xref: /netbsd-src/sys/net/bpf.c (revision 274254cdae52594c1aa480a736aef78313d15c9c)
1 /*	$NetBSD: bpf.c,v 1.144 2009/04/04 10:12:51 ad 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 1.144 2009/04/04 10:12:51 ad Exp $");
43 
44 #if defined(_KERNEL_OPT)
45 #include "opt_bpf.h"
46 #include "sl.h"
47 #include "strip.h"
48 #endif
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/mbuf.h>
53 #include <sys/buf.h>
54 #include <sys/time.h>
55 #include <sys/proc.h>
56 #include <sys/user.h>
57 #include <sys/ioctl.h>
58 #include <sys/conf.h>
59 #include <sys/vnode.h>
60 #include <sys/queue.h>
61 
62 #include <sys/file.h>
63 #include <sys/filedesc.h>
64 #include <sys/tty.h>
65 #include <sys/uio.h>
66 
67 #include <sys/protosw.h>
68 #include <sys/socket.h>
69 #include <sys/errno.h>
70 #include <sys/kernel.h>
71 #include <sys/poll.h>
72 #include <sys/sysctl.h>
73 #include <sys/kauth.h>
74 
75 #include <net/if.h>
76 #include <net/slip.h>
77 
78 #include <net/bpf.h>
79 #include <net/bpfdesc.h>
80 
81 #include <net/if_arc.h>
82 #include <net/if_ether.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/if_inarp.h>
86 
87 
88 #include <compat/sys/sockio.h>
89 
90 #ifndef BPF_BUFSIZE
91 /*
92  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
93  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
94  */
95 # define BPF_BUFSIZE 32768
96 #endif
97 
98 #define PRINET  26			/* interruptible */
99 
100 /*
101  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
102  * XXX the default values should be computed dynamically based
103  * on available memory size and available mbuf clusters.
104  */
105 int bpf_bufsize = BPF_BUFSIZE;
106 int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
107 
108 
109 /*
110  * Global BPF statistics returned by net.bpf.stats sysctl.
111  */
112 struct bpf_stat	bpf_gstats;
113 
114 /*
115  * Use a mutex to avoid a race condition between gathering the stats/peers
116  * and opening/closing the device.
117  */
118 static kmutex_t bpf_mtx;
119 
120 /*
121  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
122  *  bpf_dtab holds the descriptors, indexed by minor device #
123  */
124 struct bpf_if	*bpf_iflist;
125 LIST_HEAD(, bpf_d) bpf_list;
126 
127 static int	bpf_allocbufs(struct bpf_d *);
128 static void	bpf_deliver(struct bpf_if *,
129 		            void *(*cpfn)(void *, const void *, size_t),
130 			    void *, u_int, u_int, struct ifnet *);
131 static void	bpf_freed(struct bpf_d *);
132 static void	bpf_ifname(struct ifnet *, struct ifreq *);
133 static void	*bpf_mcpy(void *, const void *, size_t);
134 static int	bpf_movein(struct uio *, int, int,
135 			        struct mbuf **, struct sockaddr *);
136 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
137 static void	bpf_detachd(struct bpf_d *);
138 static int	bpf_setif(struct bpf_d *, struct ifreq *);
139 static void	bpf_timed_out(void *);
140 static inline void
141 		bpf_wakeup(struct bpf_d *);
142 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
143     void *(*)(void *, const void *, size_t), struct timespec *);
144 static void	reset_d(struct bpf_d *);
145 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
146 static int	bpf_setdlt(struct bpf_d *, u_int);
147 
148 static int	bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t,
149     int);
150 static int	bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t,
151     int);
152 static int	bpf_ioctl(struct file *, u_long, void *);
153 static int	bpf_poll(struct file *, int);
154 static int	bpf_close(struct file *);
155 static int	bpf_kqfilter(struct file *, struct knote *);
156 static void	bpf_softintr(void *);
157 
158 static const struct fileops bpf_fileops = {
159 	.fo_read = bpf_read,
160 	.fo_write = bpf_write,
161 	.fo_ioctl = bpf_ioctl,
162 	.fo_fcntl = fnullop_fcntl,
163 	.fo_poll = bpf_poll,
164 	.fo_stat = fbadop_stat,
165 	.fo_close = bpf_close,
166 	.fo_kqfilter = bpf_kqfilter,
167 	.fo_drain = fnullop_drain,
168 };
169 
170 dev_type_open(bpfopen);
171 
172 const struct cdevsw bpf_cdevsw = {
173 	bpfopen, noclose, noread, nowrite, noioctl,
174 	nostop, notty, nopoll, nommap, nokqfilter, D_OTHER
175 };
176 
177 static int
178 bpf_movein(struct uio *uio, int linktype, int mtu, struct mbuf **mp,
179 	   struct sockaddr *sockp)
180 {
181 	struct mbuf *m;
182 	int error;
183 	int len;
184 	int hlen;
185 	int align;
186 
187 	/*
188 	 * Build a sockaddr based on the data link layer type.
189 	 * We do this at this level because the ethernet header
190 	 * is copied directly into the data field of the sockaddr.
191 	 * In the case of SLIP, there is no header and the packet
192 	 * is forwarded as is.
193 	 * Also, we are careful to leave room at the front of the mbuf
194 	 * for the link level header.
195 	 */
196 	switch (linktype) {
197 
198 	case DLT_SLIP:
199 		sockp->sa_family = AF_INET;
200 		hlen = 0;
201 		align = 0;
202 		break;
203 
204 	case DLT_PPP:
205 		sockp->sa_family = AF_UNSPEC;
206 		hlen = 0;
207 		align = 0;
208 		break;
209 
210 	case DLT_EN10MB:
211 		sockp->sa_family = AF_UNSPEC;
212 		/* XXX Would MAXLINKHDR be better? */
213  		/* 6(dst)+6(src)+2(type) */
214 		hlen = sizeof(struct ether_header);
215 		align = 2;
216 		break;
217 
218 	case DLT_ARCNET:
219 		sockp->sa_family = AF_UNSPEC;
220 		hlen = ARC_HDRLEN;
221 		align = 5;
222 		break;
223 
224 	case DLT_FDDI:
225 		sockp->sa_family = AF_LINK;
226 		/* XXX 4(FORMAC)+6(dst)+6(src) */
227 		hlen = 16;
228 		align = 0;
229 		break;
230 
231 	case DLT_ECONET:
232 		sockp->sa_family = AF_UNSPEC;
233 		hlen = 6;
234 		align = 2;
235 		break;
236 
237 	case DLT_NULL:
238 		sockp->sa_family = AF_UNSPEC;
239 		hlen = 0;
240 		align = 0;
241 		break;
242 
243 	default:
244 		return (EIO);
245 	}
246 
247 	len = uio->uio_resid;
248 	/*
249 	 * If there aren't enough bytes for a link level header or the
250 	 * packet length exceeds the interface mtu, return an error.
251 	 */
252 	if (len < hlen || len - hlen > mtu)
253 		return (EMSGSIZE);
254 
255 	/*
256 	 * XXX Avoid complicated buffer chaining ---
257 	 * bail if it won't fit in a single mbuf.
