xref: /netbsd-src/sys/net/bpf.c (revision c38e7cc395b1472a774ff828e46123de44c628e9)
1 /*	$NetBSD: bpf.c,v 1.223 2018/01/25 02:45:02 ozaki-r 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 1.223 2018/01/25 02:45:02 ozaki-r Exp $");
43 
44 #if defined(_KERNEL_OPT)
45 #include "opt_bpf.h"
46 #include "sl.h"
47 #include "strip.h"
48 #include "opt_net_mpsafe.h"
49 #endif
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/mbuf.h>
54 #include <sys/buf.h>
55 #include <sys/time.h>
56 #include <sys/proc.h>
57 #include <sys/ioctl.h>
58 #include <sys/conf.h>
59 #include <sys/vnode.h>
60 #include <sys/queue.h>
61 #include <sys/stat.h>
62 #include <sys/module.h>
63 #include <sys/atomic.h>
64 #include <sys/cpu.h>
65 
66 #include <sys/file.h>
67 #include <sys/filedesc.h>
68 #include <sys/tty.h>
69 #include <sys/uio.h>
70 
71 #include <sys/protosw.h>
72 #include <sys/socket.h>
73 #include <sys/errno.h>
74 #include <sys/kernel.h>
75 #include <sys/poll.h>
76 #include <sys/sysctl.h>
77 #include <sys/kauth.h>
78 #include <sys/syslog.h>
79 #include <sys/percpu.h>
80 #include <sys/pserialize.h>
81 #include <sys/lwp.h>
82 
83 #include <net/if.h>
84 #include <net/slip.h>
85 
86 #include <net/bpf.h>
87 #include <net/bpfdesc.h>
88 #include <net/bpfjit.h>
89 
90 #include <net/if_arc.h>
91 #include <net/if_ether.h>
92 
93 #include <netinet/in.h>
94 #include <netinet/if_inarp.h>
95 
96 
97 #include <compat/sys/sockio.h>
98 
99 #ifndef BPF_BUFSIZE
100 /*
101  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
102  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
103  */
104 # define BPF_BUFSIZE 32768
105 #endif
106 
107 #define PRINET  26			/* interruptible */
108 
109 /*
110  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
111  * XXX the default values should be computed dynamically based
112  * on available memory size and available mbuf clusters.
113  */
114 static int bpf_bufsize = BPF_BUFSIZE;
115 static int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
116 static bool bpf_jit = false;
117 
118 struct bpfjit_ops bpfjit_module_ops = {
119 	.bj_generate_code = NULL,
120 	.bj_free_code = NULL
121 };
122 
123 /*
124  * Global BPF statistics returned by net.bpf.stats sysctl.
125  */
126 static struct percpu	*bpf_gstats_percpu; /* struct bpf_stat */
127 
128 #define BPF_STATINC(id)					\
129 	{						\
130 		struct bpf_stat *__stats =		\
131 		    percpu_getref(bpf_gstats_percpu);	\
132 		__stats->bs_##id++;			\
133 		percpu_putref(bpf_gstats_percpu);	\
134 	}
135 
136 /*
137  * Locking notes:
138  * - bpf_mtx (adaptive mutex) protects:
139  *   - Gobal lists: bpf_iflist and bpf_dlist
140  *   - struct bpf_if
141  *   - bpf_close
142  *   - bpf_psz (pserialize)
143  * - struct bpf_d has two mutexes:
144  *   - bd_buf_mtx (spin mutex) protects the buffers that can be accessed
145  *     on packet tapping
146  *   - bd_mtx (adaptive mutex) protects member variables other than the buffers
147  * - Locking order: bpf_mtx => bpf_d#bd_mtx => bpf_d#bd_buf_mtx
148  * - struct bpf_d obtained via fp->f_bpf in bpf_read and bpf_write is
149  *   never freed because struct bpf_d is only freed in bpf_close and
150  *   bpf_close never be called while executing bpf_read and bpf_write
151  * - A filter that is assigned to bpf_d can be replaced with another filter
152  *   while tapping packets, so it needs to be done atomically
153  * - struct bpf_d is iterated on bpf_dlist with psz
154  * - struct bpf_if is iterated on bpf_iflist with psz or psref
155  */
156 /*
157  * Use a mutex to avoid a race condition between gathering the stats/peers
158  * and opening/closing the device.
159  */
160 static kmutex_t bpf_mtx;
161 
162 static struct psref_class	*bpf_psref_class __read_mostly;
163 static pserialize_t		bpf_psz;
164 
165 static inline void
166 bpf_if_acquire(struct bpf_if *bp, struct psref *psref)
167 {
168 
169 	psref_acquire(psref, &bp->bif_psref, bpf_psref_class);
170 }
171 
172 static inline void
173 bpf_if_release(struct bpf_if *bp, struct psref *psref)
174 {
175 
176 	psref_release(psref, &bp->bif_psref, bpf_psref_class);
177 }
178 
179 /*
180  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
181  *  bpf_dtab holds the descriptors, indexed by minor device #
182  */
183 static struct pslist_head bpf_iflist;
184 static struct pslist_head bpf_dlist;
185 
186 /* Macros for bpf_d on bpf_dlist */
187 #define BPF_DLIST_WRITER_INSERT_HEAD(__d)				\
188 	PSLIST_WRITER_INSERT_HEAD(&bpf_dlist, (__d), bd_bpf_dlist_entry)
189 #define BPF_DLIST_READER_FOREACH(__d)					\
190 	PSLIST_READER_FOREACH((__d), &bpf_dlist, struct bpf_d,		\
191 	                      bd_bpf_dlist_entry)
192 #define BPF_DLIST_WRITER_FOREACH(__d)					\
193 	PSLIST_WRITER_FOREACH((__d), &bpf_dlist, struct bpf_d,		\
194 	                      bd_bpf_dlist_entry)
195 #define BPF_DLIST_ENTRY_INIT(__d)					\
196 	PSLIST_ENTRY_INIT((__d), bd_bpf_dlist_entry)
197 #define BPF_DLIST_WRITER_REMOVE(__d)					\
198 	PSLIST_WRITER_REMOVE((__d), bd_bpf_dlist_entry)
199 #define BPF_DLIST_ENTRY_DESTROY(__d)					\
200 	PSLIST_ENTRY_DESTROY((__d), bd_bpf_dlist_entry)
201 
202 /* Macros for bpf_if on bpf_iflist */
203 #define BPF_IFLIST_WRITER_INSERT_HEAD(__bp)				\
204 	PSLIST_WRITER_INSERT_HEAD(&bpf_iflist, (__bp), bif_iflist_entry)
205 #define BPF_IFLIST_READER_FOREACH(__bp)					\
206 	PSLIST_READER_FOREACH((__bp), &bpf_iflist, struct bpf_if,	\
207 	                      bif_iflist_entry)
208 #define BPF_IFLIST_WRITER_FOREACH(__bp)					\
209 	PSLIST_WRITER_FOREACH((__bp), &bpf_iflist, struct bpf_if,	\
210 	                      bif_iflist_entry)
211 #define BPF_IFLIST_WRITER_REMOVE(__bp)					\
212 	PSLIST_WRITER_REMOVE((__bp), bif_iflist_entry)
213 #define BPF_IFLIST_ENTRY_INIT(__bp)					\
214 	PSLIST_ENTRY_INIT((__bp), bif_iflist_entry)
215 #define BPF_IFLIST_ENTRY_DESTROY(__bp)					\
216 	PSLIST_ENTRY_DESTROY((__bp), bif_iflist_entry)
217 
218 /* Macros for bpf_d on bpf_if#bif_dlist_pslist */
219 #define BPFIF_DLIST_READER_FOREACH(__d, __bp)				\
220 	PSLIST_READER_FOREACH((__d), &(__bp)->bif_dlist_head, struct bpf_d, \
221 	                      bd_bif_dlist_entry)
222 #define BPFIF_DLIST_WRITER_INSERT_HEAD(__bp, __d)			\
223 	PSLIST_WRITER_INSERT_HEAD(&(__bp)->bif_dlist_head, (__d),	\
224 	                          bd_bif_dlist_entry)
225 #define BPFIF_DLIST_WRITER_REMOVE(__d)					\
226 	PSLIST_WRITER_REMOVE((__d), bd_bif_dlist_entry)
227 #define BPFIF_DLIST_ENTRY_INIT(__d)					\
228 	PSLIST_ENTRY_INIT((__d), bd_bif_dlist_entry)
229 #define	BPFIF_DLIST_READER_EMPTY(__bp)					\
230 	(PSLIST_READER_FIRST(&(__bp)->bif_dlist_head, struct bpf_d,	\
231 	                     bd_bif_dlist_entry) == NULL)
232 #define	BPFIF_DLIST_WRITER_EMPTY(__bp)					\
233 	(PSLIST_WRITER_FIRST(&(__bp)->bif_dlist_head, struct bpf_d,	\
234 	                     bd_bif_dlist_entry) == NULL)
235 #define BPFIF_DLIST_ENTRY_DESTROY(__d)					\
236 	PSLIST_ENTRY_DESTROY((__d), bd_bif_dlist_entry)
237 
238 static int	bpf_allocbufs(struct bpf_d *);
239 static void	bpf_deliver(struct bpf_if *,
240 		            void *(*cpfn)(void *, const void *, size_t),
241 		            void *, u_int, u_int, const bool);
242 static void	bpf_freed(struct bpf_d *);
243 static void	bpf_free_filter(struct bpf_filter *);
244 static void	bpf_ifname(struct ifnet *, struct ifreq *);
245 static void	*bpf_mcpy(void *, const void *, size_t);
246 static int	bpf_movein(struct uio *, int, uint64_t,
247 			        struct mbuf **, struct sockaddr *);
248 static void	bpf_attachd(struct bpf_d *, struct bpf_if *);
249 static void	bpf_detachd(struct bpf_d *);
250 static int	bpf_setif(struct bpf_d *, struct ifreq *);
251 static int	bpf_setf(struct bpf_d *, struct bpf_program *);
252 static void	bpf_timed_out(void *);
253 static inline void
254 		bpf_wakeup(struct bpf_d *);
255 static int	bpf_hdrlen(struct bpf_d *);
256 static void	catchpacket(struct bpf_d *, u_char *, u_int, u_int,
257     void *(*)(void *, const void *, size_t), 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 
272 static const struct fileops bpf_fileops = {
273 	.fo_name = "bpf",
274 	.fo_read = bpf_read,
275 	.fo_write = bpf_write,
276 	.fo_ioctl = bpf_ioctl,
277 	.fo_fcntl = fnullop_fcntl,
278 	.fo_poll = bpf_poll,
279 	.fo_stat = bpf_stat,
280 	.fo_close = bpf_close,
281 	.fo_kqfilter = bpf_kqfilter,
282 	.fo_restart = fnullop_restart,
283 };
284 
285 dev_type_open(bpfopen);
286 
287 const struct cdevsw bpf_cdevsw = {
288 	.d_open = bpfopen,
289 	.d_close = noclose,
290 	.d_read = noread,
291 	.d_write = nowrite,
292 	.d_ioctl = noioctl,
293 	.d_stop = nostop,
294 	.d_tty = notty,
295 	.d_poll = nopoll,
296 	.d_mmap = nommap,
297 	.d_kqfilter = nokqfilter,
298 	.d_discard = nodiscard,
299 	.d_flag = D_OTHER | D_MPSAFE
300 };
301 
302 bpfjit_func_t
303 bpf_jit_generate(bpf_ctx_t *bc, void *code, size_t size)
304 {
305 
306 	membar_consumer();
307 	if (bpfjit_module_ops.bj_generate_code != NULL) {
308 		return bpfjit_module_ops.bj_generate_code(bc, code, size);
309 	}
310 	return NULL;
311 }
312 
313 void
314 bpf_jit_freecode(bpfjit_func_t jcode)
315 {
316 	KASSERT(bpfjit_module_ops.bj_free_code != NULL);
317 	bpfjit_module_ops.bj_free_code(jcode);
318 }
319 
320 static int
321 bpf_movein(struct uio *uio, int linktype, uint64_t mtu, struct mbuf **mp,
322 	   struct sockaddr *sockp)
323 {
324 	struct mbuf *m;
325 	int error;
326 	size_t len;
327 	size_t hlen;
328 	size_t align;
329 
330 	/*
331 	 * Build a sockaddr based on the data link layer type.
