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 * Copyright (c) 2019 Andrey V. Elsukov <ae@FreeBSD.org>
7 *
8 * This code is derived from the Stanford/CMU enet packet filter,
9 * (net/enet.c) distributed as part of 4.3BSD, and code contributed
10 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
11 * Berkeley Laboratory.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 */
37
38 #include <sys/cdefs.h>
39 #include "opt_bpf.h"
40 #include "opt_netgraph.h"
41
42 #include <sys/param.h>
43 #include <sys/conf.h>
44 #include <sys/eventhandler.h>
45 #include <sys/fcntl.h>
46 #include <sys/jail.h>
47 #include <sys/ktr.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/mutex.h>
52 #include <sys/time.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/signalvar.h>
56 #include <sys/filio.h>
57 #include <sys/sockio.h>
58 #include <sys/ttycom.h>
59 #include <sys/uio.h>
60 #include <sys/sysent.h>
61 #include <sys/systm.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 #include <net/if_var.h>
72 #include <net/if_private.h>
73 #include <net/if_vlan_var.h>
74 #include <net/if_dl.h>
75 #include <net/bpf.h>
76 #include <net/bpf_buffer.h>
77 #ifdef BPF_JITTER
78 #include <net/bpf_jitter.h>
79 #endif
80 #include <net/bpf_zerocopy.h>
81 #include <net/bpfdesc.h>
82 #include <net/route.h>
83 #include <net/vnet.h>
84
85 #include <netinet/in.h>
86 #include <netinet/if_ether.h>
87 #include <sys/kernel.h>
88 #include <sys/sysctl.h>
89
90 #include <net80211/ieee80211_freebsd.h>
91
92 #include <security/mac/mac_framework.h>
93
94 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
95
96 static const struct bpf_if_ext dead_bpf_if = {
97 .bif_dlist = CK_LIST_HEAD_INITIALIZER()
98 };
99
100 struct bpf_if {
101 #define bif_next bif_ext.bif_next
102 #define bif_dlist bif_ext.bif_dlist
103 struct bpf_if_ext bif_ext; /* public members */
104 u_int bif_dlt; /* link layer type */
105 u_int bif_hdrlen; /* length of link header */
106 struct bpfd_list bif_wlist; /* writer-only list */
107 struct ifnet *bif_ifp; /* corresponding interface */
108 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
109 volatile u_int bif_refcnt;
110 struct epoch_context epoch_ctx;
111 };
112
113 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
114
115 struct bpf_program_buffer {
116 struct epoch_context epoch_ctx;
117 #ifdef BPF_JITTER
118 bpf_jit_filter *func;
119 #endif
120 void *buffer[0];
121 };
122
123 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
124
125 #define PRINET 26 /* interruptible */
126 #define BPF_PRIO_MAX 7
127
128 #define SIZEOF_BPF_HDR(type) \
129 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
130
131 #ifdef COMPAT_FREEBSD32
132 #include <sys/mount.h>
133 #include <compat/freebsd32/freebsd32.h>
134 #define BPF_ALIGNMENT32 sizeof(int32_t)
135 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
136
137 #ifndef BURN_BRIDGES
138 /*
139 * 32-bit version of structure prepended to each packet. We use this header
140 * instead of the standard one for 32-bit streams. We mark the a stream as
141 * 32-bit the first time we see a 32-bit compat ioctl request.
142 */
143 struct bpf_hdr32 {
144 struct timeval32 bh_tstamp; /* time stamp */
145 uint32_t bh_caplen; /* length of captured portion */
146 uint32_t bh_datalen; /* original length of packet */
147 uint16_t bh_hdrlen; /* length of bpf header (this struct
148 plus alignment padding) */
149 };
150 #endif
151
152 struct bpf_program32 {
153 u_int bf_len;
154 uint32_t bf_insns;
155 };
156
157 struct bpf_dltlist32 {
158 u_int bfl_len;
159 u_int bfl_list;
160 };
161
162 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
163 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
164 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
165 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
166 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
167 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
168 #endif
169
170 #define BPF_LOCK() sx_xlock(&bpf_sx)
171 #define BPF_UNLOCK() sx_xunlock(&bpf_sx)
172 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED)
173 /*
174 * bpf_iflist is a list of BPF interface structures, each corresponding to a
175 * specific DLT. The same network interface might have several BPF interface
176 * structures registered by different layers in the stack (i.e., 802.11
177 * frames, ethernet frames, etc).
178 */
179 CK_LIST_HEAD(bpf_iflist, bpf_if);
180 static struct bpf_iflist bpf_iflist = CK_LIST_HEAD_INITIALIZER();
181 static struct sx bpf_sx; /* bpf global lock */
182 static int bpf_bpfd_cnt;
183
184 static void bpfif_ref(struct bpf_if *);
185 static void bpfif_rele(struct bpf_if *);
186
187 static void bpfd_ref(struct bpf_d *);
188 static void bpfd_rele(struct bpf_d *);
189 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
190 static void bpf_detachd(struct bpf_d *, bool);
191 static void bpfd_free(epoch_context_t);
192 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
193 struct sockaddr *, int *, struct bpf_d *);
194 static int bpf_setif(struct bpf_d *, struct ifreq *);
195 static void bpf_timed_out(void *);
196 static __inline void
197 bpf_wakeup(struct bpf_d *);
198 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
199 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
200 struct bintime *);
201 static void reset_d(struct bpf_d *);
202 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
203 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
204 static int bpf_setdlt(struct bpf_d *, u_int);
205 static void filt_bpfdetach(struct knote *);
206 static int filt_bpfread(struct knote *, long);
207 static int filt_bpfwrite(struct knote *, long);
208 static void bpf_drvinit(void *);
209 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
210
211 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
212 "bpf sysctl");
213 int bpf_maxinsns = BPF_MAXINSNS;
214 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
215 &bpf_maxinsns, 0, "Maximum bpf program instructions");
216 static int bpf_zerocopy_enable = 0;
217 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
218 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
219 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
220 bpf_stats_sysctl, "bpf statistics portal");
221
222 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
223 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
224 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RWTUN,
225 &VNET_NAME(bpf_optimize_writers), 0,
226 "Do not send packets until BPF program is set");
227
228 static d_open_t bpfopen;
229 static d_read_t bpfread;
230 static d_write_t bpfwrite;
231 static d_ioctl_t bpfioctl;
232 static d_poll_t bpfpoll;
233 static d_kqfilter_t bpfkqfilter;
234
235 static struct cdevsw bpf_cdevsw = {
236 .d_version = D_VERSION,
237 .d_open = bpfopen,
238 .d_read = bpfread,
239 .d_write = bpfwrite,
240 .d_ioctl = bpfioctl,
241 .d_poll = bpfpoll,
242 .d_name = "bpf",
243 .d_kqfilter = bpfkqfilter,
244 };
245
246 static const struct filterops bpfread_filtops = {
247 .f_isfd = 1,
248 .f_detach = filt_bpfdetach,
249 .f_event = filt_bpfread,
250 .f_copy = knote_triv_copy,
251 };
252
253 static const struct filterops bpfwrite_filtops = {
254 .f_isfd = 1,
255 .f_detach = filt_bpfdetach,
256 .f_event = filt_bpfwrite,
257 .f_copy = knote_triv_copy,
258 };
259
260 /*
261 * LOCKING MODEL USED BY BPF
262 *
263 * Locks:
264 * 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
265 * every bpf_iflist changes, serializes ioctl access to bpf descriptors.
266 * 2) Descriptor lock. Mutex, used to protect BPF buffers and various
267 * structure fields used by bpf_*tap* code.
268 *
269 * Lock order: global lock, then descriptor lock.
270 *
271 * There are several possible consumers:
272 *
273 * 1. The kernel registers interface pointer with bpfattach().
274 * Each call allocates new bpf_if structure, references ifnet pointer
275 * and links bpf_if into bpf_iflist chain. This is protected with global
276 * lock.
277 *
278 * 2. An userland application uses ioctl() call to bpf_d descriptor.
279 * All such call are serialized with global lock. BPF filters can be
280 * changed, but pointer to old filter will be freed using NET_EPOCH_CALL().
281 * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
282 * filter pointers, even if change will happen during bpf_tap execution.
283 * Destroying of bpf_d descriptor also is doing using NET_EPOCH_CALL().
284 *
285 * 3. An userland application can write packets into bpf_d descriptor.
286 * There we need to be sure, that ifnet won't disappear during bpfwrite().
287 *
288 * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
289 * bif_dlist is protected with net_epoch_preempt section. So, it should
290 * be safe to make access to bpf_d descriptor inside the section.
291 *
292 * 5. The kernel invokes bpfdetach() on interface destroying. All lists
293 * are modified with global lock held and actual free() is done using
294 * NET_EPOCH_CALL().
295 */
296
297 static void
bpfif_free(epoch_context_t ctx)298 bpfif_free(epoch_context_t ctx)
299 {
300 struct bpf_if *bp;
301
302 bp = __containerof(ctx, struct bpf_if, epoch_ctx);
303 if_rele(bp->bif_ifp);
304 free(bp, M_BPF);
305 }
306
307 static void
bpfif_ref(struct bpf_if * bp)308 bpfif_ref(struct bpf_if *bp)
309 {
310
311 refcount_acquire(&bp->bif_refcnt);
312 }
313
314 static void
bpfif_rele(struct bpf_if * bp)315 bpfif_rele(struct bpf_if *bp)
316 {
317
318 if (!refcount_release(&bp->bif_refcnt))
319 return;
320 NET_EPOCH_CALL(bpfif_free, &bp->epoch_ctx);
321 }
322
323 static void
bpfd_ref(struct bpf_d * d)324 bpfd_ref(struct bpf_d *d)
325 {
326
327 refcount_acquire(&d->bd_refcnt);
328 }
329
330 static void
bpfd_rele(struct bpf_d * d)331 bpfd_rele(struct bpf_d *d)
332 {
333
334 if (!refcount_release(&d->bd_refcnt))
335 return;
336 NET_EPOCH_CALL(bpfd_free, &d->epoch_ctx);
337 }
338
339 static struct bpf_program_buffer*
bpf_program_buffer_alloc(size_t size,int flags)340 bpf_program_buffer_alloc(size_t size, int flags)
341 {
342
343 return (malloc(sizeof(struct bpf_program_buffer) + size,
344 M_BPF, flags));
345 }
346
347 static void
bpf_program_buffer_free(epoch_context_t ctx)348 bpf_program_buffer_free(epoch_context_t ctx)
349 {
350 struct bpf_program_buffer *ptr;
351
352 ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
353 #ifdef BPF_JITTER
354 if (ptr->func != NULL)
355 bpf_destroy_jit_filter(ptr->func);
356 #endif
357 free(ptr, M_BPF);
358 }
359
360 /*
361 * Wrapper functions for various buffering methods. If the set of buffer
362 * modes expands, we will probably want to introduce a switch data structure
363 * similar to protosw, et.
364 */
365 static void
bpf_append_bytes(struct bpf_d * d,caddr_t buf,u_int offset,void * src,u_int len)366 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
367 u_int len)
368 {
369
370 BPFD_LOCK_ASSERT(d);
371
372 switch (d->bd_bufmode) {
373 case BPF_BUFMODE_BUFFER:
374 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
375
376 case BPF_BUFMODE_ZBUF:
377 counter_u64_add(d->bd_zcopy, 1);
378 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
379
380 default:
381 panic("bpf_buf_append_bytes");
382 }
383 }
384
385 static void
bpf_append_mbuf(struct bpf_d * d,caddr_t buf,u_int offset,void * src,u_int len)386 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
387 u_int len)
388 {
389
390 BPFD_LOCK_ASSERT(d);
391
392 switch (d->bd_bufmode) {
393 case BPF_BUFMODE_BUFFER:
394 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
395
396 case BPF_BUFMODE_ZBUF:
397 counter_u64_add(d->bd_zcopy, 1);
398 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
399
400 default:
401 panic("bpf_buf_append_mbuf");
402 }
403 }
404
405 /*
406 * This function gets called when the free buffer is re-assigned.
407 */
408 static void
bpf_buf_reclaimed(struct bpf_d * d)409 bpf_buf_reclaimed(struct bpf_d *d)
410 {
411
412 BPFD_LOCK_ASSERT(d);
413
414 switch (d->bd_bufmode) {
415 case BPF_BUFMODE_BUFFER:
416 return;
417
418 case BPF_BUFMODE_ZBUF:
419 bpf_zerocopy_buf_reclaimed(d);
420 return;
421
422 default:
423 panic("bpf_buf_reclaimed");
424 }
425 }
426
427 /*
428 * If the buffer mechanism has a way to decide that a held buffer can be made
429 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
430 * returned if the buffer can be discarded, (0) is returned if it cannot.
