xref: /netbsd-src/sys/net/bpf.c (revision c94bf7cb489840bf1d8df08d9cb6179007cc7fe2)
1 /*	$NetBSD: bpf.c,v 1.258 2024/10/20 14:03:51 mlelstv 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 1.258 2024/10/20 14:03:51 mlelstv Exp $");
43 
44 #if defined(_KERNEL_OPT)
45 #include "opt_bpf.h"
46 #include "sl.h"
47 #include "opt_net_mpsafe.h"
48 #endif
49 
50 #include <sys/param.h>
51 
52 #include <sys/atomic.h>
53 #include <sys/buf.h>
54 #include <sys/conf.h>
55 #include <sys/cpu.h>
56 #include <sys/errno.h>
57 #include <sys/file.h>
58 #include <sys/filedesc.h>
59 #include <sys/ioctl.h>
60 #include <sys/kauth.h>
61 #include <sys/kernel.h>
62 #include <sys/lwp.h>
63 #include <sys/mbuf.h>
64 #include <sys/module.h>
65 #include <sys/percpu.h>
66 #include <sys/poll.h>
67 #include <sys/proc.h>
68 #include <sys/protosw.h>
69 #include <sys/pserialize.h>
70 #include <sys/queue.h>
71 #include <sys/socket.h>
72 #include <sys/stat.h>
73 #include <sys/sysctl.h>
74 #include <sys/syslog.h>
75 #include <sys/systm.h>
76 #include <sys/time.h>
77 #include <sys/tty.h>
78 #include <sys/uio.h>
79 #include <sys/vnode.h>
80 #include <sys/xcall.h>
81 
82 #include <net/bpf.h>
83 #include <net/bpfdesc.h>
84 #include <net/bpfjit.h>
85 #include <net/if.h>
86 #include <net/if_arc.h>
87 #include <net/if_ether.h>
88 #include <net/if_types.h>
89 #include <net/slip.h>
90 
91 #include <netinet/if_inarp.h>
92 #include <netinet/in.h>
93 
94 #include <compat/sys/sockio.h>
95 
96 #ifndef BPF_BUFSIZE
97 /*
98  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
99  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
100  */
101 # define BPF_BUFSIZE 32768
102 #endif
103 
104 #define PRINET  26			/* interruptible */
105 
106 /*
107  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
108  * XXX the default values should be computed dynamically based
109  * on available memory size and available mbuf clusters.
110  */
111 static int bpf_bufsize = BPF_BUFSIZE;
112 static int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
113 static bool bpf_jit = false;
114 
115 struct bpfjit_ops bpfjit_module_ops = {
116 	.bj_generate_code = NULL,
117 	.bj_free_code = NULL
118 };
119 
120 /*
121  * Global BPF statistics returned by net.bpf.stats sysctl.
122  */
123 static struct percpu	*bpf_gstats_percpu; /* struct bpf_stat */
124 
125 #define BPF_STATINC(id)					\
126 	{						\
127 		struct bpf_stat *__stats =		\
128 		    percpu_getref(bpf_gstats_percpu);	\
129 		__stats->bs_##id++;			\
130 		percpu_putref(bpf_gstats_percpu);	\
131 	}
132 
133 /*
134  * Locking notes:
135  * - bpf_mtx (adaptive mutex) protects:
136  *   - Gobal lists: bpf_iflist and bpf_dlist
137  *   - struct bpf_if
138  *   - bpf_close
139  *   - bpf_psz (pserialize)
140  * - struct bpf_d has two mutexes:
141  *   - bd_buf_mtx (spin mutex) protects the buffers that can be accessed
142  *     on packet tapping
143  *   - bd_mtx (adaptive mutex) protects member variables other than the buffers
144  * - Locking order: bpf_mtx => bpf_d#bd_mtx => bpf_d#bd_buf_mtx
145  * - struct bpf_d obtained via fp->f_bpf in bpf_read and bpf_write is
146  *   never freed because struct bpf_d is only freed in bpf_close and
147  *   bpf_close never be called while executing bpf_read and bpf_write
148  * - A filter that is assigned to bpf_d can be replaced with another filter
149  *   while tapping packets, so it needs to be done atomically
150  * - struct bpf_d is iterated on bpf_dlist with psz
151  * - struct bpf_if is iterated on bpf_iflist with psz or psref
152  */
153 /*
154  * Use a mutex to avoid a race condition between gathering the stats/peers
155  * and opening/closing the device.
156  */
157 static kmutex_t bpf_mtx;
158 
159 static struct psref_class	*bpf_psref_class __read_mostly;
160 static pserialize_t		bpf_psz;
161 
162 static inline void
163 bpf_if_acquire(struct bpf_if *bp, struct psref *psref)
164 {
165 
166 	psref_acquire(psref, &bp->bif_psref, bpf_psref_class);
167 }
168 
169 static inline void
170 bpf_if_release(struct bpf_if *bp, struct psref *psref)
171 {
172 
173 	psref_release(psref, &bp->bif_psref, bpf_psref_class);
174 }
175 
176 /*
177  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
178  *  bpf_dtab holds the descriptors, indexed by minor device #
179  */
180 static struct pslist_head bpf_iflist;
181 static struct pslist_head bpf_dlist;
182 
183 /* Macros for bpf_d on bpf_dlist */
184 #define BPF_DLIST_WRITER_INSERT_HEAD(__d)				\
185 	PSLIST_WRITER_INSERT_HEAD(&bpf_dlist, (__d), bd_bpf_dlist_entry)
186 #define BPF_DLIST_READER_FOREACH(__d)					\
187 	PSLIST_READER_FOREACH((__d), &bpf_dlist, struct bpf_d,		\
188 	    bd_bpf_dlist_entry)
189 #define BPF_DLIST_WRITER_FOREACH(__d)					\
190 	PSLIST_WRITER_FOREACH((__d), &bpf_dlist, struct bpf_d,		\
191 	    bd_bpf_dlist_entry)
192 #define BPF_DLIST_ENTRY_INIT(__d)					\
193 	PSLIST_ENTRY_INIT((__d), bd_bpf_dlist_entry)
194 #define BPF_DLIST_WRITER_REMOVE(__d)					\
195 	PSLIST_WRITER_REMOVE((__d), bd_bpf_dlist_entry)
196 #define BPF_DLIST_ENTRY_DESTROY(__d)					\
197 	PSLIST_ENTRY_DESTROY((__d), bd_bpf_dlist_entry)
198 
199 /* Macros for bpf_if on bpf_iflist */
200 #define BPF_IFLIST_WRITER_INSERT_HEAD(__bp)				\
201 	PSLIST_WRITER_INSERT_HEAD(&bpf_iflist, (__bp), bif_iflist_entry)
202 #define BPF_IFLIST_READER_FOREACH(__bp)					\
203 	PSLIST_READER_FOREACH((__bp), &bpf_iflist, struct bpf_if,	\
204 	    bif_iflist_entry)
205 #define BPF_IFLIST_WRITER_FOREACH(__bp)					\
206 	PSLIST_WRITER_FOREACH((__bp), &bpf_iflist, struct bpf_if,	\
207 	    bif_iflist_entry)
208 #define BPF_IFLIST_WRITER_REMOVE(__bp)					\
209 	PSLIST_WRITER_REMOVE((__bp), bif_iflist_entry)
210 #define BPF_IFLIST_ENTRY_INIT(__bp)					\
211 	PSLIST_ENTRY_INIT((__bp), bif_iflist_entry)
212 #define BPF_IFLIST_ENTRY_DESTROY(__bp)					\
213 	PSLIST_ENTRY_DESTROY((__bp), bif_iflist_entry)
214 
215 /* Macros for bpf_d on bpf_if#bif_dlist_pslist */
216 #define BPFIF_DLIST_READER_FOREACH(__d, __bp)				\
217 	PSLIST_READER_FOREACH((__d), &(__bp)->bif_dlist_head, struct bpf_d, \
218 	    bd_bif_dlist_entry)
219 #define BPFIF_DLIST_WRITER_INSERT_HEAD(__bp, __d)			\
220 	PSLIST_WRITER_INSERT_HEAD(&(__bp)->bif_dlist_head, (__d),	\
221 	    bd_bif_dlist_entry)
222 #define BPFIF_DLIST_WRITER_REMOVE(__d)					\
223 	PSLIST_WRITER_REMOVE((__d), bd_bif_dlist_entry)
224 #define BPFIF_DLIST_ENTRY_INIT(__d)					\
225 	PSLIST_ENTRY_INIT((__d), bd_bif_dlist_entry)
226 #define	BPFIF_DLIST_READER_EMPTY(__bp)					\
227 	(PSLIST_READER_FIRST(&(__bp)->bif_dlist_head, struct bpf_d,	\
228 	    bd_bif_dlist_entry) == NULL)
229 #define	BPFIF_DLIST_WRITER_EMPTY(__bp)					\
230 	(PSLIST_WRITER_FIRST(&(__bp)->bif_dlist_head, struct bpf_d,	\
231 	    bd_bif_dlist_entry) == NULL)
232 #define BPFIF_DLIST_ENTRY_DESTROY(__d)					\
233 	PSLIST_ENTRY_DESTROY((__d), bd_bif_dlist_entry)
234 
235 static int	bpf_allocbufs(struct bpf_d *);
236 static u_int	bpf_xfilter(struct bpf_filter **, void *, u_int, u_int);
237 static void	bpf_deliver(struct bpf_if *,
238 		    void *(*cpfn)(void *, const void *, size_t),
239 		    void *, u_int, u_int, const u_int);
240 static void	bpf_freed(struct bpf_d *);
241 static void	bpf_free_filter(struct bpf_filter *);
242 static void	bpf_ifname(struct ifnet *, struct ifreq *);
243 static void	*bpf_mcpy(void *, const void *, size_t);
244 static int	bpf_movein(struct ifnet *, struct uio *, int, uint64_t,
245 		    struct mbuf **, struct sockaddr *,
246 		    struct bpf_filter **);
247 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
248 static void	bpf_detachd(struct bpf_d *);
249 static int	bpf_setif(struct bpf_d *, struct ifreq *);
250 static int	bpf_setf(struct bpf_d *, struct bpf_program *, u_long);
251 static void	bpf_timed_out(void *);
252 static inline void
253 		bpf_wakeup(struct bpf_d *);
254 static int	bpf_hdrlen(struct bpf_d *);
255 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
256 		    void *(*)(void *, const void *, size_t),
257 		    struct timespec *);
258 static void	reset_d(struct bpf_d *);
259 static int	bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
260 static int	bpf_setdlt(struct bpf_d *, u_int);
261 
262 static int	bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t,
263 		    int);
264 static int	bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t,
265 		    int);
266 static int	bpf_ioctl(struct file *, u_long, void *);
267 static int	bpf_poll(struct file *, int);
268 static int	bpf_stat(struct file *, struct stat *);
269 static int	bpf_close(struct file *);
270 static int	bpf_kqfilter(struct file *, struct knote *);
271 static void	bpf_softintr(void *);
272 
273 static const struct fileops bpf_fileops = {
274 	.fo_name = "bpf",
275 	.fo_read = bpf_read,
276 	.fo_write = bpf_write,
277 	.fo_ioctl = bpf_ioctl,
278 	.fo_fcntl = fnullop_fcntl,
279 	.fo_poll = bpf_poll,
280 	.fo_stat = bpf_stat,
281 	.fo_close = bpf_close,
282 	.fo_kqfilter = bpf_kqfilter,
283 	.fo_restart = fnullop_restart,
284 };
285 
286 dev_type_open(bpfopen);
287 
288 const struct cdevsw bpf_cdevsw = {
289 	.d_open = bpfopen,
290 	.d_close = noclose,
291 	.d_read = noread,
292 	.d_write = nowrite,
293 	.d_ioctl = noioctl,
294 	.d_stop = nostop,
295 	.d_tty = notty,
296 	.d_poll = nopoll,
297 	.d_mmap = nommap,
298 	.d_kqfilter = nokqfilter,
299 	.d_discard = nodiscard,
300 	.d_flag = D_OTHER | D_MPSAFE
301 };
302 
303 bpfjit_func_t
304 bpf_jit_generate(bpf_ctx_t *bc, void *code, size_t size)
305 {
306 	struct bpfjit_ops *ops = &bpfjit_module_ops;
307 	bpfjit_func_t (*generate_code)(const bpf_ctx_t *,
308 	    const struct bpf_insn *, size_t);
309 
310 	generate_code = atomic_load_acquire(&ops->bj_generate_code);
311 	if (generate_code != NULL) {
312 		return generate_code(bc, code, size);
313 	}
314 	return NULL;
315 }
316 
317 void
318 bpf_jit_freecode(bpfjit_func_t jcode)
319 {
320 	KASSERT(bpfjit_module_ops.bj_free_code != NULL);
321 	bpfjit_module_ops.bj_free_code(jcode);
322 }
323 
324 static int
325 bpf_movein(struct ifnet *ifp, struct uio *uio, int linktype, uint64_t mtu,
326     struct mbuf **mp, struct sockaddr *sockp, struct bpf_filter **wfilter)
327 {
328 	struct mbuf *m, *m0, *n;
329 	int error;
330 	size_t len;
331 	size_t hlen;
332 	size_t align;
333 	u_int slen;
334 
335 	/*
336 	 * Build a sockaddr based on the data link layer type.
