xref: /netbsd-src/sys/net/bpf.c (revision d0fed6c87ddc40a8bffa6f99e7433ddfc864dd83)
1 /*	$NetBSD: bpf.c,v 1.35 1997/03/17 06:45:20 scottr 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. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)bpf.c	8.2 (Berkeley) 3/28/94
41  */
42 
43 #include "bpfilter.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/mbuf.h>
48 #include <sys/buf.h>
49 #include <sys/time.h>
50 #include <sys/proc.h>
51 #include <sys/user.h>
52 #include <sys/ioctl.h>
53 #include <sys/map.h>
54 #include <sys/conf.h>
55 
56 #include <sys/file.h>
57 #if defined(sparc) && BSD < 199103
58 #include <sys/stream.h>
59 #endif
60 #include <sys/tty.h>
61 #include <sys/uio.h>
62 
63 #include <sys/protosw.h>
64 #include <sys/socket.h>
65 #include <sys/errno.h>
66 #include <sys/kernel.h>
67 #include <sys/poll.h>
68 
69 #include <net/if.h>
70 
71 #include <net/bpf.h>
72 #include <net/bpfdesc.h>
73 
74 #include <net/if_arc.h>
75 #include <net/if_ether.h>
76 
77 #include <netinet/in.h>
78 #include <netinet/if_inarp.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 *, size_t));
128 static int	bpf_movein __P((struct uio *, int,
129 			        struct mbuf **, struct sockaddr *));
130 static void	bpf_attachd __P((struct bpf_d *, struct bpf_if *));
131 static void	bpf_detachd __P((struct bpf_d *));
132 static int	bpf_setif __P((struct bpf_d *, struct ifreq *));
133 int		bpfpoll __P((dev_t, int, struct proc *));
134 static __inline void
135 		bpf_wakeup __P((struct bpf_d *));
136 static void	catchpacket __P((struct bpf_d *, u_char *, size_t, size_t,
137 				 void (*)(const void *, void *, size_t)));
138 static void	reset_d __P((struct bpf_d *));
139 
140 static int
141 bpf_movein(uio, linktype, mp, sockp)
142 	register struct uio *uio;
143 	int linktype;
144 	register struct mbuf **mp;
145 	register struct sockaddr *sockp;
146 {
147 	struct mbuf *m;
148 	int error;
149 	int len;
150 	int hlen;
151 
152 	/*
153 	 * Build a sockaddr based on the data link layer type.
154 	 * We do this at this level because the ethernet header
155 	 * is copied directly into the data field of the sockaddr.
156 	 * In the case of SLIP, there is no header and the packet
157 	 * is forwarded as is.
158 	 * Also, we are careful to leave room at the front of the mbuf
159 	 * for the link level header.
160 	 */
161 	switch (linktype) {
162 
163 	case DLT_SLIP:
164 		sockp->sa_family = AF_INET;
165 		hlen = 0;
166 		break;
167 
168 	case DLT_PPP:
169 		sockp->sa_family = AF_UNSPEC;
170 		hlen = 0;
171 		break;
172 
173 	case DLT_EN10MB:
174 		sockp->sa_family = AF_UNSPEC;
175 		/* XXX Would MAXLINKHDR be better? */
176 		hlen = sizeof(struct ether_header);
177 		break;
178 
179 	case DLT_ARCNET:
180 		sockp->sa_family = AF_UNSPEC;
181 		hlen = ARC_HDRLEN;
182 		break;
183 
184 	case DLT_FDDI:
185 		sockp->sa_family = AF_UNSPEC;
186 		/* XXX 4(FORMAC)+6(dst)+6(src)+3(LLC)+5(SNAP) */
187 		hlen = 24;
188 		break;
189 
190 	case DLT_NULL:
191 		sockp->sa_family = AF_UNSPEC;
192 		hlen = 0;
193 		break;
194 
195 	default:
196 		return (EIO);
197 	}
198 
199 	len = uio->uio_resid;
200 	if ((unsigned)len > MCLBYTES)
201 		return (EIO);
202 
203 	MGETHDR(m, M_WAIT, MT_DATA);
204 	m->m_pkthdr.rcvif = 0;
205 	m->m_pkthdr.len = len - hlen;
206 
207 	if (len > MHLEN) {
208 #if BSD >= 199103
209 		MCLGET(m, M_WAIT);
210 		if ((m->m_flags & M_EXT) == 0) {
211 #else
212 		MCLGET(m);
213 		if (m->m_len != MCLBYTES) {
214 #endif
215 			error = ENOBUFS;
216 			goto bad;
217 		}
218 	}
219 	m->m_len = len;
220 	*mp = m;
221 	/*
222 	 * Make room for link header.
223 	 */
224 	if (hlen != 0) {
225 		m->m_len -= hlen;
226 #if BSD >= 199103
227 		m->m_data += hlen; /* XXX */
228 #else
229 		m->m_off += hlen;
230 #endif
231 		error = UIOMOVE((caddr_t)sockp->sa_data, hlen, UIO_WRITE, uio);
232 		if (error)
233 			goto bad;
234 	}
235 	error = UIOMOVE(mtod(m, caddr_t), len - hlen, UIO_WRITE, uio);
236 	if (!error)
237 		return (0);
238  bad:
239 	m_freem(m);
240 	return (error);
241 }
242 
243 /*
244  * Attach file to the bpf interface, i.e. make d listen on bp.
