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_ddb.h"
41 #include "opt_netgraph.h"
42
43 #include <sys/param.h>
44 #include <sys/conf.h>
45 #include <sys/eventhandler.h>
46 #include <sys/fcntl.h>
47 #include <sys/jail.h>
48 #include <sys/ktr.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/mutex.h>
53 #include <sys/time.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/signalvar.h>
57 #include <sys/filio.h>
58 #include <sys/sockio.h>
59 #include <sys/ttycom.h>
60 #include <sys/uio.h>
61 #include <sys/sysent.h>
62 #include <sys/systm.h>
63
64 #include <sys/event.h>
65 #include <sys/file.h>
66 #include <sys/poll.h>
67 #include <sys/proc.h>
68
69 #include <sys/socket.h>
70
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #endif
74
75 #include <net/if.h>
76 #include <net/if_var.h>
77 #include <net/if_private.h>
78 #include <net/if_vlan_var.h>
79 #include <net/if_dl.h>
80 #include <net/bpf.h>
81 #include <net/bpf_buffer.h>
82 #ifdef BPF_JITTER
83 #include <net/bpf_jitter.h>
84 #endif
85 #include <net/bpf_zerocopy.h>
86 #include <net/bpfdesc.h>
87 #include <net/route.h>
88 #include <net/vnet.h>
89
90 #include <netinet/in.h>
91 #include <netinet/if_ether.h>
92 #include <sys/kernel.h>
93 #include <sys/sysctl.h>
94
95 #include <net80211/ieee80211_freebsd.h>
96
97 #include <security/mac/mac_framework.h>
98
99 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
100
101 static const struct bpf_if_ext dead_bpf_if = {
102 .bif_dlist = CK_LIST_HEAD_INITIALIZER()
103 };
104
105 struct bpf_if {
106 #define bif_next bif_ext.bif_next
107 #define bif_dlist bif_ext.bif_dlist
108 struct bpf_if_ext bif_ext; /* public members */
109 u_int bif_dlt; /* link layer type */
110 u_int bif_hdrlen; /* length of link header */
111 struct bpfd_list bif_wlist; /* writer-only list */
112 struct ifnet *bif_ifp; /* corresponding interface */
113 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
114 volatile u_int bif_refcnt;
115 struct epoch_context epoch_ctx;
116 };
117
118 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
119
120 struct bpf_program_buffer {
121 struct epoch_context epoch_ctx;
122 #ifdef BPF_JITTER
123 bpf_jit_filter *func;
124 #endif
125 void *buffer[0];
126 };
127
128 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
129
130 #define PRINET 26 /* interruptible */
131 #define BPF_PRIO_MAX 7
132
133 #define SIZEOF_BPF_HDR(type) \
134 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
135
136 #ifdef COMPAT_FREEBSD32
137 #include <sys/mount.h>
138 #include <compat/freebsd32/freebsd32.h>
139 #define BPF_ALIGNMENT32 sizeof(int32_t)
140 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
141
142 #ifndef BURN_BRIDGES
143 /*
144 * 32-bit version of structure prepended to each packet. We use this header
145 * instead of the standard one for 32-bit streams. We mark the a stream as
146 * 32-bit the first time we see a 32-bit compat ioctl request.
147 */
148 struct bpf_hdr32 {
149 struct timeval32 bh_tstamp; /* time stamp */
150 uint32_t bh_caplen; /* length of captured portion */
151 uint32_t bh_datalen; /* original length of packet */
152 uint16_t bh_hdrlen; /* length of bpf header (this struct
153 plus alignment padding) */
154 };
155 #endif
156
157 struct bpf_program32 {
158 u_int bf_len;
159 uint32_t bf_insns;
160 };
161
162 struct bpf_dltlist32 {
163 u_int bfl_len;
164 u_int bfl_list;
165 };
166
167 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
168 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
169 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
170 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
171 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
172 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
173 #endif
174
175 #define BPF_LOCK() sx_xlock(&bpf_sx)
176 #define BPF_UNLOCK() sx_xunlock(&bpf_sx)
177 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED)
178 /*
179 * bpf_iflist is a list of BPF interface structures, each corresponding to a
180 * specific DLT. The same network interface might have several BPF interface
181 * structures registered by different layers in the stack (i.e., 802.11
182 * frames, ethernet frames, etc).
183 */
184 CK_LIST_HEAD(bpf_iflist, bpf_if);
185 static struct bpf_iflist bpf_iflist;
186 static struct sx bpf_sx; /* bpf global lock */
187 static int bpf_bpfd_cnt;
188
189 static void bpfif_ref(struct bpf_if *);
190 static void bpfif_rele(struct bpf_if *);
191
192 static void bpfd_ref(struct bpf_d *);
193 static void bpfd_rele(struct bpf_d *);
194 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
195 static void bpf_detachd(struct bpf_d *);
196 static void bpf_detachd_locked(struct bpf_d *, bool);
197 static void bpfd_free(epoch_context_t);
198 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
199 struct sockaddr *, int *, struct bpf_d *);
200 static int bpf_setif(struct bpf_d *, struct ifreq *);
201 static void bpf_timed_out(void *);
202 static __inline void
203 bpf_wakeup(struct bpf_d *);
204 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
205 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
206 struct bintime *);
207 static void reset_d(struct bpf_d *);
208 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
209 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
210 static int bpf_setdlt(struct bpf_d *, u_int);
211 static void filt_bpfdetach(struct knote *);
212 static int filt_bpfread(struct knote *, long);
213 static int filt_bpfwrite(struct knote *, long);
214 static void bpf_drvinit(void *);
215 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
216
217 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
218 "bpf sysctl");
219 int bpf_maxinsns = BPF_MAXINSNS;
220 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
221 &bpf_maxinsns, 0, "Maximum bpf program instructions");
222 static int bpf_zerocopy_enable = 0;
223 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
224 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
225 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
226 bpf_stats_sysctl, "bpf statistics portal");
227
228 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
229 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
230 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RWTUN,
231 &VNET_NAME(bpf_optimize_writers), 0,
232 "Do not send packets until BPF program is set");
233
234 static d_open_t bpfopen;
235 static d_read_t bpfread;
236 static d_write_t bpfwrite;
237 static d_ioctl_t bpfioctl;
238 static d_poll_t bpfpoll;
239 static d_kqfilter_t bpfkqfilter;
240
241 static struct cdevsw bpf_cdevsw = {
242 .d_version = D_VERSION,
243 .d_open = bpfopen,
244 .d_read = bpfread,
245 .d_write = bpfwrite,
246 .d_ioctl = bpfioctl,
247 .d_poll = bpfpoll,
248 .d_name = "bpf",
249 .d_kqfilter = bpfkqfilter,
250 };
251
252 static struct filterops bpfread_filtops = {
253 .f_isfd = 1,
254 .f_detach = filt_bpfdetach,
255 .f_event = filt_bpfread,
256 };
257
258 static struct filterops bpfwrite_filtops = {
259 .f_isfd = 1,
260 .f_detach = filt_bpfdetach,
261 .f_event = filt_bpfwrite,
262 };
263
264 /*
265 * LOCKING MODEL USED BY BPF
266 *
267 * Locks:
268 * 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
269 * every bpf_iflist changes, serializes ioctl access to bpf descriptors.
