xref: /netbsd-src/sys/net/bpf.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /*	$NetBSD: bpf.c,v 1.190 2014/12/29 13:38:13 ozaki-r Exp $	*/
2 
3 /*
4  * Copyright (c) 1990, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from the Stanford/CMU enet packet filter,
8  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
9  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
10  * Berkeley Laboratory.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)bpf.c	8.4 (Berkeley) 1/9/95
37  * static char rcsid[] =
38  * "Header: bpf.c,v 1.67 96/09/26 22:00:52 leres Exp ";
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: bpf.c,v 1.190 2014/12/29 13:38:13 ozaki-r 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 	.d_open = bpfopen,
185 	.d_close = noclose,
186 	.d_read = noread,
187 	.d_write = nowrite,
188 	.d_ioctl = noioctl,
189 	.d_stop = nostop,
190 	.d_tty = notty,
191 	.d_poll = nopoll,
192 	.d_mmap = nommap,
193 	.d_kqfilter = nokqfilter,
194 	.d_discard = nodiscard,
195 	.d_flag = D_OTHER
196 };
197 
198 bpfjit_func_t
199 bpf_jit_generate(bpf_ctx_t *bc, void *code, size_t size)
200 {
201 
202 	membar_consumer();
203 	if (bpfjit_module_ops.bj_generate_code != NULL) {
204 		return bpfjit_module_ops.bj_generate_code(bc, code, size);
205 	}
206 	return NULL;
207 }
208 
209 void
210 bpf_jit_freecode(bpfjit_func_t jcode)
211 {
212 	KASSERT(bpfjit_module_ops.bj_free_code != NULL);
213 	bpfjit_module_ops.bj_free_code(jcode);
214 }
215 
216 static int
217 bpf_movein(struct uio *uio, int linktype, uint64_t mtu, struct mbuf **mp,
218 	   struct sockaddr *sockp)
219 {
220 	struct mbuf *m;
221 	int error;
222 	size_t len;
223 	size_t hlen;
224 	size_t align;
225 
226 	/*
227 	 * Build a sockaddr based on the data link layer type.
228 	 * We do this at this level because the ethernet header
229 	 * is copied directly into the data field of the sockaddr.
230 	 * In the case of SLIP, there is no header and the packet
231 	 * is forwarded as is.
232 	 * Also, we are careful to leave room at the front of the mbuf
233 	 * for the link level header.
234 	 */
235 	switch (linktype) {
236 
237 	case DLT_SLIP:
238 		sockp->sa_family = AF_INET;
239 		hlen = 0;
240 		align = 0;
241 		break;
242 
243 	case DLT_PPP:
244 		sockp->sa_family = AF_UNSPEC;
245 		hlen = 0;
246 		align = 0;
247 		break;
248 
249 	case DLT_EN10MB:
250 		sockp->sa_family = AF_UNSPEC;
251 		/* XXX Would MAXLINKHDR be better? */
252  		/* 6(dst)+6(src)+2(type) */
253 		hlen = sizeof(struct ether_header);
254 		align = 2;
255 		break;
256 
257 	case DLT_ARCNET:
258 		sockp->sa_family = AF_UNSPEC;
259 		hlen = ARC_HDRLEN;
260 		align = 5;
261 		break;
262 
263 	case DLT_FDDI:
264 		sockp->sa_family = AF_LINK;
265 		/* XXX 4(FORMAC)+6(dst)+6(src) */
266 		hlen = 16;
267 		align = 0;
268 		break;
269 
270 	case DLT_ECONET:
271 		sockp->sa_family = AF_UNSPEC;
272 		hlen = 6;
273 		align = 2;
274 		break;
275 
276 	case DLT_NULL:
277 		sockp->sa_family = AF_UNSPEC;
278 		hlen = 0;
279 		align = 0;
280 		break;
281 
282 	default:
283 		return (EIO);
284 	}
285 
286 	len = uio->uio_resid;
287 	/*
288 	 * If there aren't enough bytes for a link level header or the
289 	 * packet length exceeds the interface mtu, return an error.
290 	 */
291 	if (len - hlen > mtu)
292 		return (EMSGSIZE);
293 
294 	/*
295 	 * XXX Avoid complicated buffer chaining ---
296 	 * bail if it won't fit in a single mbuf.
297 	 * (Take into account possible alignment bytes)
298 	 */
299 	if (len + align > MCLBYTES)
300 		return (EIO);
301 
302 	m = m_gethdr(M_WAIT, MT_DATA);
303 	m->m_pkthdr.rcvif = NULL;
304 	m->m_pkthdr.len = (int)(len - hlen);
305 	if (len + align > MHLEN) {
306 		m_clget(m, M_WAIT);
307 		if ((m->m_flags & M_EXT) == 0) {
308 			error = ENOBUFS;
309 			goto bad;
310 		}
311 	}
312 
313 	/* Insure the data is properly aligned */
314 	if (align > 0) {
315 		m->m_data += align;
316 		m->m_len -= (int)align;
317 	}
318 
319 	error = uiomove(mtod(m, void *), len, uio);
320 	if (error)
321 		goto bad;
322 	if (hlen != 0) {
323 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
324 		m->m_data += hlen; /* XXX */
325 		len -= hlen;
326 	}
327 	m->m_len = (int)len;
328 	*mp = m;
329 	return (0);
330 
331 bad:
332 	m_freem(m);
333 	return (error);
334 }
335 
336 /*
337  * Attach file to the bpf interface, i.e. make d listen on bp.
338  * Must be called at splnet.
339  */
340 static void
341 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
342 {
343 	/*
344 	 * Point d at bp, and add d to the interface's list of listeners.
345 	 * Finally, point the driver's bpf cookie at the interface so
346 	 * it will divert packets to bpf.
347 	 */
348 	d->bd_bif = bp;
349 	d->bd_next = bp->bif_dlist;
350 	bp->bif_dlist = d;
351 
352 	*bp->bif_driverp = bp;
353 }
354 
355 /*
356  * Detach a file from its interface.
357  */
358 static void
359 bpf_detachd(struct bpf_d *d)
360 {
361 	struct bpf_d **p;
362 	struct bpf_if *bp;
363 
364 	bp = d->bd_bif;
365 	/*
366 	 * Check if this descriptor had requested promiscuous mode.
367 	 * If so, turn it off.
368 	 */
369 	if (d->bd_promisc) {
370 		int error __diagused;
371 
372 		d->bd_promisc = 0;
373 		/*
374 		 * Take device out of promiscuous mode.  Since we were
375 		 * able to enter promiscuous mode, we should be able
376 		 * to turn it off.  But we can get an error if
377 		 * the interface was configured down, so only panic
378 		 * if we don't get an unexpected error.
379 		 */
380   		error = ifpromisc(bp->bif_ifp, 0);
381 #ifdef DIAGNOSTIC
382 		if (error)
383 			printf("%s: ifpromisc failed: %d", __func__, error);
384 #endif
385 	}
386 	/* Remove d from the interface's descriptor list. */
387 	p = &bp->bif_dlist;
388 	while (*p != d) {
389 		p = &(*p)->bd_next;
390 		if (*p == NULL)
391 			panic("%s: descriptor not in list", __func__);
392 	}
393 	*p = (*p)->bd_next;
394 	if (bp->bif_dlist == NULL)
395 		/*
396 		 * Let the driver know that there are no more listeners.
397 		 */
398 		*d->bd_bif->bif_driverp = NULL;
399 	d->bd_bif = NULL;
400 }
401 
402 static int
403 doinit(void)
404 {
405 
406 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
407 
408 	LIST_INIT(&bpf_list);
409 
410 	bpf_gstats.bs_recv = 0;
411 	bpf_gstats.bs_drop = 0;
412 	bpf_gstats.bs_capt = 0;
413 
414 	return 0;
415 }
416 
417 /*
418  * bpfilterattach() is called at boot time.
