xref: /netbsd-src/sys/net/bpf.c (revision fad4c9f71477ae11cea2ee75ec82151ac770a534)
1 /*	$NetBSD: bpf.c,v 1.118 2006/06/27 10:45:09 tron Exp $	*/
2 
3 /*
4  * Copyright (c) 1990, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from the Stanford/CMU enet packet filter,
8  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
9  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
10  * Berkeley Laboratory.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)bpf.c	8.4 (Berkeley) 1/9/95
37  * static char rcsid[] =
38  * "Header: bpf.c,v 1.67 96/09/26 22:00:52 leres Exp ";
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: bpf.c,v 1.118 2006/06/27 10:45:09 tron Exp $");
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/mbuf.h>
47 #include <sys/buf.h>
48 #include <sys/time.h>
49 #include <sys/proc.h>
50 #include <sys/user.h>
51 #include <sys/ioctl.h>
52 #include <sys/conf.h>
53 #include <sys/vnode.h>
54 #include <sys/queue.h>
55 
56 #include <sys/file.h>
57 #include <sys/filedesc.h>
58 #include <sys/tty.h>
59 #include <sys/uio.h>
60 
61 #include <sys/protosw.h>
62 #include <sys/socket.h>
63 #include <sys/errno.h>
64 #include <sys/kernel.h>
65 #include <sys/poll.h>
66 #include <sys/sysctl.h>
67 #include <sys/kauth.h>
68 
69 #include <net/if.h>
70 #include <net/slip.h>
71 
72 #include <net/bpf.h>
73 #include <net/bpfdesc.h>
74 
75 #include <net/if_arc.h>
76 #include <net/if_ether.h>
77 
78 #include <netinet/in.h>
79 #include <netinet/if_inarp.h>
80 
81 #if defined(_KERNEL_OPT)
82 #include "opt_bpf.h"
83 #include "sl.h"
84 #include "strip.h"
85 #endif
86 
87 #ifndef BPF_BUFSIZE
88 /*
89  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
90  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
91  */
92 # define BPF_BUFSIZE 32768
93 #endif
94 
95 #define PRINET  26			/* interruptible */
96 
97 /*
98  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
99  * XXX the default values should be computed dynamically based
100  * on available memory size and available mbuf clusters.
101  */
102 int bpf_bufsize = BPF_BUFSIZE;
103 int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
104 
105 
106 /*
107  * Global BPF statistics returned by net.bpf.stats sysctl.
108  */
109 struct bpf_stat	bpf_gstats;
110 
111 /*
112  * Use a mutex to avoid a race condition between gathering the stats/peers
113  * and opening/closing the device.
114  */
115 struct simplelock bpf_slock;
116 
117 /*
118  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
119  *  bpf_dtab holds the descriptors, indexed by minor device #
120  */
121 struct bpf_if	*bpf_iflist;
122 LIST_HEAD(, bpf_d) bpf_list;
123 
124 static int	bpf_allocbufs(struct bpf_d *);
125 static void	bpf_deliver(struct bpf_if *,
126 		            void *(*cpfn)(void *, const void *, size_t),
127 			    void *, u_int, u_int, struct ifnet *);
128 static void	bpf_freed(struct bpf_d *);
129 static void	bpf_ifname(struct ifnet *, struct ifreq *);
130 static void	*bpf_mcpy(void *, const void *, size_t);
131 static int	bpf_movein(struct uio *, int, int,
132 			        struct mbuf **, struct sockaddr *);
133 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
134 static void	bpf_detachd(struct bpf_d *);
135 static int	bpf_setif(struct bpf_d *, struct ifreq *);
136 static void	bpf_timed_out(void *);
137 static inline void
138 		bpf_wakeup(struct bpf_d *);
139 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
140 				 void *(*)(void *, const void *, size_t));
141 static void	reset_d(struct bpf_d *);
142 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
143 static int	bpf_setdlt(struct bpf_d *, u_int);
144 
145 static int	bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t,
146     int);
147 static int	bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t,
148     int);
149 static int	bpf_ioctl(struct file *, u_long, void *, struct lwp *);
150 static int	bpf_poll(struct file *, int, struct lwp *);
151 static int	bpf_close(struct file *, struct lwp *);
152 static int	bpf_kqfilter(struct file *, struct knote *);
153 
154 static const struct fileops bpf_fileops = {
155 	bpf_read,
156 	bpf_write,
157 	bpf_ioctl,
158 	fnullop_fcntl,
159 	bpf_poll,
160 	fbadop_stat,
161 	bpf_close,
162 	bpf_kqfilter,
163 };
164 
165 dev_type_open(bpfopen);
166 
167 const struct cdevsw bpf_cdevsw = {
168 	bpfopen, noclose, noread, nowrite, noioctl,
169 	nostop, notty, nopoll, nommap, nokqfilter,
170 };
171 
172 static int
173 bpf_movein(struct uio *uio, int linktype, int mtu, struct mbuf **mp,
174 	   struct sockaddr *sockp)
175 {
176 	struct mbuf *m;
177 	int error;
178 	int len;
179 	int hlen;
180 	int align;
181 
182 	/*
183 	 * Build a sockaddr based on the data link layer type.
184 	 * We do this at this level because the ethernet header
185 	 * is copied directly into the data field of the sockaddr.
186 	 * In the case of SLIP, there is no header and the packet
187 	 * is forwarded as is.
188 	 * Also, we are careful to leave room at the front of the mbuf
189 	 * for the link level header.
190 	 */
191 	switch (linktype) {
192 
193 	case DLT_SLIP:
194 		sockp->sa_family = AF_INET;
195 		hlen = 0;
196 		align = 0;
197 		break;
198 
199 	case DLT_PPP:
200 		sockp->sa_family = AF_UNSPEC;
201 		hlen = 0;
202 		align = 0;
203 		break;
204 
205 	case DLT_EN10MB:
206 		sockp->sa_family = AF_UNSPEC;
207 		/* XXX Would MAXLINKHDR be better? */
208  		/* 6(dst)+6(src)+2(type) */
209 		hlen = sizeof(struct ether_header);
210 		align = 2;
211 		break;
212 
213 	case DLT_ARCNET:
214 		sockp->sa_family = AF_UNSPEC;
215 		hlen = ARC_HDRLEN;
216 		align = 5;
217 		break;
218 
219 	case DLT_FDDI:
220 		sockp->sa_family = AF_LINK;
221 		/* XXX 4(FORMAC)+6(dst)+6(src) */
222 		hlen = 16;
223 		align = 0;
224 		break;
225 
226 	case DLT_ECONET:
227 		sockp->sa_family = AF_UNSPEC;
228 		hlen = 6;
229 		align = 2;
230 		break;
231 
232 	case DLT_NULL:
233 		sockp->sa_family = AF_UNSPEC;
234 		hlen = 0;
235 		align = 0;
236 		break;
237 
238 	default:
239 		return (EIO);
240 	}
241 
242 	len = uio->uio_resid;
243 	/*
244 	 * If there aren't enough bytes for a link level header or the
245 	 * packet length exceeds the interface mtu, return an error.
