xref: /netbsd-src/sys/net/bpf.c (revision d9158b13b5dfe46201430699a3f7a235ecf28df3)
1 /*
2  * Copyright (c) 1990, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from the Stanford/CMU enet packet filter,
6  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
7  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
8  * Berkeley Laboratory.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	from: @(#)bpf.c	8.2 (Berkeley) 3/28/94
39  *	$Id: bpf.c,v 1.12 1994/05/13 06:02:14 mycroft Exp $
40  */
41 
42 #include "bpfilter.h"
43 
44 #ifndef __GNUC__
45 #define inline
46 #else
47 #define inline __inline
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/user.h>
57 #include <sys/ioctl.h>
58 #include <sys/map.h>
59 
60 #include <sys/file.h>
61 #if defined(sparc) && BSD < 199103
62 #include <sys/stream.h>
63 #endif
64 #include <sys/tty.h>
65 #include <sys/uio.h>
66 
67 #include <sys/protosw.h>
68 #include <sys/socket.h>
69 #include <net/if.h>
70 
71 #include <net/bpf.h>
72 #include <net/bpfdesc.h>
73 
74 #include <sys/errno.h>
75 
76 #include <netinet/in.h>
77 #include <netinet/if_ether.h>
78 #include <sys/kernel.h>
79 
80 /*
81  * Older BSDs don't have kernel malloc.
82  */
83 #if BSD < 199103
84 extern bcopy();
85 static caddr_t bpf_alloc();
86 #include <net/bpf_compat.h>
87 #define BPF_BUFSIZE (MCLBYTES-8)
88 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, code, uio)
89 #else
90 #define BPF_BUFSIZE 4096
91 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, uio)
92 #endif
93 
94 #define PRINET  26			/* interruptible */
95 
96 /*
97  * The default read buffer size is patchable.
98  */
99 int bpf_bufsize = BPF_BUFSIZE;
100 
101 /*
102  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
103  *  bpf_dtab holds the descriptors, indexed by minor device #
104  */
105 struct bpf_if	*bpf_iflist;
106 struct bpf_d	bpf_dtab[NBPFILTER];
107 
108 #if BSD >= 199207 || NetBSD0_9 >= 2
109 /*
110  * bpfilterattach() is called at boot time in new systems.  We do
111  * nothing here since old systems will not call this.
112  */
113 /* ARGSUSED */
114 void
115 bpfilterattach(n)
116 	int n;
117 {
118 }
119 #endif
120 
121 static int	bpf_allocbufs __P((struct bpf_d *));
122 static int	bpf_allocbufs __P((struct bpf_d *));
123 static void	bpf_freed __P((struct bpf_d *));
124 static void	bpf_freed __P((struct bpf_d *));
125 static void	bpf_ifname __P((struct ifnet *, struct ifreq *));
126 static void	bpf_ifname __P((struct ifnet *, struct ifreq *));
127 static void	bpf_mcopy __P((const void *, void *, u_int));
128 static int	bpf_movein __P((struct uio *, int,
129 		    struct mbuf **, struct sockaddr *, int *));
130 static int	bpf_setif __P((struct bpf_d *, struct ifreq *));
131 static int	bpf_setif __P((struct bpf_d *, struct ifreq *));
132 static inline void
133 		bpf_wakeup __P((struct bpf_d *));
134 static void	catchpacket __P((struct bpf_d *, u_char *, u_int,
135 		    u_int, void (*)(const void *, void *, u_int)));
136 static void	reset_d __P((struct bpf_d *));
137 
138 static int
139 bpf_movein(uio, linktype, mp, sockp, datlen)
140 	register struct uio *uio;
141 	int linktype, *datlen;
142 	register struct mbuf **mp;
143 	register struct sockaddr *sockp;
144 {
145 	struct mbuf *m;
146 	int error;
147 	int len;
148 	int hlen;
149 
150 	/*
151 	 * Build a sockaddr based on the data link layer type.
152 	 * We do this at this level because the ethernet header
153 	 * is copied directly into the data field of the sockaddr.
154 	 * In the case of SLIP, there is no header and the packet
155 	 * is forwarded as is.
156 	 * Also, we are careful to leave room at the front of the mbuf
157 	 * for the link level header.
158 	 */
159 	switch (linktype) {
160 
161 	case DLT_SLIP:
162 		sockp->sa_family = AF_INET;
163 		hlen = 0;
164 		break;
165 
166 	case DLT_PPP:
167 		sockp->sa_family = AF_UNSPEC;
168 		hlen = 0;
169 		break;
170 
171 	case DLT_EN10MB:
172 		sockp->sa_family = AF_UNSPEC;
173 		/* XXX Would MAXLINKHDR be better? */
174 		hlen = sizeof(struct ether_header);
175 		break;
176 
177 	case DLT_FDDI:
178 		sockp->sa_family = AF_UNSPEC;
179 		/* XXX 4(FORMAC)+6(dst)+6(src)+3(LLC)+5(SNAP) */
180 		hlen = 24;
181 		break;
182 
183 	case DLT_NULL:
184 		sockp->sa_family = AF_UNSPEC;
185 		hlen = 0;
186 		break;
187 
188 	default:
189 		return (EIO);
190 	}
191 
192 	len = uio->uio_resid;
193 	*datlen = len - hlen;
194 	if ((unsigned)len > MCLBYTES)
195 		return (EIO);
196 
197 	MGET(m, M_WAIT, MT_DATA);
198 	if (m == 0)
199 		return (ENOBUFS);
200 	if (len > MLEN) {
201 #if BSD >= 199103
202 		MCLGET(m, M_WAIT);
203 		if ((m->m_flags & M_EXT) == 0) {
204 #else
205 		MCLGET(m);
206 		if (m->m_len != MCLBYTES) {
207 #endif
208 			error = ENOBUFS;
209 			goto bad;
210 		}
211 	}
212 	m->m_len = len;
213 	*mp = m;
214 	/*
215 	 * Make room for link header.