431 */
432 static int
bpf_canfreebuf(struct bpf_d * d)433 bpf_canfreebuf(struct bpf_d *d)
434 {
435
436 BPFD_LOCK_ASSERT(d);
437
438 switch (d->bd_bufmode) {
439 case BPF_BUFMODE_ZBUF:
440 return (bpf_zerocopy_canfreebuf(d));
441 }
442 return (0);
443 }
444
445 /*
446 * Allow the buffer model to indicate that the current store buffer is
447 * immutable, regardless of the appearance of space. Return (1) if the
448 * buffer is writable, and (0) if not.
449 */
450 static int
bpf_canwritebuf(struct bpf_d * d)451 bpf_canwritebuf(struct bpf_d *d)
452 {
453 BPFD_LOCK_ASSERT(d);
454
455 switch (d->bd_bufmode) {
456 case BPF_BUFMODE_ZBUF:
457 return (bpf_zerocopy_canwritebuf(d));
458 }
459 return (1);
460 }
461
462 /*
463 * Notify buffer model that an attempt to write to the store buffer has
464 * resulted in a dropped packet, in which case the buffer may be considered
465 * full.
466 */
467 static void
bpf_buffull(struct bpf_d * d)468 bpf_buffull(struct bpf_d *d)
469 {
470
471 BPFD_LOCK_ASSERT(d);
472
473 switch (d->bd_bufmode) {
474 case BPF_BUFMODE_ZBUF:
475 bpf_zerocopy_buffull(d);
476 break;
477 }
478 }
479
480 /*
481 * Notify the buffer model that a buffer has moved into the hold position.
482 */
483 void
bpf_bufheld(struct bpf_d * d)484 bpf_bufheld(struct bpf_d *d)
485 {
486
487 BPFD_LOCK_ASSERT(d);
488
489 switch (d->bd_bufmode) {
490 case BPF_BUFMODE_ZBUF:
491 bpf_zerocopy_bufheld(d);
492 break;
493 }
494 }
495
496 static void
bpf_free(struct bpf_d * d)497 bpf_free(struct bpf_d *d)
498 {
499
500 switch (d->bd_bufmode) {
501 case BPF_BUFMODE_BUFFER:
502 return (bpf_buffer_free(d));
503
504 case BPF_BUFMODE_ZBUF:
505 return (bpf_zerocopy_free(d));
506
507 default:
508 panic("bpf_buf_free");
509 }
510 }
511
512 static int
bpf_uiomove(struct bpf_d * d,caddr_t buf,u_int len,struct uio * uio)513 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
514 {
515
516 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
517 return (EOPNOTSUPP);
518 return (bpf_buffer_uiomove(d, buf, len, uio));
519 }
520
521 static int
bpf_ioctl_sblen(struct bpf_d * d,u_int * i)522 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
523 {
524
525 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
526 return (EOPNOTSUPP);
527 return (bpf_buffer_ioctl_sblen(d, i));
528 }
529
530 static int
bpf_ioctl_getzmax(struct thread * td,struct bpf_d * d,size_t * i)531 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
532 {
533
534 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
535 return (EOPNOTSUPP);
536 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
537 }
538
539 static int
bpf_ioctl_rotzbuf(struct thread * td,struct bpf_d * d,struct bpf_zbuf * bz)540 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
541 {
542
543 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
544 return (EOPNOTSUPP);
545 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
546 }
547
548 static int
bpf_ioctl_setzbuf(struct thread * td,struct bpf_d * d,struct bpf_zbuf * bz)549 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
550 {
551
552 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
553 return (EOPNOTSUPP);
554 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
555 }
556
557 /*
558 * General BPF functions.
559 */
560 static int
bpf_movein(struct uio * uio,int linktype,struct ifnet * ifp,struct mbuf ** mp,struct sockaddr * sockp,int * hdrlen,struct bpf_d * d)561 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
562 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
563 {
564 const struct ieee80211_bpf_params *p;
565 struct ether_header *eh;
566 struct mbuf *m;
567 int error;
568 int len;
569 int hlen;
570 int slen;
571
572 /*
573 * Build a sockaddr based on the data link layer type.
574 * We do this at this level because the ethernet header
575 * is copied directly into the data field of the sockaddr.
576 * In the case of SLIP, there is no header and the packet
577 * is forwarded as is.
578 * Also, we are careful to leave room at the front of the mbuf
579 * for the link level header.
580 */
581 switch (linktype) {
582 case DLT_SLIP:
583 sockp->sa_family = AF_INET;
584 hlen = 0;
585 break;
586
587 case DLT_EN10MB:
588 sockp->sa_family = AF_UNSPEC;
589 /* XXX Would MAXLINKHDR be better? */
590 hlen = ETHER_HDR_LEN;
591 break;
592
593 case DLT_FDDI:
594 sockp->sa_family = AF_IMPLINK;
595 hlen = 0;
596 break;
597
598 case DLT_RAW:
599 sockp->sa_family = AF_UNSPEC;
600 hlen = 0;
601 break;
602
603 case DLT_NULL:
604 /*
605 * null interface types require a 4 byte pseudo header which
606 * corresponds to the address family of the packet.
607 */
608 sockp->sa_family = AF_UNSPEC;
609 hlen = 4;
610 break;
611
612 case DLT_ATM_RFC1483:
613 /*
614 * en atm driver requires 4-byte atm pseudo header.
615 * though it isn't standard, vpi:vci needs to be
616 * specified anyway.
617 */
618 sockp->sa_family = AF_UNSPEC;
619 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
620 break;
621
622 case DLT_PPP:
623 sockp->sa_family = AF_UNSPEC;
624 hlen = 4; /* This should match PPP_HDRLEN */
625 break;
626
627 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
628 sockp->sa_family = AF_IEEE80211;
629 hlen = 0;
630 break;
631
632 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
633 sockp->sa_family = AF_IEEE80211;
634 sockp->sa_len = 12; /* XXX != 0 */
635 hlen = sizeof(struct ieee80211_bpf_params);
636 break;
637
638 default:
639 return (EIO);
640 }
641
642 len = uio->uio_resid;
643 if (len < hlen || len - hlen > ifp->if_mtu)
644 return (EMSGSIZE);
645
646 /* Allocate a mbuf, up to MJUM16BYTES bytes, for our write. */
647 m = m_get3(len, M_WAITOK, MT_DATA, M_PKTHDR);
648 if (m == NULL)
649 return (EIO);
650 m->m_pkthdr.len = m->m_len = len;
651 *mp = m;
652
653 error = uiomove(mtod(m, u_char *), len, uio);
654 if (error)
655 goto bad;
656
657 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
658 if (slen == 0) {
659 error = EPERM;
660 goto bad;
661 }
662
663 /* Check for multicast destination */
664 switch (linktype) {
665 case DLT_EN10MB:
666 eh = mtod(m, struct ether_header *);
667 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
668 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
669 ETHER_ADDR_LEN) == 0)
670 m->m_flags |= M_BCAST;
671 else
672 m->m_flags |= M_MCAST;
673 }
674 if (d->bd_hdrcmplt == 0) {
675 memcpy(eh->ether_shost, IF_LLADDR(ifp),
676 sizeof(eh->ether_shost));
677 }
678 break;
679 }
680
681 /*
682 * Make room for link header, and copy it to sockaddr
683 */
684 if (hlen != 0) {
685 if (sockp->sa_family == AF_IEEE80211) {
686 /*
687 * Collect true length from the parameter header
688 * NB: sockp is known to be zero'd so if we do a
689 * short copy unspecified parameters will be
690 * zero.
691 * NB: packet may not be aligned after stripping
692 * bpf params
693 * XXX check ibp_vers
694 */
695 p = mtod(m, const struct ieee80211_bpf_params *);
696 hlen = p->ibp_len;
697 if (hlen > sizeof(sockp->sa_data)) {
698 error = EINVAL;
699 goto bad;
700 }
701 }
702 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
703 }
704 *hdrlen = hlen;
705
706 return (0);
707 bad:
708 m_freem(m);
709 return (error);
710 }
711
712 /*
713 * Attach descriptor to the bpf interface, i.e. make d listen on bp,
714 * then reset its buffers and counters with reset_d().
715 */
716 static void
bpf_attachd(struct bpf_d * d,struct bpf_if * bp)717 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
718 {
719 int op_w;
720
721 BPF_LOCK_ASSERT();
722
723 /*
724 * Save sysctl value to protect from sysctl change
725 * between reads
726 */
727 op_w = V_bpf_optimize_writers || d->bd_writer;
728
729 if (d->bd_bif != NULL)
730 bpf_detachd(d, false);
731 /*
732 * Point d at bp, and add d to the interface's list.
733 * Since there are many applications using BPF for
734 * sending raw packets only (dhcpd, cdpd are good examples)
735 * we can delay adding d to the list of active listeners until
736 * some filter is configured.
737 */
738
739 BPFD_LOCK(d);
740 /*
741 * Hold reference to bpif while descriptor uses this interface.
742 */
743 bpfif_ref(bp);
744 d->bd_bif = bp;
745 if (op_w != 0) {
746 /* Add to writers-only list */
747 CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
748 /*
749 * We decrement bd_writer on every filter set operation.
750 * First BIOCSETF is done by pcap_open_live() to set up
751 * snap length. After that appliation usually sets its own
752 * filter.
753 */
754 d->bd_writer = 2;
755 } else
756 CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
757
758 reset_d(d);
759
760 /* Trigger EVFILT_WRITE events. */
761 bpf_wakeup(d);
762
763 BPFD_UNLOCK(d);
764 bpf_bpfd_cnt++;
765
766 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
767 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
768
769 if (op_w == 0)
770 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
771 }
772
773 /*
774 * Check if we need to upgrade our descriptor @d from write-only mode.
775 */
776 static int
bpf_check_upgrade(u_long cmd,struct bpf_d * d,struct bpf_insn * fcode,int flen)777 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
778 int flen)
779 {
780 int is_snap, need_upgrade;
781
782 /*
783 * Check if we've already upgraded or new filter is empty.
784 */
785 if (d->bd_writer == 0 || fcode == NULL)
786 return (0);
787
788 need_upgrade = 0;
789
790 /*
791 * Check if cmd looks like snaplen setting from
792 * pcap_bpf.c:pcap_open_live().
793 * Note we're not checking .k value here:
794 * while pcap_open_live() definitely sets to non-zero value,
795 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
796 * do not consider upgrading immediately
797 */
798 if (cmd == BIOCSETF && flen == 1 &&
799 fcode[0].code == (BPF_RET | BPF_K))
800 is_snap = 1;
801 else
802 is_snap = 0;
803
804 if (is_snap == 0) {
805 /*
806 * We're setting first filter and it doesn't look like
807 * setting snaplen. We're probably using bpf directly.
808 * Upgrade immediately.
809 */
810 need_upgrade = 1;
811 } else {
812 /*
813 * Do not require upgrade by first BIOCSETF
814 * (used to set snaplen) by pcap_open_live().
815 */
816
817 if (--d->bd_writer == 0) {
818 /*
819 * First snaplen filter has already
820 * been set. This is probably catch-all
821 * filter
822 */
823 need_upgrade = 1;
824 }
825 }
826
827 CTR5(KTR_NET,
828 "%s: filter function set by pid %d, "
829 "bd_writer counter %d, snap %d upgrade %d",
830 __func__, d->bd_pid, d->bd_writer,
831 is_snap, need_upgrade);
832
833 return (need_upgrade);
834 }
835
836 /*
837 * Detach a file from its interface.
838 */
839 static void
bpf_detachd(struct bpf_d * d,bool detached_ifp)840 bpf_detachd(struct bpf_d *d, bool detached_ifp)
841 {
842 struct bpf_if *bp;
843 struct ifnet *ifp;
844 int error;
845
846 BPF_LOCK_ASSERT();
847 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
848
849 /* Check if descriptor is attached */
850 if ((bp = d->bd_bif) == NULL)
851 return;
852
853 BPFD_LOCK(d);
854 /* Remove d from the interface's descriptor list. */
855 CK_LIST_REMOVE(d, bd_next);
856 /* Save bd_writer value */
857 error = d->bd_writer;
858 ifp = bp->bif_ifp;
859 d->bd_bif = NULL;
860 if (detached_ifp) {
861 /*
862 * Notify descriptor as it's detached, so that any
863 * sleepers wake up and get ENXIO.