419  */
420 /* ARGSUSED */
421 void
422 bpfilterattach(int n)
423 {
424 	static ONCE_DECL(control);
425 
426 	RUN_ONCE(&control, doinit);
427 }
428 
429 /*
430  * Open ethernet device. Clones.
431  */
432 /* ARGSUSED */
433 int
434 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
435 {
436 	struct bpf_d *d;
437 	struct file *fp;
438 	int error, fd;
439 
440 	/* falloc() will use the descriptor for us. */
441 	if ((error = fd_allocfile(&fp, &fd)) != 0)
442 		return error;
443 
444 	d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK|M_ZERO);
445 	d->bd_bufsize = bpf_bufsize;
446 	d->bd_seesent = 1;
447 	d->bd_feedback = 0;
448 	d->bd_pid = l->l_proc->p_pid;
449 #ifdef _LP64
450 	if (curproc->p_flag & PK_32)
451 		d->bd_compat32 = 1;
452 #endif
453 	getnanotime(&d->bd_btime);
454 	d->bd_atime = d->bd_mtime = d->bd_btime;
455 	callout_init(&d->bd_callout, 0);
456 	selinit(&d->bd_sel);
457 	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
458 	d->bd_jitcode = NULL;
459 
460 	mutex_enter(&bpf_mtx);
461 	LIST_INSERT_HEAD(&bpf_list, d, bd_list);
462 	mutex_exit(&bpf_mtx);
463 
464 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
465 }
466 
467 /*
468  * Close the descriptor by detaching it from its interface,
469  * deallocating its buffers, and marking it free.
470  */
471 /* ARGSUSED */
472 static int
473 bpf_close(struct file *fp)
474 {
475 	struct bpf_d *d = fp->f_bpf;
476 	int s;
477 
478 	KERNEL_LOCK(1, NULL);
479 
480 	/*
481 	 * Refresh the PID associated with this bpf file.
482 	 */
483 	d->bd_pid = curproc->p_pid;
484 
485 	s = splnet();
486 	if (d->bd_state == BPF_WAITING)
487 		callout_stop(&d->bd_callout);
488 	d->bd_state = BPF_IDLE;
489 	if (d->bd_bif)
490 		bpf_detachd(d);
491 	splx(s);
492 	bpf_freed(d);
493 	mutex_enter(&bpf_mtx);
494 	LIST_REMOVE(d, bd_list);
495 	mutex_exit(&bpf_mtx);
496 	callout_destroy(&d->bd_callout);
497 	seldestroy(&d->bd_sel);
498 	softint_disestablish(d->bd_sih);
499 	free(d, M_DEVBUF);
500 	fp->f_bpf = NULL;
501 
502 	KERNEL_UNLOCK_ONE(NULL);
503 
504 	return (0);
505 }
506 
507 /*
508  * Rotate the packet buffers in descriptor d.  Move the store buffer
509  * into the hold slot, and the free buffer into the store slot.
510  * Zero the length of the new store buffer.
511  */
512 #define ROTATE_BUFFERS(d) \
513 	(d)->bd_hbuf = (d)->bd_sbuf; \
514 	(d)->bd_hlen = (d)->bd_slen; \
515 	(d)->bd_sbuf = (d)->bd_fbuf; \
516 	(d)->bd_slen = 0; \
517 	(d)->bd_fbuf = NULL;
518 /*
519  *  bpfread - read next chunk of packets from buffers
520  */
521 static int
522 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
523     kauth_cred_t cred, int flags)
524 {
525 	struct bpf_d *d = fp->f_bpf;
526 	int timed_out;
527 	int error;
528 	int s;
529 
530 	getnanotime(&d->bd_atime);
531 	/*
532 	 * Restrict application to use a buffer the same size as
533 	 * the kernel buffers.
534 	 */
535 	if (uio->uio_resid != d->bd_bufsize)
536 		return (EINVAL);
537 
538 	KERNEL_LOCK(1, NULL);
539 	s = splnet();
540 	if (d->bd_state == BPF_WAITING)
541 		callout_stop(&d->bd_callout);
542 	timed_out = (d->bd_state == BPF_TIMED_OUT);
543 	d->bd_state = BPF_IDLE;
544 	/*
545 	 * If the hold buffer is empty, then do a timed sleep, which
546 	 * ends when the timeout expires or when enough packets
547 	 * have arrived to fill the store buffer.
548 	 */
549 	while (d->bd_hbuf == NULL) {
550 		if (fp->f_flag & FNONBLOCK) {
551 			if (d->bd_slen == 0) {
552 				splx(s);
553 				KERNEL_UNLOCK_ONE(NULL);
554 				return (EWOULDBLOCK);
555 			}
556 			ROTATE_BUFFERS(d);
557 			break;
558 		}
559 
560 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
561 			/*
562 			 * A packet(s) either arrived since the previous
563 			 * read or arrived while we were asleep.
564 			 * Rotate the buffers and return what's here.
565 			 */
566 			ROTATE_BUFFERS(d);
567 			break;
568 		}
569 		error = tsleep(d, PRINET|PCATCH, "bpf",
570 				d->bd_rtout);
571 		if (error == EINTR || error == ERESTART) {
572 			splx(s);
573 			KERNEL_UNLOCK_ONE(NULL);
574 			return (error);
575 		}
576 		if (error == EWOULDBLOCK) {
577 			/*
578 			 * On a timeout, return what's in the buffer,
579 			 * which may be nothing.  If there is something
580 			 * in the store buffer, we can rotate the buffers.
581 			 */
582 			if (d->bd_hbuf)
583 				/*
584 				 * We filled up the buffer in between
585 				 * getting the timeout and arriving
586 				 * here, so we don't need to rotate.
587 				 */
588 				break;
589 
590 			if (d->bd_slen == 0) {
591 				splx(s);
592 				KERNEL_UNLOCK_ONE(NULL);
593 				return (0);
594 			}
595 			ROTATE_BUFFERS(d);
596 			break;
597 		}
598 		if (error != 0)
599 			goto done;
600 	}
601 	/*
602 	 * At this point, we know we have something in the hold slot.
603 	 */
604 	splx(s);
605 
606 	/*
607 	 * Move data from hold buffer into user space.
608 	 * We know the entire buffer is transferred since
609 	 * we checked above that the read buffer is bpf_bufsize bytes.
610 	 */
611 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
612 
613 	s = splnet();
614 	d->bd_fbuf = d->bd_hbuf;
615 	d->bd_hbuf = NULL;
616 	d->bd_hlen = 0;
617 done:
618 	splx(s);
619 	KERNEL_UNLOCK_ONE(NULL);
620 	return (error);
621 }
622 
623 
624 /*
625  * If there are processes sleeping on this descriptor, wake them up.