216 	 */
217 	if (hlen != 0) {
218 		m->m_len -= hlen;
219 #if BSD >= 199103
220 		m->m_data += hlen; /* XXX */
221 #else
222 		m->m_off += hlen;
223 #endif
224 		error = UIOMOVE((caddr_t)sockp->sa_data, hlen, UIO_WRITE, uio);
225 		if (error)
226 			goto bad;
227 	}
228 	error = UIOMOVE(mtod(m, caddr_t), len - hlen, UIO_WRITE, uio);
229 	if (!error)
230 		return (0);
231  bad:
232 	m_freem(m);
233 	return (error);
234 }
235 
236 /*
237  * Attach file to the bpf interface, i.e. make d listen on bp.
238  * Must be called at splimp.
239  */
240 static void
241 bpf_attachd(d, bp)
242 	struct bpf_d *d;
243 	struct bpf_if *bp;
244 {
245 	/*
246 	 * Point d at bp, and add d to the interface's list of listeners.
247 	 * Finally, point the driver's bpf cookie at the interface so
248 	 * it will divert packets to bpf.
249 	 */
250 	d->bd_bif = bp;
251 	d->bd_next = bp->bif_dlist;
252 	bp->bif_dlist = d;
253 
254 	*bp->bif_driverp = bp;
255 }
256 
257 /*
258  * Detach a file from its interface.
259  */
260 static void
261 bpf_detachd(d)
262 	struct bpf_d *d;
263 {
264 	struct bpf_d **p;
265 	struct bpf_if *bp;
266 
267 	bp = d->bd_bif;
268 	/*
269 	 * Check if this descriptor had requested promiscuous mode.
270 	 * If so, turn it off.
271 	 */
272 	if (d->bd_promisc) {
273 		d->bd_promisc = 0;
274 		if (ifpromisc(bp->bif_ifp, 0))
275 			/*
276 			 * Something is really wrong if we were able to put
277 			 * the driver into promiscuous mode, but can't
278 			 * take it out.
279 			 */
280 			panic("bpf: ifpromisc failed");
281 	}
282 	/* Remove d from the interface's descriptor list. */
283 	p = &bp->bif_dlist;
284 	while (*p != d) {
285 		p = &(*p)->bd_next;
286 		if (*p == 0)
287 			panic("bpf_detachd: descriptor not in list");
288 	}
289 	*p = (*p)->bd_next;
290 	if (bp->bif_dlist == 0)
291 		/*
292 		 * Let the driver know that there are no more listeners.
293 		 */
294 		*d->bd_bif->bif_driverp = 0;
295 	d->bd_bif = 0;
296 }
297 
298 
299 /*
300  * Mark a descriptor free by making it point to itself.
301  * This is probably cheaper than marking with a constant since
302  * the address should be in a register anyway.
303  */
304 #define D_ISFREE(d) ((d) == (d)->bd_next)
305 #define D_MARKFREE(d) ((d)->bd_next = (d))
306 #define D_MARKUSED(d) ((d)->bd_next = 0)
307 
308 /*
309  * Open ethernet device.  Returns ENXIO for illegal minor device number,
310  * EBUSY if file is open by another process.
311  */
312 /* ARGSUSED */
313 int
314 bpfopen(dev, flag)
315 	dev_t dev;
316 	int flag;
317 {
318 	register struct bpf_d *d;
319 
320 	if (minor(dev) >= NBPFILTER)
321 		return (ENXIO);
322 	/*
323 	 * Each minor can be opened by only one process.  If the requested
324 	 * minor is in use, return EBUSY.
325 	 */
326 	d = &bpf_dtab[minor(dev)];
327 	if (!D_ISFREE(d))
328 		return (EBUSY);
329 
330 	/* Mark "free" and do most initialization. */
331 	bzero((char *)d, sizeof(*d));
332 	d->bd_bufsize = bpf_bufsize;
333 
334 	return (0);
335 }
336 
337 /*
338  * Close the descriptor by detaching it from its interface,
339  * deallocating its buffers, and marking it free.
340  */
341 /* ARGSUSED */
342 int
343 bpfclose(dev, flag)
344 	dev_t dev;
345 	int flag;
346 {
347 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
348 	register int s;
349 
350 	s = splimp();
351 	if (d->bd_bif)
352 		bpf_detachd(d);
353 	splx(s);
354 	bpf_freed(d);
355 
356 	return (0);
357 }
358 
359 /*
360  * Support for SunOS, which does not have tsleep.
