xref: /netbsd-src/sys/net/bpf.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: bpf.c,v 1.150 2009/12/20 09:36:06 dsl 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 1.150 2009/12/20 09:36:06 dsl 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 
62 #include <sys/file.h>
63 #include <sys/filedesc.h>
64 #include <sys/tty.h>
65 #include <sys/uio.h>
66 
67 #include <sys/protosw.h>
68 #include <sys/socket.h>
69 #include <sys/errno.h>
70 #include <sys/kernel.h>
71 #include <sys/poll.h>
72 #include <sys/sysctl.h>
73 #include <sys/kauth.h>
74 
75 #include <net/if.h>
76 #include <net/slip.h>
77 
78 #include <net/bpf.h>
79 #include <net/bpfdesc.h>
80 
81 #include <net/if_arc.h>
82 #include <net/if_ether.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/if_inarp.h>
86 
87 
88 #include <compat/sys/sockio.h>
89 
90 #ifndef BPF_BUFSIZE
91 /*
92  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
93  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
94  */
95 # define BPF_BUFSIZE 32768
96 #endif
97 
98 #define PRINET  26			/* interruptible */
99 
100 /*
101  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
102  * XXX the default values should be computed dynamically based
103  * on available memory size and available mbuf clusters.
104  */
105 int bpf_bufsize = BPF_BUFSIZE;
106 int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
107 
108 
109 /*
110  * Global BPF statistics returned by net.bpf.stats sysctl.
111  */
112 struct bpf_stat	bpf_gstats;
113 
114 /*
115  * Use a mutex to avoid a race condition between gathering the stats/peers
116  * and opening/closing the device.
117  */
118 static kmutex_t bpf_mtx;
119 
120 /*
121  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
122  *  bpf_dtab holds the descriptors, indexed by minor device #
123  */
124 struct bpf_if	*bpf_iflist;
125 LIST_HEAD(, bpf_d) bpf_list;
126 
127 static int	bpf_allocbufs(struct bpf_d *);
128 static void	bpf_deliver(struct bpf_if *,
129 		            void *(*cpfn)(void *, const void *, size_t),
130 			    void *, u_int, u_int, struct ifnet *);
131 static void	bpf_freed(struct bpf_d *);
132 static void	bpf_ifname(struct ifnet *, struct ifreq *);
133 static void	*bpf_mcpy(void *, const void *, size_t);
134 static int	bpf_movein(struct uio *, int, int,
135 			        struct mbuf **, struct sockaddr *);
136 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
137 static void	bpf_detachd(struct bpf_d *);
138 static int	bpf_setif(struct bpf_d *, struct ifreq *);
139 static void	bpf_timed_out(void *);
140 static inline void
141 		bpf_wakeup(struct bpf_d *);
142 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
143     void *(*)(void *, const void *, size_t), struct timespec *);
144 static void	reset_d(struct bpf_d *);
145 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
146 static int	bpf_setdlt(struct bpf_d *, u_int);
147 
148 static int	bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t,
149     int);
150 static int	bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t,
151     int);
152 static int	bpf_ioctl(struct file *, u_long, void *);
153 static int	bpf_poll(struct file *, int);
154 static int	bpf_stat(struct file *, struct stat *);
155 static int	bpf_close(struct file *);
156 static int	bpf_kqfilter(struct file *, struct knote *);
157 static void	bpf_softintr(void *);
158 
159 static const struct fileops bpf_fileops = {
160 	.fo_read = bpf_read,
161 	.fo_write = bpf_write,
162 	.fo_ioctl = bpf_ioctl,
163 	.fo_fcntl = fnullop_fcntl,
164 	.fo_poll = bpf_poll,
165 	.fo_stat = bpf_stat,
166 	.fo_close = bpf_close,
167 	.fo_kqfilter = bpf_kqfilter,
168 	.fo_restart = fnullop_restart,
169 };
170 
171 dev_type_open(bpfopen);
172 
173 const struct cdevsw bpf_cdevsw = {
174 	bpfopen, noclose, noread, nowrite, noioctl,
175 	nostop, notty, nopoll, nommap, nokqfilter, D_OTHER
176 };
177 
178 static int
179 bpf_movein(struct uio *uio, int linktype, int mtu, struct mbuf **mp,
180 	   struct sockaddr *sockp)
181 {
182 	struct mbuf *m;
183 	int error;
184 	int len;
185 	int hlen;
186 	int align;
187 
188 	/*
189 	 * Build a sockaddr based on the data link layer type.
190 	 * We do this at this level because the ethernet header
191 	 * is copied directly into the data field of the sockaddr.
192 	 * In the case of SLIP, there is no header and the packet
193 	 * is forwarded as is.
194 	 * Also, we are careful to leave room at the front of the mbuf
195 	 * for the link level header.
196 	 */
197 	switch (linktype) {
198 
199 	case DLT_SLIP:
200 		sockp->sa_family = AF_INET;
201 		hlen = 0;
202 		align = 0;
203 		break;
204 
205 	case DLT_PPP:
206 		sockp->sa_family = AF_UNSPEC;
207 		hlen = 0;
208 		align = 0;
209 		break;
210 
211 	case DLT_EN10MB:
212 		sockp->sa_family = AF_UNSPEC;
213 		/* XXX Would MAXLINKHDR be better? */
214  		/* 6(dst)+6(src)+2(type) */
215 		hlen = sizeof(struct ether_header);
216 		align = 2;
217 		break;
218 
219 	case DLT_ARCNET:
220 		sockp->sa_family = AF_UNSPEC;
221 		hlen = ARC_HDRLEN;
222 		align = 5;
223 		break;
224 
225 	case DLT_FDDI:
226 		sockp->sa_family = AF_LINK;
227 		/* XXX 4(FORMAC)+6(dst)+6(src) */
228 		hlen = 16;
229 		align = 0;
230 		break;
231 
232 	case DLT_ECONET:
233 		sockp->sa_family = AF_UNSPEC;
234 		hlen = 6;
235 		align = 2;
236 		break;
237 
238 	case DLT_NULL:
239 		sockp->sa_family = AF_UNSPEC;
240 		hlen = 0;
241 		align = 0;
242 		break;
243 
244 	default:
245 		return (EIO);
246 	}
247 
248 	len = uio->uio_resid;
249 	/*
250 	 * If there aren't enough bytes for a link level header or the
251 	 * packet length exceeds the interface mtu, return an error.
252 	 */
253 	if (len < hlen || len - hlen > mtu)
254 		return (EMSGSIZE);
255 
256 	/*
257 	 * XXX Avoid complicated buffer chaining ---
258 	 * bail if it won't fit in a single mbuf.
259 	 * (Take into account possible alignment bytes)
260 	 */
261 	if ((unsigned)len > MCLBYTES - align)
262 		return (EIO);
263 
264 	m = m_gethdr(M_WAIT, MT_DATA);
265 	m->m_pkthdr.rcvif = 0;
266 	m->m_pkthdr.len = len - hlen;
267 	if (len > MHLEN - align) {
268 		m_clget(m, M_WAIT);
269 		if ((m->m_flags & M_EXT) == 0) {
270 			error = ENOBUFS;
271 			goto bad;
272 		}
273 	}
274 
275 	/* Insure the data is properly aligned */
276 	if (align > 0) {
277 		m->m_data += align;
278 		m->m_len -= align;
279 	}
280 
281 	error = uiomove(mtod(m, void *), len, uio);
282 	if (error)
283 		goto bad;
284 	if (hlen != 0) {
285 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
286 		m->m_data += hlen; /* XXX */
287 		len -= hlen;
288 	}
289 	m->m_len = len;
290 	*mp = m;
291 	return (0);
292 
293 bad:
294 	m_freem(m);
295 	return (error);
296 }
297 
298 /*
299  * Attach file to the bpf interface, i.e. make d listen on bp.