864 */
865 bpf_wakeup(d);
866 }
867 BPFD_UNLOCK(d);
868 bpf_bpfd_cnt--;
869
870 /* Call event handler iff d is attached */
871 if (error == 0)
872 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
873
874 /*
875 * Check if this descriptor had requested promiscuous mode.
876 * If so and ifnet is not detached, turn it off.
877 */
878 if (d->bd_promisc && !detached_ifp) {
879 d->bd_promisc = 0;
880 CURVNET_SET(ifp->if_vnet);
881 error = ifpromisc(ifp, 0);
882 CURVNET_RESTORE();
883 if (error != 0 && error != ENXIO) {
884 /*
885 * ENXIO can happen if a pccard is unplugged
886 * Something is really wrong if we were able to put
887 * the driver into promiscuous mode, but can't
888 * take it out.
889 */
890 if_printf(bp->bif_ifp,
891 "bpf_detach: ifpromisc failed (%d)\n", error);
892 }
893 }
894 bpfif_rele(bp);
895 }
896
897 /*
898 * Close the descriptor by detaching it from its interface,
899 * deallocating its buffers, and marking it free.
900 */
901 static void
bpf_dtor(void * data)902 bpf_dtor(void *data)
903 {
904 struct bpf_d *d = data;
905
906 BPFD_LOCK(d);
907 if (d->bd_state == BPF_WAITING)
908 callout_stop(&d->bd_callout);
909 d->bd_state = BPF_IDLE;
910 BPFD_UNLOCK(d);
911 funsetown(&d->bd_sigio);
912 BPF_LOCK();
913 bpf_detachd(d, false);
914 BPF_UNLOCK();
915 #ifdef MAC
916 mac_bpfdesc_destroy(d);
917 #endif /* MAC */
918 seldrain(&d->bd_sel);
919 knlist_destroy(&d->bd_sel.si_note);
920 callout_drain(&d->bd_callout);
921 bpfd_rele(d);
922 }
923
924 /*
925 * Open ethernet device. Returns ENXIO for illegal minor device number,
926 * EBUSY if file is open by another process.
927 */
928 /* ARGSUSED */
929 static int
bpfopen(struct cdev * dev,int flags,int fmt,struct thread * td)930 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
931 {
932 struct bpf_d *d;
933 int error;
934
935 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
936 error = devfs_set_cdevpriv(d, bpf_dtor);
937 if (error != 0) {
938 free(d, M_BPF);
939 return (error);
940 }
941
942 /* Setup counters */
943 d->bd_rcount = counter_u64_alloc(M_WAITOK);
944 d->bd_dcount = counter_u64_alloc(M_WAITOK);
945 d->bd_fcount = counter_u64_alloc(M_WAITOK);
946 d->bd_wcount = counter_u64_alloc(M_WAITOK);
947 d->bd_wfcount = counter_u64_alloc(M_WAITOK);
948 d->bd_wdcount = counter_u64_alloc(M_WAITOK);
949 d->bd_zcopy = counter_u64_alloc(M_WAITOK);
950
951 /*
952 * For historical reasons, perform a one-time initialization call to
953 * the buffer routines, even though we're not yet committed to a
954 * particular buffer method.
955 */
956 bpf_buffer_init(d);
957 if ((flags & FREAD) == 0)
958 d->bd_writer = 2;
959 d->bd_hbuf_in_use = 0;
960 d->bd_bufmode = BPF_BUFMODE_BUFFER;
961 d->bd_sig = SIGIO;
962 d->bd_direction = BPF_D_INOUT;
963 refcount_init(&d->bd_refcnt, 1);
964 BPF_PID_REFRESH(d, td);
965 #ifdef MAC
966 mac_bpfdesc_init(d);
967 mac_bpfdesc_create(td->td_ucred, d);
968 #endif
969 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
970 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
971 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
972
973 /* Disable VLAN pcp tagging. */
974 d->bd_pcp = 0;
975
976 return (0);
977 }
978
979 /*
980 * bpfread - read next chunk of packets from buffers
981 */
982 static int
bpfread(struct cdev * dev,struct uio * uio,int ioflag)983 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
984 {
985 struct bpf_d *d;
986 int error;
987 int non_block;
988 int timed_out;
989
990 error = devfs_get_cdevpriv((void **)&d);
991 if (error != 0)
992 return (error);
993
994 /*
995 * Restrict application to use a buffer the same size as
996 * as kernel buffers.
997 */
998 if (uio->uio_resid != d->bd_bufsize)
999 return (EINVAL);
1000
1001 non_block = ((ioflag & O_NONBLOCK) != 0);
1002
1003 BPFD_LOCK(d);
1004 BPF_PID_REFRESH_CUR(d);
1005 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1006 BPFD_UNLOCK(d);
1007 return (EOPNOTSUPP);
1008 }
1009 if (d->bd_state == BPF_WAITING)
1010 callout_stop(&d->bd_callout);
1011 timed_out = (d->bd_state == BPF_TIMED_OUT);
1012 d->bd_state = BPF_IDLE;
1013 while (d->bd_hbuf_in_use) {
1014 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1015 PRINET | PCATCH, "bd_hbuf", 0);
1016 if (error != 0) {
1017 BPFD_UNLOCK(d);
1018 return (error);
1019 }
1020 }
1021 /*
1022 * If the hold buffer is empty, then do a timed sleep, which
1023 * ends when the timeout expires or when enough packets
1024 * have arrived to fill the store buffer.
1025 */
1026 while (d->bd_hbuf == NULL) {
1027 if (d->bd_slen != 0) {
1028 /*
1029 * A packet(s) either arrived since the previous
1030 * read or arrived while we were asleep.
1031 */
1032 if (d->bd_immediate || non_block || timed_out) {
1033 /*
1034 * Rotate the buffers and return what's here
1035 * if we are in immediate mode, non-blocking
1036 * flag is set, or this descriptor timed out.
1037 */
1038 ROTATE_BUFFERS(d);
1039 break;
1040 }
1041 }
1042
1043 /*
1044 * No data is available, check to see if the bpf device
1045 * is still pointed at a real interface. If not, return
1046 * ENXIO so that the userland process knows to rebind
1047 * it before using it again.
1048 */
1049 if (d->bd_bif == NULL) {
1050 BPFD_UNLOCK(d);
1051 return (ENXIO);
1052 }
1053
1054 if (non_block) {
1055 BPFD_UNLOCK(d);
1056 return (EWOULDBLOCK);
1057 }
1058 error = msleep(d, &d->bd_lock, PRINET | PCATCH,
1059 "bpf", d->bd_rtout);
1060 if (error == EINTR || error == ERESTART) {
1061 BPFD_UNLOCK(d);
1062 return (error);
1063 }
1064 if (error == EWOULDBLOCK) {
1065 /*
1066 * On a timeout, return what's in the buffer,
1067 * which may be nothing. If there is something
1068 * in the store buffer, we can rotate the buffers.
1069 */
1070 if (d->bd_hbuf)
1071 /*
1072 * We filled up the buffer in between
1073 * getting the timeout and arriving
1074 * here, so we don't need to rotate.
1075 */
1076 break;
1077
1078 if (d->bd_slen == 0) {
1079 BPFD_UNLOCK(d);
1080 return (0);
1081 }
1082 ROTATE_BUFFERS(d);
1083 break;
1084 }
1085 }
1086 /*
1087 * At this point, we know we have something in the hold slot.
1088 */
1089 d->bd_hbuf_in_use = 1;
1090 BPFD_UNLOCK(d);
1091
1092 /*
1093 * Move data from hold buffer into user space.
1094 * We know the entire buffer is transferred since
1095 * we checked above that the read buffer is bpf_bufsize bytes.
1096 *
1097 * We do not have to worry about simultaneous reads because
1098 * we waited for sole access to the hold buffer above.
1099 */
1100 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1101
1102 BPFD_LOCK(d);
1103 if (d->bd_hbuf_in_use) {
1104 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1105 d->bd_fbuf = d->bd_hbuf;
1106 d->bd_hbuf = NULL;
1107 d->bd_hlen = 0;
1108 bpf_buf_reclaimed(d);
1109 d->bd_hbuf_in_use = 0;
1110 wakeup(&d->bd_hbuf_in_use);
1111 }
1112 BPFD_UNLOCK(d);
1113
1114 return (error);
1115 }
1116
1117 /*
1118 * If there are processes sleeping on this descriptor, wake them up.
1119 */
1120 static __inline void
bpf_wakeup(struct bpf_d * d)1121 bpf_wakeup(struct bpf_d *d)
1122 {
1123
1124 BPFD_LOCK_ASSERT(d);
1125 if (d->bd_state == BPF_WAITING) {
1126 callout_stop(&d->bd_callout);
1127 d->bd_state = BPF_IDLE;
1128 }
1129 wakeup(d);
1130 if (d->bd_async && d->bd_sig && d->bd_sigio)
1131 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1132
1133 selwakeuppri(&d->bd_sel, PRINET);
1134 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1135 }
1136
1137 static void
bpf_timed_out(void * arg)1138 bpf_timed_out(void *arg)
1139 {
1140 struct bpf_d *d = (struct bpf_d *)arg;
1141
1142 BPFD_LOCK_ASSERT(d);
1143
1144 if (callout_pending(&d->bd_callout) ||
1145 !callout_active(&d->bd_callout))
1146 return;
1147 if (d->bd_state == BPF_WAITING) {
1148 d->bd_state = BPF_TIMED_OUT;
1149 if (d->bd_slen != 0)
1150 bpf_wakeup(d);
1151 }
1152 }
1153
1154 static int
bpf_ready(struct bpf_d * d)1155 bpf_ready(struct bpf_d *d)
1156 {
1157
1158 BPFD_LOCK_ASSERT(d);
1159
1160 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1161 return (1);
1162 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1163 d->bd_slen != 0)
1164 return (1);
1165 return (0);
1166 }
1167
1168 static int
bpfwrite(struct cdev * dev,struct uio * uio,int ioflag)1169 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1170 {
1171 struct route ro;
1172 struct sockaddr dst;
1173 struct epoch_tracker et;
1174 struct bpf_if *bp;
1175 struct bpf_d *d;
1176 struct ifnet *ifp;
1177 struct mbuf *m, *mc;
1178 int error, hlen;
1179
1180 error = devfs_get_cdevpriv((void **)&d);
1181 if (error != 0)
1182 return (error);
1183
1184 NET_EPOCH_ENTER(et);
1185 BPFD_LOCK(d);
1186 BPF_PID_REFRESH_CUR(d);
1187 counter_u64_add(d->bd_wcount, 1);
1188 if ((bp = d->bd_bif) == NULL) {
1189 error = ENXIO;
1190 goto out_locked;
1191 }
1192
1193 ifp = bp->bif_ifp;
1194 if ((ifp->if_flags & IFF_UP) == 0) {
1195 error = ENETDOWN;
1196 goto out_locked;
1197 }
1198
1199 if (uio->uio_resid == 0)
1200 goto out_locked;
1201
1202 bzero(&dst, sizeof(dst));
1203 m = NULL;
1204 hlen = 0;
1205
1206 /*
1207 * Take extra reference, unlock d and exit from epoch section,
1208 * since bpf_movein() can sleep.
1209 */
1210 bpfd_ref(d);
1211 NET_EPOCH_EXIT(et);
1212 BPFD_UNLOCK(d);
1213
1214 error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1215 &m, &dst, &hlen, d);
1216
1217 if (error != 0) {
1218 counter_u64_add(d->bd_wdcount, 1);
1219 bpfd_rele(d);
1220 return (error);
1221 }
1222
1223 BPFD_LOCK(d);
1224 /*
1225 * Check that descriptor is still attached to the interface.
1226 * This can happen on bpfdetach(). To avoid access to detached
1227 * ifnet, free mbuf and return ENXIO.