361  */
362 #if BSD < 199103
363 static
364 bpf_timeout(arg)
365 	caddr_t arg;
366 {
367 	struct bpf_d *d = (struct bpf_d *)arg;
368 	d->bd_timedout = 1;
369 	wakeup(arg);
370 }
371 
372 #define BPF_SLEEP(chan, pri, s, t) bpf_sleep((struct bpf_d *)chan)
373 
374 int
375 bpf_sleep(d)
376 	register struct bpf_d *d;
377 {
378 	register int rto = d->bd_rtout;
379 	register int st;
380 
381 	if (rto != 0) {
382 		d->bd_timedout = 0;
383 		timeout(bpf_timeout, (caddr_t)d, rto);
384 	}
385 	st = sleep((caddr_t)d, PRINET|PCATCH);
386 	if (rto != 0) {
387 		if (d->bd_timedout == 0)
388 			untimeout(bpf_timeout, (caddr_t)d);
389 		else if (st == 0)
390 			return EWOULDBLOCK;
391 	}
392 	return (st != 0) ? EINTR : 0;
393 }
394 #else
395 #define BPF_SLEEP tsleep
396 #endif
397 
398 /*
399  * Rotate the packet buffers in descriptor d.  Move the store buffer
400  * into the hold slot, and the free buffer into the store slot.
401  * Zero the length of the new store buffer.
402  */
403 #define ROTATE_BUFFERS(d) \
404 	(d)->bd_hbuf = (d)->bd_sbuf; \
405 	(d)->bd_hlen = (d)->bd_slen; \
406 	(d)->bd_sbuf = (d)->bd_fbuf; \
407 	(d)->bd_slen = 0; \
408 	(d)->bd_fbuf = 0;
409 /*
410  *  bpfread - read next chunk of packets from buffers
411  */
412 int
413 bpfread(dev, uio)
414 	dev_t dev;
415 	register struct uio *uio;
416 {
417 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
418 	int error;
419 	int s;
420 
421 	/*
422 	 * Restrict application to use a buffer the same size as
423 	 * as kernel buffers.
424 	 */
425 	if (uio->uio_resid != d->bd_bufsize)
426 		return (EINVAL);
427 
428 	s = splimp();
429 	/*
430 	 * If the hold buffer is empty, then do a timed sleep, which
431 	 * ends when the timeout expires or when enough packets
432 	 * have arrived to fill the store buffer.
433 	 */
434 	while (d->bd_hbuf == 0) {
435 		if (d->bd_immediate && d->bd_slen != 0) {
436 			/*
437 			 * A packet(s) either arrived since the previous
438 			 * read or arrived while we were asleep.
439 			 * Rotate the buffers and return what's here.
440 			 */
441 			ROTATE_BUFFERS(d);
442 			break;
443 		}
444 		error = BPF_SLEEP((caddr_t)d, PRINET|PCATCH, "bpf",
445 				  d->bd_rtout);
446 		if (error == EINTR || error == ERESTART) {
447 			splx(s);
448 			return (error);
449 		}
450 		if (error == EWOULDBLOCK) {
451 			/*
452 			 * On a timeout, return what's in the buffer,
453 			 * which may be nothing.  If there is something
454 			 * in the store buffer, we can rotate the buffers.
455 			 */
456 			if (d->bd_hbuf)
457 				/*
458 				 * We filled up the buffer in between
459 				 * getting the timeout and arriving
460 				 * here, so we don't need to rotate.
461 				 */
462 				break;
463 
464 			if (d->bd_slen == 0) {
465 				splx(s);
466 				return (0);
467 			}
468 			ROTATE_BUFFERS(d);
469 			break;
470 		}
471 	}
472 	/*
473 	 * At this point, we know we have something in the hold slot.
474 	 */
475 	splx(s);
476 
477 	/*
478 	 * Move data from hold buffer into user space.
479 	 * We know the entire buffer is transferred since
480 	 * we checked above that the read buffer is bpf_bufsize bytes.
481 	 */
482 	error = UIOMOVE(d->bd_hbuf, d->bd_hlen, UIO_READ, uio);
483 
484 	s = splimp();
485 	d->bd_fbuf = d->bd_hbuf;
486 	d->bd_hbuf = 0;
487 	d->bd_hlen = 0;
488 	splx(s);
489 
490 	return (error);
491 }
492 
493 
494 /*
495  * If there are processes sleeping on this descriptor, wake them up.