337 	 * We do this at this level because the ethernet header
338 	 * is copied directly into the data field of the sockaddr.
339 	 * In the case of SLIP, there is no header and the packet
340 	 * is forwarded as is.
341 	 * Also, we are careful to leave room at the front of the mbuf
342 	 * for the link level header.
343 	 */
344 	switch (linktype) {
345 
346 	case DLT_SLIP:
347 		sockp->sa_family = AF_INET;
348 		hlen = 0;
349 		align = 0;
350 		break;
351 
352 	case DLT_PPP:
353 		sockp->sa_family = AF_UNSPEC;
354 		hlen = 0;
355 		align = 0;
356 		break;
357 
358 	case DLT_EN10MB:
359 		sockp->sa_family = AF_UNSPEC;
360 		/* XXX Would MAXLINKHDR be better? */
361  		/* 6(dst)+6(src)+2(type) */
362 		hlen = sizeof(struct ether_header);
363 		align = 2;
364 		break;
365 
366 	case DLT_ARCNET:
367 		sockp->sa_family = AF_UNSPEC;
368 		hlen = ARC_HDRLEN;
369 		align = 5;
370 		break;
371 
372 	case DLT_FDDI:
373 		sockp->sa_family = AF_LINK;
374 		/* XXX 4(FORMAC)+6(dst)+6(src) */
375 		hlen = 16;
376 		align = 0;
377 		break;
378 
379 	case DLT_ECONET:
380 		sockp->sa_family = AF_UNSPEC;
381 		hlen = 6;
382 		align = 2;
383 		break;
384 
385 	case DLT_NULL:
386 		sockp->sa_family = AF_UNSPEC;
387 		if (ifp->if_type == IFT_LOOP) {
388 			/* Set here to apply the following validations */
389 			hlen = sizeof(uint32_t);
390 		} else
391 			hlen = 0;
392 		align = 0;
393 		break;
394 
395 	default:
396 		return (EIO);
397 	}
398 
399 	len = uio->uio_resid;
400 	/*
401 	 * If there aren't enough bytes for a link level header or the
402 	 * packet length exceeds the interface mtu, return an error.
403 	 */
404 	if (len - hlen > mtu)
405 		return (EMSGSIZE);
406 
407 	m0 = m = m_gethdr(M_WAIT, MT_DATA);
408 	m_reset_rcvif(m);
409 	m->m_pkthdr.len = (int)(len - hlen);
410 	if (len + align > MHLEN) {
411 		m_clget(m, M_WAIT);
412 		if ((m->m_flags & M_EXT) == 0) {
413 			error = ENOBUFS;
414 			goto bad;
415 		}
416 	}
417 
418 	/* Ensure the data is properly aligned */
419 	if (align > 0)
420 		m->m_data += align;
421 
422 	for (;;) {
423 		len = M_TRAILINGSPACE(m);
424 		if (len > uio->uio_resid)
425 			len = uio->uio_resid;
426 		error = uiomove(mtod(m, void *), len, uio);
427 		if (error)
428 			goto bad;
429 		m->m_len = len;
430 
431 		if (uio->uio_resid == 0)
432 			break;
433 
434 		n = m_get(M_WAIT, MT_DATA);
435 		m_clget(n, M_WAIT);	/* if fails, there is no problem */
436 		m->m_next = n;
437 		m = n;
438 	}
439 
440 	slen = bpf_xfilter(wfilter, mtod(m, u_char *), len, len);
441 	if (slen == 0) {
442 		error = EPERM;
443 		goto bad;
444 	}
445 
446 	if (hlen != 0) {
447 		if (linktype == DLT_NULL && ifp->if_type == IFT_LOOP) {
448 			uint32_t af;
449 			/* the link header indicates the address family */
450 			memcpy(&af, mtod(m0, void *), sizeof(af));
451 			sockp->sa_family = af;
452 		} else {
453 			/* move link level header in the top of mbuf to sa_data */
454 			memcpy(sockp->sa_data, mtod(m0, void *), hlen);
455 		}
456 		m0->m_data += hlen;
457 		m0->m_len -= hlen;
458 	}
459 
460 	m_claimm(m, ifp->if_mowner);
461 
462 	*mp = m0;
463 	return (0);
464 
465 bad:
466 	m_freem(m0);
467 	return (error);
468 }
469 
470 /*
471  * Attach file to the bpf interface, i.e. make d listen on bp.
472  */
473 static void
474 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
475 {
476 	struct bpf_event_tracker *t;
477 
478 	KASSERT(mutex_owned(&bpf_mtx));
479 	KASSERT(mutex_owned(d->bd_mtx));
480 	/*
481 	 * Point d at bp, and add d to the interface's list of listeners.
482 	 * Finally, point the driver's bpf cookie at the interface so
483 	 * it will divert packets to bpf.
484 	 */
485 	d->bd_bif = bp;
486 	BPFIF_DLIST_WRITER_INSERT_HEAD(bp, d);
487 
488 	*bp->bif_driverp = bp;
489 
490 	SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) {
491 		t->bet_notify(bp, bp->bif_ifp, bp->bif_dlt,
492 		    BPF_TRACK_EVENT_ATTACH);
493 	}
494 }
495 
496 /*
497  * Detach a file from its interface.
498  */
499 static void
500 bpf_detachd(struct bpf_d *d)
501 {
502 	struct bpf_if *bp;
503 	struct bpf_event_tracker *t;
504 
505 	KASSERT(mutex_owned(&bpf_mtx));
506 	KASSERT(mutex_owned(d->bd_mtx));
507 
508 	bp = d->bd_bif;
509 	/*
510 	 * Check if this descriptor had requested promiscuous mode.
511 	 * If so, turn it off.
512 	 */
513 	if (d->bd_promisc) {
514 		int error __diagused;
515 
516 		d->bd_promisc = 0;
517 		/*
518 		 * Take device out of promiscuous mode.  Since we were
519 		 * able to enter promiscuous mode, we should be able
520 		 * to turn it off.  But we can get an error if
521 		 * the interface was configured down, so only panic
522 		 * if we don't get an unexpected error.
523 		 */
524 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
525   		error = ifpromisc(bp->bif_ifp, 0);
526 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
527 #ifdef DIAGNOSTIC
528 		if (error)
529 			printf("%s: ifpromisc failed: %d", __func__, error);
530 #endif
531 	}
532 
533 	/* Remove d from the interface's descriptor list. */
534 	BPFIF_DLIST_WRITER_REMOVE(d);
535 
536 	pserialize_perform(bpf_psz);
537 
538 	if (BPFIF_DLIST_WRITER_EMPTY(bp)) {
539 		/*
540 		 * Let the driver know that there are no more listeners.
541 		 */
542 		*d->bd_bif->bif_driverp = NULL;
543 	}
544 
545 	d->bd_bif = NULL;
546 
547 	SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) {
548 		t->bet_notify(bp, bp->bif_ifp, bp->bif_dlt,
549 		    BPF_TRACK_EVENT_DETACH);
550 	}
551 }
552 
553 static void
554 bpf_init(void)
555 {
556 
557 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
558 	bpf_psz = pserialize_create();
559 	bpf_psref_class = psref_class_create("bpf", IPL_SOFTNET);
560 
561 	PSLIST_INIT(&bpf_iflist);
562 	PSLIST_INIT(&bpf_dlist);
563 
564 	bpf_gstats_percpu = percpu_alloc(sizeof(struct bpf_stat));
565 
566 	return;
567 }
568 
569 /*
570  * bpfilterattach() is called at boot time.  We don't need to do anything
571  * here, since any initialization will happen as part of module init code.
572  */
573 /* ARGSUSED */
574 void
575 bpfilterattach(int n)
576 {
577 
578 }
579 
580 /*
581  * Open ethernet device. Clones.
582  */
583 /* ARGSUSED */
584 int
585 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
586 {
587 	struct bpf_d *d;
588 	struct file *fp;
589 	int error, fd;
590 
591 	/* falloc() will fill in the descriptor for us. */
592 	if ((error = fd_allocfile(&fp, &fd)) != 0)
593 		return error;
594 
595 	d = kmem_zalloc(sizeof(*d), KM_SLEEP);
596 	d->bd_bufsize = bpf_bufsize;
597 	d->bd_direction = BPF_D_INOUT;
598 	d->bd_feedback = 0;
599 	d->bd_pid = l->l_proc->p_pid;
600 #ifdef _LP64
601 	if (curproc->p_flag & PK_32)
602 		d->bd_compat32 = 1;
603 #endif
604 	getnanotime(&d->bd_btime);
605 	d->bd_atime = d->bd_mtime = d->bd_btime;
606 	callout_init(&d->bd_callout, CALLOUT_MPSAFE);
607 	selinit(&d->bd_sel);
608 	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
609 	d->bd_jitcode = NULL;
610 	d->bd_rfilter = NULL;
611 	d->bd_wfilter = NULL;
612 	d->bd_locked = 0;
613 	BPF_DLIST_ENTRY_INIT(d);
614 	BPFIF_DLIST_ENTRY_INIT(d);
615 	d->bd_mtx = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
616 	d->bd_buf_mtx = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
617 	cv_init(&d->bd_cv, "bpf");
618 
619 	mutex_enter(&bpf_mtx);
620 	BPF_DLIST_WRITER_INSERT_HEAD(d);
621 	mutex_exit(&bpf_mtx);
622 
623 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
624 }
625 
626 /*
627  * Close the descriptor by detaching it from its interface,
628  * deallocating its buffers, and marking it free.
629  */
630 /* ARGSUSED */
631 static int
632 bpf_close(struct file *fp)
633 {
634 	struct bpf_d *d;
635 
636 	mutex_enter(&bpf_mtx);
637 
638 	if ((d = fp->f_bpf) == NULL) {
639 		mutex_exit(&bpf_mtx);
640 		return 0;
641 	}
642 
643 	/*
644 	 * Refresh the PID associated with this bpf file.
645 	 */
646 	d->bd_pid = curproc->p_pid;
647 
648 	mutex_enter(d->bd_buf_mtx);
649 	if (d->bd_state == BPF_WAITING)
650 		callout_halt(&d->bd_callout, d->bd_buf_mtx);
651 	d->bd_state = BPF_IDLE;
652 	mutex_exit(d->bd_buf_mtx);
653 	mutex_enter(d->bd_mtx);
654 	if (d->bd_bif)
655 		bpf_detachd(d);
656 	mutex_exit(d->bd_mtx);
657 
658 	BPF_DLIST_WRITER_REMOVE(d);
659 
660 	pserialize_perform(bpf_psz);
661 	mutex_exit(&bpf_mtx);
662 
663 	BPFIF_DLIST_ENTRY_DESTROY(d);
664 	BPF_DLIST_ENTRY_DESTROY(d);
665 	fp->f_bpf = NULL;
666 	bpf_freed(d);
667 	callout_destroy(&d->bd_callout);
668 	seldestroy(&d->bd_sel);
669 	softint_disestablish(d->bd_sih);
670 	mutex_obj_free(d->bd_mtx);
671 	mutex_obj_free(d->bd_buf_mtx);
672 	cv_destroy(&d->bd_cv);
673 
674 	kmem_free(d, sizeof(*d));
675 
676 	return (0);
677 }
678 
679 /*
680  * Rotate the packet buffers in descriptor d.  Move the store buffer
681  * into the hold slot, and the free buffer into the store slot.