258 	 * (Take into account possible alignment bytes)
259 	 */
260 	if ((unsigned)len > MCLBYTES - align)
261 		return (EIO);
262 
263 	m = m_gethdr(M_WAIT, MT_DATA);
264 	m->m_pkthdr.rcvif = 0;
265 	m->m_pkthdr.len = len - hlen;
266 	if (len > MHLEN - align) {
267 		m_clget(m, M_WAIT);
268 		if ((m->m_flags & M_EXT) == 0) {
269 			error = ENOBUFS;
270 			goto bad;
271 		}
272 	}
273 
274 	/* Insure the data is properly aligned */
275 	if (align > 0) {
276 		m->m_data += align;
277 		m->m_len -= align;
278 	}
279 
280 	error = uiomove(mtod(m, void *), len, uio);
281 	if (error)
282 		goto bad;
283 	if (hlen != 0) {
284 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
285 		m->m_data += hlen; /* XXX */
286 		len -= hlen;
287 	}
288 	m->m_len = len;
289 	*mp = m;
290 	return (0);
291 
292 bad:
293 	m_freem(m);
294 	return (error);
295 }
296 
297 /*
298  * Attach file to the bpf interface, i.e. make d listen on bp.
299  * Must be called at splnet.
300  */
301 static void
302 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
303 {
304 	/*
305 	 * Point d at bp, and add d to the interface's list of listeners.
306 	 * Finally, point the driver's bpf cookie at the interface so
307 	 * it will divert packets to bpf.
308 	 */
309 	d->bd_bif = bp;
310 	d->bd_next = bp->bif_dlist;
311 	bp->bif_dlist = d;
312 
313 	*bp->bif_driverp = bp;
314 }
315 
316 /*
317  * Detach a file from its interface.
318  */
319 static void
320 bpf_detachd(struct bpf_d *d)
321 {
322 	struct bpf_d **p;
323 	struct bpf_if *bp;
324 
325 	bp = d->bd_bif;
326 	/*
327 	 * Check if this descriptor had requested promiscuous mode.
328 	 * If so, turn it off.
329 	 */
330 	if (d->bd_promisc) {
331 		int error;
332 
333 		d->bd_promisc = 0;
334 		/*
335 		 * Take device out of promiscuous mode.  Since we were
336 		 * able to enter promiscuous mode, we should be able
337 		 * to turn it off.  But we can get an error if
338 		 * the interface was configured down, so only panic
339 		 * if we don't get an unexpected error.
340 		 */
341   		error = ifpromisc(bp->bif_ifp, 0);
342 		if (error && error != EINVAL)
343 			panic("bpf: ifpromisc failed");
344 	}
345 	/* Remove d from the interface's descriptor list. */
346 	p = &bp->bif_dlist;
347 	while (*p != d) {
348 		p = &(*p)->bd_next;
349 		if (*p == 0)
350 			panic("bpf_detachd: descriptor not in list");
351 	}
352 	*p = (*p)->bd_next;
353 	if (bp->bif_dlist == 0)
354 		/*
355 		 * Let the driver know that there are no more listeners.
356 		 */
357 		*d->bd_bif->bif_driverp = 0;
358 	d->bd_bif = 0;
359 }
360 
361 
362 /*
363  * Mark a descriptor free by making it point to itself.
364  * This is probably cheaper than marking with a constant since
365  * the address should be in a register anyway.
366  */
367 
368 /*
369  * bpfilterattach() is called at boot time.
370  */
371 /* ARGSUSED */
372 void
373 bpfilterattach(int n)
374 {
375 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
376 
377 	mutex_enter(&bpf_mtx);
378 	LIST_INIT(&bpf_list);
379 	mutex_exit(&bpf_mtx);
380 
381 	bpf_gstats.bs_recv = 0;
382 	bpf_gstats.bs_drop = 0;
383 	bpf_gstats.bs_capt = 0;
384 }
385 
386 /*
387  * Open ethernet device. Clones.
388  */
389 /* ARGSUSED */
390 int
391 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
392 {
393 	struct bpf_d *d;
394 	struct file *fp;
395 	int error, fd;
396 
397 	/* falloc() will use the descriptor for us. */
398 	if ((error = fd_allocfile(&fp, &fd)) != 0)
399 		return error;
400 
401 	d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK|M_ZERO);
402 	d->bd_bufsize = bpf_bufsize;
403 	d->bd_seesent = 1;
404 	d->bd_pid = l->l_proc->p_pid;
405 	callout_init(&d->bd_callout, 0);
406 	selinit(&d->bd_sel);
407 	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
408 
409 	mutex_enter(&bpf_mtx);
410 	LIST_INSERT_HEAD(&bpf_list, d, bd_list);
411 	mutex_exit(&bpf_mtx);
412 
413 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
414 }
415 
416 /*
417  * Close the descriptor by detaching it from its interface,
418  * deallocating its buffers, and marking it free.
419  */
420 /* ARGSUSED */
421 static int
422 bpf_close(struct file *fp)
423 {
424 	struct bpf_d *d = fp->f_data;
425 	int s;
426 
427 	KERNEL_LOCK(1, NULL);
428 
429 	/*
430 	 * Refresh the PID associated with this bpf file.
431 	 */
432 	d->bd_pid = curproc->p_pid;
433 
434 	s = splnet();
435 	if (d->bd_state == BPF_WAITING)
436 		callout_stop(&d->bd_callout);
437 	d->bd_state = BPF_IDLE;
438 	if (d->bd_bif)
439 		bpf_detachd(d);
440 	splx(s);
441 	bpf_freed(d);
442 	mutex_enter(&bpf_mtx);
443 	LIST_REMOVE(d, bd_list);
444 	mutex_exit(&bpf_mtx);
445 	callout_destroy(&d->bd_callout);
446 	seldestroy(&d->bd_sel);
447 	softint_disestablish(d->bd_sih);
448 	free(d, M_DEVBUF);
449 	fp->f_data = NULL;
450 
451 	KERNEL_UNLOCK_ONE(NULL);
452 
453 	return (0);
454 }
455 
456 /*
457  * Rotate the packet buffers in descriptor d.  Move the store buffer
458  * into the hold slot, and the free buffer into the store slot.
459  * Zero the length of the new store buffer.
460  */
461 #define ROTATE_BUFFERS(d) \
462 	(d)->bd_hbuf = (d)->bd_sbuf; \
463 	(d)->bd_hlen = (d)->bd_slen; \
464 	(d)->bd_sbuf = (d)->bd_fbuf; \
465 	(d)->bd_slen = 0; \
466 	(d)->bd_fbuf = 0;
467 /*
468  *  bpfread - read next chunk of packets from buffers
469  */
470 static int
471 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
472     kauth_cred_t cred, int flags)
473 {
474 	struct bpf_d *d = fp->f_data;
475 	int timed_out;
476 	int error;
477 	int s;
478 
479 	/*
480 	 * Restrict application to use a buffer the same size as
481 	 * the kernel buffers.
482 	 */
483 	if (uio->uio_resid != d->bd_bufsize)
484 		return (EINVAL);
485 
486 	KERNEL_LOCK(1, NULL);
487 	s = splnet();
488 	if (d->bd_state == BPF_WAITING)
489 		callout_stop(&d->bd_callout);
490 	timed_out = (d->bd_state == BPF_TIMED_OUT);
491 	d->bd_state = BPF_IDLE;
492 	/*
493 	 * If the hold buffer is empty, then do a timed sleep, which
494 	 * ends when the timeout expires or when enough packets
495 	 * have arrived to fill the store buffer.
