xref: /netbsd-src/sys/net/bpf.c (revision a6f3f22f245acb8ee3bbf6871d7dce989204fa97)
1 /*	$NetBSD: bpf.c,v 1.192 2015/10/14 19:40:09 christos 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 1.192 2015/10/14 19:40:09 christos 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/ioctl.h>
57 #include <sys/conf.h>
58 #include <sys/vnode.h>
59 #include <sys/queue.h>
60 #include <sys/stat.h>
61 #include <sys/module.h>
62 #include <sys/once.h>
63 #include <sys/atomic.h>
64 
65 #include <sys/file.h>
66 #include <sys/filedesc.h>
67 #include <sys/tty.h>
68 #include <sys/uio.h>
69 
70 #include <sys/protosw.h>
71 #include <sys/socket.h>
72 #include <sys/errno.h>
73 #include <sys/kernel.h>
74 #include <sys/poll.h>
75 #include <sys/sysctl.h>
76 #include <sys/kauth.h>
77 
78 #include <net/if.h>
79 #include <net/slip.h>
80 
81 #include <net/bpf.h>
82 #include <net/bpfdesc.h>
83 #include <net/bpfjit.h>
84 
85 #include <net/if_arc.h>
86 #include <net/if_ether.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/if_inarp.h>
90 
91 
92 #include <compat/sys/sockio.h>
93 
94 #ifndef BPF_BUFSIZE
95 /*
96  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
97  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
98  */
99 # define BPF_BUFSIZE 32768
100 #endif
101 
102 #define PRINET  26			/* interruptible */
103 
104 /*
105  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
106  * XXX the default values should be computed dynamically based
107  * on available memory size and available mbuf clusters.
108  */
109 int bpf_bufsize = BPF_BUFSIZE;
110 int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
111 bool bpf_jit = false;
112 
113 struct bpfjit_ops bpfjit_module_ops = {
114 	.bj_generate_code = NULL,
115 	.bj_free_code = NULL
116 };
117 
118 /*
119  * Global BPF statistics returned by net.bpf.stats sysctl.
120  */
121 struct bpf_stat	bpf_gstats;
122 
123 /*
124  * Use a mutex to avoid a race condition between gathering the stats/peers
125  * and opening/closing the device.
126  */
127 static kmutex_t bpf_mtx;
128 
129 /*
130  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
131  *  bpf_dtab holds the descriptors, indexed by minor device #
132  */
133 struct bpf_if	*bpf_iflist;
134 LIST_HEAD(, bpf_d) bpf_list;
135 
136 static int	bpf_allocbufs(struct bpf_d *);
137 static void	bpf_deliver(struct bpf_if *,
138 		            void *(*cpfn)(void *, const void *, size_t),
139 		            void *, u_int, u_int, const bool);
140 static void	bpf_freed(struct bpf_d *);
141 static void	bpf_ifname(struct ifnet *, struct ifreq *);
142 static void	*bpf_mcpy(void *, const void *, size_t);
143 static int	bpf_movein(struct uio *, int, uint64_t,
144 			        struct mbuf **, struct sockaddr *);
145 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
146 static void	bpf_detachd(struct bpf_d *);
147 static int	bpf_setif(struct bpf_d *, struct ifreq *);
148 static void	bpf_timed_out(void *);
149 static inline void
150 		bpf_wakeup(struct bpf_d *);
151 static int	bpf_hdrlen(struct bpf_d *);
152 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
153     void *(*)(void *, const void *, size_t), struct timespec *);
154 static void	reset_d(struct bpf_d *);
155 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
156 static int	bpf_setdlt(struct bpf_d *, u_int);
157 
158 static int	bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t,
159     int);
160 static int	bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t,
161     int);
162 static int	bpf_ioctl(struct file *, u_long, void *);
163 static int	bpf_poll(struct file *, int);
164 static int	bpf_stat(struct file *, struct stat *);
165 static int	bpf_close(struct file *);
166 static int	bpf_kqfilter(struct file *, struct knote *);
167 static void	bpf_softintr(void *);
168 
169 static const struct fileops bpf_fileops = {
170 	.fo_read = bpf_read,
171 	.fo_write = bpf_write,
172 	.fo_ioctl = bpf_ioctl,
173 	.fo_fcntl = fnullop_fcntl,
174 	.fo_poll = bpf_poll,
175 	.fo_stat = bpf_stat,
176 	.fo_close = bpf_close,
177 	.fo_kqfilter = bpf_kqfilter,
178 	.fo_restart = fnullop_restart,
179 };
180 
181 dev_type_open(bpfopen);
182 
183 const struct cdevsw bpf_cdevsw = {
184 	.d_open = bpfopen,
185 	.d_close = noclose,
186 	.d_read = noread,
187 	.d_write = nowrite,
188 	.d_ioctl = noioctl,
189 	.d_stop = nostop,
190 	.d_tty = notty,
191 	.d_poll = nopoll,
192 	.d_mmap = nommap,
193 	.d_kqfilter = nokqfilter,
194 	.d_discard = nodiscard,
195 	.d_flag = D_OTHER
196 };
197 
198 bpfjit_func_t
199 bpf_jit_generate(bpf_ctx_t *bc, void *code, size_t size)
200 {
201 
202 	membar_consumer();
203 	if (bpfjit_module_ops.bj_generate_code != NULL) {
204 		return bpfjit_module_ops.bj_generate_code(bc, code, size);
205 	}
206 	return NULL;
207 }
208 
209 void
210 bpf_jit_freecode(bpfjit_func_t jcode)
211 {
212 	KASSERT(bpfjit_module_ops.bj_free_code != NULL);
213 	bpfjit_module_ops.bj_free_code(jcode);
214 }
215 
216 static int
217 bpf_movein(struct uio *uio, int linktype, uint64_t mtu, struct mbuf **mp,
218 	   struct sockaddr *sockp)
219 {
220 	struct mbuf *m;
221 	int error;
222 	size_t len;
223 	size_t hlen;
224 	size_t align;
225 
226 	/*
227 	 * Build a sockaddr based on the data link layer type.
228 	 * We do this at this level because the ethernet header
229 	 * is copied directly into the data field of the sockaddr.
230 	 * In the case of SLIP, there is no header and the packet
231 	 * is forwarded as is.
232 	 * Also, we are careful to leave room at the front of the mbuf
233 	 * for the link level header.
234 	 */
235 	switch (linktype) {
236 
237 	case DLT_SLIP:
238 		sockp->sa_family = AF_INET;
239 		hlen = 0;
240 		align = 0;
241 		break;
242 
243 	case DLT_PPP:
244 		sockp->sa_family = AF_UNSPEC;
245 		hlen = 0;
246 		align = 0;
247 		break;
248 
249 	case DLT_EN10MB:
250 		sockp->sa_family = AF_UNSPEC;
251 		/* XXX Would MAXLINKHDR be better? */
252  		/* 6(dst)+6(src)+2(type) */
253 		hlen = sizeof(struct ether_header);
254 		align = 2;
255 		break;
256 
257 	case DLT_ARCNET:
258 		sockp->sa_family = AF_UNSPEC;
259 		hlen = ARC_HDRLEN;
260 		align = 5;
261 		break;
262 
263 	case DLT_FDDI:
264 		sockp->sa_family = AF_LINK;
265 		/* XXX 4(FORMAC)+6(dst)+6(src) */
266 		hlen = 16;
267 		align = 0;
268 		break;
269 
270 	case DLT_ECONET:
271 		sockp->sa_family = AF_UNSPEC;
272 		hlen = 6;
273 		align = 2;
274 		break;
275 
276 	case DLT_NULL:
277 		sockp->sa_family = AF_UNSPEC;
278 		hlen = 0;
279 		align = 0;
280 		break;
281 
282 	default:
283 		return (EIO);
284 	}
285 
286 	len = uio->uio_resid;
287 	/*
288 	 * If there aren't enough bytes for a link level header or the
289 	 * packet length exceeds the interface mtu, return an error.
290 	 */
291 	if (len - hlen > mtu)
292 		return (EMSGSIZE);
293 
294 	/*
295 	 * XXX Avoid complicated buffer chaining ---
296 	 * bail if it won't fit in a single mbuf.