332 	 * We do this at this level because the ethernet header
333 	 * is copied directly into the data field of the sockaddr.
334 	 * In the case of SLIP, there is no header and the packet
335 	 * is forwarded as is.
336 	 * Also, we are careful to leave room at the front of the mbuf
337 	 * for the link level header.
338 	 */
339 	switch (linktype) {
340 
341 	case DLT_SLIP:
342 		sockp->sa_family = AF_INET;
343 		hlen = 0;
344 		align = 0;
345 		break;
346 
347 	case DLT_PPP:
348 		sockp->sa_family = AF_UNSPEC;
349 		hlen = 0;
350 		align = 0;
351 		break;
352 
353 	case DLT_EN10MB:
354 		sockp->sa_family = AF_UNSPEC;
355 		/* XXX Would MAXLINKHDR be better? */
356  		/* 6(dst)+6(src)+2(type) */
357 		hlen = sizeof(struct ether_header);
358 		align = 2;
359 		break;
360 
361 	case DLT_ARCNET:
362 		sockp->sa_family = AF_UNSPEC;
363 		hlen = ARC_HDRLEN;
364 		align = 5;
365 		break;
366 
367 	case DLT_FDDI:
368 		sockp->sa_family = AF_LINK;
369 		/* XXX 4(FORMAC)+6(dst)+6(src) */
370 		hlen = 16;
371 		align = 0;
372 		break;
373 
374 	case DLT_ECONET:
375 		sockp->sa_family = AF_UNSPEC;
376 		hlen = 6;
377 		align = 2;
378 		break;
379 
380 	case DLT_NULL:
381 		sockp->sa_family = AF_UNSPEC;
382 		hlen = 0;
383 		align = 0;
384 		break;
385 
386 	default:
387 		return (EIO);
388 	}
389 
390 	len = uio->uio_resid;
391 	/*
392 	 * If there aren't enough bytes for a link level header or the
393 	 * packet length exceeds the interface mtu, return an error.
394 	 */
395 	if (len - hlen > mtu)
396 		return (EMSGSIZE);
397 
398 	/*
399 	 * XXX Avoid complicated buffer chaining ---
400 	 * bail if it won't fit in a single mbuf.
401 	 * (Take into account possible alignment bytes)
402 	 */
403 	if (len + align > MCLBYTES)
404 		return (EIO);
405 
406 	m = m_gethdr(M_WAIT, MT_DATA);
407 	m_reset_rcvif(m);
408 	m->m_pkthdr.len = (int)(len - hlen);
409 	if (len + align > MHLEN) {
410 		m_clget(m, M_WAIT);
411 		if ((m->m_flags & M_EXT) == 0) {
412 			error = ENOBUFS;
413 			goto bad;
414 		}
415 	}
416 
417 	/* Insure the data is properly aligned */
418 	if (align > 0) {
419 		m->m_data += align;
420 		m->m_len -= (int)align;
421 	}
422 
423 	error = uiomove(mtod(m, void *), len, uio);
424 	if (error)
425 		goto bad;
426 	if (hlen != 0) {
427 		memcpy(sockp->sa_data, mtod(m, void *), hlen);
428 		m->m_data += hlen; /* XXX */
429 		len -= hlen;
430 	}
431 	m->m_len = (int)len;
432 	*mp = m;
433 	return (0);
434 
435 bad:
436 	m_freem(m);
437 	return (error);
438 }
439 
440 /*
441  * Attach file to the bpf interface, i.e. make d listen on bp.
442  */
443 static void
444 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
445 {
446 
447 	KASSERT(mutex_owned(&bpf_mtx));
448 	KASSERT(mutex_owned(d->bd_mtx));
449 	/*
450 	 * Point d at bp, and add d to the interface's list of listeners.
451 	 * Finally, point the driver's bpf cookie at the interface so
452 	 * it will divert packets to bpf.
453 	 */
454 	d->bd_bif = bp;
455 	BPFIF_DLIST_WRITER_INSERT_HEAD(bp, d);
456 
457 	*bp->bif_driverp = bp;
458 }
459 
460 /*
461  * Detach a file from its interface.
462  */
463 static void
464 bpf_detachd(struct bpf_d *d)
465 {
466 	struct bpf_if *bp;
467 
468 	KASSERT(mutex_owned(&bpf_mtx));
469 	KASSERT(mutex_owned(d->bd_mtx));
470 
471 	bp = d->bd_bif;
472 	/*
473 	 * Check if this descriptor had requested promiscuous mode.
474 	 * If so, turn it off.
475 	 */
476 	if (d->bd_promisc) {
477 		int error __diagused;
478 
479 		d->bd_promisc = 0;
480 		/*
481 		 * Take device out of promiscuous mode.  Since we were
482 		 * able to enter promiscuous mode, we should be able
483 		 * to turn it off.  But we can get an error if
484 		 * the interface was configured down, so only panic
485 		 * if we don't get an unexpected error.
486 		 */
487 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
488   		error = ifpromisc(bp->bif_ifp, 0);
489 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
490 #ifdef DIAGNOSTIC
491 		if (error)
492 			printf("%s: ifpromisc failed: %d", __func__, error);
493 #endif
494 	}
495 
496 	/* Remove d from the interface's descriptor list. */
497 	BPFIF_DLIST_WRITER_REMOVE(d);
498 
499 	pserialize_perform(bpf_psz);
500 
501 	if (BPFIF_DLIST_WRITER_EMPTY(bp)) {
502 		/*
503 		 * Let the driver know that there are no more listeners.
504 		 */
505 		*d->bd_bif->bif_driverp = NULL;
506 	}
507 	d->bd_bif = NULL;
508 }
509 
510 static void
511 bpf_init(void)
512 {
513 
514 	mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE);
515 	bpf_psz = pserialize_create();
516 	bpf_psref_class = psref_class_create("bpf", IPL_SOFTNET);
517 
518 	PSLIST_INIT(&bpf_iflist);
519 	PSLIST_INIT(&bpf_dlist);
520 
521 	bpf_gstats_percpu = percpu_alloc(sizeof(struct bpf_stat));
522 
523 	return;
524 }
525 
526 /*
527  * bpfilterattach() is called at boot time.  We don't need to do anything
528  * here, since any initialization will happen as part of module init code.
529  */
530 /* ARGSUSED */
531 void
532 bpfilterattach(int n)
533 {
534 
535 }
536 
537 /*
538  * Open ethernet device. Clones.
539  */
540 /* ARGSUSED */
541 int
542 bpfopen(dev_t dev, int flag, int mode, struct lwp *l)
543 {
544 	struct bpf_d *d;
545 	struct file *fp;
546 	int error, fd;
547 
548 	/* falloc() will fill in the descriptor for us. */
549 	if ((error = fd_allocfile(&fp, &fd)) != 0)
550 		return error;
551 
552 	d = kmem_zalloc(sizeof(*d), KM_SLEEP);
553 	d->bd_bufsize = bpf_bufsize;
554 	d->bd_seesent = 1;
555 	d->bd_feedback = 0;
556 	d->bd_pid = l->l_proc->p_pid;
557 #ifdef _LP64
558 	if (curproc->p_flag & PK_32)
559 		d->bd_compat32 = 1;
560 #endif
561 	getnanotime(&d->bd_btime);
562 	d->bd_atime = d->bd_mtime = d->bd_btime;
563 	callout_init(&d->bd_callout, CALLOUT_MPSAFE);
564 	selinit(&d->bd_sel);
565 	d->bd_jitcode = NULL;
566 	d->bd_filter = NULL;
567 	BPF_DLIST_ENTRY_INIT(d);
568 	BPFIF_DLIST_ENTRY_INIT(d);
569 	d->bd_mtx = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
570 	d->bd_buf_mtx = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
571 	cv_init(&d->bd_cv, "bpf");
572 
573 	mutex_enter(&bpf_mtx);
574 	BPF_DLIST_WRITER_INSERT_HEAD(d);
575 	mutex_exit(&bpf_mtx);
576 
577 	return fd_clone(fp, fd, flag, &bpf_fileops, d);
578 }
579 
580 /*
581  * Close the descriptor by detaching it from its interface,
582  * deallocating its buffers, and marking it free.
583  */
584 /* ARGSUSED */
585 static int
586 bpf_close(struct file *fp)
587 {
588 	struct bpf_d *d;
589 
590 	mutex_enter(&bpf_mtx);
591 
592 	if ((d = fp->f_bpf) == NULL) {
593 		mutex_exit(&bpf_mtx);
594 		return 0;
595 	}
596 
597 	/*
598 	 * Refresh the PID associated with this bpf file.
599 	 */
600 	d->bd_pid = curproc->p_pid;
601 
602 	mutex_enter(d->bd_mtx);
603 	if (d->bd_state == BPF_WAITING)
604 		callout_halt(&d->bd_callout, d->bd_mtx);
605 	d->bd_state = BPF_IDLE;
606 	if (d->bd_bif)
607 		bpf_detachd(d);
608 	mutex_exit(d->bd_mtx);
609 
610 	BPF_DLIST_WRITER_REMOVE(d);
611 
612 	pserialize_perform(bpf_psz);
613 	mutex_exit(&bpf_mtx);
614 
615 	BPFIF_DLIST_ENTRY_DESTROY(d);
616 	BPF_DLIST_ENTRY_DESTROY(d);
617 	fp->f_bpf = NULL;
618 	bpf_freed(d);
619 	callout_destroy(&d->bd_callout);
620 	seldestroy(&d->bd_sel);
621 	mutex_obj_free(d->bd_mtx);
622 	mutex_obj_free(d->bd_buf_mtx);
623 	cv_destroy(&d->bd_cv);
624 
625 	kmem_free(d, sizeof(*d));
626 
627 	return (0);
628 }
629 
630 /*
631  * Rotate the packet buffers in descriptor d.  Move the store buffer
632  * into the hold slot, and the free buffer into the store slot.