496 	 */
497 	while (d->bd_hbuf == 0) {
498 		if (fp->f_flag & FNONBLOCK) {
499 			if (d->bd_slen == 0) {
500 				splx(s);
501 				KERNEL_UNLOCK_ONE(NULL);
502 				return (EWOULDBLOCK);
503 			}
504 			ROTATE_BUFFERS(d);
505 			break;
506 		}
507 
508 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
509 			/*
510 			 * A packet(s) either arrived since the previous
511 			 * read or arrived while we were asleep.
512 			 * Rotate the buffers and return what's here.
513 			 */
514 			ROTATE_BUFFERS(d);
515 			break;
516 		}
517 		error = tsleep(d, PRINET|PCATCH, "bpf",
518 				d->bd_rtout);
519 		if (error == EINTR || error == ERESTART) {
520 			splx(s);
521 			KERNEL_UNLOCK_ONE(NULL);
522 			return (error);
523 		}
524 		if (error == EWOULDBLOCK) {
525 			/*
526 			 * On a timeout, return what's in the buffer,
527 			 * which may be nothing.  If there is something
528 			 * in the store buffer, we can rotate the buffers.
529 			 */
530 			if (d->bd_hbuf)
531 				/*
532 				 * We filled up the buffer in between
533 				 * getting the timeout and arriving
534 				 * here, so we don't need to rotate.
535 				 */
536 				break;
537 
538 			if (d->bd_slen == 0) {
539 				splx(s);
540 				KERNEL_UNLOCK_ONE(NULL);
541 				return (0);
542 			}
543 			ROTATE_BUFFERS(d);
544 			break;
545 		}
546 		if (error != 0)
547 			goto done;
548 	}
549 	/*
550 	 * At this point, we know we have something in the hold slot.
551 	 */
552 	splx(s);
553 
554 	/*
555 	 * Move data from hold buffer into user space.
556 	 * We know the entire buffer is transferred since
557 	 * we checked above that the read buffer is bpf_bufsize bytes.
558 	 */
559 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
560 
561 	s = splnet();
562 	d->bd_fbuf = d->bd_hbuf;
563 	d->bd_hbuf = 0;
564 	d->bd_hlen = 0;
565 done:
566 	splx(s);
567 	KERNEL_UNLOCK_ONE(NULL);
568 	return (error);
569 }
570 
571 
572 /*
573  * If there are processes sleeping on this descriptor, wake them up.
574  */
575 static inline void
576 bpf_wakeup(struct bpf_d *d)
577 {
578 	wakeup(d);
579 	if (d->bd_async)
580 		softint_schedule(d->bd_sih);
581 	selnotify(&d->bd_sel, 0, 0);
582 }
583 
584 static void
585 bpf_softintr(void *cookie)
586 {
587 	struct bpf_d *d;
588 
589 	d = cookie;
590 	if (d->bd_async)
591 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
592 }
593 
594 static void
595 bpf_timed_out(void *arg)
596 {
597 	struct bpf_d *d = arg;
598 	int s;
599 
600 	s = splnet();
601 	if (d->bd_state == BPF_WAITING) {
602 		d->bd_state = BPF_TIMED_OUT;
603 		if (d->bd_slen != 0)
604 			bpf_wakeup(d);
605 	}
606 	splx(s);
607 }
608 
609 
610 static int
611 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
612     kauth_cred_t cred, int flags)
613 {
614 	struct bpf_d *d = fp->f_data;
615 	struct ifnet *ifp;
616 	struct mbuf *m;
617 	int error, s;
618 	static struct sockaddr_storage dst;
619 
620 	m = NULL;	/* XXX gcc */
621 
622 	KERNEL_LOCK(1, NULL);
623 
624 	if (d->bd_bif == 0) {
625 		KERNEL_UNLOCK_ONE(NULL);
626 		return (ENXIO);
627 	}
628 
629 	ifp = d->bd_bif->bif_ifp;
630 
631 	if (uio->uio_resid == 0) {
632 		KERNEL_UNLOCK_ONE(NULL);
633 		return (0);
634 	}
635 
636 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
637 		(struct sockaddr *) &dst);
638 	if (error) {
639 		KERNEL_UNLOCK_ONE(NULL);
640 		return (error);
641 	}
642 
643 	if (m->m_pkthdr.len > ifp->if_mtu) {
644 		KERNEL_UNLOCK_ONE(NULL);
645 		m_freem(m);
646 		return (EMSGSIZE);
647 	}
648 
649 	if (d->bd_hdrcmplt)
650 		dst.ss_family = pseudo_AF_HDRCMPLT;
651 
652 	s = splsoftnet();
653 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *) &dst, NULL);
654 	splx(s);
655 	KERNEL_UNLOCK_ONE(NULL);
656 	/*
657 	 * The driver frees the mbuf.
658 	 */
659 	return (error);
660 }
661 
662 /*
663  * Reset a descriptor by flushing its packet buffer and clearing the
664  * receive and drop counts.  Should be called at splnet.
665  */
666 static void
667 reset_d(struct bpf_d *d)
668 {
669 	if (d->bd_hbuf) {
670 		/* Free the hold buffer. */
671 		d->bd_fbuf = d->bd_hbuf;
672 		d->bd_hbuf = 0;
673 	}
674 	d->bd_slen = 0;
675 	d->bd_hlen = 0;
676 	d->bd_rcount = 0;
677 	d->bd_dcount = 0;
678 	d->bd_ccount = 0;
679 }
680 
681 /*
682  *  FIONREAD		Check for read packet available.
683  *  BIOCGBLEN		Get buffer len [for read()].
684  *  BIOCSETF		Set ethernet read filter.
685  *  BIOCFLUSH		Flush read packet buffer.
686  *  BIOCPROMISC		Put interface into promiscuous mode.
687  *  BIOCGDLT		Get link layer type.
688  *  BIOCGETIF		Get interface name.
689  *  BIOCSETIF		Set interface.
690  *  BIOCSRTIMEOUT	Set read timeout.
691  *  BIOCGRTIMEOUT	Get read timeout.
692  *  BIOCGSTATS		Get packet stats.
693  *  BIOCIMMEDIATE	Set immediate mode.
694  *  BIOCVERSION		Get filter language version.
695  *  BIOCGHDRCMPLT	Get "header already complete" flag.
696  *  BIOCSHDRCMPLT	Set "header already complete" flag.
697  */
698 /* ARGSUSED */
699 static int
700 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
701 {
702 	struct bpf_d *d = fp->f_data;
703 	int s, error = 0;
704 
705 	/*
706 	 * Refresh the PID associated with this bpf file.
707 	 */
708 	KERNEL_LOCK(1, NULL);
709 	d->bd_pid = curproc->p_pid;
710 
711 	s = splnet();
712 	if (d->bd_state == BPF_WAITING)
713 		callout_stop(&d->bd_callout);
714 	d->bd_state = BPF_IDLE;
715 	splx(s);
716 
717 	switch (cmd) {
718 
719 	default:
720 		error = EINVAL;
721 		break;
722 
723 	/*
724 	 * Check for read packet available.
725 	 */
726 	case FIONREAD:
727 		{
728 			int n;
729 
730 			s = splnet();
731 			n = d->bd_slen;
732 			if (d->bd_hbuf)
733 				n += d->bd_hlen;
734 			splx(s);
735 
736 			*(int *)addr = n;
737 			break;
738 		}
739 
740 	/*
741 	 * Get buffer len [for read()].