682  * Zero the length of the new store buffer.
683  */
684 #define ROTATE_BUFFERS(d) \
685 	(d)->bd_hbuf = (d)->bd_sbuf; \
686 	(d)->bd_hlen = (d)->bd_slen; \
687 	(d)->bd_sbuf = (d)->bd_fbuf; \
688 	(d)->bd_slen = 0; \
689 	(d)->bd_fbuf = NULL;
690 /*
691  *  bpfread - read next chunk of packets from buffers
692  */
693 static int
694 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
695     kauth_cred_t cred, int flags)
696 {
697 	struct bpf_d *d = fp->f_bpf;
698 	int timed_out;
699 	int error;
700 
701 	/*
702 	 * Refresh the PID associated with this bpf file.
703 	 */
704 	d->bd_pid = curproc->p_pid;
705 
706 	getnanotime(&d->bd_atime);
707 	/*
708 	 * Restrict application to use a buffer the same size as
709 	 * the kernel buffers.
710 	 */
711 	if (uio->uio_resid != d->bd_bufsize)
712 		return (EINVAL);
713 
714 	mutex_enter(d->bd_buf_mtx);
715 	if (d->bd_state == BPF_WAITING)
716 		callout_halt(&d->bd_callout, d->bd_buf_mtx);
717 	timed_out = (d->bd_state == BPF_TIMED_OUT);
718 	d->bd_state = BPF_IDLE;
719 	mutex_exit(d->bd_buf_mtx);
720 	/*
721 	 * If the hold buffer is empty, then do a timed sleep, which
722 	 * ends when the timeout expires or when enough packets
723 	 * have arrived to fill the store buffer.
724 	 */
725 	mutex_enter(d->bd_buf_mtx);
726 	while (d->bd_hbuf == NULL) {
727 		if (fp->f_flag & FNONBLOCK) {
728 			if (d->bd_slen == 0) {
729 				error = EWOULDBLOCK;
730 				goto out;
731 			}
732 			ROTATE_BUFFERS(d);
733 			break;
734 		}
735 
736 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
737 			/*
738 			 * A packet(s) either arrived since the previous
739 			 * read or arrived while we were asleep.
740 			 * Rotate the buffers and return what's here.
741 			 */
742 			ROTATE_BUFFERS(d);
743 			break;
744 		}
745 
746 		error = cv_timedwait_sig(&d->bd_cv, d->bd_buf_mtx, d->bd_rtout);
747 
748 		if (error == EINTR || error == ERESTART)
749 			goto out;
750 
751 		if (error == EWOULDBLOCK) {
752 			/*
753 			 * On a timeout, return what's in the buffer,
754 			 * which may be nothing.  If there is something
755 			 * in the store buffer, we can rotate the buffers.
756 			 */
757 			if (d->bd_hbuf)
758 				/*
759 				 * We filled up the buffer in between
760 				 * getting the timeout and arriving
761 				 * here, so we don't need to rotate.
762 				 */
763 				break;
764 
765 			if (d->bd_slen == 0) {
766 				error = 0;
767 				goto out;
768 			}
769 			ROTATE_BUFFERS(d);
770 			break;
771 		}
772 		if (error != 0)
773 			goto out;
774 	}
775 	/*
776 	 * At this point, we know we have something in the hold slot.
777 	 */
778 	mutex_exit(d->bd_buf_mtx);
779 
780 	/*
781 	 * Move data from hold buffer into user space.
782 	 * We know the entire buffer is transferred since
783 	 * we checked above that the read buffer is bpf_bufsize bytes.
784 	 */
785 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
786 
787 	mutex_enter(d->bd_buf_mtx);
788 	d->bd_fbuf = d->bd_hbuf;
789 	d->bd_hbuf = NULL;
790 	d->bd_hlen = 0;
791 out:
792 	mutex_exit(d->bd_buf_mtx);
793 	return (error);
794 }
795 
796 /*
797  * If there are processes sleeping on this descriptor, wake them up.
798  */
799 static inline void
800 bpf_wakeup(struct bpf_d *d)
801 {
802 
803 	KASSERT(mutex_owned(d->bd_buf_mtx));
804 
805 	cv_broadcast(&d->bd_cv);
806 
807 	if (d->bd_async)
808 		softint_schedule(d->bd_sih);
809 	selnotify(&d->bd_sel, 0, NOTE_SUBMIT);
810 }
811 
812 static void
813 bpf_softintr(void *cookie)
814 {
815 	struct bpf_d *d;
816 
817 	d = cookie;
818 	if (d->bd_async)
819 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
820 }
821 
822 static void
823 bpf_timed_out(void *arg)
824 {
825 	struct bpf_d *d = arg;
826 
827 	mutex_enter(d->bd_buf_mtx);
828 	if (d->bd_state == BPF_WAITING) {
829 		d->bd_state = BPF_TIMED_OUT;
830 		if (d->bd_slen != 0)
831 			bpf_wakeup(d);
832 	}
833 	mutex_exit(d->bd_buf_mtx);
834 }
835 
836 static int
837 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
838     kauth_cred_t cred, int flags)
839 {
840 	struct bpf_d *d = fp->f_bpf;
841 	struct bpf_if *bp;
842 	struct ifnet *ifp;
843 	struct mbuf *m, *mc;
844 	int error;
845 	static struct sockaddr_storage dst;
846 	struct psref psref;
847 	int bound;
848 
849 	/*
850 	 * Refresh the PID associated with this bpf file.
851 	 */
852 	d->bd_pid = curproc->p_pid;
853 
854 	m = NULL;	/* XXX gcc */
855 
856 	bound = curlwp_bind();
857 	mutex_enter(d->bd_mtx);
858 	bp = d->bd_bif;
859 	if (bp == NULL) {
860 		mutex_exit(d->bd_mtx);
861 		error = ENXIO;
862 		goto out_bindx;
863 	}
864 	bpf_if_acquire(bp, &psref);
865 	mutex_exit(d->bd_mtx);
866 
867 	getnanotime(&d->bd_mtime);
868 
869 	ifp = bp->bif_ifp;
870 	if (if_is_deactivated(ifp)) {
871 		error = ENXIO;
872 		goto out;
873 	}
874 
875 	if (uio->uio_resid == 0) {
876 		error = 0;
877 		goto out;
878 	}
879 
880 	error = bpf_movein(ifp, uio, (int)bp->bif_dlt, ifp->if_mtu, &m,
881 	    (struct sockaddr *) &dst, &d->bd_wfilter);
882 	if (error)
883 		goto out;
884 
885 	if (m->m_pkthdr.len > ifp->if_mtu) {
886 		m_freem(m);
887 		error = EMSGSIZE;
888 		goto out;
889 	}
890 
891 	/*
892 	 * If writing to a loopback interface, the address family has
893 	 * already been specially computed in bpf_movein(), so don't
894 	 * clobber it, or the loopback will reject it in looutput().
895 	 */
896 	if (d->bd_hdrcmplt && ifp->if_type != IFT_LOOP)
897 		dst.ss_family = pseudo_AF_HDRCMPLT;
898 
899 	if (d->bd_feedback) {
900 		mc = m_dup(m, 0, M_COPYALL, M_NOWAIT);
901 		if (mc != NULL)
902 			m_set_rcvif(mc, ifp);
903 		/* Set M_PROMISC for outgoing packets to be discarded. */
904 		if (1 /*d->bd_direction == BPF_D_INOUT*/)
905 			m->m_flags |= M_PROMISC;
906 	} else
907 		mc = NULL;
908 
909 	error = if_output_lock(ifp, ifp, m, (struct sockaddr *) &dst, NULL);
910 
911 	if (mc != NULL) {
912 		if (error == 0) {
913 			int s = splsoftnet();
914 			KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
915 			ifp->_if_input(ifp, mc);
916 			KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
917 			splx(s);
918 		} else
919 			m_freem(mc);
920 	}
921 	/*
922 	 * The driver frees the mbuf.
923 	 */
924 out:
925 	bpf_if_release(bp, &psref);
926 out_bindx:
927 	curlwp_bindx(bound);
928 	return error;
929 }
930 
931 /*
932  * Reset a descriptor by flushing its packet buffer and clearing the
933  * receive and drop counts.
934  */
935 static void
936 reset_d(struct bpf_d *d)
937 {
938 
939 	KASSERT(mutex_owned(d->bd_mtx));
940 
941 	mutex_enter(d->bd_buf_mtx);
942 	if (d->bd_hbuf) {
943 		/* Free the hold buffer. */
944 		d->bd_fbuf = d->bd_hbuf;
945 		d->bd_hbuf = NULL;
946 	}
947 	d->bd_slen = 0;
948 	d->bd_hlen = 0;
949 	d->bd_rcount = 0;
950 	d->bd_dcount = 0;
951 	d->bd_ccount = 0;
952 	mutex_exit(d->bd_buf_mtx);
953 }
954 
955 /*
956  *  FIONREAD		Check for read packet available.
957  *  BIOCGBLEN		Get buffer len [for read()].
958  *  BIOCSETF		Set ethernet read filter.
959  *  BIOCFLUSH		Flush read packet buffer.
960  *  BIOCPROMISC		Put interface into promiscuous mode.
961  *  BIOCGDLT		Get link layer type.
962  *  BIOCGETIF		Get interface name.
963  *  BIOCSETIF		Set interface.
964  *  BIOCSRTIMEOUT	Set read timeout.
965  *  BIOCGRTIMEOUT	Get read timeout.
966  *  BIOCGSTATS		Get packet stats.
967  *  BIOCIMMEDIATE	Set immediate mode.
968  *  BIOCVERSION		Get filter language version.
969  *  BIOCGHDRCMPLT	Get "header already complete" flag.
970  *  BIOCSHDRCMPLT	Set "header already complete" flag.
971  *  BIOCSFEEDBACK	Set packet feedback mode.
972  *  BIOCGFEEDBACK	Get packet feedback mode.
973  *  BIOCGDIRECTION	Get packet direction flag
974  *  BIOCSDIRECTION	Set packet direction flag
975  */
976 /* ARGSUSED */
977 static int
978 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
979 {
980 	struct bpf_d *d = fp->f_bpf;
981 	int error = 0;
982 
983 	/*
984 	 * Refresh the PID associated with this bpf file.
985 	 */
986 	d->bd_pid = curproc->p_pid;
987 #ifdef _LP64
988 	if (curproc->p_flag & PK_32)
989 		d->bd_compat32 = 1;
990 	else
991 		d->bd_compat32 = 0;
992 #endif
993 
994 	mutex_enter(d->bd_buf_mtx);
995 	if (d->bd_state == BPF_WAITING)
996 		callout_halt(&d->bd_callout, d->bd_buf_mtx);
997 	d->bd_state = BPF_IDLE;
998 	mutex_exit(d->bd_buf_mtx);
999 
1000 	if (d->bd_locked) {
1001 		switch (cmd) {
1002 		case BIOCGBLEN:		/* FALLTHROUGH */
1003 		case BIOCFLUSH:		/* FALLTHROUGH */
1004 		case BIOCGDLT:		/* FALLTHROUGH */
1005 		case BIOCGDLTLIST:	/* FALLTHROUGH */
1006 		case BIOCGETIF:		/* FALLTHROUGH */
1007 		case BIOCGRTIMEOUT:	/* FALLTHROUGH */
1008 		case BIOCGSTATS:	/* FALLTHROUGH */
1009 		case BIOCVERSION:	/* FALLTHROUGH */
1010 		case BIOCGHDRCMPLT:	/* FALLTHROUGH */
1011 		case FIONREAD:		/* FALLTHROUGH */
1012 		case BIOCLOCK:		/* FALLTHROUGH */
1013 		case BIOCSRTIMEOUT:	/* FALLTHROUGH */
1014 		case BIOCIMMEDIATE:	/* FALLTHROUGH */
1015 		case TIOCGPGRP:
1016 			break;
1017 		default:
1018 			return EPERM;
1019 		}
1020 	}
1021 
1022 	switch (cmd) {
1023 
1024 	default:
1025 		error = EINVAL;
1026 		break;
1027 
1028 	/*
1029 	 * Check for read packet available.
