FreeBSD/Linux Kernel Cross Reference
sys/net/bpf.c
1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
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 */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD: releng/12.0/sys/net/bpf.c 336676 2018-07-24 16:35:52Z andrew $");
41
42 #include "opt_bpf.h"
43 #include "opt_ddb.h"
44 #include "opt_netgraph.h"
45
46 #include <sys/types.h>
47 #include <sys/param.h>
48 #include <sys/lock.h>
49 #include <sys/rwlock.h>
50 #include <sys/systm.h>
51 #include <sys/conf.h>
52 #include <sys/fcntl.h>
53 #include <sys/jail.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/time.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/signalvar.h>
60 #include <sys/filio.h>
61 #include <sys/sockio.h>
62 #include <sys/ttycom.h>
63 #include <sys/uio.h>
64 #include <sys/sysent.h>
65
66 #include <sys/event.h>
67 #include <sys/file.h>
68 #include <sys/poll.h>
69 #include <sys/proc.h>
70
71 #include <sys/socket.h>
72
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #endif
76
77 #include <net/if.h>
78 #include <net/if_var.h>
79 #include <net/if_dl.h>
80 #include <net/bpf.h>
81 #include <net/bpf_buffer.h>
82 #ifdef BPF_JITTER
83 #include <net/bpf_jitter.h>
84 #endif
85 #include <net/bpf_zerocopy.h>
86 #include <net/bpfdesc.h>
87 #include <net/route.h>
88 #include <net/vnet.h>
89
90 #include <netinet/in.h>
91 #include <netinet/if_ether.h>
92 #include <sys/kernel.h>
93 #include <sys/sysctl.h>
94
95 #include <net80211/ieee80211_freebsd.h>
96
97 #include <security/mac/mac_framework.h>
98
99 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
100
101 static struct bpf_if_ext dead_bpf_if = {
102 .bif_dlist = LIST_HEAD_INITIALIZER()
103 };
104
105 struct bpf_if {
106 #define bif_next bif_ext.bif_next
107 #define bif_dlist bif_ext.bif_dlist
108 struct bpf_if_ext bif_ext; /* public members */
109 u_int bif_dlt; /* link layer type */
110 u_int bif_hdrlen; /* length of link header */
111 struct ifnet *bif_ifp; /* corresponding interface */
112 struct rwlock bif_lock; /* interface lock */
113 LIST_HEAD(, bpf_d) bif_wlist; /* writer-only list */
114 int bif_flags; /* Interface flags */
115 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
116 };
117
118 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
119
120 #define BPFIF_RLOCK(bif) rw_rlock(&(bif)->bif_lock)
121 #define BPFIF_RUNLOCK(bif) rw_runlock(&(bif)->bif_lock)
122 #define BPFIF_WLOCK(bif) rw_wlock(&(bif)->bif_lock)
123 #define BPFIF_WUNLOCK(bif) rw_wunlock(&(bif)->bif_lock)
124
125 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
126
127 #define PRINET 26 /* interruptible */
128
129 #define SIZEOF_BPF_HDR(type) \
130 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
131
132 #ifdef COMPAT_FREEBSD32
133 #include <sys/mount.h>
134 #include <compat/freebsd32/freebsd32.h>
135 #define BPF_ALIGNMENT32 sizeof(int32_t)
136 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
137
138 #ifndef BURN_BRIDGES
139 /*
140 * 32-bit version of structure prepended to each packet. We use this header
141 * instead of the standard one for 32-bit streams. We mark the a stream as
142 * 32-bit the first time we see a 32-bit compat ioctl request.
143 */
144 struct bpf_hdr32 {
145 struct timeval32 bh_tstamp; /* time stamp */
146 uint32_t bh_caplen; /* length of captured portion */
147 uint32_t bh_datalen; /* original length of packet */
148 uint16_t bh_hdrlen; /* length of bpf header (this struct
149 plus alignment padding) */
150 };
151 #endif
152
153 struct bpf_program32 {
154 u_int bf_len;
155 uint32_t bf_insns;
156 };
157
158 struct bpf_dltlist32 {
159 u_int bfl_len;
160 u_int bfl_list;
161 };
162
163 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
164 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
165 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
166 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
167 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
168 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
169 #endif
170
171 #define BPF_LOCK() sx_xlock(&bpf_sx)
172 #define BPF_UNLOCK() sx_xunlock(&bpf_sx)
173 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED)
174 /*
175 * bpf_iflist is a list of BPF interface structures, each corresponding to a
176 * specific DLT. The same network interface might have several BPF interface
177 * structures registered by different layers in the stack (i.e., 802.11
178 * frames, ethernet frames, etc).
179 */
180 static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist;
181 static struct sx bpf_sx; /* bpf global lock */
182 static int bpf_bpfd_cnt;
183
184 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
185 static void bpf_detachd(struct bpf_d *);
186 static void bpf_detachd_locked(struct bpf_d *);
187 static void bpf_freed(struct bpf_d *);
188 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
189 struct sockaddr *, int *, struct bpf_d *);
190 static int bpf_setif(struct bpf_d *, struct ifreq *);
191 static void bpf_timed_out(void *);
192 static __inline void
193 bpf_wakeup(struct bpf_d *);
194 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
195 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
196 struct bintime *);
197 static void reset_d(struct bpf_d *);
198 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
199 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
200 static int bpf_setdlt(struct bpf_d *, u_int);
201 static void filt_bpfdetach(struct knote *);
202 static int filt_bpfread(struct knote *, long);
203 static void bpf_drvinit(void *);
204 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
205
206 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
207 int bpf_maxinsns = BPF_MAXINSNS;
208 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
209 &bpf_maxinsns, 0, "Maximum bpf program instructions");
210 static int bpf_zerocopy_enable = 0;
211 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
212 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
213 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
214 bpf_stats_sysctl, "bpf statistics portal");
215
216 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
217 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
218 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW,
219 &VNET_NAME(bpf_optimize_writers), 0,
220 "Do not send packets until BPF program is set");
221
222 static d_open_t bpfopen;
223 static d_read_t bpfread;
224 static d_write_t bpfwrite;
225 static d_ioctl_t bpfioctl;
226 static d_poll_t bpfpoll;
227 static d_kqfilter_t bpfkqfilter;
228
229 static struct cdevsw bpf_cdevsw = {
230 .d_version = D_VERSION,
231 .d_open = bpfopen,
232 .d_read = bpfread,
233 .d_write = bpfwrite,
234 .d_ioctl = bpfioctl,
235 .d_poll = bpfpoll,
236 .d_name = "bpf",
237 .d_kqfilter = bpfkqfilter,
238 };
239
240 static struct filterops bpfread_filtops = {
241 .f_isfd = 1,
242 .f_detach = filt_bpfdetach,
243 .f_event = filt_bpfread,
244 };
245
246 eventhandler_tag bpf_ifdetach_cookie = NULL;
247
248 /*
249 * LOCKING MODEL USED BY BPF:
250 * Locks:
251 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
252 * some global counters and every bpf_if reference.
253 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
254 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
255 * used by bpf_mtap code.
256 *
257 * Lock order:
258 *
259 * Global lock, interface lock, descriptor lock
260 *
261 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
262 * working model. In many places (like bpf_detachd) we start with BPF descriptor
263 * (and we need to at least rlock it to get reliable interface pointer). This
264 * gives us potential LOR. As a result, we use global lock to protect from bpf_if
265 * change in every such place.
266 *
267 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
268 * 3) descriptor main wlock.
269 * Reading bd_bif can be protected by any of these locks, typically global lock.
270 *
271 * Changing read/write BPF filter is protected by the same three locks,
272 * the same applies for reading.
273 *
274 * Sleeping in global lock is not allowed due to bpfdetach() using it.
275 */
276
277 /*
278 * Wrapper functions for various buffering methods. If the set of buffer
279 * modes expands, we will probably want to introduce a switch data structure
280 * similar to protosw, et.
281 */
282 static void
283 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
284 u_int len)
285 {
286
287 BPFD_LOCK_ASSERT(d);
288
289 switch (d->bd_bufmode) {
290 case BPF_BUFMODE_BUFFER:
291 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
292
293 case BPF_BUFMODE_ZBUF:
294 counter_u64_add(d->bd_zcopy, 1);
295 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
296
297 default:
298 panic("bpf_buf_append_bytes");
299 }
300 }
301
302 static void
303 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
304 u_int len)
305 {
306
307 BPFD_LOCK_ASSERT(d);
308
309 switch (d->bd_bufmode) {
310 case BPF_BUFMODE_BUFFER:
311 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
312
313 case BPF_BUFMODE_ZBUF:
314 counter_u64_add(d->bd_zcopy, 1);
315 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
316
317 default:
318 panic("bpf_buf_append_mbuf");
319 }
320 }
321
322 /*
323 * This function gets called when the free buffer is re-assigned.
324 */
325 static void
326 bpf_buf_reclaimed(struct bpf_d *d)
327 {
328
329 BPFD_LOCK_ASSERT(d);
330
331 switch (d->bd_bufmode) {
332 case BPF_BUFMODE_BUFFER:
333 return;
334
335 case BPF_BUFMODE_ZBUF:
336 bpf_zerocopy_buf_reclaimed(d);
337 return;
338
339 default:
340 panic("bpf_buf_reclaimed");
341 }
342 }
343
344 /*
345 * If the buffer mechanism has a way to decide that a held buffer can be made
346 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
347 * returned if the buffer can be discarded, (0) is returned if it cannot.
348 */
349 static int
350 bpf_canfreebuf(struct bpf_d *d)
351 {
352
353 BPFD_LOCK_ASSERT(d);
354
355 switch (d->bd_bufmode) {
356 case BPF_BUFMODE_ZBUF:
357 return (bpf_zerocopy_canfreebuf(d));
358 }
359 return (0);
360 }
361
362 /*
363 * Allow the buffer model to indicate that the current store buffer is
364 * immutable, regardless of the appearance of space. Return (1) if the
365 * buffer is writable, and (0) if not.
366 */
367 static int
368 bpf_canwritebuf(struct bpf_d *d)
369 {
370 BPFD_LOCK_ASSERT(d);
371
372 switch (d->bd_bufmode) {
373 case BPF_BUFMODE_ZBUF:
374 return (bpf_zerocopy_canwritebuf(d));
375 }
376 return (1);
377 }
378
379 /*
380 * Notify buffer model that an attempt to write to the store buffer has
381 * resulted in a dropped packet, in which case the buffer may be considered
382 * full.
383 */
384 static void
385 bpf_buffull(struct bpf_d *d)
386 {
387
388 BPFD_LOCK_ASSERT(d);
389
390 switch (d->bd_bufmode) {
391 case BPF_BUFMODE_ZBUF:
392 bpf_zerocopy_buffull(d);
393 break;
394 }
395 }
396
397 /*
398 * Notify the buffer model that a buffer has moved into the hold position.
