xref: /openbsd-src/sys/net/bpf.c (revision 8500990981f885cbe5e6a4958549cacc238b5ae6)
1 /*	$OpenBSD: bpf.c,v 1.41 2003/10/24 04:26:16 canacar Exp $	*/
2 /*	$NetBSD: bpf.c,v 1.33 1997/02/21 23:59:35 thorpej Exp $	*/
3 
4 /*
5  * Copyright (c) 1990, 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from the Stanford/CMU enet packet filter,
9  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
10  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
11  * Berkeley Laboratory.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)bpf.c	8.2 (Berkeley) 3/28/94
38  */
39 
40 #include "bpfilter.h"
41 
42 #include <sys/param.h>
43 #include <sys/mbuf.h>
44 #include <sys/proc.h>
45 #include <sys/signalvar.h>
46 #include <sys/ioctl.h>
47 #include <sys/conf.h>
48 #include <sys/vnode.h>
49 #include <sys/file.h>
50 #include <sys/socket.h>
51 #include <sys/poll.h>
52 #include <sys/kernel.h>
53 
54 #include <net/if.h>
55 #include <net/bpf.h>
56 #include <net/bpfdesc.h>
57 
58 #include <netinet/in.h>
59 #include <netinet/if_arc.h>
60 #include <netinet/if_ether.h>
61 
62 #define BPF_BUFSIZE 9216	/* 8192 too small for ATM frames */
63 
64 #define PRINET  26			/* interruptible */
65 
66 /*
67  * The default read buffer size is patchable.
68  */
69 int bpf_bufsize = BPF_BUFSIZE;
70 
71 /*
72  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
73  *  bpf_dtab holds the descriptors, indexed by minor device #
74  */
75 struct bpf_if	*bpf_iflist;
76 struct bpf_d	*bpf_dtab;
77 int nbpfilter;
78 
79 int	bpf_allocbufs(struct bpf_d *);
80 void	bpf_freed(struct bpf_d *);
81 void	bpf_ifname(struct ifnet *, struct ifreq *);
82 void	bpf_mcopy(const void *, void *, size_t);
83 int	bpf_movein(struct uio *, int, struct mbuf **,
84 	    struct sockaddr *, struct bpf_insn *);
85 void	bpf_attachd(struct bpf_d *, struct bpf_if *);
86 void	bpf_detachd(struct bpf_d *);
87 int	bpf_setif(struct bpf_d *, struct ifreq *);
88 int	bpfpoll(dev_t, int, struct proc *);
89 int	bpfkqfilter(dev_t, struct knote *);
90 static __inline void bpf_wakeup(struct bpf_d *);
91 void	bpf_catchpacket(struct bpf_d *, u_char *, size_t, size_t,
92 	    void (*)(const void *, void *, size_t));
93 void	bpf_reset_d(struct bpf_d *);
94 
95 void	filt_bpfrdetach(struct knote *);
96 int	filt_bpfread(struct knote *, long);
97 
98 int
99 bpf_movein(uio, linktype, mp, sockp, filter)
100 	register struct uio *uio;
101 	int linktype;
102 	register struct mbuf **mp;
103 	register struct sockaddr *sockp;
104 	struct bpf_insn *filter;
105 {
106 	struct mbuf *m;
107 	int error;
108 	u_int hlen;
109 	u_int len;
110 	u_int slen;
111 
112 	/*
113 	 * Build a sockaddr based on the data link layer type.
114 	 * We do this at this level because the ethernet header
115 	 * is copied directly into the data field of the sockaddr.
116 	 * In the case of SLIP, there is no header and the packet
117 	 * is forwarded as is.
118 	 * Also, we are careful to leave room at the front of the mbuf
119 	 * for the link level header.
120 	 */
121 	switch (linktype) {
122 
123 	case DLT_SLIP:
124 		sockp->sa_family = AF_INET;
125 		hlen = 0;
126 		break;
127 
128 	case DLT_PPP:
129 		sockp->sa_family = AF_UNSPEC;
130 		hlen = 0;
131 		break;
132 
133 	case DLT_EN10MB:
134 		sockp->sa_family = AF_UNSPEC;
135 		/* XXX Would MAXLINKHDR be better? */
136 		hlen = sizeof(struct ether_header);
137 		break;
138 
139 	case DLT_ARCNET:
140 		sockp->sa_family = AF_UNSPEC;
141 		hlen = ARC_HDRLEN;
142 		break;
143 
144 	case DLT_FDDI:
145 		sockp->sa_family = AF_UNSPEC;
146 		/* XXX 4(FORMAC)+6(dst)+6(src)+3(LLC)+5(SNAP) */
147 		hlen = 24;
148 		break;
149 
150 	case DLT_RAW:
151 	case DLT_NULL:
152 		sockp->sa_family = AF_UNSPEC;
153 		hlen = 0;
154 		break;
155 
156 	case DLT_ATM_RFC1483:
157 		/*
158 		 * en atm driver requires 4-byte atm pseudo header.
159 		 * though it isn't standard, vpi:vci needs to be
160 		 * specified anyway.