626  */
627 static inline void
628 bpf_wakeup(struct bpf_d *d)
629 {
630 	wakeup(d);
631 	if (d->bd_async)
632 		softint_schedule(d->bd_sih);
633 	selnotify(&d->bd_sel, 0, 0);
634 }
635 
636 static void
637 bpf_softintr(void *cookie)
638 {
639 	struct bpf_d *d;
640 
641 	d = cookie;
642 	if (d->bd_async)
643 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
644 }
645 
646 static void
647 bpf_timed_out(void *arg)
648 {
649 	struct bpf_d *d = arg;
650 	int s;
651 
652 	s = splnet();
653 	if (d->bd_state == BPF_WAITING) {
654 		d->bd_state = BPF_TIMED_OUT;
655 		if (d->bd_slen != 0)
656 			bpf_wakeup(d);
657 	}
658 	splx(s);
659 }
660 
661 
662 static int
663 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
664     kauth_cred_t cred, int flags)
665 {
666 	struct bpf_d *d = fp->f_bpf;
667 	struct ifnet *ifp;
668 	struct mbuf *m, *mc;
669 	int error, s;
670 	static struct sockaddr_storage dst;
671 
672 	m = NULL;	/* XXX gcc */
673 
674 	KERNEL_LOCK(1, NULL);
675 
676 	if (d->bd_bif == NULL) {
677 		KERNEL_UNLOCK_ONE(NULL);
678 		return (ENXIO);
679 	}
680 	getnanotime(&d->bd_mtime);
681 
682 	ifp = d->bd_bif->bif_ifp;
683 
684 	if (uio->uio_resid == 0) {
685 		KERNEL_UNLOCK_ONE(NULL);
686 		return (0);
687 	}
688 
689 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
690 		(struct sockaddr *) &dst);
691 	if (error) {
692 		KERNEL_UNLOCK_ONE(NULL);
693 		return (error);
694 	}
695 
696 	if (m->m_pkthdr.len > ifp->if_mtu) {
697 		KERNEL_UNLOCK_ONE(NULL);
698 		m_freem(m);
699 		return (EMSGSIZE);
700 	}
701 
702 	if (d->bd_hdrcmplt)
703 		dst.ss_family = pseudo_AF_HDRCMPLT;
704 
705 	if (d->bd_feedback) {
706 		mc = m_dup(m, 0, M_COPYALL, M_NOWAIT);
707 		if (mc != NULL)
708 			mc->m_pkthdr.rcvif = ifp;
709 		/* Set M_PROMISC for outgoing packets to be discarded. */
710 		if (1 /*d->bd_direction == BPF_D_INOUT*/)
711 			m->m_flags |= M_PROMISC;
712 	} else
713 		mc = NULL;
714 
715 	s = splsoftnet();
716 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *) &dst, NULL);
717 
718 	if (mc != NULL) {
719 		if (error == 0)
720 			(*ifp->if_input)(ifp, mc);
721 		m_freem(mc);
722 	}
723 	splx(s);
724 	KERNEL_UNLOCK_ONE(NULL);
725 	/*
726 	 * The driver frees the mbuf.
727 	 */
728 	return (error);
729 }
730 
731 /*
732  * Reset a descriptor by flushing its packet buffer and clearing the
733  * receive and drop counts.  Should be called at splnet.
734  */
735 static void
736 reset_d(struct bpf_d *d)
737 {
738 	if (d->bd_hbuf) {
739 		/* Free the hold buffer. */
740 		d->bd_fbuf = d->bd_hbuf;
741 		d->bd_hbuf = NULL;
742 	}
743 	d->bd_slen = 0;
744 	d->bd_hlen = 0;
745 	d->bd_rcount = 0;
746 	d->bd_dcount = 0;
747 	d->bd_ccount = 0;
748 }
749 
750 /*
751  *  FIONREAD		Check for read packet available.
752  *  BIOCGBLEN		Get buffer len [for read()].
753  *  BIOCSETF		Set ethernet read filter.
754  *  BIOCFLUSH		Flush read packet buffer.
755  *  BIOCPROMISC		Put interface into promiscuous mode.
756  *  BIOCGDLT		Get link layer type.
757  *  BIOCGETIF		Get interface name.
758  *  BIOCSETIF		Set interface.
759  *  BIOCSRTIMEOUT	Set read timeout.
760  *  BIOCGRTIMEOUT	Get read timeout.
761  *  BIOCGSTATS		Get packet stats.
762  *  BIOCIMMEDIATE	Set immediate mode.
763  *  BIOCVERSION		Get filter language version.
764  *  BIOCGHDRCMPLT	Get "header already complete" flag.
765  *  BIOCSHDRCMPLT	Set "header already complete" flag.
766  *  BIOCSFEEDBACK	Set packet feedback mode.
767  *  BIOCGFEEDBACK	Get packet feedback mode.
768  *  BIOCGSEESENT  	Get "see sent packets" mode.
769  *  BIOCSSEESENT  	Set "see sent packets" mode.
770  */
771 /* ARGSUSED */
772 static int
773 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
774 {
775 	struct bpf_d *d = fp->f_bpf;
776 	int s, error = 0;
777 
778 	/*
779 	 * Refresh the PID associated with this bpf file.
780 	 */
781 	KERNEL_LOCK(1, NULL);
782 	d->bd_pid = curproc->p_pid;
783 #ifdef _LP64
784 	if (curproc->p_flag & PK_32)
785 		d->bd_compat32 = 1;
786 	else
787 		d->bd_compat32 = 0;
788 #endif
789 
790 	s = splnet();
791 	if (d->bd_state == BPF_WAITING)
792 		callout_stop(&d->bd_callout);
793 	d->bd_state = BPF_IDLE;
794 	splx(s);
795 
796 	switch (cmd) {
797 
798 	default:
799 		error = EINVAL;
800 		break;
801 
802 	/*
803 	 * Check for read packet available.
804 	 */
805 	case FIONREAD:
806 		{
807 			int n;
808 
809 			s = splnet();
810 			n = d->bd_slen;
811 			if (d->bd_hbuf)
812 				n += d->bd_hlen;
813 			splx(s);
814 
815 			*(int *)addr = n;
816 			break;
817 		}
818 
819 	/*
820 	 * Get buffer len [for read()].
821 	 */
822 	case BIOCGBLEN:
823 		*(u_int *)addr = d->bd_bufsize;
824 		break;
825 
826 	/*
827 	 * Set buffer length.
828 	 */
829 	case BIOCSBLEN:
830 		if (d->bd_bif != NULL)
831 			error = EINVAL;
832 		else {
833 			u_int size = *(u_int *)addr;
834 
835 			if (size > bpf_maxbufsize)
836 				*(u_int *)addr = size = bpf_maxbufsize;
837 			else if (size < BPF_MINBUFSIZE)
838 				*(u_int *)addr = size = BPF_MINBUFSIZE;
839 			d->bd_bufsize = size;
840 		}
841 		break;
842 
843 	/*
844 	 * Set link layer read filter.
845 	 */
846 	case BIOCSETF:
847 		error = bpf_setf(d, addr);
848 		break;
849 
850 	/*
851 	 * Flush read packet buffer.
852 	 */
853 	case BIOCFLUSH:
854 		s = splnet();
855 		reset_d(d);
856 		splx(s);
857 		break;
858 
859 	/*
860 	 * Put interface into promiscuous mode.
861 	 */
862 	case BIOCPROMISC:
863 		if (d->bd_bif == NULL) {
864 			/*
865 			 * No interface attached yet.
866 			 */
867 			error = EINVAL;
868 			break;
869 		}
870 		s = splnet();
871 		if (d->bd_promisc == 0) {
872 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
873 			if (error == 0)
874 				d->bd_promisc = 1;
875 		}
876 		splx(s);
877 		break;
878 
879 	/*
880 	 * Get device parameters.
881 	 */
882 	case BIOCGDLT:
883 		if (d->bd_bif == NULL)
884 			error = EINVAL;
885 		else
886 			*(u_int *)addr = d->bd_bif->bif_dlt;
887 		break;
888 
889 	/*
890 	 * Get a list of supported device parameters.
891 	 */
892 	case BIOCGDLTLIST:
893 		if (d->bd_bif == NULL)
894 			error = EINVAL;
895 		else
896 			error = bpf_getdltlist(d, addr);
897 		break;
898 
899 	/*
900 	 * Set device parameters.
901 	 */
902 	case BIOCSDLT:
903 		if (d->bd_bif == NULL)
904 			error = EINVAL;
905 		else
906 			error = bpf_setdlt(d, *(u_int *)addr);
907 		break;
908 
909 	/*
910 	 * Set interface name.
911 	 */
912 #ifdef OBIOCGETIF
913 	case OBIOCGETIF:
914 #endif
915 	case BIOCGETIF:
916 		if (d->bd_bif == NULL)
917 			error = EINVAL;
918 		else
919 			bpf_ifname(d->bd_bif->bif_ifp, addr);
920 		break;
921 
922 	/*
923 	 * Set interface.