633  * Zero the length of the new store buffer.
634  */
635 #define ROTATE_BUFFERS(d) \
636 	(d)->bd_hbuf = (d)->bd_sbuf; \
637 	(d)->bd_hlen = (d)->bd_slen; \
638 	(d)->bd_sbuf = (d)->bd_fbuf; \
639 	(d)->bd_slen = 0; \
640 	(d)->bd_fbuf = NULL;
641 /*
642  *  bpfread - read next chunk of packets from buffers
643  */
644 static int
645 bpf_read(struct file *fp, off_t *offp, struct uio *uio,
646     kauth_cred_t cred, int flags)
647 {
648 	struct bpf_d *d = fp->f_bpf;
649 	int timed_out;
650 	int error;
651 
652 	getnanotime(&d->bd_atime);
653 	/*
654 	 * Restrict application to use a buffer the same size as
655 	 * the kernel buffers.
656 	 */
657 	if (uio->uio_resid != d->bd_bufsize)
658 		return (EINVAL);
659 
660 	mutex_enter(d->bd_mtx);
661 	if (d->bd_state == BPF_WAITING)
662 		callout_halt(&d->bd_callout, d->bd_mtx);
663 	timed_out = (d->bd_state == BPF_TIMED_OUT);
664 	d->bd_state = BPF_IDLE;
665 	mutex_exit(d->bd_mtx);
666 	/*
667 	 * If the hold buffer is empty, then do a timed sleep, which
668 	 * ends when the timeout expires or when enough packets
669 	 * have arrived to fill the store buffer.
670 	 */
671 	mutex_enter(d->bd_buf_mtx);
672 	while (d->bd_hbuf == NULL) {
673 		if (fp->f_flag & FNONBLOCK) {
674 			if (d->bd_slen == 0) {
675 				error = EWOULDBLOCK;
676 				goto out;
677 			}
678 			ROTATE_BUFFERS(d);
679 			break;
680 		}
681 
682 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
683 			/*
684 			 * A packet(s) either arrived since the previous
685 			 * read or arrived while we were asleep.
686 			 * Rotate the buffers and return what's here.
687 			 */
688 			ROTATE_BUFFERS(d);
689 			break;
690 		}
691 
692 		error = cv_timedwait_sig(&d->bd_cv, d->bd_buf_mtx, d->bd_rtout);
693 
694 		if (error == EINTR || error == ERESTART)
695 			goto out;
696 
697 		if (error == EWOULDBLOCK) {
698 			/*
699 			 * On a timeout, return what's in the buffer,
700 			 * which may be nothing.  If there is something
701 			 * in the store buffer, we can rotate the buffers.
702 			 */
703 			if (d->bd_hbuf)
704 				/*
705 				 * We filled up the buffer in between
706 				 * getting the timeout and arriving
707 				 * here, so we don't need to rotate.
708 				 */
709 				break;
710 
711 			if (d->bd_slen == 0) {
712 				error = 0;
713 				goto out;
714 			}
715 			ROTATE_BUFFERS(d);
716 			break;
717 		}
718 		if (error != 0)
719 			goto out;
720 	}
721 	/*
722 	 * At this point, we know we have something in the hold slot.
723 	 */
724 	mutex_exit(d->bd_buf_mtx);
725 
726 	/*
727 	 * Move data from hold buffer into user space.
728 	 * We know the entire buffer is transferred since
729 	 * we checked above that the read buffer is bpf_bufsize bytes.
730 	 */
731 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
732 
733 	mutex_enter(d->bd_buf_mtx);
734 	d->bd_fbuf = d->bd_hbuf;
735 	d->bd_hbuf = NULL;
736 	d->bd_hlen = 0;
737 out:
738 	mutex_exit(d->bd_buf_mtx);
739 	return (error);
740 }
741 
742 
743 /*
744  * If there are processes sleeping on this descriptor, wake them up.
745  */
746 static inline void
747 bpf_wakeup(struct bpf_d *d)
748 {
749 
750 	mutex_enter(d->bd_buf_mtx);
751 	cv_broadcast(&d->bd_cv);
752 	mutex_exit(d->bd_buf_mtx);
753 
754 	if (d->bd_async)
755 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
756 	selnotify(&d->bd_sel, 0, 0);
757 }
758 
759 static void
760 bpf_timed_out(void *arg)
761 {
762 	struct bpf_d *d = arg;
763 
764 	mutex_enter(d->bd_mtx);
765 	if (d->bd_state == BPF_WAITING) {
766 		d->bd_state = BPF_TIMED_OUT;
767 		if (d->bd_slen != 0)
768 			bpf_wakeup(d);
769 	}
770 	mutex_exit(d->bd_mtx);
771 }
772 
773 
774 static int
775 bpf_write(struct file *fp, off_t *offp, struct uio *uio,
776     kauth_cred_t cred, int flags)
777 {
778 	struct bpf_d *d = fp->f_bpf;
779 	struct bpf_if *bp;
780 	struct ifnet *ifp;
781 	struct mbuf *m, *mc;
782 	int error;
783 	static struct sockaddr_storage dst;
784 	struct psref psref;
785 	int bound;
786 
787 	m = NULL;	/* XXX gcc */
788 
789 	bound = curlwp_bind();
790 	mutex_enter(d->bd_mtx);
791 	bp = d->bd_bif;
792 	if (bp == NULL) {
793 		mutex_exit(d->bd_mtx);
794 		error = ENXIO;
795 		goto out_bindx;
796 	}
797 	bpf_if_acquire(bp, &psref);
798 	mutex_exit(d->bd_mtx);
799 
800 	getnanotime(&d->bd_mtime);
801 
802 	ifp = bp->bif_ifp;
803 	if (if_is_deactivated(ifp)) {
804 		error = ENXIO;
805 		goto out;
806 	}
807 
808 	if (uio->uio_resid == 0) {
809 		error = 0;
810 		goto out;
811 	}
812 
813 	error = bpf_movein(uio, (int)bp->bif_dlt, ifp->if_mtu, &m,
814 		(struct sockaddr *) &dst);
815 	if (error)
816 		goto out;
817 
818 	if (m->m_pkthdr.len > ifp->if_mtu) {
819 		m_freem(m);
820 		error = EMSGSIZE;
821 		goto out;
822 	}
823 
824 	if (d->bd_hdrcmplt)
825 		dst.ss_family = pseudo_AF_HDRCMPLT;
826 
827 	if (d->bd_feedback) {
828 		mc = m_dup(m, 0, M_COPYALL, M_NOWAIT);
829 		if (mc != NULL)
830 			m_set_rcvif(mc, ifp);
831 		/* Set M_PROMISC for outgoing packets to be discarded. */
832 		if (1 /*d->bd_direction == BPF_D_INOUT*/)
833 			m->m_flags |= M_PROMISC;
834 	} else
835 		mc = NULL;
836 
837 	error = if_output_lock(ifp, ifp, m, (struct sockaddr *) &dst, NULL);
838 
839 	if (mc != NULL) {
840 		if (error == 0)
841 			ifp->_if_input(ifp, mc);
842 		else
843 			m_freem(mc);
844 	}
845 	/*
846 	 * The driver frees the mbuf.
847 	 */
848 out:
849 	bpf_if_release(bp, &psref);
850 out_bindx:
851 	curlwp_bindx(bound);
852 	return error;
853 }
854 
855 /*
856  * Reset a descriptor by flushing its packet buffer and clearing the
857  * receive and drop counts.
858  */
859 static void
860 reset_d(struct bpf_d *d)
861 {
862 
863 	KASSERT(mutex_owned(d->bd_mtx));
864 
865 	mutex_enter(d->bd_buf_mtx);
866 	if (d->bd_hbuf) {
867 		/* Free the hold buffer. */
868 		d->bd_fbuf = d->bd_hbuf;
869 		d->bd_hbuf = NULL;
870 	}
871 	d->bd_slen = 0;
872 	d->bd_hlen = 0;
873 	d->bd_rcount = 0;
874 	d->bd_dcount = 0;
875 	d->bd_ccount = 0;
876 	mutex_exit(d->bd_buf_mtx);
877 }
878 
879 /*
880  *  FIONREAD		Check for read packet available.
881  *  BIOCGBLEN		Get buffer len [for read()].
882  *  BIOCSETF		Set ethernet read filter.
883  *  BIOCFLUSH		Flush read packet buffer.
884  *  BIOCPROMISC		Put interface into promiscuous mode.
885  *  BIOCGDLT		Get link layer type.
886  *  BIOCGETIF		Get interface name.
887  *  BIOCSETIF		Set interface.
888  *  BIOCSRTIMEOUT	Set read timeout.
889  *  BIOCGRTIMEOUT	Get read timeout.
890  *  BIOCGSTATS		Get packet stats.
891  *  BIOCIMMEDIATE	Set immediate mode.
892  *  BIOCVERSION		Get filter language version.
893  *  BIOCGHDRCMPLT	Get "header already complete" flag.
894  *  BIOCSHDRCMPLT	Set "header already complete" flag.
895  *  BIOCSFEEDBACK	Set packet feedback mode.
896  *  BIOCGFEEDBACK	Get packet feedback mode.
897  *  BIOCGSEESENT  	Get "see sent packets" mode.
898  *  BIOCSSEESENT  	Set "see sent packets" mode.
899  */
900 /* ARGSUSED */
901 static int
902 bpf_ioctl(struct file *fp, u_long cmd, void *addr)
903 {
904 	struct bpf_d *d = fp->f_bpf;
905 	int error = 0;
906 
907 	/*
908 	 * Refresh the PID associated with this bpf file.
909 	 */
910 	d->bd_pid = curproc->p_pid;
911 #ifdef _LP64
912 	if (curproc->p_flag & PK_32)
913 		d->bd_compat32 = 1;
914 	else
915 		d->bd_compat32 = 0;
916 #endif
917 
918 	mutex_enter(d->bd_mtx);
919 	if (d->bd_state == BPF_WAITING)
920 		callout_halt(&d->bd_callout, d->bd_mtx);
921 	d->bd_state = BPF_IDLE;
922 	mutex_exit(d->bd_mtx);
923 
924 	switch (cmd) {
925 
926 	default:
927 		error = EINVAL;
928 		break;
929 
930 	/*
931 	 * Check for read packet available.