1228 */
1229 if (d->bd_bif == NULL) {
1230 counter_u64_add(d->bd_wdcount, 1);
1231 BPFD_UNLOCK(d);
1232 bpfd_rele(d);
1233 m_freem(m);
1234 return (ENXIO);
1235 }
1236 counter_u64_add(d->bd_wfcount, 1);
1237 if (d->bd_hdrcmplt)
1238 dst.sa_family = pseudo_AF_HDRCMPLT;
1239
1240 if (d->bd_feedback) {
1241 mc = m_dup(m, M_NOWAIT);
1242 if (mc != NULL)
1243 mc->m_pkthdr.rcvif = ifp;
1244 /* Set M_PROMISC for outgoing packets to be discarded. */
1245 if (d->bd_direction == BPF_D_INOUT)
1246 m->m_flags |= M_PROMISC;
1247 } else
1248 mc = NULL;
1249
1250 m->m_pkthdr.len -= hlen;
1251 m->m_len -= hlen;
1252 m->m_data += hlen; /* XXX */
1253
1254 CURVNET_SET(ifp->if_vnet);
1255 #ifdef MAC
1256 mac_bpfdesc_create_mbuf(d, m);
1257 if (mc != NULL)
1258 mac_bpfdesc_create_mbuf(d, mc);
1259 #endif
1260
1261 bzero(&ro, sizeof(ro));
1262 if (hlen != 0) {
1263 ro.ro_prepend = (u_char *)&dst.sa_data;
1264 ro.ro_plen = hlen;
1265 ro.ro_flags = RT_HAS_HEADER;
1266 }
1267
1268 if (d->bd_pcp != 0)
1269 vlan_set_pcp(m, d->bd_pcp);
1270
1271 /* Avoid possible recursion on BPFD_LOCK(). */
1272 NET_EPOCH_ENTER(et);
1273 BPFD_UNLOCK(d);
1274 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1275 if (error)
1276 counter_u64_add(d->bd_wdcount, 1);
1277
1278 if (mc != NULL) {
1279 if (error == 0)
1280 (*ifp->if_input)(ifp, mc);
1281 else
1282 m_freem(mc);
1283 }
1284 NET_EPOCH_EXIT(et);
1285 CURVNET_RESTORE();
1286 bpfd_rele(d);
1287 return (error);
1288
1289 out_locked:
1290 counter_u64_add(d->bd_wdcount, 1);
1291 NET_EPOCH_EXIT(et);
1292 BPFD_UNLOCK(d);
1293 return (error);
1294 }
1295
1296 /*
1297 * Reset a descriptor by flushing its packet buffer and clearing the receive
1298 * and drop counts. This is doable for kernel-only buffers, but with
1299 * zero-copy buffers, we can't write to (or rotate) buffers that are
1300 * currently owned by userspace. It would be nice if we could encapsulate
1301 * this logic in the buffer code rather than here.
1302 */
1303 static void
reset_d(struct bpf_d * d)1304 reset_d(struct bpf_d *d)
1305 {
1306
1307 BPFD_LOCK_ASSERT(d);
1308
1309 while (d->bd_hbuf_in_use)
1310 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1311 "bd_hbuf", 0);
1312 if ((d->bd_hbuf != NULL) &&
1313 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1314 /* Free the hold buffer. */
1315 d->bd_fbuf = d->bd_hbuf;
1316 d->bd_hbuf = NULL;
1317 d->bd_hlen = 0;
1318 bpf_buf_reclaimed(d);
1319 }
1320 if (bpf_canwritebuf(d))
1321 d->bd_slen = 0;
1322 counter_u64_zero(d->bd_rcount);
1323 counter_u64_zero(d->bd_dcount);
1324 counter_u64_zero(d->bd_fcount);
1325 counter_u64_zero(d->bd_wcount);
1326 counter_u64_zero(d->bd_wfcount);
1327 counter_u64_zero(d->bd_wdcount);
1328 counter_u64_zero(d->bd_zcopy);
1329 }
1330
1331 /*
1332 * FIONREAD Check for read packet available.
1333 * BIOCGBLEN Get buffer len [for read()].
1334 * BIOCSETF Set read filter.
1335 * BIOCSETFNR Set read filter without resetting descriptor.
1336 * BIOCSETWF Set write filter.
1337 * BIOCFLUSH Flush read packet buffer.
1338 * BIOCPROMISC Put interface into promiscuous mode.
1339 * BIOCGDLT Get link layer type.
1340 * BIOCGETIF Get interface name.
1341 * BIOCSETIF Set interface.
1342 * BIOCSRTIMEOUT Set read timeout.
1343 * BIOCGRTIMEOUT Get read timeout.
1344 * BIOCGSTATS Get packet stats.
1345 * BIOCIMMEDIATE Set immediate mode.
1346 * BIOCVERSION Get filter language version.
1347 * BIOCGHDRCMPLT Get "header already complete" flag
1348 * BIOCSHDRCMPLT Set "header already complete" flag
1349 * BIOCGDIRECTION Get packet direction flag
1350 * BIOCSDIRECTION Set packet direction flag
1351 * BIOCGTSTAMP Get time stamp format and resolution.
1352 * BIOCSTSTAMP Set time stamp format and resolution.
1353 * BIOCLOCK Set "locked" flag
1354 * BIOCFEEDBACK Set packet feedback mode.
1355 * BIOCSETZBUF Set current zero-copy buffer locations.
1356 * BIOCGETZMAX Get maximum zero-copy buffer size.
1357 * BIOCROTZBUF Force rotation of zero-copy buffer
1358 * BIOCSETBUFMODE Set buffer mode.
1359 * BIOCGETBUFMODE Get current buffer mode.
1360 * BIOCSETVLANPCP Set VLAN PCP tag.
1361 */
1362 /* ARGSUSED */
1363 static int
bpfioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flags,struct thread * td)1364 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1365 struct thread *td)
1366 {
1367 struct bpf_d *d;
1368 int error;
1369
1370 error = devfs_get_cdevpriv((void **)&d);
1371 if (error != 0)
1372 return (error);
1373
1374 /*
1375 * Refresh PID associated with this descriptor.
1376 */
1377 BPFD_LOCK(d);
1378 BPF_PID_REFRESH(d, td);
1379 if (d->bd_state == BPF_WAITING)
1380 callout_stop(&d->bd_callout);
1381 d->bd_state = BPF_IDLE;
1382 BPFD_UNLOCK(d);
1383
1384 if (d->bd_locked == 1) {
1385 switch (cmd) {
1386 case BIOCGBLEN:
1387 case BIOCFLUSH:
1388 case BIOCGDLT:
1389 case BIOCGDLTLIST:
1390 #ifdef COMPAT_FREEBSD32
1391 case BIOCGDLTLIST32:
1392 #endif
1393 case BIOCGETIF:
1394 case BIOCGRTIMEOUT:
1395 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1396 case BIOCGRTIMEOUT32:
1397 #endif
1398 case BIOCGSTATS:
1399 case BIOCVERSION:
1400 case BIOCGRSIG:
1401 case BIOCGHDRCMPLT:
1402 case BIOCSTSTAMP:
1403 case BIOCFEEDBACK:
1404 case FIONREAD:
1405 case BIOCLOCK:
1406 case BIOCSRTIMEOUT:
1407 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1408 case BIOCSRTIMEOUT32:
1409 #endif
1410 case BIOCIMMEDIATE:
1411 case TIOCGPGRP:
1412 case BIOCROTZBUF:
1413 break;
1414 default:
1415 return (EPERM);
1416 }
1417 }
1418 #ifdef COMPAT_FREEBSD32
1419 /*
1420 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1421 * that it will get 32-bit packet headers.
1422 */
1423 switch (cmd) {
1424 case BIOCSETF32:
1425 case BIOCSETFNR32:
1426 case BIOCSETWF32:
1427 case BIOCGDLTLIST32:
1428 case BIOCGRTIMEOUT32:
1429 case BIOCSRTIMEOUT32:
1430 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1431 BPFD_LOCK(d);
1432 d->bd_compat32 = 1;
1433 BPFD_UNLOCK(d);
1434 }
1435 }
1436 #endif
1437
1438 CURVNET_SET(TD_TO_VNET(td));
1439 switch (cmd) {
1440 default:
1441 error = EINVAL;
1442 break;
1443
1444 /*
1445 * Check for read packet available.
1446 */
1447 case FIONREAD:
1448 {
1449 int n;
1450
1451 BPFD_LOCK(d);
1452 n = d->bd_slen;
1453 while (d->bd_hbuf_in_use)
1454 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1455 PRINET, "bd_hbuf", 0);
1456 if (d->bd_hbuf)
1457 n += d->bd_hlen;
1458 BPFD_UNLOCK(d);
1459
1460 *(int *)addr = n;
1461 break;
1462 }
1463
1464 /*
1465 * Get buffer len [for read()].
1466 */
1467 case BIOCGBLEN:
1468 BPFD_LOCK(d);
1469 *(u_int *)addr = d->bd_bufsize;
1470 BPFD_UNLOCK(d);
1471 break;
1472
1473 /*
1474 * Set buffer length.
1475 */
1476 case BIOCSBLEN:
1477 error = bpf_ioctl_sblen(d, (u_int *)addr);
1478 break;
1479
1480 /*
1481 * Set link layer read filter.
1482 */
1483 case BIOCSETF:
1484 case BIOCSETFNR:
1485 case BIOCSETWF:
1486 #ifdef COMPAT_FREEBSD32
1487 case BIOCSETF32:
1488 case BIOCSETFNR32:
1489 case BIOCSETWF32:
1490 #endif
1491 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1492 break;
1493
1494 /*
1495 * Flush read packet buffer.
1496 */
1497 case BIOCFLUSH:
1498 BPFD_LOCK(d);
1499 reset_d(d);
1500 BPFD_UNLOCK(d);
1501 break;
1502
1503 /*
1504 * Put interface into promiscuous mode.
1505 */
1506 case BIOCPROMISC:
1507 BPF_LOCK();
1508 if (d->bd_bif == NULL) {
1509 /*
1510 * No interface attached yet.
1511 */
1512 error = EINVAL;
1513 } else if (d->bd_promisc == 0) {
1514 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1515 if (error == 0)
1516 d->bd_promisc = 1;
1517 }
1518 BPF_UNLOCK();
1519 break;
1520
1521 /*
1522 * Get current data link type.
1523 */
1524 case BIOCGDLT:
1525 BPF_LOCK();
1526 if (d->bd_bif == NULL)
1527 error = EINVAL;
1528 else
1529 *(u_int *)addr = d->bd_bif->bif_dlt;
1530 BPF_UNLOCK();
1531 break;
1532
1533 /*
1534 * Get a list of supported data link types.
1535 */
1536 #ifdef COMPAT_FREEBSD32
1537 case BIOCGDLTLIST32:
1538 {
1539 struct bpf_dltlist32 *list32;
1540 struct bpf_dltlist dltlist;
1541
1542 list32 = (struct bpf_dltlist32 *)addr;
1543 dltlist.bfl_len = list32->bfl_len;
1544 dltlist.bfl_list = PTRIN(list32->bfl_list);
1545 BPF_LOCK();
1546 if (d->bd_bif == NULL)
1547 error = EINVAL;
1548 else {
1549 error = bpf_getdltlist(d, &dltlist);
1550 if (error == 0)
1551 list32->bfl_len = dltlist.bfl_len;
1552 }
1553 BPF_UNLOCK();
1554 break;
1555 }
1556 #endif
1557
1558 case BIOCGDLTLIST:
1559 BPF_LOCK();
1560 if (d->bd_bif == NULL)
1561 error = EINVAL;
1562 else
1563 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1564 BPF_UNLOCK();
1565 break;
1566
1567 /*
1568 * Set data link type.
1569 */
1570 case BIOCSDLT:
1571 BPF_LOCK();
1572 if (d->bd_bif == NULL)
1573 error = EINVAL;
1574 else
1575 error = bpf_setdlt(d, *(u_int *)addr);
1576 BPF_UNLOCK();
1577 break;
1578
1579 /*
1580 * Get interface name.
1581 */
1582 case BIOCGETIF:
1583 BPF_LOCK();
1584 if (d->bd_bif == NULL)
1585 error = EINVAL;
1586 else {
1587 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1588 struct ifreq *const ifr = (struct ifreq *)addr;
1589
1590 strlcpy(ifr->ifr_name, ifp->if_xname,
1591 sizeof(ifr->ifr_name));
1592 }
1593 BPF_UNLOCK();
1594 break;
1595
1596 /*
1597 * Set interface.
1598 */
1599 case BIOCSETIF:
1600 /*
1601 * Behavior here depends on the buffering model. If we're
1602 * using kernel memory buffers, then we can allocate them here.
1603 * If we're using zero-copy, then the user process must have
1604 * registered buffers by the time we get here.
1605 */
1606 BPFD_LOCK(d);
1607 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1608 d->bd_sbuf == NULL) {
1609 u_int size;
1610
1611 size = d->bd_bufsize;
1612 BPFD_UNLOCK(d);
1613 error = bpf_buffer_ioctl_sblen(d, &size);
1614 if (error != 0)
1615 break;
1616 } else
1617 BPFD_UNLOCK(d);
1618 BPF_LOCK();
1619 error = bpf_setif(d, (struct ifreq *)addr);
1620 BPF_UNLOCK();
1621 break;
1622
1623 /*
1624 * Set read timeout.