300  * Must be called at splnet.
301  */
302 static void
303 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
304 {
305 	/*
306 	 * Point d at bp, and add d to the interface's list of listeners.
307 	 * Finally, point the driver's bpf cookie at the interface so
308 	 * it will divert packets to bpf.
309 	 */
310 	d->bd_bif = bp;
311 	d->bd_next = bp->bif_dlist;
312 	bp->bif_dlist = d;
313 
314 	*bp->bif_driverp = bp;
315 }
316 
317 /*
318  * Detach a file from its interface.
319  */
320 static void
321 bpf_detachd(struct bpf_d *d)
322 {
323 	struct bpf_d **p;
324 	struct bpf_if *bp;
325 
326 	bp = d->bd_bif;
327 	/*
328 	 * Check if this descriptor had requested promiscuous mode.
329 	 * If so, turn it off.
330 	 */
331 	if (d->bd_promisc) {
332 		int error;
333 
334 		d->bd_promisc = 0;
335 		/*
336 		 * Take device out of promiscuous mode.  Since we were
337 		 * able to enter promiscuous mode, we should be able
338 		 * to turn it off.  But we can get an error if
339 		 * the interface was configured down, so only panic
340 		 * if we don't get an unexpected error.
341 		 */
342   		error = ifpromisc(bp->bif_ifp, 0);
343 		if (error && error != EINVAL)
344 			panic("%s: ifpromisc failed: %d", __func__, error);
345 	}
346 	/* Remove d from the interface's descriptor list. */
347 	p = &bp->bif_dlist;
348 	while (*p != d) {
349 		p = &(*p)->bd_next;
350 		if (*p == 0)
351 			panic("%s: descriptor not in list", __func__);
352 	}
353 	*p = (*p)->bd_next;
354 	if (bp->bif_dlist == 0)
355 		/*
356 		 * Let the driver know that there are no more listeners.
357 		 */
358 		*d->bd_bif->bif_driverp = 0;
359 	d->bd_bif = 0;
360 }
361 
362 
363 /*
364  * Mark a descriptor free by making it point to itself.
365  * This is probably cheaper than marking with a constant since
366  * the address should be in a register anyway.
367  */
368 
369 /*
370  * bpfilterattach() is called at boot time.
371  */
372 /* ARGSUSED */
373 void
374 bpfilterattach(int n)
375 {
376 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
377 
378 	mutex_enter(&bpf_mtx);
379 	LIST_INIT(&bpf_list);
380 	mutex_exit(&bpf_mtx);
381 
382 	bpf_gstats.bs_recv = 0;
383 	bpf_gstats.bs_drop = 0;
384 	bpf_gstats.bs_capt = 0;
385 }
386 
387 /*
388  * Open ethernet device. Clones.
389  */
390 /* ARGSUSED */
391 int
392 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
393 {
394 	struct bpf_d *d;
395 	struct file *fp;
396 	int error, fd;
397 
398 	/* falloc() will use the descriptor for us. */
399 	if ((error = fd_allocfile(&fp, &fd)) != 0)
400 		return error;
401 
402 	d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK|M_ZERO);
403 	d->bd_bufsize = bpf_bufsize;
404 	d->bd_seesent = 1;
405 	d->bd_pid = l->l_proc->p_pid;
406 	getnanotime(&d->bd_btime);
407 	d->bd_atime = d->bd_mtime = d->bd_btime;
408 	callout_init(&d->bd_callout, 0);
409 	selinit(&d->bd_sel);
410 	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
411 
412 	mutex_enter(&bpf_mtx);
413 	LIST_INSERT_HEAD(&bpf_list, d, bd_list);
414 	mutex_exit(&bpf_mtx);
415 
416 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
417 }
418 
419 /*
420  * Close the descriptor by detaching it from its interface,
421  * deallocating its buffers, and marking it free.
422  */
423 /* ARGSUSED */
424 static int
425 bpf_close(struct file *fp)
426 {
427 	struct bpf_d *d = fp->f_data;
428 	int s;
429 
430 	KERNEL_LOCK(1, NULL);
431 
432 	/*
433 	 * Refresh the PID associated with this bpf file.
434 	 */
435 	d->bd_pid = curproc->p_pid;
436 
437 	s = splnet();
438 	if (d->bd_state == BPF_WAITING)
439 		callout_stop(&d->bd_callout);
440 	d->bd_state = BPF_IDLE;
441 	if (d->bd_bif)
442 		bpf_detachd(d);
443 	splx(s);
444 	bpf_freed(d);
445 	mutex_enter(&bpf_mtx);
446 	LIST_REMOVE(d, bd_list);
447 	mutex_exit(&bpf_mtx);
448 	callout_destroy(&d->bd_callout);
449 	seldestroy(&d->bd_sel);
450 	softint_disestablish(d->bd_sih);
451 	free(d, M_DEVBUF);
452 	fp->f_data = NULL;
453 
454 	KERNEL_UNLOCK_ONE(NULL);
455 
456 	return (0);
457 }
458 
459 /*
460  * Rotate the packet buffers in descriptor d.  Move the store buffer
461  * into the hold slot, and the free buffer into the store slot.
462  * Zero the length of the new store buffer.
463  */
464 #define ROTATE_BUFFERS(d) \
465 	(d)->bd_hbuf = (d)->bd_sbuf; \
466 	(d)->bd_hlen = (d)->bd_slen; \
467 	(d)->bd_sbuf = (d)->bd_fbuf; \
468 	(d)->bd_slen = 0; \
469 	(d)->bd_fbuf = 0;
470 /*
471  *  bpfread - read next chunk of packets from buffers
472  */
473 static int
474 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
475     kauth_cred_t cred, int flags)
476 {
477 	struct bpf_d *d = fp->f_data;
478 	int timed_out;
479 	int error;
480 	int s;
481 
482 	getnanotime(&d->bd_atime);
483 	/*
484 	 * Restrict application to use a buffer the same size as
485 	 * the kernel buffers.
486 	 */
487 	if (uio->uio_resid != d->bd_bufsize)
488 		return (EINVAL);
489 
490 	KERNEL_LOCK(1, NULL);
491 	s = splnet();
492 	if (d->bd_state == BPF_WAITING)
493 		callout_stop(&d->bd_callout);
494 	timed_out = (d->bd_state == BPF_TIMED_OUT);
495 	d->bd_state = BPF_IDLE;
496 	/*
497 	 * If the hold buffer is empty, then do a timed sleep, which
498 	 * ends when the timeout expires or when enough packets
499 	 * have arrived to fill the store buffer.
500 	 */
501 	while (d->bd_hbuf == 0) {
502 		if (fp->f_flag & FNONBLOCK) {
503 			if (d->bd_slen == 0) {
504 				splx(s);
505 				KERNEL_UNLOCK_ONE(NULL);
506 				return (EWOULDBLOCK);
507 			}
508 			ROTATE_BUFFERS(d);
509 			break;
510 		}
511 
512 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
513 			/*
514 			 * A packet(s) either arrived since the previous
515 			 * read or arrived while we were asleep.