932 	 */
933 	case FIONREAD:
934 		{
935 			int n;
936 
937 			mutex_enter(d->bd_buf_mtx);
938 			n = d->bd_slen;
939 			if (d->bd_hbuf)
940 				n += d->bd_hlen;
941 			mutex_exit(d->bd_buf_mtx);
942 
943 			*(int *)addr = n;
944 			break;
945 		}
946 
947 	/*
948 	 * Get buffer len [for read()].
949 	 */
950 	case BIOCGBLEN:
951 		*(u_int *)addr = d->bd_bufsize;
952 		break;
953 
954 	/*
955 	 * Set buffer length.
956 	 */
957 	case BIOCSBLEN:
958 		/*
959 		 * Forbid to change the buffer length if buffers are already
960 		 * allocated.
961 		 */
962 		mutex_enter(d->bd_mtx);
963 		mutex_enter(d->bd_buf_mtx);
964 		if (d->bd_bif != NULL || d->bd_sbuf != NULL)
965 			error = EINVAL;
966 		else {
967 			u_int size = *(u_int *)addr;
968 
969 			if (size > bpf_maxbufsize)
970 				*(u_int *)addr = size = bpf_maxbufsize;
971 			else if (size < BPF_MINBUFSIZE)
972 				*(u_int *)addr = size = BPF_MINBUFSIZE;
973 			d->bd_bufsize = size;
974 		}
975 		mutex_exit(d->bd_buf_mtx);
976 		mutex_exit(d->bd_mtx);
977 		break;
978 
979 	/*
980 	 * Set link layer read filter.
981 	 */
982 	case BIOCSETF:
983 		error = bpf_setf(d, addr);
984 		break;
985 
986 	/*
987 	 * Flush read packet buffer.
988 	 */
989 	case BIOCFLUSH:
990 		mutex_enter(d->bd_mtx);
991 		reset_d(d);
992 		mutex_exit(d->bd_mtx);
993 		break;
994 
995 	/*
996 	 * Put interface into promiscuous mode.
997 	 */
998 	case BIOCPROMISC:
999 		mutex_enter(d->bd_mtx);
1000 		if (d->bd_bif == NULL) {
1001 			mutex_exit(d->bd_mtx);
1002 			/*
1003 			 * No interface attached yet.
1004 			 */
1005 			error = EINVAL;
1006 			break;
1007 		}
1008 		if (d->bd_promisc == 0) {
1009 			KERNEL_LOCK_UNLESS_NET_MPSAFE();
1010 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
1011 			KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
1012 			if (error == 0)
1013 				d->bd_promisc = 1;
1014 		}
1015 		mutex_exit(d->bd_mtx);
1016 		break;
1017 
1018 	/*
1019 	 * Get device parameters.
1020 	 */
1021 	case BIOCGDLT:
1022 		mutex_enter(d->bd_mtx);
1023 		if (d->bd_bif == NULL)
1024 			error = EINVAL;
1025 		else
1026 			*(u_int *)addr = d->bd_bif->bif_dlt;
1027 		mutex_exit(d->bd_mtx);
1028 		break;
1029 
1030 	/*
1031 	 * Get a list of supported device parameters.
1032 	 */
1033 	case BIOCGDLTLIST:
1034 		mutex_enter(d->bd_mtx);
1035 		if (d->bd_bif == NULL)
1036 			error = EINVAL;
1037 		else
1038 			error = bpf_getdltlist(d, addr);
1039 		mutex_exit(d->bd_mtx);
1040 		break;
1041 
1042 	/*
1043 	 * Set device parameters.
1044 	 */
1045 	case BIOCSDLT:
1046 		mutex_enter(&bpf_mtx);
1047 		mutex_enter(d->bd_mtx);
1048 		if (d->bd_bif == NULL)
1049 			error = EINVAL;
1050 		else
1051 			error = bpf_setdlt(d, *(u_int *)addr);
1052 		mutex_exit(d->bd_mtx);
1053 		mutex_exit(&bpf_mtx);
1054 		break;
1055 
1056 	/*
1057 	 * Set interface name.
1058 	 */
1059 #ifdef OBIOCGETIF
1060 	case OBIOCGETIF:
1061 #endif
1062 	case BIOCGETIF:
1063 		mutex_enter(d->bd_mtx);
1064 		if (d->bd_bif == NULL)
1065 			error = EINVAL;
1066 		else
1067 			bpf_ifname(d->bd_bif->bif_ifp, addr);
1068 		mutex_exit(d->bd_mtx);
1069 		break;
1070 
1071 	/*
1072 	 * Set interface.
1073 	 */
1074 #ifdef OBIOCSETIF
1075 	case OBIOCSETIF:
1076 #endif
1077 	case BIOCSETIF:
1078 		mutex_enter(&bpf_mtx);
1079 		error = bpf_setif(d, addr);
1080 		mutex_exit(&bpf_mtx);
1081 		break;
1082 
1083 	/*
1084 	 * Set read timeout.
1085 	 */
1086 	case BIOCSRTIMEOUT:
1087 		{
1088 			struct timeval *tv = addr;
1089 
1090 			/* Compute number of ticks. */
1091 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
1092 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
1093 				d->bd_rtout = 1;
1094 			break;
1095 		}
1096 
1097 #ifdef BIOCGORTIMEOUT
1098 	/*
1099 	 * Get read timeout.
1100 	 */
1101 	case BIOCGORTIMEOUT:
1102 		{
1103 			struct timeval50 *tv = addr;
1104 
1105 			tv->tv_sec = d->bd_rtout / hz;
1106 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1107 			break;
1108 		}
1109 #endif
1110 
1111 #ifdef BIOCSORTIMEOUT
1112 	/*
1113 	 * Set read timeout.
1114 	 */
1115 	case BIOCSORTIMEOUT:
1116 		{
1117 			struct timeval50 *tv = addr;
1118 
1119 			/* Compute number of ticks. */
1120 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
1121 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
1122 				d->bd_rtout = 1;
1123 			break;
1124 		}
1125 #endif
1126 
1127 	/*
1128 	 * Get read timeout.
1129 	 */
1130 	case BIOCGRTIMEOUT:
1131 		{
1132 			struct timeval *tv = addr;
1133 
1134 			tv->tv_sec = d->bd_rtout / hz;
1135 			tv->tv_usec = (d->bd_rtout % hz) * tick;
1136 			break;
1137 		}
1138 	/*
1139 	 * Get packet stats.
1140 	 */
1141 	case BIOCGSTATS:
1142 		{
1143 			struct bpf_stat *bs = addr;
1144 
1145 			bs->bs_recv = d->bd_rcount;
1146 			bs->bs_drop = d->bd_dcount;
1147 			bs->bs_capt = d->bd_ccount;
1148 			break;
1149 		}
1150 
1151 	case BIOCGSTATSOLD:
1152 		{
1153 			struct bpf_stat_old *bs = addr;
1154 
1155 			bs->bs_recv = d->bd_rcount;
1156 			bs->bs_drop = d->bd_dcount;
1157 			break;
1158 		}
1159 
1160 	/*
1161 	 * Set immediate mode.
1162 	 */
1163 	case BIOCIMMEDIATE:
1164 		d->bd_immediate = *(u_int *)addr;
1165 		break;
1166 
1167 	case BIOCVERSION:
1168 		{
1169 			struct bpf_version *bv = addr;
1170 
1171 			bv->bv_major = BPF_MAJOR_VERSION;
1172 			bv->bv_minor = BPF_MINOR_VERSION;
1173 			break;
1174 		}
1175 
1176 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
1177 		*(u_int *)addr = d->bd_hdrcmplt;
1178 		break;
1179 
1180 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
1181 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1182 		break;
1183 
1184 	/*
1185 	 * Get "see sent packets" flag
1186 	 */
1187 	case BIOCGSEESENT:
1188 		*(u_int *)addr = d->bd_seesent;
1189 		break;
1190 
1191 	/*
1192 	 * Set "see sent" packets flag
1193 	 */
1194 	case BIOCSSEESENT:
1195 		d->bd_seesent = *(u_int *)addr;
1196 		break;
1197 
1198 	/*
1199 	 * Set "feed packets from bpf back to input" mode
1200 	 */
1201 	case BIOCSFEEDBACK:
1202 		d->bd_feedback = *(u_int *)addr;
1203 		break;
1204 
1205 	/*
1206 	 * Get "feed packets from bpf back to input" mode
1207 	 */
1208 	case BIOCGFEEDBACK:
1209 		*(u_int *)addr = d->bd_feedback;
1210 		break;
1211 
1212 	case FIONBIO:		/* Non-blocking I/O */
1213 		/*
1214 		 * No need to do anything special as we use IO_NDELAY in
1215 		 * bpfread() as an indication of whether or not to block
1216 		 * the read.
1217 		 */
1218 		break;
1219 
1220 	case FIOASYNC:		/* Send signal on receive packets */
1221 		mutex_enter(d->bd_mtx);
1222 		d->bd_async = *(int *)addr;
1223 		mutex_exit(d->bd_mtx);
1224 		break;
1225 
1226 	case TIOCSPGRP:		/* Process or group to send signals to */
1227 	case FIOSETOWN:
1228 		error = fsetown(&d->bd_pgid, cmd, addr);
1229 		break;
1230 
1231 	case TIOCGPGRP:
1232 	case FIOGETOWN:
1233 		error = fgetown(d->bd_pgid, cmd, addr);
1234 		break;
1235 	}
1236 	return (error);
1237 }
1238 
1239 /*
1240  * Set d's packet filter program to fp.  If this file already has a filter,
1241  * free it and replace it.  Returns EINVAL for bogus requests.
1242  */
1243 static int
1244 bpf_setf(struct bpf_d *d, struct bpf_program *fp)
1245 {
1246 	struct bpf_insn *fcode;
1247 	bpfjit_func_t jcode;
1248 	size_t flen, size = 0;
1249 	struct bpf_filter *oldf, *newf;
1250 
1251 	jcode = NULL;
1252 	flen = fp->bf_len;
1253 
1254 	if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) {
1255 		return EINVAL;
1256 	}
1257 
1258 	if (flen) {
1259 		/*
1260 		 * Allocate the buffer, copy the byte-code from
1261 		 * userspace and validate it.