245  * Must be called at splimp.
246  */
247 static void
248 bpf_attachd(d, bp)
249 	struct bpf_d *d;
250 	struct bpf_if *bp;
251 {
252 	/*
253 	 * Point d at bp, and add d to the interface's list of listeners.
254 	 * Finally, point the driver's bpf cookie at the interface so
255 	 * it will divert packets to bpf.
256 	 */
257 	d->bd_bif = bp;
258 	d->bd_next = bp->bif_dlist;
259 	bp->bif_dlist = d;
260 
261 	*bp->bif_driverp = bp;
262 }
263 
264 /*
265  * Detach a file from its interface.
266  */
267 static void
268 bpf_detachd(d)
269 	struct bpf_d *d;
270 {
271 	struct bpf_d **p;
272 	struct bpf_if *bp;
273 
274 	bp = d->bd_bif;
275 	/*
276 	 * Check if this descriptor had requested promiscuous mode.
277 	 * If so, turn it off.
278 	 */
279 	if (d->bd_promisc) {
280 		int error;
281 
282 		d->bd_promisc = 0;
283 		error = ifpromisc(bp->bif_ifp, 0);
284 		if (error && error != EINVAL)
285 			/*
286 			 * Something is really wrong if we were able to put
287 			 * the driver into promiscuous mode, but can't
288 			 * take it out.
289 			 */
290 			panic("bpf: ifpromisc failed");
291 	}
292 	/* Remove d from the interface's descriptor list. */
293 	p = &bp->bif_dlist;
294 	while (*p != d) {
295 		p = &(*p)->bd_next;
296 		if (*p == 0)
297 			panic("bpf_detachd: descriptor not in list");
298 	}
299 	*p = (*p)->bd_next;
300 	if (bp->bif_dlist == 0)
301 		/*
302 		 * Let the driver know that there are no more listeners.
303 		 */
304 		*d->bd_bif->bif_driverp = 0;
305 	d->bd_bif = 0;
306 }
307 
308 
309 /*
310  * Mark a descriptor free by making it point to itself.
311  * This is probably cheaper than marking with a constant since
312  * the address should be in a register anyway.
313  */
314 #define D_ISFREE(d) ((d) == (d)->bd_next)
315 #define D_MARKFREE(d) ((d)->bd_next = (d))
316 #define D_MARKUSED(d) ((d)->bd_next = 0)
317 
318 /*
319  * Open ethernet device.  Returns ENXIO for illegal minor device number,
320  * EBUSY if file is open by another process.
321  */
322 /* ARGSUSED */
323 int
324 bpfopen(dev, flag, mode, p)
325 	dev_t dev;
326 	int flag;
327 	int mode;
328 	struct proc *p;
329 {
330 	register struct bpf_d *d;
331 
332 	if (minor(dev) >= NBPFILTER)
333 		return (ENXIO);
334 	/*
335 	 * Each minor can be opened by only one process.  If the requested
336 	 * minor is in use, return EBUSY.
337 	 */
338 	d = &bpf_dtab[minor(dev)];
339 	if (!D_ISFREE(d))
340 		return (EBUSY);
341 
342 	/* Mark "free" and do most initialization. */
343 	bzero((char *)d, sizeof(*d));
344 	d->bd_bufsize = bpf_bufsize;
345 	d->bd_sig = SIGIO;
346 
347 	return (0);
348 }
349 
350 /*
351  * Close the descriptor by detaching it from its interface,
352  * deallocating its buffers, and marking it free.
353  */
354 /* ARGSUSED */
355 int
356 bpfclose(dev, flag, mode, p)
357 	dev_t dev;
358 	int flag;
359 	int mode;
360 	struct proc *p;
361 {
362 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
363 	register int s;
364 
365 	s = splimp();
366 	if (d->bd_bif)
367 		bpf_detachd(d);
368 	splx(s);
369 	bpf_freed(d);
370 
371 	return (0);
372 }
373 
374 /*
375  * Support for SunOS, which does not have tsleep.
376  */
377 #if BSD < 199103
378 static
379 bpf_timeout(arg)
380 	caddr_t arg;
381 {
382 	struct bpf_d *d = (struct bpf_d *)arg;
383 	d->bd_timedout = 1;
384 	wakeup(arg);
385 }
386 
387 #define BPF_SLEEP(chan, pri, s, t) bpf_sleep((struct bpf_d *)chan)
388 
389 int
390 bpf_sleep(d)
391 	register struct bpf_d *d;
392 {
393 	register int rto = d->bd_rtout;
394 	register int st;
395 
396 	if (rto != 0) {
397 		d->bd_timedout = 0;
398 		timeout(bpf_timeout, (caddr_t)d, rto);
399 	}
400 	st = sleep((caddr_t)d, PRINET|PCATCH);
401 	if (rto != 0) {
402 		if (d->bd_timedout == 0)
403 			untimeout(bpf_timeout, (caddr_t)d);
404 		else if (st == 0)
405 			return EWOULDBLOCK;
406 	}
407 	return (st != 0) ? EINTR : 0;
408 }
409 #else
410 #define BPF_SLEEP tsleep
411 #endif
412 
413 /*
414  * Rotate the packet buffers in descriptor d.  Move the store buffer
415  * into the hold slot, and the free buffer into the store slot.
