1 /*-
2 * Copyright (c) 1990, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from the Stanford/CMU enet packet filter,
6 * (net/enet.c) distributed as part of 4.3BSD, and code contributed
7 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
8 * Berkeley Laboratory.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)bpf.c 8.4 (Berkeley) 1/9/95
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_bpf.h"
41 #include "opt_compat.h"
42 #include "opt_netgraph.h"
43
44 #include <sys/types.h>
45 #include <sys/param.h>
46 #include <sys/lock.h>
47 #include <sys/rwlock.h>
48 #include <sys/systm.h>
49 #include <sys/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/jail.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/time.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/signalvar.h>
58 #include <sys/filio.h>
59 #include <sys/sockio.h>
60 #include <sys/ttycom.h>
61 #include <sys/uio.h>
62
63 #include <sys/event.h>
64 #include <sys/file.h>
65 #include <sys/poll.h>
66 #include <sys/proc.h>
67
68 #include <sys/socket.h>
69
70 #include <net/if.h>
71 #define BPF_INTERNAL
72 #include <net/bpf.h>
73 #include <net/bpf_buffer.h>
74 #ifdef BPF_JITTER
75 #include <net/bpf_jitter.h>
76 #endif
77 #include <net/bpf_zerocopy.h>
78 #include <net/bpfdesc.h>
79 #include <net/vnet.h>
80
81 #include <netinet/in.h>
82 #include <netinet/if_ether.h>
83 #include <sys/kernel.h>
84 #include <sys/sysctl.h>
85
86 #include <net80211/ieee80211_freebsd.h>
87
88 #include <security/mac/mac_framework.h>
89
90 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
91
92 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
93
94 #define PRINET 26 /* interruptible */
95
96 #define SIZEOF_BPF_HDR(type) \
97 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
98
99 #ifdef COMPAT_FREEBSD32
100 #include <sys/mount.h>
101 #include <compat/freebsd32/freebsd32.h>
102 #define BPF_ALIGNMENT32 sizeof(int32_t)
103 #define BPF_WORDALIGN32(x) (((x)+(BPF_ALIGNMENT32-1))&~(BPF_ALIGNMENT32-1))
104
105 #ifndef BURN_BRIDGES
106 /*
107 * 32-bit version of structure prepended to each packet. We use this header
108 * instead of the standard one for 32-bit streams. We mark the a stream as
109 * 32-bit the first time we see a 32-bit compat ioctl request.
110 */
111 struct bpf_hdr32 {
112 struct timeval32 bh_tstamp; /* time stamp */
113 uint32_t bh_caplen; /* length of captured portion */
114 uint32_t bh_datalen; /* original length of packet */
115 uint16_t bh_hdrlen; /* length of bpf header (this struct
116 plus alignment padding) */
117 };
118 #endif
119
120 struct bpf_program32 {
121 u_int bf_len;
122 uint32_t bf_insns;
123 };
124
125 struct bpf_dltlist32 {
126 u_int bfl_len;
127 u_int bfl_list;
128 };
129
130 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
131 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
132 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
133 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
134 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
135 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
136 #endif
137
138 /*
139 * bpf_iflist is a list of BPF interface structures, each corresponding to a
140 * specific DLT. The same network interface might have several BPF interface
141 * structures registered by different layers in the stack (i.e., 802.11
142 * frames, ethernet frames, etc).
143 */
144 static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist;
145 static struct mtx bpf_mtx; /* bpf global lock */
146 static int bpf_bpfd_cnt;
147
148 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
149 static void bpf_detachd(struct bpf_d *);
150 static void bpf_detachd_locked(struct bpf_d *);
151 static void bpf_freed(struct bpf_d *);
152 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
153 struct sockaddr *, int *, struct bpf_insn *);
154 static int bpf_setif(struct bpf_d *, struct ifreq *);
155 static void bpf_timed_out(void *);
156 static __inline void
157 bpf_wakeup(struct bpf_d *);
158 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
159 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
160 struct bintime *);
161 static void reset_d(struct bpf_d *);
162 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
163 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
164 static int bpf_setdlt(struct bpf_d *, u_int);
165 static void filt_bpfdetach(struct knote *);
166 static int filt_bpfread(struct knote *, long);
167 static void bpf_drvinit(void *);
168 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
169
170 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
171 int bpf_maxinsns = BPF_MAXINSNS;
172 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
173 &bpf_maxinsns, 0, "Maximum bpf program instructions");
174 static int bpf_zerocopy_enable = 0;
175 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
176 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
177 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
178 bpf_stats_sysctl, "bpf statistics portal");
179
180 static VNET_DEFINE(int, bpf_optimize_writers) = 0;
181 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
182 SYSCTL_VNET_INT(_net_bpf, OID_AUTO, optimize_writers,
183 CTLFLAG_RW, &VNET_NAME(bpf_optimize_writers), 0,
184 "Do not send packets until BPF program is set");
185
186 static d_open_t bpfopen;
187 static d_read_t bpfread;
188 static d_write_t bpfwrite;
189 static d_ioctl_t bpfioctl;
190 static d_poll_t bpfpoll;
191 static d_kqfilter_t bpfkqfilter;
192
193 static struct cdevsw bpf_cdevsw = {
194 .d_version = D_VERSION,
195 .d_open = bpfopen,
196 .d_read = bpfread,
197 .d_write = bpfwrite,
198 .d_ioctl = bpfioctl,
199 .d_poll = bpfpoll,
200 .d_name = "bpf",
201 .d_kqfilter = bpfkqfilter,
202 };
203
204 static struct filterops bpfread_filtops = {
205 .f_isfd = 1,
206 .f_detach = filt_bpfdetach,
207 .f_event = filt_bpfread,
208 };
209
210 eventhandler_tag bpf_ifdetach_cookie = NULL;
211
212 /*
213 * LOCKING MODEL USED BY BPF:
214 * Locks:
215 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
216 * some global counters and every bpf_if reference.
217 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
218 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
219 * used by bpf_mtap code.
220 *
221 * Lock order:
222 *
223 * Global lock, interface lock, descriptor lock
224 *
225 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
226 * working model. In many places (like bpf_detachd) we start with BPF descriptor
227 * (and we need to at least rlock it to get reliable interface pointer). This
228 * gives us potential LOR. As a result, we use global lock to protect from bpf_if
229 * change in every such place.
230 *
231 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
232 * 3) descriptor main wlock.
233 * Reading bd_bif can be protected by any of these locks, typically global lock.
234 *
235 * Changing read/write BPF filter is protected by the same three locks,
236 * the same applies for reading.
237 *
238 * Sleeping in global lock is not allowed due to bpfdetach() using it.
239 */
240
241 /*
242 * Wrapper functions for various buffering methods. If the set of buffer
243 * modes expands, we will probably want to introduce a switch data structure
244 * similar to protosw, et.
245 */
246 static void
bpf_append_bytes(struct bpf_d * d,caddr_t buf,u_int offset,void * src,u_int len)247 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
248 u_int len)
249 {
250
251 BPFD_LOCK_ASSERT(d);
252
253 switch (d->bd_bufmode) {
254 case BPF_BUFMODE_BUFFER:
255 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
256
257 case BPF_BUFMODE_ZBUF:
258 d->bd_zcopy++;
259 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
260
261 default:
262 panic("bpf_buf_append_bytes");
263 }
264 }
265
266 static void
bpf_append_mbuf(struct bpf_d * d,caddr_t buf,u_int offset,void * src,u_int len)267 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
268 u_int len)
269 {
270
271 BPFD_LOCK_ASSERT(d);
272
273 switch (d->bd_bufmode) {
274 case BPF_BUFMODE_BUFFER:
275 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
276
277 case BPF_BUFMODE_ZBUF:
278 d->bd_zcopy++;
279 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
280
281 default:
282 panic("bpf_buf_append_mbuf");
283 }
284 }
285
286 /*
287 * This function gets called when the free buffer is re-assigned.
288 */
289 static void
bpf_buf_reclaimed(struct bpf_d * d)290 bpf_buf_reclaimed(struct bpf_d *d)
291 {
292
293 BPFD_LOCK_ASSERT(d);
294
295 switch (d->bd_bufmode) {
296 case BPF_BUFMODE_BUFFER:
297 return;
298
299 case BPF_BUFMODE_ZBUF:
300 bpf_zerocopy_buf_reclaimed(d);
301 return;
302
303 default:
304 panic("bpf_buf_reclaimed");
305 }
306 }
307
308 /*
309 * If the buffer mechanism has a way to decide that a held buffer can be made
310 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
311 * returned if the buffer can be discarded, (0) is returned if it cannot.
312 */
313 static int
bpf_canfreebuf(struct bpf_d * d)314 bpf_canfreebuf(struct bpf_d *d)
315 {
316
317 BPFD_LOCK_ASSERT(d);
318
319 switch (d->bd_bufmode) {
320 case BPF_BUFMODE_ZBUF:
321 return (bpf_zerocopy_canfreebuf(d));
322 }
323 return (0);
324 }
325
326 /*
327 * Allow the buffer model to indicate that the current store buffer is
328 * immutable, regardless of the appearance of space. Return (1) if the
329 * buffer is writable, and (0) if not.
330 */
331 static int
bpf_canwritebuf(struct bpf_d * d)332 bpf_canwritebuf(struct bpf_d *d)
333 {
334 BPFD_LOCK_ASSERT(d);
335
336 switch (d->bd_bufmode) {
337 case BPF_BUFMODE_ZBUF:
338 return (bpf_zerocopy_canwritebuf(d));
339 }
340 return (1);
341 }
342
343 /*
344 * Notify buffer model that an attempt to write to the store buffer has
345 * resulted in a dropped packet, in which case the buffer may be considered
346 * full.
347 */
348 static void
bpf_buffull(struct bpf_d * d)349 bpf_buffull(struct bpf_d *d)
350 {
351
352 BPFD_LOCK_ASSERT(d);
353
354 switch (d->bd_bufmode) {
355 case BPF_BUFMODE_ZBUF:
356 bpf_zerocopy_buffull(d);
357 break;
358 }
359 }
360
361 /*
362 * Notify the buffer model that a buffer has moved into the hold position.