399 */
400 void
401 bpf_bufheld(struct bpf_d *d)
402 {
403
404 BPFD_LOCK_ASSERT(d);
405
406 switch (d->bd_bufmode) {
407 case BPF_BUFMODE_ZBUF:
408 bpf_zerocopy_bufheld(d);
409 break;
410 }
411 }
412
413 static void
414 bpf_free(struct bpf_d *d)
415 {
416
417 switch (d->bd_bufmode) {
418 case BPF_BUFMODE_BUFFER:
419 return (bpf_buffer_free(d));
420
421 case BPF_BUFMODE_ZBUF:
422 return (bpf_zerocopy_free(d));
423
424 default:
425 panic("bpf_buf_free");
426 }
427 }
428
429 static int
430 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
431 {
432
433 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
434 return (EOPNOTSUPP);
435 return (bpf_buffer_uiomove(d, buf, len, uio));
436 }
437
438 static int
439 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
440 {
441
442 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
443 return (EOPNOTSUPP);
444 return (bpf_buffer_ioctl_sblen(d, i));
445 }
446
447 static int
448 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
449 {
450
451 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
452 return (EOPNOTSUPP);
453 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
454 }
455
456 static int
457 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
458 {
459
460 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
461 return (EOPNOTSUPP);
462 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
463 }
464
465 static int
466 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
467 {
468
469 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
470 return (EOPNOTSUPP);
471 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
472 }
473
474 /*
475 * General BPF functions.
476 */
477 static int
478 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
479 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
480 {
481 const struct ieee80211_bpf_params *p;
482 struct ether_header *eh;
483 struct mbuf *m;
484 int error;
485 int len;
486 int hlen;
487 int slen;
488
489 /*
490 * Build a sockaddr based on the data link layer type.
491 * We do this at this level because the ethernet header
492 * is copied directly into the data field of the sockaddr.
493 * In the case of SLIP, there is no header and the packet
494 * is forwarded as is.
495 * Also, we are careful to leave room at the front of the mbuf
496 * for the link level header.
497 */
498 switch (linktype) {
499
500 case DLT_SLIP:
501 sockp->sa_family = AF_INET;
502 hlen = 0;
503 break;
504
505 case DLT_EN10MB:
506 sockp->sa_family = AF_UNSPEC;
507 /* XXX Would MAXLINKHDR be better? */
508 hlen = ETHER_HDR_LEN;
509 break;
510
511 case DLT_FDDI:
512 sockp->sa_family = AF_IMPLINK;
513 hlen = 0;
514 break;
515
516 case DLT_RAW:
517 sockp->sa_family = AF_UNSPEC;
518 hlen = 0;
519 break;
520
521 case DLT_NULL:
522 /*
523 * null interface types require a 4 byte pseudo header which
524 * corresponds to the address family of the packet.
525 */
526 sockp->sa_family = AF_UNSPEC;
527 hlen = 4;
528 break;
529
530 case DLT_ATM_RFC1483:
531 /*
532 * en atm driver requires 4-byte atm pseudo header.
533 * though it isn't standard, vpi:vci needs to be
534 * specified anyway.
535 */
536 sockp->sa_family = AF_UNSPEC;
537 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
538 break;
539
540 case DLT_PPP:
541 sockp->sa_family = AF_UNSPEC;
542 hlen = 4; /* This should match PPP_HDRLEN */
543 break;
544
545 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
546 sockp->sa_family = AF_IEEE80211;
547 hlen = 0;
548 break;
549
550 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
551 sockp->sa_family = AF_IEEE80211;
552 sockp->sa_len = 12; /* XXX != 0 */
553 hlen = sizeof(struct ieee80211_bpf_params);
554 break;
555
556 default:
557 return (EIO);
558 }
559
560 len = uio->uio_resid;
561 if (len < hlen || len - hlen > ifp->if_mtu)
562 return (EMSGSIZE);
563
564 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
565 if (m == NULL)
566 return (EIO);
567 m->m_pkthdr.len = m->m_len = len;
568 *mp = m;
569
570 error = uiomove(mtod(m, u_char *), len, uio);
571 if (error)
572 goto bad;
573
574 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
575 if (slen == 0) {
576 error = EPERM;
577 goto bad;
578 }
579
580 /* Check for multicast destination */
581 switch (linktype) {
582 case DLT_EN10MB:
583 eh = mtod(m, struct ether_header *);
584 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
585 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
586 ETHER_ADDR_LEN) == 0)
587 m->m_flags |= M_BCAST;
588 else
589 m->m_flags |= M_MCAST;
590 }
591 if (d->bd_hdrcmplt == 0) {
592 memcpy(eh->ether_shost, IF_LLADDR(ifp),
593 sizeof(eh->ether_shost));
594 }
595 break;
596 }
597
598 /*
599 * Make room for link header, and copy it to sockaddr
600 */
601 if (hlen != 0) {
602 if (sockp->sa_family == AF_IEEE80211) {
603 /*
604 * Collect true length from the parameter header
605 * NB: sockp is known to be zero'd so if we do a
606 * short copy unspecified parameters will be
607 * zero.
608 * NB: packet may not be aligned after stripping
609 * bpf params
610 * XXX check ibp_vers
611 */
612 p = mtod(m, const struct ieee80211_bpf_params *);
613 hlen = p->ibp_len;
614 if (hlen > sizeof(sockp->sa_data)) {
615 error = EINVAL;
616 goto bad;
617 }
618 }
619 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
620 }
621 *hdrlen = hlen;
622
623 return (0);
624 bad:
625 m_freem(m);
626 return (error);
627 }
628
629 /*
630 * Attach file to the bpf interface, i.e. make d listen on bp.
631 */
632 static void
633 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
634 {
635 int op_w;
636
637 BPF_LOCK_ASSERT();
638
639 /*
640 * Save sysctl value to protect from sysctl change
641 * between reads
642 */
643 op_w = V_bpf_optimize_writers || d->bd_writer;
644
645 if (d->bd_bif != NULL)
646 bpf_detachd_locked(d);
647 /*
648 * Point d at bp, and add d to the interface's list.
649 * Since there are many applications using BPF for
650 * sending raw packets only (dhcpd, cdpd are good examples)
651 * we can delay adding d to the list of active listeners until
652 * some filter is configured.
653 */
654
655 BPFIF_WLOCK(bp);
656 BPFD_LOCK(d);
657
658 d->bd_bif = bp;
659
660 if (op_w != 0) {
661 /* Add to writers-only list */
662 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
663 /*
664 * We decrement bd_writer on every filter set operation.
665 * First BIOCSETF is done by pcap_open_live() to set up
666 * snap length. After that appliation usually sets its own filter
667 */
668 d->bd_writer = 2;
669 } else
670 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
671
672 BPFD_UNLOCK(d);
673 BPFIF_WUNLOCK(bp);
674
675 bpf_bpfd_cnt++;
676
677 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
678 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
679
680 if (op_w == 0)
681 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
682 }
683
684 /*
685 * Check if we need to upgrade our descriptor @d from write-only mode.
686 */
687 static int
688 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen)
689 {
690 int is_snap, need_upgrade;
691
692 /*
693 * Check if we've already upgraded or new filter is empty.
694 */
695 if (d->bd_writer == 0 || fcode == NULL)
696 return (0);
697
698 need_upgrade = 0;
699
700 /*
701 * Check if cmd looks like snaplen setting from
702 * pcap_bpf.c:pcap_open_live().
703 * Note we're not checking .k value here:
704 * while pcap_open_live() definitely sets to non-zero value,
705 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
706 * do not consider upgrading immediately
707 */
708 if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K))
709 is_snap = 1;
710 else
711 is_snap = 0;
712
713 if (is_snap == 0) {
714 /*
715 * We're setting first filter and it doesn't look like
716 * setting snaplen. We're probably using bpf directly.
717 * Upgrade immediately.
718 */
719 need_upgrade = 1;
720 } else {
721 /*
722 * Do not require upgrade by first BIOCSETF
723 * (used to set snaplen) by pcap_open_live().
724 */
725
726 if (--d->bd_writer == 0) {
727 /*
728 * First snaplen filter has already
729 * been set. This is probably catch-all
730 * filter
731 */
732 need_upgrade = 1;
733 }
734 }
735
736 CTR5(KTR_NET,
737 "%s: filter function set by pid %d, "
738 "bd_writer counter %d, snap %d upgrade %d",
739 __func__, d->bd_pid, d->bd_writer,
740 is_snap, need_upgrade);
741
742 return (need_upgrade);
743 }
744
745 /*
746 * Add d to the list of active bp filters.
747 * Requires bpf_attachd() to be called before.
748 */
749 static void
750 bpf_upgraded(struct bpf_d *d)
751 {
752 struct bpf_if *bp;
753
754 BPF_LOCK_ASSERT();
755
756 bp = d->bd_bif;
757
758 /*
759 * Filter can be set several times without specifying interface.
760 * Mark d as reader and exit.
761 */
762 if (bp == NULL) {
763 BPFD_LOCK(d);
764 d->bd_writer = 0;
765 BPFD_UNLOCK(d);
766 return;
767 }
768
769 BPFIF_WLOCK(bp);
770 BPFD_LOCK(d);
771
772 /* Remove from writers-only list */
773 LIST_REMOVE(d, bd_next);
774 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
775 /* Mark d as reader */
776 d->bd_writer = 0;
777
778 BPFD_UNLOCK(d);
779 BPFIF_WUNLOCK(bp);
780
781 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
782
783 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
784 }
785
786 /*
787 * Detach a file from its interface.
788 */
789 static void
790 bpf_detachd(struct bpf_d *d)
791 {
792 BPF_LOCK();
793 bpf_detachd_locked(d);
794 BPF_UNLOCK();
795 }
796
797 static void
798 bpf_detachd_locked(struct bpf_d *d)
799 {
800 int error;
801 struct bpf_if *bp;
802 struct ifnet *ifp;
803
804 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
805
806 BPF_LOCK_ASSERT();
807
808 /* Check if descriptor is attached */
809 if ((bp = d->bd_bif) == NULL)
810 return;
811
812 BPFIF_WLOCK(bp);
813 BPFD_LOCK(d);
814
815 /* Save bd_writer value */
816 error = d->bd_writer;
817
818 /*
819 * Remove d from the interface's descriptor list.
820 */
821 LIST_REMOVE(d, bd_next);
822
823 ifp = bp->bif_ifp;
824 d->bd_bif = NULL;
825 BPFD_UNLOCK(d);
826 BPFIF_WUNLOCK(bp);
827
828 bpf_bpfd_cnt--;
829
830 /* Call event handler iff d is attached */
831 if (error == 0)
832 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
833
834 /*
835 * Check if this descriptor had requested promiscuous mode.
836 * If so, turn it off.
837 */
838 if (d->bd_promisc) {
839 d->bd_promisc = 0;
840 CURVNET_SET(ifp->if_vnet);
841 error = ifpromisc(ifp, 0);
842 CURVNET_RESTORE();
843 if (error != 0 && error != ENXIO) {
844 /*
845 * ENXIO can happen if a pccard is unplugged
846 * Something is really wrong if we were able to put
847 * the driver into promiscuous mode, but can't
848 * take it out.
849 */
850 if_printf(bp->bif_ifp,
851 "bpf_detach: ifpromisc failed (%d)\n", error);
852 }
853 }
854 }
855
856 /*
857 * Close the descriptor by detaching it from its interface,
858 * deallocating its buffers, and marking it free.