496  */
497 static inline void
498 bpf_wakeup(d)
499 	register struct bpf_d *d;
500 {
501 	wakeup((caddr_t)d);
502 #if BSD >= 199103
503 	selwakeup(&d->bd_sel);
504 	/* XXX */
505 	d->bd_sel.si_pid = 0;
506 #else
507 	if (d->bd_selproc) {
508 		selwakeup(d->bd_selproc, (int)d->bd_selcoll);
509 		d->bd_selcoll = 0;
510 		d->bd_selproc = 0;
511 	}
512 #endif
513 }
514 
515 int
516 bpfwrite(dev, uio)
517 	dev_t dev;
518 	struct uio *uio;
519 {
520 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
521 	struct ifnet *ifp;
522 	struct mbuf *m;
523 	int error, s;
524 	static struct sockaddr dst;
525 	int datlen;
526 
527 	if (d->bd_bif == 0)
528 		return (ENXIO);
529 
530 	ifp = d->bd_bif->bif_ifp;
531 
532 	if (uio->uio_resid == 0)
533 		return (0);
534 
535 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, &dst, &datlen);
536 	if (error)
537 		return (error);
538 
539 	if (datlen > ifp->if_mtu)
540 		return (EMSGSIZE);
541 
542 	s = splnet();
543 #if BSD >= 199103
544 	error = (*ifp->if_output)(ifp, m, &dst, (struct rtentry *)0);
545 #else
546 	error = (*ifp->if_output)(ifp, m, &dst);
547 #endif
548 	splx(s);
549 	/*
550 	 * The driver frees the mbuf.
551 	 */
552 	return (error);
553 }
554 
555 /*
556  * Reset a descriptor by flushing its packet buffer and clearing the
557  * receive and drop counts.  Should be called at splimp.
558  */
559 static void
560 reset_d(d)
561 	struct bpf_d *d;
562 {
563 	if (d->bd_hbuf) {
564 		/* Free the hold buffer. */
565 		d->bd_fbuf = d->bd_hbuf;
566 		d->bd_hbuf = 0;
567 	}
568 	d->bd_slen = 0;
569 	d->bd_hlen = 0;
570 	d->bd_rcount = 0;
571 	d->bd_dcount = 0;
572 }
573 
574 /*
575  *  FIONREAD		Check for read packet available.
576  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
577  *  BIOCGBLEN		Get buffer len [for read()].
578  *  BIOCSETF		Set ethernet read filter.
579  *  BIOCFLUSH		Flush read packet buffer.
580  *  BIOCPROMISC		Put interface into promiscuous mode.
581  *  BIOCGDLT		Get link layer type.
582  *  BIOCGETIF		Get interface name.
583  *  BIOCSETIF		Set interface.
584  *  BIOCSRTIMEOUT	Set read timeout.
585  *  BIOCGRTIMEOUT	Get read timeout.
586  *  BIOCGSTATS		Get packet stats.
587  *  BIOCIMMEDIATE	Set immediate mode.
588  *  BIOCVERSION		Get filter language version.
589  */
590 /* ARGSUSED */
591 int
592 bpfioctl(dev, cmd, addr, flag)
593 	dev_t dev;
594 	int cmd;
595 	caddr_t addr;
596 	int flag;
597 {
598 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
599 	int s, error = 0;
600 
601 	switch (cmd) {
602 
603 	default:
604 		error = EINVAL;
605 		break;
606 
607 	/*
608 	 * Check for read packet available.
609 	 */
610 	case FIONREAD:
611 		{
612 			int n;
613 
614 			s = splimp();
615 			n = d->bd_slen;
616 			if (d->bd_hbuf)
617 				n += d->bd_hlen;
618 			splx(s);
619 
620 			*(int *)addr = n;
621 			break;
622 		}
623 
624 	case SIOCGIFADDR:
625 		{
626 			struct ifnet *ifp;
627 
628 			if (d->bd_bif == 0)
629 				error = EINVAL;
630 			else {
631 				ifp = d->bd_bif->bif_ifp;
632 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
633 			}
634 			break;
635 		}
636 
637 	/*
638 	 * Get buffer len [for read()].
639 	 */
640 	case BIOCGBLEN:
641 		*(u_int *)addr = d->bd_bufsize;
642 		break;
643 
644 	/*
645 	 * Set buffer length.
646 	 */
647 	case BIOCSBLEN:
648 #if BSD < 199103
649 		error = EINVAL;
650 #else
651 		if (d->bd_bif != 0)
652 			error = EINVAL;
653 		else {
654 			register u_int size = *(u_int *)addr;
655 
656 			if (size > BPF_MAXBUFSIZE)
657 				*(u_int *)addr = size = BPF_MAXBUFSIZE;
658 			else if (size < BPF_MINBUFSIZE)
659 				*(u_int *)addr = size = BPF_MINBUFSIZE;
660 			d->bd_bufsize = size;
661 		}
662 #endif
663 		break;
664 
665 	/*
666 	 * Set link layer read filter.
667 	 */
668 	case BIOCSETF:
669 		error = bpf_setf(d, (struct bpf_program *)addr);
670 		break;
671 
672 	/*
673 	 * Flush read packet buffer.
674 	 */
675 	case BIOCFLUSH:
676 		s = splimp();
677 		reset_d(d);
678 		splx(s);
679 		break;
680 
681 	/*
682 	 * Put interface into promiscuous mode.
683 	 */
684 	case BIOCPROMISC:
685 		if (d->bd_bif == 0) {
686 			/*
687 			 * No interface attached yet.