246 	 */
247 	if (len < hlen || len - hlen > mtu)
248 		return (EMSGSIZE);
249 
250 	/*
251 	 * XXX Avoid complicated buffer chaining ---
252 	 * bail if it won't fit in a single mbuf.
253 	 * (Take into account possible alignment bytes)
254 	 */
255 	if ((unsigned)len > MCLBYTES - align)
256 		return (EIO);
257 
258 	m = m_gethdr(M_WAIT, MT_DATA);
259 	m->m_pkthdr.rcvif = 0;
260 	m->m_pkthdr.len = len - hlen;
261 	if (len > MHLEN - align) {
262 		m_clget(m, M_WAIT);
263 		if ((m->m_flags & M_EXT) == 0) {
264 			error = ENOBUFS;
265 			goto bad;
266 		}
267 	}
268 
269 	/* Insure the data is properly aligned */
270 	if (align > 0) {
271 		m->m_data += align;
272 		m->m_len -= align;
273 	}
274 
275 	error = uiomove(mtod(m, void *), len, uio);
276 	if (error)
277 		goto bad;
278 	if (hlen != 0) {
279 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
280 		m->m_data += hlen; /* XXX */
281 		len -= hlen;
282 	}
283 	m->m_len = len;
284 	*mp = m;
285 	return (0);
286 
287 bad:
288 	m_freem(m);
289 	return (error);
290 }
291 
292 /*
293  * Attach file to the bpf interface, i.e. make d listen on bp.
294  * Must be called at splnet.
295  */
296 static void
297 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
298 {
299 	/*
300 	 * Point d at bp, and add d to the interface's list of listeners.
301 	 * Finally, point the driver's bpf cookie at the interface so
302 	 * it will divert packets to bpf.
303 	 */
304 	d->bd_bif = bp;
305 	d->bd_next = bp->bif_dlist;
306 	bp->bif_dlist = d;
307 
308 	*bp->bif_driverp = bp;
309 }
310 
311 /*
312  * Detach a file from its interface.
313  */
314 static void
315 bpf_detachd(struct bpf_d *d)
316 {
317 	struct bpf_d **p;
318 	struct bpf_if *bp;
319 
320 	bp = d->bd_bif;
321 	/*
322 	 * Check if this descriptor had requested promiscuous mode.
323 	 * If so, turn it off.
324 	 */
325 	if (d->bd_promisc) {
326 		int error;
327 
328 		d->bd_promisc = 0;
329 		/*
330 		 * Take device out of promiscuous mode.  Since we were
331 		 * able to enter promiscuous mode, we should be able
332 		 * to turn it off.  But we can get an error if
333 		 * the interface was configured down, so only panic
334 		 * if we don't get an unexpected error.
335 		 */
336   		error = ifpromisc(bp->bif_ifp, 0);
337 		if (error && error != EINVAL)
338 			panic("bpf: ifpromisc failed");
339 	}
340 	/* Remove d from the interface's descriptor list. */
341 	p = &bp->bif_dlist;
342 	while (*p != d) {
343 		p = &(*p)->bd_next;
344 		if (*p == 0)
345 			panic("bpf_detachd: descriptor not in list");
346 	}
347 	*p = (*p)->bd_next;
348 	if (bp->bif_dlist == 0)
349 		/*
350 		 * Let the driver know that there are no more listeners.
351 		 */
352 		*d->bd_bif->bif_driverp = 0;
353 	d->bd_bif = 0;
354 }
355 
356 
357 /*
358  * Mark a descriptor free by making it point to itself.
359  * This is probably cheaper than marking with a constant since
360  * the address should be in a register anyway.
361  */
362 
363 /*
364  * bpfilterattach() is called at boot time.
365  */
366 /* ARGSUSED */
367 void
368 bpfilterattach(int n)
369 {
370 	simple_lock_init(&bpf_slock);
371 
372 	simple_lock(&bpf_slock);
373 	LIST_INIT(&bpf_list);
374 	simple_unlock(&bpf_slock);
375 
376 	bpf_gstats.bs_recv = 0;
377 	bpf_gstats.bs_drop = 0;
378 	bpf_gstats.bs_capt = 0;
379 }
380 
381 /*
382  * Open ethernet device. Clones.
383  */
384 /* ARGSUSED */
385 int
386 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
387 {
388 	struct bpf_d *d;
389 	struct file *fp;
390 	int error, fd;
391 
392 	/* falloc() will use the descriptor for us. */
393 	if ((error = falloc(l->l_proc, &fp, &fd)) != 0)
394 		return error;
395 
396 	d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK);
397 	(void)memset(d, 0, sizeof(*d));
398 	d->bd_bufsize = bpf_bufsize;
399 	d->bd_seesent = 1;
400 	d->bd_pid = l->l_proc->p_pid;
401 	callout_init(&d->bd_callout);
402 
403 	simple_lock(&bpf_slock);
404 	LIST_INSERT_HEAD(&bpf_list, d, bd_list);
405 	simple_unlock(&bpf_slock);
406 
407 	return fdclone(l, fp, fd, flag, &bpf_fileops, d);
408 }
409 
410 /*
411  * Close the descriptor by detaching it from its interface,
412  * deallocating its buffers, and marking it free.
413  */
414 /* ARGSUSED */
415 static int
416 bpf_close(struct file *fp, struct lwp *l)
417 {
418 	struct bpf_d *d = fp->f_data;
419 	int s;
420 
421 	/*
422 	 * Refresh the PID associated with this bpf file.
423 	 */
424 	d->bd_pid = l->l_proc->p_pid;
425 
426 	s = splnet();
427 	if (d->bd_state == BPF_WAITING)
428 		callout_stop(&d->bd_callout);
429 	d->bd_state = BPF_IDLE;
430 	if (d->bd_bif)
431 		bpf_detachd(d);
432 	splx(s);
433 	bpf_freed(d);
434 	simple_lock(&bpf_slock);
435 	LIST_REMOVE(d, bd_list);
436 	simple_unlock(&bpf_slock);
437 	free(d, M_DEVBUF);
438 	fp->f_data = NULL;
439 
440 	return (0);
441 }
442 
443 /*
444  * Rotate the packet buffers in descriptor d.  Move the store buffer
445  * into the hold slot, and the free buffer into the store slot.
446  * Zero the length of the new store buffer.
447  */
448 #define ROTATE_BUFFERS(d) \
449 	(d)->bd_hbuf = (d)->bd_sbuf; \
450 	(d)->bd_hlen = (d)->bd_slen; \
451 	(d)->bd_sbuf = (d)->bd_fbuf; \
452 	(d)->bd_slen = 0; \
453 	(d)->bd_fbuf = 0;
454 /*
455  *  bpfread - read next chunk of packets from buffers
456  */
457 static int
458 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
459 	 kauth_cred_t cred, int flags)
460 {
461 	struct bpf_d *d = fp->f_data;
462 	int timed_out;
463 	int error;
464 	int s;
465 
466 	/*
467 	 * Restrict application to use a buffer the same size as
468 	 * as kernel buffers.