297 	 * (Take into account possible alignment bytes)
298 	 */
299 	if (len + align > MCLBYTES)
300 		return (EIO);
301 
302 	m = m_gethdr(M_WAIT, MT_DATA);
303 	m->m_pkthdr.rcvif = NULL;
304 	m->m_pkthdr.len = (int)(len - hlen);
305 	if (len + align > MHLEN) {
306 		m_clget(m, M_WAIT);
307 		if ((m->m_flags & M_EXT) == 0) {
308 			error = ENOBUFS;
309 			goto bad;
310 		}
311 	}
312 
313 	/* Insure the data is properly aligned */
314 	if (align > 0) {
315 		m->m_data += align;
316 		m->m_len -= (int)align;
317 	}
318 
319 	error = uiomove(mtod(m, void *), len, uio);
320 	if (error)
321 		goto bad;
322 	if (hlen != 0) {
323 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
324 		m->m_data += hlen; /* XXX */
325 		len -= hlen;
326 	}
327 	m->m_len = (int)len;
328 	*mp = m;
329 	return (0);
330 
331 bad:
332 	m_freem(m);
333 	return (error);
334 }
335 
336 /*
337  * Attach file to the bpf interface, i.e. make d listen on bp.
338  * Must be called at splnet.
339  */
340 static void
341 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
342 {
343 	KASSERT(mutex_owned(&bpf_mtx));
344 	/*
345 	 * Point d at bp, and add d to the interface's list of listeners.
346 	 * Finally, point the driver's bpf cookie at the interface so
347 	 * it will divert packets to bpf.
348 	 */
349 	d->bd_bif = bp;
350 	d->bd_next = bp->bif_dlist;
351 	bp->bif_dlist = d;
352 
353 	*bp->bif_driverp = bp;
354 }
355 
356 /*
357  * Detach a file from its interface.
358  */
359 static void
360 bpf_detachd(struct bpf_d *d)
361 {
362 	struct bpf_d **p;
363 	struct bpf_if *bp;
364 
365 	KASSERT(mutex_owned(&bpf_mtx));
366 
367 	bp = d->bd_bif;
368 	/*
369 	 * Check if this descriptor had requested promiscuous mode.
370 	 * If so, turn it off.
371 	 */
372 	if (d->bd_promisc) {
373 		int error __diagused;
374 
375 		d->bd_promisc = 0;
376 		/*
377 		 * Take device out of promiscuous mode.  Since we were
378 		 * able to enter promiscuous mode, we should be able
379 		 * to turn it off.  But we can get an error if
380 		 * the interface was configured down, so only panic
381 		 * if we don't get an unexpected error.
382 		 */
383   		error = ifpromisc(bp->bif_ifp, 0);
384 #ifdef DIAGNOSTIC
385 		if (error)
386 			printf("%s: ifpromisc failed: %d", __func__, error);
387 #endif
388 	}
389 	/* Remove d from the interface's descriptor list. */
390 	p = &bp->bif_dlist;
391 	while (*p != d) {
392 		p = &(*p)->bd_next;
393 		if (*p == NULL)
394 			panic("%s: descriptor not in list", __func__);
395 	}
396 	*p = (*p)->bd_next;
397 	if (bp->bif_dlist == NULL)
398 		/*
399 		 * Let the driver know that there are no more listeners.
400 		 */
401 		*d->bd_bif->bif_driverp = NULL;
402 	d->bd_bif = NULL;
403 }
404 
405 static int
406 doinit(void)
407 {
408 
409 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
410 
411 	LIST_INIT(&bpf_list);
412 
413 	bpf_gstats.bs_recv = 0;
414 	bpf_gstats.bs_drop = 0;
415 	bpf_gstats.bs_capt = 0;
416 
417 	return 0;
418 }
419 
420 /*
421  * bpfilterattach() is called at boot time.
422  */
423 /* ARGSUSED */
424 void
425 bpfilterattach(int n)
426 {
427 	static ONCE_DECL(control);
428 
429 	RUN_ONCE(&control, doinit);
430 }
431 
432 /*
433  * Open ethernet device. Clones.
434  */
435 /* ARGSUSED */
436 int
437 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
438 {
439 	struct bpf_d *d;
440 	struct file *fp;
441 	int error, fd;
442 
443 	/* falloc() will fill in the descriptor for us. */
444 	if ((error = fd_allocfile(&fp, &fd)) != 0)
445 		return error;
446 
447 	d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK|M_ZERO);
448 	d->bd_bufsize = bpf_bufsize;
449 	d->bd_seesent = 1;
450 	d->bd_feedback = 0;
451 	d->bd_pid = l->l_proc->p_pid;
452 #ifdef _LP64
453 	if (curproc->p_flag & PK_32)
454 		d->bd_compat32 = 1;
455 #endif
456 	getnanotime(&d->bd_btime);
457 	d->bd_atime = d->bd_mtime = d->bd_btime;
458 	callout_init(&d->bd_callout, 0);
459 	selinit(&d->bd_sel);
460 	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
461 	d->bd_jitcode = NULL;
462 
463 	mutex_enter(&bpf_mtx);
464 	LIST_INSERT_HEAD(&bpf_list, d, bd_list);
465 	mutex_exit(&bpf_mtx);
466 
467 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
468 }
469 
470 /*
471  * Close the descriptor by detaching it from its interface,
472  * deallocating its buffers, and marking it free.
473  */
474 /* ARGSUSED */
475 static int
476 bpf_close(struct file *fp)
477 {
478 	struct bpf_d *d = fp->f_bpf;
479 	int s;
480 
481 	KERNEL_LOCK(1, NULL);
482 	mutex_enter(&bpf_mtx);
483 
484 	/*
485 	 * Refresh the PID associated with this bpf file.
486 	 */
487 	d->bd_pid = curproc->p_pid;
488 
489 	s = splnet();
490 	if (d->bd_state == BPF_WAITING)
491 		callout_stop(&d->bd_callout);
492 	d->bd_state = BPF_IDLE;
493 	if (d->bd_bif)
494 		bpf_detachd(d);
495 	splx(s);
496 	bpf_freed(d);
497 	LIST_REMOVE(d, bd_list);
498 	callout_destroy(&d->bd_callout);
499 	seldestroy(&d->bd_sel);
500 	softint_disestablish(d->bd_sih);
501 	free(d, M_DEVBUF);
502 	fp->f_bpf = NULL;
503 
504 	mutex_exit(&bpf_mtx);
505 	KERNEL_UNLOCK_ONE(NULL);
506 
507 	return (0);
508 }
509 
510 /*
511  * Rotate the packet buffers in descriptor d.  Move the store buffer
512  * into the hold slot, and the free buffer into the store slot.
513  * Zero the length of the new store buffer.
514  */
515 #define ROTATE_BUFFERS(d) \
516 	(d)->bd_hbuf = (d)->bd_sbuf; \
517 	(d)->bd_hlen = (d)->bd_slen; \
518 	(d)->bd_sbuf = (d)->bd_fbuf; \
519 	(d)->bd_slen = 0; \
520 	(d)->bd_fbuf = NULL;
521 /*
522  *  bpfread - read next chunk of packets from buffers
523  */
524 static int
525 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
526     kauth_cred_t cred, int flags)
527 {
528 	struct bpf_d *d = fp->f_bpf;
529 	int timed_out;
530 	int error;
531 	int s;
532 
533 	getnanotime(&d->bd_atime);
534 	/*
535 	 * Restrict application to use a buffer the same size as
536 	 * the kernel buffers.
537 	 */
538 	if (uio->uio_resid != d->bd_bufsize)
539 		return (EINVAL);
540 
541 	KERNEL_LOCK(1, NULL);
542 	s = splnet();
543 	if (d->bd_state == BPF_WAITING)
544 		callout_stop(&d->bd_callout);
545 	timed_out = (d->bd_state == BPF_TIMED_OUT);
546 	d->bd_state = BPF_IDLE;
547 	/*
548 	 * If the hold buffer is empty, then do a timed sleep, which
549 	 * ends when the timeout expires or when enough packets
550 	 * have arrived to fill the store buffer.
551 	 */
552 	while (d->bd_hbuf == NULL) {
553 		if (fp->f_flag & FNONBLOCK) {
554 			if (d->bd_slen == 0) {
555 				splx(s);
556 				KERNEL_UNLOCK_ONE(NULL);
557 				return (EWOULDBLOCK);
558 			}
559 			ROTATE_BUFFERS(d);
560 			break;
561 		}
562 
563 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
564 			/*
565 			 * A packet(s) either arrived since the previous
566 			 * read or arrived while we were asleep.
567 			 * Rotate the buffers and return what's here.