516 			 * Rotate the buffers and return what's here.
517 			 */
518 			ROTATE_BUFFERS(d);
519 			break;
520 		}
521 		error = tsleep(d, PRINET|PCATCH, "bpf",
522 				d->bd_rtout);
523 		if (error == EINTR || error == ERESTART) {
524 			splx(s);
525 			KERNEL_UNLOCK_ONE(NULL);
526 			return (error);
527 		}
528 		if (error == EWOULDBLOCK) {
529 			/*
530 			 * On a timeout, return what's in the buffer,
531 			 * which may be nothing.  If there is something
532 			 * in the store buffer, we can rotate the buffers.
533 			 */
534 			if (d->bd_hbuf)
535 				/*
536 				 * We filled up the buffer in between
537 				 * getting the timeout and arriving
538 				 * here, so we don't need to rotate.
539 				 */
540 				break;
541 
542 			if (d->bd_slen == 0) {
543 				splx(s);
544 				KERNEL_UNLOCK_ONE(NULL);
545 				return (0);
546 			}
547 			ROTATE_BUFFERS(d);
548 			break;
549 		}
550 		if (error != 0)
551 			goto done;
552 	}
553 	/*
554 	 * At this point, we know we have something in the hold slot.
555 	 */
556 	splx(s);
557 
558 	/*
559 	 * Move data from hold buffer into user space.
560 	 * We know the entire buffer is transferred since
561 	 * we checked above that the read buffer is bpf_bufsize bytes.
562 	 */
563 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
564 
565 	s = splnet();
566 	d->bd_fbuf = d->bd_hbuf;
567 	d->bd_hbuf = 0;
568 	d->bd_hlen = 0;
569 done:
570 	splx(s);
571 	KERNEL_UNLOCK_ONE(NULL);
572 	return (error);
573 }
574 
575 
576 /*
577  * If there are processes sleeping on this descriptor, wake them up.
578  */
579 static inline void
580 bpf_wakeup(struct bpf_d *d)
581 {
582 	wakeup(d);
583 	if (d->bd_async)
584 		softint_schedule(d->bd_sih);
585 	selnotify(&d->bd_sel, 0, 0);
586 }
587 
588 static void
589 bpf_softintr(void *cookie)
590 {
591 	struct bpf_d *d;
592 
593 	d = cookie;
594 	if (d->bd_async)
595 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
596 }
597 
598 static void
599 bpf_timed_out(void *arg)
600 {
601 	struct bpf_d *d = arg;
602 	int s;
603 
604 	s = splnet();
605 	if (d->bd_state == BPF_WAITING) {
606 		d->bd_state = BPF_TIMED_OUT;
607 		if (d->bd_slen != 0)
608 			bpf_wakeup(d);
609 	}
610 	splx(s);
611 }
612 
613 
614 static int
615 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
616     kauth_cred_t cred, int flags)
617 {
618 	struct bpf_d *d = fp->f_data;
619 	struct ifnet *ifp;
620 	struct mbuf *m;
621 	int error, s;
622 	static struct sockaddr_storage dst;
623 
624 	m = NULL;	/* XXX gcc */
625 
626 	KERNEL_LOCK(1, NULL);
627 
628 	if (d->bd_bif == 0) {
629 		KERNEL_UNLOCK_ONE(NULL);
630 		return (ENXIO);
631 	}
632 	getnanotime(&d->bd_mtime);
633 
634 	ifp = d->bd_bif->bif_ifp;
635 
636 	if (uio->uio_resid == 0) {
637 		KERNEL_UNLOCK_ONE(NULL);
638 		return (0);
639 	}
640 
641 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
642 		(struct sockaddr *) &dst);
643 	if (error) {
644 		KERNEL_UNLOCK_ONE(NULL);
645 		return (error);
646 	}
647 
648 	if (m->m_pkthdr.len > ifp->if_mtu) {
649 		KERNEL_UNLOCK_ONE(NULL);
650 		m_freem(m);
651 		return (EMSGSIZE);
652 	}
653 
654 	if (d->bd_hdrcmplt)
655 		dst.ss_family = pseudo_AF_HDRCMPLT;
656 
657 	s = splsoftnet();
658 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *) &dst, NULL);
659 	splx(s);
660 	KERNEL_UNLOCK_ONE(NULL);
661 	/*
662 	 * The driver frees the mbuf.
663 	 */
664 	return (error);
665 }
666 
667 /*
668  * Reset a descriptor by flushing its packet buffer and clearing the
669  * receive and drop counts.  Should be called at splnet.
670  */
671 static void
672 reset_d(struct bpf_d *d)
673 {
674 	if (d->bd_hbuf) {
675 		/* Free the hold buffer. */
676 		d->bd_fbuf = d->bd_hbuf;
677 		d->bd_hbuf = 0;
678 	}
679 	d->bd_slen = 0;
680 	d->bd_hlen = 0;
681 	d->bd_rcount = 0;
682 	d->bd_dcount = 0;
683 	d->bd_ccount = 0;
684 }
685 
686 /*
687  *  FIONREAD		Check for read packet available.
688  *  BIOCGBLEN		Get buffer len [for read()].
689  *  BIOCSETF		Set ethernet read filter.
690  *  BIOCFLUSH		Flush read packet buffer.
691  *  BIOCPROMISC		Put interface into promiscuous mode.
692  *  BIOCGDLT		Get link layer type.
693  *  BIOCGETIF		Get interface name.
694  *  BIOCSETIF		Set interface.
695  *  BIOCSRTIMEOUT	Set read timeout.
696  *  BIOCGRTIMEOUT	Get read timeout.
697  *  BIOCGSTATS		Get packet stats.
698  *  BIOCIMMEDIATE	Set immediate mode.
699  *  BIOCVERSION		Get filter language version.
700  *  BIOCGHDRCMPLT	Get "header already complete" flag.
701  *  BIOCSHDRCMPLT	Set "header already complete" flag.
702  */
703 /* ARGSUSED */
704 static int
705 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
706 {
707 	struct bpf_d *d = fp->f_data;
708 	int s, error = 0;
709 
710 	/*
711 	 * Refresh the PID associated with this bpf file.
712 	 */
713 	KERNEL_LOCK(1, NULL);
714 	d->bd_pid = curproc->p_pid;
715 
716 	s = splnet();
717 	if (d->bd_state == BPF_WAITING)
718 		callout_stop(&d->bd_callout);
719 	d->bd_state = BPF_IDLE;
720 	splx(s);
721 
722 	switch (cmd) {
723 
724 	default:
725 		error = EINVAL;
726 		break;
727 
728 	/*
729 	 * Check for read packet available.
730 	 */
731 	case FIONREAD:
732 		{
733 			int n;
734 
735 			s = splnet();
736 			n = d->bd_slen;
737 			if (d->bd_hbuf)
738 				n += d->bd_hlen;
739 			splx(s);
740 
741 			*(int *)addr = n;
742 			break;
743 		}
744 
745 	/*
746 	 * Get buffer len [for read()].
747 	 */
748 	case BIOCGBLEN:
749 		*(u_int *)addr = d->bd_bufsize;
750 		break;
751 
752 	/*
753 	 * Set buffer length.
