xref: /freebsd/sys/net/bpf.c (revision 6e8394b8baa7d5d9153ab90de6824bcd19b3b4e1)
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  *      @(#)bpf.c	8.2 (Berkeley) 3/28/94
39  *
40  * $Id: bpf.c,v 1.50 1999/05/30 16:53:04 phk Exp $
41  */
42 
43 #include "bpfilter.h"
44 
45 #ifndef __GNUC__
46 #define inline
47 #else
48 #define inline __inline
49 #endif
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/conf.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/time.h>
57 #include <sys/proc.h>
58 #include <sys/signalvar.h>
59 #include <sys/filio.h>
60 #include <sys/sockio.h>
61 #include <sys/ttycom.h>
62 #include <sys/filedesc.h>
63 
64 #if defined(sparc) && BSD < 199103
65 #include <sys/stream.h>
66 #endif
67 #include <sys/poll.h>
68 
69 #include <sys/socket.h>
70 #include <sys/vnode.h>
71 
72 #include <net/if.h>
73 #include <net/bpf.h>
74 #include <net/bpfdesc.h>
75 
76 #include <netinet/in.h>
77 #include <netinet/if_ether.h>
78 #include <sys/kernel.h>
79 #include <sys/sysctl.h>
80 
81 #include "opt_devfs.h"
82 
83 #ifdef DEVFS
84 #include <sys/devfsext.h>
85 #endif /*DEVFS*/
86 
87 #if NBPFILTER > 0
88 
89 /*
90  * Older BSDs don't have kernel malloc.
91  */
92 #if BSD < 199103
93 extern bcopy();
94 static caddr_t bpf_alloc();
95 #include <net/bpf_compat.h>
96 #define BPF_BUFSIZE (MCLBYTES-8)
97 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, code, uio)
98 #else
99 #define BPF_BUFSIZE 4096
100 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, uio)
101 #endif
102 
103 #define PRINET  26			/* interruptible */
104 
105 /*
106  * The default read buffer size is patchable.
107  */
108 static int bpf_bufsize = BPF_BUFSIZE;
109 SYSCTL_INT(_debug, OID_AUTO, bpf_bufsize, CTLFLAG_RW,
110 	&bpf_bufsize, 0, "");
111 
112 /*
113  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
114  *  bpf_dtab holds the descriptors, indexed by minor device #
115  */
116 static struct bpf_if	*bpf_iflist;
117 static struct bpf_d	bpf_dtab[NBPFILTER];
118 static int		bpf_dtab_init;
119 
120 static int	bpf_allocbufs __P((struct bpf_d *));
121 static void	bpf_attachd __P((struct bpf_d *d, struct bpf_if *bp));
122 static void	bpf_detachd __P((struct bpf_d *d));
123 static void	bpf_freed __P((struct bpf_d *));
124 static void	bpf_ifname __P((struct ifnet *, struct ifreq *));
125 static void	bpf_mcopy __P((const void *, void *, size_t));
126 static int	bpf_movein __P((struct uio *, int,
127 		    struct mbuf **, struct sockaddr *, int *));
128 static int	bpf_setif __P((struct bpf_d *, struct ifreq *));
129 static inline void
130 		bpf_wakeup __P((struct bpf_d *));
131 static void	catchpacket __P((struct bpf_d *, u_char *, u_int,
132 		    u_int, void (*)(const void *, void *, size_t)));
133 static void	reset_d __P((struct bpf_d *));
134 static int	 bpf_setf __P((struct bpf_d *, struct bpf_program *));
135 
136 static	d_open_t	bpfopen;
137 static	d_close_t	bpfclose;
138 static	d_read_t	bpfread;
139 static	d_write_t	bpfwrite;
140 static	d_ioctl_t	bpfioctl;
141 static	d_poll_t	bpfpoll;
142 
143 #define CDEV_MAJOR 23
144 static struct cdevsw bpf_cdevsw = {
145 	/* open */	bpfopen,
146 	/* close */	bpfclose,
147 	/* read */	bpfread,
148 	/* write */	bpfwrite,
149 	/* ioctl */	bpfioctl,
150 	/* stop */	nostop,
151 	/* reset */	noreset,
152 	/* devtotty */	nodevtotty,
153 	/* poll */	bpfpoll,
154 	/* mmap */	nommap,
155 	/* strategy */	nostrategy,
156 	/* name */	"bpf",
157 	/* parms */	noparms,
158 	/* maj */	CDEV_MAJOR,
159 	/* dump */	nodump,
160 	/* psize */	nopsize,
161 	/* flags */	0,
162 	/* maxio */	0,
163 	/* bmaj */	-1
164 };
165 
166 
167 static int
168 bpf_movein(uio, linktype, mp, sockp, datlen)
169 	register struct uio *uio;
170 	int linktype, *datlen;
171 	register struct mbuf **mp;
172 	register struct sockaddr *sockp;
173 {
174 	struct mbuf *m;
175 	int error;
176 	int len;
177 	int hlen;
178 
179 	/*
180 	 * Build a sockaddr based on the data link layer type.
181 	 * We do this at this level because the ethernet header
182 	 * is copied directly into the data field of the sockaddr.
183 	 * In the case of SLIP, there is no header and the packet
184 	 * is forwarded as is.
185 	 * Also, we are careful to leave room at the front of the mbuf
186 	 * for the link level header.