1262 		 */
1263 		size = flen * sizeof(*fp->bf_insns);
1264 		fcode = kmem_alloc(size, KM_SLEEP);
1265 		if (copyin(fp->bf_insns, fcode, size) != 0 ||
1266 		    !bpf_validate(fcode, (int)flen)) {
1267 			kmem_free(fcode, size);
1268 			return EINVAL;
1269 		}
1270 		membar_consumer();
1271 		if (bpf_jit)
1272 			jcode = bpf_jit_generate(NULL, fcode, flen);
1273 	} else {
1274 		fcode = NULL;
1275 	}
1276 
1277 	newf = kmem_alloc(sizeof(*newf), KM_SLEEP);
1278 	newf->bf_insn = fcode;
1279 	newf->bf_size = size;
1280 	newf->bf_jitcode = jcode;
1281 	d->bd_jitcode = jcode; /* XXX just for kvm(3) users */
1282 
1283 	/* Need to hold bpf_mtx for pserialize_perform */
1284 	mutex_enter(&bpf_mtx);
1285 	mutex_enter(d->bd_mtx);
1286 	oldf = d->bd_filter;
1287 	d->bd_filter = newf;
1288 	membar_producer();
1289 	reset_d(d);
1290 	pserialize_perform(bpf_psz);
1291 	mutex_exit(d->bd_mtx);
1292 	mutex_exit(&bpf_mtx);
1293 
1294 	if (oldf != NULL)
1295 		bpf_free_filter(oldf);
1296 
1297 	return 0;
1298 }
1299 
1300 /*
1301  * Detach a file from its current interface (if attached at all) and attach
1302  * to the interface indicated by the name stored in ifr.
1303  * Return an errno or 0.
1304  */
1305 static int
1306 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1307 {
1308 	struct bpf_if *bp;
1309 	char *cp;
1310 	int unit_seen, i, error;
1311 
1312 	KASSERT(mutex_owned(&bpf_mtx));
1313 	/*
1314 	 * Make sure the provided name has a unit number, and default
1315 	 * it to '0' if not specified.
1316 	 * XXX This is ugly ... do this differently?
1317 	 */
1318 	unit_seen = 0;
1319 	cp = ifr->ifr_name;
1320 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
1321 	while (*cp++)
1322 		if (*cp >= '0' && *cp <= '9')
1323 			unit_seen = 1;
1324 	if (!unit_seen) {
1325 		/* Make sure to leave room for the '\0'. */
1326 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
1327 			if ((ifr->ifr_name[i] >= 'a' &&
1328 			     ifr->ifr_name[i] <= 'z') ||
1329 			    (ifr->ifr_name[i] >= 'A' &&
1330 			     ifr->ifr_name[i] <= 'Z'))
1331 				continue;
1332 			ifr->ifr_name[i] = '0';
1333 		}
1334 	}
1335 
1336 	/*
1337 	 * Look through attached interfaces for the named one.
1338 	 */
1339 	BPF_IFLIST_WRITER_FOREACH(bp) {
1340 		struct ifnet *ifp = bp->bif_ifp;
1341 
1342 		if (ifp == NULL ||
1343 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
1344 			continue;
1345 		/* skip additional entry */
1346 		if (bp->bif_driverp != &ifp->if_bpf)
1347 			continue;
1348 		/*
1349 		 * We found the requested interface.
1350 		 * Allocate the packet buffers if we need to.
1351 		 * If we're already attached to requested interface,
1352 		 * just flush the buffer.
1353 		 */
1354 		/*
1355 		 * bpf_allocbufs is called only here. bpf_mtx ensures that
1356 		 * no race condition happen on d->bd_sbuf.
1357 		 */
1358 		if (d->bd_sbuf == NULL) {
1359 			error = bpf_allocbufs(d);
1360 			if (error != 0)
1361 				return (error);
1362 		}
1363 		mutex_enter(d->bd_mtx);
1364 		if (bp != d->bd_bif) {
1365 			if (d->bd_bif) {
1366 				/*
1367 				 * Detach if attached to something else.
1368 				 */
1369 				bpf_detachd(d);
1370 				BPFIF_DLIST_ENTRY_INIT(d);
1371 			}
1372 
1373 			bpf_attachd(d, bp);
1374 		}
1375 		reset_d(d);
1376 		mutex_exit(d->bd_mtx);
1377 		return (0);
1378 	}
1379 	/* Not found. */
1380 	return (ENXIO);
1381 }
1382 
1383 /*
1384  * Copy the interface name to the ifreq.
1385  */
1386 static void
1387 bpf_ifname(struct ifnet *ifp, struct ifreq *ifr)
1388 {
1389 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
1390 }
1391 
1392 static int
1393 bpf_stat(struct file *fp, struct stat *st)
1394 {
1395 	struct bpf_d *d = fp->f_bpf;
1396 
1397 	(void)memset(st, 0, sizeof(*st));
1398 	mutex_enter(d->bd_mtx);
1399 	st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid);
1400 	st->st_atimespec = d->bd_atime;
1401 	st->st_mtimespec = d->bd_mtime;
1402 	st->st_ctimespec = st->st_birthtimespec = d->bd_btime;
1403 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
1404 	st->st_gid = kauth_cred_getegid(fp->f_cred);
1405 	st->st_mode = S_IFCHR;
1406 	mutex_exit(d->bd_mtx);
1407 	return 0;
1408 }
1409 
1410 /*
1411  * Support for poll() system call
1412  *
1413  * Return true iff the specific operation will not block indefinitely - with
1414  * the assumption that it is safe to positively acknowledge a request for the
1415  * ability to write to the BPF device.
1416  * Otherwise, return false but make a note that a selnotify() must be done.
1417  */
1418 static int
1419 bpf_poll(struct file *fp, int events)
1420 {
1421 	struct bpf_d *d = fp->f_bpf;
1422 	int revents;
1423 
1424 	/*
1425 	 * Refresh the PID associated with this bpf file.
1426 	 */
1427 	mutex_enter(&bpf_mtx);
1428 	d->bd_pid = curproc->p_pid;
1429 
1430 	revents = events & (POLLOUT | POLLWRNORM);
1431 	if (events & (POLLIN | POLLRDNORM)) {
1432 		/*
1433 		 * An imitation of the FIONREAD ioctl code.
1434 		 */
1435 		mutex_enter(d->bd_mtx);
1436 		if (d->bd_hlen != 0 ||
1437 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1438 		     d->bd_slen != 0)) {
1439 			revents |= events & (POLLIN | POLLRDNORM);
1440 		} else {
1441 			selrecord(curlwp, &d->bd_sel);
1442 			/* Start the read timeout if necessary */
1443 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1444 				callout_reset(&d->bd_callout, d->bd_rtout,
1445 					      bpf_timed_out, d);
1446 				d->bd_state = BPF_WAITING;
1447 			}
1448 		}
1449 		mutex_exit(d->bd_mtx);
1450 	}
1451 
1452 	mutex_exit(&bpf_mtx);
1453 	return (revents);
1454 }
1455 
1456 static void
1457 filt_bpfrdetach(struct knote *kn)
1458 {
1459 	struct bpf_d *d = kn->kn_hook;
1460 
1461 	mutex_enter(d->bd_buf_mtx);
1462 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
1463 	mutex_exit(d->bd_buf_mtx);
1464 }
1465 
1466 static int
1467 filt_bpfread(struct knote *kn, long hint)
1468 {
1469 	struct bpf_d *d = kn->kn_hook;
1470 	int rv;
1471 
1472 	mutex_enter(d->bd_buf_mtx);
1473 	kn->kn_data = d->bd_hlen;
1474 	if (d->bd_immediate)
1475 		kn->kn_data += d->bd_slen;
1476 	rv = (kn->kn_data > 0);
1477 	mutex_exit(d->bd_buf_mtx);
1478 	return rv;
1479 }
1480 
1481 static const struct filterops bpfread_filtops = {
1482 	.f_isfd = 1,
1483 	.f_attach = NULL,
1484 	.f_detach = filt_bpfrdetach,
1485 	.f_event = filt_bpfread,
1486 };
1487 
1488 static int
1489 bpf_kqfilter(struct file *fp, struct knote *kn)
1490 {
1491 	struct bpf_d *d = fp->f_bpf;
1492 	struct klist *klist;
1493 
1494 	mutex_enter(d->bd_buf_mtx);
1495 	switch (kn->kn_filter) {
1496 	case EVFILT_READ:
1497 		klist = &d->bd_sel.sel_klist;
1498 		kn->kn_fop = &bpfread_filtops;
1499 		break;
1500 
1501 	default:
1502 		mutex_exit(d->bd_buf_mtx);
1503 		return (EINVAL);
1504 	}
1505 
1506 	kn->kn_hook = d;
1507 
1508 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1509 	mutex_exit(d->bd_buf_mtx);
1510 
1511 	return (0);
1512 }
1513 
1514 /*
1515  * Copy data from an mbuf chain into a buffer.  This code is derived
1516  * from m_copydata in sys/uipc_mbuf.c.
1517  */
1518 static void *
1519 bpf_mcpy(void *dst_arg, const void *src_arg, size_t len)
1520 {
1521 	const struct mbuf *m;
1522 	u_int count;
1523 	u_char *dst;
1524 
1525 	m = src_arg;
1526 	dst = dst_arg;
1527 	while (len > 0) {
1528 		if (m == NULL)
1529 			panic("bpf_mcpy");
1530 		count = min(m->m_len, len);
1531 		memcpy(dst, mtod(m, const void *), count);
1532 		m = m->m_next;
1533 		dst += count;
1534 		len -= count;
1535 	}
1536 	return dst_arg;
1537 }
1538 
1539 /*
1540  * Dispatch a packet to all the listeners on interface bp.
1541  *
1542  * pkt     pointer to the packet, either a data buffer or an mbuf chain
1543  * buflen  buffer length, if pkt is a data buffer
1544  * cpfn    a function that can copy pkt into the listener's buffer
1545  * pktlen  length of the packet
1546  * rcv     true if packet came in
1547  */
1548 static inline void
1549 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
1550     void *pkt, u_int pktlen, u_int buflen, const bool rcv)
1551 {
1552 	uint32_t mem[BPF_MEMWORDS];
1553 	bpf_args_t args = {
1554 		.pkt = (const uint8_t *)pkt,
1555 		.wirelen = pktlen,
1556 		.buflen = buflen,
1557 		.mem = mem,
1558 		.arg = NULL
1559 	};
1560 	bool gottime = false;
1561 	struct timespec ts;
1562 	struct bpf_d *d;
1563 	int s;
1564 
1565 	KASSERT(!cpu_intr_p());
1566 
1567 	/*
1568 	 * Note that the IPL does not have to be raised at this point.
1569 	 * The only problem that could arise here is that if two different
1570 	 * interfaces shared any data.  This is not the case.
1571 	 */
1572 	s = pserialize_read_enter();
1573 	BPFIF_DLIST_READER_FOREACH(d, bp) {
1574 		u_int slen = 0;
1575 		struct bpf_filter *filter;
1576 
1577 		if (!d->bd_seesent && !rcv) {
1578 			continue;
1579 		}
1580 		atomic_inc_ulong(&d->bd_rcount);
1581 		BPF_STATINC(recv);
1582 
1583 		filter = d->bd_filter;
1584 		membar_datadep_consumer();
1585 		if (filter != NULL) {
1586 			if (filter->bf_jitcode != NULL)
1587 				slen = filter->bf_jitcode(NULL, &args);
1588 			else
1589 				slen = bpf_filter_ext(NULL, filter->bf_insn,
1590 				    &args);
1591 		}
1592 
1593 		if (!slen) {
1594 			continue;
1595 		}
1596 		if (!gottime) {
1597 			gottime = true;
1598 			nanotime(&ts);
1599 		}
1600 		/* Assume catchpacket doesn't sleep */
1601 		catchpacket(d, pkt, pktlen, slen, cpfn, &ts);
1602 	}
1603 	pserialize_read_exit(s);
1604 }
1605 
1606 /*
1607  * Incoming linkage from device drivers.  Process the packet pkt, of length
1608  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
1609  * by each process' filter, and if accepted, stashed into the corresponding
1610  * buffer.