363 */
364 void
bpf_bufheld(struct bpf_d * d)365 bpf_bufheld(struct bpf_d *d)
366 {
367
368 BPFD_LOCK_ASSERT(d);
369
370 switch (d->bd_bufmode) {
371 case BPF_BUFMODE_ZBUF:
372 bpf_zerocopy_bufheld(d);
373 break;
374 }
375 }
376
377 static void
bpf_free(struct bpf_d * d)378 bpf_free(struct bpf_d *d)
379 {
380
381 switch (d->bd_bufmode) {
382 case BPF_BUFMODE_BUFFER:
383 return (bpf_buffer_free(d));
384
385 case BPF_BUFMODE_ZBUF:
386 return (bpf_zerocopy_free(d));
387
388 default:
389 panic("bpf_buf_free");
390 }
391 }
392
393 static int
bpf_uiomove(struct bpf_d * d,caddr_t buf,u_int len,struct uio * uio)394 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
395 {
396
397 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
398 return (EOPNOTSUPP);
399 return (bpf_buffer_uiomove(d, buf, len, uio));
400 }
401
402 static int
bpf_ioctl_sblen(struct bpf_d * d,u_int * i)403 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
404 {
405
406 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
407 return (EOPNOTSUPP);
408 return (bpf_buffer_ioctl_sblen(d, i));
409 }
410
411 static int
bpf_ioctl_getzmax(struct thread * td,struct bpf_d * d,size_t * i)412 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
413 {
414
415 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
416 return (EOPNOTSUPP);
417 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
418 }
419
420 static int
bpf_ioctl_rotzbuf(struct thread * td,struct bpf_d * d,struct bpf_zbuf * bz)421 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
422 {
423
424 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
425 return (EOPNOTSUPP);
426 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
427 }
428
429 static int
bpf_ioctl_setzbuf(struct thread * td,struct bpf_d * d,struct bpf_zbuf * bz)430 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
431 {
432
433 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
434 return (EOPNOTSUPP);
435 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
436 }
437
438 /*
439 * General BPF functions.
440 */
441 static int
bpf_movein(struct uio * uio,int linktype,struct ifnet * ifp,struct mbuf ** mp,struct sockaddr * sockp,int * hdrlen,struct bpf_insn * wfilter)442 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
443 struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter)
444 {
445 const struct ieee80211_bpf_params *p;
446 struct ether_header *eh;
447 struct mbuf *m;
448 int error;
449 int len;
450 int hlen;
451 int slen;
452
453 /*
454 * Build a sockaddr based on the data link layer type.
455 * We do this at this level because the ethernet header
456 * is copied directly into the data field of the sockaddr.
457 * In the case of SLIP, there is no header and the packet
458 * is forwarded as is.
459 * Also, we are careful to leave room at the front of the mbuf
460 * for the link level header.
461 */
462 switch (linktype) {
463
464 case DLT_SLIP:
465 sockp->sa_family = AF_INET;
466 hlen = 0;
467 break;
468
469 case DLT_EN10MB:
470 sockp->sa_family = AF_UNSPEC;
471 /* XXX Would MAXLINKHDR be better? */
472 hlen = ETHER_HDR_LEN;
473 break;
474
475 case DLT_FDDI:
476 sockp->sa_family = AF_IMPLINK;
477 hlen = 0;
478 break;
479
480 case DLT_RAW:
481 sockp->sa_family = AF_UNSPEC;
482 hlen = 0;
483 break;
484
485 case DLT_NULL:
486 /*
487 * null interface types require a 4 byte pseudo header which
488 * corresponds to the address family of the packet.
489 */
490 sockp->sa_family = AF_UNSPEC;
491 hlen = 4;
492 break;
493
494 case DLT_ATM_RFC1483:
495 /*
496 * en atm driver requires 4-byte atm pseudo header.
497 * though it isn't standard, vpi:vci needs to be
498 * specified anyway.
499 */
500 sockp->sa_family = AF_UNSPEC;
501 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
502 break;
503
504 case DLT_PPP:
505 sockp->sa_family = AF_UNSPEC;
506 hlen = 4; /* This should match PPP_HDRLEN */
507 break;
508
509 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
510 sockp->sa_family = AF_IEEE80211;
511 hlen = 0;
512 break;
513
514 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
515 sockp->sa_family = AF_IEEE80211;
516 sockp->sa_len = 12; /* XXX != 0 */
517 hlen = sizeof(struct ieee80211_bpf_params);
518 break;
519
520 default:
521 return (EIO);
522 }
523
524 len = uio->uio_resid;
525 if (len < hlen || len - hlen > ifp->if_mtu)
526 return (EMSGSIZE);
527
528 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
529 if (m == NULL)
530 return (EIO);
531 m->m_pkthdr.len = m->m_len = len;
532 *mp = m;
533
534 error = uiomove(mtod(m, u_char *), len, uio);
535 if (error)
536 goto bad;
537
538 slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
539 if (slen == 0) {
540 error = EPERM;
541 goto bad;
542 }
543
544 /* Check for multicast destination */
545 switch (linktype) {
546 case DLT_EN10MB:
547 eh = mtod(m, struct ether_header *);
548 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
549 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
550 ETHER_ADDR_LEN) == 0)
551 m->m_flags |= M_BCAST;
552 else
553 m->m_flags |= M_MCAST;
554 }
555 break;
556 }
557
558 /*
559 * Make room for link header, and copy it to sockaddr
560 */
561 if (hlen != 0) {
562 if (sockp->sa_family == AF_IEEE80211) {
563 /*
564 * Collect true length from the parameter header
565 * NB: sockp is known to be zero'd so if we do a
566 * short copy unspecified parameters will be
567 * zero.
568 * NB: packet may not be aligned after stripping
569 * bpf params
570 * XXX check ibp_vers
571 */
572 p = mtod(m, const struct ieee80211_bpf_params *);
573 hlen = p->ibp_len;
574 if (hlen > sizeof(sockp->sa_data)) {
575 error = EINVAL;
576 goto bad;
577 }
578 }
579 bcopy(m->m_data, sockp->sa_data, hlen);
580 }
581 *hdrlen = hlen;
582
583 return (0);
584 bad:
585 m_freem(m);
586 return (error);
587 }
588
589 /*
590 * Attach file to the bpf interface, i.e. make d listen on bp.
591 */
592 static void
bpf_attachd(struct bpf_d * d,struct bpf_if * bp)593 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
594 {
595 int op_w;
596
597 BPF_LOCK_ASSERT();
598
599 /*
600 * Save sysctl value to protect from sysctl change
601 * between reads
602 */
603 op_w = V_bpf_optimize_writers || d->bd_writer;
604
605 if (d->bd_bif != NULL)
606 bpf_detachd_locked(d);
607 /*
608 * Point d at bp, and add d to the interface's list.
609 * Since there are many applicaiotns using BPF for
610 * sending raw packets only (dhcpd, cdpd are good examples)
611 * we can delay adding d to the list of active listeners until
612 * some filter is configured.
613 */
614
615 BPFIF_WLOCK(bp);
616 BPFD_LOCK(d);
617
618 d->bd_bif = bp;
619
620 if (op_w != 0) {
621 /* Add to writers-only list */
622 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
623 /*
624 * We decrement bd_writer on every filter set operation.
625 * First BIOCSETF is done by pcap_open_live() to set up
626 * snap length. After that appliation usually sets its own filter
627 */
628 d->bd_writer = 2;
629 } else
630 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
631
632 BPFD_UNLOCK(d);
633 BPFIF_WUNLOCK(bp);
634
635 bpf_bpfd_cnt++;
636
637 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
638 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
639
640 if (op_w == 0)
641 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
642 }
643
644 /*
645 * Add d to the list of active bp filters.
646 * Reuqires bpf_attachd() to be called before
647 */
648 static void
bpf_upgraded(struct bpf_d * d)649 bpf_upgraded(struct bpf_d *d)
650 {
651 struct bpf_if *bp;
652
653 BPF_LOCK_ASSERT();
654
655 bp = d->bd_bif;
656
657 /*
658 * Filter can be set several times without specifying interface.
659 * Mark d as reader and exit.
660 */
661 if (bp == NULL) {
662 BPFD_LOCK(d);
663 d->bd_writer = 0;
664 BPFD_UNLOCK(d);
665 return;
666 }
667
668 BPFIF_WLOCK(bp);
669 BPFD_LOCK(d);
670
671 /* Remove from writers-only list */
672 LIST_REMOVE(d, bd_next);
673 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
674 /* Mark d as reader */
675 d->bd_writer = 0;
676
677 BPFD_UNLOCK(d);
678 BPFIF_WUNLOCK(bp);
679
680 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
681
682 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
683 }
684
685 /*
686 * Detach a file from its interface.
687 */
688 static void
bpf_detachd(struct bpf_d * d)689 bpf_detachd(struct bpf_d *d)
690 {
691 BPF_LOCK();
692 bpf_detachd_locked(d);
693 BPF_UNLOCK();
694 }
695
696 static void
bpf_detachd_locked(struct bpf_d * d)697 bpf_detachd_locked(struct bpf_d *d)
698 {
699 int error;
700 struct bpf_if *bp;
701 struct ifnet *ifp;
702
703 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
704
705 BPF_LOCK_ASSERT();
706
707 /* Check if descriptor is attached */
708 if ((bp = d->bd_bif) == NULL)
709 return;
710
711 BPFIF_WLOCK(bp);
712 BPFD_LOCK(d);
713
714 /* Save bd_writer value */
715 error = d->bd_writer;
716
717 /*
718 * Remove d from the interface's descriptor list.
719 */
720 LIST_REMOVE(d, bd_next);
721
722 ifp = bp->bif_ifp;
723 d->bd_bif = NULL;
724 BPFD_UNLOCK(d);
725 BPFIF_WUNLOCK(bp);
726
727 bpf_bpfd_cnt--;
728
729 /* Call event handler iff d is attached */
730 if (error == 0)
731 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
732
733 /*
734 * Check if this descriptor had requested promiscuous mode.
735 * If so, turn it off.
736 */
737 if (d->bd_promisc) {
738 d->bd_promisc = 0;
739 CURVNET_SET(ifp->if_vnet);
740 error = ifpromisc(ifp, 0);
741 CURVNET_RESTORE();
742 if (error != 0 && error != ENXIO) {
743 /*
744 * ENXIO can happen if a pccard is unplugged
745 * Something is really wrong if we were able to put
746 * the driver into promiscuous mode, but can't
747 * take it out.
748 */
749 if_printf(bp->bif_ifp,
750 "bpf_detach: ifpromisc failed (%d)\n", error);
751 }
752 }
753 }
754
755 /*
756 * Close the descriptor by detaching it from its interface,
757 * deallocating its buffers, and marking it free.
758 */
759 static void
bpf_dtor(void * data)760 bpf_dtor(void *data)
761 {
762 struct bpf_d *d = data;
763
764 BPFD_LOCK(d);
765 if (d->bd_state == BPF_WAITING)
766 callout_stop(&d->bd_callout);
767 d->bd_state = BPF_IDLE;
768 BPFD_UNLOCK(d);
769 funsetown(&d->bd_sigio);
770 bpf_detachd(d);
771 #ifdef MAC
772 mac_bpfdesc_destroy(d);
773 #endif /* MAC */
774 seldrain(&d->bd_sel);
775 knlist_destroy(&d->bd_sel.si_note);
776 callout_drain(&d->bd_callout);
777 bpf_freed(d);
778 free(d, M_BPF);
779 }
780
781 /*
782 * Open ethernet device. Returns ENXIO for illegal minor device number,
783 * EBUSY if file is open by another process.
784 */
785 /* ARGSUSED */
786 static int
bpfopen(struct cdev * dev,int flags,int fmt,struct thread * td)787 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
788 {
789 struct bpf_d *d;
790 int error, size;
791
792 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
793 error = devfs_set_cdevpriv(d, bpf_dtor);
794 if (error != 0) {
795 free(d, M_BPF);
796 return (error);
797 }
798
799 /*
800 * For historical reasons, perform a one-time initialization call to
801 * the buffer routines, even though we're not yet committed to a
802 * particular buffer method.
