BIOCIMMEDIATE:
983 		d->bd_immediate = *(u_int *)addr;
984 		break;
985 
986 	case BIOCVERSION:
987 		{
988 			struct bpf_version *bv = addr;
989 
990 			bv->bv_major = BPF_MAJOR_VERSION;
991 			bv->bv_minor = BPF_MINOR_VERSION;
992 			break;
993 		}
994 
995 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
996 		*(u_int *)addr = d->bd_hdrcmplt;
997 		break;
998 
999 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
1000 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1001 		break;
1002 
1003 	/*
1004 	 * Get "see sent packets" flag
1005 	 */
1006 	case BIOCGSEESENT:
1007 		*(u_int *)addr = d->bd_seesent;
1008 		break;
1009 
1010 	/*
1011 	 * Set "see sent" packets flag
1012 	 */
1013 	case BIOCSSEESENT:
1014 		d->bd_seesent = *(u_int *)addr;
1015 		break;
1016 
1017 	/*
1018 	 * Set "feed packets from bpf back to input" mode
1019 	 */
1020 	case BIOCSFEEDBACK:
1021 		d->bd_feedback = *(u_int *)addr;
1022 		break;
1023 
1024 	/*
1025 	 * Get "feed packets from bpf back to input" mode
1026 	 */
1027 	case BIOCGFEEDBACK:
1028 		*(u_int *)addr = d->bd_feedback;
1029 		break;
1030 
1031 	case FIONBIO:		/* Non-blocking I/O */
1032 		/*
1033 		 * No need to do anything special as we use IO_NDELAY in
1034 		 * bpfread() as an indication of whether or not to block
1035 		 * the read.
1036 		 */
1037 		break;
1038 
1039 	case FIOASYNC:		/* Send signal on receive packets */
1040 		d->bd_async = *(int *)addr;
1041 		break;
1042 
1043 	case TIOCSPGRP:		/* Process or group to send signals to */
1044 	case FIOSETOWN:
1045 		error = fsetown(&d->bd_pgid, cmd, addr);
1046 		break;
1047 
1048 	case TIOCGPGRP:
1049 	case FIOGETOWN:
1050 		error = fgetown(d->bd_pgid, cmd, addr);
1051 		break;
1052 	}
1053 	KERNEL_UNLOCK_ONE(NULL);
1054 	return (error);
1055 }
1056 
1057 /*
1058  * Set d's packet filter program to fp.  If this file already has a filter,
1059  * free it and replace it.  Returns EINVAL for bogus requests.
1060  */
1061 int
1062 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
1063 {
1064 	struct bpf_insn *fcode, *old;
1065 	bpfjit_function_t jcode, oldj;
1066 	size_t flen, size;
1067 	int s;
1068 
1069 	jcode = NULL;
1070 	flen = fp->bf_len;
1071 
1072 	if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) {
1073 		return EINVAL;
1074 	}
1075 
1076 	if (flen) {
1077 		/*
1078 		 * Allocate the buffer, copy the byte-code from
1079 		 * userspace and validate it.
1080 		 */
1081 		size = flen * sizeof(*fp->bf_insns);
1082 		fcode = malloc(size, M_DEVBUF, M_WAITOK);
1083 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1084 		    !bpf_validate(fcode, (int)flen)) {
1085 			free(fcode, M_DEVBUF);
1086 			return EINVAL;
1087 		}
1088 		membar_consumer();
1089 		if (bpf_jit && bpfjit_module_ops.bj_generate_code != NULL) {
1090 			jcode = bpfjit_module_ops.bj_generate_code(fcode, flen);
1091 		}
1092 	} else {
1093 		fcode = NULL;
1094 	}
1095 
1096 	s = splnet();
1097 	old = d->bd_filter;
1098 	d->bd_filter = fcode;
1099 	oldj = d->bd_jitcode;
1100 	d->bd_jitcode = jcode;
1101 	reset_d(d);
1102 	splx(s);
1103 
1104 	if (old) {
1105 		free(old, M_DEVBUF);
1106 	}
1107 
1108 	if (oldj != NULL) {
1109 		KASSERT(bpfjit_module_ops.bj_free_code != NULL);
1110 		bpfjit_module_ops.bj_free_code(oldj);
1111 	}
1112 
1113 	return 0;
1114 }
1115 
1116 /*
1117  * Detach a file from its current interface (if attached at all) and attach
1118  * to the interface indicated by the name stored in ifr.
1119  * Return an errno or 0.
1120  */
1121 static int
1122 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1123 {
1124 	struct bpf_if *bp;
1125 	char *cp;
1126 	int unit_seen, i, s, error;
1127 
1128 	/*
1129 	 * Make sure the provided name has a unit number, and default
1130 	 * it to '0' if not specified.