187 	 */
188 	switch (linktype) {
189 
190 	case DLT_SLIP:
191 		sockp->sa_family = AF_INET;
192 		hlen = 0;
193 		break;
194 
195 	case DLT_EN10MB:
196 		sockp->sa_family = AF_UNSPEC;
197 		/* XXX Would MAXLINKHDR be better? */
198 		hlen = sizeof(struct ether_header);
199 		break;
200 
201 	case DLT_FDDI:
202 #if defined(__FreeBSD__) || defined(__bsdi__)
203 		sockp->sa_family = AF_IMPLINK;
204 		hlen = 0;
205 #else
206 		sockp->sa_family = AF_UNSPEC;
207 		/* XXX 4(FORMAC)+6(dst)+6(src)+3(LLC)+5(SNAP) */
208 		hlen = 24;
209 #endif
210 		break;
211 
212 	case DLT_RAW:
213 	case DLT_NULL:
214 		sockp->sa_family = AF_UNSPEC;
215 		hlen = 0;
216 		break;
217 
218 #ifdef __FreeBSD__
219 	case DLT_ATM_RFC1483:
220 		/*
221 		 * en atm driver requires 4-byte atm pseudo header.
222 		 * though it isn't standard, vpi:vci needs to be
223 		 * specified anyway.
224 		 */
225 		sockp->sa_family = AF_UNSPEC;
226 		hlen = 12; 	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
227 		break;
228 #endif
229 
230 	default:
231 		return (EIO);
232 	}
233 
234 	len = uio->uio_resid;
235 	*datlen = len - hlen;
236 	if ((unsigned)len > MCLBYTES)
237 		return (EIO);
238 
239 	MGETHDR(m, M_WAIT, MT_DATA);
240 	if (m == 0)
241 		return (ENOBUFS);
242 	if (len > MHLEN) {
243 #if BSD >= 199103
244 		MCLGET(m, M_WAIT);
245 		if ((m->m_flags & M_EXT) == 0) {
246 #else
247 		MCLGET(m);
248 		if (m->m_len != MCLBYTES) {
249 #endif
250 			error = ENOBUFS;
251 			goto bad;
252 		}
253 	}
254 	m->m_pkthdr.len = m->m_len = len;
255 	m->m_pkthdr.rcvif = NULL;
256 	*mp = m;
257 	/*
258 	 * Make room for link header.
259 	 */
260 	if (hlen != 0) {
261 		m->m_pkthdr.len -= hlen;
262 		m->m_len -= hlen;
263 #if BSD >= 199103
264 		m->m_data += hlen; /* XXX */
265 #else
266 		m->m_off += hlen;
267 #endif
268 		error = UIOMOVE((caddr_t)sockp->sa_data, hlen, UIO_WRITE, uio);
269 		if (error)
270 			goto bad;
271 	}
272 	error = UIOMOVE(mtod(m, caddr_t), len - hlen, UIO_WRITE, uio);
273 	if (!error)
274 		return (0);
275  bad:
276 	m_freem(m);
277 	return (error);
278 }
279 
280 /*
281  * Attach file to the bpf interface, i.e. make d listen on bp.
282  * Must be called at splimp.
283  */
284 static void
285 bpf_attachd(d, bp)
286 	struct bpf_d *d;
287 	struct bpf_if *bp;
288 {
289 	/*
290 	 * Point d at bp, and add d to the interface's list of listeners.
291 	 * Finally, point the driver's bpf cookie at the interface so
292 	 * it will divert packets to bpf.
293 	 */
294 	d->bd_bif = bp;
295 	d->bd_next = bp->bif_dlist;
296 	bp->bif_dlist = d;
297 
298 	bp->bif_ifp->if_bpf = bp;
299 }
300 
301 /*
302  * Detach a file from its interface.
303  */
304 static void
305 bpf_detachd(d)
306 	struct bpf_d *d;
307 {
308 	struct bpf_d **p;
309 	struct bpf_if *bp;
310 
311 	bp = d->bd_bif;
312 	/*
313 	 * Check if this descriptor had requested promiscuous mode.
314 	 * If so, turn it off.
315 	 */
316 	if (d->bd_promisc) {
317 		d->bd_promisc = 0;
318 		if (ifpromisc(bp->bif_ifp, 0))
319 			/*
320 			 * Something is really wrong if we were able to put
321 			 * the driver into promiscuous mode, but can't
322 			 * take it out.
323 			 */
324 			panic("bpf: ifpromisc failed");
325 	}
326 	/* Remove d from the interface's descriptor list. */
327 	p = &bp->bif_dlist;
328 	while (*p != d) {
329 		p = &(*p)->bd_next;
330 		if (*p == 0)
331 			panic("bpf_detachd: descriptor not in list");
332 	}
333 	*p = (*p)->bd_next;
334 	if (bp->bif_dlist == 0)
335 		/*
336 		 * Let the driver know that there are no more listeners.
337 		 */
338 		d->bd_bif->bif_ifp->if_bpf = 0;
339 	d->bd_bif = 0;
340 }
341 
342 
343 /*
344  * Mark a descriptor free by making it point to itself.
345  * This is probably cheaper than marking with a constant since
346  * the address should be in a register anyway.