1625 */
1626 case BIOCSRTIMEOUT:
1627 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1628 case BIOCSRTIMEOUT32:
1629 #endif
1630 {
1631 struct timeval *tv = (struct timeval *)addr;
1632 #if defined(COMPAT_FREEBSD32)
1633 struct timeval32 *tv32;
1634 struct timeval tv64;
1635
1636 if (cmd == BIOCSRTIMEOUT32) {
1637 tv32 = (struct timeval32 *)addr;
1638 tv = &tv64;
1639 tv->tv_sec = tv32->tv_sec;
1640 tv->tv_usec = tv32->tv_usec;
1641 } else
1642 #endif
1643 tv = (struct timeval *)addr;
1644
1645 /*
1646 * Subtract 1 tick from tvtohz() since this isn't
1647 * a one-shot timer.
1648 */
1649 if ((error = itimerfix(tv)) == 0)
1650 d->bd_rtout = tvtohz(tv) - 1;
1651 break;
1652 }
1653
1654 /*
1655 * Get read timeout.
1656 */
1657 case BIOCGRTIMEOUT:
1658 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1659 case BIOCGRTIMEOUT32:
1660 #endif
1661 {
1662 struct timeval *tv;
1663 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1664 struct timeval32 *tv32;
1665 struct timeval tv64;
1666
1667 if (cmd == BIOCGRTIMEOUT32)
1668 tv = &tv64;
1669 else
1670 #endif
1671 tv = (struct timeval *)addr;
1672
1673 tv->tv_sec = d->bd_rtout / hz;
1674 tv->tv_usec = (d->bd_rtout % hz) * tick;
1675 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1676 if (cmd == BIOCGRTIMEOUT32) {
1677 tv32 = (struct timeval32 *)addr;
1678 tv32->tv_sec = tv->tv_sec;
1679 tv32->tv_usec = tv->tv_usec;
1680 }
1681 #endif
1682
1683 break;
1684 }
1685
1686 /*
1687 * Get packet stats.
1688 */
1689 case BIOCGSTATS:
1690 {
1691 struct bpf_stat *bs = (struct bpf_stat *)addr;
1692
1693 /* XXXCSJP overflow */
1694 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1695 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1696 break;
1697 }
1698
1699 /*
1700 * Set immediate mode.
1701 */
1702 case BIOCIMMEDIATE:
1703 BPFD_LOCK(d);
1704 d->bd_immediate = *(u_int *)addr;
1705 BPFD_UNLOCK(d);
1706 break;
1707
1708 case BIOCVERSION:
1709 {
1710 struct bpf_version *bv = (struct bpf_version *)addr;
1711
1712 bv->bv_major = BPF_MAJOR_VERSION;
1713 bv->bv_minor = BPF_MINOR_VERSION;
1714 break;
1715 }
1716
1717 /*
1718 * Get "header already complete" flag
1719 */
1720 case BIOCGHDRCMPLT:
1721 BPFD_LOCK(d);
1722 *(u_int *)addr = d->bd_hdrcmplt;
1723 BPFD_UNLOCK(d);
1724 break;
1725
1726 /*
1727 * Set "header already complete" flag
1728 */
1729 case BIOCSHDRCMPLT:
1730 BPFD_LOCK(d);
1731 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1732 BPFD_UNLOCK(d);
1733 break;
1734
1735 /*
1736 * Get packet direction flag
1737 */
1738 case BIOCGDIRECTION:
1739 BPFD_LOCK(d);
1740 *(u_int *)addr = d->bd_direction;
1741 BPFD_UNLOCK(d);
1742 break;
1743
1744 /*
1745 * Set packet direction flag
1746 */
1747 case BIOCSDIRECTION:
1748 {
1749 u_int direction;
1750
1751 direction = *(u_int *)addr;
1752 switch (direction) {
1753 case BPF_D_IN:
1754 case BPF_D_INOUT:
1755 case BPF_D_OUT:
1756 BPFD_LOCK(d);
1757 d->bd_direction = direction;
1758 BPFD_UNLOCK(d);
1759 break;
1760 default:
1761 error = EINVAL;
1762 }
1763 }
1764 break;
1765
1766 /*
1767 * Get packet timestamp format and resolution.
1768 */
1769 case BIOCGTSTAMP:
1770 BPFD_LOCK(d);
1771 *(u_int *)addr = d->bd_tstamp;
1772 BPFD_UNLOCK(d);
1773 break;
1774
1775 /*
1776 * Set packet timestamp format and resolution.
1777 */
1778 case BIOCSTSTAMP:
1779 {
1780 u_int func;
1781
1782 func = *(u_int *)addr;
1783 if (BPF_T_VALID(func))
1784 d->bd_tstamp = func;
1785 else
1786 error = EINVAL;
1787 }
1788 break;
1789
1790 case BIOCFEEDBACK:
1791 BPFD_LOCK(d);
1792 d->bd_feedback = *(u_int *)addr;
1793 BPFD_UNLOCK(d);
1794 break;
1795
1796 case BIOCLOCK:
1797 BPFD_LOCK(d);
1798 d->bd_locked = 1;
1799 BPFD_UNLOCK(d);
1800 break;
1801
1802 case FIONBIO: /* Non-blocking I/O */
1803 break;
1804
1805 case FIOASYNC: /* Send signal on receive packets */
1806 BPFD_LOCK(d);
1807 d->bd_async = *(int *)addr;
1808 BPFD_UNLOCK(d);
1809 break;
1810
1811 case FIOSETOWN:
1812 /*
1813 * XXX: Add some sort of locking here?
1814 * fsetown() can sleep.
1815 */
1816 error = fsetown(*(int *)addr, &d->bd_sigio);
1817 break;
1818
1819 case FIOGETOWN:
1820 BPFD_LOCK(d);
1821 *(int *)addr = fgetown(&d->bd_sigio);
1822 BPFD_UNLOCK(d);
1823 break;
1824
1825 /* This is deprecated, FIOSETOWN should be used instead. */
1826 case TIOCSPGRP:
1827 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1828 break;
1829
1830 /* This is deprecated, FIOGETOWN should be used instead. */
1831 case TIOCGPGRP:
1832 *(int *)addr = -fgetown(&d->bd_sigio);
1833 break;
1834
1835 case BIOCSRSIG: /* Set receive signal */
1836 {
1837 u_int sig;
1838
1839 sig = *(u_int *)addr;
1840
1841 if (sig >= NSIG)
1842 error = EINVAL;
1843 else {
1844 BPFD_LOCK(d);
1845 d->bd_sig = sig;
1846 BPFD_UNLOCK(d);
1847 }
1848 break;
1849 }
1850 case BIOCGRSIG:
1851 BPFD_LOCK(d);
1852 *(u_int *)addr = d->bd_sig;
1853 BPFD_UNLOCK(d);
1854 break;
1855
1856 case BIOCGETBUFMODE:
1857 BPFD_LOCK(d);
1858 *(u_int *)addr = d->bd_bufmode;
1859 BPFD_UNLOCK(d);
1860 break;
1861
1862 case BIOCSETBUFMODE:
1863 /*
1864 * Allow the buffering mode to be changed as long as we
1865 * haven't yet committed to a particular mode. Our
1866 * definition of commitment, for now, is whether or not a
1867 * buffer has been allocated or an interface attached, since
1868 * that's the point where things get tricky.
1869 */
1870 switch (*(u_int *)addr) {
1871 case BPF_BUFMODE_BUFFER:
1872 break;
1873
1874 case BPF_BUFMODE_ZBUF:
1875 if (bpf_zerocopy_enable)
1876 break;
1877 /* FALLSTHROUGH */
1878
1879 default:
1880 CURVNET_RESTORE();
1881 return (EINVAL);
1882 }
1883
1884 BPFD_LOCK(d);
1885 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1886 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1887 BPFD_UNLOCK(d);
1888 CURVNET_RESTORE();
1889 return (EBUSY);
1890 }
1891 d->bd_bufmode = *(u_int *)addr;
1892 BPFD_UNLOCK(d);
1893 break;
1894
1895 case BIOCGETZMAX:
1896 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1897 break;
1898
1899 case BIOCSETZBUF:
1900 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1901 break;
1902
1903 case BIOCROTZBUF:
1904 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1905 break;
1906
1907 case BIOCSETVLANPCP:
1908 {
1909 u_int pcp;
1910
1911 pcp = *(u_int *)addr;
1912 if (pcp > BPF_PRIO_MAX || pcp < 0) {
1913 error = EINVAL;
1914 break;
1915 }
1916 d->bd_pcp = pcp;
1917 break;
1918 }
1919 }
1920 CURVNET_RESTORE();
1921 return (error);
1922 }
1923
1924 /*
1925 * Set d's packet filter program to fp. If this file already has a filter,
1926 * free it and replace it. Returns EINVAL for bogus requests.
1927 *
1928 * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1929 * calls.
1930 */
1931 static int
bpf_setf(struct bpf_d * d,struct bpf_program * fp,u_long cmd)1932 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1933 {
1934 #ifdef COMPAT_FREEBSD32
1935 struct bpf_program fp_swab;
1936 struct bpf_program32 *fp32;
1937 #endif
1938 struct bpf_program_buffer *fcode;
1939 struct bpf_insn *filter;
1940 #ifdef BPF_JITTER
1941 bpf_jit_filter *jfunc;
1942 #endif
1943 size_t size;
1944 u_int flen;
1945 bool track_event;
1946
1947 #ifdef COMPAT_FREEBSD32
1948 switch (cmd) {
1949 case BIOCSETF32:
1950 case BIOCSETWF32:
1951 case BIOCSETFNR32:
1952 fp32 = (struct bpf_program32 *)fp;
1953 fp_swab.bf_len = fp32->bf_len;
1954 fp_swab.bf_insns =
1955 (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1956 fp = &fp_swab;
1957 switch (cmd) {
1958 case BIOCSETF32:
1959 cmd = BIOCSETF;
1960 break;
1961 case BIOCSETWF32:
1962 cmd = BIOCSETWF;
1963 break;
1964 }
1965 break;
1966 }
1967 #endif
1968
1969 filter = NULL;
1970 #ifdef BPF_JITTER
1971 jfunc = NULL;
1972 #endif
1973 /*
1974 * Check new filter validness before acquiring any locks.
1975 * Allocate memory for new filter, if needed.
1976 */
1977 flen = fp->bf_len;
1978 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1979 return (EINVAL);
1980 size = flen * sizeof(*fp->bf_insns);
1981 if (size > 0) {
1982 /* We're setting up new filter. Copy and check actual data. */
1983 fcode = bpf_program_buffer_alloc(size, M_WAITOK);
1984 filter = (struct bpf_insn *)fcode->buffer;
1985 if (copyin(fp->bf_insns, filter, size) != 0 ||
1986 !bpf_validate(filter, flen)) {
1987 free(fcode, M_BPF);
1988 return (EINVAL);
1989 }
1990 #ifdef BPF_JITTER
1991 if (cmd != BIOCSETWF) {
1992 /*
1993 * Filter is copied inside fcode and is
1994 * perfectly valid.
1995 */
1996 jfunc = bpf_jitter(filter, flen);
1997 }
1998 #endif
1999 }
2000
2001 track_event = false;
2002 fcode = NULL;
2003
2004 BPF_LOCK();
2005 BPFD_LOCK(d);
2006 /* Set up new filter. */
2007 if (cmd == BIOCSETWF) {
2008 if (d->bd_wfilter != NULL) {
2009 fcode = __containerof((void *)d->bd_wfilter,
2010 struct bpf_program_buffer, buffer);
2011 #ifdef BPF_JITTER
2012 fcode->func = NULL;
2013 #endif
2014 }
2015 d->bd_wfilter = filter;
2016 } else {
2017 if (d->bd_rfilter != NULL) {
2018 fcode = __containerof((void *)d->bd_rfilter,
2019 struct bpf_program_buffer, buffer);
2020 #ifdef BPF_JITTER
2021 fcode->func = d->bd_bfilter;
2022 #endif
2023 }
2024 d->bd_rfilter = filter;
2025 #ifdef BPF_JITTER
2026 d->bd_bfilter = jfunc;
2027 #endif
2028 if (cmd == BIOCSETF)
2029 reset_d(d);
2030
2031 if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2032 /*
2033 * Filter can be set several times without
2034 * specifying interface. In this case just mark d
2035 * as reader.