803 */
804 bpf_buffer_init(d);
805 if ((flags & FREAD) == 0)
806 d->bd_writer = 2;
807 d->bd_hbuf_in_use = 0;
808 d->bd_bufmode = BPF_BUFMODE_BUFFER;
809 d->bd_sig = SIGIO;
810 d->bd_direction = BPF_D_INOUT;
811 BPF_PID_REFRESH(d, td);
812 #ifdef MAC
813 mac_bpfdesc_init(d);
814 mac_bpfdesc_create(td->td_ucred, d);
815 #endif
816 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
817 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
818 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
819
820 /* Allocate default buffers */
821 size = d->bd_bufsize;
822 bpf_buffer_ioctl_sblen(d, &size);
823
824 return (0);
825 }
826
827 /*
828 * bpfread - read next chunk of packets from buffers
829 */
830 static int
bpfread(struct cdev * dev,struct uio * uio,int ioflag)831 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
832 {
833 struct bpf_d *d;
834 int error;
835 int non_block;
836 int timed_out;
837
838 error = devfs_get_cdevpriv((void **)&d);
839 if (error != 0)
840 return (error);
841
842 /*
843 * Restrict application to use a buffer the same size as
844 * as kernel buffers.
845 */
846 if (uio->uio_resid != d->bd_bufsize)
847 return (EINVAL);
848
849 non_block = ((ioflag & O_NONBLOCK) != 0);
850
851 BPFD_LOCK(d);
852 BPF_PID_REFRESH_CUR(d);
853 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
854 BPFD_UNLOCK(d);
855 return (EOPNOTSUPP);
856 }
857 if (d->bd_state == BPF_WAITING)
858 callout_stop(&d->bd_callout);
859 timed_out = (d->bd_state == BPF_TIMED_OUT);
860 d->bd_state = BPF_IDLE;
861 while (d->bd_hbuf_in_use) {
862 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
863 PRINET|PCATCH, "bd_hbuf", 0);
864 if (error != 0) {
865 BPFD_UNLOCK(d);
866 return (error);
867 }
868 }
869 /*
870 * If the hold buffer is empty, then do a timed sleep, which
871 * ends when the timeout expires or when enough packets
872 * have arrived to fill the store buffer.
873 */
874 while (d->bd_hbuf == NULL) {
875 if (d->bd_slen != 0) {
876 /*
877 * A packet(s) either arrived since the previous
878 * read or arrived while we were asleep.
879 */
880 if (d->bd_immediate || non_block || timed_out) {
881 /*
882 * Rotate the buffers and return what's here
883 * if we are in immediate mode, non-blocking
884 * flag is set, or this descriptor timed out.
885 */
886 ROTATE_BUFFERS(d);
887 break;
888 }
889 }
890
891 /*
892 * No data is available, check to see if the bpf device
893 * is still pointed at a real interface. If not, return
894 * ENXIO so that the userland process knows to rebind
895 * it before using it again.
896 */
897 if (d->bd_bif == NULL) {
898 BPFD_UNLOCK(d);
899 return (ENXIO);
900 }
901
902 if (non_block) {
903 BPFD_UNLOCK(d);
904 return (EWOULDBLOCK);
905 }
906 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
907 "bpf", d->bd_rtout);
908 if (error == EINTR || error == ERESTART) {
909 BPFD_UNLOCK(d);
910 return (error);
911 }
912 if (error == EWOULDBLOCK) {
913 /*
914 * On a timeout, return what's in the buffer,
915 * which may be nothing. If there is something
916 * in the store buffer, we can rotate the buffers.
917 */
918 if (d->bd_hbuf)
919 /*
920 * We filled up the buffer in between
921 * getting the timeout and arriving
922 * here, so we don't need to rotate.
923 */
924 break;
925
926 if (d->bd_slen == 0) {
927 BPFD_UNLOCK(d);
928 return (0);
929 }
930 ROTATE_BUFFERS(d);
931 break;
932 }
933 }
934 /*
935 * At this point, we know we have something in the hold slot.
936 */
937 d->bd_hbuf_in_use = 1;
938 BPFD_UNLOCK(d);
939
940 /*
941 * Move data from hold buffer into user space.
942 * We know the entire buffer is transferred since
943 * we checked above that the read buffer is bpf_bufsize bytes.
944 *
945 * We do not have to worry about simultaneous reads because
946 * we waited for sole access to the hold buffer above.
947 */
948 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
949
950 BPFD_LOCK(d);
951 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
952 d->bd_fbuf = d->bd_hbuf;
953 d->bd_hbuf = NULL;
954 d->bd_hlen = 0;
955 bpf_buf_reclaimed(d);
956 d->bd_hbuf_in_use = 0;
957 wakeup(&d->bd_hbuf_in_use);
958 BPFD_UNLOCK(d);
959
960 return (error);
961 }
962
963 /*
964 * If there are processes sleeping on this descriptor, wake them up.
965 */
966 static __inline void
bpf_wakeup(struct bpf_d * d)967 bpf_wakeup(struct bpf_d *d)
968 {
969
970 BPFD_LOCK_ASSERT(d);
971 if (d->bd_state == BPF_WAITING) {
972 callout_stop(&d->bd_callout);
973 d->bd_state = BPF_IDLE;
974 }
975 wakeup(d);
976 if (d->bd_async && d->bd_sig && d->bd_sigio)
977 pgsigio(&d->bd_sigio, d->bd_sig, 0);
978
979 selwakeuppri(&d->bd_sel, PRINET);
980 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
981 }
982
983 static void
bpf_timed_out(void * arg)984 bpf_timed_out(void *arg)
985 {
986 struct bpf_d *d = (struct bpf_d *)arg;
987
988 BPFD_LOCK_ASSERT(d);
989
990 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
991 return;
992 if (d->bd_state == BPF_WAITING) {
993 d->bd_state = BPF_TIMED_OUT;
994 if (d->bd_slen != 0)
995 bpf_wakeup(d);
996 }
997 }
998
999 static int
bpf_ready(struct bpf_d * d)1000 bpf_ready(struct bpf_d *d)
1001 {
1002
1003 BPFD_LOCK_ASSERT(d);
1004
1005 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1006 return (1);
1007 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1008 d->bd_slen != 0)
1009 return (1);
1010 return (0);
1011 }
1012
1013 static int
bpfwrite(struct cdev * dev,struct uio * uio,int ioflag)1014 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1015 {
1016 struct bpf_d *d;
1017 struct ifnet *ifp;
1018 struct mbuf *m, *mc;
1019 struct sockaddr dst;
1020 int error, hlen;
1021
1022 error = devfs_get_cdevpriv((void **)&d);
1023 if (error != 0)
1024 return (error);
1025
1026 BPF_PID_REFRESH_CUR(d);
1027 d->bd_wcount++;
1028 /* XXX: locking required */
1029 if (d->bd_bif == NULL) {
1030 d->bd_wdcount++;
1031 return (ENXIO);
1032 }
1033
1034 ifp = d->bd_bif->bif_ifp;
1035
1036 if ((ifp->if_flags & IFF_UP) == 0) {
1037 d->bd_wdcount++;
1038 return (ENETDOWN);
1039 }
1040
1041 if (uio->uio_resid == 0) {
1042 d->bd_wdcount++;
1043 return (0);
1044 }
1045
1046 bzero(&dst, sizeof(dst));
1047 m = NULL;
1048 hlen = 0;
1049 /* XXX: bpf_movein() can sleep */
1050 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1051 &m, &dst, &hlen, d->bd_wfilter);
1052 if (error) {
1053 d->bd_wdcount++;
1054 return (error);
1055 }
1056 d->bd_wfcount++;
1057 if (d->bd_hdrcmplt)
1058 dst.sa_family = pseudo_AF_HDRCMPLT;
1059
1060 if (d->bd_feedback) {
1061 mc = m_dup(m, M_NOWAIT);
1062 if (mc != NULL)
1063 mc->m_pkthdr.rcvif = ifp;
1064 /* Set M_PROMISC for outgoing packets to be discarded. */
1065 if (d->bd_direction == BPF_D_INOUT)
1066 m->m_flags |= M_PROMISC;
1067 } else
1068 mc = NULL;
1069
1070 m->m_pkthdr.len -= hlen;
1071 m->m_len -= hlen;
1072 m->m_data += hlen; /* XXX */
1073
1074 CURVNET_SET(ifp->if_vnet);
1075 #ifdef MAC
1076 BPFD_LOCK(d);
1077 mac_bpfdesc_create_mbuf(d, m);
1078 if (mc != NULL)
1079 mac_bpfdesc_create_mbuf(d, mc);
1080 BPFD_UNLOCK(d);
1081 #endif
1082
1083 error = (*ifp->if_output)(ifp, m, &dst, NULL);
1084 if (error)
1085 d->bd_wdcount++;
1086
1087 if (mc != NULL) {
1088 if (error == 0)
1089 (*ifp->if_input)(ifp, mc);
1090 else
1091 m_freem(mc);
1092 }
1093 CURVNET_RESTORE();
1094
1095 return (error);
1096 }
1097
1098 /*
1099 * Reset a descriptor by flushing its packet buffer and clearing the receive
1100 * and drop counts. This is doable for kernel-only buffers, but with
1101 * zero-copy buffers, we can't write to (or rotate) buffers that are
1102 * currently owned by userspace. It would be nice if we could encapsulate
1103 * this logic in the buffer code rather than here.
1104 */
1105 static void
reset_d(struct bpf_d * d)1106 reset_d(struct bpf_d *d)
1107 {
1108
1109 BPFD_LOCK_ASSERT(d);
1110
1111 while (d->bd_hbuf_in_use)
1112 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1113 "bd_hbuf", 0);
1114 if ((d->bd_hbuf != NULL) &&
1115 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1116 /* Free the hold buffer. */
1117 d->bd_fbuf = d->bd_hbuf;
1118 d->bd_hbuf = NULL;
1119 d->bd_hlen = 0;
1120 bpf_buf_reclaimed(d);
1121 }
1122 if (bpf_canwritebuf(d))
1123 d->bd_slen = 0;
1124 d->bd_rcount = 0;
1125 d->bd_dcount = 0;
1126 d->bd_fcount = 0;
1127 d->bd_wcount = 0;
1128 d->bd_wfcount = 0;
1129 d->bd_wdcount = 0;
1130 d->bd_zcopy = 0;
1131 }
1132
1133 /*
1134 * FIONREAD Check for read packet available.
1135 * SIOCGIFADDR Get interface address - convenient hook to driver.
1136 * BIOCGBLEN Get buffer len [for read()].
1137 * BIOCSETF Set read filter.
1138 * BIOCSETFNR Set read filter without resetting descriptor.
1139 * BIOCSETWF Set write filter.
1140 * BIOCFLUSH Flush read packet buffer.
1141 * BIOCPROMISC Put interface into promiscuous mode.
1142 * BIOCGDLT Get link layer type.
1143 * BIOCGETIF Get interface name.
1144 * BIOCSETIF Set interface.