347  */
348 #define D_ISFREE(d) ((d) == (d)->bd_next)
349 #define D_MARKFREE(d) ((d)->bd_next = (d))
350 #define D_MARKUSED(d) ((d)->bd_next = 0)
351 
352 /*
353  * Open ethernet device.  Returns ENXIO for illegal minor device number,
354  * EBUSY if file is open by another process.
355  */
356 /* ARGSUSED */
357 static	int
358 bpfopen(dev, flags, fmt, p)
359 	dev_t dev;
360 	int flags;
361 	int fmt;
362 	struct proc *p;
363 {
364 	register struct bpf_d *d;
365 
366 	if (p->p_prison)
367 		return (EPERM);
368 
369 	if (minor(dev) >= NBPFILTER)
370 		return (ENXIO);
371 	/*
372 	 * Each minor can be opened by only one process.  If the requested
373 	 * minor is in use, return EBUSY.
374 	 */
375 	d = &bpf_dtab[minor(dev)];
376 	if (!D_ISFREE(d))
377 		return (EBUSY);
378 
379 	/* Mark "free" and do most initialization. */
380 	bzero((char *)d, sizeof(*d));
381 	d->bd_bufsize = bpf_bufsize;
382 	d->bd_sig = SIGIO;
383 
384 	return (0);
385 }
386 
387 /*
388  * Close the descriptor by detaching it from its interface,
389  * deallocating its buffers, and marking it free.
390  */
391 /* ARGSUSED */
392 static	int
393 bpfclose(dev, flags, fmt, p)
394 	dev_t dev;
395 	int flags;
396 	int fmt;
397 	struct proc *p;
398 {
399 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
400 	register int s;
401 
402 	funsetown(d->bd_sigio);
403 	s = splimp();
404 	if (d->bd_bif)
405 		bpf_detachd(d);
406 	splx(s);
407 	bpf_freed(d);
408 
409 	return (0);
410 }
411 
412 /*
413  * Support for SunOS, which does not have tsleep.
414  */
415 #if BSD < 199103
416 static
417 bpf_timeout(arg)
418 	caddr_t arg;
419 {
420 	struct bpf_d *d = (struct bpf_d *)arg;
421 	d->bd_timedout = 1;
422 	wakeup(arg);
423 }
424 
425 #define BPF_SLEEP(chan, pri, s, t) bpf_sleep((struct bpf_d *)chan)
426 
427 int
428 bpf_sleep(d)
429 	register struct bpf_d *d;
430 {
431 	register int rto = d->bd_rtout;
432 	register int st;
433 
434 	if (rto != 0) {
435 		d->bd_timedout = 0;
436 		timeout(bpf_timeout, (caddr_t)d, rto);
437 	}
438 	st = sleep((caddr_t)d, PRINET|PCATCH);
439 	if (rto != 0) {
440 		if (d->bd_timedout == 0)
441 			untimeout(bpf_timeout, (caddr_t)d);
442 		else if (st == 0)
443 			return EWOULDBLOCK;
444 	}
445 	return (st != 0) ? EINTR : 0;
446 }
447 #else
448 #define BPF_SLEEP tsleep
449 #endif
450 
451 /*
452  * Rotate the packet buffers in descriptor d.  Move the store buffer
453  * into the hold slot, and the free buffer into the store slot.
454  * Zero the length of the new store buffer.
455  */
456 #define ROTATE_BUFFERS(d) \
457 	(d)->bd_hbuf = (d)->bd_sbuf; \
458 	(d)->bd_hlen = (d)->bd_slen; \
459 	(d)->bd_sbuf = (d)->bd_fbuf; \
460 	(d)->bd_slen = 0; \
461 	(d)->bd_fbuf = 0;
462 /*
463  *  bpfread - read next chunk of packets from buffers
464  */
465 static	int
466 bpfread(dev, uio, ioflag)
467 	dev_t dev;
468 	register struct uio *uio;
469 	int ioflag;
470 {
471 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
472 	int error;
473 	int s;
474 
475 	/*
476 	 * Restrict application to use a buffer the same size as
477 	 * as kernel buffers.
478 	 */
479 	if (uio->uio_resid != d->bd_bufsize)
480 		return (EINVAL);
481 
482 	s = splimp();
483 	/*
484 	 * If the hold buffer is empty, then do a timed sleep, which
485 	 * ends when the timeout expires or when enough packets
486 	 * have arrived to fill the store buffer.
487 	 */
488 	while (d->bd_hbuf == 0) {
489 		if (d->bd_immediate && d->bd_slen != 0) {
490 			/*
491 			 * A packet(s) either arrived since the previous
492 			 * read or arrived while we were asleep.
493 			 * Rotate the buffers and return what's here.
494 			 */
495 			ROTATE_BUFFERS(d);
496 			break;
497 		}
498 		if (ioflag & IO_NDELAY)
499 			error = EWOULDBLOCK;
500 		else
501 			error = BPF_SLEEP((caddr_t)d, PRINET|PCATCH, "bpf",
502 					  d->bd_rtout);
503 		if (error == EINTR || error == ERESTART) {
504 			splx(s);
505 			return (error);
506 		}
507 		if (error == EWOULDBLOCK) {
508 			/*
509 			 * On a timeout, return what's in the buffer,
510 			 * which may be nothing.  If there is something
511 			 * in the store buffer, we can rotate the buffers.