469 	 */
470 	if (uio->uio_resid != d->bd_bufsize)
471 		return (EINVAL);
472 
473 	s = splnet();
474 	if (d->bd_state == BPF_WAITING)
475 		callout_stop(&d->bd_callout);
476 	timed_out = (d->bd_state == BPF_TIMED_OUT);
477 	d->bd_state = BPF_IDLE;
478 	/*
479 	 * If the hold buffer is empty, then do a timed sleep, which
480 	 * ends when the timeout expires or when enough packets
481 	 * have arrived to fill the store buffer.
482 	 */
483 	while (d->bd_hbuf == 0) {
484 		if (fp->f_flag & FNONBLOCK) {
485 			if (d->bd_slen == 0) {
486 				splx(s);
487 				return (EWOULDBLOCK);
488 			}
489 			ROTATE_BUFFERS(d);
490 			break;
491 		}
492 
493 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
494 			/*
495 			 * A packet(s) either arrived since the previous
496 			 * read or arrived while we were asleep.
497 			 * Rotate the buffers and return what's here.
498 			 */
499 			ROTATE_BUFFERS(d);
500 			break;
501 		}
502 		error = tsleep(d, PRINET|PCATCH, "bpf",
503 				d->bd_rtout);
504 		if (error == EINTR || error == ERESTART) {
505 			splx(s);
506 			return (error);
507 		}
508 		if (error == EWOULDBLOCK) {
509 			/*
510 			 * On a timeout, return what's in the buffer,
511 			 * which may be nothing.  If there is something
512 			 * in the store buffer, we can rotate the buffers.
513 			 */
514 			if (d->bd_hbuf)
515 				/*
516 				 * We filled up the buffer in between
517 				 * getting the timeout and arriving
518 				 * here, so we don't need to rotate.
519 				 */
520 				break;
521 
522 			if (d->bd_slen == 0) {
523 				splx(s);
524 				return (0);
525 			}
526 			ROTATE_BUFFERS(d);
527 			break;
528 		}
529 		if (error != 0)
530 			goto done;
531 	}
532 	/*
533 	 * At this point, we know we have something in the hold slot.
534 	 */
535 	splx(s);
536 
537 	/*
538 	 * Move data from hold buffer into user space.
539 	 * We know the entire buffer is transferred since
540 	 * we checked above that the read buffer is bpf_bufsize bytes.
541 	 */
542 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
543 
544 	s = splnet();
545 	d->bd_fbuf = d->bd_hbuf;
546 	d->bd_hbuf = 0;
547 	d->bd_hlen = 0;
548 done:
549 	splx(s);
550 	return (error);
551 }
552 
553 
554 /*
555  * If there are processes sleeping on this descriptor, wake them up.
556  */
557 static inline void
558 bpf_wakeup(struct bpf_d *d)
559 {
560 	wakeup(d);
561 	if (d->bd_async)
562 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
563 
564 	selnotify(&d->bd_sel, 0);
565 	/* XXX */
566 	d->bd_sel.sel_pid = 0;
567 }
568 
569 
570 static void
571 bpf_timed_out(void *arg)
572 {
573 	struct bpf_d *d = arg;
574 	int s;
575 
576 	s = splnet();
577 	if (d->bd_state == BPF_WAITING) {
578 		d->bd_state = BPF_TIMED_OUT;
579 		if (d->bd_slen != 0)
580 			bpf_wakeup(d);
581 	}
582 	splx(s);
583 }
584 
585 
586 static int
587 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
588 	  kauth_cred_t cred, int flags)
589 {
590 	struct bpf_d *d = fp->f_data;
591 	struct ifnet *ifp;
592 	struct mbuf *m;
593 	int error, s;
594 	static struct sockaddr_storage dst;
595 
596 	m = NULL;	/* XXX gcc */
597 
598 	if (d->bd_bif == 0)
599 		return (ENXIO);
600 
601 	ifp = d->bd_bif->bif_ifp;
602 
603 	if (uio->uio_resid == 0)
604 		return (0);
605 
606 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
607 		(struct sockaddr *) &dst);
608 	if (error)
609 		return (error);
610 
611 	if (m->m_pkthdr.len > ifp->if_mtu) {
612 		m_freem(m);
613 		return (EMSGSIZE);
614 	}
615 
616 	if (d->bd_hdrcmplt)
617 		dst.ss_family = pseudo_AF_HDRCMPLT;
618 
619 	s = splsoftnet();
620 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *) &dst, NULL);
621 	splx(s);
622 	/*
623 	 * The driver frees the mbuf.
624 	 */
625 	return (error);
626 }
627 
628 /*
629  * Reset a descriptor by flushing its packet buffer and clearing the
630  * receive and drop counts.  Should be called at splnet.
631  */
632 static void
633 reset_d(struct bpf_d *d)
634 {
635 	if (d->bd_hbuf) {
636 		/* Free the hold buffer. */
637 		d->bd_fbuf = d->bd_hbuf;
638 		d->bd_hbuf = 0;
639 	}
640 	d->bd_slen = 0;
641 	d->bd_hlen = 0;
642 	d->bd_rcount = 0;
643 	d->bd_dcount = 0;
644 	d->bd_ccount = 0;
645 }
646 
647 /*
648  *  FIONREAD		Check for read packet available.
649  *  BIOCGBLEN		Get buffer len [for read()].
650  *  BIOCSETF		Set ethernet read filter.
651  *  BIOCFLUSH		Flush read packet buffer.
652  *  BIOCPROMISC		Put interface into promiscuous mode.
653  *  BIOCGDLT		Get link layer type.
654  *  BIOCGETIF		Get interface name.
655  *  BIOCSETIF		Set interface.
656  *  BIOCSRTIMEOUT	Set read timeout.
657  *  BIOCGRTIMEOUT	Get read timeout.
658  *  BIOCGSTATS		Get packet stats.
659  *  BIOCIMMEDIATE	Set immediate mode.
660  *  BIOCVERSION		Get filter language version.
661  *  BIOCGHDRCMPLT	Get "header already complete" flag.
662  *  BIOCSHDRCMPLT	Set "header already complete" flag.
663  */
664 /* ARGSUSED */
665 static int
666 bpf_ioctl(struct file *fp, u_long cmd, void *addr, struct lwp *l)
667 {
668 	struct bpf_d *d = fp->f_data;
669 	int s, error = 0;
670 
671 	/*
672 	 * Refresh the PID associated with this bpf file.
673 	 */
674 	d->bd_pid = l->l_proc->p_pid;
675 
676 	s = splnet();
677 	if (d->bd_state == BPF_WAITING)
678 		callout_stop(&d->bd_callout);
679 	d->bd_state = BPF_IDLE;
680 	splx(s);
681 
682 	switch (cmd) {
683 
684 	default:
685 		error = EINVAL;
686 		break;
687 
688 	/*
689 	 * Check for read packet available.