2036 */
2037 d->bd_writer = 0;
2038 if (d->bd_bif != NULL) {
2039 /*
2040 * Remove descriptor from writers-only list
2041 * and add it to active readers list.
2042 */
2043 CK_LIST_REMOVE(d, bd_next);
2044 CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2045 d, bd_next);
2046 CTR2(KTR_NET,
2047 "%s: upgrade required by pid %d",
2048 __func__, d->bd_pid);
2049 track_event = true;
2050 }
2051 }
2052 }
2053 BPFD_UNLOCK(d);
2054
2055 if (fcode != NULL)
2056 NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2057
2058 if (track_event)
2059 EVENTHANDLER_INVOKE(bpf_track,
2060 d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2061
2062 BPF_UNLOCK();
2063 return (0);
2064 }
2065
2066 /*
2067 * Detach a file from its current interface (if attached at all) and attach
2068 * to the interface indicated by the name stored in ifr.
2069 * Return an errno or 0.
2070 */
2071 static int
bpf_setif(struct bpf_d * d,struct ifreq * ifr)2072 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2073 {
2074 struct bpf_if *bp;
2075 struct ifnet *theywant;
2076
2077 BPF_LOCK_ASSERT();
2078
2079 theywant = ifunit(ifr->ifr_name);
2080 if (theywant == NULL)
2081 return (ENXIO);
2082 /*
2083 * Look through attached interfaces for the named one.
2084 */
2085 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2086 if (bp->bif_ifp == theywant &&
2087 bp->bif_bpf == &theywant->if_bpf)
2088 break;
2089 }
2090 if (bp == NULL)
2091 return (ENXIO);
2092
2093 MPASS(bp == theywant->if_bpf);
2094 /*
2095 * At this point, we expect the buffer is already allocated. If not,
2096 * return an error.
2097 */
2098 switch (d->bd_bufmode) {
2099 case BPF_BUFMODE_BUFFER:
2100 case BPF_BUFMODE_ZBUF:
2101 if (d->bd_sbuf == NULL)
2102 return (EINVAL);
2103 break;
2104
2105 default:
2106 panic("bpf_setif: bufmode %d", d->bd_bufmode);
2107 }
2108 if (bp != d->bd_bif)
2109 bpf_attachd(d, bp);
2110 else {
2111 BPFD_LOCK(d);
2112 reset_d(d);
2113 BPFD_UNLOCK(d);
2114 }
2115 return (0);
2116 }
2117
2118 /*
2119 * Support for select() and poll() system calls
2120 *
2121 * Return true iff the specific operation will not block indefinitely.
2122 * Otherwise, return false but make a note that a selwakeup() must be done.
2123 */
2124 static int
bpfpoll(struct cdev * dev,int events,struct thread * td)2125 bpfpoll(struct cdev *dev, int events, struct thread *td)
2126 {
2127 struct bpf_d *d;
2128 int revents;
2129
2130 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2131 return (events &
2132 (POLLHUP | POLLIN | POLLRDNORM | POLLOUT | POLLWRNORM));
2133
2134 /*
2135 * Refresh PID associated with this descriptor.
2136 */
2137 revents = events & (POLLOUT | POLLWRNORM);
2138 BPFD_LOCK(d);
2139 BPF_PID_REFRESH(d, td);
2140 if (events & (POLLIN | POLLRDNORM)) {
2141 if (bpf_ready(d))
2142 revents |= events & (POLLIN | POLLRDNORM);
2143 else {
2144 selrecord(td, &d->bd_sel);
2145 /* Start the read timeout if necessary. */
2146 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2147 callout_reset(&d->bd_callout, d->bd_rtout,
2148 bpf_timed_out, d);
2149 d->bd_state = BPF_WAITING;
2150 }
2151 }
2152 }
2153 BPFD_UNLOCK(d);
2154 return (revents);
2155 }
2156
2157 /*
2158 * Support for kevent() system call. Register EVFILT_READ filters and
2159 * reject all others.
2160 */
2161 int
bpfkqfilter(struct cdev * dev,struct knote * kn)2162 bpfkqfilter(struct cdev *dev, struct knote *kn)
2163 {
2164 struct bpf_d *d;
2165
2166 if (devfs_get_cdevpriv((void **)&d) != 0)
2167 return (1);
2168
2169 switch (kn->kn_filter) {
2170 case EVFILT_READ:
2171 kn->kn_fop = &bpfread_filtops;
2172 break;
2173
2174 case EVFILT_WRITE:
2175 kn->kn_fop = &bpfwrite_filtops;
2176 break;
2177
2178 default:
2179 return (1);
2180 }
2181
2182 /*
2183 * Refresh PID associated with this descriptor.
2184 */
2185 BPFD_LOCK(d);
2186 BPF_PID_REFRESH_CUR(d);
2187 kn->kn_hook = d;
2188 knlist_add(&d->bd_sel.si_note, kn, 1);
2189 BPFD_UNLOCK(d);
2190
2191 return (0);
2192 }
2193
2194 static void
filt_bpfdetach(struct knote * kn)2195 filt_bpfdetach(struct knote *kn)
2196 {
2197 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2198
2199 knlist_remove(&d->bd_sel.si_note, kn, 0);
2200 }
2201
2202 static int
filt_bpfread(struct knote * kn,long hint)2203 filt_bpfread(struct knote *kn, long hint)
2204 {
2205 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2206 int ready;
2207
2208 BPFD_LOCK_ASSERT(d);
2209 ready = bpf_ready(d);
2210 if (ready) {
2211 kn->kn_data = d->bd_slen;
2212 /*
2213 * Ignore the hold buffer if it is being copied to user space.
2214 */
2215 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2216 kn->kn_data += d->bd_hlen;
2217 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2218 callout_reset(&d->bd_callout, d->bd_rtout,
2219 bpf_timed_out, d);
2220 d->bd_state = BPF_WAITING;
2221 }
2222
2223 return (ready);
2224 }
2225
2226 static int
filt_bpfwrite(struct knote * kn,long hint)2227 filt_bpfwrite(struct knote *kn, long hint)
2228 {
2229 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2230
2231 BPFD_LOCK_ASSERT(d);
2232
2233 if (d->bd_bif == NULL) {
2234 kn->kn_data = 0;
2235 return (0);
2236 } else {
2237 kn->kn_data = d->bd_bif->bif_ifp->if_mtu;
2238 return (1);
2239 }
2240 }
2241
2242 #define BPF_TSTAMP_NONE 0
2243 #define BPF_TSTAMP_FAST 1
2244 #define BPF_TSTAMP_NORMAL 2
2245 #define BPF_TSTAMP_EXTERN 3
2246
2247 static int
bpf_ts_quality(int tstype)2248 bpf_ts_quality(int tstype)
2249 {
2250
2251 if (tstype == BPF_T_NONE)
2252 return (BPF_TSTAMP_NONE);
2253 if ((tstype & BPF_T_FAST) != 0)
2254 return (BPF_TSTAMP_FAST);
2255
2256 return (BPF_TSTAMP_NORMAL);
2257 }
2258
2259 static int
bpf_gettime(struct bintime * bt,int tstype,struct mbuf * m)2260 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2261 {
2262 struct timespec ts;
2263 struct m_tag *tag;
2264 int quality;
2265
2266 quality = bpf_ts_quality(tstype);
2267 if (quality == BPF_TSTAMP_NONE)
2268 return (quality);
2269
2270 if (m != NULL) {
2271 if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) {
2272 mbuf_tstmp2timespec(m, &ts);
2273 timespec2bintime(&ts, bt);
2274 return (BPF_TSTAMP_EXTERN);
2275 }
2276 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2277 if (tag != NULL) {
2278 *bt = *(struct bintime *)(tag + 1);
2279 return (BPF_TSTAMP_EXTERN);
2280 }
2281 }
2282 if (quality == BPF_TSTAMP_NORMAL)
2283 binuptime(bt);
2284 else
2285 getbinuptime(bt);
2286
2287 return (quality);
2288 }
2289
2290 /*
2291 * Incoming linkage from device drivers. Process the packet pkt, of length
2292 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2293 * by each process' filter, and if accepted, stashed into the corresponding
2294 * buffer.
2295 */
2296 void
bpf_tap(struct bpf_if * bp,u_char * pkt,u_int pktlen)2297 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2298 {
2299 struct epoch_tracker et;
2300 struct bintime bt;
2301 struct bpf_d *d;
2302 #ifdef BPF_JITTER
2303 bpf_jit_filter *bf;
2304 #endif
2305 u_int slen;
2306 int gottime;
2307
2308 gottime = BPF_TSTAMP_NONE;
2309 NET_EPOCH_ENTER(et);
2310 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2311 counter_u64_add(d->bd_rcount, 1);
2312 /*
2313 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2314 * is no way for the caller to indiciate to us whether this
2315 * packet is inbound or outbound. In the bpf_mtap() routines,
2316 * we use the interface pointers on the mbuf to figure it out.
2317 */
2318 #ifdef BPF_JITTER
2319 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2320 if (bf != NULL)
2321 slen = (*(bf->func))(pkt, pktlen, pktlen);
2322 else
2323 #endif
2324 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2325 if (slen != 0) {
2326 /*
2327 * Filter matches. Let's to acquire write lock.
2328 */
2329 BPFD_LOCK(d);
2330 counter_u64_add(d->bd_fcount, 1);
2331 if (gottime < bpf_ts_quality(d->bd_tstamp))
2332 gottime = bpf_gettime(&bt, d->bd_tstamp,
2333 NULL);
2334 #ifdef MAC
2335 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2336 #endif
2337 catchpacket(d, pkt, pktlen, slen,
2338 bpf_append_bytes, &bt);
2339 BPFD_UNLOCK(d);
2340 }
2341 }
2342 NET_EPOCH_EXIT(et);
2343 }
2344
2345 void
bpf_tap_if(if_t ifp,u_char * pkt,u_int pktlen)2346 bpf_tap_if(if_t ifp, u_char *pkt, u_int pktlen)
2347 {
2348 if (bpf_peers_present(ifp->if_bpf))
2349 bpf_tap(ifp->if_bpf, pkt, pktlen);
2350 }
2351
2352 #define BPF_CHECK_DIRECTION(d, r, i) \
2353 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2354 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2355
2356 /*
2357 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2358 * Locking model is explained in bpf_tap().
2359 */
2360 void
bpf_mtap(struct bpf_if * bp,struct mbuf * m)2361 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2362 {
2363 struct epoch_tracker et;
2364 struct bintime bt;
2365 struct bpf_d *d;
2366 #ifdef BPF_JITTER
2367 bpf_jit_filter *bf;
2368 #endif
2369 u_int pktlen, slen;
2370 int gottime;
2371
2372 /* Skip outgoing duplicate packets. */
2373 if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2374 m->m_flags &= ~M_PROMISC;
2375 return;
2376 }
2377
2378 pktlen = m_length(m, NULL);
2379 gottime = BPF_TSTAMP_NONE;
2380
2381 NET_EPOCH_ENTER(et);
2382 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2383 if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2384 continue;
2385 counter_u64_add(d->bd_rcount, 1);
2386 #ifdef BPF_JITTER
2387 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2388 /* XXX We cannot handle multiple mbufs. */
2389 if (bf != NULL && m->m_next == NULL)
2390 slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2391 pktlen);
2392 else
2393 #endif
2394 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2395 if (slen != 0) {
2396 BPFD_LOCK(d);
2397
2398 counter_u64_add(d->bd_fcount, 1);
2399 if (gottime < bpf_ts_quality(d->bd_tstamp))
2400 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2401 #ifdef MAC
2402 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2403 #endif
2404 catchpacket(d, (u_char *)m, pktlen, slen,
2405 bpf_append_mbuf, &bt);
2406 BPFD_UNLOCK(d);
2407 }
2408 }
2409 NET_EPOCH_EXIT(et);
2410 }
2411
2412 void
bpf_mtap_if(if_t ifp,struct mbuf * m)2413 bpf_mtap_if(if_t ifp, struct mbuf *m)
2414 {
2415 if (bpf_peers_present(ifp->if_bpf)) {
2416 M_ASSERTVALID(m);
2417 bpf_mtap(ifp->if_bpf, m);
2418 }
2419 }
2420
2421 /*
2422 * Incoming linkage from device drivers, when packet is in
2423 * an mbuf chain and to be prepended by a contiguous header.