512 			 */
513 			if (d->bd_hbuf)
514 				/*
515 				 * We filled up the buffer in between
516 				 * getting the timeout and arriving
517 				 * here, so we don't need to rotate.
518 				 */
519 				break;
520 
521 			if (d->bd_slen == 0) {
522 				splx(s);
523 				return (0);
524 			}
525 			ROTATE_BUFFERS(d);
526 			break;
527 		}
528 	}
529 	/*
530 	 * At this point, we know we have something in the hold slot.
531 	 */
532 	splx(s);
533 
534 	/*
535 	 * Move data from hold buffer into user space.
536 	 * We know the entire buffer is transferred since
537 	 * we checked above that the read buffer is bpf_bufsize bytes.
538 	 */
539 	error = UIOMOVE(d->bd_hbuf, d->bd_hlen, UIO_READ, uio);
540 
541 	s = splimp();
542 	d->bd_fbuf = d->bd_hbuf;
543 	d->bd_hbuf = 0;
544 	d->bd_hlen = 0;
545 	splx(s);
546 
547 	return (error);
548 }
549 
550 
551 /*
552  * If there are processes sleeping on this descriptor, wake them up.
553  */
554 static inline void
555 bpf_wakeup(d)
556 	register struct bpf_d *d;
557 {
558 	wakeup((caddr_t)d);
559 	if (d->bd_async && d->bd_sig && d->bd_sigio)
560 		pgsigio(d->bd_sigio, d->bd_sig, 0);
561 
562 #if BSD >= 199103
563 	selwakeup(&d->bd_sel);
564 	/* XXX */
565 	d->bd_sel.si_pid = 0;
566 #else
567 	if (d->bd_selproc) {
568 		selwakeup(d->bd_selproc, (int)d->bd_selcoll);
569 		d->bd_selcoll = 0;
570 		d->bd_selproc = 0;
571 	}
572 #endif
573 }
574 
575 static	int
576 bpfwrite(dev, uio, ioflag)
577 	dev_t dev;
578 	struct uio *uio;
579 	int ioflag;
580 {
581 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
582 	struct ifnet *ifp;
583 	struct mbuf *m;
584 	int error, s;
585 	static struct sockaddr dst;
586 	int datlen;
587 
588 	if (d->bd_bif == 0)
589 		return (ENXIO);
590 
591 	ifp = d->bd_bif->bif_ifp;
592 
593 	if (uio->uio_resid == 0)
594 		return (0);
595 
596 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, &dst, &datlen);
597 	if (error)
598 		return (error);
599 
600 	if (datlen > ifp->if_mtu)
601 		return (EMSGSIZE);
602 
603 	s = splnet();
604 #if BSD >= 199103
605 	error = (*ifp->if_output)(ifp, m, &dst, (struct rtentry *)0);
606 #else
607 	error = (*ifp->if_output)(ifp, m, &dst);
608 #endif
609 	splx(s);
610 	/*
611 	 * The driver frees the mbuf.
612 	 */
613 	return (error);
614 }
615 
616 /*
617  * Reset a descriptor by flushing its packet buffer and clearing the
618  * receive and drop counts.  Should be called at splimp.
619  */
620 static void
621 reset_d(d)
622 	struct bpf_d *d;
623 {
624 	if (d->bd_hbuf) {
625 		/* Free the hold buffer. */
626 		d->bd_fbuf = d->bd_hbuf;
627 		d->bd_hbuf = 0;
628 	}
629 	d->bd_slen = 0;
630 	d->bd_hlen = 0;
631 	d->bd_rcount = 0;
632 	d->bd_dcount = 0;
633 }
634 
635 /*
636  *  FIONREAD		Check for read packet available.
637  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
638  *  BIOCGBLEN		Get buffer len [for read()].
639  *  BIOCSETF		Set ethernet read filter.
640  *  BIOCFLUSH		Flush read packet buffer.
641  *  BIOCPROMISC		Put interface into promiscuous mode.
642  *  BIOCGDLT		Get link layer type.
643  *  BIOCGETIF		Get interface name.
644  *  BIOCSETIF		Set interface.
645  *  BIOCSRTIMEOUT	Set read timeout.
646  *  BIOCGRTIMEOUT	Get read timeout.
647  *  BIOCGSTATS		Get packet stats.
648  *  BIOCIMMEDIATE	Set immediate mode.
649  *  BIOCVERSION		Get filter language version.
650  */
651 /* ARGSUSED */
652 static	int
653 bpfioctl(dev, cmd, addr, flags, p)
654 	dev_t dev;
655 	u_long cmd;
656 	caddr_t addr;
657 	int flags;
658 	struct proc *p;
659 {
660 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
661 	int s, error = 0;
662 
663 	switch (cmd) {
664 
665 	default:
666 		error = EINVAL;
667 		break;
668 
669 	/*
670 	 * Check for read packet available.
671 	 */
672 	case FIONREAD:
673 		{
674 			int n;
675 
676 			s = splimp();
677 			n = d->bd_slen;
678 			if (d->bd_hbuf)
679 				n += d->bd_hlen;
680 			splx(s);
681 
682 			*(int *)addr = n;
683 			break;
684 		}
685 
686 	case SIOCGIFADDR:
687 		{
688 			struct ifnet *ifp;
689 
690 			if (d->bd_bif == 0)
691 				error = EINVAL;
692 			else {
693 				ifp = d->bd_bif->bif_ifp;
694 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
695 			}
696 			break;
697 		}
698 
699 	/*
700 	 * Get buffer len [for read()].