690 	 */
691 	case FIONREAD:
692 		{
693 			int n;
694 
695 			s = splnet();
696 			n = d->bd_slen;
697 			if (d->bd_hbuf)
698 				n += d->bd_hlen;
699 			splx(s);
700 
701 			*(int *)addr = n;
702 			break;
703 		}
704 
705 	/*
706 	 * Get buffer len [for read()].
707 	 */
708 	case BIOCGBLEN:
709 		*(u_int *)addr = d->bd_bufsize;
710 		break;
711 
712 	/*
713 	 * Set buffer length.
714 	 */
715 	case BIOCSBLEN:
716 		if (d->bd_bif != 0)
717 			error = EINVAL;
718 		else {
719 			u_int size = *(u_int *)addr;
720 
721 			if (size > bpf_maxbufsize)
722 				*(u_int *)addr = size = bpf_maxbufsize;
723 			else if (size < BPF_MINBUFSIZE)
724 				*(u_int *)addr = size = BPF_MINBUFSIZE;
725 			d->bd_bufsize = size;
726 		}
727 		break;
728 
729 	/*
730 	 * Set link layer read filter.
731 	 */
732 	case BIOCSETF:
733 		error = bpf_setf(d, addr);
734 		break;
735 
736 	/*
737 	 * Flush read packet buffer.
738 	 */
739 	case BIOCFLUSH:
740 		s = splnet();
741 		reset_d(d);
742 		splx(s);
743 		break;
744 
745 	/*
746 	 * Put interface into promiscuous mode.
747 	 */
748 	case BIOCPROMISC:
749 		if (d->bd_bif == 0) {
750 			/*
751 			 * No interface attached yet.
752 			 */
753 			error = EINVAL;
754 			break;
755 		}
756 		s = splnet();
757 		if (d->bd_promisc == 0) {
758 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
759 			if (error == 0)
760 				d->bd_promisc = 1;
761 		}
762 		splx(s);
763 		break;
764 
765 	/*
766 	 * Get device parameters.
767 	 */
768 	case BIOCGDLT:
769 		if (d->bd_bif == 0)
770 			error = EINVAL;
771 		else
772 			*(u_int *)addr = d->bd_bif->bif_dlt;
773 		break;
774 
775 	/*
776 	 * Get a list of supported device parameters.
777 	 */
778 	case BIOCGDLTLIST:
779 		if (d->bd_bif == 0)
780 			error = EINVAL;
781 		else
782 			error = bpf_getdltlist(d, addr);
783 		break;
784 
785 	/*
786 	 * Set device parameters.
787 	 */
788 	case BIOCSDLT:
789 		if (d->bd_bif == 0)
790 			error = EINVAL;
791 		else
792 			error = bpf_setdlt(d, *(u_int *)addr);
793 		break;
794 
795 	/*
796 	 * Set interface name.
797 	 */
798 	case BIOCGETIF:
799 		if (d->bd_bif == 0)
800 			error = EINVAL;
801 		else
802 			bpf_ifname(d->bd_bif->bif_ifp, addr);
803 		break;
804 
805 	/*
806 	 * Set interface.
807 	 */
808 	case BIOCSETIF:
809 		error = bpf_setif(d, addr);
810 		break;
811 
812 	/*
813 	 * Set read timeout.
814 	 */
815 	case BIOCSRTIMEOUT:
816 		{
817 			struct timeval *tv = addr;
818 
819 			/* Compute number of ticks. */
820 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
821 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
822 				d->bd_rtout = 1;
823 			break;
824 		}
825 
826 	/*
827 	 * Get read timeout.
828 	 */
829 	case BIOCGRTIMEOUT:
830 		{
831 			struct timeval *tv = addr;
832 
833 			tv->tv_sec = d->bd_rtout / hz;
834 			tv->tv_usec = (d->bd_rtout % hz) * tick;
835 			break;
836 		}
837 
838 	/*
839 	 * Get packet stats.
840 	 */
841 	case BIOCGSTATS:
842 		{
843 			struct bpf_stat *bs = addr;
844 
845 			bs->bs_recv = d->bd_rcount;
846 			bs->bs_drop = d->bd_dcount;
847 			bs->bs_capt = d->bd_ccount;
848 			break;
849 		}
850 
851 	case BIOCGSTATSOLD:
852 		{
853 			struct bpf_stat_old *bs = addr;
854 
855 			bs->bs_recv = d->bd_rcount;
856 			bs->bs_drop = d->bd_dcount;
857 			break;
858 		}
859 
860 	/*
861 	 * Set immediate mode.
862 	 */
863 	case BIOCIMMEDIATE:
864 		d->bd_immediate = *(u_int *)addr;
865 		break;
866 
867 	case BIOCVERSION:
868 		{
869 			struct bpf_version *bv = addr;
870 
871 			bv->bv_major = BPF_MAJOR_VERSION;
872 			bv->bv_minor = BPF_MINOR_VERSION;
873 			break;
874 		}
875 
876 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
877 		*(u_int *)addr = d->bd_hdrcmplt;
878 		break;
879 
880 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
881 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
882 		break;
883 
884 	/*
885 	 * Get "see sent packets" flag
886 	 */
887 	case BIOCGSEESENT:
888 		*(u_int *)addr = d->bd_seesent;
889 		break;
890 
891 	/*
892 	 * Set "see sent" packets flag
893 	 */
894 	case BIOCSSEESENT:
895 		d->bd_seesent = *(u_int *)addr;
896 		break;
897 
898 	case FIONBIO:		/* Non-blocking I/O */
899 		/*
900 		 * No need to do anything special as we use IO_NDELAY in
901 		 * bpfread() as an indication of whether or not to block
902 		 * the read.
903 		 */
904 		break;
905 
906 	case FIOASYNC:		/* Send signal on receive packets */
907 		d->bd_async = *(int *)addr;
908 		break;
909 
910 	case TIOCSPGRP:		/* Process or group to send signals to */
911 	case FIOSETOWN:
912 		error = fsetown(l->l_proc, &d->bd_pgid, cmd, addr);
913 		break;
914 
915 	case TIOCGPGRP:
916 	case FIOGETOWN:
917 		error = fgetown(l->l_proc, d->bd_pgid, cmd, addr);
918 		break;
919 	}
920 	return (error);
921 }
922 
923 /*
924  * Set d's packet filter program to fp.  If this file already has a filter,
925  * free it and replace it.  Returns EINVAL for bogus requests.