1030 	 */
1031 	case FIONREAD: {
1032 		int n;
1033 
1034 		mutex_enter(d->bd_buf_mtx);
1035 		n = d->bd_slen;
1036 		if (d->bd_hbuf)
1037 			n += d->bd_hlen;
1038 		mutex_exit(d->bd_buf_mtx);
1039 
1040 		*(int *)addr = n;
1041 		break;
1042 	}
1043 
1044 	/*
1045 	 * Get buffer len [for read()].
1046 	 */
1047 	case BIOCGBLEN:
1048 		*(u_int *)addr = d->bd_bufsize;
1049 		break;
1050 
1051 	/*
1052 	 * Set buffer length.
1053 	 */
1054 	case BIOCSBLEN:
1055 		/*
1056 		 * Forbid to change the buffer length if buffers are already
1057 		 * allocated.
1058 		 */
1059 		mutex_enter(d->bd_mtx);
1060 		mutex_enter(d->bd_buf_mtx);
1061 		if (d->bd_bif != NULL || d->bd_sbuf != NULL)
1062 			error = EINVAL;
1063 		else {
1064 			u_int size = *(u_int *)addr;
1065 
1066 			if (size > bpf_maxbufsize)
1067 				*(u_int *)addr = size = bpf_maxbufsize;
1068 			else if (size < BPF_MINBUFSIZE)
1069 				*(u_int *)addr = size = BPF_MINBUFSIZE;
1070 			d->bd_bufsize = size;
1071 		}
1072 		mutex_exit(d->bd_buf_mtx);
1073 		mutex_exit(d->bd_mtx);
1074 		break;
1075 
1076 	/*
1077 	 * Set link layer read filter.
1078 	 */
1079 	case BIOCSETF:		/* FALLTHROUGH */
1080 	case BIOCSETWF:
1081 		error = bpf_setf(d, addr, cmd);
1082 		break;
1083 
1084 	case BIOCLOCK:
1085 		d->bd_locked = 1;
1086 		break;
1087 
1088 	/*
1089 	 * Flush read packet buffer.
1090 	 */
1091 	case BIOCFLUSH:
1092 		mutex_enter(d->bd_mtx);
1093 		reset_d(d);
1094 		mutex_exit(d->bd_mtx);
1095 		break;
1096 
1097 	/*
1098 	 * Put interface into promiscuous mode.
1099 	 */
1100 	case BIOCPROMISC:
1101 		mutex_enter(d->bd_mtx);
1102 		if (d->bd_bif == NULL) {
1103 			mutex_exit(d->bd_mtx);
1104 			/*
1105 			 * No interface attached yet.
1106 			 */
1107 			error = EINVAL;
1108 			break;
1109 		}
1110 		if (d->bd_promisc == 0) {
1111 			KERNEL_LOCK_UNLESS_NET_MPSAFE();
1112 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1113 			KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
1114 			if (error == 0)
1115 				d->bd_promisc = 1;
1116 		}
1117 		mutex_exit(d->bd_mtx);
1118 		break;
1119 
1120 	/*
1121 	 * Get device parameters.
1122 	 */
1123 	case BIOCGDLT:
1124 		mutex_enter(d->bd_mtx);
1125 		if (d->bd_bif == NULL)
1126 			error = EINVAL;
1127 		else
1128 			*(u_int *)addr = d->bd_bif->bif_dlt;
1129 		mutex_exit(d->bd_mtx);
1130 		break;
1131 
1132 	/*
1133 	 * Get a list of supported device parameters.
1134 	 */
1135 	case BIOCGDLTLIST:
1136 		mutex_enter(d->bd_mtx);
1137 		if (d->bd_bif == NULL)
1138 			error = EINVAL;
1139 		else
1140 			error = bpf_getdltlist(d, addr);
1141 		mutex_exit(d->bd_mtx);
1142 		break;
1143 
1144 	/*
1145 	 * Set device parameters.
1146 	 */
1147 	case BIOCSDLT:
1148 		mutex_enter(&bpf_mtx);
1149 		mutex_enter(d->bd_mtx);
1150 		if (d->bd_bif == NULL)
1151 			error = EINVAL;
1152 		else
1153 			error = bpf_setdlt(d, *(u_int *)addr);
1154 		mutex_exit(d->bd_mtx);
1155 		mutex_exit(&bpf_mtx);
1156 		break;
1157 
1158 	/*
1159 	 * Set interface name.
1160 	 */
1161 #ifdef OBIOCGETIF
1162 	case OBIOCGETIF:
1163 #endif
1164 	case BIOCGETIF:
1165 		mutex_enter(d->bd_mtx);
1166 		if (d->bd_bif == NULL)
1167 			error = EINVAL;
1168 		else
1169 			bpf_ifname(d->bd_bif->bif_ifp, addr);
1170 		mutex_exit(d->bd_mtx);
1171 		break;
1172 
1173 	/*
1174 	 * Set interface.
1175 	 */
1176 #ifdef OBIOCSETIF
1177 	case OBIOCSETIF:
1178 #endif
1179 	case BIOCSETIF:
1180 		mutex_enter(&bpf_mtx);
1181 		error = bpf_setif(d, addr);
1182 		mutex_exit(&bpf_mtx);
1183 		break;
1184 
1185 	/*
1186 	 * Set read timeout.
1187 	 */
1188 	case BIOCSRTIMEOUT: {
1189 		struct timeval *tv = addr;
1190 
1191 		/* Compute number of ticks. */
1192 		if (tv->tv_sec < 0 ||
1193 		    tv->tv_usec < 0 || tv->tv_usec >= 1000000) {
1194 			error = EINVAL;
1195 			break;
1196 		} else if (tv->tv_sec > INT_MAX/hz - 1) {
1197 			d->bd_rtout = INT_MAX;
1198 		} else {
1199 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
1200 		}
1201 		if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
1202 			d->bd_rtout = 1;
1203 		break;
1204 	}
1205 
1206 #ifdef BIOCGORTIMEOUT
1207 	/*
1208 	 * Get read timeout.
1209 	 */
1210 	case BIOCGORTIMEOUT: {
1211 		struct timeval50 *tv = addr;
1212 
1213 		tv->tv_sec = d->bd_rtout / hz;
1214 		tv->tv_usec = (d->bd_rtout % hz) * tick;
1215 		break;
1216 	}
1217 #endif
1218 
1219 #ifdef BIOCSORTIMEOUT
1220 	/*
1221 	 * Set read timeout.
1222 	 */
1223 	case BIOCSORTIMEOUT: {
1224 		struct timeval50 *tv = addr;
1225 
1226 		/* Compute number of ticks. */
1227 		if (tv->tv_sec < 0 ||
1228 		    tv->tv_usec < 0 || tv->tv_usec >= 1000000) {
1229 			error = EINVAL;
1230 			break;
1231 		} else if (tv->tv_sec > INT_MAX/hz - 1) {
1232 			d->bd_rtout = INT_MAX;
1233 		} else {
1234 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
1235 		}
1236 		if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
1237 			d->bd_rtout = 1;
1238 		break;
1239 	}
1240 #endif
1241 
1242 	/*
1243 	 * Get read timeout.
1244 	 */
1245 	case BIOCGRTIMEOUT: {
1246 		struct timeval *tv = addr;
1247 
1248 		tv->tv_sec = d->bd_rtout / hz;
1249 		tv->tv_usec = (d->bd_rtout % hz) * tick;
1250 		break;
1251 	}
1252 	/*
1253 	 * Get packet stats.
1254 	 */
1255 	case BIOCGSTATS: {
1256 		struct bpf_stat *bs = addr;
1257 
1258 		bs->bs_recv = d->bd_rcount;
1259 		bs->bs_drop = d->bd_dcount;
1260 		bs->bs_capt = d->bd_ccount;
1261 		break;
1262 	}
1263 
1264 	case BIOCGSTATS_30: {
1265 		struct bpf_stat30 *bs = addr;
1266 
1267 		bs->bs_recv = d->bd_rcount;
1268 		bs->bs_drop = d->bd_dcount;
1269 		break;
1270 	}
1271 
1272 	/*
1273 	 * Set immediate mode.
1274 	 */
1275 	case BIOCIMMEDIATE:
1276 		d->bd_immediate = *(u_int *)addr;
1277 		break;
1278 
1279 	case BIOCVERSION: {
1280 		struct bpf_version *bv = addr;
1281 
1282 		bv->bv_major = BPF_MAJOR_VERSION;
1283 		bv->bv_minor = BPF_MINOR_VERSION;
1284 		break;
1285 	}
1286 
1287 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
1288 		*(u_int *)addr = d->bd_hdrcmplt;
1289 		break;
1290 
1291 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
1292 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1293 		break;
1294 
1295 	/*
1296 	 * Get packet direction flag
1297 	 */
1298 	case BIOCGDIRECTION:
1299 		*(u_int *)addr = d->bd_direction;
1300 		break;
1301 
1302 	/*
1303 	 * Set packet direction flag
1304 	 */
1305 	case BIOCSDIRECTION: {
1306 		u_int	direction;
1307 
1308 		direction = *(u_int *)addr;
1309 		switch (direction) {
1310 		case BPF_D_IN:
1311 		case BPF_D_INOUT:
1312 		case BPF_D_OUT:
1313 			d->bd_direction = direction;
1314 			break;
1315 		default:
1316 			error = EINVAL;
1317 		}
1318 	}
1319 	break;
1320 
1321 	/*
1322 	 * Set "feed packets from bpf back to input" mode
1323 	 */
1324 	case BIOCSFEEDBACK:
1325 		d->bd_feedback = *(u_int *)addr;
1326 		break;
1327 
1328 	/*
1329 	 * Get "feed packets from bpf back to input" mode
1330 	 */
1331 	case BIOCGFEEDBACK:
1332 		*(u_int *)addr = d->bd_feedback;
1333 		break;
1334 
1335 	case FIONBIO:		/* Non-blocking I/O */
1336 		/*
1337 		 * No need to do anything special as we use IO_NDELAY in
1338 		 * bpfread() as an indication of whether or not to block
1339 		 * the read.
1340 		 */
1341 		break;
1342 
1343 	case FIOASYNC:		/* Send signal on receive packets */
1344 		mutex_enter(d->bd_mtx);
1345 		d->bd_async = *(int *)addr;
1346 		mutex_exit(d->bd_mtx);
1347 		break;
1348 
1349 	case TIOCSPGRP:		/* Process or group to send signals to */
1350 	case FIOSETOWN:
1351 		error = fsetown(&d->bd_pgid, cmd, addr);
1352 		break;
1353 
1354 	case TIOCGPGRP:
1355 	case FIOGETOWN:
1356 		error = fgetown(d->bd_pgid, cmd, addr);
1357 		break;
1358 	}
1359 	return (error);
1360 }
1361 
1362 /*
1363  * Set d's packet filter program to fp.  If this file already has a filter,
1364  * free it and replace it.  Returns EINVAL for bogus requests.