270 * 2) Descriptor lock. Mutex, used to protect BPF buffers and various
271 * structure fields used by bpf_*tap* code.
272 *
273 * Lock order: global lock, then descriptor lock.
274 *
275 * There are several possible consumers:
276 *
277 * 1. The kernel registers interface pointer with bpfattach().
278 * Each call allocates new bpf_if structure, references ifnet pointer
279 * and links bpf_if into bpf_iflist chain. This is protected with global
280 * lock.
281 *
282 * 2. An userland application uses ioctl() call to bpf_d descriptor.
283 * All such call are serialized with global lock. BPF filters can be
284 * changed, but pointer to old filter will be freed using NET_EPOCH_CALL().
285 * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
286 * filter pointers, even if change will happen during bpf_tap execution.
287 * Destroying of bpf_d descriptor also is doing using NET_EPOCH_CALL().
288 *
289 * 3. An userland application can write packets into bpf_d descriptor.
290 * There we need to be sure, that ifnet won't disappear during bpfwrite().
291 *
292 * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
293 * bif_dlist is protected with net_epoch_preempt section. So, it should
294 * be safe to make access to bpf_d descriptor inside the section.
295 *
296 * 5. The kernel invokes bpfdetach() on interface destroying. All lists
297 * are modified with global lock held and actual free() is done using
298 * NET_EPOCH_CALL().
299 */
300
301 static void
bpfif_free(epoch_context_t ctx)302 bpfif_free(epoch_context_t ctx)
303 {
304 struct bpf_if *bp;
305
306 bp = __containerof(ctx, struct bpf_if, epoch_ctx);
307 if_rele(bp->bif_ifp);
308 free(bp, M_BPF);
309 }
310
311 static void
bpfif_ref(struct bpf_if * bp)312 bpfif_ref(struct bpf_if *bp)
313 {
314
315 refcount_acquire(&bp->bif_refcnt);
316 }
317
318 static void
bpfif_rele(struct bpf_if * bp)319 bpfif_rele(struct bpf_if *bp)
320 {
321
322 if (!refcount_release(&bp->bif_refcnt))
323 return;
324 NET_EPOCH_CALL(bpfif_free, &bp->epoch_ctx);
325 }
326
327 static void
bpfd_ref(struct bpf_d * d)328 bpfd_ref(struct bpf_d *d)
329 {
330
331 refcount_acquire(&d->bd_refcnt);
332 }
333
334 static void
bpfd_rele(struct bpf_d * d)335 bpfd_rele(struct bpf_d *d)
336 {
337
338 if (!refcount_release(&d->bd_refcnt))
339 return;
340 NET_EPOCH_CALL(bpfd_free, &d->epoch_ctx);
341 }
342
343 static struct bpf_program_buffer*
bpf_program_buffer_alloc(size_t size,int flags)344 bpf_program_buffer_alloc(size_t size, int flags)
345 {
346
347 return (malloc(sizeof(struct bpf_program_buffer) + size,
348 M_BPF, flags));
349 }
350
351 static void
bpf_program_buffer_free(epoch_context_t ctx)352 bpf_program_buffer_free(epoch_context_t ctx)
353 {
354 struct bpf_program_buffer *ptr;
355
356 ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
357 #ifdef BPF_JITTER
358 if (ptr->func != NULL)
359 bpf_destroy_jit_filter(ptr->func);
360 #endif
361 free(ptr, M_BPF);
362 }
363
364 /*
365 * Wrapper functions for various buffering methods. If the set of buffer
366 * modes expands, we will probably want to introduce a switch data structure
367 * similar to protosw, et.
368 */
369 static void
bpf_append_bytes(struct bpf_d * d,caddr_t buf,u_int offset,void * src,u_int len)370 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
371 u_int len)
372 {
373
374 BPFD_LOCK_ASSERT(d);
375
376 switch (d->bd_bufmode) {
377 case BPF_BUFMODE_BUFFER:
378 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
379
380 case BPF_BUFMODE_ZBUF:
381 counter_u64_add(d->bd_zcopy, 1);
382 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
383
384 default:
385 panic("bpf_buf_append_bytes");
386 }
387 }
388
389 static void
bpf_append_mbuf(struct bpf_d * d,caddr_t buf,u_int offset,void * src,u_int len)390 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
391 u_int len)
392 {
393
394 BPFD_LOCK_ASSERT(d);
395
396 switch (d->bd_bufmode) {
397 case BPF_BUFMODE_BUFFER:
398 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
399
400 case BPF_BUFMODE_ZBUF:
401 counter_u64_add(d->bd_zcopy, 1);
402 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
403
404 default:
405 panic("bpf_buf_append_mbuf");
406 }
407 }
408
409 /*
410 * This function gets called when the free buffer is re-assigned.