754 	 */
755 	case BIOCSBLEN:
756 		if (d->bd_bif != 0)
757 			error = EINVAL;
758 		else {
759 			u_int size = *(u_int *)addr;
760 
761 			if (size > bpf_maxbufsize)
762 				*(u_int *)addr = size = bpf_maxbufsize;
763 			else if (size < BPF_MINBUFSIZE)
764 				*(u_int *)addr = size = BPF_MINBUFSIZE;
765 			d->bd_bufsize = size;
766 		}
767 		break;
768 
769 	/*
770 	 * Set link layer read filter.
771 	 */
772 	case BIOCSETF:
773 		error = bpf_setf(d, addr);
774 		break;
775 
776 	/*
777 	 * Flush read packet buffer.
778 	 */
779 	case BIOCFLUSH:
780 		s = splnet();
781 		reset_d(d);
782 		splx(s);
783 		break;
784 
785 	/*
786 	 * Put interface into promiscuous mode.
787 	 */
788 	case BIOCPROMISC:
789 		if (d->bd_bif == 0) {
790 			/*
791 			 * No interface attached yet.
792 			 */
793 			error = EINVAL;
794 			break;
795 		}
796 		s = splnet();
797 		if (d->bd_promisc == 0) {
798 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
799 			if (error == 0)
800 				d->bd_promisc = 1;
801 		}
802 		splx(s);
803 		break;
804 
805 	/*
806 	 * Get device parameters.
807 	 */
808 	case BIOCGDLT:
809 		if (d->bd_bif == 0)
810 			error = EINVAL;
811 		else
812 			*(u_int *)addr = d->bd_bif->bif_dlt;
813 		break;
814 
815 	/*
816 	 * Get a list of supported device parameters.
817 	 */
818 	case BIOCGDLTLIST:
819 		if (d->bd_bif == 0)
820 			error = EINVAL;
821 		else
822 			error = bpf_getdltlist(d, addr);
823 		break;
824 
825 	/*
826 	 * Set device parameters.
827 	 */
828 	case BIOCSDLT:
829 		if (d->bd_bif == 0)
830 			error = EINVAL;
831 		else
832 			error = bpf_setdlt(d, *(u_int *)addr);
833 		break;
834 
835 	/*
836 	 * Set interface name.
837 	 */
838 #ifdef OBIOCGETIF
839 	case OBIOCGETIF:
840 #endif
841 	case BIOCGETIF:
842 		if (d->bd_bif == 0)
843 			error = EINVAL;
844 		else
845 			bpf_ifname(d->bd_bif->bif_ifp, addr);
846 		break;
847 
848 	/*
849 	 * Set interface.
850 	 */
851 #ifdef OBIOCSETIF
852 	case OBIOCSETIF:
853 #endif
854 	case BIOCSETIF:
855 		error = bpf_setif(d, addr);
856 		break;
857 
858 	/*
859 	 * Set read timeout.
860 	 */
861 	case BIOCSRTIMEOUT:
862 		{
863 			struct timeval *tv = addr;
864 
865 			/* Compute number of ticks. */
866 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
867 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
868 				d->bd_rtout = 1;
869 			break;
870 		}
871 
872 #ifdef BIOCGORTIMEOUT
873 	/*
874 	 * Get read timeout.
875 	 */
876 	case BIOCGORTIMEOUT:
877 		{
878 			struct timeval50 *tv = addr;
879 
880 			tv->tv_sec = d->bd_rtout / hz;
881 			tv->tv_usec = (d->bd_rtout % hz) * tick;
882 			break;
883 		}
884 #endif
885 
886 #ifdef BIOCSORTIMEOUT
887 	/*
888 	 * Set read timeout.
889 	 */
890 	case BIOCSORTIMEOUT:
891 		{
892 			struct timeval50 *tv = addr;
893 
894 			/* Compute number of ticks. */
895 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
896 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
897 				d->bd_rtout = 1;
898 			break;
899 		}
900 #endif
901 
902 	/*
903 	 * Get read timeout.
904 	 */
905 	case BIOCGRTIMEOUT:
906 		{
907 			struct timeval *tv = addr;
908 
909 			tv->tv_sec = d->bd_rtout / hz;
910 			tv->tv_usec = (d->bd_rtout % hz) * tick;
911 			break;
912 		}
913 	/*
914 	 * Get packet stats.
915 	 */
916 	case BIOCGSTATS:
917 		{
918 			struct bpf_stat *bs = addr;
919 
920 			bs->bs_recv = d->bd_rcount;
921 			bs->bs_drop = d->bd_dcount;
922 			bs->bs_capt = d->bd_ccount;
923 			break;
924 		}
925 
926 	case BIOCGSTATSOLD:
927 		{
928 			struct bpf_stat_old *bs = addr;
929 
930 			bs->bs_recv = d->bd_rcount;
931 			bs->bs_drop = d->bd_dcount;
932 			break;
933 		}
934 
935 	/*
936 	 * Set immediate mode.
937 	 */
938 	case BIOCIMMEDIATE:
939 		d->bd_immediate = *(u_int *)addr;
940 		break;
941 
942 	case BIOCVERSION:
943 		{
944 			struct bpf_version *bv = addr;
945 
946 			bv->bv_major = BPF_MAJOR_VERSION;
947 			bv->bv_minor = BPF_MINOR_VERSION;
948 			break;
949 		}
950 
951 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
952 		*(u_int *)addr = d->bd_hdrcmplt;
953 		break;
954 
955 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
956 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
957 		break;
958 
959 	/*
960 	 * Get "see sent packets" flag
961 	 */
962 	case BIOCGSEESENT:
963 		*(u_int *)addr = d->bd_seesent;
964 		break;
965 
966 	/*
967 	 * Set "see sent" packets flag
968 	 */
969 	case BIOCSSEESENT:
970 		d->bd_seesent = *(u_int *)addr;
971 		break;
972 
973 	case FIONBIO:		/* Non-blocking I/O */
974 		/*
975 		 * No need to do anything special as we use IO_NDELAY in
976 		 * bpfread() as an indication of whether or not to block
977 		 * the read.
978 		 */
979 		break;
980 
981 	case FIOASYNC:		/* Send signal on receive packets */
982 		d->bd_async = *(int *)addr;
983 		break;
984 
985 	case TIOCSPGRP:		/* Process or group to send signals to */
986 	case FIOSETOWN:
987 		error = fsetown(&d->bd_pgid, cmd, addr);
988 		break;
989 
990 	case TIOCGPGRP:
991 	case FIOGETOWN:
992 		error = fgetown(d->bd_pgid, cmd, addr);
993 		break;
994 	}
995 	KERNEL_UNLOCK_ONE(NULL);
996 	return (error);
997 }
998 
999 /*
1000  * Set d's packet filter program to fp.  If this file already has a filter,
1001  * free it and replace it.  Returns EINVAL for bogus requests.