742 	 */
743 	case BIOCGBLEN:
744 		*(u_int *)addr = d->bd_bufsize;
745 		break;
746 
747 	/*
748 	 * Set buffer length.
749 	 */
750 	case BIOCSBLEN:
751 		if (d->bd_bif != 0)
752 			error = EINVAL;
753 		else {
754 			u_int size = *(u_int *)addr;
755 
756 			if (size > bpf_maxbufsize)
757 				*(u_int *)addr = size = bpf_maxbufsize;
758 			else if (size < BPF_MINBUFSIZE)
759 				*(u_int *)addr = size = BPF_MINBUFSIZE;
760 			d->bd_bufsize = size;
761 		}
762 		break;
763 
764 	/*
765 	 * Set link layer read filter.
766 	 */
767 	case BIOCSETF:
768 		error = bpf_setf(d, addr);
769 		break;
770 
771 	/*
772 	 * Flush read packet buffer.
773 	 */
774 	case BIOCFLUSH:
775 		s = splnet();
776 		reset_d(d);
777 		splx(s);
778 		break;
779 
780 	/*
781 	 * Put interface into promiscuous mode.
782 	 */
783 	case BIOCPROMISC:
784 		if (d->bd_bif == 0) {
785 			/*
786 			 * No interface attached yet.
787 			 */
788 			error = EINVAL;
789 			break;
790 		}
791 		s = splnet();
792 		if (d->bd_promisc == 0) {
793 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
794 			if (error == 0)
795 				d->bd_promisc = 1;
796 		}
797 		splx(s);
798 		break;
799 
800 	/*
801 	 * Get device parameters.
802 	 */
803 	case BIOCGDLT:
804 		if (d->bd_bif == 0)
805 			error = EINVAL;
806 		else
807 			*(u_int *)addr = d->bd_bif->bif_dlt;
808 		break;
809 
810 	/*
811 	 * Get a list of supported device parameters.
812 	 */
813 	case BIOCGDLTLIST:
814 		if (d->bd_bif == 0)
815 			error = EINVAL;
816 		else
817 			error = bpf_getdltlist(d, addr);
818 		break;
819 
820 	/*
821 	 * Set device parameters.
822 	 */
823 	case BIOCSDLT:
824 		if (d->bd_bif == 0)
825 			error = EINVAL;
826 		else
827 			error = bpf_setdlt(d, *(u_int *)addr);
828 		break;
829 
830 	/*
831 	 * Set interface name.
832 	 */
833 #ifdef OBIOCGETIF
834 	case OBIOCGETIF:
835 #endif
836 	case BIOCGETIF:
837 		if (d->bd_bif == 0)
838 			error = EINVAL;
839 		else
840 			bpf_ifname(d->bd_bif->bif_ifp, addr);
841 		break;
842 
843 	/*
844 	 * Set interface.
845 	 */
846 #ifdef OBIOCSETIF
847 	case OBIOCSETIF:
848 #endif
849 	case BIOCSETIF:
850 		error = bpf_setif(d, addr);
851 		break;
852 
853 	/*
854 	 * Set read timeout.
855 	 */
856 	case BIOCSRTIMEOUT:
857 		{
858 			struct timeval *tv = addr;
859 
860 			/* Compute number of ticks. */
861 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
862 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
863 				d->bd_rtout = 1;
864 			break;
865 		}
866 
867 #ifdef BIOCGORTIMEOUT
868 	/*
869 	 * Get read timeout.
870 	 */
871 	case BIOCGORTIMEOUT:
872 		{
873 			struct timeval50 *tv = addr;
874 
875 			tv->tv_sec = d->bd_rtout / hz;
876 			tv->tv_usec = (d->bd_rtout % hz) * tick;
877 			break;
878 		}
879 #endif
880 
881 #ifdef BIOCSORTIMEOUT
882 	/*
883 	 * Set read timeout.
884 	 */
885 	case BIOCSORTIMEOUT:
886 		{
887 			struct timeval50 *tv = addr;
888 
889 			/* Compute number of ticks. */
890 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
891 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
892 				d->bd_rtout = 1;
893 			break;
894 		}
895 #endif
896 
897 	/*
898 	 * Get read timeout.
899 	 */
900 	case BIOCGRTIMEOUT:
901 		{
902 			struct timeval *tv = addr;
903 
904 			tv->tv_sec = d->bd_rtout / hz;
905 			tv->tv_usec = (d->bd_rtout % hz) * tick;
906 			break;
907 		}
908 	/*
909 	 * Get packet stats.
910 	 */
911 	case BIOCGSTATS:
912 		{
913 			struct bpf_stat *bs = addr;
914 
915 			bs->bs_recv = d->bd_rcount;
916 			bs->bs_drop = d->bd_dcount;
917 			bs->bs_capt = d->bd_ccount;
918 			break;
919 		}
920 
921 	case BIOCGSTATSOLD:
922 		{
923 			struct bpf_stat_old *bs = addr;
924 
925 			bs->bs_recv = d->bd_rcount;
926 			bs->bs_drop = d->bd_dcount;
927 			break;
928 		}
929 
930 	/*
931 	 * Set immediate mode.
932 	 */
933 	case BIOCIMMEDIATE:
934 		d->bd_immediate = *(u_int *)addr;
935 		break;
936 
937 	case BIOCVERSION:
938 		{
939 			struct bpf_version *bv = addr;
940 
941 			bv->bv_major = BPF_MAJOR_VERSION;
942 			bv->bv_minor = BPF_MINOR_VERSION;
943 			break;
944 		}
945 
946 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
947 		*(u_int *)addr = d->bd_hdrcmplt;
948 		break;
949 
950 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
951 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
952 		break;
953 
954 	/*
955 	 * Get "see sent packets" flag
956 	 */
957 	case BIOCGSEESENT:
958 		*(u_int *)addr = d->bd_seesent;
959 		break;
960 
961 	/*
962 	 * Set "see sent" packets flag
963 	 */
964 	case BIOCSSEESENT:
965 		d->bd_seesent = *(u_int *)addr;
966 		break;
967 
968 	case FIONBIO:		/* Non-blocking I/O */
969 		/*
970 		 * No need to do anything special as we use IO_NDELAY in
971 		 * bpfread() as an indication of whether or not to block
972 		 * the read.
973 		 */
974 		break;
975 
976 	case FIOASYNC:		/* Send signal on receive packets */
977 		d->bd_async = *(int *)addr;
978 		break;
979 
980 	case TIOCSPGRP:		/* Process or group to send signals to */
981 	case FIOSETOWN:
982 		error = fsetown(&d->bd_pgid, cmd, addr);
983 		break;
984 
985 	case TIOCGPGRP:
986 	case FIOGETOWN:
987 		error = fgetown(d->bd_pgid, cmd, addr);
988 		break;
989 	}
990 	KERNEL_UNLOCK_ONE(NULL);
991 	return (error);
992 }
993 
994 /*
995  * Set d's packet filter program to fp.  If this file already has a filter,
996  * free it and replace it.  Returns EINVAL for bogus requests.