411 */
412 static void
bpf_buf_reclaimed(struct bpf_d * d)413 bpf_buf_reclaimed(struct bpf_d *d)
414 {
415
416 BPFD_LOCK_ASSERT(d);
417
418 switch (d->bd_bufmode) {
419 case BPF_BUFMODE_BUFFER:
420 return;
421
422 case BPF_BUFMODE_ZBUF:
423 bpf_zerocopy_buf_reclaimed(d);
424 return;
425
426 default:
427 panic("bpf_buf_reclaimed");
428 }
429 }
430
431 /*
432 * If the buffer mechanism has a way to decide that a held buffer can be made
433 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
434 * returned if the buffer can be discarded, (0) is returned if it cannot.
435 */
436 static int
bpf_canfreebuf(struct bpf_d * d)437 bpf_canfreebuf(struct bpf_d *d)
438 {
439
440 BPFD_LOCK_ASSERT(d);
441
442 switch (d->bd_bufmode) {
443 case BPF_BUFMODE_ZBUF:
444 return (bpf_zerocopy_canfreebuf(d));
445 }
446 return (0);
447 }
448
449 /*
450 * Allow the buffer model to indicate that the current store buffer is
451 * immutable, regardless of the appearance of space. Return (1) if the
452 * buffer is writable, and (0) if not.
453 */
454 static int
bpf_canwritebuf(struct bpf_d * d)455 bpf_canwritebuf(struct bpf_d *d)
456 {
457 BPFD_LOCK_ASSERT(d);
458
459 switch (d->bd_bufmode) {
460 case BPF_BUFMODE_ZBUF:
461 return (bpf_zerocopy_canwritebuf(d));
462 }
463 return (1);
464 }
465
466 /*
467 * Notify buffer model that an attempt to write to the store buffer has
468 * resulted in a dropped packet, in which case the buffer may be considered
469 * full.
470 */
471 static void
bpf_buffull(struct bpf_d * d)472 bpf_buffull(struct bpf_d *d)
473 {
474
475 BPFD_LOCK_ASSERT(d);
476
477 switch (d->bd_bufmode) {
478 case BPF_BUFMODE_ZBUF:
479 bpf_zerocopy_buffull(d);
480 break;
481 }
482 }
483
484 /*
485 * Notify the buffer model that a buffer has moved into the hold position.
486 */
487 void
bpf_bufheld(struct bpf_d * d)488 bpf_bufheld(struct bpf_d *d)
489 {
490
491 BPFD_LOCK_ASSERT(d);
492
493 switch (d->bd_bufmode) {
494 case BPF_BUFMODE_ZBUF:
495 bpf_zerocopy_bufheld(d);
496 break;
497 }
498 }
499
500 static void
bpf_free(struct bpf_d * d)501 bpf_free(struct bpf_d *d)
502 {
503
504 switch (d->bd_bufmode) {
505 case BPF_BUFMODE_BUFFER:
506 return (bpf_buffer_free(d));
507
508 case BPF_BUFMODE_ZBUF:
509 return (bpf_zerocopy_free(d));
510
511 default:
512 panic("bpf_buf_free");
513 }
514 }
515
516 static int
bpf_uiomove(struct bpf_d * d,caddr_t buf,u_int len,struct uio * uio)517 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
518 {
519
520 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
521 return (EOPNOTSUPP);
522 return (bpf_buffer_uiomove(d, buf, len, uio));
523 }
524
525 static int
bpf_ioctl_sblen(struct bpf_d * d,u_int * i)526 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
527 {
528
529 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
530 return (EOPNOTSUPP);
531 return (bpf_buffer_ioctl_sblen(d, i));
532 }
533
534 static int
bpf_ioctl_getzmax(struct thread * td,struct bpf_d * d,size_t * i)535 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
536 {
537
538 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
539 return (EOPNOTSUPP);
540 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
541 }
542
543 static int
bpf_ioctl_rotzbuf(struct thread * td,struct bpf_d * d,struct bpf_zbuf * bz)544 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
545 {
546
547 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
548 return (EOPNOTSUPP);
549 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
550 }
551
552 static int
bpf_ioctl_setzbuf(struct thread * td,struct bpf_d * d,struct bpf_zbuf * bz)553 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
554 {
555
556 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
557 return (EOPNOTSUPP);
558 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
559 }
560
561 /*
562 * General BPF functions.
563 */
564 static int
bpf_movein(struct uio * uio,int linktype,struct ifnet * ifp,struct mbuf ** mp,struct sockaddr * sockp,int * hdrlen,struct bpf_d * d)565 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
566 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
567 {
568 const struct ieee80211_bpf_params *p;
569 struct ether_header *eh;
570 struct mbuf *m;
571 int error;
572 int len;
573 int hlen;
574 int slen;
575
576 /*
577 * Build a sockaddr based on the data link layer type.
578 * We do this at this level because the ethernet header
579 * is copied directly into the data field of the sockaddr.
580 * In the case of SLIP, there is no header and the packet
581 * is forwarded as is.
582 * Also, we are careful to leave room at the front of the mbuf
583 * for the link level header.
584 */
585 switch (linktype) {
586 case DLT_SLIP:
587 sockp->sa_family = AF_INET;
588 hlen = 0;
589 break;
590
591 case DLT_EN10MB:
592 sockp->sa_family = AF_UNSPEC;
593 /* XXX Would MAXLINKHDR be better? */
594 hlen = ETHER_HDR_LEN;
595 break;
596
597 case DLT_FDDI:
598 sockp->sa_family = AF_IMPLINK;
599 hlen = 0;
600 break;
601
602 case DLT_RAW:
603 sockp->sa_family = AF_UNSPEC;
604 hlen = 0;
605 break;
606
607 case DLT_NULL:
608 /*
609 * null interface types require a 4 byte pseudo header which
610 * corresponds to the address family of the packet.
611 */
612 sockp->sa_family = AF_UNSPEC;
613 hlen = 4;
614 break;
615
616 case DLT_ATM_RFC1483:
617 /*
618 * en atm driver requires 4-byte atm pseudo header.
619 * though it isn't standard, vpi:vci needs to be
620 * specified anyway.