924 	 */
925 #ifdef OBIOCSETIF
926 	case OBIOCSETIF:
927 #endif
928 	case BIOCSETIF:
929 		error = bpf_setif(d, addr);
930 		break;
931 
932 	/*
933 	 * Set read timeout.
934 	 */
935 	case BIOCSRTIMEOUT:
936 		{
937 			struct timeval *tv = addr;
938 
939 			/* Compute number of ticks. */
940 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
941 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
942 				d->bd_rtout = 1;
943 			break;
944 		}
945 
946 #ifdef BIOCGORTIMEOUT
947 	/*
948 	 * Get read timeout.
949 	 */
950 	case BIOCGORTIMEOUT:
951 		{
952 			struct timeval50 *tv = addr;
953 
954 			tv->tv_sec = d->bd_rtout / hz;
955 			tv->tv_usec = (d->bd_rtout % hz) * tick;
956 			break;
957 		}
958 #endif
959 
960 #ifdef BIOCSORTIMEOUT
961 	/*
962 	 * Set read timeout.
963 	 */
964 	case BIOCSORTIMEOUT:
965 		{
966 			struct timeval50 *tv = addr;
967 
968 			/* Compute number of ticks. */
969 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
970 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
971 				d->bd_rtout = 1;
972 			break;
973 		}
974 #endif
975 
976 	/*
977 	 * Get read timeout.
978 	 */
979 	case BIOCGRTIMEOUT:
980 		{
981 			struct timeval *tv = addr;
982 
983 			tv->tv_sec = d->bd_rtout / hz;
984 			tv->tv_usec = (d->bd_rtout % hz) * tick;
985 			break;
986 		}
987 	/*
988 	 * Get packet stats.
989 	 */
990 	case BIOCGSTATS:
991 		{
992 			struct bpf_stat *bs = addr;
993 
994 			bs->bs_recv = d->bd_rcount;
995 			bs->bs_drop = d->bd_dcount;
996 			bs->bs_capt = d->bd_ccount;
997 			break;
998 		}
999 
1000 	case BIOCGSTATSOLD:
1001 		{
1002 			struct bpf_stat_old *bs = addr;
1003 
1004 			bs->bs_recv = d->bd_rcount;
1005 			bs->bs_drop = d->bd_dcount;
1006 			break;
1007 		}
1008 
1009 	/*
1010 	 * Set immediate mode.
1011 	 */
1012 	case BIOCIMMEDIATE:
1013 		d->bd_immediate = *(u_int *)addr;
1014 		break;
1015 
1016 	case BIOCVERSION:
1017 		{
1018 			struct bpf_version *bv = addr;
1019 
1020 			bv->bv_major = BPF_MAJOR_VERSION;
1021 			bv->bv_minor = BPF_MINOR_VERSION;
1022 			break;
1023 		}
1024 
1025 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
1026 		*(u_int *)addr = d->bd_hdrcmplt;
1027 		break;
1028 
1029 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
1030 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1031 		break;
1032 
1033 	/*
1034 	 * Get "see sent packets" flag
1035 	 */
1036 	case BIOCGSEESENT:
1037 		*(u_int *)addr = d->bd_seesent;
1038 		break;
1039 
1040 	/*
1041 	 * Set "see sent" packets flag
1042 	 */
1043 	case BIOCSSEESENT:
1044 		d->bd_seesent = *(u_int *)addr;
1045 		break;
1046 
1047 	/*
1048 	 * Set "feed packets from bpf back to input" mode
1049 	 */
1050 	case BIOCSFEEDBACK:
1051 		d->bd_feedback = *(u_int *)addr;
1052 		break;
1053 
1054 	/*
1055 	 * Get "feed packets from bpf back to input" mode
1056 	 */
1057 	case BIOCGFEEDBACK:
1058 		*(u_int *)addr = d->bd_feedback;
1059 		break;
1060 
1061 	case FIONBIO:		/* Non-blocking I/O */
1062 		/*
1063 		 * No need to do anything special as we use IO_NDELAY in
1064 		 * bpfread() as an indication of whether or not to block
1065 		 * the read.
1066 		 */
1067 		break;
1068 
1069 	case FIOASYNC:		/* Send signal on receive packets */
1070 		d->bd_async = *(int *)addr;
1071 		break;
1072 
1073 	case TIOCSPGRP:		/* Process or group to send signals to */
1074 	case FIOSETOWN:
1075 		error = fsetown(&d->bd_pgid, cmd, addr);
1076 		break;
1077 
1078 	case TIOCGPGRP:
1079 	case FIOGETOWN:
1080 		error = fgetown(d->bd_pgid, cmd, addr);
1081 		break;
1082 	}
1083 	KERNEL_UNLOCK_ONE(NULL);
1084 	return (error);
1085 }
1086 
1087 /*
1088  * Set d's packet filter program to fp.  If this file already has a filter,
1089  * free it and replace it.  Returns EINVAL for bogus requests.
1090  */
1091 int
1092 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
1093 {
1094 	struct bpf_insn *fcode, *old;
1095 	bpfjit_func_t jcode, oldj;
1096 	size_t flen, size;
1097 	int s;
1098 
1099 	jcode = NULL;
1100 	flen = fp->bf_len;
1101 
1102 	if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) {
1103 		return EINVAL;
1104 	}
1105 
1106 	if (flen) {
1107 		/*
1108 		 * Allocate the buffer, copy the byte-code from
1109 		 * userspace and validate it.
1110 		 */
1111 		size = flen * sizeof(*fp->bf_insns);
1112 		fcode = malloc(size, M_DEVBUF, M_WAITOK);
1113 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1114 		    !bpf_validate(fcode, (int)flen)) {
1115 			free(fcode, M_DEVBUF);
1116 			return EINVAL;
1117 		}
1118 		membar_consumer();
1119 		if (bpf_jit)
1120 			jcode = bpf_jit_generate(NULL, fcode, flen);
1121 	} else {
1122 		fcode = NULL;
1123 	}
1124 
1125 	s = splnet();
1126 	old = d->bd_filter;
1127 	d->bd_filter = fcode;
1128 	oldj = d->bd_jitcode;
1129 	d->bd_jitcode = jcode;
1130 	reset_d(d);
1131 	splx(s);
1132 
1133 	if (old) {
1134 		free(old, M_DEVBUF);
1135 	}
1136 	if (oldj) {
1137 		bpf_jit_freecode(oldj);
1138 	}
1139 
1140 	return 0;
1141 }
1142 
1143 /*
1144  * Detach a file from its current interface (if attached at all) and attach
1145  * to the interface indicated by the name stored in ifr.
1146  * Return an errno or 0.
1147  */
1148 static int
1149 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1150 {
1151 	struct bpf_if *bp;
1152 	char *cp;
1153 	int unit_seen, i, s, error;
1154 
1155 	/*
1156 	 * Make sure the provided name has a unit number, and default
1157 	 * it to '0' if not specified.
1158 	 * XXX This is ugly ... do this differently?
1159 	 */
1160 	unit_seen = 0;
1161 	cp = ifr->ifr_name;
1162 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1163 	while (*cp++)
1164 		if (*cp >= '0' && *cp <= '9')
1165 			unit_seen = 1;
1166 	if (!unit_seen) {
1167 		/* Make sure to leave room for the '\0'. */
1168 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1169 			if ((ifr->ifr_name[i] >= 'a' &&
1170 			     ifr->ifr_name[i] <= 'z') ||
1171 			    (ifr->ifr_name[i] >= 'A' &&
1172 			     ifr->ifr_name[i] <= 'Z'))
1173 				continue;
1174 			ifr->ifr_name[i] = '0';
1175 		}
1176 	}
1177 
1178 	/*
1179 	 * Look through attached interfaces for the named one.