161 		 */
162 		sockp->sa_family = AF_UNSPEC;
163 		hlen = 12; 	/* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
164 		break;
165 
166 	default:
167 		return (EIO);
168 	}
169 
170 	len = uio->uio_resid;
171 	if (len > MCLBYTES)
172 		return (EIO);
173 
174 	MGETHDR(m, M_WAIT, MT_DATA);
175 	m->m_pkthdr.rcvif = 0;
176 	m->m_pkthdr.len = len - hlen;
177 
178 	if (len > MHLEN) {
179 		MCLGET(m, M_WAIT);
180 		if ((m->m_flags & M_EXT) == 0) {
181 			error = ENOBUFS;
182 			goto bad;
183 		}
184 	}
185 	m->m_len = len;
186 	*mp = m;
187 
188 	error = uiomove(mtod(m, caddr_t), len, uio);
189 	if (error)
190 		goto bad;
191 
192 	slen = bpf_filter(filter, mtod(m, u_char *), len, len);
193 	if (slen < len) {
194 		error = EPERM;
195 		goto bad;
196 	}
197 
198 	if (m->m_len < hlen) {
199 		error = EPERM;
200 		goto bad;
201 	}
202 	/*
203 	 * Make room for link header, and copy it to sockaddr
204 	 */
205 	if (hlen != 0) {
206 		bcopy(m->m_data, sockp->sa_data, hlen);
207 		m->m_len -= hlen;
208 		m->m_data += hlen; /* XXX */
209 	}
210 
211 	return (0);
212  bad:
213 	m_freem(m);
214 	return (error);
215 }
216 
217 /*
218  * Attach file to the bpf interface, i.e. make d listen on bp.
219  * Must be called at splimp.
220  */
221 void
222 bpf_attachd(d, bp)
223 	struct bpf_d *d;
224 	struct bpf_if *bp;
225 {
226 	/*
227 	 * Point d at bp, and add d to the interface's list of listeners.
228 	 * Finally, point the driver's bpf cookie at the interface so
229 	 * it will divert packets to bpf.
230 	 */
231 	d->bd_bif = bp;
232 	d->bd_next = bp->bif_dlist;
233 	bp->bif_dlist = d;
234 
235 	*bp->bif_driverp = bp;
236 }
237 
238 /*
239  * Detach a file from its interface.
240  */
241 void
242 bpf_detachd(d)
243 	struct bpf_d *d;
244 {
245 	struct bpf_d **p;
246 	struct bpf_if *bp;
247 
248 	bp = d->bd_bif;
249 	/*
250 	 * Check if this descriptor had requested promiscuous mode.
251 	 * If so, turn it off.
252 	 */
253 	if (d->bd_promisc) {
254 		int error;
255 
256 		d->bd_promisc = 0;
257 		error = ifpromisc(bp->bif_ifp, 0);
258 		if (error && !(error == EINVAL || error == ENODEV))
259 			/*
260 			 * Something is really wrong if we were able to put
261 			 * the driver into promiscuous mode, but can't
262 			 * take it out.
263 			 */
264 			panic("bpf: ifpromisc failed");
265 	}
266 	/* Remove d from the interface's descriptor list. */
267 	p = &bp->bif_dlist;
268 	while (*p != d) {
269 		p = &(*p)->bd_next;
270 		if (*p == 0)
271 			panic("bpf_detachd: descriptor not in list");
272 	}
273 	*p = (*p)->bd_next;
274 	if (bp->bif_dlist == 0)
275 		/*
276 		 * Let the driver know that there are no more listeners.
277 		 */
278 		*d->bd_bif->bif_driverp = 0;
279 	d->bd_bif = 0;
280 }
281 
282 
283 /*
284  * Mark a descriptor free by making it point to itself.
285  * This is probably cheaper than marking with a constant since
286  * the address should be in a register anyway.
287  */
288 #define D_ISFREE(d) ((d) == (d)->bd_next)
289 #define D_MARKFREE(d) ((d)->bd_next = (d))
290 #define D_MARKUSED(d) ((d)->bd_next = 0)
291 
292 /*
293  * bpfilterattach() is called at boot time in new systems.  We do
294  * nothing here since old systems will not call this.
295  */
296 /* ARGSUSED */
297 void
298 bpfilterattach(n)
299 	int n;
300 {
301 	int i;
302 
303 	bpf_dtab = malloc(n * sizeof(*bpf_dtab), M_DEVBUF, M_NOWAIT);
304 	if (!bpf_dtab)
305 		return;
306 	nbpfilter = n;
307 	bzero(bpf_dtab, n * sizeof(*bpf_dtab));
308 	/*
309 	 * Mark all the descriptors free if this hasn't been done.
310 	 */
311 	if (!D_ISFREE(&bpf_dtab[0]))
312 		for (i = 0; i < nbpfilter; ++i)
313 			D_MARKFREE(&bpf_dtab[i]);
314 }
315 
316 /*
317  * Open ethernet device.  Returns ENXIO for illegal minor device number,
318  * EBUSY if file is open by another process.
319  */
320 /* ARGSUSED */
321 int
322 bpfopen(dev, flag, mode, p)
323 	dev_t dev;
324 	int flag;
325 	int mode;
326 	struct proc *p;
327 {
328 	register struct bpf_d *d;
329 
330 	if (minor(dev) >= nbpfilter)
331 		return (ENXIO);
332 	/*
333 	 * Each minor can be opened by only one process.  If the requested
334 	 * minor is in use, return EBUSY.
335 	 */
336 	d = &bpf_dtab[minor(dev)];
337 	if (!D_ISFREE(d))
338 		return (EBUSY);
339 
340 	/* Mark "free" and do most initialization. */
341 	bzero((char *)d, sizeof(*d));
342 	d->bd_bufsize = bpf_bufsize;
343 	d->bd_sig = SIGIO;
344 
345 	return (0);
346 }
347 
348 /*
349  * Close the descriptor by detaching it from its interface,
350  * deallocating its buffers, and marking it free.