621 */
622 sockp->sa_family = AF_UNSPEC;
623 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
624 break;
625
626 case DLT_PPP:
627 sockp->sa_family = AF_UNSPEC;
628 hlen = 4; /* This should match PPP_HDRLEN */
629 break;
630
631 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
632 sockp->sa_family = AF_IEEE80211;
633 hlen = 0;
634 break;
635
636 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
637 sockp->sa_family = AF_IEEE80211;
638 sockp->sa_len = 12; /* XXX != 0 */
639 hlen = sizeof(struct ieee80211_bpf_params);
640 break;
641
642 default:
643 return (EIO);
644 }
645
646 len = uio->uio_resid;
647 if (len < hlen || len - hlen > ifp->if_mtu)
648 return (EMSGSIZE);
649
650 /* Allocate a mbuf, up to MJUM16BYTES bytes, for our write. */
651 m = m_get3(len, M_WAITOK, MT_DATA, M_PKTHDR);
652 if (m == NULL)
653 return (EIO);
654 m->m_pkthdr.len = m->m_len = len;
655 *mp = m;
656
657 error = uiomove(mtod(m, u_char *), len, uio);
658 if (error)
659 goto bad;
660
661 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
662 if (slen == 0) {
663 error = EPERM;
664 goto bad;
665 }
666
667 /* Check for multicast destination */
668 switch (linktype) {
669 case DLT_EN10MB:
670 eh = mtod(m, struct ether_header *);
671 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
672 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
673 ETHER_ADDR_LEN) == 0)
674 m->m_flags |= M_BCAST;
675 else
676 m->m_flags |= M_MCAST;
677 }
678 if (d->bd_hdrcmplt == 0) {
679 memcpy(eh->ether_shost, IF_LLADDR(ifp),
680 sizeof(eh->ether_shost));
681 }
682 break;
683 }
684
685 /*
686 * Make room for link header, and copy it to sockaddr
687 */
688 if (hlen != 0) {
689 if (sockp->sa_family == AF_IEEE80211) {
690 /*
691 * Collect true length from the parameter header
692 * NB: sockp is known to be zero'd so if we do a
693 * short copy unspecified parameters will be
694 * zero.
695 * NB: packet may not be aligned after stripping
696 * bpf params
697 * XXX check ibp_vers
698 */
699 p = mtod(m, const struct ieee80211_bpf_params *);
700 hlen = p->ibp_len;
701 if (hlen > sizeof(sockp->sa_data)) {
702 error = EINVAL;
703 goto bad;
704 }
705 }
706 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
707 }
708 *hdrlen = hlen;
709
710 return (0);
711 bad:
712 m_freem(m);
713 return (error);
714 }
715
716 /*
717 * Attach descriptor to the bpf interface, i.e. make d listen on bp,
718 * then reset its buffers and counters with reset_d().
719 */
720 static void
bpf_attachd(struct bpf_d * d,struct bpf_if * bp)721 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
722 {
723 int op_w;
724
725 BPF_LOCK_ASSERT();
726
727 /*
728 * Save sysctl value to protect from sysctl change
729 * between reads
730 */
731 op_w = V_bpf_optimize_writers || d->bd_writer;
732
733 if (d->bd_bif != NULL)
734 bpf_detachd_locked(d, false);
735 /*
736 * Point d at bp, and add d to the interface's list.
737 * Since there are many applications using BPF for
738 * sending raw packets only (dhcpd, cdpd are good examples)
739 * we can delay adding d to the list of active listeners until
740 * some filter is configured.
741 */
742
743 BPFD_LOCK(d);
744 /*
745 * Hold reference to bpif while descriptor uses this interface.
746 */
747 bpfif_ref(bp);
748 d->bd_bif = bp;
749 if (op_w != 0) {
750 /* Add to writers-only list */
751 CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
752 /*
753 * We decrement bd_writer on every filter set operation.
754 * First BIOCSETF is done by pcap_open_live() to set up
755 * snap length. After that appliation usually sets its own
756 * filter.
757 */
758 d->bd_writer = 2;
759 } else
760 CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
761
762 reset_d(d);
763
764 /* Trigger EVFILT_WRITE events. */
765 bpf_wakeup(d);
766
767 BPFD_UNLOCK(d);
768 bpf_bpfd_cnt++;
769
770 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
771 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
772
773 if (op_w == 0)
774 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
775 }
776
777 /*
778 * Check if we need to upgrade our descriptor @d from write-only mode.
779 */
780 static int
bpf_check_upgrade(u_long cmd,struct bpf_d * d,struct bpf_insn * fcode,int flen)781 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
782 int flen)
783 {
784 int is_snap, need_upgrade;
785
786 /*
787 * Check if we've already upgraded or new filter is empty.
788 */
789 if (d->bd_writer == 0 || fcode == NULL)
790 return (0);
791
792 need_upgrade = 0;
793
794 /*
795 * Check if cmd looks like snaplen setting from
796 * pcap_bpf.c:pcap_open_live().
797 * Note we're not checking .k value here:
798 * while pcap_open_live() definitely sets to non-zero value,
799 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
800 * do not consider upgrading immediately
801 */
802 if (cmd == BIOCSETF && flen == 1 &&
803 fcode[0].code == (BPF_RET | BPF_K))
804 is_snap = 1;
805 else
806 is_snap = 0;
807
808 if (is_snap == 0) {
809 /*
810 * We're setting first filter and it doesn't look like
811 * setting snaplen. We're probably using bpf directly.
812 * Upgrade immediately.
813 */
814 need_upgrade = 1;
815 } else {
816 /*
817 * Do not require upgrade by first BIOCSETF
818 * (used to set snaplen) by pcap_open_live().
819 */
820
821 if (--d->bd_writer == 0) {
822 /*
823 * First snaplen filter has already
824 * been set. This is probably catch-all
825 * filter
826 */
827 need_upgrade = 1;
828 }
829 }
830
831 CTR5(KTR_NET,
832 "%s: filter function set by pid %d, "
833 "bd_writer counter %d, snap %d upgrade %d",
834 __func__, d->bd_pid, d->bd_writer,
835 is_snap, need_upgrade);
836
837 return (need_upgrade);
838 }
839
840 /*
841 * Detach a file from its interface.