416  * Zero the length of the new store buffer.
417  */
418 #define ROTATE_BUFFERS(d) \
419 	(d)->bd_hbuf = (d)->bd_sbuf; \
420 	(d)->bd_hlen = (d)->bd_slen; \
421 	(d)->bd_sbuf = (d)->bd_fbuf; \
422 	(d)->bd_slen = 0; \
423 	(d)->bd_fbuf = 0;
424 /*
425  *  bpfread - read next chunk of packets from buffers
426  */
427 int
428 bpfread(dev, uio, ioflag)
429 	dev_t dev;
430 	register struct uio *uio;
431 	int ioflag;
432 {
433 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
434 	int error;
435 	int s;
436 
437 	/*
438 	 * Restrict application to use a buffer the same size as
439 	 * as kernel buffers.
440 	 */
441 	if (uio->uio_resid != d->bd_bufsize)
442 		return (EINVAL);
443 
444 	s = splimp();
445 	/*
446 	 * If the hold buffer is empty, then do a timed sleep, which
447 	 * ends when the timeout expires or when enough packets
448 	 * have arrived to fill the store buffer.
449 	 */
450 	while (d->bd_hbuf == 0) {
451 		if (d->bd_immediate && d->bd_slen != 0) {
452 			/*
453 			 * A packet(s) either arrived since the previous
454 			 * read or arrived while we were asleep.
455 			 * Rotate the buffers and return what's here.
456 			 */
457 			ROTATE_BUFFERS(d);
458 			break;
459 		}
460 		if (d->bd_rtout != -1)
461 			error = BPF_SLEEP((caddr_t)d, PRINET|PCATCH, "bpf",
462 					  d->bd_rtout);
463 		else
464 			error = EWOULDBLOCK; /* User requested non-blocking I/O */
465 		if (error == EINTR || error == ERESTART) {
466 			splx(s);
467 			return (error);
468 		}
469 		if (error == EWOULDBLOCK) {
470 			/*
471 			 * On a timeout, return what's in the buffer,
472 			 * which may be nothing.  If there is something
473 			 * in the store buffer, we can rotate the buffers.
474 			 */
475 			if (d->bd_hbuf)
476 				/*
477 				 * We filled up the buffer in between
478 				 * getting the timeout and arriving
479 				 * here, so we don't need to rotate.
480 				 */
481 				break;
482 
483 			if (d->bd_slen == 0) {
484 				splx(s);
485 				return (0);
486 			}
487 			ROTATE_BUFFERS(d);
488 			break;
489 		}
490 	}
491 	/*
492 	 * At this point, we know we have something in the hold slot.
493 	 */
494 	splx(s);
495 
496 	/*
497 	 * Move data from hold buffer into user space.
498 	 * We know the entire buffer is transferred since
499 	 * we checked above that the read buffer is bpf_bufsize bytes.
500 	 */
501 	error = UIOMOVE(d->bd_hbuf, d->bd_hlen, UIO_READ, uio);
502 
503 	s = splimp();
504 	d->bd_fbuf = d->bd_hbuf;
505 	d->bd_hbuf = 0;
506 	d->bd_hlen = 0;
507 	splx(s);
508 
509 	return (error);
510 }
511 
512 
513 /*
514  * If there are processes sleeping on this descriptor, wake them up.
515  */
516 static __inline void
517 bpf_wakeup(d)
518 	register struct bpf_d *d;
519 {
520 	struct proc *p;
521 
522 	wakeup((caddr_t)d);
523 	if (d->bd_async && d->bd_sig)
524 		if (d->bd_pgid > 0)
525 			gsignal (d->bd_pgid, d->bd_sig);
526 		else if ((p = pfind (-d->bd_pgid)) != NULL)
527 			psignal (p, d->bd_sig);
528 
529 #if BSD >= 199103
530 	selwakeup(&d->bd_sel);
531 	/* XXX */
532 	d->bd_sel.si_pid = 0;
533 #else
534 	if (d->bd_selproc) {
535 		selwakeup(d->bd_selproc, (int)d->bd_selcoll);
536 		d->bd_selcoll = 0;
537 		d->bd_selproc = 0;
538 	}
539 #endif
540 }
541 
542 int
543 bpfwrite(dev, uio, ioflag)
544 	dev_t dev;
545 	struct uio *uio;
546 	int ioflag;
547 {
548 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
549 	struct ifnet *ifp;
550 	struct mbuf *m;
551 	int error, s;
552 	static struct sockaddr dst;
553 
554 	if (d->bd_bif == 0)
555 		return (ENXIO);
556 
557 	ifp = d->bd_bif->bif_ifp;
558 
559 	if (uio->uio_resid == 0)
560 		return (0);
561 
562 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, &dst);
563 	if (error)
564 		return (error);
565 
566 	if (m->m_pkthdr.len > ifp->if_mtu)
567 		return (EMSGSIZE);
568 
569 	s = splsoftnet();
570 #if BSD >= 199103
571 	error = (*ifp->if_output)(ifp, m, &dst, (struct rtentry *)0);
572 #else
573 	error = (*ifp->if_output)(ifp, m, &dst);
574 #endif
575 	splx(s);
576 	/*
577 	 * The driver frees the mbuf.