1180 	 */
1181 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1182 		struct ifnet *ifp = bp->bif_ifp;
1183 
1184 		if (ifp == NULL ||
1185 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1186 			continue;
1187 		/* skip additional entry */
1188 		if (bp->bif_driverp != &ifp->if_bpf)
1189 			continue;
1190 		/*
1191 		 * We found the requested interface.
1192 		 * Allocate the packet buffers if we need to.
1193 		 * If we're already attached to requested interface,
1194 		 * just flush the buffer.
1195 		 */
1196 		if (d->bd_sbuf == NULL) {
1197 			error = bpf_allocbufs(d);
1198 			if (error != 0)
1199 				return (error);
1200 		}
1201 		s = splnet();
1202 		if (bp != d->bd_bif) {
1203 			if (d->bd_bif)
1204 				/*
1205 				 * Detach if attached to something else.
1206 				 */
1207 				bpf_detachd(d);
1208 
1209 			bpf_attachd(d, bp);
1210 		}
1211 		reset_d(d);
1212 		splx(s);
1213 		return (0);
1214 	}
1215 	/* Not found. */
1216 	return (ENXIO);
1217 }
1218 
1219 /*
1220  * Copy the interface name to the ifreq.
1221  */
1222 static void
1223 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1224 {
1225 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1226 }
1227 
1228 static int
1229 bpf_stat(struct file *fp, struct stat *st)
1230 {
1231 	struct bpf_d *d = fp->f_bpf;
1232 
1233 	(void)memset(st, 0, sizeof(*st));
1234 	KERNEL_LOCK(1, NULL);
1235 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1236 	st->st_atimespec = d->bd_atime;
1237 	st->st_mtimespec = d->bd_mtime;
1238 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1239 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1240 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1241 	st->st_mode = S_IFCHR;
1242 	KERNEL_UNLOCK_ONE(NULL);
1243 	return 0;
1244 }
1245 
1246 /*
1247  * Support for poll() system call
1248  *
1249  * Return true iff the specific operation will not block indefinitely - with
1250  * the assumption that it is safe to positively acknowledge a request for the
1251  * ability to write to the BPF device.
1252  * Otherwise, return false but make a note that a selnotify() must be done.
1253  */
1254 static int
1255 bpf_poll(struct file *fp, int events)
1256 {
1257 	struct bpf_d *d = fp->f_bpf;
1258 	int s = splnet();
1259 	int revents;
1260 
1261 	/*
1262 	 * Refresh the PID associated with this bpf file.
1263 	 */
1264 	KERNEL_LOCK(1, NULL);
1265 	d->bd_pid = curproc->p_pid;
1266 
1267 	revents = events & (POLLOUT | POLLWRNORM);
1268 	if (events & (POLLIN | POLLRDNORM)) {
1269 		/*
1270 		 * An imitation of the FIONREAD ioctl code.
1271 		 */
1272 		if (d->bd_hlen != 0 ||
1273 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1274 		     d->bd_slen != 0)) {
1275 			revents |= events & (POLLIN | POLLRDNORM);
1276 		} else {
1277 			selrecord(curlwp, &d->bd_sel);
1278 			/* Start the read timeout if necessary */
1279 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1280 				callout_reset(&d->bd_callout, d->bd_rtout,
1281 					      bpf_timed_out, d);
1282 				d->bd_state = BPF_WAITING;
1283 			}
1284 		}
1285 	}
1286 
1287 	KERNEL_UNLOCK_ONE(NULL);
1288 	splx(s);
1289 	return (revents);
1290 }
1291 
1292 static void
1293 filt_bpfrdetach(struct knote *kn)
1294 {
1295 	struct bpf_d *d = kn->kn_hook;
1296 	int s;
1297 
1298 	KERNEL_LOCK(1, NULL);
1299 	s = splnet();
1300 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1301 	splx(s);
1302 	KERNEL_UNLOCK_ONE(NULL);
1303 }
1304 
1305 static int
1306 filt_bpfread(struct knote *kn, long hint)
1307 {
1308 	struct bpf_d *d = kn->kn_hook;
1309 	int rv;
1310 
1311 	KERNEL_LOCK(1, NULL);
1312 	kn->kn_data = d->bd_hlen;
1313 	if (d->bd_immediate)
1314 		kn->kn_data += d->bd_slen;
1315 	rv = (kn->kn_data > 0);
1316 	KERNEL_UNLOCK_ONE(NULL);
1317 	return rv;
1318 }
1319 
1320 static const struct filterops bpfread_filtops =
1321 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1322 
1323 static int
1324 bpf_kqfilter(struct file *fp, struct knote *kn)
1325 {
1326 	struct bpf_d *d = fp->f_bpf;
1327 	struct klist *klist;
1328 	int s;
1329 
1330 	KERNEL_LOCK(1, NULL);
1331 
1332 	switch (kn->kn_filter) {
1333 	case EVFILT_READ:
1334 		klist = &d->bd_sel.sel_klist;
1335 		kn->kn_fop = &bpfread_filtops;
1336 		break;
1337 
1338 	default:
1339 		KERNEL_UNLOCK_ONE(NULL);
1340 		return (EINVAL);
1341 	}
1342 
1343 	kn->kn_hook = d;
1344 
1345 	s = splnet();
1346 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1347 	splx(s);
1348 	KERNEL_UNLOCK_ONE(NULL);
1349 
1350 	return (0);
1351 }
1352 
1353 /*
1354  * Copy data from an mbuf chain into a buffer.  This code is derived
1355  * from m_copydata in sys/uipc_mbuf.c.
1356  */
1357 static void *
1358 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1359 {
1360 	const struct mbuf *m;
1361 	u_int count;
1362 	u_char *dst;
1363 
1364 	m = src_arg;
1365 	dst = dst_arg;
1366 	while (len > 0) {
1367 		if (m == NULL)
1368 			panic("bpf_mcpy");
1369 		count = min(m->m_len, len);
1370 		memcpy(dst, mtod(m, const void *), count);
1371 		m = m->m_next;
1372 		dst += count;
1373 		len -= count;
1374 	}
1375 	return dst_arg;
1376 }
1377 
1378 /*
1379  * Dispatch a packet to all the listeners on interface bp.
1380  *
1381  * pkt     pointer to the packet, either a data buffer or an mbuf chain
1382  * buflen  buffer length, if pkt is a data buffer
1383  * cpfn    a function that can copy pkt into the listener's buffer
1384  * pktlen  length of the packet
1385  * rcv     true if packet came in
1386  */
1387 static inline void
1388 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1389     void *pkt, u_int pktlen, u_int buflen, const bool rcv)
1390 {
1391 	uint32_t mem[BPF_MEMWORDS];
1392 	bpf_args_t args = {
1393 		.pkt = (const uint8_t *)pkt,
1394 		.wirelen = pktlen,
1395 		.buflen = buflen,
1396 		.mem = mem,
1397 		.arg = NULL
1398 	};
1399 	bool gottime = false;
1400 	struct timespec ts;
1401 
1402 	/*
1403 	 * Note that the IPL does not have to be raised at this point.
1404 	 * The only problem that could arise here is that if two different
1405 	 * interfaces shared any data.  This is not the case.
1406 	 */
1407 	for (struct bpf_d *d = bp->bif_dlist; d != NULL; d = d->bd_next) {
1408 		u_int slen;
1409 
1410 		if (!d->bd_seesent && !rcv) {
1411 			continue;
1412 		}
1413 		d->bd_rcount++;
1414 		bpf_gstats.bs_recv++;
1415 
1416 		if (d->bd_jitcode)
1417 			slen = d->bd_jitcode(NULL, &args);
1418 		else
1419 			slen = bpf_filter_ext(NULL, d->bd_filter, &args);
1420 
1421 		if (!slen) {
1422 			continue;
1423 		}
1424 		if (!gottime) {
1425 			gottime = true;
1426 			nanotime(&ts);
1427 		}
1428 		catchpacket(d, pkt, pktlen, slen, cpfn, &ts);
1429 	}
1430 }
1431 
1432 /*
1433  * Incoming linkage from device drivers.  Process the packet pkt, of length
1434  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1435  * by each process' filter, and if accepted, stashed into the corresponding
1436  * buffer.