1365  */
1366 static int
1367 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1368 {
1369 	struct bpf_insn *fcode;
1370 	bpfjit_func_t jcode;
1371 	size_t flen, size = 0;
1372 	struct bpf_filter *oldf, *newf, **storef;
1373 
1374 	jcode = NULL;
1375 	flen = fp->bf_len;
1376 
1377 	if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) {
1378 		return EINVAL;
1379 	}
1380 
1381 	if (flen) {
1382 		/*
1383 		 * Allocate the buffer, copy the byte-code from
1384 		 * userspace and validate it.
1385 		 */
1386 		size = flen * sizeof(*fp->bf_insns);
1387 		fcode = kmem_alloc(size, KM_SLEEP);
1388 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1389 		    !bpf_validate(fcode, (int)flen)) {
1390 			kmem_free(fcode, size);
1391 			return EINVAL;
1392 		}
1393 		if (bpf_jit)
1394 			jcode = bpf_jit_generate(NULL, fcode, flen);
1395 	} else {
1396 		fcode = NULL;
1397 	}
1398 
1399 	newf = kmem_alloc(sizeof(*newf), KM_SLEEP);
1400 	newf->bf_insn = fcode;
1401 	newf->bf_size = size;
1402 	newf->bf_jitcode = jcode;
1403 	if (cmd == BIOCSETF)
1404 		d->bd_jitcode = jcode; /* XXX just for kvm(3) users */
1405 
1406 	/* Need to hold bpf_mtx for pserialize_perform */
1407 	mutex_enter(&bpf_mtx);
1408 	mutex_enter(d->bd_mtx);
1409 	if (cmd == BIOCSETWF) {
1410 		oldf = d->bd_wfilter;
1411 		storef = &d->bd_wfilter;
1412 	} else {
1413 		oldf = d->bd_rfilter;
1414 		storef = &d->bd_rfilter;
1415 	}
1416 	atomic_store_release(storef, newf);
1417 	reset_d(d);
1418 	pserialize_perform(bpf_psz);
1419 	mutex_exit(d->bd_mtx);
1420 	mutex_exit(&bpf_mtx);
1421 
1422 	if (oldf != NULL)
1423 		bpf_free_filter(oldf);
1424 
1425 	return 0;
1426 }
1427 
1428 /*
1429  * Detach a file from its current interface (if attached at all) and attach
1430  * to the interface indicated by the name stored in ifr.
1431  * Return an errno or 0.
1432  */
1433 static int
1434 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1435 {
1436 	struct bpf_if *bp;
1437 	char *cp;
1438 	int unit_seen, i, error;
1439 
1440 	KASSERT(mutex_owned(&bpf_mtx));
1441 	/*
1442 	 * Make sure the provided name has a unit number, and default
1443 	 * it to '0' if not specified.
1444 	 * XXX This is ugly ... do this differently?
1445 	 */
1446 	unit_seen = 0;
1447 	cp = ifr->ifr_name;
1448 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1449 	while (*cp++)
1450 		if (*cp >= '0' && *cp <= '9')
1451 			unit_seen = 1;
1452 	if (!unit_seen) {
1453 		/* Make sure to leave room for the '\0'. */
1454 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1455 			if ((ifr->ifr_name[i] >= 'a' &&
1456 				ifr->ifr_name[i] <= 'z') ||
1457 			    (ifr->ifr_name[i] >= 'A' &&
1458 				ifr->ifr_name[i] <= 'Z'))
1459 				continue;
1460 			ifr->ifr_name[i] = '0';
1461 		}
1462 	}
1463 
1464 	/*
1465 	 * Look through attached interfaces for the named one.
1466 	 */
1467 	BPF_IFLIST_WRITER_FOREACH(bp) {
1468 		struct ifnet *ifp = bp->bif_ifp;
1469 
1470 		if (ifp == NULL ||
1471 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1472 			continue;
1473 		/* skip additional entry */
1474 		if (bp->bif_driverp != &ifp->if_bpf)
1475 			continue;
1476 		/*
1477 		 * We found the requested interface.
1478 		 * Allocate the packet buffers if we need to.
1479 		 * If we're already attached to requested interface,
1480 		 * just flush the buffer.
1481 		 */
1482 		/*
1483 		 * bpf_allocbufs is called only here. bpf_mtx ensures that
1484 		 * no race condition happen on d->bd_sbuf.
1485 		 */
1486 		if (d->bd_sbuf == NULL) {
1487 			error = bpf_allocbufs(d);
1488 			if (error != 0)
1489 				return (error);
1490 		}
1491 		mutex_enter(d->bd_mtx);
1492 		if (bp != d->bd_bif) {
1493 			if (d->bd_bif) {
1494 				/*
1495 				 * Detach if attached to something else.
1496 				 */
1497 				bpf_detachd(d);
1498 				BPFIF_DLIST_ENTRY_INIT(d);
1499 			}
1500 
1501 			bpf_attachd(d, bp);
1502 		}
1503 		reset_d(d);
1504 		mutex_exit(d->bd_mtx);
1505 		return (0);
1506 	}
1507 	/* Not found. */
1508 	return (ENXIO);
1509 }
1510 
1511 /*
1512  * Copy the interface name to the ifreq.
1513  */
1514 static void
1515 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1516 {
1517 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1518 }
1519 
1520 static int
1521 bpf_stat(struct file *fp, struct stat *st)
1522 {
1523 	struct bpf_d *d = fp->f_bpf;
1524 
1525 	(void)memset(st, 0, sizeof(*st));
1526 	mutex_enter(d->bd_mtx);
1527 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1528 	st->st_atimespec = d->bd_atime;
1529 	st->st_mtimespec = d->bd_mtime;
1530 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1531 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1532 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1533 	st->st_mode = S_IFCHR;
1534 	mutex_exit(d->bd_mtx);
1535 	return 0;
1536 }
1537 
1538 /*
1539  * Support for poll() system call
1540  *
1541  * Return true iff the specific operation will not block indefinitely - with
1542  * the assumption that it is safe to positively acknowledge a request for the
1543  * ability to write to the BPF device.
1544  * Otherwise, return false but make a note that a selnotify() must be done.
1545  */
1546 static int
1547 bpf_poll(struct file *fp, int events)
1548 {
1549 	struct bpf_d *d = fp->f_bpf;
1550 	int revents;
1551 
1552 	/*
1553 	 * Refresh the PID associated with this bpf file.
1554 	 */
1555 	mutex_enter(&bpf_mtx);
1556 	d->bd_pid = curproc->p_pid;
1557 
1558 	revents = events & (POLLOUT | POLLWRNORM);
1559 	if (events & (POLLIN | POLLRDNORM)) {
1560 		/*
1561 		 * An imitation of the FIONREAD ioctl code.
1562 		 */
1563 		mutex_enter(d->bd_mtx);
1564 		mutex_enter(d->bd_buf_mtx);
1565 		if (d->bd_hlen != 0 ||
1566 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1567 			d->bd_slen != 0)) {
1568 			revents |= events & (POLLIN | POLLRDNORM);
1569 		} else {
1570 			selrecord(curlwp, &d->bd_sel);
1571 			/* Start the read timeout if necessary */
1572 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1573 				callout_reset(&d->bd_callout, d->bd_rtout,
1574 				    bpf_timed_out, d);
1575 				d->bd_state = BPF_WAITING;
1576 			}
1577 		}
1578 		mutex_exit(d->bd_buf_mtx);
1579 		mutex_exit(d->bd_mtx);
1580 	}
1581 
1582 	mutex_exit(&bpf_mtx);
1583 	return (revents);
1584 }
1585 
1586 static void
1587 filt_bpfrdetach(struct knote *kn)
1588 {
1589 	struct bpf_d *d = kn->kn_hook;
1590 
1591 	mutex_enter(d->bd_buf_mtx);
1592 	selremove_knote(&d->bd_sel, kn);
1593 	mutex_exit(d->bd_buf_mtx);
1594 }
1595 
1596 static int
1597 filt_bpfread(struct knote *kn, long hint)
1598 {
1599 	struct bpf_d *d = kn->kn_hook;
1600 	int rv;
1601 
1602 	/*
1603 	 * Refresh the PID associated with this bpf file.
1604 	 */
1605 	d->bd_pid = curproc->p_pid;
1606 
1607 	if (hint & NOTE_SUBMIT)
1608 		KASSERT(mutex_owned(d->bd_buf_mtx));
1609 	else
1610 		mutex_enter(d->bd_buf_mtx);
1611 	kn->kn_data = d->bd_hlen;
1612 	if (d->bd_immediate)
1613 		kn->kn_data += d->bd_slen;
1614 	rv = (kn->kn_data > 0);
1615 	if (hint & NOTE_SUBMIT)
1616 		KASSERT(mutex_owned(d->bd_buf_mtx));
1617 	else
1618 		mutex_exit(d->bd_buf_mtx);
1619 	return rv;
1620 }
1621 
1622 static const struct filterops bpfread_filtops = {
1623 	.f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE,
1624 	.f_attach = NULL,
1625 	.f_detach = filt_bpfrdetach,
1626 	.f_event = filt_bpfread,
1627 };
1628 
1629 static int
1630 bpf_kqfilter(struct file *fp, struct knote *kn)
1631 {
1632 	struct bpf_d *d = fp->f_bpf;
1633 
1634 	switch (kn->kn_filter) {
1635 	case EVFILT_READ:
1636 		kn->kn_fop = &bpfread_filtops;
1637 		break;
1638 
1639 	default:
1640 		return (EINVAL);
1641 	}
1642 
1643 	kn->kn_hook = d;
1644 
1645 	mutex_enter(d->bd_buf_mtx);
1646 	selrecord_knote(&d->bd_sel, kn);
1647 	mutex_exit(d->bd_buf_mtx);
1648 
1649 	return (0);
1650 }
1651 
1652 /*
1653  * Copy data from an mbuf chain into a buffer.  This code is derived
1654  * from m_copydata in sys/uipc_mbuf.c.
1655  */
1656 static void *
1657 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1658 {
1659 	const struct mbuf *m;
1660 	u_int count;
1661 	u_char *dst;
1662 
1663 	m = src_arg;
1664 	dst = dst_arg;
1665 	while (len > 0) {
1666 		if (m == NULL)
1667 			panic("bpf_mcpy");
1668 		count = uimin(m->m_len, len);
1669 		memcpy(dst, mtod(m, const void *), count);
1670 		m = m->m_next;
1671 		dst += count;
1672 		len -= count;
1673 	}
1674 	return dst_arg;
1675 }
1676 
1677 static inline u_int
1678 bpf_xfilter(struct bpf_filter **filter, void *pkt, u_int pktlen, u_int buflen)
1679 {
1680 	struct bpf_filter *filt;
1681 	uint32_t mem[BPF_MEMWORDS];
1682 	bpf_args_t args = {
1683 		.pkt = (const uint8_t *)pkt,
1684 		.wirelen = pktlen,
1685 		.buflen = buflen,
1686 		.mem = mem,
1687 		.arg = NULL
1688 	};
1689 	u_int slen;
1690 
1691 	filt = atomic_load_consume(filter);
1692 	if (filt == NULL) /* No filter means accept all. */
1693 		return (u_int)-1;
1694 
1695 	if (filt->bf_jitcode != NULL)
1696 		slen = filt->bf_jitcode(NULL, &args);
1697 	else
1698 		slen = bpf_filter_ext(NULL, filt->bf_insn, &args);
1699 	return slen;
1700 }
1701 
1702 /*
1703  * Dispatch a packet to all the listeners on interface bp.
1704  *
1705  * pkt       pointer to the packet, either a data buffer or an mbuf chain
1706  * buflen    buffer length, if pkt is a data buffer
1707  * cpfn      a function that can copy pkt into the listener's buffer
1708  * pktlen    length of the packet
1709  * direction BPF_D_IN or BPF_D_OUT
1710  */
1711 static inline void
1712 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1713     void *pkt, u_int pktlen, u_int buflen, const u_int direction)
1714 {
1715 	bool gottime = false;
1716 	struct timespec ts;
1717 	struct bpf_d *d;
1718 	int s;
1719 	u_int slen;
1720 
1721 	KASSERT(!cpu_intr_p());
1722 
1723 	/*
1724 	 * Note that the IPL does not have to be raised at this point.