351  */
352 /* ARGSUSED */
353 int
354 bpfclose(dev, flag, mode, p)
355 	dev_t dev;
356 	int flag;
357 	int mode;
358 	struct proc *p;
359 {
360 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
361 	register int s;
362 
363 	s = splimp();
364 	if (d->bd_bif)
365 		bpf_detachd(d);
366 	splx(s);
367 	bpf_freed(d);
368 
369 	return (0);
370 }
371 
372 /*
373  * Rotate the packet buffers in descriptor d.  Move the store buffer
374  * into the hold slot, and the free buffer into the store slot.
375  * Zero the length of the new store buffer.
376  */
377 #define ROTATE_BUFFERS(d) \
378 	(d)->bd_hbuf = (d)->bd_sbuf; \
379 	(d)->bd_hlen = (d)->bd_slen; \
380 	(d)->bd_sbuf = (d)->bd_fbuf; \
381 	(d)->bd_slen = 0; \
382 	(d)->bd_fbuf = 0;
383 /*
384  *  bpfread - read next chunk of packets from buffers
385  */
386 int
387 bpfread(dev, uio, ioflag)
388 	dev_t dev;
389 	register struct uio *uio;
390 	int ioflag;
391 {
392 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
393 	int error;
394 	int s;
395 
396 	if (d->bd_bif == 0)
397 		return (ENXIO);
398 
399 	/*
400 	 * Restrict application to use a buffer the same size as
401 	 * as kernel buffers.
402 	 */
403 	if (uio->uio_resid != d->bd_bufsize)
404 		return (EINVAL);
405 
406 	s = splimp();
407 
408 	/*
409 	 * bd_rdStart is tagged when we start the read, iff there's a timeout.
410 	 * we can then figure out when we're done reading.
411 	 */
412 	if (d->bd_rtout != -1 && d->bd_rdStart == 0)
413 		d->bd_rdStart = ticks;
414 	else
415 		d->bd_rdStart = 0;
416 
417 	/*
418 	 * If the hold buffer is empty, then do a timed sleep, which
419 	 * ends when the timeout expires or when enough packets
420 	 * have arrived to fill the store buffer.
421 	 */
422 	while (d->bd_hbuf == 0) {
423 		if (d->bd_immediate && d->bd_slen != 0) {
424 			/*
425 			 * A packet(s) either arrived since the previous
426 			 * read or arrived while we were asleep.
427 			 * Rotate the buffers and return what's here.
428 			 */
429 			ROTATE_BUFFERS(d);
430 			break;
431 		}
432 		if ((d->bd_rtout != -1) || (d->bd_rdStart + d->bd_rtout) < ticks) {
433 			error = tsleep((caddr_t)d, PRINET|PCATCH, "bpf",
434 			    d->bd_rtout);
435 		} else {
436 			if (d->bd_rtout == -1) {
437 				/* User requested non-blocking I/O */
438 				error = EWOULDBLOCK;
439 			} else
440 				error = 0;
441 		}
442 		if (error == EINTR || error == ERESTART) {
443 			splx(s);
444 			return (error);
445 		}
446 		if (error == EWOULDBLOCK) {
447 			/*
448 			 * On a timeout, return what's in the buffer,
449 			 * which may be nothing.  If there is something
450 			 * in the store buffer, we can rotate the buffers.
451 			 */
452 			if (d->bd_hbuf)
453 				/*
454 				 * We filled up the buffer in between
455 				 * getting the timeout and arriving
456 				 * here, so we don't need to rotate.
457 				 */
458 				break;
459 
460 			if (d->bd_slen == 0) {
461 				splx(s);
462 				return (0);
463 			}
464 			ROTATE_BUFFERS(d);
465 			break;
466 		}
467 	}
468 	/*
469 	 * At this point, we know we have something in the hold slot.
470 	 */
471 	splx(s);
472 
473 	/*
474 	 * Move data from hold buffer into user space.
475 	 * We know the entire buffer is transferred since
476 	 * we checked above that the read buffer is bpf_bufsize bytes.
477 	 */
478 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
479 
480 	s = splimp();
481 	d->bd_fbuf = d->bd_hbuf;
482 	d->bd_hbuf = 0;
483 	d->bd_hlen = 0;
484 	splx(s);
485 
486 	return (error);
487 }
488 
489 
490 /*
491  * If there are processes sleeping on this descriptor, wake them up.
492  */
493 static __inline void
494 bpf_wakeup(d)
495 	register struct bpf_d *d;
496 {
497 	wakeup((caddr_t)d);
498 	if (d->bd_async && d->bd_sig)
499 		csignal(d->bd_pgid, d->bd_sig,
500 		    d->bd_siguid, d->bd_sigeuid);
501 
502 	selwakeup(&d->bd_sel);
503 	/* XXX */
504 	d->bd_sel.si_selpid = 0;
505 	KNOTE(&d->bd_sel.si_note, 0);
506 }
507 
508 int
509 bpfwrite(dev, uio, ioflag)
510 	dev_t dev;
511 	struct uio *uio;
512 	int ioflag;
513 {
514 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
515 	struct ifnet *ifp;
516 	struct mbuf *m;
517 	int error, s;
518 	struct sockaddr_storage dst;
519 
520 	if (d->bd_bif == 0)
521 		return (ENXIO);
522 
523 	ifp = d->bd_bif->bif_ifp;
524 
525 	if (uio->uio_resid == 0)
526 		return (0);
527 
528 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m,
529 	    (struct sockaddr *)&dst, d->bd_wfilter);
530 	if (error)
531 		return (error);
532 
533 	if (m->m_pkthdr.len > ifp->if_mtu) {
534 		m_freem(m);
535 		return (EMSGSIZE);
536 	}
537 
538 	if (d->bd_hdrcmplt)
539 		dst.ss_family = pseudo_AF_HDRCMPLT;
540 
541 	s = splsoftnet();
542 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)&dst,
543 	    (struct rtentry *)0);
544 	splx(s);
545 	/*
546 	 * The driver frees the mbuf.