1437  */
1438 static void
1439 _bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1440 {
1441 
1442 	bpf_deliver(bp, memcpy, pkt, pktlen, pktlen, true);
1443 }
1444 
1445 /*
1446  * Incoming linkage from device drivers, when the head of the packet is in
1447  * a buffer, and the tail is in an mbuf chain.
1448  */
1449 static void
1450 _bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1451 {
1452 	u_int pktlen;
1453 	struct mbuf mb;
1454 
1455 	/* Skip outgoing duplicate packets. */
1456 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1457 		m->m_flags &= ~M_PROMISC;
1458 		return;
1459 	}
1460 
1461 	pktlen = m_length(m) + dlen;
1462 
1463 	/*
1464 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1465 	 * Note that we cut corners here; we only setup what's
1466 	 * absolutely needed--this mbuf should never go anywhere else.
1467 	 */
1468 	(void)memset(&mb, 0, sizeof(mb));
1469 	mb.m_next = m;
1470 	mb.m_data = data;
1471 	mb.m_len = dlen;
1472 
1473 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif != NULL);
1474 }
1475 
1476 /*
1477  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1478  */
1479 static void
1480 _bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1481 {
1482 	void *(*cpfn)(void *, const void *, size_t);
1483 	u_int pktlen, buflen;
1484 	void *marg;
1485 
1486 	/* Skip outgoing duplicate packets. */
1487 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1488 		m->m_flags &= ~M_PROMISC;
1489 		return;
1490 	}
1491 
1492 	pktlen = m_length(m);
1493 
1494 	if (pktlen == m->m_len) {
1495 		cpfn = (void *)memcpy;
1496 		marg = mtod(m, void *);
1497 		buflen = pktlen;
1498 	} else {
1499 		cpfn = bpf_mcpy;
1500 		marg = m;
1501 		buflen = 0;
1502 	}
1503 
1504 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif != NULL);
1505 }
1506 
1507 /*
1508  * We need to prepend the address family as
1509  * a four byte field.  Cons up a dummy header
1510  * to pacify bpf.  This is safe because bpf
1511  * will only read from the mbuf (i.e., it won't
1512  * try to free it or keep a pointer a to it).
1513  */
1514 static void
1515 _bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m)
1516 {
1517 	struct mbuf m0;
1518 
1519 	m0.m_flags = 0;
1520 	m0.m_next = m;
1521 	m0.m_len = 4;
1522 	m0.m_data = (char *)&af;
1523 
1524 	_bpf_mtap(bp, &m0);
1525 }
1526 
1527 /*
1528  * Put the SLIP pseudo-"link header" in place.
1529  * Note this M_PREPEND() should never fail,
1530  * swince we know we always have enough space
1531  * in the input buffer.
1532  */
1533 static void
1534 _bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m)
1535 {
1536 	int s;
1537 	u_char *hp;
1538 
1539 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1540 	if (*m == NULL)
1541 		return;
1542 
1543 	hp = mtod(*m, u_char *);
1544 	hp[SLX_DIR] = SLIPDIR_IN;
1545 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1546 
1547 	s = splnet();
1548 	_bpf_mtap(bp, *m);
1549 	splx(s);
1550 
1551 	m_adj(*m, SLIP_HDRLEN);
1552 }
1553 
1554 /*
1555  * Put the SLIP pseudo-"link header" in
1556  * place.  The compressed header is now
1557  * at the beginning of the mbuf.
1558  */
1559 static void
1560 _bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m)
1561 {
1562 	struct mbuf m0;
1563 	u_char *hp;
1564 	int s;
1565 
1566 	m0.m_flags = 0;
1567 	m0.m_next = m;
1568 	m0.m_data = m0.m_dat;
1569 	m0.m_len = SLIP_HDRLEN;
1570 
1571 	hp = mtod(&m0, u_char *);
1572 
1573 	hp[SLX_DIR] = SLIPDIR_OUT;
1574 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1575 
1576 	s = splnet();
1577 	_bpf_mtap(bp, &m0);
1578 	splx(s);
1579 	m_freem(m);
1580 }
1581 
1582 static int
1583 bpf_hdrlen(struct bpf_d *d)
1584 {
1585 	int hdrlen = d->bd_bif->bif_hdrlen;
1586 	/*
1587 	 * Compute the length of the bpf header.  This is not necessarily
1588 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1589 	 * that the network layer header begins on a longword boundary (for
1590 	 * performance reasons and to alleviate alignment restrictions).
1591 	 */
1592 #ifdef _LP64
1593 	if (d->bd_compat32)
1594 		return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen);
1595 	else
1596 #endif
1597 		return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen);
1598 }
1599 
1600 /*
1601  * Move the packet data from interface memory (pkt) into the
1602  * store buffer. Call the wakeup functions if it's time to wakeup
1603  * a listener (buffer full), "cpfn" is the routine called to do the
1604  * actual data transfer. memcpy is passed in to copy contiguous chunks,
1605  * while bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1606  * pkt is really an mbuf.
1607  */
1608 static void
1609 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1610     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1611 {
1612 	char *h;
1613 	int totlen, curlen, caplen;
1614 	int hdrlen = bpf_hdrlen(d);
1615 	int do_wakeup = 0;
1616 
1617 	++d->bd_ccount;
1618 	++bpf_gstats.bs_capt;
1619 	/*
1620 	 * Figure out how many bytes to move.  If the packet is
1621 	 * greater or equal to the snapshot length, transfer that
1622 	 * much.  Otherwise, transfer the whole packet (unless
1623 	 * we hit the buffer size limit).
1624 	 */
1625 	totlen = hdrlen + min(snaplen, pktlen);
1626 	if (totlen > d->bd_bufsize)
1627 		totlen = d->bd_bufsize;
1628 	/*
1629 	 * If we adjusted totlen to fit the bufsize, it could be that
1630 	 * totlen is smaller than hdrlen because of the link layer header.
1631 	 */
1632 	caplen = totlen - hdrlen;
1633 	if (caplen < 0)
1634 		caplen = 0;
1635 
1636 	/*
1637 	 * Round up the end of the previous packet to the next longword.
1638 	 */
1639 #ifdef _LP64
1640 	if (d->bd_compat32)
1641 		curlen = BPF_WORDALIGN32(d->bd_slen);
1642 	else
1643 #endif
1644 		curlen = BPF_WORDALIGN(d->bd_slen);
1645 	if (curlen + totlen > d->bd_bufsize) {
1646 		/*
1647 		 * This packet will overflow the storage buffer.
1648 		 * Rotate the buffers if we can, then wakeup any
1649 		 * pending reads.
1650 		 */
1651 		if (d->bd_fbuf == NULL) {
1652 			/*
1653 			 * We haven't completed the previous read yet,
1654 			 * so drop the packet.
1655 			 */
1656 			++d->bd_dcount;
1657 			++bpf_gstats.bs_drop;
1658 			return;
1659 		}
1660 		ROTATE_BUFFERS(d);
1661 		do_wakeup = 1;
1662 		curlen = 0;
1663 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1664 		/*
1665 		 * Immediate mode is set, or the read timeout has
1666 		 * already expired during a select call.  A packet
1667 		 * arrived, so the reader should be woken up.
1668 		 */
1669 		do_wakeup = 1;
1670 	}
1671 
1672 	/*
1673 	 * Append the bpf header.