842 */
843 static void
bpf_detachd(struct bpf_d * d)844 bpf_detachd(struct bpf_d *d)
845 {
846 BPF_LOCK();
847 bpf_detachd_locked(d, false);
848 BPF_UNLOCK();
849 }
850
851 static void
bpf_detachd_locked(struct bpf_d * d,bool detached_ifp)852 bpf_detachd_locked(struct bpf_d *d, bool detached_ifp)
853 {
854 struct bpf_if *bp;
855 struct ifnet *ifp;
856 int error;
857
858 BPF_LOCK_ASSERT();
859 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
860
861 /* Check if descriptor is attached */
862 if ((bp = d->bd_bif) == NULL)
863 return;
864
865 BPFD_LOCK(d);
866 /* Remove d from the interface's descriptor list. */
867 CK_LIST_REMOVE(d, bd_next);
868 /* Save bd_writer value */
869 error = d->bd_writer;
870 ifp = bp->bif_ifp;
871 d->bd_bif = NULL;
872 if (detached_ifp) {
873 /*
874 * Notify descriptor as it's detached, so that any
875 * sleepers wake up and get ENXIO.
876 */
877 bpf_wakeup(d);
878 }
879 BPFD_UNLOCK(d);
880 bpf_bpfd_cnt--;
881
882 /* Call event handler iff d is attached */
883 if (error == 0)
884 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
885
886 /*
887 * Check if this descriptor had requested promiscuous mode.
888 * If so and ifnet is not detached, turn it off.
889 */
890 if (d->bd_promisc && !detached_ifp) {
891 d->bd_promisc = 0;
892 CURVNET_SET(ifp->if_vnet);
893 error = ifpromisc(ifp, 0);
894 CURVNET_RESTORE();
895 if (error != 0 && error != ENXIO) {
896 /*
897 * ENXIO can happen if a pccard is unplugged
898 * Something is really wrong if we were able to put
899 * the driver into promiscuous mode, but can't
900 * take it out.
901 */
902 if_printf(bp->bif_ifp,
903 "bpf_detach: ifpromisc failed (%d)\n", error);
904 }
905 }
906 bpfif_rele(bp);
907 }
908
909 /*
910 * Close the descriptor by detaching it from its interface,
911 * deallocating its buffers, and marking it free.
912 */
913 static void
bpf_dtor(void * data)914 bpf_dtor(void *data)
915 {
916 struct bpf_d *d = data;
917
918 BPFD_LOCK(d);
919 if (d->bd_state == BPF_WAITING)
920 callout_stop(&d->bd_callout);
921 d->bd_state = BPF_IDLE;
922 BPFD_UNLOCK(d);
923 funsetown(&d->bd_sigio);
924 bpf_detachd(d);
925 #ifdef MAC
926 mac_bpfdesc_destroy(d);
927 #endif /* MAC */
928 seldrain(&d->bd_sel);
929 knlist_destroy(&d->bd_sel.si_note);
930 callout_drain(&d->bd_callout);
931 bpfd_rele(d);
932 }
933
934 /*
935 * Open ethernet device. Returns ENXIO for illegal minor device number,
936 * EBUSY if file is open by another process.
937 */
938 /* ARGSUSED */
939 static int
bpfopen(struct cdev * dev,int flags,int fmt,struct thread * td)940 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
941 {
942 struct bpf_d *d;
943 int error;
944
945 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
946 error = devfs_set_cdevpriv(d, bpf_dtor);
947 if (error != 0) {
948 free(d, M_BPF);
949 return (error);
950 }
951
952 /* Setup counters */
953 d->bd_rcount = counter_u64_alloc(M_WAITOK);
954 d->bd_dcount = counter_u64_alloc(M_WAITOK);
955 d->bd_fcount = counter_u64_alloc(M_WAITOK);
956 d->bd_wcount = counter_u64_alloc(M_WAITOK);
957 d->bd_wfcount = counter_u64_alloc(M_WAITOK);
958 d->bd_wdcount = counter_u64_alloc(M_WAITOK);
959 d->bd_zcopy = counter_u64_alloc(M_WAITOK);
960
961 /*
962 * For historical reasons, perform a one-time initialization call to
963 * the buffer routines, even though we're not yet committed to a
964 * particular buffer method.
965 */
966 bpf_buffer_init(d);
967 if ((flags & FREAD) == 0)
968 d->bd_writer = 2;
969 d->bd_hbuf_in_use = 0;
970 d->bd_bufmode = BPF_BUFMODE_BUFFER;
971 d->bd_sig = SIGIO;
972 d->bd_direction = BPF_D_INOUT;
973 refcount_init(&d->bd_refcnt, 1);
974 BPF_PID_REFRESH(d, td);
975 #ifdef MAC
976 mac_bpfdesc_init(d);
977 mac_bpfdesc_create(td->td_ucred, d);
978 #endif
979 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
980 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
981 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
982
983 /* Disable VLAN pcp tagging. */
984 d->bd_pcp = 0;
985
986 return (0);
987 }
988
989 /*
990 * bpfread - read next chunk of packets from buffers
991 */
992 static int
bpfread(struct cdev * dev,struct uio * uio,int ioflag)993 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
994 {
995 struct bpf_d *d;
996 int error;
997 int non_block;
998 int timed_out;
999
1000 error = devfs_get_cdevpriv((void **)&d);
1001 if (error != 0)
1002 return (error);
1003
1004 /*
1005 * Restrict application to use a buffer the same size as
1006 * as kernel buffers.