1131 	 * XXX This is ugly ... do this differently?
1132 	 */
1133 	unit_seen = 0;
1134 	cp = ifr->ifr_name;
1135 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1136 	while (*cp++)
1137 		if (*cp >= '0' && *cp <= '9')
1138 			unit_seen = 1;
1139 	if (!unit_seen) {
1140 		/* Make sure to leave room for the '\0'. */
1141 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1142 			if ((ifr->ifr_name[i] >= 'a' &&
1143 			     ifr->ifr_name[i] <= 'z') ||
1144 			    (ifr->ifr_name[i] >= 'A' &&
1145 			     ifr->ifr_name[i] <= 'Z'))
1146 				continue;
1147 			ifr->ifr_name[i] = '0';
1148 		}
1149 	}
1150 
1151 	/*
1152 	 * Look through attached interfaces for the named one.
1153 	 */
1154 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
1155 		struct ifnet *ifp = bp->bif_ifp;
1156 
1157 		if (ifp == 0 ||
1158 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1159 			continue;
1160 		/* skip additional entry */
1161 		if (bp->bif_driverp != &ifp->if_bpf)
1162 			continue;
1163 		/*
1164 		 * We found the requested interface.
1165 		 * Allocate the packet buffers if we need to.
1166 		 * If we're already attached to requested interface,
1167 		 * just flush the buffer.
1168 		 */
1169 		if (d->bd_sbuf == 0) {
1170 			error = bpf_allocbufs(d);
1171 			if (error != 0)
1172 				return (error);
1173 		}
1174 		s = splnet();
1175 		if (bp != d->bd_bif) {
1176 			if (d->bd_bif)
1177 				/*
1178 				 * Detach if attached to something else.
1179 				 */
1180 				bpf_detachd(d);
1181 
1182 			bpf_attachd(d, bp);
1183 		}
1184 		reset_d(d);
1185 		splx(s);
1186 		return (0);
1187 	}
1188 	/* Not found. */
1189 	return (ENXIO);
1190 }
1191 
1192 /*
1193  * Copy the interface name to the ifreq.
1194  */
1195 static void
1196 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1197 {
1198 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1199 }
1200 
1201 static int
1202 bpf_stat(struct file *fp, struct stat *st)
1203 {
1204 	struct bpf_d *d = fp->f_data;
1205 
1206 	(void)memset(st, 0, sizeof(*st));
1207 	KERNEL_LOCK(1, NULL);
1208 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1209 	st->st_atimespec = d->bd_atime;
1210 	st->st_mtimespec = d->bd_mtime;
1211 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1212 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1213 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1214 	st->st_mode = S_IFCHR;
1215 	KERNEL_UNLOCK_ONE(NULL);
1216 	return 0;
1217 }
1218 
1219 /*
1220  * Support for poll() system call
1221  *
1222  * Return true iff the specific operation will not block indefinitely - with
1223  * the assumption that it is safe to positively acknowledge a request for the
1224  * ability to write to the BPF device.
1225  * Otherwise, return false but make a note that a selnotify() must be done.
1226  */
1227 static int
1228 bpf_poll(struct file *fp, int events)
1229 {
1230 	struct bpf_d *d = fp->f_data;
1231 	int s = splnet();
1232 	int revents;
1233 
1234 	/*
1235 	 * Refresh the PID associated with this bpf file.
1236 	 */
1237 	KERNEL_LOCK(1, NULL);
1238 	d->bd_pid = curproc->p_pid;
1239 
1240 	revents = events & (POLLOUT | POLLWRNORM);
1241 	if (events & (POLLIN | POLLRDNORM)) {
1242 		/*
1243 		 * An imitation of the FIONREAD ioctl code.