688 			 */
689 			error = EINVAL;
690 			break;
691 		}
692 		s = splimp();
693 		if (d->bd_promisc == 0) {
694 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
695 			if (error == 0)
696 				d->bd_promisc = 1;
697 		}
698 		splx(s);
699 		break;
700 
701 	/*
702 	 * Get device parameters.
703 	 */
704 	case BIOCGDLT:
705 		if (d->bd_bif == 0)
706 			error = EINVAL;
707 		else
708 			*(u_int *)addr = d->bd_bif->bif_dlt;
709 		break;
710 
711 	/*
712 	 * Set interface name.
713 	 */
714 	case BIOCGETIF:
715 		if (d->bd_bif == 0)
716 			error = EINVAL;
717 		else
718 			bpf_ifname(d->bd_bif->bif_ifp, (struct ifreq *)addr);
719 		break;
720 
721 	/*
722 	 * Set interface.
723 	 */
724 	case BIOCSETIF:
725 		error = bpf_setif(d, (struct ifreq *)addr);
726 		break;
727 
728 	/*
729 	 * Set read timeout.
730 	 */
731 	case BIOCSRTIMEOUT:
732 		{
733 			struct timeval *tv = (struct timeval *)addr;
734 			u_long msec;
735 
736 			/* Compute number of milliseconds. */
737 			msec = tv->tv_sec * 1000 + tv->tv_usec / 1000;
738 			/* Scale milliseconds to ticks.  Assume hard
739 			   clock has millisecond or greater resolution
740 			   (i.e. tick >= 1000).  For 10ms hardclock,
741 			   tick/1000 = 10, so rtout<-msec/10. */
742 			d->bd_rtout = msec / (tick / 1000);
743 			break;
744 		}
745 
746 	/*
747 	 * Get read timeout.
748 	 */
749 	case BIOCGRTIMEOUT:
750 		{
751 			struct timeval *tv = (struct timeval *)addr;
752 			u_long msec = d->bd_rtout;
753 
754 			msec *= tick / 1000;
755 			tv->tv_sec = msec / 1000;
756 			tv->tv_usec = msec % 1000;
757 			break;
758 		}
759 
760 	/*
761 	 * Get packet stats.
762 	 */
763 	case BIOCGSTATS:
764 		{
765 			struct bpf_stat *bs = (struct bpf_stat *)addr;
766 
767 			bs->bs_recv = d->bd_rcount;
768 			bs->bs_drop = d->bd_dcount;
769 			break;
770 		}
771 
772 	/*
773 	 * Set immediate mode.
774 	 */
775 	case BIOCIMMEDIATE:
776 		d->bd_immediate = *(u_int *)addr;
777 		break;
778 
779 	case BIOCVERSION:
780 		{
781 			struct bpf_version *bv = (struct bpf_version *)addr;
782 
783 			bv->bv_major = BPF_MAJOR_VERSION;
784 			bv->bv_minor = BPF_MINOR_VERSION;
785 			break;
786 		}
787 	}
788 	return (error);
789 }
790 
791 /*
792  * Set d's packet filter program to fp.  If this file already has a filter,
793  * free it and replace it.  Returns EINVAL for bogus requests.
794  */
795 int
796 bpf_setf(d, fp)
797 	struct bpf_d *d;
798 	struct bpf_program *fp;
799 {
800 	struct bpf_insn *fcode, *old;
801 	u_int flen, size;
802 	int s;
803 
804 	old = d->bd_filter;
805 	if (fp->bf_insns == 0) {
806 		if (fp->bf_len != 0)
807 			return (EINVAL);
808 		s = splimp();
809 		d->bd_filter = 0;
810 		reset_d(d);
811 		splx(s);
812 		if (old != 0)
813 			free((caddr_t)old, M_DEVBUF);
814 		return (0);
815 	}
816 	flen = fp->bf_len;
817 	if (flen > BPF_MAXINSNS)
818 		return (EINVAL);
819 
820 	size = flen * sizeof(*fp->bf_insns);
821 	fcode = (struct bpf_insn *)malloc(size, M_DEVBUF, M_WAITOK);
822 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
823 	    bpf_validate(fcode, (int)flen)) {
824 		s = splimp();
825 		d->bd_filter = fcode;
826 		reset_d(d);
827 		splx(s);
828 		if (old != 0)
829 			free((caddr_t)old, M_DEVBUF);
830 
831 		return (0);
832 	}
833 	free((caddr_t)fcode, M_DEVBUF);
834 	return (EINVAL);
835 }
836 
837 /*
838  * Detach a file from its current interface (if attached at all) and attach
839  * to the interface indicated by the name stored in ifr.
840  * Return an errno or 0.
841  */
842 static int
843 bpf_setif(d, ifr)
844 	struct bpf_d *d;
845 	struct ifreq *ifr;
846 {
847 	struct bpf_if *bp;
848 	char *cp;
849 	int unit, s, error;
850 
851 	/*
852 	 * Separate string into name part and unit number.  Put a null
853 	 * byte at the end of the name part, and compute the number.
854 	 * If the a unit number is unspecified, the default is 0,
855 	 * as initialized above.  XXX This should be common code.