859 */
860 static void
861 bpf_dtor(void *data)
862 {
863 struct bpf_d *d = data;
864
865 BPFD_LOCK(d);
866 if (d->bd_state == BPF_WAITING)
867 callout_stop(&d->bd_callout);
868 d->bd_state = BPF_IDLE;
869 BPFD_UNLOCK(d);
870 funsetown(&d->bd_sigio);
871 bpf_detachd(d);
872 #ifdef MAC
873 mac_bpfdesc_destroy(d);
874 #endif /* MAC */
875 seldrain(&d->bd_sel);
876 knlist_destroy(&d->bd_sel.si_note);
877 callout_drain(&d->bd_callout);
878 bpf_freed(d);
879 free(d, M_BPF);
880 }
881
882 /*
883 * Open ethernet device. Returns ENXIO for illegal minor device number,
884 * EBUSY if file is open by another process.
885 */
886 /* ARGSUSED */
887 static int
888 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
889 {
890 struct bpf_d *d;
891 int error;
892
893 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
894 error = devfs_set_cdevpriv(d, bpf_dtor);
895 if (error != 0) {
896 free(d, M_BPF);
897 return (error);
898 }
899
900 /* Setup counters */
901 d->bd_rcount = counter_u64_alloc(M_WAITOK);
902 d->bd_dcount = counter_u64_alloc(M_WAITOK);
903 d->bd_fcount = counter_u64_alloc(M_WAITOK);
904 d->bd_wcount = counter_u64_alloc(M_WAITOK);
905 d->bd_wfcount = counter_u64_alloc(M_WAITOK);
906 d->bd_wdcount = counter_u64_alloc(M_WAITOK);
907 d->bd_zcopy = counter_u64_alloc(M_WAITOK);
908
909 /*
910 * For historical reasons, perform a one-time initialization call to
911 * the buffer routines, even though we're not yet committed to a
912 * particular buffer method.
913 */
914 bpf_buffer_init(d);
915 if ((flags & FREAD) == 0)
916 d->bd_writer = 2;
917 d->bd_hbuf_in_use = 0;
918 d->bd_bufmode = BPF_BUFMODE_BUFFER;
919 d->bd_sig = SIGIO;
920 d->bd_direction = BPF_D_INOUT;
921 BPF_PID_REFRESH(d, td);
922 #ifdef MAC
923 mac_bpfdesc_init(d);
924 mac_bpfdesc_create(td->td_ucred, d);
925 #endif
926 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
927 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
928 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
929
930 return (0);
931 }
932
933 /*
934 * bpfread - read next chunk of packets from buffers
935 */
936 static int
937 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
938 {
939 struct bpf_d *d;
940 int error;
941 int non_block;
942 int timed_out;
943
944 error = devfs_get_cdevpriv((void **)&d);
945 if (error != 0)
946 return (error);
947
948 /*
949 * Restrict application to use a buffer the same size as
950 * as kernel buffers.
951 */
952 if (uio->uio_resid != d->bd_bufsize)
953 return (EINVAL);
954
955 non_block = ((ioflag & O_NONBLOCK) != 0);
956
957 BPFD_LOCK(d);
958 BPF_PID_REFRESH_CUR(d);
959 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
960 BPFD_UNLOCK(d);
961 return (EOPNOTSUPP);
962 }
963 if (d->bd_state == BPF_WAITING)
964 callout_stop(&d->bd_callout);
965 timed_out = (d->bd_state == BPF_TIMED_OUT);
966 d->bd_state = BPF_IDLE;
967 while (d->bd_hbuf_in_use) {
968 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
969 PRINET|PCATCH, "bd_hbuf", 0);
970 if (error != 0) {
971 BPFD_UNLOCK(d);
972 return (error);
973 }
974 }
975 /*
976 * If the hold buffer is empty, then do a timed sleep, which
977 * ends when the timeout expires or when enough packets
978 * have arrived to fill the store buffer.
979 */
980 while (d->bd_hbuf == NULL) {
981 if (d->bd_slen != 0) {
982 /*
983 * A packet(s) either arrived since the previous
984 * read or arrived while we were asleep.
985 */
986 if (d->bd_immediate || non_block || timed_out) {
987 /*
988 * Rotate the buffers and return what's here
989 * if we are in immediate mode, non-blocking
990 * flag is set, or this descriptor timed out.
991 */
992 ROTATE_BUFFERS(d);
993 break;
994 }
995 }
996
997 /*
998 * No data is available, check to see if the bpf device
999 * is still pointed at a real interface. If not, return
1000 * ENXIO so that the userland process knows to rebind
1001 * it before using it again.
1002 */
1003 if (d->bd_bif == NULL) {
1004 BPFD_UNLOCK(d);
1005 return (ENXIO);
1006 }
1007
1008 if (non_block) {
1009 BPFD_UNLOCK(d);
1010 return (EWOULDBLOCK);
1011 }
1012 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
1013 "bpf", d->bd_rtout);
1014 if (error == EINTR || error == ERESTART) {
1015 BPFD_UNLOCK(d);
1016 return (error);
1017 }
1018 if (error == EWOULDBLOCK) {
1019 /*
1020 * On a timeout, return what's in the buffer,
1021 * which may be nothing. If there is something
1022 * in the store buffer, we can rotate the buffers.
1023 */
1024 if (d->bd_hbuf)
1025 /*
1026 * We filled up the buffer in between
1027 * getting the timeout and arriving
1028 * here, so we don't need to rotate.
1029 */
1030 break;
1031
1032 if (d->bd_slen == 0) {
1033 BPFD_UNLOCK(d);
1034 return (0);
1035 }
1036 ROTATE_BUFFERS(d);
1037 break;
1038 }
1039 }
1040 /*
1041 * At this point, we know we have something in the hold slot.
1042 */
1043 d->bd_hbuf_in_use = 1;
1044 BPFD_UNLOCK(d);
1045
1046 /*
1047 * Move data from hold buffer into user space.
1048 * We know the entire buffer is transferred since
1049 * we checked above that the read buffer is bpf_bufsize bytes.
1050 *
1051 * We do not have to worry about simultaneous reads because
1052 * we waited for sole access to the hold buffer above.
1053 */
1054 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1055
1056 BPFD_LOCK(d);
1057 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1058 d->bd_fbuf = d->bd_hbuf;
1059 d->bd_hbuf = NULL;
1060 d->bd_hlen = 0;
1061 bpf_buf_reclaimed(d);
1062 d->bd_hbuf_in_use = 0;
1063 wakeup(&d->bd_hbuf_in_use);
1064 BPFD_UNLOCK(d);
1065
1066 return (error);
1067 }
1068
1069 /*
1070 * If there are processes sleeping on this descriptor, wake them up.
1071 */
1072 static __inline void
1073 bpf_wakeup(struct bpf_d *d)
1074 {
1075
1076 BPFD_LOCK_ASSERT(d);
1077 if (d->bd_state == BPF_WAITING) {
1078 callout_stop(&d->bd_callout);
1079 d->bd_state = BPF_IDLE;
1080 }
1081 wakeup(d);
1082 if (d->bd_async && d->bd_sig && d->bd_sigio)
1083 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1084
1085 selwakeuppri(&d->bd_sel, PRINET);
1086 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1087 }
1088
1089 static void
1090 bpf_timed_out(void *arg)
1091 {
1092 struct bpf_d *d = (struct bpf_d *)arg;
1093
1094 BPFD_LOCK_ASSERT(d);
1095
1096 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
1097 return;
1098 if (d->bd_state == BPF_WAITING) {
1099 d->bd_state = BPF_TIMED_OUT;
1100 if (d->bd_slen != 0)
1101 bpf_wakeup(d);
1102 }
1103 }
1104
1105 static int
1106 bpf_ready(struct bpf_d *d)
1107 {
1108
1109 BPFD_LOCK_ASSERT(d);
1110
1111 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1112 return (1);
1113 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1114 d->bd_slen != 0)
1115 return (1);
1116 return (0);
1117 }
1118
1119 static int
1120 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1121 {
1122 struct bpf_d *d;
1123 struct ifnet *ifp;
1124 struct mbuf *m, *mc;
1125 struct sockaddr dst;
1126 struct route ro;
1127 int error, hlen;
1128
1129 error = devfs_get_cdevpriv((void **)&d);
1130 if (error != 0)
1131 return (error);
1132
1133 BPF_PID_REFRESH_CUR(d);
1134 counter_u64_add(d->bd_wcount, 1);
1135 /* XXX: locking required */
1136 if (d->bd_bif == NULL) {
1137 counter_u64_add(d->bd_wdcount, 1);
1138 return (ENXIO);
1139 }
1140
1141 ifp = d->bd_bif->bif_ifp;
1142
1143 if ((ifp->if_flags & IFF_UP) == 0) {
1144 counter_u64_add(d->bd_wdcount, 1);
1145 return (ENETDOWN);
1146 }
1147
1148 if (uio->uio_resid == 0) {
1149 counter_u64_add(d->bd_wdcount, 1);
1150 return (0);
1151 }
1152
1153 bzero(&dst, sizeof(dst));
1154 m = NULL;
1155 hlen = 0;
1156 /* XXX: bpf_movein() can sleep */
1157 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1158 &m, &dst, &hlen, d);
1159 if (error) {
1160 counter_u64_add(d->bd_wdcount, 1);
1161 return (error);
1162 }
1163 counter_u64_add(d->bd_wfcount, 1);
1164 if (d->bd_hdrcmplt)
1165 dst.sa_family = pseudo_AF_HDRCMPLT;
1166
1167 if (d->bd_feedback) {
1168 mc = m_dup(m, M_NOWAIT);
1169 if (mc != NULL)
1170 mc->m_pkthdr.rcvif = ifp;
1171 /* Set M_PROMISC for outgoing packets to be discarded. */
1172 if (d->bd_direction == BPF_D_INOUT)
1173 m->m_flags |= M_PROMISC;
1174 } else
1175 mc = NULL;
1176
1177 m->m_pkthdr.len -= hlen;
1178 m->m_len -= hlen;
1179 m->m_data += hlen; /* XXX */
1180
1181 CURVNET_SET(ifp->if_vnet);
1182 #ifdef MAC
1183 BPFD_LOCK(d);
1184 mac_bpfdesc_create_mbuf(d, m);
1185 if (mc != NULL)
1186 mac_bpfdesc_create_mbuf(d, mc);
1187 BPFD_UNLOCK(d);
1188 #endif
1189
1190 bzero(&ro, sizeof(ro));
1191 if (hlen != 0) {
1192 ro.ro_prepend = (u_char *)&dst.sa_data;
1193 ro.ro_plen = hlen;
1194 ro.ro_flags = RT_HAS_HEADER;
1195 }
1196
1197 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1198 if (error)
1199 counter_u64_add(d->bd_wdcount, 1);
1200
1201 if (mc != NULL) {
1202 if (error == 0)
1203 (*ifp->if_input)(ifp, mc);
1204 else
1205 m_freem(mc);
1206 }
1207 CURVNET_RESTORE();
1208
1209 return (error);
1210 }
1211
1212 /*
1213 * Reset a descriptor by flushing its packet buffer and clearing the receive
1214 * and drop counts. This is doable for kernel-only buffers, but with
1215 * zero-copy buffers, we can't write to (or rotate) buffers that are
1216 * currently owned by userspace. It would be nice if we could encapsulate
1217 * this logic in the buffer code rather than here.