926  */
927 int
928 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
929 {
930 	struct bpf_insn *fcode, *old;
931 	u_int flen, size;
932 	int s;
933 
934 	old = d->bd_filter;
935 	if (fp->bf_insns == 0) {
936 		if (fp->bf_len != 0)
937 			return (EINVAL);
938 		s = splnet();
939 		d->bd_filter = 0;
940 		reset_d(d);
941 		splx(s);
942 		if (old != 0)
943 			free(old, M_DEVBUF);
944 		return (0);
945 	}
946 	flen = fp->bf_len;
947 	if (flen > BPF_MAXINSNS)
948 		return (EINVAL);
949 
950 	size = flen * sizeof(*fp->bf_insns);
951 	fcode = malloc(size, M_DEVBUF, M_WAITOK);
952 	if (copyin(fp->bf_insns, fcode, size) == 0 &&
953 	    bpf_validate(fcode, (int)flen)) {
954 		s = splnet();
955 		d->bd_filter = fcode;
956 		reset_d(d);
957 		splx(s);
958 		if (old != 0)
959 			free(old, M_DEVBUF);
960 
961 		return (0);
962 	}
963 	free(fcode, M_DEVBUF);
964 	return (EINVAL);
965 }
966 
967 /*
968  * Detach a file from its current interface (if attached at all) and attach
969  * to the interface indicated by the name stored in ifr.
970  * Return an errno or 0.
971  */
972 static int
973 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
974 {
975 	struct bpf_if *bp;
976 	char *cp;
977 	int unit_seen, i, s, error;
978 
979 	/*
980 	 * Make sure the provided name has a unit number, and default
981 	 * it to '0' if not specified.
982 	 * XXX This is ugly ... do this differently?
983 	 */
984 	unit_seen = 0;
985 	cp = ifr->ifr_name;
986 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
987 	while (*cp++)
988 		if (*cp >= '0' && *cp <= '9')
989 			unit_seen = 1;
990 	if (!unit_seen) {
991 		/* Make sure to leave room for the '\0'. */
992 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
993 			if ((ifr->ifr_name[i] >= 'a' &&
994 			     ifr->ifr_name[i] <= 'z') ||
995 			    (ifr->ifr_name[i] >= 'A' &&
996 			     ifr->ifr_name[i] <= 'Z'))
997 				continue;
998 			ifr->ifr_name[i] = '0';
999 		}
1000 	}
1001 
1002 	/*
1003 	 * Look through attached interfaces for the named one.
1004 	 */
1005 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
1006 		struct ifnet *ifp = bp->bif_ifp;
1007 
1008 		if (ifp == 0 ||
1009 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1010 			continue;
1011 		/* skip additional entry */
1012 		if ((caddr_t *)bp->bif_driverp != &ifp->if_bpf)
1013 			continue;
1014 		/*
1015 		 * We found the requested interface.
1016 		 * Allocate the packet buffers if we need to.
1017 		 * If we're already attached to requested interface,
1018 		 * just flush the buffer.
1019 		 */
1020 		if (d->bd_sbuf == 0) {
1021 			error = bpf_allocbufs(d);
1022 			if (error != 0)
1023 				return (error);
1024 		}
1025 		s = splnet();
1026 		if (bp != d->bd_bif) {
1027 			if (d->bd_bif)
1028 				/*
1029 				 * Detach if attached to something else.
1030 				 */
1031 				bpf_detachd(d);
1032 
1033 			bpf_attachd(d, bp);
1034 		}
1035 		reset_d(d);
1036 		splx(s);
1037 		return (0);
1038 	}
1039 	/* Not found. */
1040 	return (ENXIO);
1041 }
1042 
1043 /*
1044  * Copy the interface name to the ifreq.
1045  */
1046 static void
1047 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1048 {
1049 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1050 }
1051 
1052 /*
1053  * Support for poll() system call
1054  *
1055  * Return true iff the specific operation will not block indefinitely - with
1056  * the assumption that it is safe to positively acknowledge a request for the
1057  * ability to write to the BPF device.
1058  * Otherwise, return false but make a note that a selwakeup() must be done.
1059  */
1060 static int
1061 bpf_poll(struct file *fp, int events, struct lwp *l)
1062 {
1063 	struct bpf_d *d = fp->f_data;
1064 	int s = splnet();
1065 	int revents;
1066 
1067 	/*
1068 	 * Refresh the PID associated with this bpf file.
1069 	 */
1070 	d->bd_pid = l->l_proc->p_pid;
1071 
1072 	revents = events & (POLLOUT | POLLWRNORM);
1073 	if (events & (POLLIN | POLLRDNORM)) {
1074 		/*
1075 		 * An imitation of the FIONREAD ioctl code.
1076 		 */
1077 		if ((d->bd_hlen != 0) ||
1078 		    (d->bd_immediate && d->bd_slen != 0)) {
1079 			revents |= events & (POLLIN | POLLRDNORM);
1080 		} else if (d->bd_state == BPF_TIMED_OUT) {
1081 			if (d->bd_slen != 0)
1082 				revents |= events & (POLLIN | POLLRDNORM);
1083 			else
1084 				revents |= events & POLLIN;
1085 		} else {
1086 			selrecord(l, &d->bd_sel);
1087 			/* Start the read timeout if necessary */
1088 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1089 				callout_reset(&d->bd_callout, d->bd_rtout,
1090 					      bpf_timed_out, d);
1091 				d->bd_state = BPF_WAITING;
1092 			}
1093 		}
1094 	}
1095 
1096 	splx(s);
1097 	return (revents);
1098 }
1099 
1100 static void
1101 filt_bpfrdetach(struct knote *kn)
1102 {
1103 	struct bpf_d *d = kn->kn_hook;
1104 	int s;
1105 
1106 	s = splnet();
1107 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1108 	splx(s);
1109 }
1110 
1111 static int
1112 filt_bpfread(struct knote *kn, long hint)
1113 {
1114 	struct bpf_d *d = kn->kn_hook;
1115 
1116 	kn->kn_data = d->bd_hlen;
1117 	if (d->bd_immediate)
1118 		kn->kn_data += d->bd_slen;
1119 	return (kn->kn_data > 0);
1120 }
1121 
1122 static const struct filterops bpfread_filtops =
1123 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1124 
1125 static int
1126 bpf_kqfilter(struct file *fp, struct knote *kn)
1127 {
1128 	struct bpf_d *d = fp->f_data;
1129 	struct klist *klist;
1130 	int s;
1131 
1132 	switch (kn->kn_filter) {
1133 	case EVFILT_READ:
1134 		klist = &d->bd_sel.sel_klist;
1135 		kn->kn_fop = &bpfread_filtops;
1136 		break;
1137 
1138 	default:
1139 		return (1);
1140 	}
1141 
1142 	kn->kn_hook = d;
1143 
1144 	s = splnet();
1145 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1146 	splx(s);
1147 
1148 	return (0);
1149 }
1150 
1151 /*
1152  * Incoming linkage from device drivers.  Process the packet pkt, of length
1153  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1154  * by each process' filter, and if accepted, stashed into the corresponding
1155  * buffer.
1156  */
1157 void
1158 bpf_tap(void *arg, u_char *pkt, u_int pktlen)
1159 {
1160 	struct bpf_if *bp;
1161 	struct bpf_d *d;
1162 	u_int slen;
1163 	/*
1164 	 * Note that the ipl does not have to be raised at this point.
1165 	 * The only problem that could arise here is that if two different
1166 	 * interfaces shared any data.  This is not the case.