1244 		 */
1245 		if (d->bd_hlen != 0 ||
1246 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1247 		     d->bd_slen != 0)) {
1248 			revents |= events & (POLLIN | POLLRDNORM);
1249 		} else {
1250 			selrecord(curlwp, &d->bd_sel);
1251 			/* Start the read timeout if necessary */
1252 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1253 				callout_reset(&d->bd_callout, d->bd_rtout,
1254 					      bpf_timed_out, d);
1255 				d->bd_state = BPF_WAITING;
1256 			}
1257 		}
1258 	}
1259 
1260 	KERNEL_UNLOCK_ONE(NULL);
1261 	splx(s);
1262 	return (revents);
1263 }
1264 
1265 static void
1266 filt_bpfrdetach(struct knote *kn)
1267 {
1268 	struct bpf_d *d = kn->kn_hook;
1269 	int s;
1270 
1271 	KERNEL_LOCK(1, NULL);
1272 	s = splnet();
1273 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1274 	splx(s);
1275 	KERNEL_UNLOCK_ONE(NULL);
1276 }
1277 
1278 static int
1279 filt_bpfread(struct knote *kn, long hint)
1280 {
1281 	struct bpf_d *d = kn->kn_hook;
1282 	int rv;
1283 
1284 	KERNEL_LOCK(1, NULL);
1285 	kn->kn_data = d->bd_hlen;
1286 	if (d->bd_immediate)
1287 		kn->kn_data += d->bd_slen;
1288 	rv = (kn->kn_data > 0);
1289 	KERNEL_UNLOCK_ONE(NULL);
1290 	return rv;
1291 }
1292 
1293 static const struct filterops bpfread_filtops =
1294 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1295 
1296 static int
1297 bpf_kqfilter(struct file *fp, struct knote *kn)
1298 {
1299 	struct bpf_d *d = fp->f_data;
1300 	struct klist *klist;
1301 	int s;
1302 
1303 	KERNEL_LOCK(1, NULL);
1304 
1305 	switch (kn->kn_filter) {
1306 	case EVFILT_READ:
1307 		klist = &d->bd_sel.sel_klist;
1308 		kn->kn_fop = &bpfread_filtops;
1309 		break;
1310 
1311 	default:
1312 		KERNEL_UNLOCK_ONE(NULL);
1313 		return (EINVAL);
1314 	}
1315 
1316 	kn->kn_hook = d;
1317 
1318 	s = splnet();
1319 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1320 	splx(s);
1321 	KERNEL_UNLOCK_ONE(NULL);
1322 
1323 	return (0);
1324 }
1325 
1326 /*
1327  * Copy data from an mbuf chain into a buffer.  This code is derived
1328  * from m_copydata in sys/uipc_mbuf.c.
1329  */
1330 static void *
1331 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1332 {
1333 	const struct mbuf *m;
1334 	u_int count;
1335 	u_char *dst;
1336 
1337 	m = src_arg;
1338 	dst = dst_arg;
1339 	while (len > 0) {
1340 		if (m == NULL)
1341 			panic("bpf_mcpy");
1342 		count = min(m->m_len, len);
1343 		memcpy(dst, mtod(m, const void *), count);
1344 		m = m->m_next;
1345 		dst += count;
1346 		len -= count;
1347 	}
1348 	return dst_arg;
1349 }
1350 
1351 /*
1352  * Dispatch a packet to all the listeners on interface bp.
1353  *
1354  * pkt     pointer to the packet, either a data buffer or an mbuf chain
1355  * buflen  buffer length, if pkt is a data buffer
1356  * cpfn    a function that can copy pkt into the listener's buffer
1357  * pktlen  length of the packet
1358  * rcv     true if packet came in
1359  */
1360 static inline void
1361 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1362     void *pkt, u_int pktlen, u_int buflen, const bool rcv)
1363 {
1364 	struct bpf_d *d;
1365 	struct timespec ts;
1366 	bool gottime = false;
1367 
1368 	/*
1369 	 * Note that the IPL does not have to be raised at this point.
1370 	 * The only problem that could arise here is that if two different
1371 	 * interfaces shared any data.  This is not the case.
1372 	 */
1373 	for (d = bp->bif_dlist; d != NULL; d = d->bd_next) {
1374 		u_int slen;
1375 
1376 		if (!d->bd_seesent && !rcv) {
1377 			continue;
1378 		}
1379 		d->bd_rcount++;
1380 		bpf_gstats.bs_recv++;
1381 
1382 		if (d->bd_jitcode != NULL)
1383 			slen = d->bd_jitcode(pkt, pktlen, buflen);
1384 		else
1385 			slen = bpf_filter(d->bd_filter, pkt, pktlen, buflen);
1386 
1387 		if (!slen) {
1388 			continue;
1389 		}
1390 		if (!gottime) {
1391 			gottime = true;
1392 			nanotime(&ts);
1393 		}
1394 		catchpacket(d, pkt, pktlen, slen, cpfn, &ts);
1395 	}
1396 }
1397 
1398 /*
1399  * Incoming linkage from device drivers.  Process the packet pkt, of length
1400  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1401  * by each process' filter, and if accepted, stashed into the corresponding
1402  * buffer.
1403  */
1404 static void
1405 _bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1406 {
1407 
1408 	bpf_deliver(bp, memcpy, pkt, pktlen, pktlen, true);
1409 }
1410 
1411 /*
1412  * Incoming linkage from device drivers, when the head of the packet is in
1413  * a buffer, and the tail is in an mbuf chain.