856 	 */
857 	unit = 0;
858 	cp = ifr->ifr_name;
859 	cp[sizeof(ifr->ifr_name) - 1] = '\0';
860 	while (*cp++) {
861 		if (*cp >= '0' && *cp <= '9') {
862 			unit = *cp - '0';
863 			*cp++ = '\0';
864 			while (*cp)
865 				unit = 10 * unit + *cp++ - '0';
866 			break;
867 		}
868 	}
869 	/*
870 	 * Look through attached interfaces for the named one.
871 	 */
872 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
873 		struct ifnet *ifp = bp->bif_ifp;
874 
875 		if (ifp == 0 || unit != ifp->if_unit
876 		    || strcmp(ifp->if_name, ifr->ifr_name) != 0)
877 			continue;
878 		/*
879 		 * We found the requested interface.
880 		 * If it's not up, return an error.
881 		 * Allocate the packet buffers if we need to.
882 		 * If we're already attached to requested interface,
883 		 * just flush the buffer.
884 		 */
885 		if ((ifp->if_flags & IFF_UP) == 0)
886 			return (ENETDOWN);
887 
888 		if (d->bd_sbuf == 0) {
889 			error = bpf_allocbufs(d);
890 			if (error != 0)
891 				return (error);
892 		}
893 		s = splimp();
894 		if (bp != d->bd_bif) {
895 			if (d->bd_bif)
896 				/*
897 				 * Detach if attached to something else.
898 				 */
899 				bpf_detachd(d);
900 
901 			bpf_attachd(d, bp);
902 		}
903 		reset_d(d);
904 		splx(s);
905 		return (0);
906 	}
907 	/* Not found. */
908 	return (ENXIO);
909 }
910 
911 /*
912  * Convert an interface name plus unit number of an ifp to a single
913  * name which is returned in the ifr.
914  */
915 static void
916 bpf_ifname(ifp, ifr)
917 	struct ifnet *ifp;
918 	struct ifreq *ifr;
919 {
920 	char *s = ifp->if_name;
921 	char *d = ifr->ifr_name;
922 
923 	while (*d++ = *s++)
924 		continue;
925 	/* XXX Assume that unit number is less than 10. */
926 	*d++ = ifp->if_unit + '0';
927 	*d = '\0';
928 }
929 
930 /*
931  * The new select interface passes down the proc pointer; the old select
932  * stubs had to grab it out of the user struct.  This glue allows either case.
933  */
934 #if BSD >= 199103
935 #define bpf_select bpfselect
936 #else
937 int
938 bpfselect(dev, rw)
939 	register dev_t dev;
940 	int rw;
941 {
942 	return (bpf_select(dev, rw, u.u_procp));
943 }
944 #endif
945 
946 /*
947  * Support for select() system call
948  *
949  * Return true iff the specific operation will not block indefinitely.
950  * Otherwise, return false but make a note that a selwakeup() must be done.
951  */
952 int
953 bpf_select(dev, rw, p)
954 	register dev_t dev;
955 	int rw;
956 	struct proc *p;
957 {
958 	register struct bpf_d *d;
959 	register int s;
960 
961 	if (rw != FREAD)
962 		return (0);
963 	/*
964 	 * An imitation of the FIONREAD ioctl code.
965 	 */
966 	d = &bpf_dtab[minor(dev)];
967 
968 	s = splimp();
969 	if (d->bd_hlen != 0 || (d->bd_immediate && d->bd_slen != 0)) {
970 		/*
971 		 * There is data waiting.
972 		 */
973 		splx(s);
974 		return (1);
975 	}
976 #if BSD >= 199103
977 	selrecord(p, &d->bd_sel);
978 #else
979 	/*
980 	 * No data ready.  If there's already a select() waiting on this
981 	 * minor device then this is a collision.  This shouldn't happen
982 	 * because minors really should not be shared, but if a process
983 	 * forks while one of these is open, it is possible that both
984 	 * processes could select on the same descriptor.
985 	 */
986 	if (d->bd_selproc && d->bd_selproc->p_wchan == (caddr_t)&selwait)
987 		d->bd_selcoll = 1;
988 	else
989 		d->bd_selproc = p;
990 #endif
991 	splx(s);
992 	return (0);
993 }
994 
995 /*
996  * Incoming linkage from device drivers.  Process the packet pkt, of length
997  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
998  * by each process' filter, and if accepted, stashed into the corresponding
999  * buffer.
1000  */
1001 void
1002 bpf_tap(arg, pkt, pktlen)
1003 	caddr_t arg;
1004 	register u_char *pkt;
1005 	register u_int pktlen;
1006 {
1007 	struct bpf_if *bp;
1008 	register struct bpf_d *d;
1009 	register u_int slen;
1010 	/*
1011 	 * Note that the ipl does not have to be raised at this point.
1012 	 * The only problem that could arise here is that if two different
1013 	 * interfaces shared any data.  This is not the case.
1014 	 */
1015 	bp = (struct bpf_if *)arg;
1016 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1017 		++d->bd_rcount;
1018 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
1019 		if (slen != 0)
1020 			catchpacket(d, pkt, pktlen, slen, bcopy);
1021 	}
1022 }
1023 
1024 /*
1025  * Copy data from an mbuf chain into a buffer.  This code is derived
1026  * from m_copydata in sys/uipc_mbuf.c.