1218 */
1219 static void
1220 reset_d(struct bpf_d *d)
1221 {
1222
1223 BPFD_LOCK_ASSERT(d);
1224
1225 while (d->bd_hbuf_in_use)
1226 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1227 "bd_hbuf", 0);
1228 if ((d->bd_hbuf != NULL) &&
1229 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1230 /* Free the hold buffer. */
1231 d->bd_fbuf = d->bd_hbuf;
1232 d->bd_hbuf = NULL;
1233 d->bd_hlen = 0;
1234 bpf_buf_reclaimed(d);
1235 }
1236 if (bpf_canwritebuf(d))
1237 d->bd_slen = 0;
1238 counter_u64_zero(d->bd_rcount);
1239 counter_u64_zero(d->bd_dcount);
1240 counter_u64_zero(d->bd_fcount);
1241 counter_u64_zero(d->bd_wcount);
1242 counter_u64_zero(d->bd_wfcount);
1243 counter_u64_zero(d->bd_wdcount);
1244 counter_u64_zero(d->bd_zcopy);
1245 }
1246
1247 /*
1248 * FIONREAD Check for read packet available.
1249 * BIOCGBLEN Get buffer len [for read()].
1250 * BIOCSETF Set read filter.
1251 * BIOCSETFNR Set read filter without resetting descriptor.
1252 * BIOCSETWF Set write filter.
1253 * BIOCFLUSH Flush read packet buffer.
1254 * BIOCPROMISC Put interface into promiscuous mode.
1255 * BIOCGDLT Get link layer type.
1256 * BIOCGETIF Get interface name.
1257 * BIOCSETIF Set interface.
1258 * BIOCSRTIMEOUT Set read timeout.
1259 * BIOCGRTIMEOUT Get read timeout.
1260 * BIOCGSTATS Get packet stats.
1261 * BIOCIMMEDIATE Set immediate mode.
1262 * BIOCVERSION Get filter language version.
1263 * BIOCGHDRCMPLT Get "header already complete" flag
1264 * BIOCSHDRCMPLT Set "header already complete" flag
1265 * BIOCGDIRECTION Get packet direction flag
1266 * BIOCSDIRECTION Set packet direction flag
1267 * BIOCGTSTAMP Get time stamp format and resolution.
1268 * BIOCSTSTAMP Set time stamp format and resolution.
1269 * BIOCLOCK Set "locked" flag
1270 * BIOCFEEDBACK Set packet feedback mode.
1271 * BIOCSETZBUF Set current zero-copy buffer locations.
1272 * BIOCGETZMAX Get maximum zero-copy buffer size.
1273 * BIOCROTZBUF Force rotation of zero-copy buffer
1274 * BIOCSETBUFMODE Set buffer mode.
1275 * BIOCGETBUFMODE Get current buffer mode.
1276 */
1277 /* ARGSUSED */
1278 static int
1279 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1280 struct thread *td)
1281 {
1282 struct bpf_d *d;
1283 int error;
1284
1285 error = devfs_get_cdevpriv((void **)&d);
1286 if (error != 0)
1287 return (error);
1288
1289 /*
1290 * Refresh PID associated with this descriptor.
1291 */
1292 BPFD_LOCK(d);
1293 BPF_PID_REFRESH(d, td);
1294 if (d->bd_state == BPF_WAITING)
1295 callout_stop(&d->bd_callout);
1296 d->bd_state = BPF_IDLE;
1297 BPFD_UNLOCK(d);
1298
1299 if (d->bd_locked == 1) {
1300 switch (cmd) {
1301 case BIOCGBLEN:
1302 case BIOCFLUSH:
1303 case BIOCGDLT:
1304 case BIOCGDLTLIST:
1305 #ifdef COMPAT_FREEBSD32
1306 case BIOCGDLTLIST32:
1307 #endif
1308 case BIOCGETIF:
1309 case BIOCGRTIMEOUT:
1310 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1311 case BIOCGRTIMEOUT32:
1312 #endif
1313 case BIOCGSTATS:
1314 case BIOCVERSION:
1315 case BIOCGRSIG:
1316 case BIOCGHDRCMPLT:
1317 case BIOCSTSTAMP:
1318 case BIOCFEEDBACK:
1319 case FIONREAD:
1320 case BIOCLOCK:
1321 case BIOCSRTIMEOUT:
1322 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1323 case BIOCSRTIMEOUT32:
1324 #endif
1325 case BIOCIMMEDIATE:
1326 case TIOCGPGRP:
1327 case BIOCROTZBUF:
1328 break;
1329 default:
1330 return (EPERM);
1331 }
1332 }
1333 #ifdef COMPAT_FREEBSD32
1334 /*
1335 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1336 * that it will get 32-bit packet headers.
1337 */
1338 switch (cmd) {
1339 case BIOCSETF32:
1340 case BIOCSETFNR32:
1341 case BIOCSETWF32:
1342 case BIOCGDLTLIST32:
1343 case BIOCGRTIMEOUT32:
1344 case BIOCSRTIMEOUT32:
1345 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1346 BPFD_LOCK(d);
1347 d->bd_compat32 = 1;
1348 BPFD_UNLOCK(d);
1349 }
1350 }
1351 #endif
1352
1353 CURVNET_SET(TD_TO_VNET(td));
1354 switch (cmd) {
1355
1356 default:
1357 error = EINVAL;
1358 break;
1359
1360 /*
1361 * Check for read packet available.
1362 */
1363 case FIONREAD:
1364 {
1365 int n;
1366
1367 BPFD_LOCK(d);
1368 n = d->bd_slen;
1369 while (d->bd_hbuf_in_use)
1370 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1371 PRINET, "bd_hbuf", 0);
1372 if (d->bd_hbuf)
1373 n += d->bd_hlen;
1374 BPFD_UNLOCK(d);
1375
1376 *(int *)addr = n;
1377 break;
1378 }
1379
1380 /*
1381 * Get buffer len [for read()].
1382 */
1383 case BIOCGBLEN:
1384 BPFD_LOCK(d);
1385 *(u_int *)addr = d->bd_bufsize;
1386 BPFD_UNLOCK(d);
1387 break;
1388
1389 /*
1390 * Set buffer length.
1391 */
1392 case BIOCSBLEN:
1393 error = bpf_ioctl_sblen(d, (u_int *)addr);
1394 break;
1395
1396 /*
1397 * Set link layer read filter.
1398 */
1399 case BIOCSETF:
1400 case BIOCSETFNR:
1401 case BIOCSETWF:
1402 #ifdef COMPAT_FREEBSD32
1403 case BIOCSETF32:
1404 case BIOCSETFNR32:
1405 case BIOCSETWF32:
1406 #endif
1407 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1408 break;
1409
1410 /*
1411 * Flush read packet buffer.
1412 */
1413 case BIOCFLUSH:
1414 BPFD_LOCK(d);
1415 reset_d(d);
1416 BPFD_UNLOCK(d);
1417 break;
1418
1419 /*
1420 * Put interface into promiscuous mode.
1421 */
1422 case BIOCPROMISC:
1423 if (d->bd_bif == NULL) {
1424 /*
1425 * No interface attached yet.
1426 */
1427 error = EINVAL;
1428 break;
1429 }
1430 if (d->bd_promisc == 0) {
1431 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1432 if (error == 0)
1433 d->bd_promisc = 1;
1434 }
1435 break;
1436
1437 /*
1438 * Get current data link type.
1439 */
1440 case BIOCGDLT:
1441 BPF_LOCK();
1442 if (d->bd_bif == NULL)
1443 error = EINVAL;
1444 else
1445 *(u_int *)addr = d->bd_bif->bif_dlt;
1446 BPF_UNLOCK();
1447 break;
1448
1449 /*
1450 * Get a list of supported data link types.
1451 */
1452 #ifdef COMPAT_FREEBSD32
1453 case BIOCGDLTLIST32:
1454 {
1455 struct bpf_dltlist32 *list32;
1456 struct bpf_dltlist dltlist;
1457
1458 list32 = (struct bpf_dltlist32 *)addr;
1459 dltlist.bfl_len = list32->bfl_len;
1460 dltlist.bfl_list = PTRIN(list32->bfl_list);
1461 BPF_LOCK();
1462 if (d->bd_bif == NULL)
1463 error = EINVAL;
1464 else {
1465 error = bpf_getdltlist(d, &dltlist);
1466 if (error == 0)
1467 list32->bfl_len = dltlist.bfl_len;
1468 }
1469 BPF_UNLOCK();
1470 break;
1471 }
1472 #endif
1473
1474 case BIOCGDLTLIST:
1475 BPF_LOCK();
1476 if (d->bd_bif == NULL)
1477 error = EINVAL;
1478 else
1479 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1480 BPF_UNLOCK();
1481 break;
1482
1483 /*
1484 * Set data link type.
1485 */
1486 case BIOCSDLT:
1487 BPF_LOCK();
1488 if (d->bd_bif == NULL)
1489 error = EINVAL;
1490 else
1491 error = bpf_setdlt(d, *(u_int *)addr);
1492 BPF_UNLOCK();
1493 break;
1494
1495 /*
1496 * Get interface name.
1497 */
1498 case BIOCGETIF:
1499 BPF_LOCK();
1500 if (d->bd_bif == NULL)
1501 error = EINVAL;
1502 else {
1503 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1504 struct ifreq *const ifr = (struct ifreq *)addr;
1505
1506 strlcpy(ifr->ifr_name, ifp->if_xname,
1507 sizeof(ifr->ifr_name));
1508 }
1509 BPF_UNLOCK();
1510 break;
1511
1512 /*
1513 * Set interface.
1514 */
1515 case BIOCSETIF:
1516 {
1517 int alloc_buf, size;
1518
1519 /*
1520 * Behavior here depends on the buffering model. If
1521 * we're using kernel memory buffers, then we can
1522 * allocate them here. If we're using zero-copy,
1523 * then the user process must have registered buffers
1524 * by the time we get here.
1525 */
1526 alloc_buf = 0;
1527 BPFD_LOCK(d);
1528 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1529 d->bd_sbuf == NULL)
1530 alloc_buf = 1;
1531 BPFD_UNLOCK(d);
1532 if (alloc_buf) {
1533 size = d->bd_bufsize;
1534 error = bpf_buffer_ioctl_sblen(d, &size);
1535 if (error != 0)
1536 break;
1537 }
1538 BPF_LOCK();
1539 error = bpf_setif(d, (struct ifreq *)addr);
1540 BPF_UNLOCK();
1541 break;
1542 }
1543
1544 /*
1545 * Set read timeout.
1546 */
1547 case BIOCSRTIMEOUT:
1548 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1549 case BIOCSRTIMEOUT32:
1550 #endif
1551 {
1552 struct timeval *tv = (struct timeval *)addr;
1553 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1554 struct timeval32 *tv32;
1555 struct timeval tv64;
1556
1557 if (cmd == BIOCSRTIMEOUT32) {
1558 tv32 = (struct timeval32 *)addr;
1559 tv = &tv64;
1560 tv->tv_sec = tv32->tv_sec;
1561 tv->tv_usec = tv32->tv_usec;
1562 } else
1563 #endif
1564 tv = (struct timeval *)addr;
1565
1566 /*
1567 * Subtract 1 tick from tvtohz() since this isn't
1568 * a one-shot timer.
1569 */
1570 if ((error = itimerfix(tv)) == 0)
1571 d->bd_rtout = tvtohz(tv) - 1;
1572 break;
1573 }
1574
1575 /*
1576 * Get read timeout.