578 	 */
579 	return (error);
580 }
581 
582 /*
583  * Reset a descriptor by flushing its packet buffer and clearing the
584  * receive and drop counts.  Should be called at splimp.
585  */
586 static void
587 reset_d(d)
588 	struct bpf_d *d;
589 {
590 	if (d->bd_hbuf) {
591 		/* Free the hold buffer. */
592 		d->bd_fbuf = d->bd_hbuf;
593 		d->bd_hbuf = 0;
594 	}
595 	d->bd_slen = 0;
596 	d->bd_hlen = 0;
597 	d->bd_rcount = 0;
598 	d->bd_dcount = 0;
599 }
600 
601 /*
602  *  FIONREAD		Check for read packet available.
603  *  BIOCGBLEN		Get buffer len [for read()].
604  *  BIOCSETF		Set ethernet read filter.
605  *  BIOCFLUSH		Flush read packet buffer.
606  *  BIOCPROMISC		Put interface into promiscuous mode.
607  *  BIOCGDLT		Get link layer type.
608  *  BIOCGETIF		Get interface name.
609  *  BIOCSETIF		Set interface.
610  *  BIOCSRTIMEOUT	Set read timeout.
611  *  BIOCGRTIMEOUT	Get read timeout.
612  *  BIOCGSTATS		Get packet stats.
613  *  BIOCIMMEDIATE	Set immediate mode.
614  *  BIOCVERSION		Get filter language version.
615  */
616 /* ARGSUSED */
617 int
618 bpfioctl(dev, cmd, addr, flag, p)
619 	dev_t dev;
620 	u_long cmd;
621 	caddr_t addr;
622 	int flag;
623 	struct proc *p;
624 {
625 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
626 	int s, error = 0;
627 
628 	switch (cmd) {
629 
630 	default:
631 		error = EINVAL;
632 		break;
633 
634 	/*
635 	 * Check for read packet available.
636 	 */
637 	case FIONREAD:
638 		{
639 			int n;
640 
641 			s = splimp();
642 			n = d->bd_slen;
643 			if (d->bd_hbuf)
644 				n += d->bd_hlen;
645 			splx(s);
646 
647 			*(int *)addr = n;
648 			break;
649 		}
650 
651 	/*
652 	 * Get buffer len [for read()].
653 	 */
654 	case BIOCGBLEN:
655 		*(u_int *)addr = d->bd_bufsize;
656 		break;
657 
658 	/*
659 	 * Set buffer length.
660 	 */
661 	case BIOCSBLEN:
662 #if BSD < 199103
663 		error = EINVAL;
664 #else
665 		if (d->bd_bif != 0)
666 			error = EINVAL;
667 		else {
668 			register u_int size = *(u_int *)addr;
669 
670 			if (size > BPF_MAXBUFSIZE)
671 				*(u_int *)addr = size = BPF_MAXBUFSIZE;
672 			else if (size < BPF_MINBUFSIZE)
673 				*(u_int *)addr = size = BPF_MINBUFSIZE;
674 			d->bd_bufsize = size;
675 		}
676 #endif
677 		break;
678 
679 	/*
680 	 * Set link layer read filter.
681 	 */
682 	case BIOCSETF:
683 		error = bpf_setf(d, (struct bpf_program *)addr);
684 		break;
685 
686 	/*
687 	 * Flush read packet buffer.
688 	 */
689 	case BIOCFLUSH:
690 		s = splimp();
691 		reset_d(d);
692 		splx(s);
693 		break;
694 
695 	/*
696 	 * Put interface into promiscuous mode.
697 	 */
698 	case BIOCPROMISC:
699 		if (d->bd_bif == 0) {
700 			/*
701 			 * No interface attached yet.
702 			 */
703 			error = EINVAL;
704 			break;
705 		}
706 		s = splimp();
707 		if (d->bd_promisc == 0) {
708 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
709 			if (error == 0)
710 				d->bd_promisc = 1;
711 		}
712 		splx(s);
713 		break;
714 
715 	/*
716 	 * Get device parameters.
717 	 */
718 	case BIOCGDLT:
719 		if (d->bd_bif == 0)
720 			error = EINVAL;
721 		else
722 			*(u_int *)addr = d->bd_bif->bif_dlt;
723 		break;
724 
725 	/*
726 	 * Set interface name.