1725 	 * The only problem that could arise here is that if two different
1726 	 * interfaces shared any data.  This is not the case.
1727 	 */
1728 	s = pserialize_read_enter();
1729 	BPFIF_DLIST_READER_FOREACH(d, bp) {
1730 		if (direction == BPF_D_IN) {
1731 			if (d->bd_direction == BPF_D_OUT)
1732 				continue;
1733 		} else { /* BPF_D_OUT */
1734 			if (d->bd_direction == BPF_D_IN)
1735 				continue;
1736 		}
1737 
1738 		atomic_inc_ulong(&d->bd_rcount);
1739 		BPF_STATINC(recv);
1740 
1741 		slen = bpf_xfilter(&d->bd_rfilter, pkt, pktlen, buflen);
1742 		if (slen == 0)
1743 			continue;
1744 
1745 		if (!gottime) {
1746 			gottime = true;
1747 			nanotime(&ts);
1748 		}
1749 		/* Assume catchpacket doesn't sleep */
1750 		catchpacket(d, pkt, pktlen, slen, cpfn, &ts);
1751 	}
1752 	pserialize_read_exit(s);
1753 }
1754 
1755 /*
1756  * Incoming linkage from device drivers, when the head of the packet is in
1757  * a buffer, and the tail is in an mbuf chain.
1758  */
1759 static void
1760 _bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m,
1761     u_int direction)
1762 {
1763 	u_int pktlen;
1764 	struct mbuf mb;
1765 
1766 	/* Skip outgoing duplicate packets. */
1767 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) {
1768 		m->m_flags &= ~M_PROMISC;
1769 		return;
1770 	}
1771 
1772 	pktlen = m_length(m) + dlen;
1773 
1774 	/*
1775 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1776 	 * Note that we cut corners here; we only set up what's
1777 	 * absolutely needed--this mbuf should never go anywhere else.
1778 	 */
1779 	(void)memset(&mb, 0, sizeof(mb));
1780 	mb.m_type = MT_DATA;
1781 	mb.m_next = m;
1782 	mb.m_data = data;
1783 	mb.m_len = dlen;
1784 
1785 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, direction);
1786 }
1787 
1788 /*
1789  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1790  */
1791 static void
1792 _bpf_mtap(struct bpf_if *bp, struct mbuf *m, u_int direction)
1793 {
1794 	void *(*cpfn)(void *, const void *, size_t);
1795 	u_int pktlen, buflen;
1796 	void *marg;
1797 
1798 	/* Skip outgoing duplicate packets. */
1799 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) {
1800 		m->m_flags &= ~M_PROMISC;
1801 		return;
1802 	}
1803 
1804 	pktlen = m_length(m);
1805 
1806 	/* Skip zero-sized packets. */
1807 	if (__predict_false(pktlen == 0)) {
1808 		return;
1809 	}
1810 
1811 	if (pktlen == m->m_len) {
1812 		cpfn = (void *)memcpy;
1813 		marg = mtod(m, void *);
1814 		buflen = pktlen;
1815 		KASSERT(buflen != 0);
1816 	} else {
1817 		cpfn = bpf_mcpy;
1818 		marg = m;
1819 		buflen = 0;
1820 	}
1821 
1822 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, direction);
1823 }
1824 
1825 /*
1826  * We need to prepend the address family as
1827  * a four byte field.  Cons up a dummy header
1828  * to pacify bpf.  This is safe because bpf
1829  * will only read from the mbuf (i.e., it won't
1830  * try to free it or keep a pointer a to it).
1831  */
1832 static void
1833 _bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m, u_int direction)
1834 {
1835 	struct mbuf m0;
1836 
1837 	m0.m_type = MT_DATA;
1838 	m0.m_flags = 0;
1839 	m0.m_next = m;
1840 	m0.m_nextpkt = NULL;
1841 	m0.m_owner = NULL;
1842 	m0.m_len = 4;
1843 	m0.m_data = (char *)&af;
1844 
1845 	_bpf_mtap(bp, &m0, direction);
1846 }
1847 
1848 /*
1849  * Put the SLIP pseudo-"link header" in place.
1850  * Note this M_PREPEND() should never fail,
1851  * since we know we always have enough space
1852  * in the input buffer.
1853  */
1854 static void
1855 _bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m)
1856 {
1857 	u_char *hp;
1858 
1859 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1860 	if (*m == NULL)
1861 		return;
1862 
1863 	hp = mtod(*m, u_char *);
1864 	hp[SLX_DIR] = SLIPDIR_IN;
1865 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1866 
1867 	_bpf_mtap(bp, *m, BPF_D_IN);
1868 
1869 	m_adj(*m, SLIP_HDRLEN);
1870 }
1871 
1872 /*
1873  * Put the SLIP pseudo-"link header" in
1874  * place.  The compressed header is now
1875  * at the beginning of the mbuf.
1876  */
1877 static void
1878 _bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m)
1879 {
1880 	struct mbuf m0;
1881 	u_char *hp;
1882 
1883 	m0.m_type = MT_DATA;
1884 	m0.m_flags = 0;
1885 	m0.m_next = m;
1886 	m0.m_nextpkt = NULL;
1887 	m0.m_owner = NULL;
1888 	m0.m_data = m0.m_dat;
1889 	m0.m_len = SLIP_HDRLEN;
1890 
1891 	hp = mtod(&m0, u_char *);
1892 
1893 	hp[SLX_DIR] = SLIPDIR_OUT;
1894 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1895 
1896 	_bpf_mtap(bp, &m0, BPF_D_OUT);
1897 	m_freem(m);
1898 }
1899 
1900 static struct mbuf *
1901 bpf_mbuf_enqueue(struct bpf_if *bp, struct mbuf *m)
1902 {
1903 	struct mbuf *dup;
1904 
1905 	dup = m_dup(m, 0, M_COPYALL, M_NOWAIT);
1906 	if (dup == NULL)
1907 		return NULL;
1908 
1909 	if (bp->bif_mbuf_tail != NULL) {
1910 		bp->bif_mbuf_tail->m_nextpkt = dup;
1911 	} else {
1912 		bp->bif_mbuf_head = dup;
1913 	}
1914 	bp->bif_mbuf_tail = dup;
1915 #ifdef BPF_MTAP_SOFTINT_DEBUG
1916 	log(LOG_DEBUG, "%s: enqueued mbuf=%p to %s\n",
1917 	    __func__, dup, bp->bif_ifp->if_xname);
1918 #endif
1919 
1920 	return dup;
1921 }
1922 
1923 static struct mbuf *
1924 bpf_mbuf_dequeue(struct bpf_if *bp)
1925 {
1926 	struct mbuf *m;
1927 	int s;
1928 
1929 	/* XXX NOMPSAFE: assumed running on one CPU */
1930 	s = splnet();
1931 	m = bp->bif_mbuf_head;
1932 	if (m != NULL) {
1933 		bp->bif_mbuf_head = m->m_nextpkt;
1934 		m->m_nextpkt = NULL;
1935 
1936 		if (bp->bif_mbuf_head == NULL)
1937 			bp->bif_mbuf_tail = NULL;
1938 #ifdef BPF_MTAP_SOFTINT_DEBUG
1939 		log(LOG_DEBUG, "%s: dequeued mbuf=%p from %s\n",
1940 		    __func__, m, bp->bif_ifp->if_xname);
1941 #endif
1942 	}
1943 	splx(s);
1944 
1945 	return m;
1946 }
1947 
1948 static void
1949 bpf_mtap_si(void *arg)
1950 {
1951 	struct bpf_if *bp = arg;
1952 	struct mbuf *m;
1953 
1954 	while ((m = bpf_mbuf_dequeue(bp)) != NULL) {
1955 #ifdef BPF_MTAP_SOFTINT_DEBUG
1956 		log(LOG_DEBUG, "%s: tapping mbuf=%p on %s\n",
1957 		    __func__, m, bp->bif_ifp->if_xname);
1958 #endif
1959 		bpf_ops->bpf_mtap(bp, m, BPF_D_IN);
1960 		m_freem(m);
1961 	}
1962 }
1963 
1964 static void
1965 _bpf_mtap_softint(struct ifnet *ifp, struct mbuf *m)
1966 {
1967 	struct bpf_if *bp = ifp->if_bpf;
1968 	struct mbuf *dup;
1969 
1970 	KASSERT(cpu_intr_p());
1971 
1972 	/* To avoid extra invocations of the softint */
1973 	if (BPFIF_DLIST_READER_EMPTY(bp))
1974 		return;
1975 	KASSERT(bp->bif_si != NULL);
1976 
1977 	dup = bpf_mbuf_enqueue(bp, m);
1978 	if (dup != NULL)
1979 		softint_schedule(bp->bif_si);
1980 }
1981 
1982 static int
1983 bpf_hdrlen(struct bpf_d *d)
1984 {
1985 	int hdrlen = d->bd_bif->bif_hdrlen;
1986 	/*
1987 	 * Compute the length of the bpf header.  This is not necessarily
1988 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1989 	 * that the network layer header begins on a longword boundary (for
1990 	 * performance reasons and to alleviate alignment restrictions).
1991 	 */
1992 #ifdef _LP64
1993 	if (d->bd_compat32)
1994 		return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen);
1995 	else
1996 #endif
1997 		return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen);
1998 }
1999 
2000 /*
2001  * Move the packet data from interface memory (pkt) into the
2002  * store buffer. Call the wakeup functions if it's time to wake up
2003  * a listener (buffer full), "cpfn" is the routine called to do the
2004  * actual data transfer. memcpy is passed in to copy contiguous chunks,
2005  * while bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
2006  * pkt is really an mbuf.
2007  */
2008 static void
2009 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2010     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
2011 {
2012 	char *h;
2013 	int totlen, curlen, caplen;
2014 	int hdrlen = bpf_hdrlen(d);
2015 	int do_wakeup = 0;
2016 
2017 	atomic_inc_ulong(&d->bd_ccount);
2018 	BPF_STATINC(capt);
2019 	/*
2020 	 * Figure out how many bytes to move.  If the packet is
2021 	 * greater or equal to the snapshot length, transfer that
2022 	 * much.  Otherwise, transfer the whole packet (unless
2023 	 * we hit the buffer size limit).
2024 	 */
2025 	totlen = hdrlen + uimin(snaplen, pktlen);
2026 	if (totlen > d->bd_bufsize)
2027 		totlen = d->bd_bufsize;
2028 	/*
2029 	 * If we adjusted totlen to fit the bufsize, it could be that
2030 	 * totlen is smaller than hdrlen because of the link layer header.
2031 	 */
2032 	caplen = totlen - hdrlen;
2033 	if (caplen < 0)
2034 		caplen = 0;
2035 
2036 	mutex_enter(d->bd_buf_mtx);
2037 	/*
2038 	 * Round up the end of the previous packet to the next longword.
2039 	 */
2040 #ifdef _LP64
2041 	if (d->bd_compat32)
2042 		curlen = BPF_WORDALIGN32(d->bd_slen);
2043 	else
2044 #endif
2045 		curlen = BPF_WORDALIGN(d->bd_slen);
2046 	if (curlen + totlen > d->bd_bufsize) {
2047 		/*
2048 		 * This packet will overflow the storage buffer.
2049 		 * Rotate the buffers if we can, then wakeup any
2050 		 * pending reads.
2051 		 */
2052 		if (d->bd_fbuf == NULL) {
2053 			mutex_exit(d->bd_buf_mtx);
2054 			/*
2055 			 * We haven't completed the previous read yet,
2056 			 * so drop the packet.
2057 			 */
2058 			atomic_inc_ulong(&d->bd_dcount);
2059 			BPF_STATINC(drop);
2060 			return;
2061 		}
2062 		ROTATE_BUFFERS(d);
2063 		do_wakeup = 1;
2064 		curlen = 0;
2065 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
2066 		/*
2067 		 * Immediate mode is set, or the read timeout has
2068 		 * already expired during a select call.  A packet
2069 		 * arrived, so the reader should be woken up.