547 	 */
548 	return (error);
549 }
550 
551 /*
552  * Reset a descriptor by flushing its packet buffer and clearing the
553  * receive and drop counts.  Should be called at splimp.
554  */
555 void
556 bpf_reset_d(d)
557 	struct bpf_d *d;
558 {
559 	if (d->bd_hbuf) {
560 		/* Free the hold buffer. */
561 		d->bd_fbuf = d->bd_hbuf;
562 		d->bd_hbuf = 0;
563 	}
564 	d->bd_slen = 0;
565 	d->bd_hlen = 0;
566 	d->bd_rcount = 0;
567 	d->bd_dcount = 0;
568 }
569 
570 /*
571  *  FIONREAD		Check for read packet available.
572  *  BIOCGBLEN		Get buffer len [for read()].
573  *  BIOCSETF		Set ethernet read filter.
574  *  BIOCFLUSH		Flush read packet buffer.
575  *  BIOCPROMISC		Put interface into promiscuous mode.
576  *  BIOCGDLT		Get link layer type.
577  *  BIOCGETIF		Get interface name.
578  *  BIOCSETIF		Set interface.
579  *  BIOCSRTIMEOUT	Set read timeout.
580  *  BIOCGRTIMEOUT	Get read timeout.
581  *  BIOCGSTATS		Get packet stats.
582  *  BIOCIMMEDIATE	Set immediate mode.
583  *  BIOCVERSION		Get filter language version.
584  *  BIOCGHDRCMPLT	Get "header already complete" flag
585  *  BIOCSHDRCMPLT	Set "header already complete" flag
586  */
587 /* ARGSUSED */
588 int
589 bpfioctl(dev, cmd, addr, flag, p)
590 	dev_t dev;
591 	u_long cmd;
592 	caddr_t addr;
593 	int flag;
594 	struct proc *p;
595 {
596 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
597 	int s, error = 0;
598 
599 	if (d->bd_locked && suser(p, 0) != 0) {
600 		/* list of allowed ioctls when locked and not root */
601 		switch (cmd) {
602 		case BIOCGBLEN:
603 		case BIOCFLUSH:
604 		case BIOCGDLT:
605 		case BIOCGETIF:
606 		case BIOCGRTIMEOUT:
607 		case BIOCGSTATS:
608 		case BIOCVERSION:
609 		case BIOCGRSIG:
610 		case BIOCGHDRCMPLT:
611 		case FIONREAD:
612 		case BIOCLOCK:
613 		case BIOCSRTIMEOUT:
614 		case BIOCIMMEDIATE:
615 		case TIOCGPGRP:
616 			break;
617 		default:
618 			return (EPERM);
619 		}
620 	}
621 
622 	switch (cmd) {
623 
624 	default:
625 		error = EINVAL;
626 		break;
627 
628 	/*
629 	 * Check for read packet available.
630 	 */
631 	case FIONREAD:
632 		{
633 			int n;
634 
635 			s = splimp();
636 			n = d->bd_slen;
637 			if (d->bd_hbuf)
638 				n += d->bd_hlen;
639 			splx(s);
640 
641 			*(int *)addr = n;
642 			break;
643 		}
644 
645 	/*
646 	 * Get buffer len [for read()].
647 	 */
648 	case BIOCGBLEN:
649 		*(u_int *)addr = d->bd_bufsize;
650 		break;
651 
652 	/*
653 	 * Set buffer length.
654 	 */
655 	case BIOCSBLEN:
656 		if (d->bd_bif != 0)
657 			error = EINVAL;
658 		else {
659 			register u_int size = *(u_int *)addr;
660 
661 			if (size > BPF_MAXBUFSIZE)
662 				*(u_int *)addr = size = BPF_MAXBUFSIZE;
663 			else if (size < BPF_MINBUFSIZE)
664 				*(u_int *)addr = size = BPF_MINBUFSIZE;
665 			d->bd_bufsize = size;
666 		}
667 		break;
668 
669 	/*
670 	 * Set link layer read filter.
671 	 */
672 	case BIOCSETF:
673 		error = bpf_setf(d, (struct bpf_program *)addr, 0);
674 		break;
675 
676 	/*
677 	 * Set link layer write filter.
678 	 */
679 	case BIOCSETWF:
680 		error = bpf_setf(d, (struct bpf_program *)addr, 1);
681 		break;
682 
683 	/*
684 	 * Flush read packet buffer.
685 	 */
686 	case BIOCFLUSH:
687 		s = splimp();
688 		bpf_reset_d(d);
689 		splx(s);
690 		break;
691 
692 	/*
693 	 * Put interface into promiscuous mode.
694 	 */
695 	case BIOCPROMISC:
696 		if (d->bd_bif == 0) {
697 			/*
698 			 * No interface attached yet.
699 			 */
700 			error = EINVAL;
701 			break;
702 		}
703 		s = splimp();
704 		if (d->bd_promisc == 0) {
705 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
706 			if (error == 0)
707 				d->bd_promisc = 1;
708 		}
709 		splx(s);
710 		break;
711 
712 	/*
713 	 * Get device parameters.