1007 */
1008 if (uio->uio_resid != d->bd_bufsize)
1009 return (EINVAL);
1010
1011 non_block = ((ioflag & O_NONBLOCK) != 0);
1012
1013 BPFD_LOCK(d);
1014 BPF_PID_REFRESH_CUR(d);
1015 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1016 BPFD_UNLOCK(d);
1017 return (EOPNOTSUPP);
1018 }
1019 if (d->bd_state == BPF_WAITING)
1020 callout_stop(&d->bd_callout);
1021 timed_out = (d->bd_state == BPF_TIMED_OUT);
1022 d->bd_state = BPF_IDLE;
1023 while (d->bd_hbuf_in_use) {
1024 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1025 PRINET | PCATCH, "bd_hbuf", 0);
1026 if (error != 0) {
1027 BPFD_UNLOCK(d);
1028 return (error);
1029 }
1030 }
1031 /*
1032 * If the hold buffer is empty, then do a timed sleep, which
1033 * ends when the timeout expires or when enough packets
1034 * have arrived to fill the store buffer.
1035 */
1036 while (d->bd_hbuf == NULL) {
1037 if (d->bd_slen != 0) {
1038 /*
1039 * A packet(s) either arrived since the previous
1040 * read or arrived while we were asleep.
1041 */
1042 if (d->bd_immediate || non_block || timed_out) {
1043 /*
1044 * Rotate the buffers and return what's here
1045 * if we are in immediate mode, non-blocking
1046 * flag is set, or this descriptor timed out.
1047 */
1048 ROTATE_BUFFERS(d);
1049 break;
1050 }
1051 }
1052
1053 /*
1054 * No data is available, check to see if the bpf device
1055 * is still pointed at a real interface. If not, return
1056 * ENXIO so that the userland process knows to rebind
1057 * it before using it again.
1058 */
1059 if (d->bd_bif == NULL) {
1060 BPFD_UNLOCK(d);
1061 return (ENXIO);
1062 }
1063
1064 if (non_block) {
1065 BPFD_UNLOCK(d);
1066 return (EWOULDBLOCK);
1067 }
1068 error = msleep(d, &d->bd_lock, PRINET | PCATCH,
1069 "bpf", d->bd_rtout);
1070 if (error == EINTR || error == ERESTART) {
1071 BPFD_UNLOCK(d);
1072 return (error);
1073 }
1074 if (error == EWOULDBLOCK) {
1075 /*
1076 * On a timeout, return what's in the buffer,
1077 * which may be nothing. If there is something
1078 * in the store buffer, we can rotate the buffers.
1079 */
1080 if (d->bd_hbuf)
1081 /*
1082 * We filled up the buffer in between
1083 * getting the timeout and arriving
1084 * here, so we don't need to rotate.
1085 */
1086 break;
1087
1088 if (d->bd_slen == 0) {
1089 BPFD_UNLOCK(d);
1090 return (0);
1091 }
1092 ROTATE_BUFFERS(d);
1093 break;
1094 }
1095 }
1096 /*
1097 * At this point, we know we have something in the hold slot.
1098 */
1099 d->bd_hbuf_in_use = 1;
1100 BPFD_UNLOCK(d);
1101
1102 /*
1103 * Move data from hold buffer into user space.
1104 * We know the entire buffer is transferred since
1105 * we checked above that the read buffer is bpf_bufsize bytes.
1106 *
1107 * We do not have to worry about simultaneous reads because
1108 * we waited for sole access to the hold buffer above.
1109 */
1110 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1111
1112 BPFD_LOCK(d);
1113 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1114 d->bd_fbuf = d->bd_hbuf;
1115 d->bd_hbuf = NULL;
1116 d->bd_hlen = 0;
1117 bpf_buf_reclaimed(d);
1118 d->bd_hbuf_in_use = 0;
1119 wakeup(&d->bd_hbuf_in_use);
1120 BPFD_UNLOCK(d);
1121
1122 return (error);
1123 }
1124
1125 /*
1126 * If there are processes sleeping on this descriptor, wake them up.
1127 */
1128 static __inline void
bpf_wakeup(struct bpf_d * d)1129 bpf_wakeup(struct bpf_d *d)
1130 {
1131
1132 BPFD_LOCK_ASSERT(d);
1133 if (d->bd_state == BPF_WAITING) {
1134 callout_stop(&d->bd_callout);
1135 d->bd_state = BPF_IDLE;
1136 }
1137 wakeup(d);
1138 if (d->bd_async && d->bd_sig && d->bd_sigio)
1139 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1140
1141 selwakeuppri(&d->bd_sel, PRINET);
1142 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1143 }
1144
1145 static void
bpf_timed_out(void * arg)1146 bpf_timed_out(void *arg)
1147 {
1148 struct bpf_d *d = (struct bpf_d *)arg;
1149
1150 BPFD_LOCK_ASSERT(d);
1151
1152 if (callout_pending(&d->bd_callout) ||
1153 !callout_active(&d->bd_callout))
1154 return;
1155 if (d->bd_state == BPF_WAITING) {
1156 d->bd_state = BPF_TIMED_OUT;
1157 if (d->bd_slen != 0)
1158 bpf_wakeup(d);
1159 }
1160 }
1161
1162 static int
bpf_ready(struct bpf_d * d)1163 bpf_ready(struct bpf_d *d)
1164 {
1165
1166 BPFD_LOCK_ASSERT(d);
1167
1168 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1169 return (1);
1170 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1171 d->bd_slen != 0)
1172 return (1);
1173 return (0);
1174 }
1175
1176 static int
bpfwrite(struct cdev * dev,struct uio * uio,int ioflag)1177 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1178 {
1179 struct route ro;
1180 struct sockaddr dst;
1181 struct epoch_tracker et;
1182 struct bpf_if *bp;
1183 struct bpf_d *d;
1184 struct ifnet *ifp;
1185 struct mbuf *m, *mc;
1186 int error, hlen;
1187
1188 error = devfs_get_cdevpriv((void **)&d);
1189 if (error != 0)
1190 return (error);
1191
1192 NET_EPOCH_ENTER(et);
1193 BPFD_LOCK(d);
1194 BPF_PID_REFRESH_CUR(d);
1195 counter_u64_add(d->bd_wcount, 1);
1196 if ((bp = d->bd_bif) == NULL) {
1197 error = ENXIO;
1198 goto out_locked;
1199 }
1200
1201 ifp = bp->bif_ifp;
1202 if ((ifp->if_flags & IFF_UP) == 0) {
1203 error = ENETDOWN;
1204 goto out_locked;
1205 }
1206
1207 if (uio->uio_resid == 0)
1208 goto out_locked;
1209
1210 bzero(&dst, sizeof(dst));
1211 m = NULL;
1212 hlen = 0;
1213
1214 /*
1215 * Take extra reference, unlock d and exit from epoch section,
1216 * since bpf_movein() can sleep.