1414  */
1415 static void
1416 _bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1417 {
1418 	u_int pktlen;
1419 	struct mbuf mb;
1420 
1421 	/* Skip outgoing duplicate packets. */
1422 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1423 		m->m_flags &= ~M_PROMISC;
1424 		return;
1425 	}
1426 
1427 	pktlen = m_length(m) + dlen;
1428 
1429 	/*
1430 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1431 	 * Note that we cut corners here; we only setup what's
1432 	 * absolutely needed--this mbuf should never go anywhere else.
1433 	 */
1434 	(void)memset(&mb, 0, sizeof(mb));
1435 	mb.m_next = m;
1436 	mb.m_data = data;
1437 	mb.m_len = dlen;
1438 
1439 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif != NULL);
1440 }
1441 
1442 /*
1443  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1444  */
1445 static void
1446 _bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1447 {
1448 	void *(*cpfn)(void *, const void *, size_t);
1449 	u_int pktlen, buflen;
1450 	void *marg;
1451 
1452 	/* Skip outgoing duplicate packets. */
1453 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1454 		m->m_flags &= ~M_PROMISC;
1455 		return;
1456 	}
1457 
1458 	pktlen = m_length(m);
1459 
1460 	if (pktlen == m->m_len) {
1461 		cpfn = (void *)memcpy;
1462 		marg = mtod(m, void *);
1463 		buflen = pktlen;
1464 	} else {
1465 		cpfn = bpf_mcpy;
1466 		marg = m;
1467 		buflen = 0;
1468 	}
1469 
1470 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif != NULL);
1471 }
1472 
1473 /*
1474  * We need to prepend the address family as
1475  * a four byte field.  Cons up a dummy header
1476  * to pacify bpf.  This is safe because bpf
1477  * will only read from the mbuf (i.e., it won't
1478  * try to free it or keep a pointer a to it).
1479  */
1480 static void
1481 _bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m)
1482 {
1483 	struct mbuf m0;
1484 
1485 	m0.m_flags = 0;
1486 	m0.m_next = m;
1487 	m0.m_len = 4;
1488 	m0.m_data = (char *)&af;
1489 
1490 	_bpf_mtap(bp, &m0);
1491 }
1492 
1493 /*
1494  * Put the SLIP pseudo-"link header" in place.
1495  * Note this M_PREPEND() should never fail,
1496  * swince we know we always have enough space
1497  * in the input buffer.
1498  */
1499 static void
1500 _bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m)
1501 {
1502 	int s;
1503 	u_char *hp;
1504 
1505 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1506 	if (*m == NULL)
1507 		return;
1508 
1509 	hp = mtod(*m, u_char *);
1510 	hp[SLX_DIR] = SLIPDIR_IN;
1511 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1512 
1513 	s = splnet();
1514 	_bpf_mtap(bp, *m);
1515 	splx(s);
1516 
1517 	m_adj(*m, SLIP_HDRLEN);
1518 }
1519 
1520 /*
1521  * Put the SLIP pseudo-"link header" in
1522  * place.  The compressed header is now
1523  * at the beginning of the mbuf.
1524  */
1525 static void
1526 _bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m)
1527 {
1528 	struct mbuf m0;
1529 	u_char *hp;
1530 	int s;
1531 
1532 	m0.m_flags = 0;
1533 	m0.m_next = m;
1534 	m0.m_data = m0.m_dat;
1535 	m0.m_len = SLIP_HDRLEN;
1536 
1537 	hp = mtod(&m0, u_char *);
1538 
1539 	hp[SLX_DIR] = SLIPDIR_OUT;
1540 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1541 
1542 	s = splnet();
1543 	_bpf_mtap(bp, &m0);
1544 	splx(s);
1545 	m_freem(m);
1546 }
1547 
1548 static int
1549 bpf_hdrlen(struct bpf_d *d)
1550 {
1551 	int hdrlen = d->bd_bif->bif_hdrlen;
1552 	/*
1553 	 * Compute the length of the bpf header.  This is not necessarily
1554 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1555 	 * that the network layer header begins on a longword boundary (for
1556 	 * performance reasons and to alleviate alignment restrictions).
1557 	 */
1558 #ifdef _LP64
1559 	if (d->bd_compat32)
1560 		return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen);
1561 	else
1562 #endif
1563 		return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen);
1564 }
1565 
1566 /*
1567  * Move the packet data from interface memory (pkt) into the
1568  * store buffer. Call the wakeup functions if it's time to wakeup
1569  * a listener (buffer full), "cpfn" is the routine called to do the
1570  * actual data transfer. memcpy is passed in to copy contiguous chunks,
1571  * while bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1572  * pkt is really an mbuf.