727 	 */
728 	case BIOCGETIF:
729 		if (d->bd_bif == 0)
730 			error = EINVAL;
731 		else
732 			bpf_ifname(d->bd_bif->bif_ifp, (struct ifreq *)addr);
733 		break;
734 
735 	/*
736 	 * Set interface.
737 	 */
738 	case BIOCSETIF:
739 		error = bpf_setif(d, (struct ifreq *)addr);
740 		break;
741 
742 	/*
743 	 * Set read timeout.
744 	 */
745 	case BIOCSRTIMEOUT:
746 		{
747 			struct timeval *tv = (struct timeval *)addr;
748 
749 			/* Compute number of ticks. */
750 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
751 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
752 				d->bd_rtout = 1;
753 			break;
754 		}
755 
756 	/*
757 	 * Get read timeout.
758 	 */
759 	case BIOCGRTIMEOUT:
760 		{
761 			struct timeval *tv = (struct timeval *)addr;
762 
763 			tv->tv_sec = d->bd_rtout / hz;
764 			tv->tv_usec = (d->bd_rtout % hz) * tick;
765 			break;
766 		}
767 
768 	/*
769 	 * Get packet stats.
770 	 */
771 	case BIOCGSTATS:
772 		{
773 			struct bpf_stat *bs = (struct bpf_stat *)addr;
774 
775 			bs->bs_recv = d->bd_rcount;
776 			bs->bs_drop = d->bd_dcount;
777 			break;
778 		}
779 
780 	/*
781 	 * Set immediate mode.
782 	 */
783 	case BIOCIMMEDIATE:
784 		d->bd_immediate = *(u_int *)addr;
785 		break;
786 
787 	case BIOCVERSION:
788 		{
789 			struct bpf_version *bv = (struct bpf_version *)addr;
790 
791 			bv->bv_major = BPF_MAJOR_VERSION;
792 			bv->bv_minor = BPF_MINOR_VERSION;
793 			break;
794 		}
795 
796 
797 	case FIONBIO:		/* Non-blocking I/O */
798 		if (*(int *)addr)
799 			d->bd_rtout = -1;
800 		else
801 			d->bd_rtout = 0;
802 		break;
803 
804 	case FIOASYNC:		/* Send signal on receive packets */
805 		d->bd_async = *(int *)addr;
806 		break;
807 
808 	/*
809 	 * N.B.  ioctl (FIOSETOWN) and fcntl (F_SETOWN) both end up doing
810 	 * the equivalent of a TIOCSPGRP and hence end up here.  *However*
811 	 * TIOCSPGRP's arg is a process group if it's positive and a process
812 	 * id if it's negative.  This is exactly the opposite of what the
813 	 * other two functions want!  Therefore there is code in ioctl and
814 	 * fcntl to negate the arg before calling here.
815 	 */
816 	case TIOCSPGRP:		/* Process or group to send signals to */
817 		d->bd_pgid = *(int *)addr;
818 		break;
819 
820 	case TIOCGPGRP:
821 		*(int *)addr = d->bd_pgid;
822 		break;
823 
824 	case BIOCSRSIG:		/* Set receive signal */
825 		{
826 		 	u_int sig;
827 
828 			sig = *(u_int *)addr;
829 
830 			if (sig >= NSIG)
831 				error = EINVAL;
832 			else
833 				d->bd_sig = sig;
834 			break;
835 		}
836 	case BIOCGRSIG:
837 		*(u_int *)addr = d->bd_sig;
838 		break;
839 	}
840 	return (error);
841 }
842 
843 /*
844  * Set d's packet filter program to fp.  If this file already has a filter,
845  * free it and replace it.  Returns EINVAL for bogus requests.
846  */
847 int
848 bpf_setf(d, fp)
849 	struct bpf_d *d;
850 	struct bpf_program *fp;
851 {
852 	struct bpf_insn *fcode, *old;
853 	u_int flen, size;
854 	int s;
855 
856 	old = d->bd_filter;
857 	if (fp->bf_insns == 0) {
858 		if (fp->bf_len != 0)
859 			return (EINVAL);
860 		s = splimp();
861 		d->bd_filter = 0;
862 		reset_d(d);
863 		splx(s);
864 		if (old != 0)
865 			free((caddr_t)old, M_DEVBUF);
866 		return (0);
867 	}
868 	flen = fp->bf_len;
869 	if (flen > BPF_MAXINSNS)
870 		return (EINVAL);
871 
872 	size = flen * sizeof(*fp->bf_insns);
873 	fcode = (struct bpf_insn *)malloc(size, M_DEVBUF, M_WAITOK);
874 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
875 	    bpf_validate(fcode, (int)flen)) {
876 		s = splimp();
877 		d->bd_filter = fcode;
878 		reset_d(d);
879 		splx(s);
880 		if (old != 0)
881 			free((caddr_t)old, M_DEVBUF);
882 
883 		return (0);
884 	}
885 	free((caddr_t)fcode, M_DEVBUF);
886 	return (EINVAL);
887 }
888 
889 /*
890  * Detach a file from its current interface (if attached at all) and attach
891  * to the interface indicated by the name stored in ifr.