997  */
998 int
999 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
1000 {
1001 	struct bpf_insn *fcode, *old;
1002 	u_int flen, size;
1003 	int s;
1004 
1005 	old = d->bd_filter;
1006 	if (fp->bf_insns == 0) {
1007 		if (fp->bf_len != 0)
1008 			return (EINVAL);
1009 		s = splnet();
1010 		d->bd_filter = 0;
1011 		reset_d(d);
1012 		splx(s);
1013 		if (old != 0)
1014 			free(old, M_DEVBUF);
1015 		return (0);
1016 	}
1017 	flen = fp->bf_len;
1018 	if (flen > BPF_MAXINSNS)
1019 		return (EINVAL);
1020 
1021 	size = flen * sizeof(*fp->bf_insns);
1022 	fcode = malloc(size, M_DEVBUF, M_WAITOK);
1023 	if (copyin(fp->bf_insns, fcode, size) == 0 &&
1024 	    bpf_validate(fcode, (int)flen)) {
1025 		s = splnet();
1026 		d->bd_filter = fcode;
1027 		reset_d(d);
1028 		splx(s);
1029 		if (old != 0)
1030 			free(old, M_DEVBUF);
1031 
1032 		return (0);
1033 	}
1034 	free(fcode, M_DEVBUF);
1035 	return (EINVAL);
1036 }
1037 
1038 /*
1039  * Detach a file from its current interface (if attached at all) and attach
1040  * to the interface indicated by the name stored in ifr.
1041  * Return an errno or 0.
1042  */
1043 static int
1044 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1045 {
1046 	struct bpf_if *bp;
1047 	char *cp;
1048 	int unit_seen, i, s, error;
1049 
1050 	/*
1051 	 * Make sure the provided name has a unit number, and default
1052 	 * it to '0' if not specified.
1053 	 * XXX This is ugly ... do this differently?
1054 	 */
1055 	unit_seen = 0;
1056 	cp = ifr->ifr_name;
1057 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1058 	while (*cp++)
1059 		if (*cp >= '0' && *cp <= '9')
1060 			unit_seen = 1;
1061 	if (!unit_seen) {
1062 		/* Make sure to leave room for the '\0'. */
1063 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1064 			if ((ifr->ifr_name[i] >= 'a' &&
1065 			     ifr->ifr_name[i] <= 'z') ||
1066 			    (ifr->ifr_name[i] >= 'A' &&
1067 			     ifr->ifr_name[i] <= 'Z'))
1068 				continue;
1069 			ifr->ifr_name[i] = '0';
1070 		}
1071 	}
1072 
1073 	/*
1074 	 * Look through attached interfaces for the named one.
1075 	 */
1076 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
1077 		struct ifnet *ifp = bp->bif_ifp;
1078 
1079 		if (ifp == 0 ||
1080 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1081 			continue;
1082 		/* skip additional entry */
1083 		if ((void **)bp->bif_driverp != &ifp->if_bpf)
1084 			continue;
1085 		/*
1086 		 * We found the requested interface.
1087 		 * Allocate the packet buffers if we need to.
1088 		 * If we're already attached to requested interface,
1089 		 * just flush the buffer.
1090 		 */
1091 		if (d->bd_sbuf == 0) {
1092 			error = bpf_allocbufs(d);
1093 			if (error != 0)
1094 				return (error);
1095 		}
1096 		s = splnet();
1097 		if (bp != d->bd_bif) {
1098 			if (d->bd_bif)
1099 				/*
1100 				 * Detach if attached to something else.
1101 				 */
1102 				bpf_detachd(d);
1103 
1104 			bpf_attachd(d, bp);
1105 		}
1106 		reset_d(d);
1107 		splx(s);
1108 		return (0);
1109 	}
1110 	/* Not found. */
1111 	return (ENXIO);
1112 }
1113 
1114 /*
1115  * Copy the interface name to the ifreq.
1116  */
1117 static void
1118 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1119 {
1120 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1121 }
1122 
1123 /*
1124  * Support for poll() system call
1125  *
1126  * Return true iff the specific operation will not block indefinitely - with
1127  * the assumption that it is safe to positively acknowledge a request for the
1128  * ability to write to the BPF device.
1129  * Otherwise, return false but make a note that a selnotify() must be done.
1130  */
1131 static int
1132 bpf_poll(struct file *fp, int events)
1133 {
1134 	struct bpf_d *d = fp->f_data;
1135 	int s = splnet();
1136 	int revents;
1137 
1138 	/*
1139 	 * Refresh the PID associated with this bpf file.
1140 	 */
1141 	KERNEL_LOCK(1, NULL);
1142 	d->bd_pid = curproc->p_pid;
1143 
1144 	revents = events & (POLLOUT | POLLWRNORM);
1145 	if (events & (POLLIN | POLLRDNORM)) {
1146 		/*
1147 		 * An imitation of the FIONREAD ioctl code.
1148 		 */
1149 		if (d->bd_hlen != 0 ||
1150 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1151 		     d->bd_slen != 0)) {
1152 			revents |= events & (POLLIN | POLLRDNORM);
1153 		} else {
1154 			selrecord(curlwp, &d->bd_sel);
1155 			/* Start the read timeout if necessary */
1156 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1157 				callout_reset(&d->bd_callout, d->bd_rtout,
1158 					      bpf_timed_out, d);
1159 				d->bd_state = BPF_WAITING;
1160 			}
1161 		}
1162 	}
1163 
1164 	KERNEL_UNLOCK_ONE(NULL);
1165 	splx(s);
1166 	return (revents);
1167 }
1168 
1169 static void
1170 filt_bpfrdetach(struct knote *kn)
1171 {
1172 	struct bpf_d *d = kn->kn_hook;
1173 	int s;
1174 
1175 	KERNEL_LOCK(1, NULL);
1176 	s = splnet();
1177 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1178 	splx(s);
1179 	KERNEL_UNLOCK_ONE(NULL);
1180 }
1181 
1182 static int
1183 filt_bpfread(struct knote *kn, long hint)
1184 {
1185 	struct bpf_d *d = kn->kn_hook;
1186 	int rv;
1187 
1188 	KERNEL_LOCK(1, NULL);
1189 	kn->kn_data = d->bd_hlen;
1190 	if (d->bd_immediate)
1191 		kn->kn_data += d->bd_slen;
1192 	rv = (kn->kn_data > 0);
1193 	KERNEL_UNLOCK_ONE(NULL);
1194 	return rv;
1195 }
1196 
1197 static const struct filterops bpfread_filtops =
1198 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1199 
1200 static int
1201 bpf_kqfilter(struct file *fp, struct knote *kn)
1202 {
1203 	struct bpf_d *d = fp->f_data;
1204 	struct klist *klist;
1205 	int s;
1206 
1207 	KERNEL_LOCK(1, NULL);
1208 
1209 	switch (kn->kn_filter) {
1210 	case EVFILT_READ:
1211 		klist = &d->bd_sel.sel_klist;
1212 		kn->kn_fop = &bpfread_filtops;
1213 		break;
1214 
1215 	default:
1216 		KERNEL_UNLOCK_ONE(NULL);
1217 		return (EINVAL);
1218 	}
1219 
1220 	kn->kn_hook = d;
1221 
1222 	s = splnet();
1223 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1224 	splx(s);
1225 	KERNEL_UNLOCK_ONE(NULL);
1226 
1227 	return (0);
1228 }
1229 
1230 /*
1231  * Incoming linkage from device drivers.  Process the packet pkt, of length
1232  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1233  * by each process' filter, and if accepted, stashed into the corresponding
1234  * buffer.