701 	 */
702 	case BIOCGBLEN:
703 		*(u_int *)addr = d->bd_bufsize;
704 		break;
705 
706 	/*
707 	 * Set buffer length.
708 	 */
709 	case BIOCSBLEN:
710 #if BSD < 199103
711 		error = EINVAL;
712 #else
713 		if (d->bd_bif != 0)
714 			error = EINVAL;
715 		else {
716 			register u_int size = *(u_int *)addr;
717 
718 			if (size > BPF_MAXBUFSIZE)
719 				*(u_int *)addr = size = BPF_MAXBUFSIZE;
720 			else if (size < BPF_MINBUFSIZE)
721 				*(u_int *)addr = size = BPF_MINBUFSIZE;
722 			d->bd_bufsize = size;
723 		}
724 #endif
725 		break;
726 
727 	/*
728 	 * Set link layer read filter.
729 	 */
730 	case BIOCSETF:
731 		error = bpf_setf(d, (struct bpf_program *)addr);
732 		break;
733 
734 	/*
735 	 * Flush read packet buffer.
736 	 */
737 	case BIOCFLUSH:
738 		s = splimp();
739 		reset_d(d);
740 		splx(s);
741 		break;
742 
743 	/*
744 	 * Put interface into promiscuous mode.
745 	 */
746 	case BIOCPROMISC:
747 		if (d->bd_bif == 0) {
748 			/*
749 			 * No interface attached yet.
750 			 */
751 			error = EINVAL;
752 			break;
753 		}
754 		s = splimp();
755 		if (d->bd_promisc == 0) {
756 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
757 			if (error == 0)
758 				d->bd_promisc = 1;
759 		}
760 		splx(s);
761 		break;
762 
763 	/*
764 	 * Get device parameters.
765 	 */
766 	case BIOCGDLT:
767 		if (d->bd_bif == 0)
768 			error = EINVAL;
769 		else
770 			*(u_int *)addr = d->bd_bif->bif_dlt;
771 		break;
772 
773 	/*
774 	 * Set interface name.
775 	 */
776 	case BIOCGETIF:
777 		if (d->bd_bif == 0)
778 			error = EINVAL;
779 		else
780 			bpf_ifname(d->bd_bif->bif_ifp, (struct ifreq *)addr);
781 		break;
782 
783 	/*
784 	 * Set interface.
785 	 */
786 	case BIOCSETIF:
787 		error = bpf_setif(d, (struct ifreq *)addr);
788 		break;
789 
790 	/*
791 	 * Set read timeout.
792 	 */
793 	case BIOCSRTIMEOUT:
794 		{
795 			struct timeval *tv = (struct timeval *)addr;
796 
797 			/*
798 			 * Subtract 1 tick from tvtohz() since this isn't
799 			 * a one-shot timer.
800 			 */
801 			if ((error = itimerfix(tv)) == 0)
802 				d->bd_rtout = tvtohz(tv) - 1;
803 			break;
804 		}
805 
806 	/*
807 	 * Get read timeout.
808 	 */
809 	case BIOCGRTIMEOUT:
810 		{
811 			struct timeval *tv = (struct timeval *)addr;
812 
813 			tv->tv_sec = d->bd_rtout / hz;
814 			tv->tv_usec = (d->bd_rtout % hz) * tick;
815 			break;
816 		}
817 
818 	/*
819 	 * Get packet stats.
820 	 */
821 	case BIOCGSTATS:
822 		{
823 			struct bpf_stat *bs = (struct bpf_stat *)addr;
824 
825 			bs->bs_recv = d->bd_rcount;
826 			bs->bs_drop = d->bd_dcount;
827 			break;
828 		}
829 
830 	/*
831 	 * Set immediate mode.
832 	 */
833 	case BIOCIMMEDIATE:
834 		d->bd_immediate = *(u_int *)addr;
835 		break;
836 
837 	case BIOCVERSION:
838 		{
839 			struct bpf_version *bv = (struct bpf_version *)addr;
840 
841 			bv->bv_major = BPF_MAJOR_VERSION;
842 			bv->bv_minor = BPF_MINOR_VERSION;
843 			break;
844 		}
845 
846 	case FIONBIO:		/* Non-blocking I/O */
847 		break;
848 
849 	case FIOASYNC:		/* Send signal on receive packets */
850 		d->bd_async = *(int *)addr;
851 		break;
852 
853 	case FIOSETOWN:
854 		error = fsetown(*(int *)addr, &d->bd_sigio);
855 		break;
856 
857 	case FIOGETOWN:
858 		*(int *)addr = fgetown(d->bd_sigio);
859 		break;
860 
861 	/* This is deprecated, FIOSETOWN should be used instead. */
862 	case TIOCSPGRP:
863 		error = fsetown(-(*(int *)addr), &d->bd_sigio);
864 		break;
865 
866 	/* This is deprecated, FIOGETOWN should be used instead. */
867 	case TIOCGPGRP:
868 		*(int *)addr = -fgetown(d->bd_sigio);
869 		break;
870 
871 	case BIOCSRSIG:		/* Set receive signal */
872 		{
873 		 	u_int sig;
874 
875 			sig = *(u_int *)addr;
876 
877 			if (sig >= NSIG)
878 				error = EINVAL;
879 			else
880 				d->bd_sig = sig;
881 			break;
882 		}
883 	case BIOCGRSIG:
884 		*(u_int *)addr = d->bd_sig;
885 		break;
886 	}
887 	return (error);
888 }
889 
890 /*
891  * Set d's packet filter program to fp.  If this file already has a filter,
892  * free it and replace it.  Returns EINVAL for bogus requests.