1002  */
1003 int
1004 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
1005 {
1006 	struct bpf_insn *fcode, *old;
1007 	u_int flen, size;
1008 	int s;
1009 
1010 	old = d->bd_filter;
1011 	if (fp->bf_insns == 0) {
1012 		if (fp->bf_len != 0)
1013 			return (EINVAL);
1014 		s = splnet();
1015 		d->bd_filter = 0;
1016 		reset_d(d);
1017 		splx(s);
1018 		if (old != 0)
1019 			free(old, M_DEVBUF);
1020 		return (0);
1021 	}
1022 	flen = fp->bf_len;
1023 	if (flen > BPF_MAXINSNS)
1024 		return (EINVAL);
1025 
1026 	size = flen * sizeof(*fp->bf_insns);
1027 	fcode = malloc(size, M_DEVBUF, M_WAITOK);
1028 	if (copyin(fp->bf_insns, fcode, size) == 0 &&
1029 	    bpf_validate(fcode, (int)flen)) {
1030 		s = splnet();
1031 		d->bd_filter = fcode;
1032 		reset_d(d);
1033 		splx(s);
1034 		if (old != 0)
1035 			free(old, M_DEVBUF);
1036 
1037 		return (0);
1038 	}
1039 	free(fcode, M_DEVBUF);
1040 	return (EINVAL);
1041 }
1042 
1043 /*
1044  * Detach a file from its current interface (if attached at all) and attach
1045  * to the interface indicated by the name stored in ifr.
1046  * Return an errno or 0.
1047  */
1048 static int
1049 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1050 {
1051 	struct bpf_if *bp;
1052 	char *cp;
1053 	int unit_seen, i, s, error;
1054 
1055 	/*
1056 	 * Make sure the provided name has a unit number, and default
1057 	 * it to '0' if not specified.
1058 	 * XXX This is ugly ... do this differently?
1059 	 */
1060 	unit_seen = 0;
1061 	cp = ifr->ifr_name;
1062 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1063 	while (*cp++)
1064 		if (*cp >= '0' && *cp <= '9')
1065 			unit_seen = 1;
1066 	if (!unit_seen) {
1067 		/* Make sure to leave room for the '\0'. */
1068 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1069 			if ((ifr->ifr_name[i] >= 'a' &&
1070 			     ifr->ifr_name[i] <= 'z') ||
1071 			    (ifr->ifr_name[i] >= 'A' &&
1072 			     ifr->ifr_name[i] <= 'Z'))
1073 				continue;
1074 			ifr->ifr_name[i] = '0';
1075 		}
1076 	}
1077 
1078 	/*
1079 	 * Look through attached interfaces for the named one.
1080 	 */
1081 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
1082 		struct ifnet *ifp = bp->bif_ifp;
1083 
1084 		if (ifp == 0 ||
1085 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1086 			continue;
1087 		/* skip additional entry */
1088 		if ((void **)bp->bif_driverp != &ifp->if_bpf)
1089 			continue;
1090 		/*
1091 		 * We found the requested interface.
1092 		 * Allocate the packet buffers if we need to.
1093 		 * If we're already attached to requested interface,
1094 		 * just flush the buffer.
1095 		 */
1096 		if (d->bd_sbuf == 0) {
1097 			error = bpf_allocbufs(d);
1098 			if (error != 0)
1099 				return (error);
1100 		}
1101 		s = splnet();
1102 		if (bp != d->bd_bif) {
1103 			if (d->bd_bif)
1104 				/*
1105 				 * Detach if attached to something else.
1106 				 */
1107 				bpf_detachd(d);
1108 
1109 			bpf_attachd(d, bp);
1110 		}
1111 		reset_d(d);
1112 		splx(s);
1113 		return (0);
1114 	}
1115 	/* Not found. */
1116 	return (ENXIO);
1117 }
1118 
1119 /*
1120  * Copy the interface name to the ifreq.
1121  */
1122 static void
1123 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1124 {
1125 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1126 }
1127 
1128 static int
1129 bpf_stat(struct file *fp, struct stat *st)
1130 {
1131 	struct bpf_d *d = fp->f_data;
1132 
1133 	(void)memset(st, 0, sizeof(*st));
1134 	KERNEL_LOCK(1, NULL);
1135 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1136 	st->st_atimespec = d->bd_atime;
1137 	st->st_mtimespec = d->bd_mtime;
1138 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1139 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1140 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1141 	KERNEL_UNLOCK_ONE(NULL);
1142 	return 0;
1143 }
1144 
1145 /*
1146  * Support for poll() system call
1147  *
1148  * Return true iff the specific operation will not block indefinitely - with
1149  * the assumption that it is safe to positively acknowledge a request for the
1150  * ability to write to the BPF device.
1151  * Otherwise, return false but make a note that a selnotify() must be done.
1152  */
1153 static int
1154 bpf_poll(struct file *fp, int events)
1155 {
1156 	struct bpf_d *d = fp->f_data;
1157 	int s = splnet();
1158 	int revents;
1159 
1160 	/*
1161 	 * Refresh the PID associated with this bpf file.
1162 	 */
1163 	KERNEL_LOCK(1, NULL);
1164 	d->bd_pid = curproc->p_pid;
1165 
1166 	revents = events & (POLLOUT | POLLWRNORM);
1167 	if (events & (POLLIN | POLLRDNORM)) {
1168 		/*
1169 		 * An imitation of the FIONREAD ioctl code.
1170 		 */
1171 		if (d->bd_hlen != 0 ||
1172 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1173 		     d->bd_slen != 0)) {
1174 			revents |= events & (POLLIN | POLLRDNORM);
1175 		} else {
1176 			selrecord(curlwp, &d->bd_sel);
1177 			/* Start the read timeout if necessary */
1178 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1179 				callout_reset(&d->bd_callout, d->bd_rtout,
1180 					      bpf_timed_out, d);
1181 				d->bd_state = BPF_WAITING;
1182 			}
1183 		}
1184 	}
1185 
1186 	KERNEL_UNLOCK_ONE(NULL);
1187 	splx(s);
1188 	return (revents);
1189 }
1190 
1191 static void
1192 filt_bpfrdetach(struct knote *kn)
1193 {
1194 	struct bpf_d *d = kn->kn_hook;
1195 	int s;
1196 
1197 	KERNEL_LOCK(1, NULL);
1198 	s = splnet();
1199 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1200 	splx(s);
1201 	KERNEL_UNLOCK_ONE(NULL);
1202 }
1203 
1204 static int
1205 filt_bpfread(struct knote *kn, long hint)
1206 {
1207 	struct bpf_d *d = kn->kn_hook;
1208 	int rv;
1209 
1210 	KERNEL_LOCK(1, NULL);
1211 	kn->kn_data = d->bd_hlen;
1212 	if (d->bd_immediate)
1213 		kn->kn_data += d->bd_slen;
1214 	rv = (kn->kn_data > 0);
1215 	KERNEL_UNLOCK_ONE(NULL);
1216 	return rv;
1217 }
1218 
1219 static const struct filterops bpfread_filtops =
1220 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1221 
1222 static int
1223 bpf_kqfilter(struct file *fp, struct knote *kn)
1224 {
1225 	struct bpf_d *d = fp->f_data;
1226 	struct klist *klist;
1227 	int s;
1228 
1229 	KERNEL_LOCK(1, NULL);
1230 
1231 	switch (kn->kn_filter) {
1232 	case EVFILT_READ:
1233 		klist = &d->bd_sel.sel_klist;
1234 		kn->kn_fop = &bpfread_filtops;
1235 		break;
1236 
1237 	default:
1238 		KERNEL_UNLOCK_ONE(NULL);
1239 		return (EINVAL);
1240 	}
1241 
1242 	kn->kn_hook = d;
1243 
1244 	s = splnet();
1245 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1246 	splx(s);
1247 	KERNEL_UNLOCK_ONE(NULL);
1248 
1249 	return (0);
1250 }
1251 
1252 /*
1253  * Incoming linkage from device drivers.  Process the packet pkt, of length
1254  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1255  * by each process' filter, and if accepted, stashed into the corresponding
1256  * buffer.