1027  */
1028 static void
1029 bpf_mcopy(src_arg, dst_arg, len)
1030 	const void *src_arg;
1031 	void *dst_arg;
1032 	register u_int len;
1033 {
1034 	register const struct mbuf *m;
1035 	register u_int count;
1036 	u_char *dst;
1037 
1038 	m = src_arg;
1039 	dst = dst_arg;
1040 	while (len > 0) {
1041 		if (m == 0)
1042 			panic("bpf_mcopy");
1043 		count = min(m->m_len, len);
1044 		bcopy(mtod(m, caddr_t), (caddr_t)dst, count);
1045 		m = m->m_next;
1046 		dst += count;
1047 		len -= count;
1048 	}
1049 }
1050 
1051 /*
1052  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1053  */
1054 void
1055 bpf_mtap(arg, m)
1056 	caddr_t arg;
1057 	struct mbuf *m;
1058 {
1059 	struct bpf_if *bp = (struct bpf_if *)arg;
1060 	struct bpf_d *d;
1061 	u_int pktlen, slen;
1062 	struct mbuf *m0;
1063 
1064 	pktlen = 0;
1065 	for (m0 = m; m0 != 0; m0 = m0->m_next)
1066 		pktlen += m0->m_len;
1067 
1068 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1069 		++d->bd_rcount;
1070 		slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
1071 		if (slen != 0)
1072 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
1073 	}
1074 }
1075 
1076 /*
1077  * Move the packet data from interface memory (pkt) into the
1078  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1079  * otherwise 0.  "copy" is the routine called to do the actual data
1080  * transfer.  bcopy is passed in to copy contiguous chunks, while
1081  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1082  * pkt is really an mbuf.
1083  */
1084 static void
1085 catchpacket(d, pkt, pktlen, snaplen, cpfn)
1086 	register struct bpf_d *d;
1087 	register u_char *pkt;
1088 	register u_int pktlen, snaplen;
1089 	register void (*cpfn) __P((const void *, void *, u_int));
1090 {
1091 	register struct bpf_hdr *hp;
1092 	register int totlen, curlen;
1093 	register int hdrlen = d->bd_bif->bif_hdrlen;
1094 	/*
1095 	 * Figure out how many bytes to move.  If the packet is
1096 	 * greater or equal to the snapshot length, transfer that
1097 	 * much.  Otherwise, transfer the whole packet (unless
1098 	 * we hit the buffer size limit).
1099 	 */
1100 	totlen = hdrlen + min(snaplen, pktlen);
1101 	if (totlen > d->bd_bufsize)
1102 		totlen = d->bd_bufsize;
1103 
1104 	/*
1105 	 * Round up the end of the previous packet to the next longword.
1106 	 */
1107 	curlen = BPF_WORDALIGN(d->bd_slen);
1108 	if (curlen + totlen > d->bd_bufsize) {
1109 		/*
1110 		 * This packet will overflow the storage buffer.
1111 		 * Rotate the buffers if we can, then wakeup any
1112 		 * pending reads.
1113 		 */
1114 		if (d->bd_fbuf == 0) {
1115 			/*
1116 			 * We haven't completed the previous read yet,
1117 			 * so drop the packet.
1118 			 */
1119 			++d->bd_dcount;
1120 			return;
1121 		}
1122 		ROTATE_BUFFERS(d);
1123 		bpf_wakeup(d);
1124 		curlen = 0;
1125 	}
1126 	else if (d->bd_immediate)
1127 		/*
1128 		 * Immediate mode is set.  A packet arrived so any
1129 		 * reads should be woken up.
1130 		 */
1131 		bpf_wakeup(d);
1132 
1133 	/*
1134 	 * Append the bpf header.
1135 	 */
1136 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1137 #if BSD >= 199103
1138 	microtime(&hp->bh_tstamp);
1139 #elif defined(sun)
1140 	uniqtime(&hp->bh_tstamp);
1141 #else
1142 	hp->bh_tstamp = time;
1143 #endif
1144 	hp->bh_datalen = pktlen;
1145 	hp->bh_hdrlen = hdrlen;
1146 	/*
1147 	 * Copy the packet data into the store buffer and update its length.
1148 	 */
1149 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1150 	d->bd_slen = curlen + totlen;
1151 }
1152 
1153 /*
1154  * Initialize all nonzero fields of a descriptor.
1155  */
1156 static int
1157 bpf_allocbufs(d)
1158 	register struct bpf_d *d;
1159 {
1160 	d->bd_fbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
1161 	if (d->bd_fbuf == 0)
1162 		return (ENOBUFS);
1163 
1164 	d->bd_sbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
1165 	if (d->bd_sbuf == 0) {
1166 		free(d->bd_fbuf, M_DEVBUF);
1167 		return (ENOBUFS);
1168 	}
1169 	d->bd_slen = 0;
1170 	d->bd_hlen = 0;
1171 	return (0);
1172 }
1173 
1174 /*
1175  * Free buffers currently in use by a descriptor.
1176  * Called on close.