568 			 */
569 			ROTATE_BUFFERS(d);
570 			break;
571 		}
572 		error = tsleep(d, PRINET|PCATCH, "bpf",
573 				d->bd_rtout);
574 		if (error == EINTR || error == ERESTART) {
575 			splx(s);
576 			KERNEL_UNLOCK_ONE(NULL);
577 			return (error);
578 		}
579 		if (error == EWOULDBLOCK) {
580 			/*
581 			 * On a timeout, return what's in the buffer,
582 			 * which may be nothing.  If there is something
583 			 * in the store buffer, we can rotate the buffers.
584 			 */
585 			if (d->bd_hbuf)
586 				/*
587 				 * We filled up the buffer in between
588 				 * getting the timeout and arriving
589 				 * here, so we don't need to rotate.
590 				 */
591 				break;
592 
593 			if (d->bd_slen == 0) {
594 				splx(s);
595 				KERNEL_UNLOCK_ONE(NULL);
596 				return (0);
597 			}
598 			ROTATE_BUFFERS(d);
599 			break;
600 		}
601 		if (error != 0)
602 			goto done;
603 	}
604 	/*
605 	 * At this point, we know we have something in the hold slot.
606 	 */
607 	splx(s);
608 
609 	/*
610 	 * Move data from hold buffer into user space.
611 	 * We know the entire buffer is transferred since
612 	 * we checked above that the read buffer is bpf_bufsize bytes.
613 	 */
614 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
615 
616 	s = splnet();
617 	d->bd_fbuf = d->bd_hbuf;
618 	d->bd_hbuf = NULL;
619 	d->bd_hlen = 0;
620 done:
621 	splx(s);
622 	KERNEL_UNLOCK_ONE(NULL);
623 	return (error);
624 }
625 
626 
627 /*
628  * If there are processes sleeping on this descriptor, wake them up.
629  */
630 static inline void
631 bpf_wakeup(struct bpf_d *d)
632 {
633 	wakeup(d);
634 	if (d->bd_async)
635 		softint_schedule(d->bd_sih);
636 	selnotify(&d->bd_sel, 0, 0);
637 }
638 
639 static void
640 bpf_softintr(void *cookie)
641 {
642 	struct bpf_d *d;
643 
644 	d = cookie;
645 	if (d->bd_async)
646 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
647 }
648 
649 static void
650 bpf_timed_out(void *arg)
651 {
652 	struct bpf_d *d = arg;
653 	int s;
654 
655 	s = splnet();
656 	if (d->bd_state == BPF_WAITING) {
657 		d->bd_state = BPF_TIMED_OUT;
658 		if (d->bd_slen != 0)
659 			bpf_wakeup(d);
660 	}
661 	splx(s);
662 }
663 
664 
665 static int
666 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
667     kauth_cred_t cred, int flags)
668 {
669 	struct bpf_d *d = fp->f_bpf;
670 	struct ifnet *ifp;
671 	struct mbuf *m, *mc;
672 	int error, s;
673 	static struct sockaddr_storage dst;
674 
675 	m = NULL;	/* XXX gcc */
676 
677 	KERNEL_LOCK(1, NULL);
678 
679 	if (d->bd_bif == NULL) {
680 		KERNEL_UNLOCK_ONE(NULL);
681 		return (ENXIO);
682 	}
683 	getnanotime(&d->bd_mtime);
684 
685 	ifp = d->bd_bif->bif_ifp;
686 
687 	if (uio->uio_resid == 0) {
688 		KERNEL_UNLOCK_ONE(NULL);
689 		return (0);
690 	}
691 
692 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
693 		(struct sockaddr *) &dst);
694 	if (error) {
695 		KERNEL_UNLOCK_ONE(NULL);
696 		return (error);
697 	}
698 
699 	if (m->m_pkthdr.len > ifp->if_mtu) {
700 		KERNEL_UNLOCK_ONE(NULL);
701 		m_freem(m);
702 		return (EMSGSIZE);
703 	}
704 
705 	if (d->bd_hdrcmplt)
706 		dst.ss_family = pseudo_AF_HDRCMPLT;
707 
708 	if (d->bd_feedback) {
709 		mc = m_dup(m, 0, M_COPYALL, M_NOWAIT);
710 		if (mc != NULL)
711 			mc->m_pkthdr.rcvif = ifp;
712 		/* Set M_PROMISC for outgoing packets to be discarded. */
713 		if (1 /*d->bd_direction == BPF_D_INOUT*/)
714 			m->m_flags |= M_PROMISC;
715 	} else
716 		mc = NULL;
717 
718 	s = splsoftnet();
719 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *) &dst, NULL);
720 
721 	if (mc != NULL) {
722 		if (error == 0)
723 			(*ifp->if_input)(ifp, mc);
724 		m_freem(mc);
725 	}
726 	splx(s);
727 	KERNEL_UNLOCK_ONE(NULL);
728 	/*
729 	 * The driver frees the mbuf.
730 	 */
731 	return (error);
732 }
733 
734 /*
735  * Reset a descriptor by flushing its packet buffer and clearing the
736  * receive and drop counts.  Should be called at splnet.
737  */
738 static void
739 reset_d(struct bpf_d *d)
740 {
741 	if (d->bd_hbuf) {
742 		/* Free the hold buffer. */
743 		d->bd_fbuf = d->bd_hbuf;
744 		d->bd_hbuf = NULL;
745 	}
746 	d->bd_slen = 0;
747 	d->bd_hlen = 0;
748 	d->bd_rcount = 0;
749 	d->bd_dcount = 0;
750 	d->bd_ccount = 0;
751 }
752 
753 /*
754  *  FIONREAD		Check for read packet available.
755  *  BIOCGBLEN		Get buffer len [for read()].
756  *  BIOCSETF		Set ethernet read filter.
757  *  BIOCFLUSH		Flush read packet buffer.
758  *  BIOCPROMISC		Put interface into promiscuous mode.
759  *  BIOCGDLT		Get link layer type.
760  *  BIOCGETIF		Get interface name.
761  *  BIOCSETIF		Set interface.
762  *  BIOCSRTIMEOUT	Set read timeout.
763  *  BIOCGRTIMEOUT	Get read timeout.
764  *  BIOCGSTATS		Get packet stats.
765  *  BIOCIMMEDIATE	Set immediate mode.
766  *  BIOCVERSION		Get filter language version.
767  *  BIOCGHDRCMPLT	Get "header already complete" flag.
768  *  BIOCSHDRCMPLT	Set "header already complete" flag.
769  *  BIOCSFEEDBACK	Set packet feedback mode.
770  *  BIOCGFEEDBACK	Get packet feedback mode.
771  *  BIOCGSEESENT  	Get "see sent packets" mode.
772  *  BIOCSSEESENT  	Set "see sent packets" mode.
773  */
774 /* ARGSUSED */
775 static int
776 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
777 {
778 	struct bpf_d *d = fp->f_bpf;
779 	int s, error = 0;
780 
781 	/*
782 	 * Refresh the PID associated with this bpf file.
783 	 */
784 	KERNEL_LOCK(1, NULL);
785 	d->bd_pid = curproc->p_pid;
786 #ifdef _LP64
787 	if (curproc->p_flag & PK_32)
788 		d->bd_compat32 = 1;
789 	else
790 		d->bd_compat32 = 0;
791 #endif
792 
793 	s = splnet();
794 	if (d->bd_state == BPF_WAITING)
795 		callout_stop(&d->bd_callout);
796 	d->bd_state = BPF_IDLE;
797 	splx(s);
798 
799 	switch (cmd) {
800 
801 	default:
802 		error = EINVAL;
803 		break;
804 
805 	/*
806 	 * Check for read packet available.
807 	 */
808 	case FIONREAD:
809 		{
810 			int n;
811 
812 			s = splnet();
813 			n = d->bd_slen;
814 			if (d->bd_hbuf)
815 				n += d->bd_hlen;
816 			splx(s);
817 
818 			*(int *)addr = n;
819 			break;
820 		}
821 
822 	/*
823 	 * Get buffer len [for read()].
824 	 */
825 	case BIOCGBLEN:
826 		*(u_int *)addr = d->bd_bufsize;
827 		break;
828 
829 	/*
830 	 * Set buffer length.