1167 	 */
1168 	bp = arg;
1169 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1170 		++d->bd_rcount;
1171 		++bpf_gstats.bs_recv;
1172 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
1173 		if (slen != 0)
1174 			catchpacket(d, pkt, pktlen, slen, memcpy);
1175 	}
1176 }
1177 
1178 /*
1179  * Copy data from an mbuf chain into a buffer.  This code is derived
1180  * from m_copydata in sys/uipc_mbuf.c.
1181  */
1182 static void *
1183 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1184 {
1185 	const struct mbuf *m;
1186 	u_int count;
1187 	u_char *dst;
1188 
1189 	m = src_arg;
1190 	dst = dst_arg;
1191 	while (len > 0) {
1192 		if (m == 0)
1193 			panic("bpf_mcpy");
1194 		count = min(m->m_len, len);
1195 		memcpy(dst, mtod(m, void *), count);
1196 		m = m->m_next;
1197 		dst += count;
1198 		len -= count;
1199 	}
1200 	return (dst_arg);
1201 }
1202 
1203 /*
1204  * Dispatch a packet to all the listeners on interface bp.
1205  *
1206  * marg    pointer to the packet, either a data buffer or an mbuf chain
1207  * buflen  buffer length, if marg is a data buffer
1208  * cpfn    a function that can copy marg into the listener's buffer
1209  * pktlen  length of the packet
1210  * rcvif   either NULL or the interface the packet came in on.
1211  */
1212 static inline void
1213 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1214 	    void *marg, u_int pktlen, u_int buflen, struct ifnet *rcvif)
1215 {
1216 	u_int slen;
1217 	struct bpf_d *d;
1218 
1219 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1220 		if (!d->bd_seesent && (rcvif == NULL))
1221 			continue;
1222 		++d->bd_rcount;
1223 		++bpf_gstats.bs_recv;
1224 		slen = bpf_filter(d->bd_filter, marg, pktlen, buflen);
1225 		if (slen != 0)
1226 			catchpacket(d, marg, pktlen, slen, cpfn);
1227 	}
1228 }
1229 
1230 /*
1231  * Incoming linkage from device drivers, when the head of the packet is in
1232  * a buffer, and the tail is in an mbuf chain.
1233  */
1234 void
1235 bpf_mtap2(void *arg, void *data, u_int dlen, struct mbuf *m)
1236 {
1237 	struct bpf_if *bp = arg;
1238 	u_int pktlen;
1239 	struct mbuf mb;
1240 
1241 	pktlen = m_length(m) + dlen;
1242 
1243 	/*
1244 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1245 	 * Note that we cut corners here; we only setup what's
1246 	 * absolutely needed--this mbuf should never go anywhere else.
1247 	 */
1248 	(void)memset(&mb, 0, sizeof(mb));
1249 	mb.m_next = m;
1250 	mb.m_data = data;
1251 	mb.m_len = dlen;
1252 
1253 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif);
1254 }
1255 
1256 /*
1257  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1258  */
1259 void
1260 bpf_mtap(void *arg, struct mbuf *m)
1261 {
1262 	void *(*cpfn)(void *, const void *, size_t);
1263 	struct bpf_if *bp = arg;
1264 	u_int pktlen, buflen;
1265 	void *marg;
1266 
1267 	pktlen = m_length(m);
1268 
1269 	if (pktlen == m->m_len) {
1270 		cpfn = memcpy;
1271 		marg = mtod(m, void *);
1272 		buflen = pktlen;
1273 	} else {
1274 		cpfn = bpf_mcpy;
1275 		marg = m;
1276 		buflen = 0;
1277 	}
1278 
1279 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif);
1280 }
1281 
1282 /*
1283  * We need to prepend the address family as
1284  * a four byte field.  Cons up a dummy header
1285  * to pacify bpf.  This is safe because bpf
1286  * will only read from the mbuf (i.e., it won't
1287  * try to free it or keep a pointer a to it).
1288  */
1289 void
1290 bpf_mtap_af(void *arg, u_int32_t af, struct mbuf *m)
1291 {
1292 	struct mbuf m0;
1293 
1294 	m0.m_flags = 0;
1295 	m0.m_next = m;
1296 	m0.m_len = 4;
1297 	m0.m_data = (char *)&af;
1298 
1299 	bpf_mtap(arg, &m0);
1300 }
1301 
1302 void
1303 bpf_mtap_et(void *arg, u_int16_t et, struct mbuf *m)
1304 {
1305 	struct mbuf m0;
1306 
1307 	m0.m_flags = 0;
1308 	m0.m_next = m;
1309 	m0.m_len = 14;
1310 	m0.m_data = m0.m_dat;
1311 
1312 	((u_int32_t *)m0.m_data)[0] = 0;
1313 	((u_int32_t *)m0.m_data)[1] = 0;
1314 	((u_int32_t *)m0.m_data)[2] = 0;
1315 	((u_int16_t *)m0.m_data)[6] = et;
1316 
1317 	bpf_mtap(arg, &m0);
1318 }
1319 
1320 #if NSL > 0 || NSTRIP > 0
1321 /*
1322  * Put the SLIP pseudo-"link header" in place.
1323  * Note this M_PREPEND() should never fail,
1324  * swince we know we always have enough space
1325  * in the input buffer.
1326  */
1327 void
1328 bpf_mtap_sl_in(void *arg, u_char *chdr, struct mbuf **m)
1329 {
1330 	int s;
1331 	u_char *hp;
1332 
1333 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1334 	if (*m == NULL)
1335 		return;
1336 
1337 	hp = mtod(*m, u_char *);
1338 	hp[SLX_DIR] = SLIPDIR_IN;
1339 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1340 
1341 	s = splnet();
1342 	bpf_mtap(arg, *m);
1343 	splx(s);
1344 
1345 	m_adj(*m, SLIP_HDRLEN);
1346 }
1347 
1348 /*
1349  * Put the SLIP pseudo-"link header" in
1350  * place.  The compressed header is now
1351  * at the beginning of the mbuf.
1352  */
1353 void
1354 bpf_mtap_sl_out(void *arg, u_char *chdr, struct mbuf *m)
1355 {
1356 	struct mbuf m0;
1357 	u_char *hp;
1358 	int s;
1359 
1360 	m0.m_flags = 0;
1361 	m0.m_next = m;
1362 	m0.m_data = m0.m_dat;
1363 	m0.m_len = SLIP_HDRLEN;
1364 
1365 	hp = mtod(&m0, u_char *);
1366 
1367 	hp[SLX_DIR] = SLIPDIR_OUT;
1368 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1369 
1370 	s = splnet();
1371 	bpf_mtap(arg, &m0);
1372 	splx(s);
1373 	m_freem(m);
1374 }
1375 #endif
1376 
1377 /*
1378  * Move the packet data from interface memory (pkt) into the
1379  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1380  * otherwise 0.  "copy" is the routine called to do the actual data
1381  * transfer.  memcpy is passed in to copy contiguous chunks, while
1382  * bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1383  * pkt is really an mbuf.