1145 * BIOCSRTIMEOUT Set read timeout.
1146 * BIOCGRTIMEOUT Get read timeout.
1147 * BIOCGSTATS Get packet stats.
1148 * BIOCIMMEDIATE Set immediate mode.
1149 * BIOCVERSION Get filter language version.
1150 * BIOCGHDRCMPLT Get "header already complete" flag
1151 * BIOCSHDRCMPLT Set "header already complete" flag
1152 * BIOCGDIRECTION Get packet direction flag
1153 * BIOCSDIRECTION Set packet direction flag
1154 * BIOCGTSTAMP Get time stamp format and resolution.
1155 * BIOCSTSTAMP Set time stamp format and resolution.
1156 * BIOCLOCK Set "locked" flag
1157 * BIOCFEEDBACK Set packet feedback mode.
1158 * BIOCSETZBUF Set current zero-copy buffer locations.
1159 * BIOCGETZMAX Get maximum zero-copy buffer size.
1160 * BIOCROTZBUF Force rotation of zero-copy buffer
1161 * BIOCSETBUFMODE Set buffer mode.
1162 * BIOCGETBUFMODE Get current buffer mode.
1163 */
1164 /* ARGSUSED */
1165 static int
bpfioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flags,struct thread * td)1166 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1167 struct thread *td)
1168 {
1169 struct bpf_d *d;
1170 int error;
1171
1172 error = devfs_get_cdevpriv((void **)&d);
1173 if (error != 0)
1174 return (error);
1175
1176 /*
1177 * Refresh PID associated with this descriptor.
1178 */
1179 BPFD_LOCK(d);
1180 BPF_PID_REFRESH(d, td);
1181 if (d->bd_state == BPF_WAITING)
1182 callout_stop(&d->bd_callout);
1183 d->bd_state = BPF_IDLE;
1184 BPFD_UNLOCK(d);
1185
1186 if (d->bd_locked == 1) {
1187 switch (cmd) {
1188 case BIOCGBLEN:
1189 case BIOCFLUSH:
1190 case BIOCGDLT:
1191 case BIOCGDLTLIST:
1192 #ifdef COMPAT_FREEBSD32
1193 case BIOCGDLTLIST32:
1194 #endif
1195 case BIOCGETIF:
1196 case BIOCGRTIMEOUT:
1197 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1198 case BIOCGRTIMEOUT32:
1199 #endif
1200 case BIOCGSTATS:
1201 case BIOCVERSION:
1202 case BIOCGRSIG:
1203 case BIOCGHDRCMPLT:
1204 case BIOCSTSTAMP:
1205 case BIOCFEEDBACK:
1206 case FIONREAD:
1207 case BIOCLOCK:
1208 case BIOCSRTIMEOUT:
1209 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1210 case BIOCSRTIMEOUT32:
1211 #endif
1212 case BIOCIMMEDIATE:
1213 case TIOCGPGRP:
1214 case BIOCROTZBUF:
1215 break;
1216 default:
1217 return (EPERM);
1218 }
1219 }
1220 #ifdef COMPAT_FREEBSD32
1221 /*
1222 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1223 * that it will get 32-bit packet headers.
1224 */
1225 switch (cmd) {
1226 case BIOCSETF32:
1227 case BIOCSETFNR32:
1228 case BIOCSETWF32:
1229 case BIOCGDLTLIST32:
1230 case BIOCGRTIMEOUT32:
1231 case BIOCSRTIMEOUT32:
1232 BPFD_LOCK(d);
1233 d->bd_compat32 = 1;
1234 BPFD_UNLOCK(d);
1235 }
1236 #endif
1237
1238 CURVNET_SET(TD_TO_VNET(td));
1239 switch (cmd) {
1240
1241 default:
1242 error = EINVAL;
1243 break;
1244
1245 /*
1246 * Check for read packet available.
1247 */
1248 case FIONREAD:
1249 {
1250 int n;
1251
1252 BPFD_LOCK(d);
1253 n = d->bd_slen;
1254 while (d->bd_hbuf_in_use)
1255 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1256 PRINET, "bd_hbuf", 0);
1257 if (d->bd_hbuf)
1258 n += d->bd_hlen;
1259 BPFD_UNLOCK(d);
1260
1261 *(int *)addr = n;
1262 break;
1263 }
1264
1265 case SIOCGIFADDR:
1266 {
1267 struct ifnet *ifp;
1268
1269 if (d->bd_bif == NULL)
1270 error = EINVAL;
1271 else {
1272 ifp = d->bd_bif->bif_ifp;
1273 error = (*ifp->if_ioctl)(ifp, cmd, addr);
1274 }
1275 break;
1276 }
1277
1278 /*
1279 * Get buffer len [for read()].
1280 */
1281 case BIOCGBLEN:
1282 BPFD_LOCK(d);
1283 *(u_int *)addr = d->bd_bufsize;
1284 BPFD_UNLOCK(d);
1285 break;
1286
1287 /*
1288 * Set buffer length.
1289 */
1290 case BIOCSBLEN:
1291 error = bpf_ioctl_sblen(d, (u_int *)addr);
1292 break;
1293
1294 /*
1295 * Set link layer read filter.
1296 */
1297 case BIOCSETF:
1298 case BIOCSETFNR:
1299 case BIOCSETWF:
1300 #ifdef COMPAT_FREEBSD32
1301 case BIOCSETF32:
1302 case BIOCSETFNR32:
1303 case BIOCSETWF32:
1304 #endif
1305 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1306 break;
1307
1308 /*
1309 * Flush read packet buffer.
1310 */
1311 case BIOCFLUSH:
1312 BPFD_LOCK(d);
1313 reset_d(d);
1314 BPFD_UNLOCK(d);
1315 break;
1316
1317 /*
1318 * Put interface into promiscuous mode.
1319 */
1320 case BIOCPROMISC:
1321 if (d->bd_bif == NULL) {
1322 /*
1323 * No interface attached yet.
1324 */
1325 error = EINVAL;
1326 break;
1327 }
1328 if (d->bd_promisc == 0) {
1329 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1330 if (error == 0)
1331 d->bd_promisc = 1;
1332 }
1333 break;
1334
1335 /*
1336 * Get current data link type.
1337 */
1338 case BIOCGDLT:
1339 BPF_LOCK();
1340 if (d->bd_bif == NULL)
1341 error = EINVAL;
1342 else
1343 *(u_int *)addr = d->bd_bif->bif_dlt;
1344 BPF_UNLOCK();
1345 break;
1346
1347 /*
1348 * Get a list of supported data link types.
1349 */
1350 #ifdef COMPAT_FREEBSD32
1351 case BIOCGDLTLIST32:
1352 {
1353 struct bpf_dltlist32 *list32;
1354 struct bpf_dltlist dltlist;
1355
1356 list32 = (struct bpf_dltlist32 *)addr;
1357 dltlist.bfl_len = list32->bfl_len;
1358 dltlist.bfl_list = PTRIN(list32->bfl_list);
1359 BPF_LOCK();
1360 if (d->bd_bif == NULL)
1361 error = EINVAL;
1362 else {
1363 error = bpf_getdltlist(d, &dltlist);
1364 if (error == 0)
1365 list32->bfl_len = dltlist.bfl_len;
1366 }
1367 BPF_UNLOCK();
1368 break;
1369 }
1370 #endif
1371
1372 case BIOCGDLTLIST:
1373 BPF_LOCK();
1374 if (d->bd_bif == NULL)
1375 error = EINVAL;
1376 else
1377 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1378 BPF_UNLOCK();
1379 break;
1380
1381 /*
1382 * Set data link type.
1383 */
1384 case BIOCSDLT:
1385 BPF_LOCK();
1386 if (d->bd_bif == NULL)
1387 error = EINVAL;
1388 else
1389 error = bpf_setdlt(d, *(u_int *)addr);
1390 BPF_UNLOCK();
1391 break;
1392
1393 /*
1394 * Get interface name.
1395 */
1396 case BIOCGETIF:
1397 BPF_LOCK();
1398 if (d->bd_bif == NULL)
1399 error = EINVAL;
1400 else {
1401 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1402 struct ifreq *const ifr = (struct ifreq *)addr;
1403
1404 strlcpy(ifr->ifr_name, ifp->if_xname,
1405 sizeof(ifr->ifr_name));
1406 }
1407 BPF_UNLOCK();
1408 break;
1409
1410 /*
1411 * Set interface.
1412 */
1413 case BIOCSETIF:
1414 BPF_LOCK();
1415 error = bpf_setif(d, (struct ifreq *)addr);
1416 BPF_UNLOCK();
1417 break;
1418
1419 /*
1420 * Set read timeout.
1421 */
1422 case BIOCSRTIMEOUT:
1423 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1424 case BIOCSRTIMEOUT32:
1425 #endif
1426 {
1427 struct timeval *tv = (struct timeval *)addr;
1428 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1429 struct timeval32 *tv32;
1430 struct timeval tv64;
1431
1432 if (cmd == BIOCSRTIMEOUT32) {
1433 tv32 = (struct timeval32 *)addr;
1434 tv = &tv64;
1435 tv->tv_sec = tv32->tv_sec;
1436 tv->tv_usec = tv32->tv_usec;
1437 } else
1438 #endif
1439 tv = (struct timeval *)addr;
1440
1441 /*
1442 * Subtract 1 tick from tvtohz() since this isn't
1443 * a one-shot timer.
1444 */
1445 if ((error = itimerfix(tv)) == 0)
1446 d->bd_rtout = tvtohz(tv) - 1;
1447 break;
1448 }
1449
1450 /*
1451 * Get read timeout.
1452 */
1453 case BIOCGRTIMEOUT:
1454 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1455 case BIOCGRTIMEOUT32:
1456 #endif
1457 {
1458 struct timeval *tv;
1459 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1460 struct timeval32 *tv32;
1461 struct timeval tv64;
1462
1463 if (cmd == BIOCGRTIMEOUT32)
1464 tv = &tv64;
1465 else
1466 #endif
1467 tv = (struct timeval *)addr;
1468
1469 tv->tv_sec = d->bd_rtout / hz;
1470 tv->tv_usec = (d->bd_rtout % hz) * tick;
1471 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1472 if (cmd == BIOCGRTIMEOUT32) {
1473 tv32 = (struct timeval32 *)addr;
1474 tv32->tv_sec = tv->tv_sec;
1475 tv32->tv_usec = tv->tv_usec;
1476 }
1477 #endif
1478
1479 break;
1480 }
1481
1482 /*
1483 * Get packet stats.
1484 */
1485 case BIOCGSTATS:
1486 {
1487 struct bpf_stat *bs = (struct bpf_stat *)addr;
1488
1489 /* XXXCSJP overflow */
1490 bs->bs_recv = d->bd_rcount;
1491 bs->bs_drop = d->bd_dcount;
1492 break;
1493 }
1494
1495 /*
1496 * Set immediate mode.