893  */
894 static int
895 bpf_setf(d, fp)
896 	struct bpf_d *d;
897 	struct bpf_program *fp;
898 {
899 	struct bpf_insn *fcode, *old;
900 	u_int flen, size;
901 	int s;
902 
903 	old = d->bd_filter;
904 	if (fp->bf_insns == 0) {
905 		if (fp->bf_len != 0)
906 			return (EINVAL);
907 		s = splimp();
908 		d->bd_filter = 0;
909 		reset_d(d);
910 		splx(s);
911 		if (old != 0)
912 			free((caddr_t)old, M_DEVBUF);
913 		return (0);
914 	}
915 	flen = fp->bf_len;
916 	if (flen > BPF_MAXINSNS)
917 		return (EINVAL);
918 
919 	size = flen * sizeof(*fp->bf_insns);
920 	fcode = (struct bpf_insn *)malloc(size, M_DEVBUF, M_WAITOK);
921 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
922 	    bpf_validate(fcode, (int)flen)) {
923 		s = splimp();
924 		d->bd_filter = fcode;
925 		reset_d(d);
926 		splx(s);
927 		if (old != 0)
928 			free((caddr_t)old, M_DEVBUF);
929 
930 		return (0);
931 	}
932 	free((caddr_t)fcode, M_DEVBUF);
933 	return (EINVAL);
934 }
935 
936 /*
937  * Detach a file from its current interface (if attached at all) and attach
938  * to the interface indicated by the name stored in ifr.
939  * Return an errno or 0.
940  */
941 static int
942 bpf_setif(d, ifr)
943 	struct bpf_d *d;
944 	struct ifreq *ifr;
945 {
946 	struct bpf_if *bp;
947 	int s, error;
948 	struct ifnet *theywant;
949 
950 	theywant = ifunit(ifr->ifr_name);
951 	if (theywant == 0)
952 		return ENXIO;
953 
954 	/*
955 	 * Look through attached interfaces for the named one.
956 	 */
957 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
958 		struct ifnet *ifp = bp->bif_ifp;
959 
960 		if (ifp == 0 || ifp != theywant)
961 			continue;
962 		/*
963 		 * We found the requested interface.
964 		 * If it's not up, return an error.
965 		 * Allocate the packet buffers if we need to.
966 		 * If we're already attached to requested interface,
967 		 * just flush the buffer.
968 		 */
969 		if ((ifp->if_flags & IFF_UP) == 0)
970 			return (ENETDOWN);
971 
972 		if (d->bd_sbuf == 0) {
973 			error = bpf_allocbufs(d);
974 			if (error != 0)
975 				return (error);
976 		}
977 		s = splimp();
978 		if (bp != d->bd_bif) {
979 			if (d->bd_bif)
980 				/*
981 				 * Detach if attached to something else.
982 				 */
983 				bpf_detachd(d);
984 
985 			bpf_attachd(d, bp);
986 		}
987 		reset_d(d);
988 		splx(s);
989 		return (0);
990 	}
991 	/* Not found. */
992 	return (ENXIO);
993 }
994 
995 /*
996  * Convert an interface name plus unit number of an ifp to a single
997  * name which is returned in the ifr.
998  */
999 static void
1000 bpf_ifname(ifp, ifr)
1001 	struct ifnet *ifp;
1002 	struct ifreq *ifr;
1003 {
1004 	char *s = ifp->if_name;
1005 	char *d = ifr->ifr_name;
1006 
1007 	while ((*d++ = *s++) != 0)
1008 		continue;
1009 	d--; /* back to the null */
1010 	/* XXX Assume that unit number is less than 10. */
1011 	*d++ = ifp->if_unit + '0';
1012 	*d = '\0';
1013 }
1014 
1015 /*
1016  * Support for select() and poll() system calls
1017  *
1018  * Return true iff the specific operation will not block indefinitely.
1019  * Otherwise, return false but make a note that a selwakeup() must be done.
1020  */
1021 int
1022 bpfpoll(dev, events, p)
1023 	register dev_t dev;
1024 	int events;
1025 	struct proc *p;
1026 {
1027 	register struct bpf_d *d;
1028 	register int s;
1029 	int revents = 0;
1030 
1031 	/*
1032 	 * An imitation of the FIONREAD ioctl code.
1033 	 */
1034 	d = &bpf_dtab[minor(dev)];
1035 
1036 	s = splimp();
1037 	if (events & (POLLIN | POLLRDNORM)) {
1038 		if (d->bd_hlen != 0 || (d->bd_immediate && d->bd_slen != 0))
1039 			revents |= events & (POLLIN | POLLRDNORM);
1040 		else
1041 			selrecord(p, &d->bd_sel);
1042 	}
1043 	splx(s);
1044 	return (revents);
1045 }
1046 
1047 /*
1048  * Incoming linkage from device drivers.  Process the packet pkt, of length
1049  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1050  * by each process' filter, and if accepted, stashed into the corresponding
1051  * buffer.