1235  */
1236 void
1237 bpf_tap(void *arg, u_char *pkt, u_int pktlen)
1238 {
1239 	struct bpf_if *bp;
1240 	struct bpf_d *d;
1241 	u_int slen;
1242 	struct timespec ts;
1243 	int gottime=0;
1244 
1245 	/*
1246 	 * Note that the ipl does not have to be raised at this point.
1247 	 * The only problem that could arise here is that if two different
1248 	 * interfaces shared any data.  This is not the case.
1249 	 */
1250 	bp = arg;
1251 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1252 		++d->bd_rcount;
1253 		++bpf_gstats.bs_recv;
1254 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
1255 		if (slen != 0) {
1256 			if (!gottime) {
1257 				nanotime(&ts);
1258 				gottime = 1;
1259 			}
1260 			catchpacket(d, pkt, pktlen, slen, memcpy, &ts);
1261 		}
1262 	}
1263 }
1264 
1265 /*
1266  * Copy data from an mbuf chain into a buffer.  This code is derived
1267  * from m_copydata in sys/uipc_mbuf.c.
1268  */
1269 static void *
1270 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1271 {
1272 	const struct mbuf *m;
1273 	u_int count;
1274 	u_char *dst;
1275 
1276 	m = src_arg;
1277 	dst = dst_arg;
1278 	while (len > 0) {
1279 		if (m == NULL)
1280 			panic("bpf_mcpy");
1281 		count = min(m->m_len, len);
1282 		memcpy(dst, mtod(m, const void *), count);
1283 		m = m->m_next;
1284 		dst += count;
1285 		len -= count;
1286 	}
1287 	return dst_arg;
1288 }
1289 
1290 /*
1291  * Dispatch a packet to all the listeners on interface bp.
1292  *
1293  * marg    pointer to the packet, either a data buffer or an mbuf chain
1294  * buflen  buffer length, if marg is a data buffer
1295  * cpfn    a function that can copy marg into the listener's buffer
1296  * pktlen  length of the packet
1297  * rcvif   either NULL or the interface the packet came in on.
1298  */
1299 static inline void
1300 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1301 	    void *marg, u_int pktlen, u_int buflen, struct ifnet *rcvif)
1302 {
1303 	u_int slen;
1304 	struct bpf_d *d;
1305 	struct timespec ts;
1306 	int gottime = 0;
1307 
1308 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1309 		if (!d->bd_seesent && (rcvif == NULL))
1310 			continue;
1311 		++d->bd_rcount;
1312 		++bpf_gstats.bs_recv;
1313 		slen = bpf_filter(d->bd_filter, marg, pktlen, buflen);
1314 		if (slen != 0) {
1315 			if(!gottime) {
1316 				nanotime(&ts);
1317 				gottime = 1;
1318 			}
1319 			catchpacket(d, marg, pktlen, slen, cpfn, &ts);
1320 		}
1321 	}
1322 }
1323 
1324 /*
1325  * Incoming linkage from device drivers, when the head of the packet is in
1326  * a buffer, and the tail is in an mbuf chain.
1327  */
1328 void
1329 bpf_mtap2(void *arg, void *data, u_int dlen, struct mbuf *m)
1330 {
1331 	struct bpf_if *bp = arg;
1332 	u_int pktlen;
1333 	struct mbuf mb;
1334 
1335 	pktlen = m_length(m) + dlen;
1336 
1337 	/*
1338 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1339 	 * Note that we cut corners here; we only setup what's
1340 	 * absolutely needed--this mbuf should never go anywhere else.
1341 	 */
1342 	(void)memset(&mb, 0, sizeof(mb));
1343 	mb.m_next = m;
1344 	mb.m_data = data;
1345 	mb.m_len = dlen;
1346 
1347 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif);
1348 }
1349 
1350 /*
1351  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1352  */
1353 void
1354 bpf_mtap(void *arg, struct mbuf *m)
1355 {
1356 	void *(*cpfn)(void *, const void *, size_t);
1357 	struct bpf_if *bp = arg;
1358 	u_int pktlen, buflen;
1359 	void *marg;
1360 
1361 	pktlen = m_length(m);
1362 
1363 	if (pktlen == m->m_len) {
1364 		cpfn = (void *)memcpy;
1365 		marg = mtod(m, void *);
1366 		buflen = pktlen;
1367 	} else {
1368 /*###1299 [cc] warning: assignment from incompatible pointer type%%%*/
1369 		cpfn = bpf_mcpy;
1370 		marg = m;
1371 		buflen = 0;
1372 	}
1373 
1374 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif);
1375 }
1376 
1377 /*
1378  * We need to prepend the address family as
1379  * a four byte field.  Cons up a dummy header
1380  * to pacify bpf.  This is safe because bpf
1381  * will only read from the mbuf (i.e., it won't
1382  * try to free it or keep a pointer a to it).
1383  */
1384 void
1385 bpf_mtap_af(void *arg, uint32_t af, struct mbuf *m)
1386 {
1387 	struct mbuf m0;
1388 
1389 	m0.m_flags = 0;
1390 	m0.m_next = m;
1391 	m0.m_len = 4;
1392 	m0.m_data = (char *)&af;
1393 
1394 	bpf_mtap(arg, &m0);
1395 }
1396 
1397 void
1398 bpf_mtap_et(void *arg, uint16_t et, struct mbuf *m)
1399 {
1400 	struct mbuf m0;
1401 
1402 	m0.m_flags = 0;
1403 	m0.m_next = m;
1404 	m0.m_len = 14;
1405 	m0.m_data = m0.m_dat;
1406 
1407 	((uint32_t *)m0.m_data)[0] = 0;
1408 	((uint32_t *)m0.m_data)[1] = 0;
1409 	((uint32_t *)m0.m_data)[2] = 0;
1410 	((uint16_t *)m0.m_data)[6] = et;
1411 
1412 	bpf_mtap(arg, &m0);
1413 }
1414 
1415 #if NSL > 0 || NSTRIP > 0
1416 /*
1417  * Put the SLIP pseudo-"link header" in place.
1418  * Note this M_PREPEND() should never fail,
1419  * swince we know we always have enough space
1420  * in the input buffer.
1421  */
1422 void
1423 bpf_mtap_sl_in(void *arg, u_char *chdr, struct mbuf **m)
1424 {
1425 	int s;
1426 	u_char *hp;
1427 
1428 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1429 	if (*m == NULL)
1430 		return;
1431 
1432 	hp = mtod(*m, u_char *);
1433 	hp[SLX_DIR] = SLIPDIR_IN;
1434 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1435 
1436 	s = splnet();
1437 	bpf_mtap(arg, *m);
1438 	splx(s);
1439 
1440 	m_adj(*m, SLIP_HDRLEN);
1441 }
1442 
1443 /*
1444  * Put the SLIP pseudo-"link header" in
1445  * place.  The compressed header is now
1446  * at the beginning of the mbuf.