1573  */
1574 static void
1575 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1576     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1577 {
1578 	struct bpf_hdr *hp;
1579 #ifdef _LP64
1580 	struct bpf_hdr32 *hp32;
1581 #endif
1582 	int totlen, curlen;
1583 	int hdrlen = bpf_hdrlen(d);
1584 	int do_wakeup = 0;
1585 
1586 	++d->bd_ccount;
1587 	++bpf_gstats.bs_capt;
1588 	/*
1589 	 * Figure out how many bytes to move.  If the packet is
1590 	 * greater or equal to the snapshot length, transfer that
1591 	 * much.  Otherwise, transfer the whole packet (unless
1592 	 * we hit the buffer size limit).
1593 	 */
1594 	totlen = hdrlen + min(snaplen, pktlen);
1595 	if (totlen > d->bd_bufsize)
1596 		totlen = d->bd_bufsize;
1597 
1598 	/*
1599 	 * Round up the end of the previous packet to the next longword.
1600 	 */
1601 #ifdef _LP64
1602 	if (d->bd_compat32)
1603 		curlen = BPF_WORDALIGN32(d->bd_slen);
1604 	else
1605 #endif
1606 		curlen = BPF_WORDALIGN(d->bd_slen);
1607 	if (curlen + totlen > d->bd_bufsize) {
1608 		/*
1609 		 * This packet will overflow the storage buffer.
1610 		 * Rotate the buffers if we can, then wakeup any
1611 		 * pending reads.
1612 		 */
1613 		if (d->bd_fbuf == 0) {
1614 			/*
1615 			 * We haven't completed the previous read yet,
1616 			 * so drop the packet.
1617 			 */
1618 			++d->bd_dcount;
1619 			++bpf_gstats.bs_drop;
1620 			return;
1621 		}
1622 		ROTATE_BUFFERS(d);
1623 		do_wakeup = 1;
1624 		curlen = 0;
1625 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1626 		/*
1627 		 * Immediate mode is set, or the read timeout has
1628 		 * already expired during a select call.  A packet
1629 		 * arrived, so the reader should be woken up.
1630 		 */
1631 		do_wakeup = 1;
1632 	}
1633 
1634 	/*
1635 	 * Append the bpf header.
1636 	 */
1637 #ifdef _LP64
1638 	if (d->bd_compat32) {
1639 		hp32 = (struct bpf_hdr32 *)((char *)d->bd_sbuf + curlen);
1640 		hp32->bh_tstamp.tv_sec = ts->tv_sec;
1641 		hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1642 		hp32->bh_datalen = pktlen;
1643 		hp32->bh_hdrlen = hdrlen;
1644 		/*
1645 		 * Copy the packet data into the store buffer and update its length.
1646 		 */
1647 		(*cpfn)((u_char *)hp32 + hdrlen, pkt,
1648 		    (hp32->bh_caplen = totlen - hdrlen));
1649 	} else
1650 #endif
1651 	{
1652 		hp = (struct bpf_hdr *)((char *)d->bd_sbuf + curlen);
1653 		hp->bh_tstamp.tv_sec = ts->tv_sec;
1654 		hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1655 		hp->bh_datalen = pktlen;
1656 		hp->bh_hdrlen = hdrlen;
1657 		/*
1658 		 * Copy the packet data into the store buffer and update
1659 		 * its length.
1660 		 */
1661 		(*cpfn)((u_char *)hp + hdrlen, pkt,
1662 		    (hp->bh_caplen = totlen - hdrlen));
1663 	}
1664 	d->bd_slen = curlen + totlen;
1665 
1666 	/*
1667 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1668 	 * will cause filt_bpfread() to be called with it adjusted.
1669 	 */
1670 	if (do_wakeup)
1671 		bpf_wakeup(d);
1672 }
1673 
1674 /*
1675  * Initialize all nonzero fields of a descriptor.
1676  */
1677 static int
1678 bpf_allocbufs(struct bpf_d *d)
1679 {
1680 
1681 	d->bd_fbuf = malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK | M_CANFAIL);
1682 	if (!d->bd_fbuf)
1683 		return (ENOBUFS);
1684 	d->bd_sbuf = malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK | M_CANFAIL);
1685 	if (!d->bd_sbuf) {
1686 		free(d->bd_fbuf, M_DEVBUF);
1687 		return (ENOBUFS);
1688 	}
1689 	d->bd_slen = 0;
1690 	d->bd_hlen = 0;
1691 	return (0);
1692 }
1693 
1694 /*
1695  * Free buffers currently in use by a descriptor.
1696  * Called on close.