1384  */
1385 static void
1386 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1387 	    void *(*cpfn)(void *, const void *, size_t))
1388 {
1389 	struct bpf_hdr *hp;
1390 	int totlen, curlen;
1391 	int hdrlen = d->bd_bif->bif_hdrlen;
1392 
1393 	++d->bd_ccount;
1394 	++bpf_gstats.bs_capt;
1395 	/*
1396 	 * Figure out how many bytes to move.  If the packet is
1397 	 * greater or equal to the snapshot length, transfer that
1398 	 * much.  Otherwise, transfer the whole packet (unless
1399 	 * we hit the buffer size limit).
1400 	 */
1401 	totlen = hdrlen + min(snaplen, pktlen);
1402 	if (totlen > d->bd_bufsize)
1403 		totlen = d->bd_bufsize;
1404 
1405 	/*
1406 	 * Round up the end of the previous packet to the next longword.
1407 	 */
1408 	curlen = BPF_WORDALIGN(d->bd_slen);
1409 	if (curlen + totlen > d->bd_bufsize) {
1410 		/*
1411 		 * This packet will overflow the storage buffer.
1412 		 * Rotate the buffers if we can, then wakeup any
1413 		 * pending reads.
1414 		 */
1415 		if (d->bd_fbuf == 0) {
1416 			/*
1417 			 * We haven't completed the previous read yet,
1418 			 * so drop the packet.
1419 			 */
1420 			++d->bd_dcount;
1421 			++bpf_gstats.bs_drop;
1422 			return;
1423 		}
1424 		ROTATE_BUFFERS(d);
1425 		bpf_wakeup(d);
1426 		curlen = 0;
1427 	}
1428 
1429 	/*
1430 	 * Append the bpf header.
1431 	 */
1432 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1433 	microtime(&hp->bh_tstamp);
1434 	hp->bh_datalen = pktlen;
1435 	hp->bh_hdrlen = hdrlen;
1436 	/*
1437 	 * Copy the packet data into the store buffer and update its length.
1438 	 */
1439 	(*cpfn)((u_char *)hp + hdrlen, pkt, (hp->bh_caplen = totlen - hdrlen));
1440 	d->bd_slen = curlen + totlen;
1441 
1442 	/*
1443 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1444 	 * will cause filt_bpfread() to be called with it adjusted.
1445 	 */
1446 	if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
1447 		/*
1448 		 * Immediate mode is set, or the read timeout has
1449 		 * already expired during a select call.  A packet
1450 		 * arrived, so the reader should be woken up.
1451 		 */
1452 		bpf_wakeup(d);
1453 }
1454 
1455 /*
1456  * Initialize all nonzero fields of a descriptor.
1457  */
1458 static int
1459 bpf_allocbufs(struct bpf_d *d)
1460 {
1461 
1462 	d->bd_fbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1463 	if (!d->bd_fbuf)
1464 		return (ENOBUFS);
1465 	d->bd_sbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1466 	if (!d->bd_sbuf) {
1467 		free(d->bd_fbuf, M_DEVBUF);
1468 		return (ENOBUFS);
1469 	}
1470 	d->bd_slen = 0;
1471 	d->bd_hlen = 0;
1472 	return (0);
1473 }
1474 
1475 /*
1476  * Free buffers currently in use by a descriptor.
1477  * Called on close.
1478  */
1479 static void
1480 bpf_freed(struct bpf_d *d)
1481 {
1482 	/*
1483 	 * We don't need to lock out interrupts since this descriptor has
1484 	 * been detached from its interface and it yet hasn't been marked
1485 	 * free.
1486 	 */
1487 	if (d->bd_sbuf != 0) {
1488 		free(d->bd_sbuf, M_DEVBUF);
1489 		if (d->bd_hbuf != 0)
1490 			free(d->bd_hbuf, M_DEVBUF);
1491 		if (d->bd_fbuf != 0)
1492 			free(d->bd_fbuf, M_DEVBUF);
1493 	}
1494 	if (d->bd_filter)
1495 		free(d->bd_filter, M_DEVBUF);
1496 }
1497 
1498 /*
1499  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
1500  * fixed size of the link header (variable length headers not yet supported).
1501  */
1502 void
1503 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1504 {
1505 
1506 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
1507 }
1508 
1509 /*
1510  * Attach additional dlt for a interface to bpf.  dlt is the link layer type;
1511  * hdrlen is the fixed size of the link header for the specified dlt
1512  * (variable length headers not yet supported).
1513  */
1514 void
1515 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, void *driverp)
1516 {
1517 	struct bpf_if *bp;
1518 	bp = malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1519 	if (bp == 0)
1520 		panic("bpfattach");
1521 
1522 	bp->bif_dlist = 0;
1523 	bp->bif_driverp = driverp;
1524 	bp->bif_ifp = ifp;
1525 	bp->bif_dlt = dlt;
1526 
1527 	bp->bif_next = bpf_iflist;
1528 	bpf_iflist = bp;
1529 
1530 	*bp->bif_driverp = 0;
1531 
1532 	/*
1533 	 * Compute the length of the bpf header.  This is not necessarily
1534 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1535 	 * that the network layer header begins on a longword boundary (for
1536 	 * performance reasons and to alleviate alignment restrictions).
1537 	 */
1538 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1539 
1540 #if 0
1541 	printf("bpf: %s attached\n", ifp->if_xname);
1542 #endif
1543 }
1544 
1545 /*
1546  * Remove an interface from bpf.
1547  */
1548 void
1549 bpfdetach(struct ifnet *ifp)
1550 {
1551 	struct bpf_if *bp, **pbp;
1552 	struct bpf_d *d;
1553 	int s;
1554 
1555 	/* Nuke the vnodes for any open instances */
1556 	for (d = LIST_FIRST(&bpf_list); d != NULL; d = LIST_NEXT(d, bd_list)) {
1557 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
1558 			/*
1559 			 * Detach the descriptor from an interface now.
1560 			 * It will be free'ed later by close routine.
1561 			 */
1562 			s = splnet();
1563 			d->bd_promisc = 0;	/* we can't touch device. */
1564 			bpf_detachd(d);
1565 			splx(s);
1566 		}
1567 	}
1568 
1569   again:
1570 	for (bp = bpf_iflist, pbp = &bpf_iflist;
1571 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
1572 		if (bp->bif_ifp == ifp) {
1573 			*pbp = bp->bif_next;
1574 			free(bp, M_DEVBUF);
1575 			goto again;
1576 		}
1577 	}
1578 }
1579 
1580 /*
1581  * Change the data link type of a interface.
1582  */
1583 void
1584 bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1585 {
1586 	struct bpf_if *bp;
1587 
1588 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1589 		if ((caddr_t *)bp->bif_driverp == &ifp->if_bpf)
1590 			break;
1591 	}
1592 	if (bp == NULL)
1593 		panic("bpf_change_type");
1594 
1595 	bp->bif_dlt = dlt;
1596 
1597 	/*
1598 	 * Compute the length of the bpf header.  This is not necessarily
1599 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1600 	 * that the network layer header begins on a longword boundary (for
1601 	 * performance reasons and to alleviate alignment restrictions).