1447  */
1448 void
1449 bpf_mtap_sl_out(void *arg, u_char *chdr, struct mbuf *m)
1450 {
1451 	struct mbuf m0;
1452 	u_char *hp;
1453 	int s;
1454 
1455 	m0.m_flags = 0;
1456 	m0.m_next = m;
1457 	m0.m_data = m0.m_dat;
1458 	m0.m_len = SLIP_HDRLEN;
1459 
1460 	hp = mtod(&m0, u_char *);
1461 
1462 	hp[SLX_DIR] = SLIPDIR_OUT;
1463 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1464 
1465 	s = splnet();
1466 	bpf_mtap(arg, &m0);
1467 	splx(s);
1468 	m_freem(m);
1469 }
1470 #endif
1471 
1472 /*
1473  * Move the packet data from interface memory (pkt) into the
1474  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
1475  * otherwise 0.  "copy" is the routine called to do the actual data
1476  * transfer.  memcpy is passed in to copy contiguous chunks, while
1477  * bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1478  * pkt is really an mbuf.
1479  */
1480 static void
1481 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1482     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1483 {
1484 	struct bpf_hdr *hp;
1485 	int totlen, curlen;
1486 	int hdrlen = d->bd_bif->bif_hdrlen;
1487 	int do_wakeup = 0;
1488 
1489 	++d->bd_ccount;
1490 	++bpf_gstats.bs_capt;
1491 	/*
1492 	 * Figure out how many bytes to move.  If the packet is
1493 	 * greater or equal to the snapshot length, transfer that
1494 	 * much.  Otherwise, transfer the whole packet (unless
1495 	 * we hit the buffer size limit).
1496 	 */
1497 	totlen = hdrlen + min(snaplen, pktlen);
1498 	if (totlen > d->bd_bufsize)
1499 		totlen = d->bd_bufsize;
1500 
1501 	/*
1502 	 * Round up the end of the previous packet to the next longword.
1503 	 */
1504 	curlen = BPF_WORDALIGN(d->bd_slen);
1505 	if (curlen + totlen > d->bd_bufsize) {
1506 		/*
1507 		 * This packet will overflow the storage buffer.
1508 		 * Rotate the buffers if we can, then wakeup any
1509 		 * pending reads.
1510 		 */
1511 		if (d->bd_fbuf == 0) {
1512 			/*
1513 			 * We haven't completed the previous read yet,
1514 			 * so drop the packet.
1515 			 */
1516 			++d->bd_dcount;
1517 			++bpf_gstats.bs_drop;
1518 			return;
1519 		}
1520 		ROTATE_BUFFERS(d);
1521 		do_wakeup = 1;
1522 		curlen = 0;
1523 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1524 		/*
1525 		 * Immediate mode is set, or the read timeout has
1526 		 * already expired during a select call.  A packet
1527 		 * arrived, so the reader should be woken up.
1528 		 */
1529 		do_wakeup = 1;
1530 	}
1531 
1532 	/*
1533 	 * Append the bpf header.
1534 	 */
1535 	hp = (struct bpf_hdr *)((char *)d->bd_sbuf + curlen);
1536 	hp->bh_tstamp.tv_sec = ts->tv_sec;
1537 	hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1538 	hp->bh_datalen = pktlen;
1539 	hp->bh_hdrlen = hdrlen;
1540 	/*
1541 	 * Copy the packet data into the store buffer and update its length.
1542 	 */
1543 	(*cpfn)((u_char *)hp + hdrlen, pkt, (hp->bh_caplen = totlen - hdrlen));
1544 	d->bd_slen = curlen + totlen;
1545 
1546 	/*
1547 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1548 	 * will cause filt_bpfread() to be called with it adjusted.
1549 	 */
1550 	if (do_wakeup)
1551 		bpf_wakeup(d);
1552 }
1553 
1554 /*
1555  * Initialize all nonzero fields of a descriptor.
1556  */
1557 static int
1558 bpf_allocbufs(struct bpf_d *d)
1559 {
1560 
1561 	d->bd_fbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1562 	if (!d->bd_fbuf)
1563 		return (ENOBUFS);
1564 	d->bd_sbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1565 	if (!d->bd_sbuf) {
1566 		free(d->bd_fbuf, M_DEVBUF);
1567 		return (ENOBUFS);
1568 	}
1569 	d->bd_slen = 0;
1570 	d->bd_hlen = 0;
1571 	return (0);
1572 }
1573 
1574 /*
1575  * Free buffers currently in use by a descriptor.
1576  * Called on close.
1577  */
1578 static void
1579 bpf_freed(struct bpf_d *d)
1580 {
1581 	/*
1582 	 * We don't need to lock out interrupts since this descriptor has
1583 	 * been detached from its interface and it yet hasn't been marked
1584 	 * free.
1585 	 */
1586 	if (d->bd_sbuf != 0) {
1587 		free(d->bd_sbuf, M_DEVBUF);
1588 		if (d->bd_hbuf != 0)
1589 			free(d->bd_hbuf, M_DEVBUF);
1590 		if (d->bd_fbuf != 0)
1591 			free(d->bd_fbuf, M_DEVBUF);
1592 	}
1593 	if (d->bd_filter)
1594 		free(d->bd_filter, M_DEVBUF);
1595 }
1596 
1597 /*
1598  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
1599  * fixed size of the link header (variable length headers not yet supported).
1600  */
1601 void
1602 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1603 {
1604 
1605 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
1606 }
1607 
1608 /*
1609  * Attach additional dlt for a interface to bpf.  dlt is the link layer type;
1610  * hdrlen is the fixed size of the link header for the specified dlt
1611  * (variable length headers not yet supported).
1612  */
1613 void
1614 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, void *driverp)
1615 {
1616 	struct bpf_if *bp;
1617 	bp = malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1618 	if (bp == 0)
1619 		panic("bpfattach");
1620 
1621 	bp->bif_dlist = 0;
1622 	bp->bif_driverp = driverp;
1623 	bp->bif_ifp = ifp;
1624 	bp->bif_dlt = dlt;
1625 
1626 	bp->bif_next = bpf_iflist;
1627 	bpf_iflist = bp;
1628 
1629 	*bp->bif_driverp = 0;
1630 
1631 	/*
1632 	 * Compute the length of the bpf header.  This is not necessarily
1633 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1634 	 * that the network layer header begins on a longword boundary (for
1635 	 * performance reasons and to alleviate alignment restrictions).
1636 	 */
1637 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1638 
1639 #if 0
1640 	printf("bpf: %s attached\n", ifp->if_xname);
1641 #endif
1642 }
1643 
1644 /*
1645  * Remove an interface from bpf.
1646  */
1647 void
1648 bpfdetach(struct ifnet *ifp)
1649 {
1650 	struct bpf_if *bp, **pbp;
1651 	struct bpf_d *d;
1652 	int s;
1653 
1654 	/* Nuke the vnodes for any open instances */
1655 	LIST_FOREACH(d, &bpf_list, bd_list) {
1656 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
1657 			/*
1658 			 * Detach the descriptor from an interface now.
1659 			 * It will be free'ed later by close routine.
1660 			 */
1661 			s = splnet();
1662 			d->bd_promisc = 0;	/* we can't touch device. */
1663 			bpf_detachd(d);
1664 			splx(s);
1665 		}
1666 	}
1667 
1668   again:
1669 	for (bp = bpf_iflist, pbp = &bpf_iflist;
1670 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
1671 		if (bp->bif_ifp == ifp) {
1672 			*pbp = bp->bif_next;
1673 			free(bp, M_DEVBUF);
1674 			goto again;
1675 		}
1676 	}
1677 }
1678 
1679 /*
1680  * Change the data link type of a interface.