1577 */
1578 case BIOCGRTIMEOUT:
1579 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1580 case BIOCGRTIMEOUT32:
1581 #endif
1582 {
1583 struct timeval *tv;
1584 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1585 struct timeval32 *tv32;
1586 struct timeval tv64;
1587
1588 if (cmd == BIOCGRTIMEOUT32)
1589 tv = &tv64;
1590 else
1591 #endif
1592 tv = (struct timeval *)addr;
1593
1594 tv->tv_sec = d->bd_rtout / hz;
1595 tv->tv_usec = (d->bd_rtout % hz) * tick;
1596 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1597 if (cmd == BIOCGRTIMEOUT32) {
1598 tv32 = (struct timeval32 *)addr;
1599 tv32->tv_sec = tv->tv_sec;
1600 tv32->tv_usec = tv->tv_usec;
1601 }
1602 #endif
1603
1604 break;
1605 }
1606
1607 /*
1608 * Get packet stats.
1609 */
1610 case BIOCGSTATS:
1611 {
1612 struct bpf_stat *bs = (struct bpf_stat *)addr;
1613
1614 /* XXXCSJP overflow */
1615 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1616 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1617 break;
1618 }
1619
1620 /*
1621 * Set immediate mode.
1622 */
1623 case BIOCIMMEDIATE:
1624 BPFD_LOCK(d);
1625 d->bd_immediate = *(u_int *)addr;
1626 BPFD_UNLOCK(d);
1627 break;
1628
1629 case BIOCVERSION:
1630 {
1631 struct bpf_version *bv = (struct bpf_version *)addr;
1632
1633 bv->bv_major = BPF_MAJOR_VERSION;
1634 bv->bv_minor = BPF_MINOR_VERSION;
1635 break;
1636 }
1637
1638 /*
1639 * Get "header already complete" flag
1640 */
1641 case BIOCGHDRCMPLT:
1642 BPFD_LOCK(d);
1643 *(u_int *)addr = d->bd_hdrcmplt;
1644 BPFD_UNLOCK(d);
1645 break;
1646
1647 /*
1648 * Set "header already complete" flag
1649 */
1650 case BIOCSHDRCMPLT:
1651 BPFD_LOCK(d);
1652 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1653 BPFD_UNLOCK(d);
1654 break;
1655
1656 /*
1657 * Get packet direction flag
1658 */
1659 case BIOCGDIRECTION:
1660 BPFD_LOCK(d);
1661 *(u_int *)addr = d->bd_direction;
1662 BPFD_UNLOCK(d);
1663 break;
1664
1665 /*
1666 * Set packet direction flag
1667 */
1668 case BIOCSDIRECTION:
1669 {
1670 u_int direction;
1671
1672 direction = *(u_int *)addr;
1673 switch (direction) {
1674 case BPF_D_IN:
1675 case BPF_D_INOUT:
1676 case BPF_D_OUT:
1677 BPFD_LOCK(d);
1678 d->bd_direction = direction;
1679 BPFD_UNLOCK(d);
1680 break;
1681 default:
1682 error = EINVAL;
1683 }
1684 }
1685 break;
1686
1687 /*
1688 * Get packet timestamp format and resolution.
1689 */
1690 case BIOCGTSTAMP:
1691 BPFD_LOCK(d);
1692 *(u_int *)addr = d->bd_tstamp;
1693 BPFD_UNLOCK(d);
1694 break;
1695
1696 /*
1697 * Set packet timestamp format and resolution.
1698 */
1699 case BIOCSTSTAMP:
1700 {
1701 u_int func;
1702
1703 func = *(u_int *)addr;
1704 if (BPF_T_VALID(func))
1705 d->bd_tstamp = func;
1706 else
1707 error = EINVAL;
1708 }
1709 break;
1710
1711 case BIOCFEEDBACK:
1712 BPFD_LOCK(d);
1713 d->bd_feedback = *(u_int *)addr;
1714 BPFD_UNLOCK(d);
1715 break;
1716
1717 case BIOCLOCK:
1718 BPFD_LOCK(d);
1719 d->bd_locked = 1;
1720 BPFD_UNLOCK(d);
1721 break;
1722
1723 case FIONBIO: /* Non-blocking I/O */
1724 break;
1725
1726 case FIOASYNC: /* Send signal on receive packets */
1727 BPFD_LOCK(d);
1728 d->bd_async = *(int *)addr;
1729 BPFD_UNLOCK(d);
1730 break;
1731
1732 case FIOSETOWN:
1733 /*
1734 * XXX: Add some sort of locking here?
1735 * fsetown() can sleep.
1736 */
1737 error = fsetown(*(int *)addr, &d->bd_sigio);
1738 break;
1739
1740 case FIOGETOWN:
1741 BPFD_LOCK(d);
1742 *(int *)addr = fgetown(&d->bd_sigio);
1743 BPFD_UNLOCK(d);
1744 break;
1745
1746 /* This is deprecated, FIOSETOWN should be used instead. */
1747 case TIOCSPGRP:
1748 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1749 break;
1750
1751 /* This is deprecated, FIOGETOWN should be used instead. */
1752 case TIOCGPGRP:
1753 *(int *)addr = -fgetown(&d->bd_sigio);
1754 break;
1755
1756 case BIOCSRSIG: /* Set receive signal */
1757 {
1758 u_int sig;
1759
1760 sig = *(u_int *)addr;
1761
1762 if (sig >= NSIG)
1763 error = EINVAL;
1764 else {
1765 BPFD_LOCK(d);
1766 d->bd_sig = sig;
1767 BPFD_UNLOCK(d);
1768 }
1769 break;
1770 }
1771 case BIOCGRSIG:
1772 BPFD_LOCK(d);
1773 *(u_int *)addr = d->bd_sig;
1774 BPFD_UNLOCK(d);
1775 break;
1776
1777 case BIOCGETBUFMODE:
1778 BPFD_LOCK(d);
1779 *(u_int *)addr = d->bd_bufmode;
1780 BPFD_UNLOCK(d);
1781 break;
1782
1783 case BIOCSETBUFMODE:
1784 /*
1785 * Allow the buffering mode to be changed as long as we
1786 * haven't yet committed to a particular mode. Our
1787 * definition of commitment, for now, is whether or not a
1788 * buffer has been allocated or an interface attached, since
1789 * that's the point where things get tricky.
1790 */
1791 switch (*(u_int *)addr) {
1792 case BPF_BUFMODE_BUFFER:
1793 break;
1794
1795 case BPF_BUFMODE_ZBUF:
1796 if (bpf_zerocopy_enable)
1797 break;
1798 /* FALLSTHROUGH */
1799
1800 default:
1801 CURVNET_RESTORE();
1802 return (EINVAL);
1803 }
1804
1805 BPFD_LOCK(d);
1806 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1807 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1808 BPFD_UNLOCK(d);
1809 CURVNET_RESTORE();
1810 return (EBUSY);
1811 }
1812 d->bd_bufmode = *(u_int *)addr;
1813 BPFD_UNLOCK(d);
1814 break;
1815
1816 case BIOCGETZMAX:
1817 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1818 break;
1819
1820 case BIOCSETZBUF:
1821 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1822 break;
1823
1824 case BIOCROTZBUF:
1825 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1826 break;
1827 }
1828 CURVNET_RESTORE();
1829 return (error);
1830 }
1831
1832 /*
1833 * Set d's packet filter program to fp. If this file already has a filter,
1834 * free it and replace it. Returns EINVAL for bogus requests.
1835 *
1836 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1837 * since reading d->bd_bif can't be protected by d or interface lock due to
1838 * lock order.
1839 *
1840 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1841 * interface read lock to read all filers.
1842 *
1843 */
1844 static int
1845 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1846 {
1847 #ifdef COMPAT_FREEBSD32
1848 struct bpf_program fp_swab;
1849 struct bpf_program32 *fp32;
1850 #endif
1851 struct bpf_insn *fcode, *old;
1852 #ifdef BPF_JITTER
1853 bpf_jit_filter *jfunc, *ofunc;
1854 #endif
1855 size_t size;
1856 u_int flen;
1857 int need_upgrade;
1858
1859 #ifdef COMPAT_FREEBSD32
1860 switch (cmd) {
1861 case BIOCSETF32:
1862 case BIOCSETWF32:
1863 case BIOCSETFNR32:
1864 fp32 = (struct bpf_program32 *)fp;
1865 fp_swab.bf_len = fp32->bf_len;
1866 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1867 fp = &fp_swab;
1868 switch (cmd) {
1869 case BIOCSETF32:
1870 cmd = BIOCSETF;
1871 break;
1872 case BIOCSETWF32:
1873 cmd = BIOCSETWF;
1874 break;
1875 }
1876 break;
1877 }
1878 #endif
1879
1880 fcode = NULL;
1881 #ifdef BPF_JITTER
1882 jfunc = ofunc = NULL;
1883 #endif
1884 need_upgrade = 0;
1885
1886 /*
1887 * Check new filter validness before acquiring any locks.
1888 * Allocate memory for new filter, if needed.
1889 */
1890 flen = fp->bf_len;
1891 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1892 return (EINVAL);
1893 size = flen * sizeof(*fp->bf_insns);
1894 if (size > 0) {
1895 /* We're setting up new filter. Copy and check actual data. */
1896 fcode = malloc(size, M_BPF, M_WAITOK);
1897 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1898 !bpf_validate(fcode, flen)) {
1899 free(fcode, M_BPF);
1900 return (EINVAL);
1901 }
1902 #ifdef BPF_JITTER
1903 if (cmd != BIOCSETWF) {
1904 /*
1905 * Filter is copied inside fcode and is
1906 * perfectly valid.
1907 */
1908 jfunc = bpf_jitter(fcode, flen);
1909 }
1910 #endif
1911 }
1912
1913 BPF_LOCK();
1914
1915 /*
1916 * Set up new filter.
1917 * Protect filter change by interface lock.
1918 * Additionally, we are protected by global lock here.
1919 */
1920 if (d->bd_bif != NULL)
1921 BPFIF_WLOCK(d->bd_bif);
1922 BPFD_LOCK(d);
1923 if (cmd == BIOCSETWF) {
1924 old = d->bd_wfilter;
1925 d->bd_wfilter = fcode;
1926 } else {
1927 old = d->bd_rfilter;
1928 d->bd_rfilter = fcode;
1929 #ifdef BPF_JITTER
1930 ofunc = d->bd_bfilter;
1931 d->bd_bfilter = jfunc;
1932 #endif
1933 if (cmd == BIOCSETF)
1934 reset_d(d);
1935
1936 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen);
1937 }
1938 BPFD_UNLOCK(d);
1939 if (d->bd_bif != NULL)
1940 BPFIF_WUNLOCK(d->bd_bif);
1941 if (old != NULL)
1942 free(old, M_BPF);
1943 #ifdef BPF_JITTER
1944 if (ofunc != NULL)
1945 bpf_destroy_jit_filter(ofunc);
1946 #endif
1947
1948 /* Move d to active readers list. */
1949 if (need_upgrade != 0)
1950 bpf_upgraded(d);
1951
1952 BPF_UNLOCK();
1953 return (0);
1954 }
1955
1956 /*
1957 * Detach a file from its current interface (if attached at all) and attach
1958 * to the interface indicated by the name stored in ifr.