1497 */
1498 case BIOCIMMEDIATE:
1499 BPFD_LOCK(d);
1500 d->bd_immediate = *(u_int *)addr;
1501 BPFD_UNLOCK(d);
1502 break;
1503
1504 case BIOCVERSION:
1505 {
1506 struct bpf_version *bv = (struct bpf_version *)addr;
1507
1508 bv->bv_major = BPF_MAJOR_VERSION;
1509 bv->bv_minor = BPF_MINOR_VERSION;
1510 break;
1511 }
1512
1513 /*
1514 * Get "header already complete" flag
1515 */
1516 case BIOCGHDRCMPLT:
1517 BPFD_LOCK(d);
1518 *(u_int *)addr = d->bd_hdrcmplt;
1519 BPFD_UNLOCK(d);
1520 break;
1521
1522 /*
1523 * Set "header already complete" flag
1524 */
1525 case BIOCSHDRCMPLT:
1526 BPFD_LOCK(d);
1527 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1528 BPFD_UNLOCK(d);
1529 break;
1530
1531 /*
1532 * Get packet direction flag
1533 */
1534 case BIOCGDIRECTION:
1535 BPFD_LOCK(d);
1536 *(u_int *)addr = d->bd_direction;
1537 BPFD_UNLOCK(d);
1538 break;
1539
1540 /*
1541 * Set packet direction flag
1542 */
1543 case BIOCSDIRECTION:
1544 {
1545 u_int direction;
1546
1547 direction = *(u_int *)addr;
1548 switch (direction) {
1549 case BPF_D_IN:
1550 case BPF_D_INOUT:
1551 case BPF_D_OUT:
1552 BPFD_LOCK(d);
1553 d->bd_direction = direction;
1554 BPFD_UNLOCK(d);
1555 break;
1556 default:
1557 error = EINVAL;
1558 }
1559 }
1560 break;
1561
1562 /*
1563 * Get packet timestamp format and resolution.
1564 */
1565 case BIOCGTSTAMP:
1566 BPFD_LOCK(d);
1567 *(u_int *)addr = d->bd_tstamp;
1568 BPFD_UNLOCK(d);
1569 break;
1570
1571 /*
1572 * Set packet timestamp format and resolution.
1573 */
1574 case BIOCSTSTAMP:
1575 {
1576 u_int func;
1577
1578 func = *(u_int *)addr;
1579 if (BPF_T_VALID(func))
1580 d->bd_tstamp = func;
1581 else
1582 error = EINVAL;
1583 }
1584 break;
1585
1586 case BIOCFEEDBACK:
1587 BPFD_LOCK(d);
1588 d->bd_feedback = *(u_int *)addr;
1589 BPFD_UNLOCK(d);
1590 break;
1591
1592 case BIOCLOCK:
1593 BPFD_LOCK(d);
1594 d->bd_locked = 1;
1595 BPFD_UNLOCK(d);
1596 break;
1597
1598 case FIONBIO: /* Non-blocking I/O */
1599 break;
1600
1601 case FIOASYNC: /* Send signal on receive packets */
1602 BPFD_LOCK(d);
1603 d->bd_async = *(int *)addr;
1604 BPFD_UNLOCK(d);
1605 break;
1606
1607 case FIOSETOWN:
1608 /*
1609 * XXX: Add some sort of locking here?
1610 * fsetown() can sleep.
1611 */
1612 error = fsetown(*(int *)addr, &d->bd_sigio);
1613 break;
1614
1615 case FIOGETOWN:
1616 BPFD_LOCK(d);
1617 *(int *)addr = fgetown(&d->bd_sigio);
1618 BPFD_UNLOCK(d);
1619 break;
1620
1621 /* This is deprecated, FIOSETOWN should be used instead. */
1622 case TIOCSPGRP:
1623 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1624 break;
1625
1626 /* This is deprecated, FIOGETOWN should be used instead. */
1627 case TIOCGPGRP:
1628 *(int *)addr = -fgetown(&d->bd_sigio);
1629 break;
1630
1631 case BIOCSRSIG: /* Set receive signal */
1632 {
1633 u_int sig;
1634
1635 sig = *(u_int *)addr;
1636
1637 if (sig >= NSIG)
1638 error = EINVAL;
1639 else {
1640 BPFD_LOCK(d);
1641 d->bd_sig = sig;
1642 BPFD_UNLOCK(d);
1643 }
1644 break;
1645 }
1646 case BIOCGRSIG:
1647 BPFD_LOCK(d);
1648 *(u_int *)addr = d->bd_sig;
1649 BPFD_UNLOCK(d);
1650 break;
1651
1652 case BIOCGETBUFMODE:
1653 BPFD_LOCK(d);
1654 *(u_int *)addr = d->bd_bufmode;
1655 BPFD_UNLOCK(d);
1656 break;
1657
1658 case BIOCSETBUFMODE:
1659 /*
1660 * Allow the buffering mode to be changed as long as we
1661 * haven't yet committed to a particular mode. Our
1662 * definition of commitment, for now, is whether or not a
1663 * buffer has been allocated or an interface attached, since
1664 * that's the point where things get tricky.
1665 */
1666 switch (*(u_int *)addr) {
1667 case BPF_BUFMODE_BUFFER:
1668 break;
1669
1670 case BPF_BUFMODE_ZBUF:
1671 if (bpf_zerocopy_enable)
1672 break;
1673 /* FALLSTHROUGH */
1674
1675 default:
1676 CURVNET_RESTORE();
1677 return (EINVAL);
1678 }
1679
1680 BPFD_LOCK(d);
1681 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1682 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1683 BPFD_UNLOCK(d);
1684 CURVNET_RESTORE();
1685 return (EBUSY);
1686 }
1687 d->bd_bufmode = *(u_int *)addr;
1688 BPFD_UNLOCK(d);
1689 break;
1690
1691 case BIOCGETZMAX:
1692 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1693 break;
1694
1695 case BIOCSETZBUF:
1696 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1697 break;
1698
1699 case BIOCROTZBUF:
1700 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1701 break;
1702 }
1703 CURVNET_RESTORE();
1704 return (error);
1705 }
1706
1707 /*
1708 * Set d's packet filter program to fp. If this file already has a filter,
1709 * free it and replace it. Returns EINVAL for bogus requests.
1710 *
1711 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1712 * since reading d->bd_bif can't be protected by d or interface lock due to
1713 * lock order.
1714 *
1715 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1716 * interface read lock to read all filers.
1717 *
1718 */
1719 static int
bpf_setf(struct bpf_d * d,struct bpf_program * fp,u_long cmd)1720 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1721 {
1722 #ifdef COMPAT_FREEBSD32
1723 struct bpf_program fp_swab;
1724 struct bpf_program32 *fp32;
1725 #endif
1726 struct bpf_insn *fcode, *old;
1727 #ifdef BPF_JITTER
1728 bpf_jit_filter *jfunc, *ofunc;
1729 #endif
1730 size_t size;
1731 u_int flen;
1732 int need_upgrade;
1733
1734 #ifdef COMPAT_FREEBSD32
1735 switch (cmd) {
1736 case BIOCSETF32:
1737 case BIOCSETWF32:
1738 case BIOCSETFNR32:
1739 fp32 = (struct bpf_program32 *)fp;
1740 fp_swab.bf_len = fp32->bf_len;
1741 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1742 fp = &fp_swab;
1743 switch (cmd) {
1744 case BIOCSETF32:
1745 cmd = BIOCSETF;
1746 break;
1747 case BIOCSETWF32:
1748 cmd = BIOCSETWF;
1749 break;
1750 }
1751 break;
1752 }
1753 #endif
1754
1755 fcode = NULL;
1756 #ifdef BPF_JITTER
1757 jfunc = ofunc = NULL;
1758 #endif
1759 need_upgrade = 0;
1760
1761 /*
1762 * Check new filter validness before acquiring any locks.
1763 * Allocate memory for new filter, if needed.
1764 */
1765 flen = fp->bf_len;
1766 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1767 return (EINVAL);
1768 size = flen * sizeof(*fp->bf_insns);
1769 if (size > 0) {
1770 /* We're setting up new filter. Copy and check actual data. */
1771 fcode = malloc(size, M_BPF, M_WAITOK);
1772 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1773 !bpf_validate(fcode, flen)) {
1774 free(fcode, M_BPF);
1775 return (EINVAL);
1776 }
1777 #ifdef BPF_JITTER
1778 /* Filter is copied inside fcode and is perfectly valid. */
1779 jfunc = bpf_jitter(fcode, flen);
1780 #endif
1781 }
1782
1783 BPF_LOCK();
1784
1785 /*
1786 * Set up new filter.
1787 * Protect filter change by interface lock.
1788 * Additionally, we are protected by global lock here.
1789 */
1790 if (d->bd_bif != NULL)
1791 BPFIF_WLOCK(d->bd_bif);
1792 BPFD_LOCK(d);
1793 if (cmd == BIOCSETWF) {
1794 old = d->bd_wfilter;
1795 d->bd_wfilter = fcode;
1796 } else {
1797 old = d->bd_rfilter;
1798 d->bd_rfilter = fcode;
1799 #ifdef BPF_JITTER
1800 ofunc = d->bd_bfilter;
1801 d->bd_bfilter = jfunc;
1802 #endif
1803 if (cmd == BIOCSETF)
1804 reset_d(d);
1805
1806 if (fcode != NULL) {
1807 /*
1808 * Do not require upgrade by first BIOCSETF
1809 * (used to set snaplen) by pcap_open_live().
1810 */
1811 if (d->bd_writer != 0 && --d->bd_writer == 0)
1812 need_upgrade = 1;
1813 CTR4(KTR_NET, "%s: filter function set by pid %d, "
1814 "bd_writer counter %d, need_upgrade %d",
1815 __func__, d->bd_pid, d->bd_writer, need_upgrade);
1816 }
1817 }
1818 BPFD_UNLOCK(d);
1819 if (d->bd_bif != NULL)
1820 BPFIF_WUNLOCK(d->bd_bif);
1821 if (old != NULL)
1822 free(old, M_BPF);
1823 #ifdef BPF_JITTER
1824 if (ofunc != NULL)
1825 bpf_destroy_jit_filter(ofunc);
1826 #endif
1827
1828 /* Move d to active readers list. */
1829 if (need_upgrade)
1830 bpf_upgraded(d);
1831
1832 BPF_UNLOCK();
1833 return (0);
1834 }
1835
1836 /*
1837 * Detach a file from its current interface (if attached at all) and attach
1838 * to the interface indicated by the name stored in ifr.
1839 * Return an errno or 0.
1840 */
1841 static int
bpf_setif(struct bpf_d * d,struct ifreq * ifr)1842 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1843 {
1844 struct bpf_if *bp;
1845 struct ifnet *theywant;
1846
1847 BPF_LOCK_ASSERT();
1848
1849 theywant = ifunit(ifr->ifr_name);
1850 if (theywant == NULL || theywant->if_bpf == NULL)
1851 return (ENXIO);
1852
1853 bp = theywant->if_bpf;
1854
1855 /* Check if interface is not being detached from BPF */
1856 BPFIF_RLOCK(bp);
1857 if (bp->flags & BPFIF_FLAG_DYING) {
1858 BPFIF_RUNLOCK(bp);
1859 return (ENXIO);
1860 }
1861 BPFIF_RUNLOCK(bp);
1862
1863 /*
1864 * Behavior here depends on the buffering model. If we're using
1865 * kernel memory buffers, then we can allocate them here. If we're
1866 * using zero-copy, then the user process must have registered
1867 * buffers by the time we get here. If not, return an error.