1611  */
1612 static void
1613 _bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
1614 {
1615 
1616 	bpf_deliver(bp, memcpy, pkt, pktlen, pktlen, true);
1617 }
1618 
1619 /*
1620  * Incoming linkage from device drivers, when the head of the packet is in
1621  * a buffer, and the tail is in an mbuf chain.
1622  */
1623 static void
1624 _bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
1625 {
1626 	u_int pktlen;
1627 	struct mbuf mb;
1628 
1629 	/* Skip outgoing duplicate packets. */
1630 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) {
1631 		m->m_flags &= ~M_PROMISC;
1632 		return;
1633 	}
1634 
1635 	pktlen = m_length(m) + dlen;
1636 
1637 	/*
1638 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
1639 	 * Note that we cut corners here; we only setup what's
1640 	 * absolutely needed--this mbuf should never go anywhere else.
1641 	 */
1642 	(void)memset(&mb, 0, sizeof(mb));
1643 	mb.m_next = m;
1644 	mb.m_data = data;
1645 	mb.m_len = dlen;
1646 
1647 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif_index != 0);
1648 }
1649 
1650 /*
1651  * Incoming linkage from device drivers, when packet is in an mbuf chain.
1652  */
1653 static void
1654 _bpf_mtap(struct bpf_if *bp, struct mbuf *m)
1655 {
1656 	void *(*cpfn)(void *, const void *, size_t);
1657 	u_int pktlen, buflen;
1658 	void *marg;
1659 
1660 	/* Skip outgoing duplicate packets. */
1661 	if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) {
1662 		m->m_flags &= ~M_PROMISC;
1663 		return;
1664 	}
1665 
1666 	pktlen = m_length(m);
1667 
1668 	if (pktlen == m->m_len) {
1669 		cpfn = (void *)memcpy;
1670 		marg = mtod(m, void *);
1671 		buflen = pktlen;
1672 	} else {
1673 		cpfn = bpf_mcpy;
1674 		marg = m;
1675 		buflen = 0;
1676 	}
1677 
1678 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif_index != 0);
1679 }
1680 
1681 /*
1682  * We need to prepend the address family as
1683  * a four byte field.  Cons up a dummy header
1684  * to pacify bpf.  This is safe because bpf
1685  * will only read from the mbuf (i.e., it won't
1686  * try to free it or keep a pointer a to it).
1687  */
1688 static void
1689 _bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m)
1690 {
1691 	struct mbuf m0;
1692 
1693 	m0.m_flags = 0;
1694 	m0.m_next = m;
1695 	m0.m_len = 4;
1696 	m0.m_data = (char *)&af;
1697 
1698 	_bpf_mtap(bp, &m0);
1699 }
1700 
1701 /*
1702  * Put the SLIP pseudo-"link header" in place.
1703  * Note this M_PREPEND() should never fail,
1704  * swince we know we always have enough space
1705  * in the input buffer.
1706  */
1707 static void
1708 _bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m)
1709 {
1710 	u_char *hp;
1711 
1712 	M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT);
1713 	if (*m == NULL)
1714 		return;
1715 
1716 	hp = mtod(*m, u_char *);
1717 	hp[SLX_DIR] = SLIPDIR_IN;
1718 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1719 
1720 	_bpf_mtap(bp, *m);
1721 
1722 	m_adj(*m, SLIP_HDRLEN);
1723 }
1724 
1725 /*
1726  * Put the SLIP pseudo-"link header" in
1727  * place.  The compressed header is now
1728  * at the beginning of the mbuf.
1729  */
1730 static void
1731 _bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m)
1732 {
1733 	struct mbuf m0;
1734 	u_char *hp;
1735 
1736 	m0.m_flags = 0;
1737 	m0.m_next = m;
1738 	m0.m_data = m0.m_dat;
1739 	m0.m_len = SLIP_HDRLEN;
1740 
1741 	hp = mtod(&m0, u_char *);
1742 
1743 	hp[SLX_DIR] = SLIPDIR_OUT;
1744 	(void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN);
1745 
1746 	_bpf_mtap(bp, &m0);
1747 	m_freem(m);
1748 }
1749 
1750 static struct mbuf *
1751 bpf_mbuf_enqueue(struct bpf_if *bp, struct mbuf *m)
1752 {
1753 	struct mbuf *dup;
1754 
1755 	dup = m_dup(m, 0, M_COPYALL, M_NOWAIT);
1756 	if (dup == NULL)
1757 		return NULL;
1758 
1759 	if (bp->bif_mbuf_tail != NULL) {
1760 		bp->bif_mbuf_tail->m_nextpkt = dup;
1761 	} else {
1762 		bp->bif_mbuf_head = dup;
1763 	}
1764 	bp->bif_mbuf_tail = dup;
1765 #ifdef BPF_MTAP_SOFTINT_DEBUG
1766 	log(LOG_DEBUG, "%s: enqueued mbuf=%p to %s\n",
1767 	    __func__, dup, bp->bif_ifp->if_xname);
1768 #endif
1769 
1770 	return dup;
1771 }
1772 
1773 static struct mbuf *
1774 bpf_mbuf_dequeue(struct bpf_if *bp)
1775 {
1776 	struct mbuf *m;
1777 	int s;
1778 
1779 	/* XXX NOMPSAFE: assumed running on one CPU */
1780 	s = splnet();
1781 	m = bp->bif_mbuf_head;
1782 	if (m != NULL) {
1783 		bp->bif_mbuf_head = m->m_nextpkt;
1784 		m->m_nextpkt = NULL;
1785 
1786 		if (bp->bif_mbuf_head == NULL)
1787 			bp->bif_mbuf_tail = NULL;
1788 #ifdef BPF_MTAP_SOFTINT_DEBUG
1789 		log(LOG_DEBUG, "%s: dequeued mbuf=%p from %s\n",
1790 		    __func__, m, bp->bif_ifp->if_xname);
1791 #endif
1792 	}
1793 	splx(s);
1794 
1795 	return m;
1796 }
1797 
1798 static void
1799 bpf_mtap_si(void *arg)
1800 {
1801 	struct bpf_if *bp = arg;
1802 	struct mbuf *m;
1803 
1804 	while ((m = bpf_mbuf_dequeue(bp)) != NULL) {
1805 #ifdef BPF_MTAP_SOFTINT_DEBUG
1806 		log(LOG_DEBUG, "%s: tapping mbuf=%p on %s\n",
1807 		    __func__, m, bp->bif_ifp->if_xname);
1808 #endif
1809 		bpf_ops->bpf_mtap(bp, m);
1810 		m_freem(m);
1811 	}
1812 }
1813 
1814 static void
1815 _bpf_mtap_softint(struct ifnet *ifp, struct mbuf *m)
1816 {
1817 	struct bpf_if *bp = ifp->if_bpf;
1818 	struct mbuf *dup;
1819 
1820 	KASSERT(cpu_intr_p());
1821 
1822 	/* To avoid extra invocations of the softint */
1823 	if (BPFIF_DLIST_READER_EMPTY(bp))
1824 		return;
1825 	KASSERT(bp->bif_si != NULL);
1826 
1827 	dup = bpf_mbuf_enqueue(bp, m);
1828 	if (dup != NULL)
1829 		softint_schedule(bp->bif_si);
1830 }
1831 
1832 static int
1833 bpf_hdrlen(struct bpf_d *d)
1834 {
1835 	int hdrlen = d->bd_bif->bif_hdrlen;
1836 	/*
1837 	 * Compute the length of the bpf header.  This is not necessarily
1838 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1839 	 * that the network layer header begins on a longword boundary (for
1840 	 * performance reasons and to alleviate alignment restrictions).
1841 	 */
1842 #ifdef _LP64
1843 	if (d->bd_compat32)
1844 		return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen);
1845 	else
1846 #endif
1847 		return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen);
1848 }
1849 
1850 /*
1851  * Move the packet data from interface memory (pkt) into the
1852  * store buffer. Call the wakeup functions if it's time to wakeup
1853  * a listener (buffer full), "cpfn" is the routine called to do the
1854  * actual data transfer. memcpy is passed in to copy contiguous chunks,
1855  * while bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
1856  * pkt is really an mbuf.
1857  */
1858 static void
1859 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
1860     void *(*cpfn)(void *, const void *, size_t), struct timespec *ts)
1861 {
1862 	char *h;
1863 	int totlen, curlen, caplen;
1864 	int hdrlen = bpf_hdrlen(d);
1865 	int do_wakeup = 0;
1866 
1867 	atomic_inc_ulong(&d->bd_ccount);
1868 	BPF_STATINC(capt);
1869 	/*
1870 	 * Figure out how many bytes to move.  If the packet is
1871 	 * greater or equal to the snapshot length, transfer that
1872 	 * much.  Otherwise, transfer the whole packet (unless
1873 	 * we hit the buffer size limit).
1874 	 */
1875 	totlen = hdrlen + min(snaplen, pktlen);
1876 	if (totlen > d->bd_bufsize)
1877 		totlen = d->bd_bufsize;
1878 	/*
1879 	 * If we adjusted totlen to fit the bufsize, it could be that
1880 	 * totlen is smaller than hdrlen because of the link layer header.
1881 	 */
1882 	caplen = totlen - hdrlen;
1883 	if (caplen < 0)
1884 		caplen = 0;
1885 
1886 	mutex_enter(d->bd_buf_mtx);
1887 	/*
1888 	 * Round up the end of the previous packet to the next longword.
1889 	 */
1890 #ifdef _LP64
1891 	if (d->bd_compat32)
1892 		curlen = BPF_WORDALIGN32(d->bd_slen);
1893 	else
1894 #endif
1895 		curlen = BPF_WORDALIGN(d->bd_slen);
1896 	if (curlen + totlen > d->bd_bufsize) {
1897 		/*
1898 		 * This packet will overflow the storage buffer.
1899 		 * Rotate the buffers if we can, then wakeup any
1900 		 * pending reads.
1901 		 */
1902 		if (d->bd_fbuf == NULL) {
1903 			mutex_exit(d->bd_buf_mtx);
1904 			/*
1905 			 * We haven't completed the previous read yet,
1906 			 * so drop the packet.