892  * Return an errno or 0.
893  */
894 static int
895 bpf_setif(d, ifr)
896 	struct bpf_d *d;
897 	struct ifreq *ifr;
898 {
899 	struct bpf_if *bp;
900 	char *cp;
901 	int unit_seen, i, s, error;
902 
903 	/*
904 	 * Make sure the provided name has a unit number, and default
905 	 * it to '0' if not specified.
906 	 * XXX This is ugly ... do this differently?
907 	 */
908 	unit_seen = 0;
909 	cp = ifr->ifr_name;
910 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
911 	while (*cp++)
912 		if (*cp >= '0' && *cp <= '9')
913 			unit_seen = 1;
914 	if (!unit_seen) {
915 		/* Make sure to leave room for the '\0'. */
916 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
917 			if ((ifr->ifr_name[i] >= 'a' &&
918 			     ifr->ifr_name[i] <= 'z') ||
919 			    (ifr->ifr_name[i] >= 'A' &&
920 			     ifr->ifr_name[i] <= 'Z'))
921 				continue;
922 			ifr->ifr_name[i] = '0';
923 		}
924 	}
925 
926 	/*
927 	 * Look through attached interfaces for the named one.
928 	 */
929 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
930 		struct ifnet *ifp = bp->bif_ifp;
931 
932 		if (ifp == 0 ||
933 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
934 			continue;
935 		/*
936 		 * We found the requested interface.
937 		 * If it's not up, return an error.
938 		 * Allocate the packet buffers if we need to.
939 		 * If we're already attached to requested interface,
940 		 * just flush the buffer.
941 		 */
942 		if ((ifp->if_flags & IFF_UP) == 0)
943 			return (ENETDOWN);
944 
945 		if (d->bd_sbuf == 0) {
946 			error = bpf_allocbufs(d);
947 			if (error != 0)
948 				return (error);
949 		}
950 		s = splimp();
951 		if (bp != d->bd_bif) {
952 			if (d->bd_bif)
953 				/*
954 				 * Detach if attached to something else.
955 				 */
956 				bpf_detachd(d);
957 
958 			bpf_attachd(d, bp);
959 		}
960 		reset_d(d);
961 		splx(s);
962 		return (0);
963 	}
964 	/* Not found. */
965 	return (ENXIO);
966 }
967 
968 /*
969  * Copy the interface name to the ifreq.
970  */
971 static void
972 bpf_ifname(ifp, ifr)
973 	struct ifnet *ifp;
974 	struct ifreq *ifr;
975 {
976 
977 	bcopy(ifp->if_xname, ifr->ifr_name, IFNAMSIZ);
978 }
979 
980 /*
981  * Support for select() system call
982  *
983  * Return true iff the specific operation will not block indefinitely.
984  * Otherwise, return false but make a note that a selwakeup() must be done.
985  */
986 int
987 bpfpoll(dev, events, p)
988 	register dev_t dev;
989 	int events;
990 	struct proc *p;
991 {
992 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
993 	int revents = 0;
994 	register int s = splimp();
995 
996 	/*
997 	 * An imitation of the FIONREAD ioctl code.
998 	 */
999 	if (events & (POLLIN | POLLRDNORM))
1000 		if (d->bd_hlen != 0 || (d->bd_immediate && d->bd_slen != 0))
1001 			revents |= events & (POLLIN | POLLRDNORM);
1002 		else
1003 			selrecord(p, &d->bd_sel);
1004 
1005 	splx(s);
1006 	return (revents);
1007 }
1008 
1009 /*
1010  * Incoming linkage from device drivers.  Process the packet pkt, of length
1011  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1012  * by each process' filter, and if accepted, stashed into the corresponding
1013  * buffer.
1014  */
1015 void
1016 bpf_tap(arg, pkt, pktlen)
1017 	caddr_t arg;
1018 	register u_char *pkt;
1019 	register u_int pktlen;
1020 {
1021 	struct bpf_if *bp;
1022 	register struct bpf_d *d;
1023 	register size_t slen;
1024 	/*
1025 	 * Note that the ipl does not have to be raised at this point.
1026 	 * The only problem that could arise here is that if two different
1027 	 * interfaces shared any data.  This is not the case.
1028 	 */
1029 	bp = (struct bpf_if *)arg;
1030 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1031 		++d->bd_rcount;
1032 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
1033 		if (slen != 0)
1034 			catchpacket(d, pkt, pktlen, slen, bcopy);
1035 	}
1036 }
1037 
1038 /*
1039  * Copy data from an mbuf chain into a buffer.  This code is derived
1040  * from m_copydata in sys/uipc_mbuf.c.