1217 */
1218 bpfd_ref(d);
1219 NET_EPOCH_EXIT(et);
1220 BPFD_UNLOCK(d);
1221
1222 error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1223 &m, &dst, &hlen, d);
1224
1225 if (error != 0) {
1226 counter_u64_add(d->bd_wdcount, 1);
1227 bpfd_rele(d);
1228 return (error);
1229 }
1230
1231 BPFD_LOCK(d);
1232 /*
1233 * Check that descriptor is still attached to the interface.
1234 * This can happen on bpfdetach(). To avoid access to detached
1235 * ifnet, free mbuf and return ENXIO.
1236 */
1237 if (d->bd_bif == NULL) {
1238 counter_u64_add(d->bd_wdcount, 1);
1239 BPFD_UNLOCK(d);
1240 bpfd_rele(d);
1241 m_freem(m);
1242 return (ENXIO);
1243 }
1244 counter_u64_add(d->bd_wfcount, 1);
1245 if (d->bd_hdrcmplt)
1246 dst.sa_family = pseudo_AF_HDRCMPLT;
1247
1248 if (d->bd_feedback) {
1249 mc = m_dup(m, M_NOWAIT);
1250 if (mc != NULL)
1251 mc->m_pkthdr.rcvif = ifp;
1252 /* Set M_PROMISC for outgoing packets to be discarded. */
1253 if (d->bd_direction == BPF_D_INOUT)
1254 m->m_flags |= M_PROMISC;
1255 } else
1256 mc = NULL;
1257
1258 m->m_pkthdr.len -= hlen;
1259 m->m_len -= hlen;
1260 m->m_data += hlen; /* XXX */
1261
1262 CURVNET_SET(ifp->if_vnet);
1263 #ifdef MAC
1264 mac_bpfdesc_create_mbuf(d, m);
1265 if (mc != NULL)
1266 mac_bpfdesc_create_mbuf(d, mc);
1267 #endif
1268
1269 bzero(&ro, sizeof(ro));
1270 if (hlen != 0) {
1271 ro.ro_prepend = (u_char *)&dst.sa_data;
1272 ro.ro_plen = hlen;
1273 ro.ro_flags = RT_HAS_HEADER;
1274 }
1275
1276 if (d->bd_pcp != 0)
1277 vlan_set_pcp(m, d->bd_pcp);
1278
1279 /* Avoid possible recursion on BPFD_LOCK(). */
1280 NET_EPOCH_ENTER(et);
1281 BPFD_UNLOCK(d);
1282 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1283 if (error)
1284 counter_u64_add(d->bd_wdcount, 1);
1285
1286 if (mc != NULL) {
1287 if (error == 0)
1288 (*ifp->if_input)(ifp, mc);
1289 else
1290 m_freem(mc);
1291 }
1292 NET_EPOCH_EXIT(et);
1293 CURVNET_RESTORE();
1294 bpfd_rele(d);
1295 return (error);
1296
1297 out_locked:
1298 counter_u64_add(d->bd_wdcount, 1);
1299 NET_EPOCH_EXIT(et);
1300 BPFD_UNLOCK(d);
1301 return (error);
1302 }
1303
1304 /*
1305 * Reset a descriptor by flushing its packet buffer and clearing the receive
1306 * and drop counts. This is doable for kernel-only buffers, but with
1307 * zero-copy buffers, we can't write to (or rotate) buffers that are
1308 * currently owned by userspace. It would be nice if we could encapsulate
1309 * this logic in the buffer code rather than here.
1310 */
1311 static void
reset_d(struct bpf_d * d)1312 reset_d(struct bpf_d *d)
1313 {
1314
1315 BPFD_LOCK_ASSERT(d);
1316
1317 while (d->bd_hbuf_in_use)
1318 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1319 "bd_hbuf", 0);
1320 if ((d->bd_hbuf != NULL) &&
1321 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1322 /* Free the hold buffer. */
1323 d->bd_fbuf = d->bd_hbuf;
1324 d->bd_hbuf = NULL;
1325 d->bd_hlen = 0;
1326 bpf_buf_reclaimed(d);
1327 }
1328 if (bpf_canwritebuf(d))
1329 d->bd_slen = 0;
1330 counter_u64_zero(d->bd_rcount);
1331 counter_u64_zero(d->bd_dcount);
1332 counter_u64_zero(d->bd_fcount);
1333 counter_u64_zero(d->bd_wcount);
1334 counter_u64_zero(d->bd_wfcount);
1335 counter_u64_zero(d->bd_wdcount);
1336 counter_u64_zero(d->bd_zcopy);
1337 }
1338
1339 /*
1340 * FIONREAD Check for read packet available.
1341 * BIOCGBLEN Get buffer len [for read()].
1342 * BIOCSETF Set read filter.
1343 * BIOCSETFNR Set read filter without resetting descriptor.
1344 * BIOCSETWF Set write filter.
1345 * BIOCFLUSH Flush read packet buffer.
1346 * BIOCPROMISC Put interface into promiscuous mode.
1347 * BIOCGDLT Get link layer type.
1348 * BIOCGETIF Get interface name.
1349 * BIOCSETIF Set interface.
1350 * BIOCSRTIMEOUT Set read timeout.
1351 * BIOCGRTIMEOUT Get read timeout.
1352 * BIOCGSTATS Get packet stats.
1353 * BIOCIMMEDIATE Set immediate mode.
1354 * BIOCVERSION Get filter language version.
1355 * BIOCGHDRCMPLT Get "header already complete" flag
1356 * BIOCSHDRCMPLT Set "header already complete" flag
1357 * BIOCGDIRECTION Get packet direction flag
1358 * BIOCSDIRECTION Set packet direction flag
1359 * BIOCGTSTAMP Get time stamp format and resolution.