1907 			 */
1908 			atomic_inc_ulong(&d->bd_dcount);
1909 			BPF_STATINC(drop);
1910 			return;
1911 		}
1912 		ROTATE_BUFFERS(d);
1913 		do_wakeup = 1;
1914 		curlen = 0;
1915 	} else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
1916 		/*
1917 		 * Immediate mode is set, or the read timeout has
1918 		 * already expired during a select call.  A packet
1919 		 * arrived, so the reader should be woken up.
1920 		 */
1921 		do_wakeup = 1;
1922 	}
1923 
1924 	/*
1925 	 * Append the bpf header.
1926 	 */
1927 	h = (char *)d->bd_sbuf + curlen;
1928 #ifdef _LP64
1929 	if (d->bd_compat32) {
1930 		struct bpf_hdr32 *hp32;
1931 
1932 		hp32 = (struct bpf_hdr32 *)h;
1933 		hp32->bh_tstamp.tv_sec = ts->tv_sec;
1934 		hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1935 		hp32->bh_datalen = pktlen;
1936 		hp32->bh_hdrlen = hdrlen;
1937 		hp32->bh_caplen = caplen;
1938 	} else
1939 #endif
1940 	{
1941 		struct bpf_hdr *hp;
1942 
1943 		hp = (struct bpf_hdr *)h;
1944 		hp->bh_tstamp.tv_sec = ts->tv_sec;
1945 		hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000;
1946 		hp->bh_datalen = pktlen;
1947 		hp->bh_hdrlen = hdrlen;
1948 		hp->bh_caplen = caplen;
1949 	}
1950 
1951 	/*
1952 	 * Copy the packet data into the store buffer and update its length.
1953 	 */
1954 	(*cpfn)(h + hdrlen, pkt, caplen);
1955 	d->bd_slen = curlen + totlen;
1956 	mutex_exit(d->bd_buf_mtx);
1957 
1958 	/*
1959 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
1960 	 * will cause filt_bpfread() to be called with it adjusted.
1961 	 */
1962 	if (do_wakeup)
1963 		bpf_wakeup(d);
1964 }
1965 
1966 /*
1967  * Initialize all nonzero fields of a descriptor.
1968  */
1969 static int
1970 bpf_allocbufs(struct bpf_d *d)
1971 {
1972 
1973 	d->bd_fbuf = kmem_alloc(d->bd_bufsize, KM_NOSLEEP);
1974 	if (!d->bd_fbuf)
1975 		return (ENOBUFS);
1976 	d->bd_sbuf = kmem_alloc(d->bd_bufsize, KM_NOSLEEP);
1977 	if (!d->bd_sbuf) {
1978 		kmem_free(d->bd_fbuf, d->bd_bufsize);
1979 		return (ENOBUFS);
1980 	}
1981 	d->bd_slen = 0;
1982 	d->bd_hlen = 0;
1983 	return (0);
1984 }
1985 
1986 static void
1987 bpf_free_filter(struct bpf_filter *filter)
1988 {
1989 
1990 	KASSERT(filter != NULL);
1991 	KASSERT(filter->bf_insn != NULL);
1992 
1993 	kmem_free(filter->bf_insn, filter->bf_size);
1994 	if (filter->bf_jitcode != NULL)
1995 		bpf_jit_freecode(filter->bf_jitcode);
1996 	kmem_free(filter, sizeof(*filter));
1997 }
1998 
1999 /*
2000  * Free buffers currently in use by a descriptor.
2001  * Called on close.
2002  */
2003 static void
2004 bpf_freed(struct bpf_d *d)
2005 {
2006 	/*
2007 	 * We don't need to lock out interrupts since this descriptor has
2008 	 * been detached from its interface and it yet hasn't been marked
2009 	 * free.
2010 	 */
2011 	if (d->bd_sbuf != NULL) {
2012 		kmem_free(d->bd_sbuf, d->bd_bufsize);
2013 		if (d->bd_hbuf != NULL)
2014 			kmem_free(d->bd_hbuf, d->bd_bufsize);
2015 		if (d->bd_fbuf != NULL)
2016 			kmem_free(d->bd_fbuf, d->bd_bufsize);
2017 	}
2018 	if (d->bd_filter != NULL) {
2019 		bpf_free_filter(d->bd_filter);
2020 		d->bd_filter = NULL;
2021 	}
2022 	d->bd_jitcode = NULL;
2023 }
2024 
2025 /*
2026  * Attach an interface to bpf.  dlt is the link layer type;
2027  * hdrlen is the fixed size of the link header for the specified dlt
2028  * (variable length headers not yet supported).
2029  */
2030 static void
2031 _bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2032 {
2033 	struct bpf_if *bp;
2034 	bp = kmem_alloc(sizeof(*bp), KM_NOSLEEP);
2035 	if (bp == NULL)
2036 		panic("bpfattach");
2037 
2038 	mutex_enter(&bpf_mtx);
2039 	bp->bif_driverp = driverp;
2040 	bp->bif_ifp = ifp;
2041 	bp->bif_dlt = dlt;
2042 	bp->bif_si = NULL;
2043 	BPF_IFLIST_ENTRY_INIT(bp);
2044 	PSLIST_INIT(&bp->bif_dlist_head);
2045 	psref_target_init(&bp->bif_psref, bpf_psref_class);
2046 
2047 	BPF_IFLIST_WRITER_INSERT_HEAD(bp);
2048 
2049 	*bp->bif_driverp = NULL;
2050 
2051 	bp->bif_hdrlen = hdrlen;
2052 	mutex_exit(&bpf_mtx);
2053 #if 0
2054 	printf("bpf: %s attached\n", ifp->if_xname);
2055 #endif
2056 }
2057 
2058 static void
2059 _bpf_mtap_softint_init(struct ifnet *ifp)
2060 {
2061 	struct bpf_if *bp;
2062 
2063 	mutex_enter(&bpf_mtx);
2064 	BPF_IFLIST_WRITER_FOREACH(bp) {
2065 		if (bp->bif_ifp != ifp)
2066 			continue;
2067 
2068 		bp->bif_mbuf_head = NULL;
2069 		bp->bif_mbuf_tail = NULL;
2070 		bp->bif_si = softint_establish(SOFTINT_NET, bpf_mtap_si, bp);
2071 		if (bp->bif_si == NULL)
2072 			panic("%s: softint_establish() failed", __func__);
2073 		break;
2074 	}
2075 	mutex_exit(&bpf_mtx);
2076 
2077 	if (bp == NULL)
2078 		panic("%s: no bpf_if found for %s", __func__, ifp->if_xname);
2079 }
2080 
2081 /*
2082  * Remove an interface from bpf.
2083  */
2084 static void
2085 _bpfdetach(struct ifnet *ifp)
2086 {
2087 	struct bpf_if *bp;
2088 	struct bpf_d *d;
2089 	int s;
2090 
2091 	mutex_enter(&bpf_mtx);
2092 	/* Nuke the vnodes for any open instances */
2093   again_d:
2094 	BPF_DLIST_WRITER_FOREACH(d) {
2095 		mutex_enter(d->bd_mtx);
2096 		if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) {
2097 			/*
2098 			 * Detach the descriptor from an interface now.
2099 			 * It will be free'ed later by close routine.
2100 			 */
2101 			d->bd_promisc = 0;	/* we can't touch device. */
2102 			bpf_detachd(d);
2103 			mutex_exit(d->bd_mtx);
2104 			goto again_d;
2105 		}
2106 		mutex_exit(d->bd_mtx);
2107 	}
2108 
2109   again:
2110 	BPF_IFLIST_WRITER_FOREACH(bp) {
2111 		if (bp->bif_ifp == ifp) {
2112 			BPF_IFLIST_WRITER_REMOVE(bp);
2113 
2114 			pserialize_perform(bpf_psz);
2115 			psref_target_destroy(&bp->bif_psref, bpf_psref_class);
2116 
2117 			BPF_IFLIST_ENTRY_DESTROY(bp);
2118 			if (bp->bif_si != NULL) {
2119 				/* XXX NOMPSAFE: assumed running on one CPU */
2120 				s = splnet();
2121 				while (bp->bif_mbuf_head != NULL) {
2122 					struct mbuf *m = bp->bif_mbuf_head;
2123 					bp->bif_mbuf_head = m->m_nextpkt;
2124 					m_freem(m);
2125 				}
2126 				splx(s);
2127 				softint_disestablish(bp->bif_si);
2128 			}
2129 			kmem_free(bp, sizeof(*bp));
2130 			goto again;
2131 		}
2132 	}
2133 	mutex_exit(&bpf_mtx);
2134 }
2135 
2136 /*
2137  * Change the data link type of a interface.
2138  */
2139 static void
2140 _bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2141 {
2142 	struct bpf_if *bp;
2143 
2144 	mutex_enter(&bpf_mtx);
2145 	BPF_IFLIST_WRITER_FOREACH(bp) {
2146 		if (bp->bif_driverp == &ifp->if_bpf)
2147 			break;
2148 	}
2149 	if (bp == NULL)
2150 		panic("bpf_change_type");
2151 
2152 	bp->bif_dlt = dlt;
2153 
2154 	bp->bif_hdrlen = hdrlen;
2155 	mutex_exit(&bpf_mtx);
2156 }
2157 
2158 /*
2159  * Get a list of available data link type of the interface.
2160  */
2161 static int
2162 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2163 {
2164 	int n, error;
2165 	struct ifnet *ifp;
2166 	struct bpf_if *bp;
2167 	int s, bound;
2168 
2169 	KASSERT(mutex_owned(d->bd_mtx));
2170 
2171 	ifp = d->bd_bif->bif_ifp;
2172 	n = 0;
2173 	error = 0;
2174 
2175 	bound = curlwp_bind();
2176 	s = pserialize_read_enter();
2177 	BPF_IFLIST_READER_FOREACH(bp) {
2178 		if (bp->bif_ifp != ifp)
2179 			continue;
2180 		if (bfl->bfl_list != NULL) {
2181 			struct psref psref;
2182 
2183 			if (n >= bfl->bfl_len) {
2184 				pserialize_read_exit(s);
2185 				return ENOMEM;
2186 			}
2187 
2188 			bpf_if_acquire(bp, &psref);
2189 			pserialize_read_exit(s);
2190 
2191 			error = copyout(&bp->bif_dlt,
2192 			    bfl->bfl_list + n, sizeof(u_int));
2193 
2194 			s = pserialize_read_enter();
2195 			bpf_if_release(bp, &psref);
2196 		}
2197 		n++;
2198 	}
2199 	pserialize_read_exit(s);
2200 	curlwp_bindx(bound);
2201 
2202 	bfl->bfl_len = n;
2203 	return error;
2204 }
2205 
2206 /*
2207  * Set the data link type of a BPF instance.