1959 * Return an errno or 0.
1960 */
1961 static int
1962 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1963 {
1964 struct bpf_if *bp;
1965 struct ifnet *theywant;
1966
1967 BPF_LOCK_ASSERT();
1968
1969 theywant = ifunit(ifr->ifr_name);
1970 if (theywant == NULL || theywant->if_bpf == NULL)
1971 return (ENXIO);
1972
1973 bp = theywant->if_bpf;
1974
1975 /* Check if interface is not being detached from BPF */
1976 BPFIF_RLOCK(bp);
1977 if (bp->bif_flags & BPFIF_FLAG_DYING) {
1978 BPFIF_RUNLOCK(bp);
1979 return (ENXIO);
1980 }
1981 BPFIF_RUNLOCK(bp);
1982
1983 /*
1984 * At this point, we expect the buffer is already allocated. If not,
1985 * return an error.
1986 */
1987 switch (d->bd_bufmode) {
1988 case BPF_BUFMODE_BUFFER:
1989 case BPF_BUFMODE_ZBUF:
1990 if (d->bd_sbuf == NULL)
1991 return (EINVAL);
1992 break;
1993
1994 default:
1995 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1996 }
1997 if (bp != d->bd_bif)
1998 bpf_attachd(d, bp);
1999 BPFD_LOCK(d);
2000 reset_d(d);
2001 BPFD_UNLOCK(d);
2002 return (0);
2003 }
2004
2005 /*
2006 * Support for select() and poll() system calls
2007 *
2008 * Return true iff the specific operation will not block indefinitely.
2009 * Otherwise, return false but make a note that a selwakeup() must be done.
2010 */
2011 static int
2012 bpfpoll(struct cdev *dev, int events, struct thread *td)
2013 {
2014 struct bpf_d *d;
2015 int revents;
2016
2017 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2018 return (events &
2019 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
2020
2021 /*
2022 * Refresh PID associated with this descriptor.
2023 */
2024 revents = events & (POLLOUT | POLLWRNORM);
2025 BPFD_LOCK(d);
2026 BPF_PID_REFRESH(d, td);
2027 if (events & (POLLIN | POLLRDNORM)) {
2028 if (bpf_ready(d))
2029 revents |= events & (POLLIN | POLLRDNORM);
2030 else {
2031 selrecord(td, &d->bd_sel);
2032 /* Start the read timeout if necessary. */
2033 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2034 callout_reset(&d->bd_callout, d->bd_rtout,
2035 bpf_timed_out, d);
2036 d->bd_state = BPF_WAITING;
2037 }
2038 }
2039 }
2040 BPFD_UNLOCK(d);
2041 return (revents);
2042 }
2043
2044 /*
2045 * Support for kevent() system call. Register EVFILT_READ filters and
2046 * reject all others.
2047 */
2048 int
2049 bpfkqfilter(struct cdev *dev, struct knote *kn)
2050 {
2051 struct bpf_d *d;
2052
2053 if (devfs_get_cdevpriv((void **)&d) != 0 ||
2054 kn->kn_filter != EVFILT_READ)
2055 return (1);
2056
2057 /*
2058 * Refresh PID associated with this descriptor.
2059 */
2060 BPFD_LOCK(d);
2061 BPF_PID_REFRESH_CUR(d);
2062 kn->kn_fop = &bpfread_filtops;
2063 kn->kn_hook = d;
2064 knlist_add(&d->bd_sel.si_note, kn, 1);
2065 BPFD_UNLOCK(d);
2066
2067 return (0);
2068 }
2069
2070 static void
2071 filt_bpfdetach(struct knote *kn)
2072 {
2073 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2074
2075 knlist_remove(&d->bd_sel.si_note, kn, 0);
2076 }
2077
2078 static int
2079 filt_bpfread(struct knote *kn, long hint)
2080 {
2081 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2082 int ready;
2083
2084 BPFD_LOCK_ASSERT(d);
2085 ready = bpf_ready(d);
2086 if (ready) {
2087 kn->kn_data = d->bd_slen;
2088 /*
2089 * Ignore the hold buffer if it is being copied to user space.
2090 */
2091 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2092 kn->kn_data += d->bd_hlen;
2093 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2094 callout_reset(&d->bd_callout, d->bd_rtout,
2095 bpf_timed_out, d);
2096 d->bd_state = BPF_WAITING;
2097 }
2098
2099 return (ready);
2100 }
2101
2102 #define BPF_TSTAMP_NONE 0
2103 #define BPF_TSTAMP_FAST 1
2104 #define BPF_TSTAMP_NORMAL 2
2105 #define BPF_TSTAMP_EXTERN 3
2106
2107 static int
2108 bpf_ts_quality(int tstype)
2109 {
2110
2111 if (tstype == BPF_T_NONE)
2112 return (BPF_TSTAMP_NONE);
2113 if ((tstype & BPF_T_FAST) != 0)
2114 return (BPF_TSTAMP_FAST);
2115
2116 return (BPF_TSTAMP_NORMAL);
2117 }
2118
2119 static int
2120 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2121 {
2122 struct m_tag *tag;
2123 int quality;
2124
2125 quality = bpf_ts_quality(tstype);
2126 if (quality == BPF_TSTAMP_NONE)
2127 return (quality);
2128
2129 if (m != NULL) {
2130 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2131 if (tag != NULL) {
2132 *bt = *(struct bintime *)(tag + 1);
2133 return (BPF_TSTAMP_EXTERN);
2134 }
2135 }
2136 if (quality == BPF_TSTAMP_NORMAL)
2137 binuptime(bt);
2138 else
2139 getbinuptime(bt);
2140
2141 return (quality);
2142 }
2143
2144 /*
2145 * Incoming linkage from device drivers. Process the packet pkt, of length
2146 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2147 * by each process' filter, and if accepted, stashed into the corresponding
2148 * buffer.
2149 */
2150 void
2151 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2152 {
2153 struct bintime bt;
2154 struct bpf_d *d;
2155 #ifdef BPF_JITTER
2156 bpf_jit_filter *bf;
2157 #endif
2158 u_int slen;
2159 int gottime;
2160
2161 gottime = BPF_TSTAMP_NONE;
2162
2163 BPFIF_RLOCK(bp);
2164
2165 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2166 /*
2167 * We are not using any locks for d here because:
2168 * 1) any filter change is protected by interface
2169 * write lock
2170 * 2) destroying/detaching d is protected by interface
2171 * write lock, too
2172 */
2173
2174 counter_u64_add(d->bd_rcount, 1);
2175 /*
2176 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2177 * way for the caller to indiciate to us whether this packet
2178 * is inbound or outbound. In the bpf_mtap() routines, we use
2179 * the interface pointers on the mbuf to figure it out.
2180 */
2181 #ifdef BPF_JITTER
2182 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2183 if (bf != NULL)
2184 slen = (*(bf->func))(pkt, pktlen, pktlen);
2185 else
2186 #endif
2187 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2188 if (slen != 0) {
2189 /*
2190 * Filter matches. Let's to acquire write lock.
2191 */
2192 BPFD_LOCK(d);
2193
2194 counter_u64_add(d->bd_fcount, 1);
2195 if (gottime < bpf_ts_quality(d->bd_tstamp))
2196 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2197 #ifdef MAC
2198 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2199 #endif
2200 catchpacket(d, pkt, pktlen, slen,
2201 bpf_append_bytes, &bt);
2202 BPFD_UNLOCK(d);
2203 }
2204 }
2205 BPFIF_RUNLOCK(bp);
2206 }
2207
2208 #define BPF_CHECK_DIRECTION(d, r, i) \
2209 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2210 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2211
2212 /*
2213 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2214 * Locking model is explained in bpf_tap().
2215 */
2216 void
2217 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2218 {
2219 struct bintime bt;
2220 struct bpf_d *d;
2221 #ifdef BPF_JITTER
2222 bpf_jit_filter *bf;
2223 #endif
2224 u_int pktlen, slen;
2225 int gottime;
2226
2227 /* Skip outgoing duplicate packets. */
2228 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2229 m->m_flags &= ~M_PROMISC;
2230 return;
2231 }
2232
2233 pktlen = m_length(m, NULL);
2234 gottime = BPF_TSTAMP_NONE;
2235
2236 BPFIF_RLOCK(bp);
2237
2238 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2239 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2240 continue;
2241 counter_u64_add(d->bd_rcount, 1);
2242 #ifdef BPF_JITTER
2243 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2244 /* XXX We cannot handle multiple mbufs. */
2245 if (bf != NULL && m->m_next == NULL)
2246 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2247 else
2248 #endif
2249 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2250 if (slen != 0) {
2251 BPFD_LOCK(d);
2252
2253 counter_u64_add(d->bd_fcount, 1);
2254 if (gottime < bpf_ts_quality(d->bd_tstamp))
2255 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2256 #ifdef MAC
2257 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2258 #endif
2259 catchpacket(d, (u_char *)m, pktlen, slen,
2260 bpf_append_mbuf, &bt);
2261 BPFD_UNLOCK(d);
2262 }
2263 }
2264 BPFIF_RUNLOCK(bp);
2265 }
2266
2267 /*
2268 * Incoming linkage from device drivers, when packet is in
2269 * an mbuf chain and to be prepended by a contiguous header.
2270 */
2271 void
2272 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2273 {
2274 struct bintime bt;
2275 struct mbuf mb;
2276 struct bpf_d *d;
2277 u_int pktlen, slen;
2278 int gottime;
2279
2280 /* Skip outgoing duplicate packets. */
2281 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2282 m->m_flags &= ~M_PROMISC;
2283 return;
2284 }
2285
2286 pktlen = m_length(m, NULL);
2287 /*
2288 * Craft on-stack mbuf suitable for passing to bpf_filter.
2289 * Note that we cut corners here; we only setup what's
2290 * absolutely needed--this mbuf should never go anywhere else.