1052  */
1053 void
1054 bpf_tap(ifp, pkt, pktlen)
1055 	struct ifnet *ifp;
1056 	register u_char *pkt;
1057 	register u_int pktlen;
1058 {
1059 	struct bpf_if *bp;
1060 	register struct bpf_d *d;
1061 	register u_int slen;
1062 	/*
1063 	 * Note that the ipl does not have to be raised at this point.
1064 	 * The only problem that could arise here is that if two different
1065 	 * interfaces shared any data.  This is not the case.
1066 	 */
1067 	bp = ifp->if_bpf;
1068 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1069 		++d->bd_rcount;
1070 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
1071 		if (slen != 0)
1072 			catchpacket(d, pkt, pktlen, slen, bcopy);
1073 	}
1074 }
1075 
1076 /*
1077  * Copy data from an mbuf chain into a buffer.  This code is derived
1078  * from m_copydata in sys/uipc_mbuf.c.
1079  */
1080 static void
1081 bpf_mcopy(src_arg, dst_arg, len)
1082 	const void *src_arg;
1083 	void *dst_arg;
1084 	register size_t len;
1085 {
1086 	register const struct mbuf *m;
1087 	register u_int count;
1088 	u_char *dst;
1089 
1090 	m = src_arg;
1091 	dst = dst_arg;
1092 	while (len > 0) {
1093 		if (m == 0)
1094 			panic("bpf_mcopy");
1095 		count = min(m->m_len, len);
1096 		bcopy(mtod(m, void *), dst, count);
1097 		m = m->m_next;
1098 		dst += count;
1099 		len -= count;
1100 	}
1101 }
1102 
1103 /*
1104  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1105  */
1106 void
1107 bpf_mtap(ifp, m)
1108 	struct ifnet *ifp;
1109 	struct mbuf *m;
1110 {
1111 	struct bpf_if *bp = ifp->if_bpf;
1112 	struct bpf_d *d;
1113 	u_int pktlen, slen;
1114 	struct mbuf *m0;
1115 
1116 	pktlen = 0;
1117 	for (m0 = m; m0 != 0; m0 = m0->m_next)
1118 		pktlen += m0->m_len;
1119 
1120 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1121 		++d->bd_rcount;
1122 		slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
1123 		if (slen != 0)
1124 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
1125 	}
1126 }
1127 
1128 /*
1129  * Move the packet data from interface memory (pkt) into the
1130  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1131  * otherwise 0.  "copy" is the routine called to do the actual data
1132  * transfer.  bcopy is passed in to copy contiguous chunks, while
1133  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1134  * pkt is really an mbuf.
1135  */
1136 static void
1137 catchpacket(d, pkt, pktlen, snaplen, cpfn)
1138 	register struct bpf_d *d;
1139 	register u_char *pkt;
1140 	register u_int pktlen, snaplen;
1141 	register void (*cpfn) __P((const void *, void *, size_t));
1142 {
1143 	register struct bpf_hdr *hp;
1144 	register int totlen, curlen;
1145 	register int hdrlen = d->bd_bif->bif_hdrlen;
1146 	/*
1147 	 * Figure out how many bytes to move.  If the packet is
1148 	 * greater or equal to the snapshot length, transfer that
1149 	 * much.  Otherwise, transfer the whole packet (unless
1150 	 * we hit the buffer size limit).
1151 	 */
1152 	totlen = hdrlen + min(snaplen, pktlen);
1153 	if (totlen > d->bd_bufsize)
1154 		totlen = d->bd_bufsize;
1155 
1156 	/*
1157 	 * Round up the end of the previous packet to the next longword.
1158 	 */
1159 	curlen = BPF_WORDALIGN(d->bd_slen);
1160 	if (curlen + totlen > d->bd_bufsize) {
1161 		/*
1162 		 * This packet will overflow the storage buffer.
1163 		 * Rotate the buffers if we can, then wakeup any
1164 		 * pending reads.
1165 		 */
1166 		if (d->bd_fbuf == 0) {
1167 			/*
1168 			 * We haven't completed the previous read yet,
1169 			 * so drop the packet.
1170 			 */
1171 			++d->bd_dcount;
1172 			return;
1173 		}
1174 		ROTATE_BUFFERS(d);
1175 		bpf_wakeup(d);
1176 		curlen = 0;
1177 	}
1178 	else if (d->bd_immediate)
1179 		/*
1180 		 * Immediate mode is set.  A packet arrived so any
1181 		 * reads should be woken up.
1182 		 */
1183 		bpf_wakeup(d);
1184 
1185 	/*
1186 	 * Append the bpf header.
1187 	 */
1188 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1189 #if BSD >= 199103
1190 	microtime(&hp->bh_tstamp);
1191 #elif defined(sun)
1192 	uniqtime(&hp->bh_tstamp);
1193 #else
1194 	hp->bh_tstamp = time;
1195 #endif
1196 	hp->bh_datalen = pktlen;
1197 	hp->bh_hdrlen = hdrlen;
1198 	/*
1199 	 * Copy the packet data into the store buffer and update its length.
1200 	 */
1201 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1202 	d->bd_slen = curlen + totlen;
1203 }
1204 
1205 /*
1206  * Initialize all nonzero fields of a descriptor.