831 	 */
832 	case BIOCSBLEN:
833 		if (d->bd_bif != NULL)
834 			error = EINVAL;
835 		else {
836 			u_int size = *(u_int *)addr;
837 
838 			if (size > bpf_maxbufsize)
839 				*(u_int *)addr = size = bpf_maxbufsize;
840 			else if (size < BPF_MINBUFSIZE)
841 				*(u_int *)addr = size = BPF_MINBUFSIZE;
842 			d->bd_bufsize = size;
843 		}
844 		break;
845 
846 	/*
847 	 * Set link layer read filter.
848 	 */
849 	case BIOCSETF:
850 		error = bpf_setf(d, addr);
851 		break;
852 
853 	/*
854 	 * Flush read packet buffer.
855 	 */
856 	case BIOCFLUSH:
857 		s = splnet();
858 		reset_d(d);
859 		splx(s);
860 		break;
861 
862 	/*
863 	 * Put interface into promiscuous mode.
864 	 */
865 	case BIOCPROMISC:
866 		if (d->bd_bif == NULL) {
867 			/*
868 			 * No interface attached yet.
869 			 */
870 			error = EINVAL;
871 			break;
872 		}
873 		s = splnet();
874 		if (d->bd_promisc == 0) {
875 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
876 			if (error == 0)
877 				d->bd_promisc = 1;
878 		}
879 		splx(s);
880 		break;
881 
882 	/*
883 	 * Get device parameters.
884 	 */
885 	case BIOCGDLT:
886 		if (d->bd_bif == NULL)
887 			error = EINVAL;
888 		else
889 			*(u_int *)addr = d->bd_bif->bif_dlt;
890 		break;
891 
892 	/*
893 	 * Get a list of supported device parameters.
894 	 */
895 	case BIOCGDLTLIST:
896 		if (d->bd_bif == NULL)
897 			error = EINVAL;
898 		else
899 			error = bpf_getdltlist(d, addr);
900 		break;
901 
902 	/*
903 	 * Set device parameters.
904 	 */
905 	case BIOCSDLT:
906 		mutex_enter(&bpf_mtx);
907 		if (d->bd_bif == NULL)
908 			error = EINVAL;
909 		else
910 			error = bpf_setdlt(d, *(u_int *)addr);
911 		mutex_exit(&bpf_mtx);
912 		break;
913 
914 	/*
915 	 * Set interface name.
916 	 */
917 #ifdef OBIOCGETIF
918 	case OBIOCGETIF:
919 #endif
920 	case BIOCGETIF:
921 		if (d->bd_bif == NULL)
922 			error = EINVAL;
923 		else
924 			bpf_ifname(d->bd_bif->bif_ifp, addr);
925 		break;
926 
927 	/*
928 	 * Set interface.
929 	 */
930 #ifdef OBIOCSETIF
931 	case OBIOCSETIF:
932 #endif
933 	case BIOCSETIF:
934 		mutex_enter(&bpf_mtx);
935 		error = bpf_setif(d, addr);
936 		mutex_exit(&bpf_mtx);
937 		break;
938 
939 	/*
940 	 * Set read timeout.
941 	 */
942 	case BIOCSRTIMEOUT:
943 		{
944 			struct timeval *tv = addr;
945 
946 			/* Compute number of ticks. */
947 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
948 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
949 				d->bd_rtout = 1;
950 			break;
951 		}
952 
953 #ifdef BIOCGORTIMEOUT
954 	/*
955 	 * Get read timeout.
956 	 */
957 	case BIOCGORTIMEOUT:
958 		{
959 			struct timeval50 *tv = addr;
960 
961 			tv->tv_sec = d->bd_rtout / hz;
962 			tv->tv_usec = (d->bd_rtout % hz) * tick;
963 			break;
964 		}
965 #endif
966 
967 #ifdef BIOCSORTIMEOUT
968 	/*
969 	 * Set read timeout.
970 	 */
971 	case BIOCSORTIMEOUT:
972 		{
973 			struct timeval50 *tv = addr;
974 
975 			/* Compute number of ticks. */
976 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
977 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
978 				d->bd_rtout = 1;
979 			break;
980 		}
981 #endif
982 
983 	/*
984 	 * Get read timeout.
985 	 */
986 	case BIOCGRTIMEOUT:
987 		{
988 			struct timeval *tv = addr;
989 
990 			tv->tv_sec = d->bd_rtout / hz;
991 			tv->tv_usec = (d->bd_rtout % hz) * tick;
992 			break;
993 		}
994 	/*
995 	 * Get packet stats.
996 	 */
997 	case BIOCGSTATS:
998 		{
999 			struct bpf_stat *bs = addr;
1000 
1001 			bs->bs_recv = d->bd_rcount;
1002 			bs->bs_drop = d->bd_dcount;
1003 			bs->bs_capt = d->bd_ccount;
1004 			break;
1005 		}
1006 
1007 	case BIOCGSTATSOLD:
1008 		{
1009 			struct bpf_stat_old *bs = addr;
1010 
1011 			bs->bs_recv = d->bd_rcount;
1012 			bs->bs_drop = d->bd_dcount;
1013 			break;
1014 		}
1015 
1016 	/*
1017 	 * Set immediate mode.
1018 	 */
1019 	case BIOCIMMEDIATE:
1020 		d->bd_immediate = *(u_int *)addr;
1021 		break;
1022 
1023 	case BIOCVERSION:
1024 		{
1025 			struct bpf_version *bv = addr;
1026 
1027 			bv->bv_major = BPF_MAJOR_VERSION;
1028 			bv->bv_minor = BPF_MINOR_VERSION;
1029 			break;
1030 		}
1031 
1032 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
1033 		*(u_int *)addr = d->bd_hdrcmplt;
1034 		break;
1035 
1036 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
1037 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1038 		break;
1039 
1040 	/*
1041 	 * Get "see sent packets" flag
1042 	 */
1043 	case BIOCGSEESENT:
1044 		*(u_int *)addr = d->bd_seesent;
1045 		break;
1046 
1047 	/*
1048 	 * Set "see sent" packets flag
1049 	 */
1050 	case BIOCSSEESENT:
1051 		d->bd_seesent = *(u_int *)addr;
1052 		break;
1053 
1054 	/*
1055 	 * Set "feed packets from bpf back to input" mode
1056 	 */
1057 	case BIOCSFEEDBACK:
1058 		d->bd_feedback = *(u_int *)addr;
1059 		break;
1060 
1061 	/*
1062 	 * Get "feed packets from bpf back to input" mode
1063 	 */
1064 	case BIOCGFEEDBACK:
1065 		*(u_int *)addr = d->bd_feedback;
1066 		break;
1067 
1068 	case FIONBIO:		/* Non-blocking I/O */
1069 		/*
1070 		 * No need to do anything special as we use IO_NDELAY in
1071 		 * bpfread() as an indication of whether or not to block
1072 		 * the read.
1073 		 */
1074 		break;
1075 
1076 	case FIOASYNC:		/* Send signal on receive packets */
1077 		d->bd_async = *(int *)addr;
1078 		break;
1079 
1080 	case TIOCSPGRP:		/* Process or group to send signals to */
1081 	case FIOSETOWN:
1082 		error = fsetown(&d->bd_pgid, cmd, addr);
1083 		break;
1084 
1085 	case TIOCGPGRP:
1086 	case FIOGETOWN:
1087 		error = fgetown(d->bd_pgid, cmd, addr);
1088 		break;
1089 	}
1090 	KERNEL_UNLOCK_ONE(NULL);
1091 	return (error);
1092 }
1093 
1094 /*
1095  * Set d's packet filter program to fp.  If this file already has a filter,
1096  * free it and replace it.  Returns EINVAL for bogus requests.
1097  */
1098 int
1099 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
1100 {
1101 	struct bpf_insn *fcode, *old;
1102 	bpfjit_func_t jcode, oldj;
1103 	size_t flen, size;
1104 	int s;
1105 
1106 	jcode = NULL;
1107 	flen = fp->bf_len;
1108 
1109 	if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) {
1110 		return EINVAL;
1111 	}
1112 
1113 	if (flen) {
1114 		/*
1115 		 * Allocate the buffer, copy the byte-code from
1116 		 * userspace and validate it.