1257  */
1258 void
1259 bpf_tap(void *arg, u_char *pkt, u_int pktlen)
1260 {
1261 	struct bpf_if *bp;
1262 	struct bpf_d *d;
1263 	u_int slen;
1264 	struct timespec ts;
1265 	int gottime=0;
1266 
1267 	/*
1268 	 * Note that the ipl does not have to be raised at this point.
1269 	 * The only problem that could arise here is that if two different
1270 	 * interfaces shared any data.  This is not the case.
1271 	 */
1272 	bp = arg;
1273 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1274 		++d->bd_rcount;
1275 		++bpf_gstats.bs_recv;
1276 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
1277 		if (slen != 0) {
1278 			if (!gottime) {
1279 				nanotime(&ts);
1280 				gottime = 1;
1281 			}
1282 			catchpacket(d, pkt, pktlen, slen, memcpy, &ts);
1283 		}
1284 	}
1285 }
1286 
1287 /*
1288  * Copy data from an mbuf chain into a buffer.  This code is derived
1289  * from m_copydata in sys/uipc_mbuf.c.
1290  */
1291 static void *
1292 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1293 {
1294 	const struct mbuf *m;
1295 	u_int count;
1296 	u_char *dst;
1297 
1298 	m = src_arg;
1299 	dst = dst_arg;
1300 	while (len > 0) {
1301 		if (m == NULL)
1302 			panic("bpf_mcpy");
1303 		count = min(m->m_len, len);
1304 		memcpy(dst, mtod(m, const void *), count);
1305 		m = m->m_next;
1306 		dst += count;
1307 		len -= count;
1308 	}
1309 	return dst_arg;
1310 }
1311 
1312 /*
1313  * Dispatch a packet to all the listeners on interface bp.
1314  *
1315  * marg    pointer to the packet, either a data buffer or an mbuf chain
1316  * buflen  buffer length, if marg is a data buffer
1317  * cpfn    a function that can copy marg into the listener's buffer
1318  * pktlen  length of the packet
1319  * rcvif   either NULL or the interface the packet came in on.
1320  */
1321 static inline void
1322 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1323 	    void *marg, u_int pktlen, u_int buflen, struct ifnet *rcvif)
1324 {
1325 	u_int slen;
1326 	struct bpf_d *d;
1327 	struct timespec ts;
1328 	int gottime = 0;
1329 
1330 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1331 		if (!d->bd_seesent && (rcvif == NULL))
1332 			continue;
1333 		++d->bd_rcount;
1334 		++bpf_gstats.bs_recv;
1335 		slen = bpf_filter(d->bd_filter, marg, pktlen, buflen);
1336 		if (slen != 0) {
1337 			if(!gottime) {
1338 				nanotime(&ts);
1339 				gottime = 1;
1340 			}
1341 			catchpacket(d, marg, pktlen, slen, cpfn, &ts);
1342 		}
1343 	}
1344 }
1345 
1346 /*
1347  * Incoming linkage from device drivers, when the head of the packet is in
1348  * a buffer, and the tail is in an mbuf chain.
1349  */
1350 void
1351 bpf_mtap2(void *arg, void *data, u_int dlen, struct mbuf *m)
1352 {
1353 	struct bpf_if *bp = arg;
1354 	u_int pktlen;
1355 	struct mbuf mb;
1356 
1357 	pktlen = m_length(m) + dlen;
1358 
1359 	/*
1360 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1361 	 * Note that we cut corners here; we only setup what's
1362 	 * absolutely needed--this mbuf should never go anywhere else.
1363 	 */
1364 	(void)memset(&mb, 0, sizeof(mb));
1365 	mb.m_next = m;
1366 	mb.m_data = data;
1367 	mb.m_len = dlen;
1368 
1369 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif);
1370 }
1371 
1372 /*
1373  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1374  */
1375 void
1376 bpf_mtap(void *arg, struct mbuf *m)
1377 {
1378 	void *(*cpfn)(void *, const void *, size_t);
1379 	struct bpf_if *bp = arg;
1380 	u_int pktlen, buflen;
1381 	void *marg;
1382 
1383 	pktlen = m_length(m);
1384 
1385 	if (pktlen == m->m_len) {
1386 		cpfn = (void *)memcpy;
1387 		marg = mtod(m, void *);
1388 		buflen = pktlen;
1389 	} else {
1390 /*###1299 [cc] warning: assignment from incompatible pointer type%%%*/
1391 		cpfn = bpf_mcpy;
1392 		marg = m;
1393 		buflen = 0;
1394 	}
1395 
1396 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif);
1397 }
1398 
1399 /*
1400  * We need to prepend the address family as
1401  * a four byte field.  Cons up a dummy header
1402  * to pacify bpf.  This is safe because bpf
1403  * will only read from the mbuf (i.e., it won't
1404  * try to free it or keep a pointer a to it).
1405  */
1406 void
1407 bpf_mtap_af(void *arg, uint32_t af, struct mbuf *m)
1408 {
1409 	struct mbuf m0;
1410 
1411 	m0.m_flags = 0;
1412 	m0.m_next = m;
1413 	m0.m_len = 4;
1414 	m0.m_data = (char *)&af;
1415 
1416 	bpf_mtap(arg, &m0);
1417 }
1418 
1419 void
1420 bpf_mtap_et(void *arg, uint16_t et, struct mbuf *m)
1421 {
1422 	struct mbuf m0;
1423 
1424 	m0.m_flags = 0;
1425 	m0.m_next = m;
1426 	m0.m_len = 14;
1427 	m0.m_data = m0.m_dat;
1428 
1429 	((uint32_t *)m0.m_data)[0] = 0;
1430 	((uint32_t *)m0.m_data)[1] = 0;
1431 	((uint32_t *)m0.m_data)[2] = 0;
1432 	((uint16_t *)m0.m_data)[6] = et;
1433 
1434 	bpf_mtap(arg, &m0);
1435 }
1436 
1437 #if NSL > 0 || NSTRIP > 0
1438 /*
1439  * Put the SLIP pseudo-"link header" in place.
1440  * Note this M_PREPEND() should never fail,
1441  * swince we know we always have enough space
1442  * in the input buffer.
1443  */
1444 void
1445 bpf_mtap_sl_in(void *arg, u_char *chdr, struct mbuf **m)
1446 {
1447 	int s;
1448 	u_char *hp;
1449 
1450 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1451 	if (*m == NULL)
1452 		return;
1453 
1454 	hp = mtod(*m, u_char *);
1455 	hp[SLX_DIR] = SLIPDIR_IN;
1456 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1457 
1458 	s = splnet();
1459 	bpf_mtap(arg, *m);
1460 	splx(s);
1461 
1462 	m_adj(*m, SLIP_HDRLEN);
1463 }
1464 
1465 /*
1466  * Put the SLIP pseudo-"link header" in
1467  * place.  The compressed header is now
1468  * at the beginning of the mbuf.