1868 */
1869 switch (d->bd_bufmode) {
1870 case BPF_BUFMODE_BUFFER:
1871 case BPF_BUFMODE_ZBUF:
1872 if (d->bd_sbuf == NULL)
1873 return (EINVAL);
1874 break;
1875
1876 default:
1877 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1878 }
1879 if (bp != d->bd_bif)
1880 bpf_attachd(d, bp);
1881 BPFD_LOCK(d);
1882 reset_d(d);
1883 BPFD_UNLOCK(d);
1884 return (0);
1885 }
1886
1887 /*
1888 * Support for select() and poll() system calls
1889 *
1890 * Return true iff the specific operation will not block indefinitely.
1891 * Otherwise, return false but make a note that a selwakeup() must be done.
1892 */
1893 static int
bpfpoll(struct cdev * dev,int events,struct thread * td)1894 bpfpoll(struct cdev *dev, int events, struct thread *td)
1895 {
1896 struct bpf_d *d;
1897 int revents;
1898
1899 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1900 return (events &
1901 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1902
1903 /*
1904 * Refresh PID associated with this descriptor.
1905 */
1906 revents = events & (POLLOUT | POLLWRNORM);
1907 BPFD_LOCK(d);
1908 BPF_PID_REFRESH(d, td);
1909 if (events & (POLLIN | POLLRDNORM)) {
1910 if (bpf_ready(d))
1911 revents |= events & (POLLIN | POLLRDNORM);
1912 else {
1913 selrecord(td, &d->bd_sel);
1914 /* Start the read timeout if necessary. */
1915 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1916 callout_reset(&d->bd_callout, d->bd_rtout,
1917 bpf_timed_out, d);
1918 d->bd_state = BPF_WAITING;
1919 }
1920 }
1921 }
1922 BPFD_UNLOCK(d);
1923 return (revents);
1924 }
1925
1926 /*
1927 * Support for kevent() system call. Register EVFILT_READ filters and
1928 * reject all others.
1929 */
1930 int
bpfkqfilter(struct cdev * dev,struct knote * kn)1931 bpfkqfilter(struct cdev *dev, struct knote *kn)
1932 {
1933 struct bpf_d *d;
1934
1935 if (devfs_get_cdevpriv((void **)&d) != 0 ||
1936 kn->kn_filter != EVFILT_READ)
1937 return (1);
1938
1939 /*
1940 * Refresh PID associated with this descriptor.
1941 */
1942 BPFD_LOCK(d);
1943 BPF_PID_REFRESH_CUR(d);
1944 kn->kn_fop = &bpfread_filtops;
1945 kn->kn_hook = d;
1946 knlist_add(&d->bd_sel.si_note, kn, 1);
1947 BPFD_UNLOCK(d);
1948
1949 return (0);
1950 }
1951
1952 static void
filt_bpfdetach(struct knote * kn)1953 filt_bpfdetach(struct knote *kn)
1954 {
1955 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1956
1957 knlist_remove(&d->bd_sel.si_note, kn, 0);
1958 }
1959
1960 static int
filt_bpfread(struct knote * kn,long hint)1961 filt_bpfread(struct knote *kn, long hint)
1962 {
1963 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1964 int ready;
1965
1966 BPFD_LOCK_ASSERT(d);
1967 ready = bpf_ready(d);
1968 if (ready) {
1969 kn->kn_data = d->bd_slen;
1970 while (d->bd_hbuf_in_use)
1971 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1972 PRINET, "bd_hbuf", 0);
1973 if (d->bd_hbuf)
1974 kn->kn_data += d->bd_hlen;
1975 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1976 callout_reset(&d->bd_callout, d->bd_rtout,
1977 bpf_timed_out, d);
1978 d->bd_state = BPF_WAITING;
1979 }
1980
1981 return (ready);
1982 }
1983
1984 #define BPF_TSTAMP_NONE 0
1985 #define BPF_TSTAMP_FAST 1
1986 #define BPF_TSTAMP_NORMAL 2
1987 #define BPF_TSTAMP_EXTERN 3
1988
1989 static int
bpf_ts_quality(int tstype)1990 bpf_ts_quality(int tstype)
1991 {
1992
1993 if (tstype == BPF_T_NONE)
1994 return (BPF_TSTAMP_NONE);
1995 if ((tstype & BPF_T_FAST) != 0)
1996 return (BPF_TSTAMP_FAST);
1997
1998 return (BPF_TSTAMP_NORMAL);
1999 }
2000
2001 static int
bpf_gettime(struct bintime * bt,int tstype,struct mbuf * m)2002 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2003 {
2004 struct m_tag *tag;
2005 int quality;
2006
2007 quality = bpf_ts_quality(tstype);
2008 if (quality == BPF_TSTAMP_NONE)
2009 return (quality);
2010
2011 if (m != NULL) {
2012 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2013 if (tag != NULL) {
2014 *bt = *(struct bintime *)(tag + 1);
2015 return (BPF_TSTAMP_EXTERN);
2016 }
2017 }
2018 if (quality == BPF_TSTAMP_NORMAL)
2019 binuptime(bt);
2020 else
2021 getbinuptime(bt);
2022
2023 return (quality);
2024 }
2025
2026 /*
2027 * Incoming linkage from device drivers. Process the packet pkt, of length
2028 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2029 * by each process' filter, and if accepted, stashed into the corresponding
2030 * buffer.
2031 */
2032 void
bpf_tap(struct bpf_if * bp,u_char * pkt,u_int pktlen)2033 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2034 {
2035 struct bintime bt;
2036 struct bpf_d *d;
2037 #ifdef BPF_JITTER
2038 bpf_jit_filter *bf;
2039 #endif
2040 u_int slen;
2041 int gottime;
2042
2043 gottime = BPF_TSTAMP_NONE;
2044
2045 BPFIF_RLOCK(bp);
2046
2047 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2048 /*
2049 * We are not using any locks for d here because:
2050 * 1) any filter change is protected by interface
2051 * write lock
2052 * 2) destroying/detaching d is protected by interface
2053 * write lock, too
2054 */
2055
2056 /* XXX: Do not protect counter for the sake of performance. */
2057 ++d->bd_rcount;
2058 /*
2059 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2060 * way for the caller to indiciate to us whether this packet
2061 * is inbound or outbound. In the bpf_mtap() routines, we use
2062 * the interface pointers on the mbuf to figure it out.
2063 */
2064 #ifdef BPF_JITTER
2065 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2066 if (bf != NULL)
2067 slen = (*(bf->func))(pkt, pktlen, pktlen);
2068 else
2069 #endif
2070 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2071 if (slen != 0) {
2072 /*
2073 * Filter matches. Let's to acquire write lock.
2074 */
2075 BPFD_LOCK(d);
2076
2077 d->bd_fcount++;
2078 if (gottime < bpf_ts_quality(d->bd_tstamp))
2079 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2080 #ifdef MAC
2081 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2082 #endif
2083 catchpacket(d, pkt, pktlen, slen,
2084 bpf_append_bytes, &bt);
2085 BPFD_UNLOCK(d);
2086 }
2087 }
2088 BPFIF_RUNLOCK(bp);
2089 }
2090
2091 #define BPF_CHECK_DIRECTION(d, r, i) \
2092 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2093 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2094
2095 /*
2096 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2097 * Locking model is explained in bpf_tap().
2098 */
2099 void
bpf_mtap(struct bpf_if * bp,struct mbuf * m)2100 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2101 {
2102 struct bintime bt;
2103 struct bpf_d *d;
2104 #ifdef BPF_JITTER
2105 bpf_jit_filter *bf;
2106 #endif
2107 u_int pktlen, slen;
2108 int gottime;
2109
2110 /* Skip outgoing duplicate packets. */
2111 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2112 m->m_flags &= ~M_PROMISC;
2113 return;
2114 }
2115
2116 pktlen = m_length(m, NULL);
2117 gottime = BPF_TSTAMP_NONE;
2118
2119 BPFIF_RLOCK(bp);
2120
2121 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2122 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2123 continue;
2124 ++d->bd_rcount;
2125 #ifdef BPF_JITTER
2126 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2127 /* XXX We cannot handle multiple mbufs. */
2128 if (bf != NULL && m->m_next == NULL)
2129 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2130 else
2131 #endif
2132 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2133 if (slen != 0) {
2134 BPFD_LOCK(d);
2135
2136 d->bd_fcount++;
2137 if (gottime < bpf_ts_quality(d->bd_tstamp))
2138 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2139 #ifdef MAC
2140 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2141 #endif
2142 catchpacket(d, (u_char *)m, pktlen, slen,
2143 bpf_append_mbuf, &bt);
2144 BPFD_UNLOCK(d);
2145 }
2146 }
2147 BPFIF_RUNLOCK(bp);
2148 }
2149
2150 /*
2151 * Incoming linkage from device drivers, when packet is in
2152 * an mbuf chain and to be prepended by a contiguous header.
2153 */
2154 void
bpf_mtap2(struct bpf_if * bp,void * data,u_int dlen,struct mbuf * m)2155 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2156 {
2157 struct bintime bt;
2158 struct mbuf mb;
2159 struct bpf_d *d;
2160 u_int pktlen, slen;
2161 int gottime;
2162
2163 /* Skip outgoing duplicate packets. */
2164 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2165 m->m_flags &= ~M_PROMISC;
2166 return;
2167 }
2168
2169 pktlen = m_length(m, NULL);
2170 /*
2171 * Craft on-stack mbuf suitable for passing to bpf_filter.
2172 * Note that we cut corners here; we only setup what's
2173 * absolutely needed--this mbuf should never go anywhere else.