1177  */
1178 static void
1179 bpf_freed(d)
1180 	register struct bpf_d *d;
1181 {
1182 	/*
1183 	 * We don't need to lock out interrupts since this descriptor has
1184 	 * been detached from its interface and it yet hasn't been marked
1185 	 * free.
1186 	 */
1187 	if (d->bd_sbuf != 0) {
1188 		free(d->bd_sbuf, M_DEVBUF);
1189 		if (d->bd_hbuf != 0)
1190 			free(d->bd_hbuf, M_DEVBUF);
1191 		if (d->bd_fbuf != 0)
1192 			free(d->bd_fbuf, M_DEVBUF);
1193 	}
1194 	if (d->bd_filter)
1195 		free((caddr_t)d->bd_filter, M_DEVBUF);
1196 
1197 	D_MARKFREE(d);
1198 }
1199 
1200 /*
1201  * Attach an interface to bpf.  driverp is a pointer to a (struct bpf_if *)
1202  * in the driver's softc; dlt is the link layer type; hdrlen is the fixed
1203  * size of the link header (variable length headers not yet supported).
1204  */
1205 void
1206 bpfattach(driverp, ifp, dlt, hdrlen)
1207 	caddr_t *driverp;
1208 	struct ifnet *ifp;
1209 	u_int dlt, hdrlen;
1210 {
1211 	struct bpf_if *bp;
1212 	int i;
1213 #if BSD < 199103
1214 	static struct bpf_if bpf_ifs[NBPFILTER];
1215 	static int bpfifno;
1216 
1217 	bp = (bpfifno < NBPFILTER) ? &bpf_ifs[bpfifno++] : 0;
1218 #else
1219 	bp = (struct bpf_if *)malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1220 #endif
1221 	if (bp == 0)
1222 		panic("bpfattach");
1223 
1224 	bp->bif_dlist = 0;
1225 	bp->bif_driverp = (struct bpf_if **)driverp;
1226 	bp->bif_ifp = ifp;
1227 	bp->bif_dlt = dlt;
1228 
1229 	bp->bif_next = bpf_iflist;
1230 	bpf_iflist = bp;
1231 
1232 	*bp->bif_driverp = 0;
1233 
1234 	/*
1235 	 * Compute the length of the bpf header.  This is not necessarily
1236 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1237 	 * that the network layer header begins on a longword boundary (for
1238 	 * performance reasons and to alleviate alignment restrictions).
1239 	 */
1240 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1241 
1242 	/*
1243 	 * Mark all the descriptors free if this hasn't been done.
1244 	 */
1245 	if (!D_ISFREE(&bpf_dtab[0]))
1246 		for (i = 0; i < NBPFILTER; ++i)
1247 			D_MARKFREE(&bpf_dtab[i]);
1248 
1249 #if 0
1250 	printf("bpf: %s%d attached\n", ifp->if_name, ifp->if_unit);
1251 #endif
1252 }
1253 
1254 #if BSD >= 199103
1255 /* XXX This routine belongs in net/if.c. */
1256 /*
1257  * Set/clear promiscuous mode on interface ifp based on the truth value
1258  * of pswitch.  The calls are reference counted so that only the first
1259  * "on" request actually has an effect, as does the final "off" request.
1260  * Results are undefined if the "off" and "on" requests are not matched.
1261  */
1262 int
1263 ifpromisc(ifp, pswitch)
1264 	struct ifnet *ifp;
1265 	int pswitch;
1266 {
1267 	struct ifreq ifr;
1268 	/*
1269 	 * If the device is not configured up, we cannot put it in
1270 	 * promiscuous mode.
1271 	 */
1272 	if ((ifp->if_flags & IFF_UP) == 0)
1273 		return (ENETDOWN);
1274 
1275 	if (pswitch) {
1276 		if (ifp->if_pcount++ != 0)
1277 			return (0);
1278 		ifp->if_flags |= IFF_PROMISC;
1279 	} else {
1280 		if (--ifp->if_pcount > 0)
1281 			return (0);
1282 		ifp->if_flags &= ~IFF_PROMISC;
1283 	}
1284 	ifr.ifr_flags = ifp->if_flags;
1285 	return ((*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr));
1286 }
1287 #endif
1288 
1289 #if BSD < 199103
1290 /*
1291  * Allocate some memory for bpf.  This is temporary SunOS support, and
1292  * is admittedly a hack.
1293  * If resources unavaiable, return 0.
1294  */
1295 static caddr_t
1296 bpf_alloc(size, canwait)
1297 	register int size;
1298 	register int canwait;
1299 {
1300 	register struct mbuf *m;
1301 
1302 	if ((unsigned)size > (MCLBYTES-8))
1303 		return 0;
1304 
1305 	MGET(m, canwait, MT_DATA);
1306 	if (m == 0)
1307 		return 0;
1308 	if ((unsigned)size > (MLEN-8)) {
1309 		MCLGET(m);
1310 		if (m->m_len != MCLBYTES) {
1311 			m_freem(m);
1312 			return 0;
1313 		}
1314 	}
1315 	*mtod(m, struct mbuf **) = m;
1316 	return mtod(m, caddr_t) + 8;
1317 }
1318 #endif
1319