1469  */
1470 void
1471 bpf_mtap_sl_out(void *arg, u_char *chdr, struct mbuf *m)
1472 {
1473 	struct mbuf m0;
1474 	u_char *hp;
1475 	int s;
1476 
1477 	m0.m_flags = 0;
1478 	m0.m_next = m;
1479 	m0.m_data = m0.m_dat;
1480 	m0.m_len = SLIP_HDRLEN;
1481 
1482 	hp = mtod(&m0, u_char *);
1483 
1484 	hp[SLX_DIR] = SLIPDIR_OUT;
1485 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1486 
1487 	s = splnet();
1488 	bpf_mtap(arg, &m0);
1489 	splx(s);
1490 	m_freem(m);
1491 }
1492 #endif
1493 
1494 /*
1495  * Move the packet data from interface memory (pkt) into the
1496  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1497  * otherwise 0.  "copy" is the routine called to do the actual data
1498  * transfer.  memcpy is passed in to copy contiguous chunks, while
1499  * bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1500  * pkt is really an mbuf.
1501  */
1502 static void
1503 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1504     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1505 {
1506 	struct bpf_hdr *hp;
1507 	int totlen, curlen;
1508 	int hdrlen = d->bd_bif->bif_hdrlen;
1509 	int do_wakeup = 0;
1510 
1511 	++d->bd_ccount;
1512 	++bpf_gstats.bs_capt;
1513 	/*
1514 	 * Figure out how many bytes to move.  If the packet is
1515 	 * greater or equal to the snapshot length, transfer that
1516 	 * much.  Otherwise, transfer the whole packet (unless
1517 	 * we hit the buffer size limit).
1518 	 */
1519 	totlen = hdrlen + min(snaplen, pktlen);
1520 	if (totlen > d->bd_bufsize)
1521 		totlen = d->bd_bufsize;
1522 
1523 	/*
1524 	 * Round up the end of the previous packet to the next longword.
1525 	 */
1526 	curlen = BPF_WORDALIGN(d->bd_slen);
1527 	if (curlen + totlen > d->bd_bufsize) {
1528 		/*
1529 		 * This packet will overflow the storage buffer.
1530 		 * Rotate the buffers if we can, then wakeup any
1531 		 * pending reads.
1532 		 */
1533 		if (d->bd_fbuf == 0) {
1534 			/*
1535 			 * We haven't completed the previous read yet,
1536 			 * so drop the packet.
1537 			 */
1538 			++d->bd_dcount;
1539 			++bpf_gstats.bs_drop;
1540 			return;
1541 		}
1542 		ROTATE_BUFFERS(d);
1543 		do_wakeup = 1;
1544 		curlen = 0;
1545 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1546 		/*
1547 		 * Immediate mode is set, or the read timeout has
1548 		 * already expired during a select call.  A packet
1549 		 * arrived, so the reader should be woken up.
1550 		 */
1551 		do_wakeup = 1;
1552 	}
1553 
1554 	/*
1555 	 * Append the bpf header.
1556 	 */
1557 	hp = (struct bpf_hdr *)((char *)d->bd_sbuf + curlen);
1558 	hp->bh_tstamp.tv_sec = ts->tv_sec;
1559 	hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1560 	hp->bh_datalen = pktlen;
1561 	hp->bh_hdrlen = hdrlen;
1562 	/*
1563 	 * Copy the packet data into the store buffer and update its length.
1564 	 */
1565 	(*cpfn)((u_char *)hp + hdrlen, pkt, (hp->bh_caplen = totlen - hdrlen));
1566 	d->bd_slen = curlen + totlen;
1567 
1568 	/*
1569 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1570 	 * will cause filt_bpfread() to be called with it adjusted.
1571 	 */
1572 	if (do_wakeup)
1573 		bpf_wakeup(d);
1574 }
1575 
1576 /*
1577  * Initialize all nonzero fields of a descriptor.
1578  */
1579 static int
1580 bpf_allocbufs(struct bpf_d *d)
1581 {
1582 
1583 	d->bd_fbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1584 	if (!d->bd_fbuf)
1585 		return (ENOBUFS);
1586 	d->bd_sbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1587 	if (!d->bd_sbuf) {
1588 		free(d->bd_fbuf, M_DEVBUF);
1589 		return (ENOBUFS);
1590 	}
1591 	d->bd_slen = 0;
1592 	d->bd_hlen = 0;
1593 	return (0);
1594 }
1595 
1596 /*
1597  * Free buffers currently in use by a descriptor.
1598  * Called on close.
1599  */
1600 static void
1601 bpf_freed(struct bpf_d *d)
1602 {
1603 	/*
1604 	 * We don't need to lock out interrupts since this descriptor has
1605 	 * been detached from its interface and it yet hasn't been marked
1606 	 * free.
1607 	 */
1608 	if (d->bd_sbuf != 0) {
1609 		free(d->bd_sbuf, M_DEVBUF);
1610 		if (d->bd_hbuf != 0)
1611 			free(d->bd_hbuf, M_DEVBUF);
1612 		if (d->bd_fbuf != 0)
1613 			free(d->bd_fbuf, M_DEVBUF);
1614 	}
1615 	if (d->bd_filter)
1616 		free(d->bd_filter, M_DEVBUF);
1617 }
1618 
1619 /*
1620  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
1621  * fixed size of the link header (variable length headers not yet supported).
1622  */
1623 void
1624 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1625 {
1626 
1627 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
1628 }
1629 
1630 /*
1631  * Attach additional dlt for a interface to bpf.  dlt is the link layer type;
1632  * hdrlen is the fixed size of the link header for the specified dlt
1633  * (variable length headers not yet supported).
1634  */
1635 void
1636 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, void *driverp)
1637 {
1638 	struct bpf_if *bp;
1639 	bp = malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1640 	if (bp == 0)
1641 		panic("bpfattach");
1642 
1643 	bp->bif_dlist = 0;
1644 	bp->bif_driverp = driverp;
1645 	bp->bif_ifp = ifp;
1646 	bp->bif_dlt = dlt;
1647 
1648 	bp->bif_next = bpf_iflist;
1649 	bpf_iflist = bp;
1650 
1651 	*bp->bif_driverp = 0;
1652 
1653 	/*
1654 	 * Compute the length of the bpf header.  This is not necessarily
1655 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1656 	 * that the network layer header begins on a longword boundary (for
1657 	 * performance reasons and to alleviate alignment restrictions).
1658 	 */
1659 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1660 
1661 #if 0
1662 	printf("bpf: %s attached\n", ifp->if_xname);
1663 #endif
1664 }
1665 
1666 /*
1667  * Remove an interface from bpf.
1668  */
1669 void
1670 bpfdetach(struct ifnet *ifp)
1671 {
1672 	struct bpf_if *bp, **pbp;
1673 	struct bpf_d *d;
1674 	int s;
1675 
1676 	/* Nuke the vnodes for any open instances */
1677 	LIST_FOREACH(d, &bpf_list, bd_list) {
1678 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
1679 			/*
1680 			 * Detach the descriptor from an interface now.
1681 			 * It will be free'ed later by close routine.
1682 			 */
1683 			s = splnet();
1684 			d->bd_promisc = 0;	/* we can't touch device. */
1685 			bpf_detachd(d);
1686 			splx(s);
1687 		}
1688 	}
1689 
1690   again:
1691 	for (bp = bpf_iflist, pbp = &bpf_iflist;
1692 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
1693 		if (bp->bif_ifp == ifp) {
1694 			*pbp = bp->bif_next;
1695 			free(bp, M_DEVBUF);
1696 			goto again;
1697 		}
1698 	}
1699 }
1700 
1701 /*
1702  * Change the data link type of a interface.