714 	 */
715 	case BIOCGDLT:
716 		if (d->bd_bif == 0)
717 			error = EINVAL;
718 		else
719 			*(u_int *)addr = d->bd_bif->bif_dlt;
720 		break;
721 
722 	/*
723 	 * Set interface name.
724 	 */
725 	case BIOCGETIF:
726 		if (d->bd_bif == 0)
727 			error = EINVAL;
728 		else
729 			bpf_ifname(d->bd_bif->bif_ifp, (struct ifreq *)addr);
730 		break;
731 
732 	/*
733 	 * Set interface.
734 	 */
735 	case BIOCSETIF:
736 		error = bpf_setif(d, (struct ifreq *)addr);
737 		break;
738 
739 	/*
740 	 * Set read timeout.
741 	 */
742 	case BIOCSRTIMEOUT:
743 		{
744 			struct timeval *tv = (struct timeval *)addr;
745 
746 			/* Compute number of ticks. */
747 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
748 			if (d->bd_rtout == 0 && tv->tv_usec != 0)
749 				d->bd_rtout = 1;
750 			break;
751 		}
752 
753 	/*
754 	 * Get read timeout.
755 	 */
756 	case BIOCGRTIMEOUT:
757 		{
758 			struct timeval *tv = (struct timeval *)addr;
759 
760 			tv->tv_sec = d->bd_rtout / hz;
761 			tv->tv_usec = (d->bd_rtout % hz) * tick;
762 			break;
763 		}
764 
765 	/*
766 	 * Get packet stats.
767 	 */
768 	case BIOCGSTATS:
769 		{
770 			struct bpf_stat *bs = (struct bpf_stat *)addr;
771 
772 			bs->bs_recv = d->bd_rcount;
773 			bs->bs_drop = d->bd_dcount;
774 			break;
775 		}
776 
777 	/*
778 	 * Set immediate mode.
779 	 */
780 	case BIOCIMMEDIATE:
781 		d->bd_immediate = *(u_int *)addr;
782 		break;
783 
784 	case BIOCVERSION:
785 		{
786 			struct bpf_version *bv = (struct bpf_version *)addr;
787 
788 			bv->bv_major = BPF_MAJOR_VERSION;
789 			bv->bv_minor = BPF_MINOR_VERSION;
790 			break;
791 		}
792 
793 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
794 		*(u_int *)addr = d->bd_hdrcmplt;
795 		break;
796 
797 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
798 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
799 		break;
800 
801 	case BIOCLOCK:		/* set "locked" flag (no reset) */
802 		d->bd_locked = 1;
803 		break;
804 
805 	case FIONBIO:		/* Non-blocking I/O */
806 		if (*(int *)addr)
807 			d->bd_rtout = -1;
808 		else
809 			d->bd_rtout = 0;
810 		break;
811 
812 	case FIOASYNC:		/* Send signal on receive packets */
813 		d->bd_async = *(int *)addr;
814 		break;
815 
816 	/*
817 	 * N.B.  ioctl (FIOSETOWN) and fcntl (F_SETOWN) both end up doing
818 	 * the equivalent of a TIOCSPGRP and hence end up here.  *However*
819 	 * TIOCSPGRP's arg is a process group if it's positive and a process
820 	 * id if it's negative.  This is exactly the opposite of what the
821 	 * other two functions want!  Therefore there is code in ioctl and
822 	 * fcntl to negate the arg before calling here.
823 	 */
824 	case TIOCSPGRP:		/* Process or group to send signals to */
825 		d->bd_pgid = *(int *)addr;
826 		d->bd_siguid = p->p_cred->p_ruid;
827 		d->bd_sigeuid = p->p_ucred->cr_uid;
828 		break;
829 
830 	case TIOCGPGRP:
831 		*(int *)addr = d->bd_pgid;
832 		break;
833 
834 	case BIOCSRSIG:		/* Set receive signal */
835 		{
836 		 	u_int sig;
837 
838 			sig = *(u_int *)addr;
839 
840 			if (sig >= NSIG)
841 				error = EINVAL;
842 			else
843 				d->bd_sig = sig;
844 			break;
845 		}
846 	case BIOCGRSIG:
847 		*(u_int *)addr = d->bd_sig;
848 		break;
849 	}
850 	return (error);
851 }
852 
853 /*
854  * Set d's packet filter program to fp.  If this file already has a filter,
855  * free it and replace it.  Returns EINVAL for bogus requests.
856  */
857 int
858 bpf_setf(d, fp, wf)
859 	struct bpf_d *d;
860 	struct bpf_program *fp;
861 	int wf;
862 {
863 	struct bpf_insn *fcode, *old;
864 	u_int flen, size;
865 	int s;
866 
867 	old = wf ? d->bd_wfilter : d->bd_rfilter;
868 	if (fp->bf_insns == 0) {
869 		if (fp->bf_len != 0)
870 			return (EINVAL);
871 		s = splimp();
872 		if (wf)
873 			d->bd_wfilter = 0;
874 		else
875 			d->bd_rfilter = 0;
876 		bpf_reset_d(d);
877 		splx(s);
878 		if (old != 0)
879 			free((caddr_t)old, M_DEVBUF);
880 		return (0);
881 	}
882 	flen = fp->bf_len;
883 	if (flen > BPF_MAXINSNS)
884 		return (EINVAL);
885 
886 	size = flen * sizeof(*fp->bf_insns);
887 	fcode = (struct bpf_insn *)malloc(size, M_DEVBUF, M_WAITOK);
888 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
889 	    bpf_validate(fcode, (int)flen)) {
890 		s = splimp();
891 		if (wf)
892 			d->bd_wfilter = fcode;
893 		else
894 			d->bd_rfilter = fcode;
895 		bpf_reset_d(d);
896 		splx(s);
897 		if (old != 0)
898 			free((caddr_t)old, M_DEVBUF);
899 
900 		return (0);
901 	}
902 	free((caddr_t)fcode, M_DEVBUF);
903 	return (EINVAL);
904 }
905 
906 /*
907  * Detach a file from its current interface (if attached at all) and attach
908  * to the interface indicated by the name stored in ifr.