1681  */
1682 void
1683 bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1684 {
1685 	struct bpf_if *bp;
1686 
1687 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1688 		if ((void **)bp->bif_driverp == &ifp->if_bpf)
1689 			break;
1690 	}
1691 	if (bp == NULL)
1692 		panic("bpf_change_type");
1693 
1694 	bp->bif_dlt = dlt;
1695 
1696 	/*
1697 	 * Compute the length of the bpf header.  This is not necessarily
1698 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1699 	 * that the network layer header begins on a longword boundary (for
1700 	 * performance reasons and to alleviate alignment restrictions).
1701 	 */
1702 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1703 }
1704 
1705 /*
1706  * Get a list of available data link type of the interface.
1707  */
1708 static int
1709 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1710 {
1711 	int n, error;
1712 	struct ifnet *ifp;
1713 	struct bpf_if *bp;
1714 
1715 	ifp = d->bd_bif->bif_ifp;
1716 	n = 0;
1717 	error = 0;
1718 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1719 		if (bp->bif_ifp != ifp)
1720 			continue;
1721 		if (bfl->bfl_list != NULL) {
1722 			if (n >= bfl->bfl_len)
1723 				return ENOMEM;
1724 			error = copyout(&bp->bif_dlt,
1725 			    bfl->bfl_list + n, sizeof(u_int));
1726 		}
1727 		n++;
1728 	}
1729 	bfl->bfl_len = n;
1730 	return error;
1731 }
1732 
1733 /*
1734  * Set the data link type of a BPF instance.
1735  */
1736 static int
1737 bpf_setdlt(struct bpf_d *d, u_int dlt)
1738 {
1739 	int s, error, opromisc;
1740 	struct ifnet *ifp;
1741 	struct bpf_if *bp;
1742 
1743 	if (d->bd_bif->bif_dlt == dlt)
1744 		return 0;
1745 	ifp = d->bd_bif->bif_ifp;
1746 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1747 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1748 			break;
1749 	}
1750 	if (bp == NULL)
1751 		return EINVAL;
1752 	s = splnet();
1753 	opromisc = d->bd_promisc;
1754 	bpf_detachd(d);
1755 	bpf_attachd(d, bp);
1756 	reset_d(d);
1757 	if (opromisc) {
1758 		error = ifpromisc(bp->bif_ifp, 1);
1759 		if (error)
1760 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
1761 			    bp->bif_ifp->if_xname, error);
1762 		else
1763 			d->bd_promisc = 1;
1764 	}
1765 	splx(s);
1766 	return 0;
1767 }
1768 
1769 static int
1770 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
1771 {
1772 	int newsize, error;
1773 	struct sysctlnode node;
1774 
1775 	node = *rnode;
1776 	node.sysctl_data = &newsize;
1777 	newsize = bpf_maxbufsize;
1778 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1779 	if (error || newp == NULL)
1780 		return (error);
1781 
1782 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
1783 		return (EINVAL);
1784 
1785 	bpf_maxbufsize = newsize;
1786 
1787 	return (0);
1788 }
1789 
1790 static int
1791 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
1792 {
1793 	int    error, elem_count;
1794 	struct bpf_d	 *dp;
1795 	struct bpf_d_ext  dpe;
1796 	size_t len, needed, elem_size, out_size;
1797 	char   *sp;
1798 
1799 	if (namelen == 1 && name[0] == CTL_QUERY)
1800 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
1801 
1802 	if (namelen != 2)
1803 		return (EINVAL);
1804 
1805 	/* BPF peers is privileged information. */
1806 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
1807 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
1808 	if (error)
1809 		return (EPERM);
1810 
1811 	len = (oldp != NULL) ? *oldlenp : 0;
1812 	sp = oldp;
1813 	elem_size = name[0];
1814 	elem_count = name[1];
1815 	out_size = MIN(sizeof(dpe), elem_size);
1816 	needed = 0;
1817 
1818 	if (elem_size < 1 || elem_count < 0)
1819 		return (EINVAL);
1820 
1821 	mutex_enter(&bpf_mtx);
1822 	LIST_FOREACH(dp, &bpf_list, bd_list) {
1823 		if (len >= elem_size && elem_count > 0) {
1824 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
1825 			BPF_EXT(bufsize);
1826 			BPF_EXT(promisc);
1827 			BPF_EXT(promisc);
1828 			BPF_EXT(state);
1829 			BPF_EXT(immediate);
1830 			BPF_EXT(hdrcmplt);
1831 			BPF_EXT(seesent);
1832 			BPF_EXT(pid);
1833 			BPF_EXT(rcount);
1834 			BPF_EXT(dcount);
1835 			BPF_EXT(ccount);
1836 #undef BPF_EXT
1837 			if (dp->bd_bif)
1838 				(void)strlcpy(dpe.bde_ifname,
1839 				    dp->bd_bif->bif_ifp->if_xname,
1840 				    IFNAMSIZ - 1);
1841 			else
1842 				dpe.bde_ifname[0] = '\0';
1843 
1844 			error = copyout(&dpe, sp, out_size);
1845 			if (error)
1846 				break;
1847 			sp += elem_size;
1848 			len -= elem_size;
1849 		}
1850 		needed += elem_size;
1851 		if (elem_count > 0 && elem_count != INT_MAX)
1852 			elem_count--;
1853 	}
1854 	mutex_exit(&bpf_mtx);
1855 
1856 	*oldlenp = needed;
1857 
1858 	return (error);
1859 }
1860 
1861 SYSCTL_SETUP(sysctl_net_bpf_setup, "sysctl net.bpf subtree setup")
1862 {
1863 	const struct sysctlnode *node;
1864 
1865 	sysctl_createv(clog, 0, NULL, NULL,
1866 		       CTLFLAG_PERMANENT,
1867 		       CTLTYPE_NODE, "net", NULL,
1868 		       NULL, 0, NULL, 0,
1869 		       CTL_NET, CTL_EOL);
1870 
1871 	node = NULL;
1872 	sysctl_createv(clog, 0, NULL, &node,
1873 		       CTLFLAG_PERMANENT,
1874 		       CTLTYPE_NODE, "bpf",
1875 		       SYSCTL_DESCR("BPF options"),
1876 		       NULL, 0, NULL, 0,
1877 		       CTL_NET, CTL_CREATE, CTL_EOL);
1878 	if (node != NULL) {
1879 		sysctl_createv(clog, 0, NULL, NULL,
1880 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1881 			CTLTYPE_INT, "maxbufsize",
1882 			SYSCTL_DESCR("Maximum size for data capture buffer"),
1883 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
1884 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1885 		sysctl_createv(clog, 0, NULL, NULL,
1886 			CTLFLAG_PERMANENT,
1887 			CTLTYPE_STRUCT, "stats",
1888 			SYSCTL_DESCR("BPF stats"),
1889 			NULL, 0, &bpf_gstats, sizeof(bpf_gstats),
1890 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1891 		sysctl_createv(clog, 0, NULL, NULL,
1892 			CTLFLAG_PERMANENT,
1893 			CTLTYPE_STRUCT, "peers",
1894 			SYSCTL_DESCR("BPF peers"),
1895 			sysctl_net_bpf_peers, 0, NULL, 0,
1896 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1897 	}
1898 
1899 }
1900