1674 	 */
1675 	h = (char *)d->bd_sbuf + curlen;
1676 #ifdef _LP64
1677 	if (d->bd_compat32) {
1678 		struct bpf_hdr32 *hp32;
1679 
1680 		hp32 = (struct bpf_hdr32 *)h;
1681 		hp32->bh_tstamp.tv_sec = ts->tv_sec;
1682 		hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1683 		hp32->bh_datalen = pktlen;
1684 		hp32->bh_hdrlen = hdrlen;
1685 		hp32->bh_caplen = caplen;
1686 	} else
1687 #endif
1688 	{
1689 		struct bpf_hdr *hp;
1690 
1691 		hp = (struct bpf_hdr *)h;
1692 		hp->bh_tstamp.tv_sec = ts->tv_sec;
1693 		hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1694 		hp->bh_datalen = pktlen;
1695 		hp->bh_hdrlen = hdrlen;
1696 		hp->bh_caplen = caplen;
1697 	}
1698 
1699 	/*
1700 	 * Copy the packet data into the store buffer and update its length.
1701 	 */
1702 	(*cpfn)(h + hdrlen, pkt, caplen);
1703 	d->bd_slen = curlen + totlen;
1704 
1705 	/*
1706 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1707 	 * will cause filt_bpfread() to be called with it adjusted.
1708 	 */
1709 	if (do_wakeup)
1710 		bpf_wakeup(d);
1711 }
1712 
1713 /*
1714  * Initialize all nonzero fields of a descriptor.
1715  */
1716 static int
1717 bpf_allocbufs(struct bpf_d *d)
1718 {
1719 
1720 	d->bd_fbuf = malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK | M_CANFAIL);
1721 	if (!d->bd_fbuf)
1722 		return (ENOBUFS);
1723 	d->bd_sbuf = malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK | M_CANFAIL);
1724 	if (!d->bd_sbuf) {
1725 		free(d->bd_fbuf, M_DEVBUF);
1726 		return (ENOBUFS);
1727 	}
1728 	d->bd_slen = 0;
1729 	d->bd_hlen = 0;
1730 	return (0);
1731 }
1732 
1733 /*
1734  * Free buffers currently in use by a descriptor.
1735  * Called on close.
1736  */
1737 static void
1738 bpf_freed(struct bpf_d *d)
1739 {
1740 	/*
1741 	 * We don't need to lock out interrupts since this descriptor has
1742 	 * been detached from its interface and it yet hasn't been marked
1743 	 * free.
1744 	 */
1745 	if (d->bd_sbuf != NULL) {
1746 		free(d->bd_sbuf, M_DEVBUF);
1747 		if (d->bd_hbuf != NULL)
1748 			free(d->bd_hbuf, M_DEVBUF);
1749 		if (d->bd_fbuf != NULL)
1750 			free(d->bd_fbuf, M_DEVBUF);
1751 	}
1752 	if (d->bd_filter)
1753 		free(d->bd_filter, M_DEVBUF);
1754 
1755 	if (d->bd_jitcode != NULL) {
1756 		bpf_jit_freecode(d->bd_jitcode);
1757 	}
1758 }
1759 
1760 /*
1761  * Attach an interface to bpf.  dlt is the link layer type;
1762  * hdrlen is the fixed size of the link header for the specified dlt
1763  * (variable length headers not yet supported).
1764  */
1765 static void
1766 _bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1767 {
1768 	struct bpf_if *bp;
1769 	bp = malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1770 	if (bp == NULL)
1771 		panic("bpfattach");
1772 
1773 	bp->bif_dlist = NULL;
1774 	bp->bif_driverp = driverp;
1775 	bp->bif_ifp = ifp;
1776 	bp->bif_dlt = dlt;
1777 
1778 	bp->bif_next = bpf_iflist;
1779 	bpf_iflist = bp;
1780 
1781 	*bp->bif_driverp = NULL;
1782 
1783 	bp->bif_hdrlen = hdrlen;
1784 #if 0
1785 	printf("bpf: %s attached\n", ifp->if_xname);
1786 #endif
1787 }
1788 
1789 /*
1790  * Remove an interface from bpf.
1791  */
1792 static void
1793 _bpfdetach(struct ifnet *ifp)
1794 {
1795 	struct bpf_if *bp, **pbp;
1796 	struct bpf_d *d;
1797 	int s;
1798 
1799 	/* Nuke the vnodes for any open instances */
1800 	LIST_FOREACH(d, &bpf_list, bd_list) {
1801 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
1802 			/*
1803 			 * Detach the descriptor from an interface now.
1804 			 * It will be free'ed later by close routine.
1805 			 */
1806 			s = splnet();
1807 			d->bd_promisc = 0;	/* we can't touch device. */
1808 			bpf_detachd(d);
1809 			splx(s);
1810 		}
1811 	}
1812 
1813   again:
1814 	for (bp = bpf_iflist, pbp = &bpf_iflist;
1815 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
1816 		if (bp->bif_ifp == ifp) {
1817 			*pbp = bp->bif_next;
1818 			free(bp, M_DEVBUF);
1819 			goto again;
1820 		}
1821 	}
1822 }
1823 
1824 /*
1825  * Change the data link type of a interface.
1826  */
1827 static void
1828 _bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1829 {
1830 	struct bpf_if *bp;
1831 
1832 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1833 		if (bp->bif_driverp == &ifp->if_bpf)
1834 			break;
1835 	}
1836 	if (bp == NULL)
1837 		panic("bpf_change_type");
1838 
1839 	bp->bif_dlt = dlt;
1840 
1841 	bp->bif_hdrlen = hdrlen;
1842 }
1843 
1844 /*
1845  * Get a list of available data link type of the interface.
1846  */
1847 static int
1848 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1849 {
1850 	int n, error;
1851 	struct ifnet *ifp;
1852 	struct bpf_if *bp;
1853 
1854 	ifp = d->bd_bif->bif_ifp;
1855 	n = 0;
1856 	error = 0;
1857 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1858 		if (bp->bif_ifp != ifp)
1859 			continue;
1860 		if (bfl->bfl_list != NULL) {
1861 			if (n >= bfl->bfl_len)
1862 				return ENOMEM;
1863 			error = copyout(&bp->bif_dlt,
1864 			    bfl->bfl_list + n, sizeof(u_int));
1865 		}
1866 		n++;
1867 	}
1868 	bfl->bfl_len = n;
1869 	return error;
1870 }
1871 
1872 /*
1873  * Set the data link type of a BPF instance.
1874  */
1875 static int
1876 bpf_setdlt(struct bpf_d *d, u_int dlt)
1877 {
1878 	int s, error, opromisc;
1879 	struct ifnet *ifp;
1880 	struct bpf_if *bp;
1881 
1882 	if (d->bd_bif->bif_dlt == dlt)
1883 		return 0;
1884 	ifp = d->bd_bif->bif_ifp;
1885 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1886 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1887 			break;
1888 	}
1889 	if (bp == NULL)
1890 		return EINVAL;
1891 	s = splnet();
1892 	opromisc = d->bd_promisc;
1893 	bpf_detachd(d);
1894 	bpf_attachd(d, bp);
1895 	reset_d(d);
1896 	if (opromisc) {
1897 		error = ifpromisc(bp->bif_ifp, 1);
1898 		if (error)
1899 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
1900 			    bp->bif_ifp->if_xname, error);
1901 		else
1902 			d->bd_promisc = 1;
1903 	}
1904 	splx(s);
1905 	return 0;
1906 }
1907 
1908 static int
1909 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
1910 {
1911 	int newsize, error;
1912 	struct sysctlnode node;
1913 
1914 	node = *rnode;
1915 	node.sysctl_data = &newsize;
1916 	newsize = bpf_maxbufsize;
1917 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1918 	if (error || newp == NULL)
1919 		return (error);
1920 
1921 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
1922 		return (EINVAL);
1923 
1924 	bpf_maxbufsize = newsize;
1925 
1926 	return (0);
1927 }
1928 
1929 #if defined(MODULAR) || defined(BPFJIT)
1930 static int
1931 sysctl_net_bpf_jit(SYSCTLFN_ARGS)
1932 {
1933 	bool newval;
1934 	int error;
1935 	struct sysctlnode node;
1936 
1937 	node = *rnode;
1938 	node.sysctl_data = &newval;
1939 	newval = bpf_jit;
1940 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1941 	if (error != 0 || newp == NULL)
1942 		return error;
1943 
1944 	bpf_jit = newval;
1945 
1946 	/*
1947 	 * Do a full sync to publish new bpf_jit value and
1948 	 * update bpfjit_module_ops.bj_generate_code variable.