1697  */
1698 static void
1699 bpf_freed(struct bpf_d *d)
1700 {
1701 	/*
1702 	 * We don't need to lock out interrupts since this descriptor has
1703 	 * been detached from its interface and it yet hasn't been marked
1704 	 * free.
1705 	 */
1706 	if (d->bd_sbuf != NULL) {
1707 		free(d->bd_sbuf, M_DEVBUF);
1708 		if (d->bd_hbuf != NULL)
1709 			free(d->bd_hbuf, M_DEVBUF);
1710 		if (d->bd_fbuf != NULL)
1711 			free(d->bd_fbuf, M_DEVBUF);
1712 	}
1713 	if (d->bd_filter)
1714 		free(d->bd_filter, M_DEVBUF);
1715 
1716 	if (d->bd_jitcode != NULL) {
1717 		KASSERT(bpfjit_module_ops.bj_free_code != NULL);
1718 		bpfjit_module_ops.bj_free_code(d->bd_jitcode);
1719 	}
1720 }
1721 
1722 /*
1723  * Attach an interface to bpf.  dlt is the link layer type;
1724  * hdrlen is the fixed size of the link header for the specified dlt
1725  * (variable length headers not yet supported).
1726  */
1727 static void
1728 _bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1729 {
1730 	struct bpf_if *bp;
1731 	bp = malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1732 	if (bp == 0)
1733 		panic("bpfattach");
1734 
1735 	bp->bif_dlist = 0;
1736 	bp->bif_driverp = driverp;
1737 	bp->bif_ifp = ifp;
1738 	bp->bif_dlt = dlt;
1739 
1740 	bp->bif_next = bpf_iflist;
1741 	bpf_iflist = bp;
1742 
1743 	*bp->bif_driverp = 0;
1744 
1745 	bp->bif_hdrlen = hdrlen;
1746 #if 0
1747 	printf("bpf: %s attached\n", ifp->if_xname);
1748 #endif
1749 }
1750 
1751 /*
1752  * Remove an interface from bpf.
1753  */
1754 static void
1755 _bpfdetach(struct ifnet *ifp)
1756 {
1757 	struct bpf_if *bp, **pbp;
1758 	struct bpf_d *d;
1759 	int s;
1760 
1761 	/* Nuke the vnodes for any open instances */
1762 	LIST_FOREACH(d, &bpf_list, bd_list) {
1763 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
1764 			/*
1765 			 * Detach the descriptor from an interface now.
1766 			 * It will be free'ed later by close routine.
1767 			 */
1768 			s = splnet();
1769 			d->bd_promisc = 0;	/* we can't touch device. */
1770 			bpf_detachd(d);
1771 			splx(s);
1772 		}
1773 	}
1774 
1775   again:
1776 	for (bp = bpf_iflist, pbp = &bpf_iflist;
1777 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
1778 		if (bp->bif_ifp == ifp) {
1779 			*pbp = bp->bif_next;
1780 			free(bp, M_DEVBUF);
1781 			goto again;
1782 		}
1783 	}
1784 }
1785 
1786 /*
1787  * Change the data link type of a interface.
1788  */
1789 static void
1790 _bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1791 {
1792 	struct bpf_if *bp;
1793 
1794 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1795 		if (bp->bif_driverp == &ifp->if_bpf)
1796 			break;
1797 	}
1798 	if (bp == NULL)
1799 		panic("bpf_change_type");
1800 
1801 	bp->bif_dlt = dlt;
1802 
1803 	bp->bif_hdrlen = hdrlen;
1804 }
1805 
1806 /*
1807  * Get a list of available data link type of the interface.
1808  */
1809 static int
1810 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1811 {
1812 	int n, error;
1813 	struct ifnet *ifp;
1814 	struct bpf_if *bp;
1815 
1816 	ifp = d->bd_bif->bif_ifp;
1817 	n = 0;
1818 	error = 0;
1819 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1820 		if (bp->bif_ifp != ifp)
1821 			continue;
1822 		if (bfl->bfl_list != NULL) {
1823 			if (n >= bfl->bfl_len)
1824 				return ENOMEM;
1825 			error = copyout(&bp->bif_dlt,
1826 			    bfl->bfl_list + n, sizeof(u_int));
1827 		}
1828 		n++;
1829 	}
1830 	bfl->bfl_len = n;
1831 	return error;
1832 }
1833 
1834 /*
1835  * Set the data link type of a BPF instance.