2208  */
2209 static int
2210 bpf_setdlt(struct bpf_d *d, u_int dlt)
2211 {
2212 	int error, opromisc;
2213 	struct ifnet *ifp;
2214 	struct bpf_if *bp;
2215 
2216 	KASSERT(mutex_owned(&bpf_mtx));
2217 	KASSERT(mutex_owned(d->bd_mtx));
2218 
2219 	if (d->bd_bif->bif_dlt == dlt)
2220 		return 0;
2221 	ifp = d->bd_bif->bif_ifp;
2222 	BPF_IFLIST_WRITER_FOREACH(bp) {
2223 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2224 			break;
2225 	}
2226 	if (bp == NULL)
2227 		return EINVAL;
2228 	opromisc = d->bd_promisc;
2229 	bpf_detachd(d);
2230 	BPFIF_DLIST_ENTRY_INIT(d);
2231 	bpf_attachd(d, bp);
2232 	reset_d(d);
2233 	if (opromisc) {
2234 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
2235 		error = ifpromisc(bp->bif_ifp, 1);
2236 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2237 		if (error)
2238 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
2239 			    bp->bif_ifp->if_xname, error);
2240 		else
2241 			d->bd_promisc = 1;
2242 	}
2243 	return 0;
2244 }
2245 
2246 static int
2247 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
2248 {
2249 	int newsize, error;
2250 	struct sysctlnode node;
2251 
2252 	node = *rnode;
2253 	node.sysctl_data = &newsize;
2254 	newsize = bpf_maxbufsize;
2255 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2256 	if (error || newp == NULL)
2257 		return (error);
2258 
2259 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
2260 		return (EINVAL);
2261 
2262 	bpf_maxbufsize = newsize;
2263 
2264 	return (0);
2265 }
2266 
2267 #if defined(MODULAR) || defined(BPFJIT)
2268 static int
2269 sysctl_net_bpf_jit(SYSCTLFN_ARGS)
2270 {
2271 	bool newval;
2272 	int error;
2273 	struct sysctlnode node;
2274 
2275 	node = *rnode;
2276 	node.sysctl_data = &newval;
2277 	newval = bpf_jit;
2278 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2279 	if (error != 0 || newp == NULL)
2280 		return error;
2281 
2282 	bpf_jit = newval;
2283 
2284 	/*
2285 	 * Do a full sync to publish new bpf_jit value and
2286 	 * update bpfjit_module_ops.bj_generate_code variable.
2287 	 */
2288 	membar_sync();
2289 
2290 	if (newval && bpfjit_module_ops.bj_generate_code == NULL) {
2291 		printf("JIT compilation is postponed "
2292 		    "until after bpfjit module is loaded\n");
2293 	}
2294 
2295 	return 0;
2296 }
2297 #endif
2298 
2299 static int
2300 sysctl_net_bpf_peers(SYSCTLFN_ARGS)
2301 {
2302 	int    error, elem_count;
2303 	struct bpf_d	 *dp;
2304 	struct bpf_d_ext  dpe;
2305 	size_t len, needed, elem_size, out_size;
2306 	char   *sp;
2307 
2308 	if (namelen == 1 && name[0] == CTL_QUERY)
2309 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
2310 
2311 	if (namelen != 2)
2312 		return (EINVAL);
2313 
2314 	/* BPF peers is privileged information. */
2315 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
2316 	    KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL);
2317 	if (error)
2318 		return (EPERM);
2319 
2320 	len = (oldp != NULL) ? *oldlenp : 0;
2321 	sp = oldp;
2322 	elem_size = name[0];
2323 	elem_count = name[1];
2324 	out_size = MIN(sizeof(dpe), elem_size);
2325 	needed = 0;
2326 
2327 	if (elem_size < 1 || elem_count < 0)
2328 		return (EINVAL);
2329 
2330 	mutex_enter(&bpf_mtx);
2331 	BPF_DLIST_WRITER_FOREACH(dp) {
2332 		if (len >= elem_size && elem_count > 0) {
2333 #define BPF_EXT(field)	dpe.bde_ ## field = dp->bd_ ## field
2334 			BPF_EXT(bufsize);
2335 			BPF_EXT(promisc);
2336 			BPF_EXT(state);
2337 			BPF_EXT(immediate);
2338 			BPF_EXT(hdrcmplt);
2339 			BPF_EXT(seesent);
2340 			BPF_EXT(pid);
2341 			BPF_EXT(rcount);
2342 			BPF_EXT(dcount);
2343 			BPF_EXT(ccount);
2344 #undef BPF_EXT
2345 			mutex_enter(dp->bd_mtx);
2346 			if (dp->bd_bif)
2347 				(void)strlcpy(dpe.bde_ifname,
2348 				    dp->bd_bif->bif_ifp->if_xname,
2349 				    IFNAMSIZ - 1);
2350 			else
2351 				dpe.bde_ifname[0] = '\0';
2352 			mutex_exit(dp->bd_mtx);
2353 
2354 			error = copyout(&dpe, sp, out_size);
2355 			if (error)
2356 				break;
2357 			sp += elem_size;
2358 			len -= elem_size;
2359 		}
2360 		needed += elem_size;
2361 		if (elem_count > 0 && elem_count != INT_MAX)
2362 			elem_count--;
2363 	}
2364 	mutex_exit(&bpf_mtx);
2365 
2366 	*oldlenp = needed;
2367 
2368 	return (error);
2369 }
2370 
2371 static void
2372 bpf_stats(void *p, void *arg, struct cpu_info *ci __unused)
2373 {
2374 	struct bpf_stat *const stats = p;
2375 	struct bpf_stat *sum = arg;
2376 
2377 	sum->bs_recv += stats->bs_recv;
2378 	sum->bs_drop += stats->bs_drop;
2379 	sum->bs_capt += stats->bs_capt;
2380 }
2381 
2382 static int
2383 bpf_sysctl_gstats_handler(SYSCTLFN_ARGS)
2384 {
2385 	struct sysctlnode node;
2386 	int error;
2387 	struct bpf_stat sum;
2388 
2389 	memset(&sum, 0, sizeof(sum));
2390 	node = *rnode;
2391 
2392 	percpu_foreach(bpf_gstats_percpu, bpf_stats, &sum);
2393 
2394 	node.sysctl_data = &sum;
2395 	node.sysctl_size = sizeof(sum);
2396 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2397 	if (error != 0 || newp == NULL)
2398 		return error;
2399 
2400 	return 0;
2401 }
2402 
2403 static struct sysctllog *bpf_sysctllog;
2404 static void
2405 sysctl_net_bpf_setup(void)
2406 {
2407 	const struct sysctlnode *node;
2408 
2409 	node = NULL;
2410 	sysctl_createv(&bpf_sysctllog, 0, NULL, &node,
2411 		       CTLFLAG_PERMANENT,
2412 		       CTLTYPE_NODE, "bpf",
2413 		       SYSCTL_DESCR("BPF options"),
2414 		       NULL, 0, NULL, 0,
2415 		       CTL_NET, CTL_CREATE, CTL_EOL);
2416 	if (node != NULL) {
2417 #if defined(MODULAR) || defined(BPFJIT)
2418 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2419 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2420 			CTLTYPE_BOOL, "jit",
2421 			SYSCTL_DESCR("Toggle Just-In-Time compilation"),
2422 			sysctl_net_bpf_jit, 0, &bpf_jit, 0,
2423 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2424 #endif
2425 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2426 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2427 			CTLTYPE_INT, "maxbufsize",
2428 			SYSCTL_DESCR("Maximum size for data capture buffer"),
2429 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
2430 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2431 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2432 			CTLFLAG_PERMANENT,
2433 			CTLTYPE_STRUCT, "stats",
2434 			SYSCTL_DESCR("BPF stats"),
2435 			bpf_sysctl_gstats_handler, 0, NULL, 0,
2436 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2437 		sysctl_createv(&bpf_sysctllog, 0, NULL, NULL,
2438 			CTLFLAG_PERMANENT,
2439 			CTLTYPE_STRUCT, "peers",
2440 			SYSCTL_DESCR("BPF peers"),
2441 			sysctl_net_bpf_peers, 0, NULL, 0,
2442 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2443 	}
2444 
2445 }
2446 
2447 struct bpf_ops bpf_ops_kernel = {
2448 	.bpf_attach =		_bpfattach,
2449 	.bpf_detach =		_bpfdetach,
2450 	.bpf_change_type =	_bpf_change_type,
2451 
2452 	.bpf_tap =		_bpf_tap,
2453 	.bpf_mtap =		_bpf_mtap,
2454 	.bpf_mtap2 =		_bpf_mtap2,
2455 	.bpf_mtap_af =		_bpf_mtap_af,
2456 	.bpf_mtap_sl_in =	_bpf_mtap_sl_in,
2457 	.bpf_mtap_sl_out =	_bpf_mtap_sl_out,
2458 
2459 	.bpf_mtap_softint =		_bpf_mtap_softint,
2460 	.bpf_mtap_softint_init =	_bpf_mtap_softint_init,
2461 };
2462 
2463 MODULE(MODULE_CLASS_DRIVER, bpf, "bpf_filter");
2464 
2465 static int
2466 bpf_modcmd(modcmd_t cmd, void *arg)
2467 {
2468 #ifdef _MODULE
2469 	devmajor_t bmajor, cmajor;
2470 #endif
2471 	int error = 0;
2472 
2473 	switch (cmd) {
2474 	case MODULE_CMD_INIT:
2475 		bpf_init();
2476 #ifdef _MODULE
2477 		bmajor = cmajor = NODEVMAJOR;
2478 		error = devsw_attach("bpf", NULL, &bmajor,
2479 		    &bpf_cdevsw, &cmajor);
2480 		if (error)
2481 			break;
2482 #endif
2483 
2484 		bpf_ops_handover_enter(&bpf_ops_kernel);
2485 		atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel);
2486 		bpf_ops_handover_exit();
2487 		sysctl_net_bpf_setup();
2488 		break;
2489 
2490 	case MODULE_CMD_FINI:
2491 		/*
2492 		 * While there is no reference counting for bpf callers,
2493 		 * unload could at least in theory be done similarly to
2494 		 * system call disestablishment.  This should even be
2495 		 * a little simpler:
2496 		 *
2497 		 * 1) replace op vector with stubs
2498 		 * 2) post update to all cpus with xc
2499 		 * 3) check that nobody is in bpf anymore
2500 		 *    (it's doubtful we'd want something like l_sysent,
2501 		 *     but we could do something like *signed* percpu
2502 		 *     counters.  if the sum is 0, we're good).
2503 		 * 4) if fail, unroll changes
2504 		 *
2505 		 * NOTE: change won't be atomic to the outside.  some
2506 		 * packets may be not captured even if unload is
2507 		 * not succesful.  I think packet capture not working
2508 		 * is a perfectly logical consequence of trying to
2509 		 * disable packet capture.
2510 		 */
2511 		error = EOPNOTSUPP;
2512 		/* insert sysctl teardown */
2513 		break;
2514 
2515 	default:
2516 		error = ENOTTY;
2517 		break;
2518 	}
2519 
2520 	return error;
2521 }
2522