2070 		 */
2071 		do_wakeup = 1;
2072 	}
2073 
2074 	/*
2075 	 * Append the bpf header.
2076 	 */
2077 	h = (char *)d->bd_sbuf + curlen;
2078 #ifdef _LP64
2079 	if (d->bd_compat32) {
2080 		struct bpf_hdr32 *hp32;
2081 
2082 		hp32 = (struct bpf_hdr32 *)h;
2083 		hp32->bh_tstamp.tv_sec = ts->tv_sec;
2084 		hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
2085 		hp32->bh_datalen = pktlen;
2086 		hp32->bh_hdrlen = hdrlen;
2087 		hp32->bh_caplen = caplen;
2088 	} else
2089 #endif
2090 	{
2091 		struct bpf_hdr *hp;
2092 
2093 		hp = (struct bpf_hdr *)h;
2094 		hp->bh_tstamp.tv_sec = ts->tv_sec;
2095 		hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
2096 		hp->bh_datalen = pktlen;
2097 		hp->bh_hdrlen = hdrlen;
2098 		hp->bh_caplen = caplen;
2099 	}
2100 
2101 	/*
2102 	 * Copy the packet data into the store buffer and update its length.
2103 	 */
2104 	(*cpfn)(h + hdrlen, pkt, caplen);
2105 	d->bd_slen = curlen + totlen;
2106 
2107 	/*
2108 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
2109 	 * will cause filt_bpfread() to be called with it adjusted.
2110 	 */
2111 	if (do_wakeup)
2112 		bpf_wakeup(d);
2113 
2114 	mutex_exit(d->bd_buf_mtx);
2115 }
2116 
2117 /*
2118  * Initialize all nonzero fields of a descriptor.
2119  */
2120 static int
2121 bpf_allocbufs(struct bpf_d *d)
2122 {
2123 
2124 	d->bd_fbuf = kmem_zalloc(d->bd_bufsize, KM_NOSLEEP);
2125 	if (!d->bd_fbuf)
2126 		return (ENOBUFS);
2127 	d->bd_sbuf = kmem_zalloc(d->bd_bufsize, KM_NOSLEEP);
2128 	if (!d->bd_sbuf) {
2129 		kmem_free(d->bd_fbuf, d->bd_bufsize);
2130 		return (ENOBUFS);
2131 	}
2132 	d->bd_slen = 0;
2133 	d->bd_hlen = 0;
2134 	return (0);
2135 }
2136 
2137 static void
2138 bpf_free_filter(struct bpf_filter *filter)
2139 {
2140 
2141 	KASSERT(filter != NULL);
2142 
2143 	if (filter->bf_insn != NULL)
2144 		kmem_free(filter->bf_insn, filter->bf_size);
2145 	if (filter->bf_jitcode != NULL)
2146 		bpf_jit_freecode(filter->bf_jitcode);
2147 	kmem_free(filter, sizeof(*filter));
2148 }
2149 
2150 /*
2151  * Free buffers currently in use by a descriptor.
2152  * Called on close.
2153  */
2154 static void
2155 bpf_freed(struct bpf_d *d)
2156 {
2157 	/*
2158 	 * We don't need to lock out interrupts since this descriptor has
2159 	 * been detached from its interface and it yet hasn't been marked
2160 	 * free.
2161 	 */
2162 	if (d->bd_sbuf != NULL) {
2163 		kmem_free(d->bd_sbuf, d->bd_bufsize);
2164 		if (d->bd_hbuf != NULL)
2165 			kmem_free(d->bd_hbuf, d->bd_bufsize);
2166 		if (d->bd_fbuf != NULL)
2167 			kmem_free(d->bd_fbuf, d->bd_bufsize);
2168 	}
2169 	if (d->bd_rfilter != NULL) {
2170 		bpf_free_filter(d->bd_rfilter);
2171 		d->bd_rfilter = NULL;
2172 	}
2173 	if (d->bd_wfilter != NULL) {
2174 		bpf_free_filter(d->bd_wfilter);
2175 		d->bd_wfilter = NULL;
2176 	}
2177 	d->bd_jitcode = NULL;
2178 }
2179 
2180 /*
2181  * Attach an interface to bpf.  dlt is the link layer type;
2182  * hdrlen is the fixed size of the link header for the specified dlt
2183  * (variable length headers not yet supported).
2184  */
2185 static void
2186 _bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2187 {
2188 	struct bpf_if *bp;
2189 
2190 	bp = kmem_alloc(sizeof(*bp), KM_SLEEP);
2191 
2192 	mutex_enter(&bpf_mtx);
2193 	bp->bif_driverp = driverp;
2194 	bp->bif_ifp = ifp;
2195 	bp->bif_dlt = dlt;
2196 	bp->bif_si = NULL;
2197 	BPF_IFLIST_ENTRY_INIT(bp);
2198 	PSLIST_INIT(&bp->bif_dlist_head);
2199 	psref_target_init(&bp->bif_psref, bpf_psref_class);
2200 	SLIST_INIT(&bp->bif_trackers);
2201 
2202 	BPF_IFLIST_WRITER_INSERT_HEAD(bp);
2203 
2204 	*bp->bif_driverp = NULL;
2205 
2206 	bp->bif_hdrlen = hdrlen;
2207 	mutex_exit(&bpf_mtx);
2208 #if 0
2209 	printf("bpf: %s attached with dlt %x\n", ifp->if_xname, dlt);
2210 #endif
2211 }
2212 
2213 static void
2214 _bpf_mtap_softint_init(struct ifnet *ifp)
2215 {
2216 	struct bpf_if *bp;
2217 
2218 	mutex_enter(&bpf_mtx);
2219 	BPF_IFLIST_WRITER_FOREACH(bp) {
2220 		if (bp->bif_ifp != ifp)
2221 			continue;
2222 
2223 		bp->bif_mbuf_head = NULL;
2224 		bp->bif_mbuf_tail = NULL;
2225 		bp->bif_si = softint_establish(SOFTINT_NET, bpf_mtap_si, bp);
2226 		if (bp->bif_si == NULL)
2227 			panic("%s: softint_establish() failed", __func__);
2228 		break;
2229 	}
2230 	mutex_exit(&bpf_mtx);
2231 
2232 	if (bp == NULL)
2233 		panic("%s: no bpf_if found for %s", __func__, ifp->if_xname);
2234 }
2235 
2236 /*
2237  * Remove an interface from bpf.
2238  */
2239 static void
2240 _bpfdetach(struct ifnet *ifp)
2241 {
2242 	struct bpf_if *bp;
2243 	struct bpf_d *d;
2244 	int s;
2245 
2246 	mutex_enter(&bpf_mtx);
2247 	/* Nuke the vnodes for any open instances */
2248 again_d:
2249 	BPF_DLIST_WRITER_FOREACH(d) {
2250 		mutex_enter(d->bd_mtx);
2251 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
2252 			/*
2253 			 * Detach the descriptor from an interface now.
2254 			 * It will be free'ed later by close routine.
2255 			 */
2256 			bpf_detachd(d);
2257 			mutex_exit(d->bd_mtx);
2258 			goto again_d;
2259 		}
2260 		mutex_exit(d->bd_mtx);
2261 	}
2262 
2263 again:
2264 	BPF_IFLIST_WRITER_FOREACH(bp) {
2265 		if (bp->bif_ifp == ifp) {
2266 			BPF_IFLIST_WRITER_REMOVE(bp);
2267 
2268 			pserialize_perform(bpf_psz);
2269 			psref_target_destroy(&bp->bif_psref, bpf_psref_class);
2270 
2271 			while (!SLIST_EMPTY(&bp->bif_trackers)) {
2272 				struct bpf_event_tracker *t =
2273 				    SLIST_FIRST(&bp->bif_trackers);
2274 				SLIST_REMOVE_HEAD(&bp->bif_trackers,
2275 				    bet_entries);
2276 				kmem_free(t, sizeof(*t));
2277 			}
2278 
2279 			BPF_IFLIST_ENTRY_DESTROY(bp);
2280 			if (bp->bif_si != NULL) {
2281 				/* XXX NOMPSAFE: assumed running on one CPU */
2282 				s = splnet();
2283 				while (bp->bif_mbuf_head != NULL) {
2284 					struct mbuf *m = bp->bif_mbuf_head;
2285 					bp->bif_mbuf_head = m->m_nextpkt;
2286 					m_freem(m);
2287 				}
2288 				splx(s);
2289 				softint_disestablish(bp->bif_si);
2290 			}
2291 			kmem_free(bp, sizeof(*bp));
2292 			goto again;
2293 		}
2294 	}
2295 	mutex_exit(&bpf_mtx);
2296 }
2297 
2298 /*
2299  * Change the data link type of a interface.
2300  */
2301 static void
2302 _bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2303 {
2304 	struct bpf_if *bp;
2305 
2306 	mutex_enter(&bpf_mtx);
2307 	BPF_IFLIST_WRITER_FOREACH(bp) {
2308 		if (bp->bif_driverp == &ifp->if_bpf)
2309 			break;
2310 	}
2311 	if (bp == NULL)
2312 		panic("bpf_change_type");
2313 
2314 	bp->bif_dlt = dlt;
2315 
2316 	bp->bif_hdrlen = hdrlen;
2317 	mutex_exit(&bpf_mtx);
2318 }
2319 
2320 /*
2321  * Get a list of available data link type of the interface.
2322  */
2323 static int
2324 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2325 {
2326 	int n, error;
2327 	struct ifnet *ifp;
2328 	struct bpf_if *bp;
2329 	int s, bound;
2330 
2331 	KASSERT(mutex_owned(d->bd_mtx));
2332 
2333 	ifp = d->bd_bif->bif_ifp;
2334 	n = 0;
2335 	error = 0;
2336 
2337 	bound = curlwp_bind();
2338 	s = pserialize_read_enter();
2339 	BPF_IFLIST_READER_FOREACH(bp) {
2340 		if (bp->bif_ifp != ifp)
2341 			continue;
2342 		if (bfl->bfl_list != NULL) {
2343 			struct psref psref;
2344 
2345 			if (n >= bfl->bfl_len) {
2346 				pserialize_read_exit(s);
2347 				return ENOMEM;
2348 			}
2349 
2350 			bpf_if_acquire(bp, &psref);
2351 			pserialize_read_exit(s);
2352 
2353 			error = copyout(&bp->bif_dlt,
2354 			    bfl->bfl_list + n, sizeof(u_int));
2355 
2356 			s = pserialize_read_enter();
2357 			bpf_if_release(bp, &psref);
2358 		}
2359 		n++;
2360 	}
2361 	pserialize_read_exit(s);
2362 	curlwp_bindx(bound);
2363 
2364 	bfl->bfl_len = n;
2365 	return error;
2366 }
2367 
2368 /*
2369  * Set the data link type of a BPF instance.