1117 		 */
1118 		size = flen * sizeof(*fp->bf_insns);
1119 		fcode = malloc(size, M_DEVBUF, M_WAITOK);
1120 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1121 		    !bpf_validate(fcode, (int)flen)) {
1122 			free(fcode, M_DEVBUF);
1123 			return EINVAL;
1124 		}
1125 		membar_consumer();
1126 		if (bpf_jit)
1127 			jcode = bpf_jit_generate(NULL, fcode, flen);
1128 	} else {
1129 		fcode = NULL;
1130 	}
1131 
1132 	s = splnet();
1133 	old = d->bd_filter;
1134 	d->bd_filter = fcode;
1135 	oldj = d->bd_jitcode;
1136 	d->bd_jitcode = jcode;
1137 	reset_d(d);
1138 	splx(s);
1139 
1140 	if (old) {
1141 		free(old, M_DEVBUF);
1142 	}
1143 	if (oldj) {
1144 		bpf_jit_freecode(oldj);
1145 	}
1146 
1147 	return 0;
1148 }
1149 
1150 /*
1151  * Detach a file from its current interface (if attached at all) and attach
1152  * to the interface indicated by the name stored in ifr.
1153  * Return an errno or 0.
1154  */
1155 static int
1156 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1157 {
1158 	struct bpf_if *bp;
1159 	char *cp;
1160 	int unit_seen, i, s, error;
1161 
1162 	KASSERT(mutex_owned(&bpf_mtx));
1163 	/*
1164 	 * Make sure the provided name has a unit number, and default
1165 	 * it to '0' if not specified.
1166 	 * XXX This is ugly ... do this differently?
1167 	 */
1168 	unit_seen = 0;
1169 	cp = ifr->ifr_name;
1170 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1171 	while (*cp++)
1172 		if (*cp >= '0' && *cp <= '9')
1173 			unit_seen = 1;
1174 	if (!unit_seen) {
1175 		/* Make sure to leave room for the '\0'. */
1176 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1177 			if ((ifr->ifr_name[i] >= 'a' &&
1178 			     ifr->ifr_name[i] <= 'z') ||
1179 			    (ifr->ifr_name[i] >= 'A' &&
1180 			     ifr->ifr_name[i] <= 'Z'))
1181 				continue;
1182 			ifr->ifr_name[i] = '0';
1183 		}
1184 	}
1185 
1186 	/*
1187 	 * Look through attached interfaces for the named one.
1188 	 */
1189 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1190 		struct ifnet *ifp = bp->bif_ifp;
1191 
1192 		if (ifp == NULL ||
1193 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1194 			continue;
1195 		/* skip additional entry */
1196 		if (bp->bif_driverp != &ifp->if_bpf)
1197 			continue;
1198 		/*
1199 		 * We found the requested interface.
1200 		 * Allocate the packet buffers if we need to.
1201 		 * If we're already attached to requested interface,
1202 		 * just flush the buffer.
1203 		 */
1204 		if (d->bd_sbuf == NULL) {
1205 			error = bpf_allocbufs(d);
1206 			if (error != 0)
1207 				return (error);
1208 		}
1209 		s = splnet();
1210 		if (bp != d->bd_bif) {
1211 			if (d->bd_bif)
1212 				/*
1213 				 * Detach if attached to something else.
1214 				 */
1215 				bpf_detachd(d);
1216 
1217 			bpf_attachd(d, bp);
1218 		}
1219 		reset_d(d);
1220 		splx(s);
1221 		return (0);
1222 	}
1223 	/* Not found. */
1224 	return (ENXIO);
1225 }
1226 
1227 /*
1228  * Copy the interface name to the ifreq.
1229  */
1230 static void
1231 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1232 {
1233 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1234 }
1235 
1236 static int
1237 bpf_stat(struct file *fp, struct stat *st)
1238 {
1239 	struct bpf_d *d = fp->f_bpf;
1240 
1241 	(void)memset(st, 0, sizeof(*st));
1242 	KERNEL_LOCK(1, NULL);
1243 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1244 	st->st_atimespec = d->bd_atime;
1245 	st->st_mtimespec = d->bd_mtime;
1246 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1247 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1248 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1249 	st->st_mode = S_IFCHR;
1250 	KERNEL_UNLOCK_ONE(NULL);
1251 	return 0;
1252 }
1253 
1254 /*
1255  * Support for poll() system call
1256  *
1257  * Return true iff the specific operation will not block indefinitely - with
1258  * the assumption that it is safe to positively acknowledge a request for the
1259  * ability to write to the BPF device.
1260  * Otherwise, return false but make a note that a selnotify() must be done.
1261  */
1262 static int
1263 bpf_poll(struct file *fp, int events)
1264 {
1265 	struct bpf_d *d = fp->f_bpf;
1266 	int s = splnet();
1267 	int revents;
1268 
1269 	/*
1270 	 * Refresh the PID associated with this bpf file.
1271 	 */
1272 	KERNEL_LOCK(1, NULL);
1273 	d->bd_pid = curproc->p_pid;
1274 
1275 	revents = events & (POLLOUT | POLLWRNORM);
1276 	if (events & (POLLIN | POLLRDNORM)) {
1277 		/*
1278 		 * An imitation of the FIONREAD ioctl code.
1279 		 */
1280 		if (d->bd_hlen != 0 ||
1281 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1282 		     d->bd_slen != 0)) {
1283 			revents |= events & (POLLIN | POLLRDNORM);
1284 		} else {
1285 			selrecord(curlwp, &d->bd_sel);
1286 			/* Start the read timeout if necessary */
1287 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1288 				callout_reset(&d->bd_callout, d->bd_rtout,
1289 					      bpf_timed_out, d);
1290 				d->bd_state = BPF_WAITING;
1291 			}
1292 		}
1293 	}
1294 
1295 	KERNEL_UNLOCK_ONE(NULL);
1296 	splx(s);
1297 	return (revents);
1298 }
1299 
1300 static void
1301 filt_bpfrdetach(struct knote *kn)
1302 {
1303 	struct bpf_d *d = kn->kn_hook;
1304 	int s;
1305 
1306 	KERNEL_LOCK(1, NULL);
1307 	s = splnet();
1308 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1309 	splx(s);
1310 	KERNEL_UNLOCK_ONE(NULL);
1311 }
1312 
1313 static int
1314 filt_bpfread(struct knote *kn, long hint)
1315 {
1316 	struct bpf_d *d = kn->kn_hook;
1317 	int rv;
1318 
1319 	KERNEL_LOCK(1, NULL);
1320 	kn->kn_data = d->bd_hlen;
1321 	if (d->bd_immediate)
1322 		kn->kn_data += d->bd_slen;
1323 	rv = (kn->kn_data > 0);
1324 	KERNEL_UNLOCK_ONE(NULL);
1325 	return rv;
1326 }
1327 
1328 static const struct filterops bpfread_filtops =
1329 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
1330 
1331 static int
1332 bpf_kqfilter(struct file *fp, struct knote *kn)
1333 {
1334 	struct bpf_d *d = fp->f_bpf;
1335 	struct klist *klist;
1336 	int s;
1337 
1338 	KERNEL_LOCK(1, NULL);
1339 
1340 	switch (kn->kn_filter) {
1341 	case EVFILT_READ:
1342 		klist = &d->bd_sel.sel_klist;
1343 		kn->kn_fop = &bpfread_filtops;
1344 		break;
1345 
1346 	default:
1347 		KERNEL_UNLOCK_ONE(NULL);
1348 		return (EINVAL);
1349 	}
1350 
1351 	kn->kn_hook = d;
1352 
1353 	s = splnet();
1354 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1355 	splx(s);
1356 	KERNEL_UNLOCK_ONE(NULL);
1357 
1358 	return (0);
1359 }
1360 
1361 /*
1362  * Copy data from an mbuf chain into a buffer.  This code is derived
1363  * from m_copydata in sys/uipc_mbuf.c.
1364  */
1365 static void *
1366 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1367 {
1368 	const struct mbuf *m;
1369 	u_int count;
1370 	u_char *dst;
1371 
1372 	m = src_arg;
1373 	dst = dst_arg;
1374 	while (len > 0) {
1375 		if (m == NULL)
1376 			panic("bpf_mcpy");
1377 		count = min(m->m_len, len);
1378 		memcpy(dst, mtod(m, const void *), count);
1379 		m = m->m_next;
1380 		dst += count;
1381 		len -= count;
1382 	}
1383 	return dst_arg;
1384 }
1385 
1386 /*
1387  * Dispatch a packet to all the listeners on interface bp.