2174 */
2175 mb.m_next = m;
2176 mb.m_data = data;
2177 mb.m_len = dlen;
2178 pktlen += dlen;
2179
2180 gottime = BPF_TSTAMP_NONE;
2181
2182 BPFIF_RLOCK(bp);
2183
2184 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2185 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2186 continue;
2187 ++d->bd_rcount;
2188 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2189 if (slen != 0) {
2190 BPFD_LOCK(d);
2191
2192 d->bd_fcount++;
2193 if (gottime < bpf_ts_quality(d->bd_tstamp))
2194 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2195 #ifdef MAC
2196 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2197 #endif
2198 catchpacket(d, (u_char *)&mb, pktlen, slen,
2199 bpf_append_mbuf, &bt);
2200 BPFD_UNLOCK(d);
2201 }
2202 }
2203 BPFIF_RUNLOCK(bp);
2204 }
2205
2206 #undef BPF_CHECK_DIRECTION
2207
2208 #undef BPF_TSTAMP_NONE
2209 #undef BPF_TSTAMP_FAST
2210 #undef BPF_TSTAMP_NORMAL
2211 #undef BPF_TSTAMP_EXTERN
2212
2213 static int
bpf_hdrlen(struct bpf_d * d)2214 bpf_hdrlen(struct bpf_d *d)
2215 {
2216 int hdrlen;
2217
2218 hdrlen = d->bd_bif->bif_hdrlen;
2219 #ifndef BURN_BRIDGES
2220 if (d->bd_tstamp == BPF_T_NONE ||
2221 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2222 #ifdef COMPAT_FREEBSD32
2223 if (d->bd_compat32)
2224 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2225 else
2226 #endif
2227 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2228 else
2229 #endif
2230 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2231 #ifdef COMPAT_FREEBSD32
2232 if (d->bd_compat32)
2233 hdrlen = BPF_WORDALIGN32(hdrlen);
2234 else
2235 #endif
2236 hdrlen = BPF_WORDALIGN(hdrlen);
2237
2238 return (hdrlen - d->bd_bif->bif_hdrlen);
2239 }
2240
2241 static void
bpf_bintime2ts(struct bintime * bt,struct bpf_ts * ts,int tstype)2242 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2243 {
2244 struct bintime bt2;
2245 struct timeval tsm;
2246 struct timespec tsn;
2247
2248 if ((tstype & BPF_T_MONOTONIC) == 0) {
2249 bt2 = *bt;
2250 bintime_add(&bt2, &boottimebin);
2251 bt = &bt2;
2252 }
2253 switch (BPF_T_FORMAT(tstype)) {
2254 case BPF_T_MICROTIME:
2255 bintime2timeval(bt, &tsm);
2256 ts->bt_sec = tsm.tv_sec;
2257 ts->bt_frac = tsm.tv_usec;
2258 break;
2259 case BPF_T_NANOTIME:
2260 bintime2timespec(bt, &tsn);
2261 ts->bt_sec = tsn.tv_sec;
2262 ts->bt_frac = tsn.tv_nsec;
2263 break;
2264 case BPF_T_BINTIME:
2265 ts->bt_sec = bt->sec;
2266 ts->bt_frac = bt->frac;
2267 break;
2268 }
2269 }
2270
2271 /*
2272 * Move the packet data from interface memory (pkt) into the
2273 * store buffer. "cpfn" is the routine called to do the actual data
2274 * transfer. bcopy is passed in to copy contiguous chunks, while
2275 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2276 * pkt is really an mbuf.
2277 */
2278 static void
catchpacket(struct bpf_d * d,u_char * pkt,u_int pktlen,u_int snaplen,void (* cpfn)(struct bpf_d *,caddr_t,u_int,void *,u_int),struct bintime * bt)2279 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2280 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2281 struct bintime *bt)
2282 {
2283 struct bpf_xhdr hdr;
2284 #ifndef BURN_BRIDGES
2285 struct bpf_hdr hdr_old;
2286 #ifdef COMPAT_FREEBSD32
2287 struct bpf_hdr32 hdr32_old;
2288 #endif
2289 #endif
2290 int caplen, curlen, hdrlen, totlen;
2291 int do_wakeup = 0;
2292 int do_timestamp;
2293 int tstype;
2294
2295 BPFD_LOCK_ASSERT(d);
2296
2297 /*
2298 * Detect whether user space has released a buffer back to us, and if
2299 * so, move it from being a hold buffer to a free buffer. This may
2300 * not be the best place to do it (for example, we might only want to
2301 * run this check if we need the space), but for now it's a reliable
2302 * spot to do it.
2303 */
2304 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2305 while (d->bd_hbuf_in_use)
2306 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2307 PRINET, "bd_hbuf", 0);
2308 d->bd_fbuf = d->bd_hbuf;
2309 d->bd_hbuf = NULL;
2310 d->bd_hlen = 0;
2311 bpf_buf_reclaimed(d);
2312 }
2313
2314 /*
2315 * Figure out how many bytes to move. If the packet is
2316 * greater or equal to the snapshot length, transfer that
2317 * much. Otherwise, transfer the whole packet (unless
2318 * we hit the buffer size limit).
2319 */
2320 hdrlen = bpf_hdrlen(d);
2321 totlen = hdrlen + min(snaplen, pktlen);
2322 if (totlen > d->bd_bufsize)
2323 totlen = d->bd_bufsize;
2324
2325 /*
2326 * Round up the end of the previous packet to the next longword.
2327 *
2328 * Drop the packet if there's no room and no hope of room
2329 * If the packet would overflow the storage buffer or the storage
2330 * buffer is considered immutable by the buffer model, try to rotate
2331 * the buffer and wakeup pending processes.
2332 */
2333 #ifdef COMPAT_FREEBSD32
2334 if (d->bd_compat32)
2335 curlen = BPF_WORDALIGN32(d->bd_slen);
2336 else
2337 #endif
2338 curlen = BPF_WORDALIGN(d->bd_slen);
2339 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2340 if (d->bd_fbuf == NULL) {
2341 /*
2342 * There's no room in the store buffer, and no
2343 * prospect of room, so drop the packet. Notify the
2344 * buffer model.
2345 */
2346 bpf_buffull(d);
2347 ++d->bd_dcount;
2348 return;
2349 }
2350 while (d->bd_hbuf_in_use)
2351 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2352 PRINET, "bd_hbuf", 0);
2353 ROTATE_BUFFERS(d);
2354 do_wakeup = 1;
2355 curlen = 0;
2356 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2357 /*
2358 * Immediate mode is set, or the read timeout has already
2359 * expired during a select call. A packet arrived, so the
2360 * reader should be woken up.
2361 */
2362 do_wakeup = 1;
2363 caplen = totlen - hdrlen;
2364 tstype = d->bd_tstamp;
2365 do_timestamp = tstype != BPF_T_NONE;
2366 #ifndef BURN_BRIDGES
2367 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2368 struct bpf_ts ts;
2369 if (do_timestamp)
2370 bpf_bintime2ts(bt, &ts, tstype);
2371 #ifdef COMPAT_FREEBSD32
2372 if (d->bd_compat32) {
2373 bzero(&hdr32_old, sizeof(hdr32_old));
2374 if (do_timestamp) {
2375 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2376 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2377 }
2378 hdr32_old.bh_datalen = pktlen;
2379 hdr32_old.bh_hdrlen = hdrlen;
2380 hdr32_old.bh_caplen = caplen;
2381 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2382 sizeof(hdr32_old));
2383 goto copy;
2384 }
2385 #endif
2386 bzero(&hdr_old, sizeof(hdr_old));
2387 if (do_timestamp) {
2388 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2389 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2390 }
2391 hdr_old.bh_datalen = pktlen;
2392 hdr_old.bh_hdrlen = hdrlen;
2393 hdr_old.bh_caplen = caplen;
2394 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2395 sizeof(hdr_old));
2396 goto copy;
2397 }
2398 #endif
2399
2400 /*
2401 * Append the bpf header. Note we append the actual header size, but
2402 * move forward the length of the header plus padding.
2403 */
2404 bzero(&hdr, sizeof(hdr));
2405 if (do_timestamp)
2406 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2407 hdr.bh_datalen = pktlen;
2408 hdr.bh_hdrlen = hdrlen;
2409 hdr.bh_caplen = caplen;
2410 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2411
2412 /*
2413 * Copy the packet data into the store buffer and update its length.
2414 */
2415 #ifndef BURN_BRIDGES
2416 copy:
2417 #endif
2418 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2419 d->bd_slen = curlen + totlen;
2420
2421 if (do_wakeup)
2422 bpf_wakeup(d);
2423 }
2424
2425 /*
2426 * Free buffers currently in use by a descriptor.
2427 * Called on close.
2428 */
2429 static void
bpf_freed(struct bpf_d * d)2430 bpf_freed(struct bpf_d *d)
2431 {
2432
2433 /*
2434 * We don't need to lock out interrupts since this descriptor has
2435 * been detached from its interface and it yet hasn't been marked
2436 * free.
2437 */
2438 bpf_free(d);
2439 if (d->bd_rfilter != NULL) {
2440 free((caddr_t)d->bd_rfilter, M_BPF);
2441 #ifdef BPF_JITTER
2442 if (d->bd_bfilter != NULL)
2443 bpf_destroy_jit_filter(d->bd_bfilter);
2444 #endif
2445 }
2446 if (d->bd_wfilter != NULL)
2447 free((caddr_t)d->bd_wfilter, M_BPF);
2448 mtx_destroy(&d->bd_lock);
2449 }
2450
2451 /*
2452 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2453 * fixed size of the link header (variable length headers not yet supported).
2454 */
2455 void
bpfattach(struct ifnet * ifp,u_int dlt,u_int hdrlen)2456 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2457 {
2458
2459 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2460 }
2461
2462 /*
2463 * Attach an interface to bpf. ifp is a pointer to the structure
2464 * defining the interface to be attached, dlt is the link layer type,
2465 * and hdrlen is the fixed size of the link header (variable length
2466 * headers are not yet supporrted).
2467 */
2468 void
bpfattach2(struct ifnet * ifp,u_int dlt,u_int hdrlen,struct bpf_if ** driverp)2469 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2470 {
2471 struct bpf_if *bp;
2472
2473 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2474 if (bp == NULL)
2475 panic("bpfattach");
2476
2477 LIST_INIT(&bp->bif_dlist);
2478 LIST_INIT(&bp->bif_wlist);
2479 bp->bif_ifp = ifp;
2480 bp->bif_dlt = dlt;
2481 rw_init(&bp->bif_lock, "bpf interface lock");
2482 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2483 *driverp = bp;
2484
2485 BPF_LOCK();
2486 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2487 BPF_UNLOCK();
2488
2489 bp->bif_hdrlen = hdrlen;
2490
2491 if (bootverbose)
2492 if_printf(ifp, "bpf attached\n");
2493 }
2494
2495 /*
2496 * Detach bpf from an interface. This involves detaching each descriptor
2497 * associated with the interface. Notify each descriptor as it's detached
2498 * so that any sleepers wake up and get ENXIO.
2499 */
2500 void
bpfdetach(struct ifnet * ifp)2501 bpfdetach(struct ifnet *ifp)
2502 {
2503 struct bpf_if *bp, *bp_temp;
2504 struct bpf_d *d;
2505 int ndetached;
2506
2507 ndetached = 0;
2508
2509 BPF_LOCK();
2510 /* Find all bpf_if struct's which reference ifp and detach them. */
2511 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2512 if (ifp != bp->bif_ifp)
2513 continue;
2514
2515 LIST_REMOVE(bp, bif_next);
2516 /* Add to to-be-freed list */
2517 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next);
2518
2519 ndetached++;
2520 /*
2521 * Delay freeing bp till interface is detached
2522 * and all routes through this interface are removed.
2523 * Mark bp as detached to restrict new consumers.