2370  */
2371 static int
2372 bpf_setdlt(struct bpf_d *d, u_int dlt)
2373 {
2374 	int error, opromisc;
2375 	struct ifnet *ifp;
2376 	struct bpf_if *bp;
2377 
2378 	KASSERT(mutex_owned(&bpf_mtx));
2379 	KASSERT(mutex_owned(d->bd_mtx));
2380 
2381 	if (d->bd_bif->bif_dlt == dlt)
2382 		return 0;
2383 	ifp = d->bd_bif->bif_ifp;
2384 	BPF_IFLIST_WRITER_FOREACH(bp) {
2385 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2386 			break;
2387 	}
2388 	if (bp == NULL)
2389 		return EINVAL;
2390 	opromisc = d->bd_promisc;
2391 	bpf_detachd(d);
2392 	BPFIF_DLIST_ENTRY_INIT(d);
2393 	bpf_attachd(d, bp);
2394 	reset_d(d);
2395 	if (opromisc) {
2396 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
2397 		error = ifpromisc(bp->bif_ifp, 1);
2398 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2399 		if (error)
2400 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
2401 			    bp->bif_ifp->if_xname, error);
2402 		else
2403 			d->bd_promisc = 1;
2404 	}
2405 	return 0;
2406 }
2407 
2408 static int
2409 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
2410 {
2411 	int newsize, error;
2412 	struct sysctlnode node;
2413 
2414 	node = *rnode;
2415 	node.sysctl_data = &newsize;
2416 	newsize = bpf_maxbufsize;
2417 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2418 	if (error || newp == NULL)
2419 		return (error);
2420 
2421 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
2422 		return (EINVAL);
2423 
2424 	bpf_maxbufsize = newsize;
2425 
2426 	return (0);
2427 }
2428 
2429 #if defined(MODULAR) || defined(BPFJIT)
2430 static int
2431 sysctl_net_bpf_jit(SYSCTLFN_ARGS)
2432 {
2433 	bool newval;
2434 	int error;
2435 	struct sysctlnode node;
2436 
2437 	node = *rnode;
2438 	node.sysctl_data = &newval;
2439 	newval = bpf_jit;
2440 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2441 	if (error != 0 || newp == NULL)
2442 		return error;
2443 
2444 	bpf_jit = newval;
2445 	if (newval && bpfjit_module_ops.bj_generate_code == NULL) {
2446 		printf("JIT compilation is postponed "
2447 		    "until after bpfjit module is loaded\n");
2448 	}
2449 
2450 	return 0;
2451 }
2452 #endif
2453 
2454 static int
2455 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
2456 {
2457 	int    error, elem_count;
2458 	struct bpf_d	 *dp;
2459 	struct bpf_d_ext  dpe;
2460 	size_t len, needed, elem_size, out_size;
2461 	char   *sp;
2462 
2463 	if (namelen == 1 && name[0] == CTL_QUERY)
2464 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
2465 
2466 	if (namelen != 2)
2467 		return (EINVAL);
2468 
2469 	/* BPF peers is privileged information. */
2470 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
2471 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
2472 	if (error)
2473 		return (EPERM);
2474 
2475 	len = (oldp != NULL) ? *oldlenp : 0;
2476 	sp = oldp;
2477 	elem_size = name[0];
2478 	elem_count = name[1];
2479 	out_size = MIN(sizeof(dpe), elem_size);
2480 	needed = 0;
2481 
2482 	if (elem_size < 1 || elem_count < 0)
2483 		return (EINVAL);
2484 
2485 	mutex_enter(&bpf_mtx);
2486 	BPF_DLIST_WRITER_FOREACH(dp) {
2487 		if (len >= elem_size && elem_count > 0) {
2488 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
2489 			BPF_EXT(bufsize);
2490 			BPF_EXT(promisc);
2491 			BPF_EXT(state);
2492 			BPF_EXT(immediate);
2493 			BPF_EXT(hdrcmplt);
2494 			BPF_EXT(direction);
2495 			BPF_EXT(pid);
2496 			BPF_EXT(rcount);
2497 			BPF_EXT(dcount);
2498 			BPF_EXT(ccount);
2499 #undef BPF_EXT
2500 			mutex_enter(dp->bd_mtx);
2501 			if (dp->bd_bif)
2502 				(void)strlcpy(dpe.bde_ifname,
2503 				    dp->bd_bif->bif_ifp->if_xname,
2504 				    IFNAMSIZ - 1);
2505 			else
2506 				dpe.bde_ifname[0] = '\0';
2507 			dpe.bde_locked = dp->bd_locked;
2508 			mutex_exit(dp->bd_mtx);
2509 
2510 			error = copyout(&dpe, sp, out_size);
2511 			if (error)
2512 				break;
2513 			sp += elem_size;
2514 			len -= elem_size;
2515 		}
2516 		needed += elem_size;
2517 		if (elem_count > 0 && elem_count != INT_MAX)
2518 			elem_count--;
2519 	}
2520 	mutex_exit(&bpf_mtx);
2521 
2522 	*oldlenp = needed;
2523 
2524 	return (error);
2525 }
2526 
2527 static void
2528 bpf_stats(void *p, void *arg, struct cpu_info *ci __unused)
2529 {
2530 	struct bpf_stat *const stats = p;
2531 	struct bpf_stat *sum = arg;
2532 
2533 	int s = splnet();
2534 
2535 	sum->bs_recv += stats->bs_recv;
2536 	sum->bs_drop += stats->bs_drop;
2537 	sum->bs_capt += stats->bs_capt;
2538 
2539 	splx(s);
2540 }
2541 
2542 static int
2543 bpf_sysctl_gstats_handler(SYSCTLFN_ARGS)
2544 {
2545 	struct sysctlnode node;
2546 	int error;
2547 	struct bpf_stat sum;
2548 
2549 	memset(&sum, 0, sizeof(sum));
2550 	node = *rnode;
2551 
2552 	percpu_foreach_xcall(bpf_gstats_percpu, XC_HIGHPRI_IPL(IPL_SOFTNET),
2553 	    bpf_stats, &sum);
2554 
2555 	node.sysctl_data = &sum;
2556 	node.sysctl_size = sizeof(sum);
2557 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2558 	if (error != 0 || newp == NULL)
2559 		return error;
2560 
2561 	return 0;
2562 }
2563 
2564 SYSCTL_SETUP(sysctl_net_bpf_setup, "bpf sysctls")
2565 {
2566 	const struct sysctlnode *node;
2567 
2568 	node = NULL;
2569 	sysctl_createv(clog, 0, NULL, &node,
2570 	    CTLFLAG_PERMANENT,
2571 	    CTLTYPE_NODE, "bpf",
2572 	    SYSCTL_DESCR("BPF options"),
2573 	    NULL, 0, NULL, 0,
2574 	    CTL_NET, CTL_CREATE, CTL_EOL);
2575 	if (node != NULL) {
2576 #if defined(MODULAR) || defined(BPFJIT)
2577 		sysctl_createv(clog, 0, NULL, NULL,
2578 		    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2579 		    CTLTYPE_BOOL, "jit",
2580 		    SYSCTL_DESCR("Toggle Just-In-Time compilation"),
2581 		    sysctl_net_bpf_jit, 0, &bpf_jit, 0,
2582 		    CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2583 #endif
2584 		sysctl_createv(clog, 0, NULL, NULL,
2585 		    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2586 		    CTLTYPE_INT, "maxbufsize",
2587 		    SYSCTL_DESCR("Maximum size for data capture buffer"),
2588 		    sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
2589 		    CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2590 		sysctl_createv(clog, 0, NULL, NULL,
2591 		    CTLFLAG_PERMANENT,
2592 		    CTLTYPE_STRUCT, "stats",
2593 		    SYSCTL_DESCR("BPF stats"),
2594 		    bpf_sysctl_gstats_handler, 0, NULL, 0,
2595 		    CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2596 		sysctl_createv(clog, 0, NULL, NULL,
2597 		    CTLFLAG_PERMANENT,
2598 		    CTLTYPE_STRUCT, "peers",
2599 		    SYSCTL_DESCR("BPF peers"),
2600 		    sysctl_net_bpf_peers, 0, NULL, 0,
2601 		    CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2602 	}
2603 
2604 }
2605 
2606 static int
2607 _bpf_register_track_event(struct bpf_if **driverp,
2608     void (*_fun)(struct bpf_if *, struct ifnet *, int, int))
2609 {
2610 	struct bpf_if *bp;
2611 	struct bpf_event_tracker *t;
2612 	int ret = ENOENT;
2613 
2614 	t = kmem_zalloc(sizeof(*t), KM_SLEEP);
2615 	if (!t)
2616 		return ENOMEM;
2617 	t->bet_notify = _fun;
2618 
2619 	mutex_enter(&bpf_mtx);
2620 	BPF_IFLIST_WRITER_FOREACH(bp) {
2621 		if (bp->bif_driverp != driverp)
2622 			continue;
2623 		SLIST_INSERT_HEAD(&bp->bif_trackers, t, bet_entries);
2624 		ret = 0;
2625 		break;
2626 	}
2627 	mutex_exit(&bpf_mtx);
2628 
2629 	return ret;
2630 }
2631 
2632 static int
2633 _bpf_deregister_track_event(struct bpf_if **driverp,
2634     void (*_fun)(struct bpf_if *, struct ifnet *, int, int))
2635 {
2636 	struct bpf_if *bp;
2637 	struct bpf_event_tracker *t = NULL;
2638 	int ret = ENOENT;
2639 
2640 	mutex_enter(&bpf_mtx);
2641 	BPF_IFLIST_WRITER_FOREACH(bp) {
2642 		if (bp->bif_driverp != driverp)
2643 			continue;
2644 		SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) {
2645 			if (t->bet_notify == _fun) {
2646 				ret = 0;
2647 				break;
2648 			}
2649 		}
2650 		if (ret == 0)
2651 			break;
2652 	}
2653 	if (ret == 0 && t && t->bet_notify == _fun) {
2654 		SLIST_REMOVE(&bp->bif_trackers, t, bpf_event_tracker,
2655 		    bet_entries);
2656 	}
2657 	mutex_exit(&bpf_mtx);
2658 	if (ret == 0)
2659 		kmem_free(t, sizeof(*t));
2660 	return ret;
2661 }
2662 
2663 struct bpf_ops bpf_ops_kernel = {
2664 	.bpf_attach =		_bpfattach,
2665 	.bpf_detach =		_bpfdetach,
2666 	.bpf_change_type =	_bpf_change_type,
2667 	.bpf_register_track_event = _bpf_register_track_event,
2668 	.bpf_deregister_track_event = _bpf_deregister_track_event,
2669 
2670 	.bpf_mtap =		_bpf_mtap,
2671 	.bpf_mtap2 =		_bpf_mtap2,
2672 	.bpf_mtap_af =		_bpf_mtap_af,
2673 	.bpf_mtap_sl_in =	_bpf_mtap_sl_in,
2674 	.bpf_mtap_sl_out =	_bpf_mtap_sl_out,
2675 
2676 	.bpf_mtap_softint =		_bpf_mtap_softint,
2677 	.bpf_mtap_softint_init =	_bpf_mtap_softint_init,
2678 };
2679 
2680 MODULE(MODULE_CLASS_DRIVER, bpf, "bpf_filter");
2681 
2682 static int
2683 bpf_modcmd(modcmd_t cmd, void *arg)
2684 {
2685 #ifdef _MODULE
2686 	devmajor_t bmajor, cmajor;
2687 #endif
2688 	int error = 0;
2689 
2690 	switch (cmd) {
2691 	case MODULE_CMD_INIT:
2692 		bpf_init();
2693 #ifdef _MODULE
2694 		bmajor = cmajor = NODEVMAJOR;
2695 		error = devsw_attach("bpf", NULL, &bmajor,
2696 		    &bpf_cdevsw, &cmajor);
2697 		if (error)
2698 			break;
2699 #endif
2700 
2701 		bpf_ops_handover_enter(&bpf_ops_kernel);
2702 		atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel);
2703 		bpf_ops_handover_exit();
2704 		break;
2705 
2706 	case MODULE_CMD_FINI:
2707 		/*
2708 		 * While there is no reference counting for bpf callers,
2709 		 * unload could at least in theory be done similarly to
2710 		 * system call disestablishment.  This should even be
2711 		 * a little simpler:
2712 		 *
2713 		 * 1) replace op vector with stubs
2714 		 * 2) post update to all cpus with xc
2715 		 * 3) check that nobody is in bpf anymore
2716 		 *    (it's doubtful we'd want something like l_sysent,
2717 		 *     but we could do something like *signed* percpu
2718 		 *     counters.  if the sum is 0, we're good).
2719 		 * 4) if fail, unroll changes
2720 		 *
2721 		 * NOTE: change won't be atomic to the outside.  some
2722 		 * packets may be not captured even if unload is
2723 		 * not successful.  I think packet capture not working
2724 		 * is a perfectly logical consequence of trying to
2725 		 * disable packet capture.
2726 		 */
2727 		error = EOPNOTSUPP;
2728 		break;
2729 
2730 	default:
2731 		error = ENOTTY;
2732 		break;
2733 	}
2734 
2735 	return error;
2736 }
2737