909  * Return an errno or 0.
910  */
911 int
912 bpf_setif(d, ifr)
913 	struct bpf_d *d;
914 	struct ifreq *ifr;
915 {
916 	struct bpf_if *bp;
917 	char *cp;
918 	int unit_seen, i, s, error;
919 
920 	/*
921 	 * Make sure the provided name has a unit number, and default
922 	 * it to '0' if not specified.
923 	 * XXX This is ugly ... do this differently?
924 	 */
925 	unit_seen = 0;
926 	cp = ifr->ifr_name;
927 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
928 	while (*cp++)
929 		if (*cp >= '0' && *cp <= '9')
930 			unit_seen = 1;
931 	if (!unit_seen) {
932 		/* Make sure to leave room for the '\0'. */
933 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
934 			if ((ifr->ifr_name[i] >= 'a' &&
935 			     ifr->ifr_name[i] <= 'z') ||
936 			    (ifr->ifr_name[i] >= 'A' &&
937 			     ifr->ifr_name[i] <= 'Z'))
938 				continue;
939 			ifr->ifr_name[i] = '0';
940 		}
941 	}
942 
943 	/*
944 	 * Look through attached interfaces for the named one.
945 	 */
946 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
947 		struct ifnet *ifp = bp->bif_ifp;
948 
949 		if (ifp == 0 ||
950 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
951 			continue;
952 		/*
953 		 * We found the requested interface.
954 		 * If it's not up, return an error.
955 		 * Allocate the packet buffers if we need to.
956 		 * If we're already attached to requested interface,
957 		 * just flush the buffer.
958 		 */
959 		if ((ifp->if_flags & IFF_UP) == 0)
960 			return (ENETDOWN);
961 
962 		if (d->bd_sbuf == 0) {
963 			error = bpf_allocbufs(d);
964 			if (error != 0)
965 				return (error);
966 		}
967 		s = splimp();
968 		if (bp != d->bd_bif) {
969 			if (d->bd_bif)
970 				/*
971 				 * Detach if attached to something else.
972 				 */
973 				bpf_detachd(d);
974 
975 			bpf_attachd(d, bp);
976 		}
977 		bpf_reset_d(d);
978 		splx(s);
979 		return (0);
980 	}
981 	/* Not found. */
982 	return (ENXIO);
983 }
984 
985 /*
986  * Copy the interface name to the ifreq.
987  */
988 void
989 bpf_ifname(ifp, ifr)
990 	struct ifnet *ifp;
991 	struct ifreq *ifr;
992 {
993 	bcopy(ifp->if_xname, ifr->ifr_name, IFNAMSIZ);
994 }
995 
996 /*
997  * Support for poll() system call
998  */
999 int
1000 bpfpoll(dev, events, p)
1001 	register dev_t dev;
1002 	int events;
1003 	struct proc *p;
1004 {
1005 	register struct bpf_d *d;
1006 	register int s, revents;
1007 
1008 	revents = events & (POLLIN | POLLRDNORM);
1009 	if (revents == 0)
1010 		return (0);		/* only support reading */
1011 
1012 	/*
1013 	 * An imitation of the FIONREAD ioctl code.
1014 	 */
1015 	d = &bpf_dtab[minor(dev)];
1016 	s = splimp();
1017 	if (d->bd_hlen == 0 && (!d->bd_immediate || d->bd_slen == 0)) {
1018 		revents = 0;		/* no data waiting */
1019 		/*
1020 		 * if there's a timeout, mark the time we started waiting.
1021 		 */
1022 		if (d->bd_rtout != -1 && d->bd_rdStart == 0)
1023 			d->bd_rdStart = ticks;
1024 		selrecord(p, &d->bd_sel);
1025 	}
1026 	splx(s);
1027 	return (revents);
1028 }
1029 
1030 struct filterops bpfread_filtops =
1031 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1032 
1033 int
1034 bpfkqfilter(dev_t dev,struct knote *kn)
1035 {
1036 	struct bpf_d *d = &bpf_dtab[minor(dev)];
1037 	struct klist *klist;
1038 	int s;
1039 
1040 	switch (kn->kn_filter) {
1041 	case EVFILT_READ:
1042 		klist = &d->bd_sel.si_note;
1043 		kn->kn_fop = &bpfread_filtops;
1044 		break;
1045 	case EVFILT_WRITE:
1046 	default:
1047 		return (1);
1048 	}
1049 
1050 	kn->kn_hook = (caddr_t)((u_long)dev);
1051 
1052 	s = splimp();
1053 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1054 	splx(s);
1055 
1056 	return (0);
1057 }
1058 
1059 void
1060 filt_bpfrdetach(struct knote *kn)
1061 {
1062 	dev_t dev = (dev_t)((u_long)kn->kn_hook);
1063 	struct bpf_d *d = &bpf_dtab[minor(dev)];
1064 	int s = splimp();
1065 
1066 	SLIST_REMOVE(&d->bd_sel.si_note, kn, knote, kn_selnext);
1067 	splx(s);
1068 }
1069 
1070 int
1071 filt_bpfread(struct knote *kn, long hint)
1072 {
1073 	dev_t dev = (dev_t)((u_long)kn->kn_hook);
1074 	struct bpf_d *d = &bpf_dtab[minor(dev)];
1075 	int res, s;
1076 
1077 	kn->kn_data = 0;
1078 
1079 	s = splimp();
1080 	res = d->bd_hlen != 0 || (d->bd_immediate && d->bd_slen != 0);
1081 	splx(s);
1082 	return (res);
1083 }
1084 
1085 /*
1086  * Incoming linkage from device drivers.  Process the packet pkt, of length
1087  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1088  * by each process' filter, and if accepted, stashed into the corresponding
1089  * buffer.