1388  *
1389  * pkt     pointer to the packet, either a data buffer or an mbuf chain
1390  * buflen  buffer length, if pkt is a data buffer
1391  * cpfn    a function that can copy pkt into the listener's buffer
1392  * pktlen  length of the packet
1393  * rcv     true if packet came in
1394  */
1395 static inline void
1396 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1397     void *pkt, u_int pktlen, u_int buflen, const bool rcv)
1398 {
1399 	uint32_t mem[BPF_MEMWORDS];
1400 	bpf_args_t args = {
1401 		.pkt = (const uint8_t *)pkt,
1402 		.wirelen = pktlen,
1403 		.buflen = buflen,
1404 		.mem = mem,
1405 		.arg = NULL
1406 	};
1407 	bool gottime = false;
1408 	struct timespec ts;
1409 
1410 	/*
1411 	 * Note that the IPL does not have to be raised at this point.
1412 	 * The only problem that could arise here is that if two different
1413 	 * interfaces shared any data.  This is not the case.
1414 	 */
1415 	for (struct bpf_d *d = bp->bif_dlist; d != NULL; d = d->bd_next) {
1416 		u_int slen;
1417 
1418 		if (!d->bd_seesent && !rcv) {
1419 			continue;
1420 		}
1421 		d->bd_rcount++;
1422 		bpf_gstats.bs_recv++;
1423 
1424 		if (d->bd_jitcode)
1425 			slen = d->bd_jitcode(NULL, &args);
1426 		else
1427 			slen = bpf_filter_ext(NULL, d->bd_filter, &args);
1428 
1429 		if (!slen) {
1430 			continue;
1431 		}
1432 		if (!gottime) {
1433 			gottime = true;
1434 			nanotime(&ts);
1435 		}
1436 		catchpacket(d, pkt, pktlen, slen, cpfn, &ts);
1437 	}
1438 }
1439 
1440 /*
1441  * Incoming linkage from device drivers.  Process the packet pkt, of length
1442  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1443  * by each process' filter, and if accepted, stashed into the corresponding
1444  * buffer.
1445  */
1446 static void
1447 _bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1448 {
1449 
1450 	bpf_deliver(bp, memcpy, pkt, pktlen, pktlen, true);
1451 }
1452 
1453 /*
1454  * Incoming linkage from device drivers, when the head of the packet is in
1455  * a buffer, and the tail is in an mbuf chain.
1456  */
1457 static void
1458 _bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1459 {
1460 	u_int pktlen;
1461 	struct mbuf mb;
1462 
1463 	/* Skip outgoing duplicate packets. */
1464 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1465 		m->m_flags &= ~M_PROMISC;
1466 		return;
1467 	}
1468 
1469 	pktlen = m_length(m) + dlen;
1470 
1471 	/*
1472 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1473 	 * Note that we cut corners here; we only setup what's
1474 	 * absolutely needed--this mbuf should never go anywhere else.
1475 	 */
1476 	(void)memset(&mb, 0, sizeof(mb));
1477 	mb.m_next = m;
1478 	mb.m_data = data;
1479 	mb.m_len = dlen;
1480 
1481 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif != NULL);
1482 }
1483 
1484 /*
1485  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1486  */
1487 static void
1488 _bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1489 {
1490 	void *(*cpfn)(void *, const void *, size_t);
1491 	u_int pktlen, buflen;
1492 	void *marg;
1493 
1494 	/* Skip outgoing duplicate packets. */
1495 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
1496 		m->m_flags &= ~M_PROMISC;
1497 		return;
1498 	}
1499 
1500 	pktlen = m_length(m);
1501 
1502 	if (pktlen == m->m_len) {
1503 		cpfn = (void *)memcpy;
1504 		marg = mtod(m, void *);
1505 		buflen = pktlen;
1506 	} else {
1507 		cpfn = bpf_mcpy;
1508 		marg = m;
1509 		buflen = 0;
1510 	}
1511 
1512 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif != NULL);
1513 }
1514 
1515 /*
1516  * We need to prepend the address family as
1517  * a four byte field.  Cons up a dummy header
1518  * to pacify bpf.  This is safe because bpf
1519  * will only read from the mbuf (i.e., it won't
1520  * try to free it or keep a pointer a to it).
1521  */
1522 static void
1523 _bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m)
1524 {
1525 	struct mbuf m0;
1526 
1527 	m0.m_flags = 0;
1528 	m0.m_next = m;
1529 	m0.m_len = 4;
1530 	m0.m_data = (char *)&af;
1531 
1532 	_bpf_mtap(bp, &m0);
1533 }
1534 
1535 /*
1536  * Put the SLIP pseudo-"link header" in place.
1537  * Note this M_PREPEND() should never fail,
1538  * swince we know we always have enough space
1539  * in the input buffer.
1540  */
1541 static void
1542 _bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m)
1543 {
1544 	int s;
1545 	u_char *hp;
1546 
1547 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1548 	if (*m == NULL)
1549 		return;
1550 
1551 	hp = mtod(*m, u_char *);
1552 	hp[SLX_DIR] = SLIPDIR_IN;
1553 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1554 
1555 	s = splnet();
1556 	_bpf_mtap(bp, *m);
1557 	splx(s);
1558 
1559 	m_adj(*m, SLIP_HDRLEN);
1560 }
1561 
1562 /*
1563  * Put the SLIP pseudo-"link header" in
1564  * place.  The compressed header is now
1565  * at the beginning of the mbuf.
1566  */
1567 static void
1568 _bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m)
1569 {
1570 	struct mbuf m0;
1571 	u_char *hp;
1572 	int s;
1573 
1574 	m0.m_flags = 0;
1575 	m0.m_next = m;
1576 	m0.m_data = m0.m_dat;
1577 	m0.m_len = SLIP_HDRLEN;
1578 
1579 	hp = mtod(&m0, u_char *);
1580 
1581 	hp[SLX_DIR] = SLIPDIR_OUT;
1582 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1583 
1584 	s = splnet();
1585 	_bpf_mtap(bp, &m0);
1586 	splx(s);
1587 	m_freem(m);
1588 }
1589 
1590 static int
1591 bpf_hdrlen(struct bpf_d *d)
1592 {
1593 	int hdrlen = d->bd_bif->bif_hdrlen;
1594 	/*
1595 	 * Compute the length of the bpf header.  This is not necessarily
1596 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1597 	 * that the network layer header begins on a longword boundary (for
1598 	 * performance reasons and to alleviate alignment restrictions).
1599 	 */
1600 #ifdef _LP64
1601 	if (d->bd_compat32)
1602 		return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen);
1603 	else
1604 #endif
1605 		return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen);
1606 }
1607 
1608 /*
1609  * Move the packet data from interface memory (pkt) into the
1610  * store buffer. Call the wakeup functions if it's time to wakeup
1611  * a listener (buffer full), "cpfn" is the routine called to do the
1612  * actual data transfer. memcpy is passed in to copy contiguous chunks,
1613  * while bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1614  * pkt is really an mbuf.
1615  */
1616 static void
1617 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1618     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1619 {
1620 	char *h;
1621 	int totlen, curlen, caplen;
1622 	int hdrlen = bpf_hdrlen(d);
1623 	int do_wakeup = 0;
1624 
1625 	++d->bd_ccount;
1626 	++bpf_gstats.bs_capt;
1627 	/*
1628 	 * Figure out how many bytes to move.  If the packet is
1629 	 * greater or equal to the snapshot length, transfer that
1630 	 * much.  Otherwise, transfer the whole packet (unless
1631 	 * we hit the buffer size limit).
1632 	 */
1633 	totlen = hdrlen + min(snaplen, pktlen);
1634 	if (totlen > d->bd_bufsize)
1635 		totlen = d->bd_bufsize;
1636 	/*
1637 	 * If we adjusted totlen to fit the bufsize, it could be that
1638 	 * totlen is smaller than hdrlen because of the link layer header.
1639 	 */
1640 	caplen = totlen - hdrlen;
1641 	if (caplen < 0)
1642 		caplen = 0;
1643 
1644 	/*
1645 	 * Round up the end of the previous packet to the next longword.
1646 	 */
1647 #ifdef _LP64
1648 	if (d->bd_compat32)
1649 		curlen = BPF_WORDALIGN32(d->bd_slen);
1650 	else
1651 #endif
1652 		curlen = BPF_WORDALIGN(d->bd_slen);
1653 	if (curlen + totlen > d->bd_bufsize) {
1654 		/*
1655 		 * This packet will overflow the storage buffer.
1656 		 * Rotate the buffers if we can, then wakeup any
1657 		 * pending reads.