2291 */
2292 mb.m_next = m;
2293 mb.m_data = data;
2294 mb.m_len = dlen;
2295 pktlen += dlen;
2296
2297 gottime = BPF_TSTAMP_NONE;
2298
2299 BPFIF_RLOCK(bp);
2300
2301 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2302 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2303 continue;
2304 counter_u64_add(d->bd_rcount, 1);
2305 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2306 if (slen != 0) {
2307 BPFD_LOCK(d);
2308
2309 counter_u64_add(d->bd_fcount, 1);
2310 if (gottime < bpf_ts_quality(d->bd_tstamp))
2311 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2312 #ifdef MAC
2313 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2314 #endif
2315 catchpacket(d, (u_char *)&mb, pktlen, slen,
2316 bpf_append_mbuf, &bt);
2317 BPFD_UNLOCK(d);
2318 }
2319 }
2320 BPFIF_RUNLOCK(bp);
2321 }
2322
2323 #undef BPF_CHECK_DIRECTION
2324
2325 #undef BPF_TSTAMP_NONE
2326 #undef BPF_TSTAMP_FAST
2327 #undef BPF_TSTAMP_NORMAL
2328 #undef BPF_TSTAMP_EXTERN
2329
2330 static int
2331 bpf_hdrlen(struct bpf_d *d)
2332 {
2333 int hdrlen;
2334
2335 hdrlen = d->bd_bif->bif_hdrlen;
2336 #ifndef BURN_BRIDGES
2337 if (d->bd_tstamp == BPF_T_NONE ||
2338 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2339 #ifdef COMPAT_FREEBSD32
2340 if (d->bd_compat32)
2341 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2342 else
2343 #endif
2344 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2345 else
2346 #endif
2347 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2348 #ifdef COMPAT_FREEBSD32
2349 if (d->bd_compat32)
2350 hdrlen = BPF_WORDALIGN32(hdrlen);
2351 else
2352 #endif
2353 hdrlen = BPF_WORDALIGN(hdrlen);
2354
2355 return (hdrlen - d->bd_bif->bif_hdrlen);
2356 }
2357
2358 static void
2359 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2360 {
2361 struct bintime bt2, boottimebin;
2362 struct timeval tsm;
2363 struct timespec tsn;
2364
2365 if ((tstype & BPF_T_MONOTONIC) == 0) {
2366 bt2 = *bt;
2367 getboottimebin(&boottimebin);
2368 bintime_add(&bt2, &boottimebin);
2369 bt = &bt2;
2370 }
2371 switch (BPF_T_FORMAT(tstype)) {
2372 case BPF_T_MICROTIME:
2373 bintime2timeval(bt, &tsm);
2374 ts->bt_sec = tsm.tv_sec;
2375 ts->bt_frac = tsm.tv_usec;
2376 break;
2377 case BPF_T_NANOTIME:
2378 bintime2timespec(bt, &tsn);
2379 ts->bt_sec = tsn.tv_sec;
2380 ts->bt_frac = tsn.tv_nsec;
2381 break;
2382 case BPF_T_BINTIME:
2383 ts->bt_sec = bt->sec;
2384 ts->bt_frac = bt->frac;
2385 break;
2386 }
2387 }
2388
2389 /*
2390 * Move the packet data from interface memory (pkt) into the
2391 * store buffer. "cpfn" is the routine called to do the actual data
2392 * transfer. bcopy is passed in to copy contiguous chunks, while
2393 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2394 * pkt is really an mbuf.
2395 */
2396 static void
2397 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2398 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2399 struct bintime *bt)
2400 {
2401 struct bpf_xhdr hdr;
2402 #ifndef BURN_BRIDGES
2403 struct bpf_hdr hdr_old;
2404 #ifdef COMPAT_FREEBSD32
2405 struct bpf_hdr32 hdr32_old;
2406 #endif
2407 #endif
2408 int caplen, curlen, hdrlen, totlen;
2409 int do_wakeup = 0;
2410 int do_timestamp;
2411 int tstype;
2412
2413 BPFD_LOCK_ASSERT(d);
2414
2415 /*
2416 * Detect whether user space has released a buffer back to us, and if
2417 * so, move it from being a hold buffer to a free buffer. This may
2418 * not be the best place to do it (for example, we might only want to
2419 * run this check if we need the space), but for now it's a reliable
2420 * spot to do it.
2421 */
2422 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2423 d->bd_fbuf = d->bd_hbuf;
2424 d->bd_hbuf = NULL;
2425 d->bd_hlen = 0;
2426 bpf_buf_reclaimed(d);
2427 }
2428
2429 /*
2430 * Figure out how many bytes to move. If the packet is
2431 * greater or equal to the snapshot length, transfer that
2432 * much. Otherwise, transfer the whole packet (unless
2433 * we hit the buffer size limit).
2434 */
2435 hdrlen = bpf_hdrlen(d);
2436 totlen = hdrlen + min(snaplen, pktlen);
2437 if (totlen > d->bd_bufsize)
2438 totlen = d->bd_bufsize;
2439
2440 /*
2441 * Round up the end of the previous packet to the next longword.
2442 *
2443 * Drop the packet if there's no room and no hope of room
2444 * If the packet would overflow the storage buffer or the storage
2445 * buffer is considered immutable by the buffer model, try to rotate
2446 * the buffer and wakeup pending processes.
2447 */
2448 #ifdef COMPAT_FREEBSD32
2449 if (d->bd_compat32)
2450 curlen = BPF_WORDALIGN32(d->bd_slen);
2451 else
2452 #endif
2453 curlen = BPF_WORDALIGN(d->bd_slen);
2454 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2455 if (d->bd_fbuf == NULL) {
2456 /*
2457 * There's no room in the store buffer, and no
2458 * prospect of room, so drop the packet. Notify the
2459 * buffer model.
2460 */
2461 bpf_buffull(d);
2462 counter_u64_add(d->bd_dcount, 1);
2463 return;
2464 }
2465 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2466 ROTATE_BUFFERS(d);
2467 do_wakeup = 1;
2468 curlen = 0;
2469 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2470 /*
2471 * Immediate mode is set, or the read timeout has already
2472 * expired during a select call. A packet arrived, so the
2473 * reader should be woken up.
2474 */
2475 do_wakeup = 1;
2476 caplen = totlen - hdrlen;
2477 tstype = d->bd_tstamp;
2478 do_timestamp = tstype != BPF_T_NONE;
2479 #ifndef BURN_BRIDGES
2480 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2481 struct bpf_ts ts;
2482 if (do_timestamp)
2483 bpf_bintime2ts(bt, &ts, tstype);
2484 #ifdef COMPAT_FREEBSD32
2485 if (d->bd_compat32) {
2486 bzero(&hdr32_old, sizeof(hdr32_old));
2487 if (do_timestamp) {
2488 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2489 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2490 }
2491 hdr32_old.bh_datalen = pktlen;
2492 hdr32_old.bh_hdrlen = hdrlen;
2493 hdr32_old.bh_caplen = caplen;
2494 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2495 sizeof(hdr32_old));
2496 goto copy;
2497 }
2498 #endif
2499 bzero(&hdr_old, sizeof(hdr_old));
2500 if (do_timestamp) {
2501 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2502 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2503 }
2504 hdr_old.bh_datalen = pktlen;
2505 hdr_old.bh_hdrlen = hdrlen;
2506 hdr_old.bh_caplen = caplen;
2507 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2508 sizeof(hdr_old));
2509 goto copy;
2510 }
2511 #endif
2512
2513 /*
2514 * Append the bpf header. Note we append the actual header size, but
2515 * move forward the length of the header plus padding.
2516 */
2517 bzero(&hdr, sizeof(hdr));
2518 if (do_timestamp)
2519 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2520 hdr.bh_datalen = pktlen;
2521 hdr.bh_hdrlen = hdrlen;
2522 hdr.bh_caplen = caplen;
2523 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2524
2525 /*
2526 * Copy the packet data into the store buffer and update its length.
2527 */
2528 #ifndef BURN_BRIDGES
2529 copy:
2530 #endif
2531 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2532 d->bd_slen = curlen + totlen;
2533
2534 if (do_wakeup)
2535 bpf_wakeup(d);
2536 }
2537
2538 /*
2539 * Free buffers currently in use by a descriptor.
2540 * Called on close.
2541 */
2542 static void
2543 bpf_freed(struct bpf_d *d)
2544 {
2545
2546 /*
2547 * We don't need to lock out interrupts since this descriptor has
2548 * been detached from its interface and it yet hasn't been marked
2549 * free.
2550 */
2551 bpf_free(d);
2552 if (d->bd_rfilter != NULL) {
2553 free((caddr_t)d->bd_rfilter, M_BPF);
2554 #ifdef BPF_JITTER
2555 if (d->bd_bfilter != NULL)
2556 bpf_destroy_jit_filter(d->bd_bfilter);
2557 #endif
2558 }
2559 if (d->bd_wfilter != NULL)
2560 free((caddr_t)d->bd_wfilter, M_BPF);
2561 mtx_destroy(&d->bd_lock);
2562
2563 counter_u64_free(d->bd_rcount);
2564 counter_u64_free(d->bd_dcount);
2565 counter_u64_free(d->bd_fcount);
2566 counter_u64_free(d->bd_wcount);
2567 counter_u64_free(d->bd_wfcount);
2568 counter_u64_free(d->bd_wdcount);
2569 counter_u64_free(d->bd_zcopy);
2570
2571 }
2572
2573 /*
2574 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2575 * fixed size of the link header (variable length headers not yet supported).
2576 */
2577 void
2578 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2579 {
2580
2581 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2582 }
2583
2584 /*
2585 * Attach an interface to bpf. ifp is a pointer to the structure
2586 * defining the interface to be attached, dlt is the link layer type,
2587 * and hdrlen is the fixed size of the link header (variable length
2588 * headers are not yet supporrted).
2589 */
2590 void
2591 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2592 {
2593 struct bpf_if *bp;
2594
2595 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2596 if (bp == NULL)
2597 panic("bpfattach");
2598
2599 LIST_INIT(&bp->bif_dlist);
2600 LIST_INIT(&bp->bif_wlist);
2601 bp->bif_ifp = ifp;
2602 bp->bif_dlt = dlt;
2603 rw_init(&bp->bif_lock, "bpf interface lock");
2604 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2605 bp->bif_bpf = driverp;
2606 *driverp = bp;
2607
2608 BPF_LOCK();
2609 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2610 BPF_UNLOCK();
2611
2612 bp->bif_hdrlen = hdrlen;
2613
2614 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2615 if_printf(ifp, "bpf attached\n");
2616 }
2617
2618 #ifdef VIMAGE
2619 /*
2620 * When moving interfaces between vnet instances we need a way to
2621 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2622 * after the vmove. We unfortunately have no device driver infrastructure
2623 * to query the interface for these values after creation/attach, thus
2624 * add this as a workaround.
2625 */
2626 int
2627 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2628 {
2629
2630 if (bp == NULL)
2631 return (ENXIO);
2632 if (bif_dlt == NULL && bif_hdrlen == NULL)
2633 return (0);
2634
2635 if (bif_dlt != NULL)
2636 *bif_dlt = bp->bif_dlt;
2637 if (bif_hdrlen != NULL)
2638 *bif_hdrlen = bp->bif_hdrlen;
2639
2640 return (0);
2641 }
2642 #endif
2643
2644 /*
2645 * Detach bpf from an interface. This involves detaching each descriptor
2646 * associated with the interface. Notify each descriptor as it's detached
2647 * so that any sleepers wake up and get ENXIO.
2648 */
2649 void
2650 bpfdetach(struct ifnet *ifp)
2651 {
2652 struct bpf_if *bp, *bp_temp;
2653 struct bpf_d *d;
2654 int ndetached;
2655
2656 ndetached = 0;
2657
2658 BPF_LOCK();
2659 /* Find all bpf_if struct's which reference ifp and detach them. */
2660 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2661 if (ifp != bp->bif_ifp)
2662 continue;
2663
2664 LIST_REMOVE(bp, bif_next);
2665 /* Add to to-be-freed list */
2666 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next);
2667
2668 ndetached++;
2669 /*
2670 * Delay freeing bp till interface is detached
2671 * and all routes through this interface are removed.
2672 * Mark bp as detached to restrict new consumers.