1090  */
1091 void
1092 bpf_tap(arg, pkt, pktlen)
1093 	caddr_t arg;
1094 	register u_char *pkt;
1095 	register u_int pktlen;
1096 {
1097 	struct bpf_if *bp;
1098 	register struct bpf_d *d;
1099 	register size_t slen;
1100 	/*
1101 	 * Note that the ipl does not have to be raised at this point.
1102 	 * The only problem that could arise here is that if two different
1103 	 * interfaces shared any data.  This is not the case.
1104 	 */
1105 	bp = (struct bpf_if *)arg;
1106 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1107 		++d->bd_rcount;
1108 		slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
1109 		if (slen != 0)
1110 			bpf_catchpacket(d, pkt, pktlen, slen, bcopy);
1111 	}
1112 }
1113 
1114 /*
1115  * Copy data from an mbuf chain into a buffer.  This code is derived
1116  * from m_copydata in sys/uipc_mbuf.c.
1117  */
1118 void
1119 bpf_mcopy(src_arg, dst_arg, len)
1120 	const void *src_arg;
1121 	void *dst_arg;
1122 	register size_t len;
1123 {
1124 	register const struct mbuf *m;
1125 	register u_int count;
1126 	u_char *dst;
1127 
1128 	m = src_arg;
1129 	dst = dst_arg;
1130 	while (len > 0) {
1131 		if (m == 0)
1132 			panic("bpf_mcopy");
1133 		count = min(m->m_len, len);
1134 		bcopy(mtod(m, caddr_t), (caddr_t)dst, count);
1135 		m = m->m_next;
1136 		dst += count;
1137 		len -= count;
1138 	}
1139 }
1140 
1141 /*
1142  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1143  */
1144 void
1145 bpf_mtap(arg, m)
1146 	caddr_t arg;
1147 	struct mbuf *m;
1148 {
1149 	struct bpf_if *bp = (struct bpf_if *)arg;
1150 	struct bpf_d *d;
1151 	size_t pktlen, slen;
1152 	struct mbuf *m0;
1153 
1154 	if (m == NULL)
1155 		return;
1156 
1157 	pktlen = 0;
1158 	for (m0 = m; m0 != 0; m0 = m0->m_next)
1159 		pktlen += m0->m_len;
1160 
1161 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1162 		++d->bd_rcount;
1163 		slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
1164 		if (slen != 0)
1165 			bpf_catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
1166 	}
1167 }
1168 
1169 /*
1170  * Move the packet data from interface memory (pkt) into the
1171  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1172  * otherwise 0.  "copy" is the routine called to do the actual data
1173  * transfer.  bcopy is passed in to copy contiguous chunks, while
1174  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
1175  * pkt is really an mbuf.
1176  */
1177 void
1178 bpf_catchpacket(d, pkt, pktlen, snaplen, cpfn)
1179 	register struct bpf_d *d;
1180 	register u_char *pkt;
1181 	register size_t pktlen, snaplen;
1182 	register void (*cpfn)(const void *, void *, size_t);
1183 {
1184 	register struct bpf_hdr *hp;
1185 	register int totlen, curlen;
1186 	register int hdrlen = d->bd_bif->bif_hdrlen;
1187 	struct timeval tv;
1188 
1189 	/*
1190 	 * Figure out how many bytes to move.  If the packet is
1191 	 * greater or equal to the snapshot length, transfer that
1192 	 * much.  Otherwise, transfer the whole packet (unless
1193 	 * we hit the buffer size limit).
1194 	 */
1195 	totlen = hdrlen + min(snaplen, pktlen);
1196 	if (totlen > d->bd_bufsize)
1197 		totlen = d->bd_bufsize;
1198 
1199 	/*
1200 	 * Round up the end of the previous packet to the next longword.
1201 	 */
1202 	curlen = BPF_WORDALIGN(d->bd_slen);
1203 	if (curlen + totlen > d->bd_bufsize) {
1204 		/*
1205 		 * This packet will overflow the storage buffer.
1206 		 * Rotate the buffers if we can, then wakeup any
1207 		 * pending reads.
1208 		 */
1209 		if (d->bd_fbuf == 0) {
1210 			/*
1211 			 * We haven't completed the previous read yet,
1212 			 * so drop the packet.
1213 			 */
1214 			++d->bd_dcount;
1215 			return;
1216 		}
1217 		ROTATE_BUFFERS(d);
1218 		bpf_wakeup(d);
1219 		curlen = 0;
1220 	}
1221 
1222 	/*
1223 	 * Append the bpf header.