1658 		 */
1659 		if (d->bd_fbuf == NULL) {
1660 			/*
1661 			 * We haven't completed the previous read yet,
1662 			 * so drop the packet.
1663 			 */
1664 			++d->bd_dcount;
1665 			++bpf_gstats.bs_drop;
1666 			return;
1667 		}
1668 		ROTATE_BUFFERS(d);
1669 		do_wakeup = 1;
1670 		curlen = 0;
1671 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1672 		/*
1673 		 * Immediate mode is set, or the read timeout has
1674 		 * already expired during a select call.  A packet
1675 		 * arrived, so the reader should be woken up.
1676 		 */
1677 		do_wakeup = 1;
1678 	}
1679 
1680 	/*
1681 	 * Append the bpf header.
1682 	 */
1683 	h = (char *)d->bd_sbuf + curlen;
1684 #ifdef _LP64
1685 	if (d->bd_compat32) {
1686 		struct bpf_hdr32 *hp32;
1687 
1688 		hp32 = (struct bpf_hdr32 *)h;
1689 		hp32->bh_tstamp.tv_sec = ts->tv_sec;
1690 		hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1691 		hp32->bh_datalen = pktlen;
1692 		hp32->bh_hdrlen = hdrlen;
1693 		hp32->bh_caplen = caplen;
1694 	} else
1695 #endif
1696 	{
1697 		struct bpf_hdr *hp;
1698 
1699 		hp = (struct bpf_hdr *)h;
1700 		hp->bh_tstamp.tv_sec = ts->tv_sec;
1701 		hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1702 		hp->bh_datalen = pktlen;
1703 		hp->bh_hdrlen = hdrlen;
1704 		hp->bh_caplen = caplen;
1705 	}
1706 
1707 	/*
1708 	 * Copy the packet data into the store buffer and update its length.
1709 	 */
1710 	(*cpfn)(h + hdrlen, pkt, caplen);
1711 	d->bd_slen = curlen + totlen;
1712 
1713 	/*
1714 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1715 	 * will cause filt_bpfread() to be called with it adjusted.
1716 	 */
1717 	if (do_wakeup)
1718 		bpf_wakeup(d);
1719 }
1720 
1721 /*
1722  * Initialize all nonzero fields of a descriptor.
1723  */
1724 static int
1725 bpf_allocbufs(struct bpf_d *d)
1726 {
1727 
1728 	d->bd_fbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1729 	if (!d->bd_fbuf)
1730 		return (ENOBUFS);
1731 	d->bd_sbuf = malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
1732 	if (!d->bd_sbuf) {
1733 		free(d->bd_fbuf, M_DEVBUF);
1734 		return (ENOBUFS);
1735 	}
1736 	d->bd_slen = 0;
1737 	d->bd_hlen = 0;
1738 	return (0);
1739 }
1740 
1741 /*
1742  * Free buffers currently in use by a descriptor.
1743  * Called on close.
1744  */
1745 static void
1746 bpf_freed(struct bpf_d *d)
1747 {
1748 	/*
1749 	 * We don't need to lock out interrupts since this descriptor has
1750 	 * been detached from its interface and it yet hasn't been marked
1751 	 * free.
1752 	 */
1753 	if (d->bd_sbuf != NULL) {
1754 		free(d->bd_sbuf, M_DEVBUF);
1755 		if (d->bd_hbuf != NULL)
1756 			free(d->bd_hbuf, M_DEVBUF);
1757 		if (d->bd_fbuf != NULL)
1758 			free(d->bd_fbuf, M_DEVBUF);
1759 	}
1760 	if (d->bd_filter)
1761 		free(d->bd_filter, M_DEVBUF);
1762 
1763 	if (d->bd_jitcode != NULL) {
1764 		bpf_jit_freecode(d->bd_jitcode);
1765 	}
1766 }
1767 
1768 /*
1769  * Attach an interface to bpf.  dlt is the link layer type;
1770  * hdrlen is the fixed size of the link header for the specified dlt
1771  * (variable length headers not yet supported).
1772  */
1773 static void
1774 _bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
1775 {
1776 	struct bpf_if *bp;
1777 	bp = malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1778 	if (bp == NULL)
1779 		panic("bpfattach");
1780 
1781 	mutex_enter(&bpf_mtx);
1782 	bp->bif_dlist = NULL;
1783 	bp->bif_driverp = driverp;
1784 	bp->bif_ifp = ifp;
1785 	bp->bif_dlt = dlt;
1786 
1787 	bp->bif_next = bpf_iflist;
1788 	bpf_iflist = bp;
1789 
1790 	*bp->bif_driverp = NULL;
1791 
1792 	bp->bif_hdrlen = hdrlen;
1793 	mutex_exit(&bpf_mtx);
1794 #if 0
1795 	printf("bpf: %s attached\n", ifp->if_xname);
1796 #endif
1797 }
1798 
1799 /*
1800  * Remove an interface from bpf.
1801  */
1802 static void
1803 _bpfdetach(struct ifnet *ifp)
1804 {
1805 	struct bpf_if *bp, **pbp;
1806 	struct bpf_d *d;
1807 	int s;
1808 
1809 	mutex_enter(&bpf_mtx);
1810 	/* Nuke the vnodes for any open instances */
1811 	LIST_FOREACH(d, &bpf_list, bd_list) {
1812 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
1813 			/*
1814 			 * Detach the descriptor from an interface now.
1815 			 * It will be free'ed later by close routine.
1816 			 */
1817 			s = splnet();
1818 			d->bd_promisc = 0;	/* we can't touch device. */
1819 			bpf_detachd(d);
1820 			splx(s);
1821 		}
1822 	}
1823 
1824   again:
1825 	for (bp = bpf_iflist, pbp = &bpf_iflist;
1826 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
1827 		if (bp->bif_ifp == ifp) {
1828 			*pbp = bp->bif_next;
1829 			free(bp, M_DEVBUF);
1830 			goto again;
1831 		}
1832 	}
1833 	mutex_exit(&bpf_mtx);
1834 }
1835 
1836 /*
1837  * Change the data link type of a interface.
1838  */
1839 static void
1840 _bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
1841 {
1842 	struct bpf_if *bp;
1843 
1844 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1845 		if (bp->bif_driverp == &ifp->if_bpf)
1846 			break;
1847 	}
1848 	if (bp == NULL)
1849 		panic("bpf_change_type");
1850 
1851 	bp->bif_dlt = dlt;
1852 
1853 	bp->bif_hdrlen = hdrlen;
1854 }
1855 
1856 /*
1857  * Get a list of available data link type of the interface.
1858  */
1859 static int
1860 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
1861 {
1862 	int n, error;
1863 	struct ifnet *ifp;
1864 	struct bpf_if *bp;
1865 
1866 	ifp = d->bd_bif->bif_ifp;
1867 	n = 0;
1868 	error = 0;
1869 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1870 		if (bp->bif_ifp != ifp)
1871 			continue;
1872 		if (bfl->bfl_list != NULL) {
1873 			if (n >= bfl->bfl_len)
1874 				return ENOMEM;
1875 			error = copyout(&bp->bif_dlt,
1876 			    bfl->bfl_list + n, sizeof(u_int));
1877 		}
1878 		n++;
1879 	}
1880 	bfl->bfl_len = n;
1881 	return error;
1882 }
1883 
1884 /*
1885  * Set the data link type of a BPF instance.