1703  */
1704 void
1705 bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1706 {
1707 	struct bpf_if *bp;
1708 
1709 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1710 		if ((void **)bp->bif_driverp == &ifp->if_bpf)
1711 			break;
1712 	}
1713 	if (bp == NULL)
1714 		panic("bpf_change_type");
1715 
1716 	bp->bif_dlt = dlt;
1717 
1718 	/*
1719 	 * Compute the length of the bpf header.  This is not necessarily
1720 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1721 	 * that the network layer header begins on a longword boundary (for
1722 	 * performance reasons and to alleviate alignment restrictions).
1723 	 */
1724 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1725 }
1726 
1727 /*
1728  * Get a list of available data link type of the interface.
1729  */
1730 static int
1731 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1732 {
1733 	int n, error;
1734 	struct ifnet *ifp;
1735 	struct bpf_if *bp;
1736 
1737 	ifp = d->bd_bif->bif_ifp;
1738 	n = 0;
1739 	error = 0;
1740 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1741 		if (bp->bif_ifp != ifp)
1742 			continue;
1743 		if (bfl->bfl_list != NULL) {
1744 			if (n >= bfl->bfl_len)
1745 				return ENOMEM;
1746 			error = copyout(&bp->bif_dlt,
1747 			    bfl->bfl_list + n, sizeof(u_int));
1748 		}
1749 		n++;
1750 	}
1751 	bfl->bfl_len = n;
1752 	return error;
1753 }
1754 
1755 /*
1756  * Set the data link type of a BPF instance.
1757  */
1758 static int
1759 bpf_setdlt(struct bpf_d *d, u_int dlt)
1760 {
1761 	int s, error, opromisc;
1762 	struct ifnet *ifp;
1763 	struct bpf_if *bp;
1764 
1765 	if (d->bd_bif->bif_dlt == dlt)
1766 		return 0;
1767 	ifp = d->bd_bif->bif_ifp;
1768 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1769 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1770 			break;
1771 	}
1772 	if (bp == NULL)
1773 		return EINVAL;
1774 	s = splnet();
1775 	opromisc = d->bd_promisc;
1776 	bpf_detachd(d);
1777 	bpf_attachd(d, bp);
1778 	reset_d(d);
1779 	if (opromisc) {
1780 		error = ifpromisc(bp->bif_ifp, 1);
1781 		if (error)
1782 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
1783 			    bp->bif_ifp->if_xname, error);
1784 		else
1785 			d->bd_promisc = 1;
1786 	}
1787 	splx(s);
1788 	return 0;
1789 }
1790 
1791 static int
1792 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
1793 {
1794 	int newsize, error;
1795 	struct sysctlnode node;
1796 
1797 	node = *rnode;
1798 	node.sysctl_data = &newsize;
1799 	newsize = bpf_maxbufsize;
1800 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1801 	if (error || newp == NULL)
1802 		return (error);
1803 
1804 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
1805 		return (EINVAL);
1806 
1807 	bpf_maxbufsize = newsize;
1808 
1809 	return (0);
1810 }
1811 
1812 static int
1813 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
1814 {
1815 	int    error, elem_count;
1816 	struct bpf_d	 *dp;
1817 	struct bpf_d_ext  dpe;
1818 	size_t len, needed, elem_size, out_size;
1819 	char   *sp;
1820 
1821 	if (namelen == 1 && name[0] == CTL_QUERY)
1822 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
1823 
1824 	if (namelen != 2)
1825 		return (EINVAL);
1826 
1827 	/* BPF peers is privileged information. */
1828 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
1829 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
1830 	if (error)
1831 		return (EPERM);
1832 
1833 	len = (oldp != NULL) ? *oldlenp : 0;
1834 	sp = oldp;
1835 	elem_size = name[0];
1836 	elem_count = name[1];
1837 	out_size = MIN(sizeof(dpe), elem_size);
1838 	needed = 0;
1839 
1840 	if (elem_size < 1 || elem_count < 0)
1841 		return (EINVAL);
1842 
1843 	mutex_enter(&bpf_mtx);
1844 	LIST_FOREACH(dp, &bpf_list, bd_list) {
1845 		if (len >= elem_size && elem_count > 0) {
1846 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
1847 			BPF_EXT(bufsize);
1848 			BPF_EXT(promisc);
1849 			BPF_EXT(promisc);
1850 			BPF_EXT(state);
1851 			BPF_EXT(immediate);
1852 			BPF_EXT(hdrcmplt);
1853 			BPF_EXT(seesent);
1854 			BPF_EXT(pid);
1855 			BPF_EXT(rcount);
1856 			BPF_EXT(dcount);
1857 			BPF_EXT(ccount);
1858 #undef BPF_EXT
1859 			if (dp->bd_bif)
1860 				(void)strlcpy(dpe.bde_ifname,
1861 				    dp->bd_bif->bif_ifp->if_xname,
1862 				    IFNAMSIZ - 1);
1863 			else
1864 				dpe.bde_ifname[0] = '\0';
1865 
1866 			error = copyout(&dpe, sp, out_size);
1867 			if (error)
1868 				break;
1869 			sp += elem_size;
1870 			len -= elem_size;
1871 		}
1872 		needed += elem_size;
1873 		if (elem_count > 0 && elem_count != INT_MAX)
1874 			elem_count--;
1875 	}
1876 	mutex_exit(&bpf_mtx);
1877 
1878 	*oldlenp = needed;
1879 
1880 	return (error);
1881 }
1882 
1883 SYSCTL_SETUP(sysctl_net_bpf_setup, "sysctl net.bpf subtree setup")
1884 {
1885 	const struct sysctlnode *node;
1886 
1887 	sysctl_createv(clog, 0, NULL, NULL,
1888 		       CTLFLAG_PERMANENT,
1889 		       CTLTYPE_NODE, "net", NULL,
1890 		       NULL, 0, NULL, 0,
1891 		       CTL_NET, CTL_EOL);
1892 
1893 	node = NULL;
1894 	sysctl_createv(clog, 0, NULL, &node,
1895 		       CTLFLAG_PERMANENT,
1896 		       CTLTYPE_NODE, "bpf",
1897 		       SYSCTL_DESCR("BPF options"),
1898 		       NULL, 0, NULL, 0,
1899 		       CTL_NET, CTL_CREATE, CTL_EOL);
1900 	if (node != NULL) {
1901 		sysctl_createv(clog, 0, NULL, NULL,
1902 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1903 			CTLTYPE_INT, "maxbufsize",
1904 			SYSCTL_DESCR("Maximum size for data capture buffer"),
1905 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
1906 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1907 		sysctl_createv(clog, 0, NULL, NULL,
1908 			CTLFLAG_PERMANENT,
1909 			CTLTYPE_STRUCT, "stats",
1910 			SYSCTL_DESCR("BPF stats"),
1911 			NULL, 0, &bpf_gstats, sizeof(bpf_gstats),
1912 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1913 		sysctl_createv(clog, 0, NULL, NULL,
1914 			CTLFLAG_PERMANENT,
1915 			CTLTYPE_STRUCT, "peers",
1916 			SYSCTL_DESCR("BPF peers"),
1917 			sysctl_net_bpf_peers, 0, NULL, 0,
1918 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1919 	}
1920 
1921 }
1922