1224 	 */
1225 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1226 	microtime(&tv);
1227 	hp->bh_tstamp.tv_sec = tv.tv_sec;
1228 	hp->bh_tstamp.tv_usec = tv.tv_usec;
1229 	hp->bh_datalen = pktlen;
1230 	hp->bh_hdrlen = hdrlen;
1231 	/*
1232 	 * Copy the packet data into the store buffer and update its length.
1233 	 */
1234 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1235 	d->bd_slen = curlen + totlen;
1236 
1237 	if (d->bd_immediate) {
1238 		/*
1239 		 * Immediate mode is set.  A packet arrived so any
1240 		 * reads should be woken up.
1241 		 */
1242 		bpf_wakeup(d);
1243 	}
1244 
1245 	if (d->bd_rdStart && (d->bd_rtout + d->bd_rdStart < ticks)) {
1246 		/*
1247 		 * we could be selecting on the bpf, and we
1248 		 * may have timeouts set.  We got here by getting
1249 		 * a packet, so wake up the reader.
1250 		 */
1251 		if (d->bd_fbuf) {
1252 			d->bd_rdStart = 0;
1253 			ROTATE_BUFFERS(d);
1254 			bpf_wakeup(d);
1255 			curlen = 0;
1256 		}
1257 	}
1258 }
1259 
1260 /*
1261  * Initialize all nonzero fields of a descriptor.
1262  */
1263 int
1264 bpf_allocbufs(d)
1265 	register struct bpf_d *d;
1266 {
1267 	d->bd_fbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1268 	if (d->bd_fbuf == NULL)
1269 		return (ENOBUFS);
1270 	d->bd_sbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1271 	if (d->bd_sbuf == NULL) {
1272 		free(d->bd_fbuf, M_DEVBUF);
1273 		return (ENOBUFS);
1274 	}
1275 	d->bd_slen = 0;
1276 	d->bd_hlen = 0;
1277 	return (0);
1278 }
1279 
1280 /*
1281  * Free buffers currently in use by a descriptor.
1282  * Called on close.
1283  */
1284 void
1285 bpf_freed(d)
1286 	register struct bpf_d *d;
1287 {
1288 	/*
1289 	 * We don't need to lock out interrupts since this descriptor has
1290 	 * been detached from its interface and it yet hasn't been marked
1291 	 * free.
1292 	 */
1293 	if (d->bd_sbuf != 0) {
1294 		free(d->bd_sbuf, M_DEVBUF);
1295 		if (d->bd_hbuf != 0)
1296 			free(d->bd_hbuf, M_DEVBUF);
1297 		if (d->bd_fbuf != 0)
1298 			free(d->bd_fbuf, M_DEVBUF);
1299 	}
1300 	if (d->bd_rfilter)
1301 		free((caddr_t)d->bd_rfilter, M_DEVBUF);
1302 	if (d->bd_wfilter)
1303 		free((caddr_t)d->bd_wfilter, M_DEVBUF);
1304 
1305 	D_MARKFREE(d);
1306 }
1307 
1308 /*
1309  * Attach an interface to bpf.  driverp is a pointer to a (struct bpf_if *)
1310  * in the driver's softc; dlt is the link layer type; hdrlen is the fixed
1311  * size of the link header (variable length headers not yet supported).
1312  */
1313 void
1314 bpfattach(driverp, ifp, dlt, hdrlen)
1315 	caddr_t *driverp;
1316 	struct ifnet *ifp;
1317 	u_int dlt, hdrlen;
1318 {
1319 	struct bpf_if *bp;
1320 	bp = (struct bpf_if *)malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1321 
1322 	if (bp == 0)
1323 		panic("bpfattach");
1324 
1325 	bp->bif_dlist = 0;
1326 	bp->bif_driverp = (struct bpf_if **)driverp;
1327 	bp->bif_ifp = ifp;
1328 	bp->bif_dlt = dlt;
1329 
1330 	bp->bif_next = bpf_iflist;
1331 	bpf_iflist = bp;
1332 
1333 	*bp->bif_driverp = 0;
1334 
1335 	/*
1336 	 * Compute the length of the bpf header.  This is not necessarily
1337 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1338 	 * that the network layer header begins on a longword boundary (for
1339 	 * performance reasons and to alleviate alignment restrictions).
1340 	 */
1341 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1342 }
1343 
1344 /* Detach an interface from its attached bpf device.  */
1345 void
1346 bpfdetach(ifp)
1347 	struct ifnet *ifp;
1348 {
1349 	struct bpf_if *bp, *nbp, **pbp = &bpf_iflist;
1350 	struct bpf_d *bd;
1351 	int maj, mn;
1352 
1353 	for (bp = bpf_iflist; bp; bp = nbp) {
1354 		nbp= bp->bif_next;
1355 		if (bp->bif_ifp == ifp) {
1356 			*pbp = nbp;
1357 
1358 			/* Locate the major number. */
1359 			for (maj = 0; maj < nchrdev; maj++)
1360 				if (cdevsw[maj].d_open == bpfopen)
1361 					break;
1362 
1363 			for (bd = bp->bif_dlist; bd; bd = bp->bif_dlist)
1364 				/*
1365 				 * Locate the minor number and nuke the vnode
1366 				 * for any open instance.
1367 				 */
1368 				for (mn = 0; mn < nbpfilter; mn++)
1369 					if (&bpf_dtab[mn] == bd) {
1370 						vdevgone(maj, mn, mn, VCHR);
1371 						break;
1372 					}
1373 
1374 			free(bp, M_DEVBUF);
1375 		} else
1376 			pbp = &bp->bif_next;
1377 	}
1378 	ifp->if_bpf = NULL;
1379 }
1380