1041  */
1042 static void
1043 bpf_mcopy(src_arg, dst_arg, len)
1044 	const void *src_arg;
1045 	void *dst_arg;
1046 	register size_t len;
1047 {
1048 	register const struct mbuf *m;
1049 	register u_int count;
1050 	u_char *dst;
1051 
1052 	m = src_arg;
1053 	dst = dst_arg;
1054 	while (len > 0) {
1055 		if (m == 0)
1056 			panic("bpf_mcopy");
1057 		count = min(m->m_len, len);
1058 		bcopy(mtod(m, caddr_t), (caddr_t)dst, count);
1059 		m = m->m_next;
1060 		dst += count;
1061 		len -= count;
1062 	}
1063 }
1064 
1065 /*
1066  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1067  */
1068 void
1069 bpf_mtap(arg, m)
1070 	caddr_t arg;
1071 	struct mbuf *m;
1072 {
1073 	struct bpf_if *bp = (struct bpf_if *)arg;
1074 	struct bpf_d *d;
1075 	size_t pktlen, slen;
1076 	struct mbuf *m0;
1077 
1078 	pktlen = 0;
1079 	for (m0 = m; m0 != 0; m0 = m0->m_next)
1080 		pktlen += m0->m_len;
1081 
1082 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1083 		++d->bd_rcount;
1084 		slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
1085 		if (slen != 0)
1086 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
1087 	}
1088 }
1089 
1090 /*
1091  * Move the packet data from interface memory (pkt) into the
1092  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1093  * otherwise 0.  "copy" is the routine called to do the actual data
1094  * transfer.  bcopy is passed in to copy contiguous chunks, while
1095  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1096  * pkt is really an mbuf.
1097  */
1098 static void
1099 catchpacket(d, pkt, pktlen, snaplen, cpfn)
1100 	register struct bpf_d *d;
1101 	register u_char *pkt;
1102 	register size_t pktlen, snaplen;
1103 	register void (*cpfn) __P((const void *, void *, size_t));
1104 {
1105 	register struct bpf_hdr *hp;
1106 	register int totlen, curlen;
1107 	register int hdrlen = d->bd_bif->bif_hdrlen;
1108 	/*
1109 	 * Figure out how many bytes to move.  If the packet is
1110 	 * greater or equal to the snapshot length, transfer that
1111 	 * much.  Otherwise, transfer the whole packet (unless
1112 	 * we hit the buffer size limit).
1113 	 */
1114 	totlen = hdrlen + min(snaplen, pktlen);
1115 	if (totlen > d->bd_bufsize)
1116 		totlen = d->bd_bufsize;
1117 
1118 	/*
1119 	 * Round up the end of the previous packet to the next longword.
1120 	 */
1121 	curlen = BPF_WORDALIGN(d->bd_slen);
1122 	if (curlen + totlen > d->bd_bufsize) {
1123 		/*
1124 		 * This packet will overflow the storage buffer.
1125 		 * Rotate the buffers if we can, then wakeup any
1126 		 * pending reads.
1127 		 */
1128 		if (d->bd_fbuf == 0) {
1129 			/*
1130 			 * We haven't completed the previous read yet,
1131 			 * so drop the packet.
1132 			 */
1133 			++d->bd_dcount;
1134 			return;
1135 		}
1136 		ROTATE_BUFFERS(d);
1137 		bpf_wakeup(d);
1138 		curlen = 0;
1139 	}
1140 	else if (d->bd_immediate)
1141 		/*
1142 		 * Immediate mode is set.  A packet arrived so any
1143 		 * reads should be woken up.
1144 		 */
1145 		bpf_wakeup(d);
1146 
1147 	/*
1148 	 * Append the bpf header.
1149 	 */
1150 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1151 #if BSD >= 199103
1152 	microtime(&hp->bh_tstamp);
1153 #elif defined(sun)
1154 	uniqtime(&hp->bh_tstamp);
1155 #else
1156 	hp->bh_tstamp = time;
1157 #endif
1158 	hp->bh_datalen = pktlen;
1159 	hp->bh_hdrlen = hdrlen;
1160 	/*
1161 	 * Copy the packet data into the store buffer and update its length.
1162 	 */
1163 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1164 	d->bd_slen = curlen + totlen;
1165 }
1166 
1167 /*
1168  * Initialize all nonzero fields of a descriptor.
1169  */
1170 static int
1171 bpf_allocbufs(d)
1172 	register struct bpf_d *d;
1173 {
1174 	d->bd_fbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
1175 	if (d->bd_fbuf == 0)
1176 		return (ENOBUFS);
1177 
1178 	d->bd_sbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
1179 	if (d->bd_sbuf == 0) {
1180 		free(d->bd_fbuf, M_DEVBUF);
1181 		return (ENOBUFS);
1182 	}
1183 	d->bd_slen = 0;
1184 	d->bd_hlen = 0;
1185 	return (0);
1186 }
1187 
1188 /*
1189  * Free buffers currently in use by a descriptor.
1190  * Called on close.
1191  */
1192 static void
1193 bpf_freed(d)
1194 	register struct bpf_d *d;
1195 {
1196 	/*
1197 	 * We don't need to lock out interrupts since this descriptor has
1198 	 * been detached from its interface and it yet hasn't been marked
1199 	 * free.