1949 	 */
1950 	membar_sync();
1951 
1952 	if (newval && bpfjit_module_ops.bj_generate_code == NULL) {
1953 		printf("JIT compilation is postponed "
1954 		    "until after bpfjit module is loaded\n");
1955 	}
1956 
1957 	return 0;
1958 }
1959 #endif
1960 
1961 static int
1962 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
1963 {
1964 	int    error, elem_count;
1965 	struct bpf_d	 *dp;
1966 	struct bpf_d_ext  dpe;
1967 	size_t len, needed, elem_size, out_size;
1968 	char   *sp;
1969 
1970 	if (namelen == 1 && name[0] == CTL_QUERY)
1971 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
1972 
1973 	if (namelen != 2)
1974 		return (EINVAL);
1975 
1976 	/* BPF peers is privileged information. */
1977 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
1978 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
1979 	if (error)
1980 		return (EPERM);
1981 
1982 	len = (oldp != NULL) ? *oldlenp : 0;
1983 	sp = oldp;
1984 	elem_size = name[0];
1985 	elem_count = name[1];
1986 	out_size = MIN(sizeof(dpe), elem_size);
1987 	needed = 0;
1988 
1989 	if (elem_size < 1 || elem_count < 0)
1990 		return (EINVAL);
1991 
1992 	mutex_enter(&bpf_mtx);
1993 	LIST_FOREACH(dp, &bpf_list, bd_list) {
1994 		if (len >= elem_size && elem_count > 0) {
1995 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
1996 			BPF_EXT(bufsize);
1997 			BPF_EXT(promisc);
1998 			BPF_EXT(state);
1999 			BPF_EXT(immediate);
2000 			BPF_EXT(hdrcmplt);
2001 			BPF_EXT(seesent);
2002 			BPF_EXT(pid);
2003 			BPF_EXT(rcount);
2004 			BPF_EXT(dcount);
2005 			BPF_EXT(ccount);
2006 #undef BPF_EXT
2007 			if (dp->bd_bif)
2008 				(void)strlcpy(dpe.bde_ifname,
2009 				    dp->bd_bif->bif_ifp->if_xname,
2010 				    IFNAMSIZ - 1);
2011 			else
2012 				dpe.bde_ifname[0] = '\0';
2013 
2014 			error = copyout(&dpe, sp, out_size);
2015 			if (error)
2016 				break;
2017 			sp += elem_size;
2018 			len -= elem_size;
2019 		}
2020 		needed += elem_size;
2021 		if (elem_count > 0 && elem_count != INT_MAX)
2022 			elem_count--;
2023 	}
2024 	mutex_exit(&bpf_mtx);
2025 
2026 	*oldlenp = needed;
2027 
2028 	return (error);
2029 }
2030 
2031 static struct sysctllog *bpf_sysctllog;
2032 static void
2033 sysctl_net_bpf_setup(void)
2034 {
2035 	const struct sysctlnode *node;
2036 
2037 	node = NULL;
2038 	sysctl_createv(&bpf_sysctllog, 0, NULL, &node,
2039 		       CTLFLAG_PERMANENT,
2040 		       CTLTYPE_NODE, "bpf",
2041 		       SYSCTL_DESCR("BPF options"),
2042 		       NULL, 0, NULL, 0,
2043 		       CTL_NET, CTL_CREATE, CTL_EOL);
2044 	if (node != NULL) {
2045 #if defined(MODULAR) || defined(BPFJIT)
2046 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2047 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2048 			CTLTYPE_BOOL, "jit",
2049 			SYSCTL_DESCR("Toggle Just-In-Time compilation"),
2050 			sysctl_net_bpf_jit, 0, &bpf_jit, 0,
2051 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2052 #endif
2053 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2054 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2055 			CTLTYPE_INT, "maxbufsize",
2056 			SYSCTL_DESCR("Maximum size for data capture buffer"),
2057 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
2058 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2059 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2060 			CTLFLAG_PERMANENT,
2061 			CTLTYPE_STRUCT, "stats",
2062 			SYSCTL_DESCR("BPF stats"),
2063 			NULL, 0, &bpf_gstats, sizeof(bpf_gstats),
2064 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2065 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2066 			CTLFLAG_PERMANENT,
2067 			CTLTYPE_STRUCT, "peers",
2068 			SYSCTL_DESCR("BPF peers"),
2069 			sysctl_net_bpf_peers, 0, NULL, 0,
2070 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2071 	}
2072 
2073 }
2074 
2075 struct bpf_ops bpf_ops_kernel = {
2076 	.bpf_attach =		_bpfattach,
2077 	.bpf_detach =		_bpfdetach,
2078 	.bpf_change_type =	_bpf_change_type,
2079 
2080 	.bpf_tap =		_bpf_tap,
2081 	.bpf_mtap =		_bpf_mtap,
2082 	.bpf_mtap2 =		_bpf_mtap2,
2083 	.bpf_mtap_af =		_bpf_mtap_af,
2084 	.bpf_mtap_sl_in =	_bpf_mtap_sl_in,
2085 	.bpf_mtap_sl_out =	_bpf_mtap_sl_out,
2086 };
2087 
2088 MODULE(MODULE_CLASS_DRIVER, bpf, NULL);
2089 
2090 static int
2091 bpf_modcmd(modcmd_t cmd, void *arg)
2092 {
2093 	devmajor_t bmajor, cmajor;
2094 	int error;
2095 
2096 	bmajor = cmajor = NODEVMAJOR;
2097 
2098 	switch (cmd) {
2099 	case MODULE_CMD_INIT:
2100 		bpfilterattach(0);
2101 		error = devsw_attach("bpf", NULL, &bmajor,
2102 		    &bpf_cdevsw, &cmajor);
2103 		if (error == EEXIST)
2104 			error = 0; /* maybe built-in ... improve eventually */
2105 		if (error)
2106 			break;
2107 
2108 		bpf_ops_handover_enter(&bpf_ops_kernel);
2109 		atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel);
2110 		bpf_ops_handover_exit();
2111 		sysctl_net_bpf_setup();
2112 		break;
2113 
2114 	case MODULE_CMD_FINI:
2115 		/*
2116 		 * While there is no reference counting for bpf callers,
2117 		 * unload could at least in theory be done similarly to
2118 		 * system call disestablishment.  This should even be
2119 		 * a little simpler:
2120 		 *
2121 		 * 1) replace op vector with stubs
2122 		 * 2) post update to all cpus with xc
2123 		 * 3) check that nobody is in bpf anymore
2124 		 *    (it's doubtful we'd want something like l_sysent,
2125 		 *     but we could do something like *signed* percpu
2126 		 *     counters.  if the sum is 0, we're good).
2127 		 * 4) if fail, unroll changes
2128 		 *
2129 		 * NOTE: change won't be atomic to the outside.  some
2130 		 * packets may be not captured even if unload is
2131 		 * not succesful.  I think packet capture not working
2132 		 * is a perfectly logical consequence of trying to
2133 		 * disable packet capture.
2134 		 */
2135 		error = EOPNOTSUPP;
2136 		/* insert sysctl teardown */
2137 		break;
2138 
2139 	default:
2140 		error = ENOTTY;
2141 		break;
2142 	}
2143 
2144 	return error;
2145 }
2146