1836  */
1837 static int
1838 bpf_setdlt(struct bpf_d *d, u_int dlt)
1839 {
1840 	int s, error, opromisc;
1841 	struct ifnet *ifp;
1842 	struct bpf_if *bp;
1843 
1844 	if (d->bd_bif->bif_dlt == dlt)
1845 		return 0;
1846 	ifp = d->bd_bif->bif_ifp;
1847 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1848 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1849 			break;
1850 	}
1851 	if (bp == NULL)
1852 		return EINVAL;
1853 	s = splnet();
1854 	opromisc = d->bd_promisc;
1855 	bpf_detachd(d);
1856 	bpf_attachd(d, bp);
1857 	reset_d(d);
1858 	if (opromisc) {
1859 		error = ifpromisc(bp->bif_ifp, 1);
1860 		if (error)
1861 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
1862 			    bp->bif_ifp->if_xname, error);
1863 		else
1864 			d->bd_promisc = 1;
1865 	}
1866 	splx(s);
1867 	return 0;
1868 }
1869 
1870 static int
1871 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
1872 {
1873 	int newsize, error;
1874 	struct sysctlnode node;
1875 
1876 	node = *rnode;
1877 	node.sysctl_data = &newsize;
1878 	newsize = bpf_maxbufsize;
1879 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1880 	if (error || newp == NULL)
1881 		return (error);
1882 
1883 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
1884 		return (EINVAL);
1885 
1886 	bpf_maxbufsize = newsize;
1887 
1888 	return (0);
1889 }
1890 
1891 static int
1892 sysctl_net_bpf_jit(SYSCTLFN_ARGS)
1893 {
1894 	bool newval;
1895 	int error;
1896 	struct sysctlnode node;
1897 
1898 	node = *rnode;
1899 	node.sysctl_data = &newval;
1900 	newval = bpf_jit;
1901 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1902 	if (error != 0 || newp == NULL)
1903 		return error;
1904 
1905 	bpf_jit = newval;
1906 
1907 	/*
1908 	 * Do a full sync to publish new bpf_jit value and
1909 	 * update bpfjit_module_ops.bj_generate_code variable.
1910 	 */
1911 	membar_sync();
1912 
1913 	if (newval && bpfjit_module_ops.bj_generate_code == NULL) {
1914 		printf("WARNING jit activation is postponed "
1915 		    "until after bpfjit module is loaded\n");
1916 	}
1917 
1918 	return 0;
1919 }
1920 
1921 static int
1922 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
1923 {
1924 	int    error, elem_count;
1925 	struct bpf_d	 *dp;
1926 	struct bpf_d_ext  dpe;
1927 	size_t len, needed, elem_size, out_size;
1928 	char   *sp;
1929 
1930 	if (namelen == 1 && name[0] == CTL_QUERY)
1931 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
1932 
1933 	if (namelen != 2)
1934 		return (EINVAL);
1935 
1936 	/* BPF peers is privileged information. */
1937 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
1938 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
1939 	if (error)
1940 		return (EPERM);
1941 
1942 	len = (oldp != NULL) ? *oldlenp : 0;
1943 	sp = oldp;
1944 	elem_size = name[0];
1945 	elem_count = name[1];
1946 	out_size = MIN(sizeof(dpe), elem_size);
1947 	needed = 0;
1948 
1949 	if (elem_size < 1 || elem_count < 0)
1950 		return (EINVAL);
1951 
1952 	mutex_enter(&bpf_mtx);
1953 	LIST_FOREACH(dp, &bpf_list, bd_list) {
1954 		if (len >= elem_size && elem_count > 0) {
1955 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
1956 			BPF_EXT(bufsize);
1957 			BPF_EXT(promisc);
1958 			BPF_EXT(state);
1959 			BPF_EXT(immediate);
1960 			BPF_EXT(hdrcmplt);
1961 			BPF_EXT(seesent);
1962 			BPF_EXT(pid);
1963 			BPF_EXT(rcount);
1964 			BPF_EXT(dcount);
1965 			BPF_EXT(ccount);
1966 #undef BPF_EXT
1967 			if (dp->bd_bif)
1968 				(void)strlcpy(dpe.bde_ifname,
1969 				    dp->bd_bif->bif_ifp->if_xname,
1970 				    IFNAMSIZ - 1);
1971 			else
1972 				dpe.bde_ifname[0] = '\0';
1973 
1974 			error = copyout(&dpe, sp, out_size);
1975 			if (error)
1976 				break;
1977 			sp += elem_size;
1978 			len -= elem_size;
1979 		}
1980 		needed += elem_size;
1981 		if (elem_count > 0 && elem_count != INT_MAX)
1982 			elem_count--;
1983 	}
1984 	mutex_exit(&bpf_mtx);
1985 
1986 	*oldlenp = needed;
1987 
1988 	return (error);
1989 }
1990 