2524 */
2525 BPFIF_WLOCK(bp);
2526 bp->flags |= BPFIF_FLAG_DYING;
2527 BPFIF_WUNLOCK(bp);
2528
2529 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p",
2530 __func__, bp->bif_dlt, bp, ifp);
2531
2532 /* Free common descriptors */
2533 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2534 bpf_detachd_locked(d);
2535 BPFD_LOCK(d);
2536 bpf_wakeup(d);
2537 BPFD_UNLOCK(d);
2538 }
2539
2540 /* Free writer-only descriptors */
2541 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2542 bpf_detachd_locked(d);
2543 BPFD_LOCK(d);
2544 bpf_wakeup(d);
2545 BPFD_UNLOCK(d);
2546 }
2547 }
2548 BPF_UNLOCK();
2549
2550 #ifdef INVARIANTS
2551 if (ndetached == 0)
2552 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2553 #endif
2554 }
2555
2556 /*
2557 * Interface departure handler.
2558 * Note departure event does not guarantee interface is going down.
2559 * Interface renaming is currently done via departure/arrival event set.
2560 *
2561 * Departure handled is called after all routes pointing to
2562 * given interface are removed and interface is in down state
2563 * restricting any packets to be sent/received. We assume it is now safe
2564 * to free data allocated by BPF.
2565 */
2566 static void
bpf_ifdetach(void * arg __unused,struct ifnet * ifp)2567 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2568 {
2569 struct bpf_if *bp, *bp_temp;
2570 int nmatched = 0;
2571
2572 BPF_LOCK();
2573 /*
2574 * Find matching entries in free list.
2575 * Nothing should be found if bpfdetach() was not called.
2576 */
2577 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) {
2578 if (ifp != bp->bif_ifp)
2579 continue;
2580
2581 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p",
2582 __func__, bp, ifp);
2583
2584 LIST_REMOVE(bp, bif_next);
2585
2586 rw_destroy(&bp->bif_lock);
2587 free(bp, M_BPF);
2588
2589 nmatched++;
2590 }
2591 BPF_UNLOCK();
2592
2593 /*
2594 * Note that we cannot zero other pointers to
2595 * custom DLTs possibly used by given interface.
2596 */
2597 if (nmatched != 0)
2598 ifp->if_bpf = NULL;
2599 }
2600
2601 /*
2602 * Get a list of available data link type of the interface.
2603 */
2604 static int
bpf_getdltlist(struct bpf_d * d,struct bpf_dltlist * bfl)2605 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2606 {
2607 int n, error;
2608 struct ifnet *ifp;
2609 struct bpf_if *bp;
2610
2611 BPF_LOCK_ASSERT();
2612
2613 ifp = d->bd_bif->bif_ifp;
2614 n = 0;
2615 error = 0;
2616 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2617 if (bp->bif_ifp != ifp)
2618 continue;
2619 if (bfl->bfl_list != NULL) {
2620 if (n >= bfl->bfl_len)
2621 return (ENOMEM);
2622 error = copyout(&bp->bif_dlt,
2623 bfl->bfl_list + n, sizeof(u_int));
2624 }
2625 n++;
2626 }
2627 bfl->bfl_len = n;
2628 return (error);
2629 }
2630
2631 /*
2632 * Set the data link type of a BPF instance.
2633 */
2634 static int
bpf_setdlt(struct bpf_d * d,u_int dlt)2635 bpf_setdlt(struct bpf_d *d, u_int dlt)
2636 {
2637 int error, opromisc;
2638 struct ifnet *ifp;
2639 struct bpf_if *bp;
2640
2641 BPF_LOCK_ASSERT();
2642
2643 if (d->bd_bif->bif_dlt == dlt)
2644 return (0);
2645 ifp = d->bd_bif->bif_ifp;
2646
2647 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2648 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2649 break;
2650 }
2651
2652 if (bp != NULL) {
2653 opromisc = d->bd_promisc;
2654 bpf_attachd(d, bp);
2655 BPFD_LOCK(d);
2656 reset_d(d);
2657 BPFD_UNLOCK(d);
2658 if (opromisc) {
2659 error = ifpromisc(bp->bif_ifp, 1);
2660 if (error)
2661 if_printf(bp->bif_ifp,
2662 "bpf_setdlt: ifpromisc failed (%d)\n",
2663 error);
2664 else
2665 d->bd_promisc = 1;
2666 }
2667 }
2668 return (bp == NULL ? EINVAL : 0);
2669 }
2670
2671 static void
bpf_drvinit(void * unused)2672 bpf_drvinit(void *unused)
2673 {
2674 struct cdev *dev;
2675
2676 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2677 LIST_INIT(&bpf_iflist);
2678 LIST_INIT(&bpf_freelist);
2679
2680 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2681 /* For compatibility */
2682 make_dev_alias(dev, "bpf0");
2683
2684 /* Register interface departure handler */
2685 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2686 ifnet_departure_event, bpf_ifdetach, NULL,
2687 EVENTHANDLER_PRI_ANY);
2688 }
2689
2690 /*
2691 * Zero out the various packet counters associated with all of the bpf
2692 * descriptors. At some point, we will probably want to get a bit more
2693 * granular and allow the user to specify descriptors to be zeroed.
2694 */
2695 static void
bpf_zero_counters(void)2696 bpf_zero_counters(void)
2697 {
2698 struct bpf_if *bp;
2699 struct bpf_d *bd;
2700
2701 BPF_LOCK();
2702 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2703 BPFIF_RLOCK(bp);
2704 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2705 BPFD_LOCK(bd);
2706 bd->bd_rcount = 0;
2707 bd->bd_dcount = 0;
2708 bd->bd_fcount = 0;
2709 bd->bd_wcount = 0;
2710 bd->bd_wfcount = 0;
2711 bd->bd_zcopy = 0;
2712 BPFD_UNLOCK(bd);
2713 }
2714 BPFIF_RUNLOCK(bp);
2715 }
2716 BPF_UNLOCK();
2717 }
2718
2719 /*
2720 * Fill filter statistics
2721 */
2722 static void
bpfstats_fill_xbpf(struct xbpf_d * d,struct bpf_d * bd)2723 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2724 {
2725
2726 bzero(d, sizeof(*d));
2727 BPFD_LOCK_ASSERT(bd);
2728 d->bd_structsize = sizeof(*d);
2729 /* XXX: reading should be protected by global lock */
2730 d->bd_immediate = bd->bd_immediate;
2731 d->bd_promisc = bd->bd_promisc;
2732 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2733 d->bd_direction = bd->bd_direction;
2734 d->bd_feedback = bd->bd_feedback;
2735 d->bd_async = bd->bd_async;
2736 d->bd_rcount = bd->bd_rcount;
2737 d->bd_dcount = bd->bd_dcount;
2738 d->bd_fcount = bd->bd_fcount;
2739 d->bd_sig = bd->bd_sig;
2740 d->bd_slen = bd->bd_slen;
2741 d->bd_hlen = bd->bd_hlen;
2742 d->bd_bufsize = bd->bd_bufsize;
2743 d->bd_pid = bd->bd_pid;
2744 strlcpy(d->bd_ifname,
2745 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2746 d->bd_locked = bd->bd_locked;
2747 d->bd_wcount = bd->bd_wcount;
2748 d->bd_wdcount = bd->bd_wdcount;
2749 d->bd_wfcount = bd->bd_wfcount;
2750 d->bd_zcopy = bd->bd_zcopy;
2751 d->bd_bufmode = bd->bd_bufmode;
2752 }
2753
2754 /*
2755 * Handle `netstat -B' stats request
2756 */
2757 static int
bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)2758 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2759 {
2760 static const struct xbpf_d zerostats;
2761 struct xbpf_d *xbdbuf, *xbd, tempstats;
2762 int index, error;
2763 struct bpf_if *bp;
2764 struct bpf_d *bd;
2765
2766 /*
2767 * XXX This is not technically correct. It is possible for non
2768 * privileged users to open bpf devices. It would make sense
2769 * if the users who opened the devices were able to retrieve
2770 * the statistics for them, too.
2771 */
2772 error = priv_check(req->td, PRIV_NET_BPF);
2773 if (error)
2774 return (error);
2775 /*
2776 * Check to see if the user is requesting that the counters be
2777 * zeroed out. Explicitly check that the supplied data is zeroed,
2778 * as we aren't allowing the user to set the counters currently.
2779 */
2780 if (req->newptr != NULL) {
2781 if (req->newlen != sizeof(tempstats))
2782 return (EINVAL);
2783 memset(&tempstats, 0, sizeof(tempstats));
2784 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2785 if (error)
2786 return (error);
2787 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2788 return (EINVAL);
2789 bpf_zero_counters();
2790 return (0);
2791 }
2792 if (req->oldptr == NULL)
2793 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2794 if (bpf_bpfd_cnt == 0)
2795 return (SYSCTL_OUT(req, 0, 0));
2796 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2797 BPF_LOCK();
2798 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2799 BPF_UNLOCK();
2800 free(xbdbuf, M_BPF);
2801 return (ENOMEM);
2802 }
2803 index = 0;
2804 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2805 BPFIF_RLOCK(bp);
2806 /* Send writers-only first */
2807 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2808 xbd = &xbdbuf[index++];
2809 BPFD_LOCK(bd);
2810 bpfstats_fill_xbpf(xbd, bd);
2811 BPFD_UNLOCK(bd);
2812 }
2813 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2814 xbd = &xbdbuf[index++];
2815 BPFD_LOCK(bd);
2816 bpfstats_fill_xbpf(xbd, bd);
2817 BPFD_UNLOCK(bd);
2818 }
2819 BPFIF_RUNLOCK(bp);
2820 }
2821 BPF_UNLOCK();
2822 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2823 free(xbdbuf, M_BPF);
2824 return (error);
2825 }
2826
2827 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2828
2829 #else /* !DEV_BPF && !NETGRAPH_BPF */
2830 /*
2831 * NOP stubs to allow bpf-using drivers to load and function.
2832 *
2833 * A 'better' implementation would allow the core bpf functionality
2834 * to be loaded at runtime.
2835 */
2836 static struct bpf_if bp_null;
2837
2838 void
bpf_tap(struct bpf_if * bp,u_char * pkt,u_int pktlen)2839 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2840 {
2841 }
2842
2843 void
bpf_mtap(struct bpf_if * bp,struct mbuf * m)2844 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2845 {
2846 }
2847
2848 void
bpf_mtap2(struct bpf_if * bp,void * d,u_int l,struct mbuf * m)2849 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2850 {
2851 }
2852
2853 void
bpfattach(struct ifnet * ifp,u_int dlt,u_int hdrlen)2854 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2855 {
2856
2857 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2858 }
2859
2860 void
bpfattach2(struct ifnet * ifp,u_int dlt,u_int hdrlen,struct bpf_if ** driverp)2861 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2862 {
2863
2864 *driverp = &bp_null;
2865 }
2866
2867 void
bpfdetach(struct ifnet * ifp)2868 bpfdetach(struct ifnet *ifp)
2869 {
2870 }
2871
2872 u_int
bpf_filter(const struct bpf_insn * pc,u_char * p,u_int wirelen,u_int buflen)2873 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2874 {
2875 return -1; /* "no filter" behaviour */
2876 }
2877
2878 int
bpf_validate(const struct bpf_insn * f,int len)2879 bpf_validate(const struct bpf_insn *f, int len)
2880 {
2881 return 0; /* false */
2882 }
2883
2884 #endif /* !DEV_BPF && !NETGRAPH_BPF */
2885