1602 	 */
1603 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1604 }
1605 
1606 /*
1607  * Get a list of available data link type of the interface.
1608  */
1609 static int
1610 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1611 {
1612 	int n, error;
1613 	struct ifnet *ifp;
1614 	struct bpf_if *bp;
1615 
1616 	ifp = d->bd_bif->bif_ifp;
1617 	n = 0;
1618 	error = 0;
1619 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1620 		if (bp->bif_ifp != ifp)
1621 			continue;
1622 		if (bfl->bfl_list != NULL) {
1623 			if (n >= bfl->bfl_len)
1624 				return ENOMEM;
1625 			error = copyout(&bp->bif_dlt,
1626 			    bfl->bfl_list + n, sizeof(u_int));
1627 		}
1628 		n++;
1629 	}
1630 	bfl->bfl_len = n;
1631 	return error;
1632 }
1633 
1634 /*
1635  * Set the data link type of a BPF instance.
1636  */
1637 static int
1638 bpf_setdlt(struct bpf_d *d, u_int dlt)
1639 {
1640 	int s, error, opromisc;
1641 	struct ifnet *ifp;
1642 	struct bpf_if *bp;
1643 
1644 	if (d->bd_bif->bif_dlt == dlt)
1645 		return 0;
1646 	ifp = d->bd_bif->bif_ifp;
1647 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1648 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1649 			break;
1650 	}
1651 	if (bp == NULL)
1652 		return EINVAL;
1653 	s = splnet();
1654 	opromisc = d->bd_promisc;
1655 	bpf_detachd(d);
1656 	bpf_attachd(d, bp);
1657 	reset_d(d);
1658 	if (opromisc) {
1659 		error = ifpromisc(bp->bif_ifp, 1);
1660 		if (error)
1661 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
1662 			    bp->bif_ifp->if_xname, error);
1663 		else
1664 			d->bd_promisc = 1;
1665 	}
1666 	splx(s);
1667 	return 0;
1668 }
1669 
1670 static int
1671 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
1672 {
1673 	int newsize, error;
1674 	struct sysctlnode node;
1675 
1676 	node = *rnode;
1677 	node.sysctl_data = &newsize;
1678 	newsize = bpf_maxbufsize;
1679 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1680 	if (error || newp == NULL)
1681 		return (error);
1682 
1683 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
1684 		return (EINVAL);
1685 
1686 	bpf_maxbufsize = newsize;
1687 
1688 	return (0);
1689 }
1690 
1691 static int
1692 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
1693 {
1694 	int    error, elem_count;
1695 	struct bpf_d	 *dp;
1696 	struct bpf_d_ext  dpe;
1697 	size_t len, needed, elem_size, out_size;
1698 	char   *sp;
1699 
1700 	if (namelen == 1 && name[0] == CTL_QUERY)
1701 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
1702 
1703 	if (namelen != 2)
1704 		return (EINVAL);
1705 
1706 	if ((error = kauth_authorize_generic(l->l_proc->p_cred,
1707 				       KAUTH_GENERIC_ISSUSER,
1708 				       &l->l_proc->p_acflag)))
1709 		return (error);
1710 
1711 	len = (oldp != NULL) ? *oldlenp : 0;
1712 	sp = oldp;
1713 	elem_size = name[0];
1714 	elem_count = name[1];
1715 	out_size = MIN(sizeof(dpe), elem_size);
1716 	needed = 0;
1717 
1718 	if (elem_size < 1 || elem_count < 0)
1719 		return (EINVAL);
1720 
1721 	simple_lock(&bpf_slock);
1722 	LIST_FOREACH(dp, &bpf_list, bd_list) {
1723 		if (len >= elem_size && elem_count > 0) {
1724 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
1725 			BPF_EXT(bufsize);
1726 			BPF_EXT(promisc);
1727 			BPF_EXT(promisc);
1728 			BPF_EXT(state);
1729 			BPF_EXT(immediate);
1730 			BPF_EXT(hdrcmplt);
1731 			BPF_EXT(seesent);
1732 			BPF_EXT(pid);
1733 			BPF_EXT(rcount);
1734 			BPF_EXT(dcount);
1735 			BPF_EXT(ccount);
1736 #undef BPF_EXT
1737 			if (dp->bd_bif)
1738 				(void)strlcpy(dpe.bde_ifname,
1739 				    dp->bd_bif->bif_ifp->if_xname,
1740 				    IFNAMSIZ - 1);
1741 			else
1742 				dpe.bde_ifname[0] = '\0';
1743 
1744 			error = copyout(&dpe, sp, out_size);
1745 			if (error)
1746 				break;
1747 			sp += elem_size;
1748 			len -= elem_size;
1749 		}
1750 		if (elem_count > 0) {
1751 			needed += elem_size;
1752 			if (elem_count != INT_MAX)
1753 				elem_count--;
1754 		}
1755 	}
1756 	simple_unlock(&bpf_slock);
1757 
1758 	*oldlenp = needed;
1759 
1760 	return (error);
1761 }
1762 
1763 SYSCTL_SETUP(sysctl_net_bpf_setup, "sysctl net.bpf subtree setup")
1764 {
1765 	const struct sysctlnode *node;
1766 
1767 	sysctl_createv(clog, 0, NULL, NULL,
1768 		       CTLFLAG_PERMANENT,
1769 		       CTLTYPE_NODE, "net", NULL,
1770 		       NULL, 0, NULL, 0,
1771 		       CTL_NET, CTL_EOL);
1772 
1773 	node = NULL;
1774 	sysctl_createv(clog, 0, NULL, &node,
1775 		       CTLFLAG_PERMANENT,
1776 		       CTLTYPE_NODE, "bpf",
1777 		       SYSCTL_DESCR("BPF options"),
1778 		       NULL, 0, NULL, 0,
1779 		       CTL_NET, CTL_CREATE, CTL_EOL);
1780 	if (node != NULL) {
1781 		sysctl_createv(clog, 0, NULL, NULL,
1782 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1783 			CTLTYPE_INT, "maxbufsize",
1784 			SYSCTL_DESCR("Maximum size for data capture buffer"),
1785 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
1786 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1787 		sysctl_createv(clog, 0, NULL, NULL,
1788 			CTLFLAG_PERMANENT,
1789 			CTLTYPE_STRUCT, "stats",
1790 			SYSCTL_DESCR("BPF stats"),
1791 			NULL, 0, &bpf_gstats, sizeof(bpf_gstats),
1792 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1793 		sysctl_createv(clog, 0, NULL, NULL,
1794 			CTLFLAG_PERMANENT,
1795 			CTLTYPE_STRUCT, "peers",
1796 			SYSCTL_DESCR("BPF peers"),
1797 			sysctl_net_bpf_peers, 0, NULL, 0,
1798 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1799 	}
1800 
1801 }
1802