1991 static struct sysctllog *bpf_sysctllog;
1992 static void
1993 sysctl_net_bpf_setup(void)
1994 {
1995 	const struct sysctlnode *node;
1996 
1997 	sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
1998 		       CTLFLAG_PERMANENT,
1999 		       CTLTYPE_NODE, "net", NULL,
2000 		       NULL, 0, NULL, 0,
2001 		       CTL_NET, CTL_EOL);
2002 
2003 	node = NULL;
2004 	sysctl_createv(&bpf_sysctllog, 0, NULL, &node,
2005 		       CTLFLAG_PERMANENT,
2006 		       CTLTYPE_NODE, "bpf",
2007 		       SYSCTL_DESCR("BPF options"),
2008 		       NULL, 0, NULL, 0,
2009 		       CTL_NET, CTL_CREATE, CTL_EOL);
2010 	if (node != NULL) {
2011 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2012 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2013 			CTLTYPE_BOOL, "jit",
2014 			SYSCTL_DESCR("Toggle Just-In-Time compilation"),
2015 			sysctl_net_bpf_jit, 0, &bpf_jit, 0,
2016 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2017 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2018 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2019 			CTLTYPE_INT, "maxbufsize",
2020 			SYSCTL_DESCR("Maximum size for data capture buffer"),
2021 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
2022 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2023 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2024 			CTLFLAG_PERMANENT,
2025 			CTLTYPE_STRUCT, "stats",
2026 			SYSCTL_DESCR("BPF stats"),
2027 			NULL, 0, &bpf_gstats, sizeof(bpf_gstats),
2028 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2029 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2030 			CTLFLAG_PERMANENT,
2031 			CTLTYPE_STRUCT, "peers",
2032 			SYSCTL_DESCR("BPF peers"),
2033 			sysctl_net_bpf_peers, 0, NULL, 0,
2034 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2035 	}
2036 
2037 }
2038 
2039 struct bpf_ops bpf_ops_kernel = {
2040 	.bpf_attach =		_bpfattach,
2041 	.bpf_detach =		_bpfdetach,
2042 	.bpf_change_type =	_bpf_change_type,
2043 
2044 	.bpf_tap =		_bpf_tap,
2045 	.bpf_mtap =		_bpf_mtap,
2046 	.bpf_mtap2 =		_bpf_mtap2,
2047 	.bpf_mtap_af =		_bpf_mtap_af,
2048 	.bpf_mtap_sl_in =	_bpf_mtap_sl_in,
2049 	.bpf_mtap_sl_out =	_bpf_mtap_sl_out,
2050 };
2051 
2052 MODULE(MODULE_CLASS_DRIVER, bpf, NULL);
2053 
2054 static int
2055 bpf_modcmd(modcmd_t cmd, void *arg)
2056 {
2057 	devmajor_t bmajor, cmajor;
2058 	int error;
2059 
2060 	bmajor = cmajor = NODEVMAJOR;
2061 
2062 	switch (cmd) {
2063 	case MODULE_CMD_INIT:
2064 		bpfilterattach(0);
2065 		error = devsw_attach("bpf", NULL, &bmajor,
2066 		    &bpf_cdevsw, &cmajor);
2067 		if (error == EEXIST)
2068 			error = 0; /* maybe built-in ... improve eventually */
2069 		if (error)
2070 			break;
2071 
2072 		bpf_ops_handover_enter(&bpf_ops_kernel);
2073 		atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel);
2074 		bpf_ops_handover_exit();
2075 		sysctl_net_bpf_setup();
2076 		break;
2077 
2078 	case MODULE_CMD_FINI:
2079 		/*
2080 		 * While there is no reference counting for bpf callers,
2081 		 * unload could at least in theory be done similarly to
2082 		 * system call disestablishment.  This should even be
2083 		 * a little simpler:
2084 		 *
2085 		 * 1) replace op vector with stubs
2086 		 * 2) post update to all cpus with xc
2087 		 * 3) check that nobody is in bpf anymore
2088 		 *    (it's doubtful we'd want something like l_sysent,
2089 		 *     but we could do something like *signed* percpu
2090 		 *     counters.  if the sum is 0, we're good).
2091 		 * 4) if fail, unroll changes
2092 		 *
2093 		 * NOTE: change won't be atomic to the outside.  some
2094 		 * packets may be not captured even if unload is
2095 		 * not succesful.  I think packet capture not working
2096 		 * is a perfectly logical consequence of trying to
2097 		 * disable packet capture.
2098 		 */
2099 		error = EOPNOTSUPP;
2100 		/* insert sysctl teardown */
2101 		break;
2102 
2103 	default:
2104 		error = ENOTTY;
2105 		break;
2106 	}
2107 
2108 	return error;
2109 }
2110