1886  */
1887 static int
1888 bpf_setdlt(struct bpf_d *d, u_int dlt)
1889 {
1890 	int s, error, opromisc;
1891 	struct ifnet *ifp;
1892 	struct bpf_if *bp;
1893 
1894 	KASSERT(mutex_owned(&bpf_mtx));
1895 
1896 	if (d->bd_bif->bif_dlt == dlt)
1897 		return 0;
1898 	ifp = d->bd_bif->bif_ifp;
1899 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
1900 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
1901 			break;
1902 	}
1903 	if (bp == NULL)
1904 		return EINVAL;
1905 	s = splnet();
1906 	opromisc = d->bd_promisc;
1907 	bpf_detachd(d);
1908 	bpf_attachd(d, bp);
1909 	reset_d(d);
1910 	if (opromisc) {
1911 		error = ifpromisc(bp->bif_ifp, 1);
1912 		if (error)
1913 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
1914 			    bp->bif_ifp->if_xname, error);
1915 		else
1916 			d->bd_promisc = 1;
1917 	}
1918 	splx(s);
1919 	return 0;
1920 }
1921 
1922 static int
1923 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
1924 {
1925 	int newsize, error;
1926 	struct sysctlnode node;
1927 
1928 	node = *rnode;
1929 	node.sysctl_data = &newsize;
1930 	newsize = bpf_maxbufsize;
1931 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1932 	if (error || newp == NULL)
1933 		return (error);
1934 
1935 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
1936 		return (EINVAL);
1937 
1938 	bpf_maxbufsize = newsize;
1939 
1940 	return (0);
1941 }
1942 
1943 #if defined(MODULAR) || defined(BPFJIT)
1944 static int
1945 sysctl_net_bpf_jit(SYSCTLFN_ARGS)
1946 {
1947 	bool newval;
1948 	int error;
1949 	struct sysctlnode node;
1950 
1951 	node = *rnode;
1952 	node.sysctl_data = &newval;
1953 	newval = bpf_jit;
1954 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
1955 	if (error != 0 || newp == NULL)
1956 		return error;
1957 
1958 	bpf_jit = newval;
1959 
1960 	/*
1961 	 * Do a full sync to publish new bpf_jit value and
1962 	 * update bpfjit_module_ops.bj_generate_code variable.
1963 	 */
1964 	membar_sync();
1965 
1966 	if (newval && bpfjit_module_ops.bj_generate_code == NULL) {
1967 		printf("JIT compilation is postponed "
1968 		    "until after bpfjit module is loaded\n");
1969 	}
1970 
1971 	return 0;
1972 }
1973 #endif
1974 
1975 static int
1976 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
1977 {
1978 	int    error, elem_count;
1979 	struct bpf_d	 *dp;
1980 	struct bpf_d_ext  dpe;
1981 	size_t len, needed, elem_size, out_size;
1982 	char   *sp;
1983 
1984 	if (namelen == 1 && name[0] == CTL_QUERY)
1985 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
1986 
1987 	if (namelen != 2)
1988 		return (EINVAL);
1989 
1990 	/* BPF peers is privileged information. */
1991 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
1992 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
1993 	if (error)
1994 		return (EPERM);
1995 
1996 	len = (oldp != NULL) ? *oldlenp : 0;
1997 	sp = oldp;
1998 	elem_size = name[0];
1999 	elem_count = name[1];
2000 	out_size = MIN(sizeof(dpe), elem_size);
2001 	needed = 0;
2002 
2003 	if (elem_size < 1 || elem_count < 0)
2004 		return (EINVAL);
2005 
2006 	mutex_enter(&bpf_mtx);
2007 	LIST_FOREACH(dp, &bpf_list, bd_list) {
2008 		if (len >= elem_size && elem_count > 0) {
2009 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
2010 			BPF_EXT(bufsize);
2011 			BPF_EXT(promisc);
2012 			BPF_EXT(state);
2013 			BPF_EXT(immediate);
2014 			BPF_EXT(hdrcmplt);
2015 			BPF_EXT(seesent);
2016 			BPF_EXT(pid);
2017 			BPF_EXT(rcount);
2018 			BPF_EXT(dcount);
2019 			BPF_EXT(ccount);
2020 #undef BPF_EXT
2021 			if (dp->bd_bif)
2022 				(void)strlcpy(dpe.bde_ifname,
2023 				    dp->bd_bif->bif_ifp->if_xname,
2024 				    IFNAMSIZ - 1);
2025 			else
2026 				dpe.bde_ifname[0] = '\0';
2027 
2028 			error = copyout(&dpe, sp, out_size);
2029 			if (error)
2030 				break;
2031 			sp += elem_size;
2032 			len -= elem_size;
2033 		}
2034 		needed += elem_size;
2035 		if (elem_count > 0 && elem_count != INT_MAX)
2036 			elem_count--;
2037 	}
2038 	mutex_exit(&bpf_mtx);
2039 
2040 	*oldlenp = needed;
2041 
2042 	return (error);
2043 }
2044 
2045 static struct sysctllog *bpf_sysctllog;
2046 static void
2047 sysctl_net_bpf_setup(void)
2048 {
2049 	const struct sysctlnode *node;
2050 
2051 	node = NULL;
2052 	sysctl_createv(&bpf_sysctllog, 0, NULL, &node,
2053 		       CTLFLAG_PERMANENT,
2054 		       CTLTYPE_NODE, "bpf",
2055 		       SYSCTL_DESCR("BPF options"),
2056 		       NULL, 0, NULL, 0,
2057 		       CTL_NET, CTL_CREATE, CTL_EOL);
2058 	if (node != NULL) {
2059 #if defined(MODULAR) || defined(BPFJIT)
2060 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2061 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2062 			CTLTYPE_BOOL, "jit",
2063 			SYSCTL_DESCR("Toggle Just-In-Time compilation"),
2064 			sysctl_net_bpf_jit, 0, &bpf_jit, 0,
2065 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2066 #endif
2067 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2068 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2069 			CTLTYPE_INT, "maxbufsize",
2070 			SYSCTL_DESCR("Maximum size for data capture buffer"),
2071 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
2072 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2073 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2074 			CTLFLAG_PERMANENT,
2075 			CTLTYPE_STRUCT, "stats",
2076 			SYSCTL_DESCR("BPF stats"),
2077 			NULL, 0, &bpf_gstats, sizeof(bpf_gstats),
2078 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2079 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2080 			CTLFLAG_PERMANENT,
2081 			CTLTYPE_STRUCT, "peers",
2082 			SYSCTL_DESCR("BPF peers"),
2083 			sysctl_net_bpf_peers, 0, NULL, 0,
2084 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2085 	}
2086 
2087 }
2088 
2089 struct bpf_ops bpf_ops_kernel = {
2090 	.bpf_attach =		_bpfattach,
2091 	.bpf_detach =		_bpfdetach,
2092 	.bpf_change_type =	_bpf_change_type,
2093 
2094 	.bpf_tap =		_bpf_tap,
2095 	.bpf_mtap =		_bpf_mtap,
2096 	.bpf_mtap2 =		_bpf_mtap2,
2097 	.bpf_mtap_af =		_bpf_mtap_af,
2098 	.bpf_mtap_sl_in =	_bpf_mtap_sl_in,
2099 	.bpf_mtap_sl_out =	_bpf_mtap_sl_out,
2100 };
2101 
2102 MODULE(MODULE_CLASS_DRIVER, bpf, NULL);
2103 
2104 static int
2105 bpf_modcmd(modcmd_t cmd, void *arg)
2106 {
2107 	devmajor_t bmajor, cmajor;
2108 	int error;
2109 
2110 	bmajor = cmajor = NODEVMAJOR;
2111 
2112 	switch (cmd) {
2113 	case MODULE_CMD_INIT:
2114 		bpfilterattach(0);
2115 		error = devsw_attach("bpf", NULL, &bmajor,
2116 		    &bpf_cdevsw, &cmajor);
2117 		if (error == EEXIST)
2118 			error = 0; /* maybe built-in ... improve eventually */
2119 		if (error)
2120 			break;
2121 
2122 		bpf_ops_handover_enter(&bpf_ops_kernel);
2123 		atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel);
2124 		bpf_ops_handover_exit();
2125 		sysctl_net_bpf_setup();
2126 		break;
2127 
2128 	case MODULE_CMD_FINI:
2129 		/*
2130 		 * While there is no reference counting for bpf callers,
2131 		 * unload could at least in theory be done similarly to
2132 		 * system call disestablishment.  This should even be
2133 		 * a little simpler:
2134 		 *
2135 		 * 1) replace op vector with stubs
2136 		 * 2) post update to all cpus with xc
2137 		 * 3) check that nobody is in bpf anymore
2138 		 *    (it's doubtful we'd want something like l_sysent,
2139 		 *     but we could do something like *signed* percpu
2140 		 *     counters.  if the sum is 0, we're good).
2141 		 * 4) if fail, unroll changes
2142 		 *
2143 		 * NOTE: change won't be atomic to the outside.  some
2144 		 * packets may be not captured even if unload is
2145 		 * not succesful.  I think packet capture not working
2146 		 * is a perfectly logical consequence of trying to
2147 		 * disable packet capture.
2148 		 */
2149 		error = EOPNOTSUPP;
2150 		/* insert sysctl teardown */
2151 		break;
2152 
2153 	default:
2154 		error = ENOTTY;
2155 		break;
2156 	}
2157 
2158 	return error;
2159 }
2160