2673 */
2674 BPFIF_WLOCK(bp);
2675 bp->bif_flags |= BPFIF_FLAG_DYING;
2676 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
2677 BPFIF_WUNLOCK(bp);
2678
2679 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p",
2680 __func__, bp->bif_dlt, bp, ifp);
2681
2682 /* Free common descriptors */
2683 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2684 bpf_detachd_locked(d);
2685 BPFD_LOCK(d);
2686 bpf_wakeup(d);
2687 BPFD_UNLOCK(d);
2688 }
2689
2690 /* Free writer-only descriptors */
2691 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2692 bpf_detachd_locked(d);
2693 BPFD_LOCK(d);
2694 bpf_wakeup(d);
2695 BPFD_UNLOCK(d);
2696 }
2697 }
2698 BPF_UNLOCK();
2699
2700 #ifdef INVARIANTS
2701 if (ndetached == 0)
2702 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2703 #endif
2704 }
2705
2706 /*
2707 * Interface departure handler.
2708 * Note departure event does not guarantee interface is going down.
2709 * Interface renaming is currently done via departure/arrival event set.
2710 *
2711 * Departure handled is called after all routes pointing to
2712 * given interface are removed and interface is in down state
2713 * restricting any packets to be sent/received. We assume it is now safe
2714 * to free data allocated by BPF.
2715 */
2716 static void
2717 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2718 {
2719 struct bpf_if *bp, *bp_temp;
2720 int nmatched = 0;
2721
2722 /* Ignore ifnet renaming. */
2723 if (ifp->if_flags & IFF_RENAMING)
2724 return;
2725
2726 BPF_LOCK();
2727 /*
2728 * Find matching entries in free list.
2729 * Nothing should be found if bpfdetach() was not called.
2730 */
2731 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) {
2732 if (ifp != bp->bif_ifp)
2733 continue;
2734
2735 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p",
2736 __func__, bp, ifp);
2737
2738 LIST_REMOVE(bp, bif_next);
2739
2740 rw_destroy(&bp->bif_lock);
2741 free(bp, M_BPF);
2742
2743 nmatched++;
2744 }
2745 BPF_UNLOCK();
2746 }
2747
2748 /*
2749 * Get a list of available data link type of the interface.
2750 */
2751 static int
2752 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2753 {
2754 struct ifnet *ifp;
2755 struct bpf_if *bp;
2756 u_int *lst;
2757 int error, n, n1;
2758
2759 BPF_LOCK_ASSERT();
2760
2761 ifp = d->bd_bif->bif_ifp;
2762 again:
2763 n1 = 0;
2764 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2765 if (bp->bif_ifp == ifp)
2766 n1++;
2767 }
2768 if (bfl->bfl_list == NULL) {
2769 bfl->bfl_len = n1;
2770 return (0);
2771 }
2772 if (n1 > bfl->bfl_len)
2773 return (ENOMEM);
2774 BPF_UNLOCK();
2775 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2776 n = 0;
2777 BPF_LOCK();
2778 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2779 if (bp->bif_ifp != ifp)
2780 continue;
2781 if (n >= n1) {
2782 free(lst, M_TEMP);
2783 goto again;
2784 }
2785 lst[n] = bp->bif_dlt;
2786 n++;
2787 }
2788 BPF_UNLOCK();
2789 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2790 free(lst, M_TEMP);
2791 BPF_LOCK();
2792 bfl->bfl_len = n;
2793 return (error);
2794 }
2795
2796 /*
2797 * Set the data link type of a BPF instance.
2798 */
2799 static int
2800 bpf_setdlt(struct bpf_d *d, u_int dlt)
2801 {
2802 int error, opromisc;
2803 struct ifnet *ifp;
2804 struct bpf_if *bp;
2805
2806 BPF_LOCK_ASSERT();
2807
2808 if (d->bd_bif->bif_dlt == dlt)
2809 return (0);
2810 ifp = d->bd_bif->bif_ifp;
2811
2812 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2813 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2814 break;
2815 }
2816
2817 if (bp != NULL) {
2818 opromisc = d->bd_promisc;
2819 bpf_attachd(d, bp);
2820 BPFD_LOCK(d);
2821 reset_d(d);
2822 BPFD_UNLOCK(d);
2823 if (opromisc) {
2824 error = ifpromisc(bp->bif_ifp, 1);
2825 if (error)
2826 if_printf(bp->bif_ifp,
2827 "bpf_setdlt: ifpromisc failed (%d)\n",
2828 error);
2829 else
2830 d->bd_promisc = 1;
2831 }
2832 }
2833 return (bp == NULL ? EINVAL : 0);
2834 }
2835
2836 static void
2837 bpf_drvinit(void *unused)
2838 {
2839 struct cdev *dev;
2840
2841 sx_init(&bpf_sx, "bpf global lock");
2842 LIST_INIT(&bpf_iflist);
2843 LIST_INIT(&bpf_freelist);
2844
2845 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2846 /* For compatibility */
2847 make_dev_alias(dev, "bpf0");
2848
2849 /* Register interface departure handler */
2850 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2851 ifnet_departure_event, bpf_ifdetach, NULL,
2852 EVENTHANDLER_PRI_ANY);
2853 }
2854
2855 /*
2856 * Zero out the various packet counters associated with all of the bpf
2857 * descriptors. At some point, we will probably want to get a bit more
2858 * granular and allow the user to specify descriptors to be zeroed.
2859 */
2860 static void
2861 bpf_zero_counters(void)
2862 {
2863 struct bpf_if *bp;
2864 struct bpf_d *bd;
2865
2866 BPF_LOCK();
2867 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2868 BPFIF_RLOCK(bp);
2869 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2870 BPFD_LOCK(bd);
2871 counter_u64_zero(bd->bd_rcount);
2872 counter_u64_zero(bd->bd_dcount);
2873 counter_u64_zero(bd->bd_fcount);
2874 counter_u64_zero(bd->bd_wcount);
2875 counter_u64_zero(bd->bd_wfcount);
2876 counter_u64_zero(bd->bd_zcopy);
2877 BPFD_UNLOCK(bd);
2878 }
2879 BPFIF_RUNLOCK(bp);
2880 }
2881 BPF_UNLOCK();
2882 }
2883
2884 /*
2885 * Fill filter statistics
2886 */
2887 static void
2888 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2889 {
2890
2891 bzero(d, sizeof(*d));
2892 BPFD_LOCK_ASSERT(bd);
2893 d->bd_structsize = sizeof(*d);
2894 /* XXX: reading should be protected by global lock */
2895 d->bd_immediate = bd->bd_immediate;
2896 d->bd_promisc = bd->bd_promisc;
2897 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2898 d->bd_direction = bd->bd_direction;
2899 d->bd_feedback = bd->bd_feedback;
2900 d->bd_async = bd->bd_async;
2901 d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
2902 d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
2903 d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
2904 d->bd_sig = bd->bd_sig;
2905 d->bd_slen = bd->bd_slen;
2906 d->bd_hlen = bd->bd_hlen;
2907 d->bd_bufsize = bd->bd_bufsize;
2908 d->bd_pid = bd->bd_pid;
2909 strlcpy(d->bd_ifname,
2910 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2911 d->bd_locked = bd->bd_locked;
2912 d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
2913 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
2914 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
2915 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
2916 d->bd_bufmode = bd->bd_bufmode;
2917 }
2918
2919 /*
2920 * Handle `netstat -B' stats request
2921 */
2922 static int
2923 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2924 {
2925 static const struct xbpf_d zerostats;
2926 struct xbpf_d *xbdbuf, *xbd, tempstats;
2927 int index, error;
2928 struct bpf_if *bp;
2929 struct bpf_d *bd;
2930
2931 /*
2932 * XXX This is not technically correct. It is possible for non
2933 * privileged users to open bpf devices. It would make sense
2934 * if the users who opened the devices were able to retrieve
2935 * the statistics for them, too.
2936 */
2937 error = priv_check(req->td, PRIV_NET_BPF);
2938 if (error)
2939 return (error);
2940 /*
2941 * Check to see if the user is requesting that the counters be
2942 * zeroed out. Explicitly check that the supplied data is zeroed,
2943 * as we aren't allowing the user to set the counters currently.
2944 */
2945 if (req->newptr != NULL) {
2946 if (req->newlen != sizeof(tempstats))
2947 return (EINVAL);
2948 memset(&tempstats, 0, sizeof(tempstats));
2949 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2950 if (error)
2951 return (error);
2952 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2953 return (EINVAL);
2954 bpf_zero_counters();
2955 return (0);
2956 }
2957 if (req->oldptr == NULL)
2958 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2959 if (bpf_bpfd_cnt == 0)
2960 return (SYSCTL_OUT(req, 0, 0));
2961 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2962 BPF_LOCK();
2963 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2964 BPF_UNLOCK();
2965 free(xbdbuf, M_BPF);
2966 return (ENOMEM);
2967 }
2968 index = 0;
2969 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2970 BPFIF_RLOCK(bp);
2971 /* Send writers-only first */
2972 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2973 xbd = &xbdbuf[index++];
2974 BPFD_LOCK(bd);
2975 bpfstats_fill_xbpf(xbd, bd);
2976 BPFD_UNLOCK(bd);
2977 }
2978 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2979 xbd = &xbdbuf[index++];
2980 BPFD_LOCK(bd);
2981 bpfstats_fill_xbpf(xbd, bd);
2982 BPFD_UNLOCK(bd);
2983 }
2984 BPFIF_RUNLOCK(bp);
2985 }
2986 BPF_UNLOCK();
2987 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2988 free(xbdbuf, M_BPF);
2989 return (error);
2990 }
2991
2992 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2993
2994 #else /* !DEV_BPF && !NETGRAPH_BPF */
2995
2996 /*
2997 * NOP stubs to allow bpf-using drivers to load and function.
2998 *
2999 * A 'better' implementation would allow the core bpf functionality
3000 * to be loaded at runtime.
3001 */
3002
3003 void
3004 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3005 {
3006 }
3007
3008 void
3009 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3010 {
3011 }
3012
3013 void
3014 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3015 {
3016 }
3017
3018 void
3019 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3020 {
3021
3022 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3023 }
3024
3025 void
3026 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3027 {
3028
3029 *driverp = (struct bpf_if *)&dead_bpf_if;
3030 }
3031
3032 void
3033 bpfdetach(struct ifnet *ifp)
3034 {
3035 }
3036
3037 u_int
3038 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3039 {
3040 return -1; /* "no filter" behaviour */
3041 }
3042
3043 int
3044 bpf_validate(const struct bpf_insn *f, int len)
3045 {
3046 return 0; /* false */
3047 }
3048
3049 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3050
3051 #ifdef DDB
3052 static void
3053 bpf_show_bpf_if(struct bpf_if *bpf_if)
3054 {
3055
3056 if (bpf_if == NULL)
3057 return;
3058 db_printf("%p:\n", bpf_if);
3059 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3060 /* bif_ext.bif_next */
3061 /* bif_ext.bif_dlist */
3062 BPF_DB_PRINTF("%#x", bif_dlt);
3063 BPF_DB_PRINTF("%u", bif_hdrlen);
3064 BPF_DB_PRINTF("%p", bif_ifp);
3065 /* bif_lock */
3066 /* bif_wlist */
3067 BPF_DB_PRINTF("%#x", bif_flags);
3068 }
3069
3070 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3071 {
3072
3073 if (!have_addr) {
3074 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3075 return;
3076 }
3077
3078 bpf_show_bpf_if((struct bpf_if *)addr);
3079 }
3080 #endif
Cache object: 7f6df27b228cf785cc21800c7ba02cca
|