1207  */
1208 static int
1209 bpf_allocbufs(d)
1210 	register struct bpf_d *d;
1211 {
1212 	d->bd_fbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
1213 	if (d->bd_fbuf == 0)
1214 		return (ENOBUFS);
1215 
1216 	d->bd_sbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
1217 	if (d->bd_sbuf == 0) {
1218 		free(d->bd_fbuf, M_DEVBUF);
1219 		return (ENOBUFS);
1220 	}
1221 	d->bd_slen = 0;
1222 	d->bd_hlen = 0;
1223 	return (0);
1224 }
1225 
1226 /*
1227  * Free buffers currently in use by a descriptor.
1228  * Called on close.
1229  */
1230 static void
1231 bpf_freed(d)
1232 	register struct bpf_d *d;
1233 {
1234 	/*
1235 	 * We don't need to lock out interrupts since this descriptor has
1236 	 * been detached from its interface and it yet hasn't been marked
1237 	 * free.
1238 	 */
1239 	if (d->bd_sbuf != 0) {
1240 		free(d->bd_sbuf, M_DEVBUF);
1241 		if (d->bd_hbuf != 0)
1242 			free(d->bd_hbuf, M_DEVBUF);
1243 		if (d->bd_fbuf != 0)
1244 			free(d->bd_fbuf, M_DEVBUF);
1245 	}
1246 	if (d->bd_filter)
1247 		free((caddr_t)d->bd_filter, M_DEVBUF);
1248 
1249 	D_MARKFREE(d);
1250 }
1251 
1252 /*
1253  * Attach an interface to bpf.  driverp is a pointer to a (struct bpf_if *)
1254  * in the driver's softc; dlt is the link layer type; hdrlen is the fixed
1255  * size of the link header (variable length headers not yet supported).
1256  */
1257 void
1258 bpfattach(ifp, dlt, hdrlen)
1259 	struct ifnet *ifp;
1260 	u_int dlt, hdrlen;
1261 {
1262 	struct bpf_if *bp;
1263 	int i;
1264 	bp = (struct bpf_if *)malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1265 	if (bp == 0)
1266 		panic("bpfattach");
1267 
1268 	bp->bif_dlist = 0;
1269 	bp->bif_ifp = ifp;
1270 	bp->bif_dlt = dlt;
1271 
1272 	bp->bif_next = bpf_iflist;
1273 	bpf_iflist = bp;
1274 
1275 	bp->bif_ifp->if_bpf = 0;
1276 
1277 	/*
1278 	 * Compute the length of the bpf header.  This is not necessarily
1279 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1280 	 * that the network layer header begins on a longword boundary (for
1281 	 * performance reasons and to alleviate alignment restrictions).
1282 	 */
1283 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1284 
1285 	/*
1286 	 * Mark all the descriptors free if this hasn't been done.
1287 	 */
1288 	if (!bpf_dtab_init) {
1289 		for (i = 0; i < NBPFILTER; ++i)
1290 			D_MARKFREE(&bpf_dtab[i]);
1291 		bpf_dtab_init = 1;
1292 	}
1293 
1294 	if (bootverbose)
1295 		printf("bpf: %s%d attached\n", ifp->if_name, ifp->if_unit);
1296 }
1297 
1298 #ifdef DEVFS
1299 static	void *bpf_devfs_token[NBPFILTER];
1300 #endif
1301 
1302 static	int bpf_devsw_installed;
1303 
1304 static void bpf_drvinit __P((void *unused));
1305 static void
1306 bpf_drvinit(unused)
1307 	void *unused;
1308 {
1309 #ifdef DEVFS
1310 	int i;
1311 #endif
1312 
1313 	if( ! bpf_devsw_installed ) {
1314 		cdevsw_add(&bpf_cdevsw);
1315 		bpf_devsw_installed = 1;
1316 #ifdef DEVFS
1317 
1318 		for ( i = 0 ; i < NBPFILTER ; i++ ) {
1319 			bpf_devfs_token[i] =
1320 				devfs_add_devswf(&bpf_cdevsw, i, DV_CHR, 0, 0,
1321 						 0600, "bpf%d", i);
1322 		}
1323 #endif
1324     	}
1325 }
1326 
1327 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvinit,NULL)
1328 
1329 #else /* !BPFILTER */
1330 /*
1331  * NOP stubs to allow bpf-using drivers to load and function.
1332  *
1333  * A 'better' implementation would allow the core bpf functionality
1334  * to be loaded at runtime.
1335  */
1336 
1337 void
1338 bpf_tap(ifp, pkt, pktlen)
1339 	struct ifnet *ifp;
1340 	register u_char *pkt;
1341 	register u_int pktlen;
1342 {
1343 }
1344 
1345 void
1346 bpf_mtap(ifp, m)
1347 	struct ifnet *ifp;
1348 	struct mbuf *m;
1349 {
1350 }
1351 
1352 void
1353 bpfattach(ifp, dlt, hdrlen)
1354 	struct ifnet *ifp;
1355 	u_int dlt, hdrlen;
1356 {
1357 }
1358 
1359 u_int
1360 bpf_filter(pc, p, wirelen, buflen)
1361 	register struct bpf_insn *pc;
1362 	register u_char *p;
1363 	u_int wirelen;
1364 	register u_int buflen;
1365 {
1366 	return -1;	/* "no filter" behaviour */
1367 }
1368 
1369 #endif
1370