1200 	 */
1201 	if (d->bd_sbuf != 0) {
1202 		free(d->bd_sbuf, M_DEVBUF);
1203 		if (d->bd_hbuf != 0)
1204 			free(d->bd_hbuf, M_DEVBUF);
1205 		if (d->bd_fbuf != 0)
1206 			free(d->bd_fbuf, M_DEVBUF);
1207 	}
1208 	if (d->bd_filter)
1209 		free((caddr_t)d->bd_filter, M_DEVBUF);
1210 
1211 	D_MARKFREE(d);
1212 }
1213 
1214 /*
1215  * Attach an interface to bpf.  driverp is a pointer to a (struct bpf_if *)
1216  * in the driver's softc; dlt is the link layer type; hdrlen is the fixed
1217  * size of the link header (variable length headers not yet supported).
1218  */
1219 void
1220 bpfattach(driverp, ifp, dlt, hdrlen)
1221 	caddr_t *driverp;
1222 	struct ifnet *ifp;
1223 	u_int dlt, hdrlen;
1224 {
1225 	struct bpf_if *bp;
1226 	int i;
1227 #if BSD < 199103
1228 	static struct bpf_if bpf_ifs[NBPFILTER];
1229 	static int bpfifno;
1230 
1231 	bp = (bpfifno < NBPFILTER) ? &bpf_ifs[bpfifno++] : 0;
1232 #else
1233 	bp = (struct bpf_if *)malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1234 #endif
1235 	if (bp == 0)
1236 		panic("bpfattach");
1237 
1238 	bp->bif_dlist = 0;
1239 	bp->bif_driverp = (struct bpf_if **)driverp;
1240 	bp->bif_ifp = ifp;
1241 	bp->bif_dlt = dlt;
1242 
1243 	bp->bif_next = bpf_iflist;
1244 	bpf_iflist = bp;
1245 
1246 	*bp->bif_driverp = 0;
1247 
1248 	/*
1249 	 * Compute the length of the bpf header.  This is not necessarily
1250 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1251 	 * that the network layer header begins on a longword boundary (for
1252 	 * performance reasons and to alleviate alignment restrictions).
1253 	 */
1254 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1255 
1256 	/*
1257 	 * Mark all the descriptors free if this hasn't been done.
1258 	 */
1259 	if (!D_ISFREE(&bpf_dtab[0]))
1260 		for (i = 0; i < NBPFILTER; ++i)
1261 			D_MARKFREE(&bpf_dtab[i]);
1262 
1263 #if 0
1264 	printf("bpf: %s attached\n", ifp->if_xname);
1265 #endif
1266 }
1267 
1268 #if BSD >= 199103
1269 /* XXX This routine belongs in net/if.c. */
1270 /*
1271  * Set/clear promiscuous mode on interface ifp based on the truth value
1272  * of pswitch.  The calls are reference counted so that only the first
1273  * "on" request actually has an effect, as does the final "off" request.
1274  * Results are undefined if the "off" and "on" requests are not matched.
1275  */
1276 int
1277 ifpromisc(ifp, pswitch)
1278 	struct ifnet *ifp;
1279 	int pswitch;
1280 {
1281 	struct ifreq ifr;
1282 
1283 	if (pswitch) {
1284 		/*
1285 		 * If the device is not configured up, we cannot put it in
1286 		 * promiscuous mode.
1287 		 */
1288 		if ((ifp->if_flags & IFF_UP) == 0)
1289 			return (ENETDOWN);
1290 		if (ifp->if_pcount++ != 0)
1291 			return (0);
1292 		ifp->if_flags |= IFF_PROMISC;
1293 	} else {
1294 		if (--ifp->if_pcount > 0)
1295 			return (0);
1296 		ifp->if_flags &= ~IFF_PROMISC;
1297 		/*
1298 		 * If the device is not configured up, we should not need to
1299 		 * turn off promiscuous mode (device should have turned it
1300 		 * off when interface went down; and will look at IFF_PROMISC
1301 		 * again next time interface comes up).
1302 		 */
1303 		if ((ifp->if_flags & IFF_UP) == 0)
1304 			return (0);
1305 	}
1306 	ifr.ifr_flags = ifp->if_flags;
1307 	return ((*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr));
1308 }
1309 #endif
1310 
1311 #if BSD < 199103
1312 /*
1313  * Allocate some memory for bpf.  This is temporary SunOS support, and
1314  * is admittedly a hack.
1315  * If resources unavaiable, return 0.
1316  */
1317 static caddr_t
1318 bpf_alloc(size, canwait)
1319 	register int size;
1320 	register int canwait;
1321 {
1322 	register struct mbuf *m;
1323 
1324 	if ((unsigned)size > (MCLBYTES-8))
1325 		return 0;
1326 
1327 	MGET(m, canwait, MT_DATA);
1328 	if (m == 0)
1329 		return 0;
1330 	if ((unsigned)size > (MLEN-8)) {
1331 		MCLGET(m);
1332 		if (m->m_len != MCLBYTES) {
1333 			m_freem(m);
1334 			return 0;
1335 		}
1336 	}
1337 	*mtod(m, struct mbuf **) = m;
1338 	return mtod(m, caddr_t) + 8;
1339 }
1340 #endif
1341