2424 */
2425 void
bpf_mtap2(struct bpf_if * bp,void * data,u_int dlen,struct mbuf * m)2426 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2427 {
2428 struct epoch_tracker et;
2429 struct bintime bt;
2430 struct mbuf mb;
2431 struct bpf_d *d;
2432 u_int pktlen, slen;
2433 int gottime;
2434
2435 /* Skip outgoing duplicate packets. */
2436 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2437 m->m_flags &= ~M_PROMISC;
2438 return;
2439 }
2440
2441 pktlen = m_length(m, NULL);
2442 /*
2443 * Craft on-stack mbuf suitable for passing to bpf_filter.
2444 * Note that we cut corners here; we only setup what's
2445 * absolutely needed--this mbuf should never go anywhere else.
2446 */
2447 mb.m_flags = 0;
2448 mb.m_next = m;
2449 mb.m_data = data;
2450 mb.m_len = dlen;
2451 pktlen += dlen;
2452
2453 gottime = BPF_TSTAMP_NONE;
2454
2455 NET_EPOCH_ENTER(et);
2456 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2457 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2458 continue;
2459 counter_u64_add(d->bd_rcount, 1);
2460 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2461 if (slen != 0) {
2462 BPFD_LOCK(d);
2463
2464 counter_u64_add(d->bd_fcount, 1);
2465 if (gottime < bpf_ts_quality(d->bd_tstamp))
2466 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2467 #ifdef MAC
2468 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2469 #endif
2470 catchpacket(d, (u_char *)&mb, pktlen, slen,
2471 bpf_append_mbuf, &bt);
2472 BPFD_UNLOCK(d);
2473 }
2474 }
2475 NET_EPOCH_EXIT(et);
2476 }
2477
2478 void
bpf_mtap2_if(if_t ifp,void * data,u_int dlen,struct mbuf * m)2479 bpf_mtap2_if(if_t ifp, void *data, u_int dlen, struct mbuf *m)
2480 {
2481 if (bpf_peers_present(ifp->if_bpf)) {
2482 M_ASSERTVALID(m);
2483 bpf_mtap2(ifp->if_bpf, data, dlen, m);
2484 }
2485 }
2486
2487 #undef BPF_CHECK_DIRECTION
2488 #undef BPF_TSTAMP_NONE
2489 #undef BPF_TSTAMP_FAST
2490 #undef BPF_TSTAMP_NORMAL
2491 #undef BPF_TSTAMP_EXTERN
2492
2493 static int
bpf_hdrlen(struct bpf_d * d)2494 bpf_hdrlen(struct bpf_d *d)
2495 {
2496 int hdrlen;
2497
2498 hdrlen = d->bd_bif->bif_hdrlen;
2499 #ifndef BURN_BRIDGES
2500 if (d->bd_tstamp == BPF_T_NONE ||
2501 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2502 #ifdef COMPAT_FREEBSD32
2503 if (d->bd_compat32)
2504 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2505 else
2506 #endif
2507 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2508 else
2509 #endif
2510 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2511 #ifdef COMPAT_FREEBSD32
2512 if (d->bd_compat32)
2513 hdrlen = BPF_WORDALIGN32(hdrlen);
2514 else
2515 #endif
2516 hdrlen = BPF_WORDALIGN(hdrlen);
2517
2518 return (hdrlen - d->bd_bif->bif_hdrlen);
2519 }
2520
2521 static void
bpf_bintime2ts(struct bintime * bt,struct bpf_ts * ts,int tstype)2522 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2523 {
2524 struct bintime bt2, boottimebin;
2525 struct timeval tsm;
2526 struct timespec tsn;
2527
2528 if ((tstype & BPF_T_MONOTONIC) == 0) {
2529 bt2 = *bt;
2530 getboottimebin(&boottimebin);
2531 bintime_add(&bt2, &boottimebin);
2532 bt = &bt2;
2533 }
2534 switch (BPF_T_FORMAT(tstype)) {
2535 case BPF_T_MICROTIME:
2536 bintime2timeval(bt, &tsm);
2537 ts->bt_sec = tsm.tv_sec;
2538 ts->bt_frac = tsm.tv_usec;
2539 break;
2540 case BPF_T_NANOTIME:
2541 bintime2timespec(bt, &tsn);
2542 ts->bt_sec = tsn.tv_sec;
2543 ts->bt_frac = tsn.tv_nsec;
2544 break;
2545 case BPF_T_BINTIME:
2546 ts->bt_sec = bt->sec;
2547 ts->bt_frac = bt->frac;
2548 break;
2549 }
2550 }
2551
2552 /*
2553 * Move the packet data from interface memory (pkt) into the
2554 * store buffer. "cpfn" is the routine called to do the actual data
2555 * transfer. bcopy is passed in to copy contiguous chunks, while
2556 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2557 * pkt is really an mbuf.
2558 */
2559 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)2560 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2561 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2562 struct bintime *bt)
2563 {
2564 static char zeroes[BPF_ALIGNMENT];
2565 struct bpf_xhdr hdr;
2566 #ifndef BURN_BRIDGES
2567 struct bpf_hdr hdr_old;
2568 #ifdef COMPAT_FREEBSD32
2569 struct bpf_hdr32 hdr32_old;
2570 #endif
2571 #endif
2572 int caplen, curlen, hdrlen, pad, totlen;
2573 int do_wakeup = 0;
2574 int do_timestamp;
2575 int tstype;
2576
2577 BPFD_LOCK_ASSERT(d);
2578 if (d->bd_bif == NULL) {
2579 /* Descriptor was detached in concurrent thread */
2580 counter_u64_add(d->bd_dcount, 1);
2581 return;
2582 }
2583
2584 /*
2585 * Detect whether user space has released a buffer back to us, and if
2586 * so, move it from being a hold buffer to a free buffer. This may
2587 * not be the best place to do it (for example, we might only want to
2588 * run this check if we need the space), but for now it's a reliable
2589 * spot to do it.
2590 */
2591 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2592 d->bd_fbuf = d->bd_hbuf;
2593 d->bd_hbuf = NULL;
2594 d->bd_hlen = 0;
2595 bpf_buf_reclaimed(d);
2596 }
2597
2598 /*
2599 * Figure out how many bytes to move. If the packet is
2600 * greater or equal to the snapshot length, transfer that
2601 * much. Otherwise, transfer the whole packet (unless
2602 * we hit the buffer size limit).
2603 */
2604 hdrlen = bpf_hdrlen(d);
2605 totlen = hdrlen + min(snaplen, pktlen);
2606 if (totlen > d->bd_bufsize)
2607 totlen = d->bd_bufsize;
2608
2609 /*
2610 * Round up the end of the previous packet to the next longword.
2611 *
2612 * Drop the packet if there's no room and no hope of room
2613 * If the packet would overflow the storage buffer or the storage
2614 * buffer is considered immutable by the buffer model, try to rotate
2615 * the buffer and wakeup pending processes.
2616 */
2617 #ifdef COMPAT_FREEBSD32
2618 if (d->bd_compat32)
2619 curlen = BPF_WORDALIGN32(d->bd_slen);
2620 else
2621 #endif
2622 curlen = BPF_WORDALIGN(d->bd_slen);
2623 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2624 if (d->bd_fbuf == NULL) {
2625 /*
2626 * There's no room in the store buffer, and no
2627 * prospect of room, so drop the packet. Notify the
2628 * buffer model.
2629 */
2630 bpf_buffull(d);
2631 counter_u64_add(d->bd_dcount, 1);
2632 return;
2633 }
2634 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2635 ROTATE_BUFFERS(d);
2636 do_wakeup = 1;
2637 curlen = 0;
2638 } else {
2639 if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
2640 /*
2641 * Immediate mode is set, or the read timeout has
2642 * already expired during a select call. A packet
2643 * arrived, so the reader should be woken up.
2644 */
2645 do_wakeup = 1;
2646 }
2647 pad = curlen - d->bd_slen;
2648 KASSERT(pad >= 0 && pad <= sizeof(zeroes),
2649 ("%s: invalid pad byte count %d", __func__, pad));
2650 if (pad > 0) {
2651 /* Zero pad bytes. */
2652 bpf_append_bytes(d, d->bd_sbuf, d->bd_slen, zeroes,
2653 pad);
2654 }
2655 }
2656
2657 caplen = totlen - hdrlen;
2658 tstype = d->bd_tstamp;
2659 do_timestamp = tstype != BPF_T_NONE;
2660 #ifndef BURN_BRIDGES
2661 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2662 struct bpf_ts ts;
2663 if (do_timestamp)
2664 bpf_bintime2ts(bt, &ts, tstype);
2665 #ifdef COMPAT_FREEBSD32
2666 if (d->bd_compat32) {
2667 bzero(&hdr32_old, sizeof(hdr32_old));
2668 if (do_timestamp) {
2669 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2670 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2671 }
2672 hdr32_old.bh_datalen = pktlen;
2673 hdr32_old.bh_hdrlen = hdrlen;
2674 hdr32_old.bh_caplen = caplen;
2675 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2676 sizeof(hdr32_old));
2677 goto copy;
2678 }
2679 #endif
2680 bzero(&hdr_old, sizeof(hdr_old));
2681 if (do_timestamp) {
2682 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2683 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2684 }
2685 hdr_old.bh_datalen = pktlen;
2686 hdr_old.bh_hdrlen = hdrlen;
2687 hdr_old.bh_caplen = caplen;
2688 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2689 sizeof(hdr_old));
2690 goto copy;
2691 }
2692 #endif
2693
2694 /*
2695 * Append the bpf header. Note we append the actual header size, but
2696 * move forward the length of the header plus padding.
2697 */
2698 bzero(&hdr, sizeof(hdr));
2699 if (do_timestamp)
2700 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2701 hdr.bh_datalen = pktlen;
2702 hdr.bh_hdrlen = hdrlen;
2703 hdr.bh_caplen = caplen;
2704 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2705
2706 /*
2707 * Copy the packet data into the store buffer and update its length.
2708 */
2709 #ifndef BURN_BRIDGES
2710 copy:
2711 #endif
2712 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2713 d->bd_slen = curlen + totlen;
2714
2715 if (do_wakeup)
2716 bpf_wakeup(d);
2717 }
2718
2719 /*
2720 * Free buffers currently in use by a descriptor.
2721 * Called on close.
2722 */
2723 static void
bpfd_free(epoch_context_t ctx)2724 bpfd_free(epoch_context_t ctx)
2725 {
2726 struct bpf_d *d;
2727 struct bpf_program_buffer *p;
2728
2729 /*
2730 * We don't need to lock out interrupts since this descriptor has
2731 * been detached from its interface and it yet hasn't been marked
2732 * free.
2733 */
2734 d = __containerof(ctx, struct bpf_d, epoch_ctx);
2735 bpf_free(d);
2736 if (d->bd_rfilter != NULL) {
2737 p = __containerof((void *)d->bd_rfilter,
2738 struct bpf_program_buffer, buffer);
2739 #ifdef BPF_JITTER
2740 p->func = d->bd_bfilter;
2741 #endif
2742 bpf_program_buffer_free(&p->epoch_ctx);
2743 }
2744 if (d->bd_wfilter != NULL) {
2745 p = __containerof((void *)d->bd_wfilter,
2746 struct bpf_program_buffer, buffer);
2747 #ifdef BPF_JITTER
2748 p->func = NULL;
2749 #endif
2750 bpf_program_buffer_free(&p->epoch_ctx);
2751 }
2752
2753 mtx_destroy(&d->bd_lock);
2754 counter_u64_free(d->bd_rcount);
2755 counter_u64_free(d->bd_dcount);
2756 counter_u64_free(d->bd_fcount);
2757 counter_u64_free(d->bd_wcount);
2758 counter_u64_free(d->bd_wfcount);
2759 counter_u64_free(d->bd_wdcount);
2760 counter_u64_free(d->bd_zcopy);
2761 free(d, M_BPF);
2762 }
2763
2764 /*
2765 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2766 * fixed size of the link header (variable length headers not yet supported).
2767 */
2768 void
bpfattach(struct ifnet * ifp,u_int dlt,u_int hdrlen)2769 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2770 {
2771
2772 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2773 }
2774
2775 /*
2776 * Attach an interface to bpf. ifp is a pointer to the structure
2777 * defining the interface to be attached, dlt is the link layer type,
2778 * and hdrlen is the fixed size of the link header (variable length
2779 * headers are not yet supporrted).
2780 */
2781 void
bpfattach2(struct ifnet * ifp,u_int dlt,u_int hdrlen,struct bpf_if ** driverp)2782 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2783 struct bpf_if **driverp)
2784 {
2785 struct bpf_if *bp;
2786
2787 KASSERT(*driverp == NULL,
2788 ("bpfattach2: driverp already initialized"));
2789
2790 bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2791
2792 CK_LIST_INIT(&bp->bif_dlist);
2793 CK_LIST_INIT(&bp->bif_wlist);
2794 bp->bif_ifp = ifp;
2795 bp->bif_dlt = dlt;
2796 bp->bif_hdrlen = hdrlen;
2797 bp->bif_bpf = driverp;
2798 refcount_init(&bp->bif_refcnt, 1);
2799 *driverp = bp;
2800 /*
2801 * Reference ifnet pointer, so it won't freed until
2802 * we release it.
2803 */
2804 if_ref(ifp);
2805 BPF_LOCK();
2806 CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2807 BPF_UNLOCK();
2808
2809 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2810 if_printf(ifp, "bpf attached\n");
2811 }
2812
2813 #ifdef VIMAGE
2814 /*
2815 * Detach descriptors on interface's vmove event.
2816 */
2817 void
bpf_ifdetach(struct ifnet * ifp)2818 bpf_ifdetach(struct ifnet *ifp)
2819 {
2820 struct bpf_if *bp;
2821 struct bpf_d *d;
2822
2823 BPF_LOCK();
2824 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2825 if (bp->bif_ifp != ifp)
2826 continue;
2827
2828 /* Detach common descriptors */
2829 while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2830 bpf_detachd(d, true);
2831 }
2832
2833 /* Detach writer-only descriptors */
2834 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2835 bpf_detachd(d, true);
2836 }
2837 }
2838 BPF_UNLOCK();
2839 }
2840 #endif
2841
2842 /*
2843 * Detach bpf from an interface. This involves detaching each descriptor
2844 * associated with the interface. Notify each descriptor as it's detached
2845 * so that any sleepers wake up and get ENXIO.
2846 */
2847 void
bpfdetach(struct ifnet * ifp)2848 bpfdetach(struct ifnet *ifp)
2849 {
2850 struct bpf_if *bp, *bp_temp;
2851 struct bpf_d *d;
2852
2853 BPF_LOCK();
2854 /* Find all bpf_if struct's which reference ifp and detach them. */
2855 CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2856 if (ifp != bp->bif_ifp)
2857 continue;
2858
2859 CK_LIST_REMOVE(bp, bif_next);
2860 *bp->bif_bpf = __DECONST(struct bpf_if *, &dead_bpf_if);
2861
2862 CTR4(KTR_NET,
2863 "%s: sheduling free for encap %d (%p) for if %p",
2864 __func__, bp->bif_dlt, bp, ifp);
2865
2866 /* Detach common descriptors */
2867 while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2868 bpf_detachd(d, true);
2869 }
2870
2871 /* Detach writer-only descriptors */
2872 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2873 bpf_detachd(d, true);
2874 }
2875 bpfif_rele(bp);
2876 }
2877 BPF_UNLOCK();
2878 }
2879
2880 bool
bpf_peers_present_if(struct ifnet * ifp)2881 bpf_peers_present_if(struct ifnet *ifp)
2882 {
2883 return (bpf_peers_present(ifp->if_bpf));
2884 }
2885
2886 /*
2887 * Get a list of available data link type of the interface.
2888 */
2889 static int
bpf_getdltlist(struct bpf_d * d,struct bpf_dltlist * bfl)2890 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2891 {
2892 struct ifnet *ifp;
2893 struct bpf_if *bp;
2894 u_int *lst;
2895 int error, n, n1;
2896
2897 BPF_LOCK_ASSERT();
2898
2899 ifp = d->bd_bif->bif_ifp;
2900 n1 = 0;
2901 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2902 if (bp->bif_ifp == ifp)
2903 n1++;
2904 }
2905 if (bfl->bfl_list == NULL) {
2906 bfl->bfl_len = n1;
2907 return (0);
2908 }
2909 if (n1 > bfl->bfl_len)
2910 return (ENOMEM);
2911
2912 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2913 n = 0;
2914 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2915 if (bp->bif_ifp != ifp)
2916 continue;
2917 lst[n++] = bp->bif_dlt;
2918 }
2919 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2920 free(lst, M_TEMP);
2921 bfl->bfl_len = n;
2922 return (error);
2923 }
2924
2925 /*
2926 * Set the data link type of a BPF instance.
2927 */
2928 static int
bpf_setdlt(struct bpf_d * d,u_int dlt)2929 bpf_setdlt(struct bpf_d *d, u_int dlt)
2930 {
2931 int error, opromisc;
2932 struct ifnet *ifp;
2933 struct bpf_if *bp;
2934
2935 BPF_LOCK_ASSERT();
2936 MPASS(d->bd_bif != NULL);
2937
2938 /*
2939 * It is safe to check bd_bif without BPFD_LOCK, it can not be
2940 * changed while we hold global lock.
2941 */
2942 if (d->bd_bif->bif_dlt == dlt)
2943 return (0);
2944
2945 ifp = d->bd_bif->bif_ifp;
2946 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2947 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2948 break;
2949 }
2950 if (bp == NULL)
2951 return (EINVAL);
2952
2953 opromisc = d->bd_promisc;
2954 bpf_attachd(d, bp);
2955 if (opromisc) {
2956 error = ifpromisc(bp->bif_ifp, 1);
2957 if (error)
2958 if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
2959 __func__, error);
2960 else
2961 d->bd_promisc = 1;
2962 }
2963 return (0);
2964 }
2965
2966 static void
bpf_drvinit(void * unused)2967 bpf_drvinit(void *unused)
2968 {
2969 struct cdev *dev;
2970
2971 sx_init(&bpf_sx, "bpf global lock");
2972 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2973 /* For compatibility */
2974 make_dev_alias(dev, "bpf0");
2975 }
2976
2977 /*
2978 * Zero out the various packet counters associated with all of the bpf
2979 * descriptors. At some point, we will probably want to get a bit more
2980 * granular and allow the user to specify descriptors to be zeroed.
2981 */
2982 static void
bpf_zero_counters(void)2983 bpf_zero_counters(void)
2984 {
2985 struct bpf_if *bp;
2986 struct bpf_d *bd;
2987
2988 BPF_LOCK();
2989 /*
2990 * We are protected by global lock here, interfaces and
2991 * descriptors can not be deleted while we hold it.
2992 */
2993 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2994 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2995 counter_u64_zero(bd->bd_rcount);
2996 counter_u64_zero(bd->bd_dcount);
2997 counter_u64_zero(bd->bd_fcount);
2998 counter_u64_zero(bd->bd_wcount);
2999 counter_u64_zero(bd->bd_wfcount);
3000 counter_u64_zero(bd->bd_zcopy);
3001 }
3002 }
3003 BPF_UNLOCK();
3004 }
3005
3006 /*
3007 * Fill filter statistics
3008 */
3009 static void
bpfstats_fill_xbpf(struct xbpf_d * d,struct bpf_d * bd)3010 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
3011 {
3012
3013 BPF_LOCK_ASSERT();
3014 bzero(d, sizeof(*d));
3015 d->bd_structsize = sizeof(*d);
3016 d->bd_immediate = bd->bd_immediate;
3017 d->bd_promisc = bd->bd_promisc;
3018 d->bd_hdrcmplt = bd->bd_hdrcmplt;
3019 d->bd_direction = bd->bd_direction;
3020 d->bd_feedback = bd->bd_feedback;
3021 d->bd_async = bd->bd_async;
3022 d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
3023 d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
3024 d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
3025 d->bd_sig = bd->bd_sig;
3026 d->bd_slen = bd->bd_slen;
3027 d->bd_hlen = bd->bd_hlen;
3028 d->bd_bufsize = bd->bd_bufsize;
3029 d->bd_pid = bd->bd_pid;
3030 strlcpy(d->bd_ifname,
3031 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
3032 d->bd_locked = bd->bd_locked;
3033 d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
3034 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
3035 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
3036 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
3037 d->bd_bufmode = bd->bd_bufmode;
3038 }
3039
3040 /*
3041 * Handle `netstat -B' stats request
3042 */
3043 static int
bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)3044 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
3045 {
3046 static const struct xbpf_d zerostats;
3047 struct xbpf_d *xbdbuf, *xbd, tempstats;
3048 int index, error;
3049 struct bpf_if *bp;
3050 struct bpf_d *bd;
3051
3052 /*
3053 * XXX This is not technically correct. It is possible for non
3054 * privileged users to open bpf devices. It would make sense
3055 * if the users who opened the devices were able to retrieve
3056 * the statistics for them, too.
3057 */
3058 error = priv_check(req->td, PRIV_NET_BPF);
3059 if (error)
3060 return (error);
3061 /*
3062 * Check to see if the user is requesting that the counters be
3063 * zeroed out. Explicitly check that the supplied data is zeroed,
3064 * as we aren't allowing the user to set the counters currently.
3065 */
3066 if (req->newptr != NULL) {
3067 if (req->newlen != sizeof(tempstats))
3068 return (EINVAL);
3069 memset(&tempstats, 0, sizeof(tempstats));
3070 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
3071 if (error)
3072 return (error);
3073 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
3074 return (EINVAL);
3075 bpf_zero_counters();
3076 return (0);
3077 }
3078 if (req->oldptr == NULL)
3079 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
3080 if (bpf_bpfd_cnt == 0)
3081 return (SYSCTL_OUT(req, 0, 0));
3082 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
3083 BPF_LOCK();
3084 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
3085 BPF_UNLOCK();
3086 free(xbdbuf, M_BPF);
3087 return (ENOMEM);
3088 }
3089 index = 0;
3090 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
3091 /* Send writers-only first */
3092 CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
3093 xbd = &xbdbuf[index++];
3094 bpfstats_fill_xbpf(xbd, bd);
3095 }
3096 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
3097 xbd = &xbdbuf[index++];
3098 bpfstats_fill_xbpf(xbd, bd);
3099 }
3100 }
3101 BPF_UNLOCK();
3102 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
3103 free(xbdbuf, M_BPF);
3104 return (error);
3105 }
3106
3107 SYSINIT(bpfdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, bpf_drvinit, NULL);
3108
3109 #else /* !DEV_BPF && !NETGRAPH_BPF */
3110
3111 /*
3112 * NOP stubs to allow bpf-using drivers to load and function.
3113 *
3114 * A 'better' implementation would allow the core bpf functionality
3115 * to be loaded at runtime.
3116 */
3117
3118 void
bpf_tap(struct bpf_if * bp,u_char * pkt,u_int pktlen)3119 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3120 {
3121 }
3122
3123 void
bpf_tap_if(if_t ifp,u_char * pkt,u_int pktlen)3124 bpf_tap_if(if_t ifp, u_char *pkt, u_int pktlen)
3125 {
3126 }
3127
3128 void
bpf_mtap(struct bpf_if * bp,struct mbuf * m)3129 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3130 {
3131 }
3132
3133 void
bpf_mtap_if(if_t ifp,struct mbuf * m)3134 bpf_mtap_if(if_t ifp, struct mbuf *m)
3135 {
3136 }
3137
3138 void
bpf_mtap2(struct bpf_if * bp,void * d,u_int l,struct mbuf * m)3139 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3140 {
3141 }
3142
3143 void
bpf_mtap2_if(if_t ifp,void * data,u_int dlen,struct mbuf * m)3144 bpf_mtap2_if(if_t ifp, void *data, u_int dlen, struct mbuf *m)
3145 {
3146 }
3147
3148 void
bpfattach(struct ifnet * ifp,u_int dlt,u_int hdrlen)3149 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3150 {
3151
3152 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3153 }
3154
3155 void
bpfattach2(struct ifnet * ifp,u_int dlt,u_int hdrlen,struct bpf_if ** driverp)3156 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3157 {
3158
3159 *driverp = __DECONST(struct bpf_if *, &dead_bpf_if);
3160 }
3161
3162 void
bpfdetach(struct ifnet * ifp)3163 bpfdetach(struct ifnet *ifp)
3164 {
3165 }
3166
3167 bool
bpf_peers_present_if(struct ifnet * ifp)3168 bpf_peers_present_if(struct ifnet *ifp)
3169 {
3170 return (false);
3171 }
3172
3173 u_int
bpf_filter(const struct bpf_insn * pc,u_char * p,u_int wirelen,u_int buflen)3174 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3175 {
3176 return (-1); /* "no filter" behaviour */
3177 }
3178
3179 int
bpf_validate(const struct bpf_insn * f,int len)3180 bpf_validate(const struct bpf_insn *f, int len)
3181 {
3182 return (0); /* false */
3183 }
3184
3185 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3186