1360 * BIOCSTSTAMP Set time stamp format and resolution.
1361 * BIOCLOCK Set "locked" flag
1362 * BIOCFEEDBACK Set packet feedback mode.
1363 * BIOCSETZBUF Set current zero-copy buffer locations.
1364 * BIOCGETZMAX Get maximum zero-copy buffer size.
1365 * BIOCROTZBUF Force rotation of zero-copy buffer
1366 * BIOCSETBUFMODE Set buffer mode.
1367 * BIOCGETBUFMODE Get current buffer mode.
1368 * BIOCSETVLANPCP Set VLAN PCP tag.
1369 */
1370 /* ARGSUSED */
1371 static int
bpfioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flags,struct thread * td)1372 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1373 struct thread *td)
1374 {
1375 struct bpf_d *d;
1376 int error;
1377
1378 error = devfs_get_cdevpriv((void **)&d);
1379 if (error != 0)
1380 return (error);
1381
1382 /*
1383 * Refresh PID associated with this descriptor.
1384 */
1385 BPFD_LOCK(d);
1386 BPF_PID_REFRESH(d, td);
1387 if (d->bd_state == BPF_WAITING)
1388 callout_stop(&d->bd_callout);
1389 d->bd_state = BPF_IDLE;
1390 BPFD_UNLOCK(d);
1391
1392 if (d->bd_locked == 1) {
1393 switch (cmd) {
1394 case BIOCGBLEN:
1395 case BIOCFLUSH:
1396 case BIOCGDLT:
1397 case BIOCGDLTLIST:
1398 #ifdef COMPAT_FREEBSD32
1399 case BIOCGDLTLIST32:
1400 #endif
1401 case BIOCGETIF:
1402 case BIOCGRTIMEOUT:
1403 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1404 case BIOCGRTIMEOUT32:
1405 #endif
1406 case BIOCGSTATS:
1407 case BIOCVERSION:
1408 case BIOCGRSIG:
1409 case BIOCGHDRCMPLT:
1410 case BIOCSTSTAMP:
1411 case BIOCFEEDBACK:
1412 case FIONREAD:
1413 case BIOCLOCK:
1414 case BIOCSRTIMEOUT:
1415 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1416 case BIOCSRTIMEOUT32:
1417 #endif
1418 case BIOCIMMEDIATE:
1419 case TIOCGPGRP:
1420 case BIOCROTZBUF:
1421 break;
1422 default:
1423 return (EPERM);
1424 }
1425 }
1426 #ifdef COMPAT_FREEBSD32
1427 /*
1428 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1429 * that it will get 32-bit packet headers.
1430 */
1431 switch (cmd) {
1432 case BIOCSETF32:
1433 case BIOCSETFNR32:
1434 case BIOCSETWF32:
1435 case BIOCGDLTLIST32:
1436 case BIOCGRTIMEOUT32:
1437 case BIOCSRTIMEOUT32:
1438 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1439 BPFD_LOCK(d);
1440 d->bd_compat32 = 1;
1441 BPFD_UNLOCK(d);
1442 }
1443 }
1444 #endif
1445
1446 CURVNET_SET(TD_TO_VNET(td));
1447 switch (cmd) {
1448 default:
1449 error = EINVAL;
1450 break;
1451
1452 /*
1453 * Check for read packet available.
1454 */
1455 case FIONREAD:
1456 {
1457 int n;
1458
1459 BPFD_LOCK(d);
1460 n = d->bd_slen;
1461 while (d->bd_hbuf_in_use)
1462 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1463 PRINET, "bd_hbuf", 0);
1464 if (d->bd_hbuf)
1465 n += d->bd_hlen;
1466 BPFD_UNLOCK(d);
1467
1468 *(int *)addr = n;
1469 break;
1470 }
1471
1472 /*
1473 * Get buffer len [for read()].
1474 */
1475 case BIOCGBLEN:
1476 BPFD_LOCK(d);
1477 *(u_int *)addr = d->bd_bufsize;
1478 BPFD_UNLOCK(d);
1479 break;
1480
1481 /*
1482 * Set buffer length.
1483 */
1484 case BIOCSBLEN:
1485 error = bpf_ioctl_sblen(d, (u_int *)addr);
1486 break;
1487
1488 /*
1489 * Set link layer read filter.
1490 */
1491 case BIOCSETF:
1492 case BIOCSETFNR:
1493 case BIOCSETWF:
1494 #ifdef COMPAT_FREEBSD32
1495 case BIOCSETF32:
1496 case BIOCSETFNR32:
1497 case BIOCSETWF32:
1498 #endif
1499 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1500 break;
1501
1502 /*
1503 * Flush read packet buffer.
1504 */
1505 case BIOCFLUSH:
1506 BPFD_LOCK(d);
1507 reset_d(d);
1508 BPFD_UNLOCK(d);
1509 break;
1510
1511 /*
1512 * Put interface into promiscuous mode.
1513 */
1514 case BIOCPROMISC:
1515 BPF_LOCK();
1516 if (d->bd_bif == NULL) {
1517 /*
1518 * No interface attached yet.
1519 */
1520 error = EINVAL;
1521 } else if (d->bd_promisc == 0) {
1522 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1523 if (error == 0)
1524 d->bd_promisc = 1;
1525 }
1526 BPF_UNLOCK();
1527 break;
1528
1529 /*
1530 * Get current data link type.
1531 */
1532 case BIOCGDLT:
1533 BPF_LOCK();
1534 if (d->bd_bif == NULL)
1535 error = EINVAL;
1536 else
1537 *(u_int *)addr = d->bd_bif->bif_dlt;
1538 BPF_UNLOCK();
1539 break;
1540
1541 /*
1542 * Get a list of supported data link types.
1543 */
1544 #ifdef COMPAT_FREEBSD32
1545 case BIOCGDLTLIST32:
1546 {
1547 struct bpf_dltlist32 *list32;
1548 struct bpf_dltlist dltlist;
1549
1550 list32 = (struct bpf_dltlist32 *)addr;
1551 dltlist.bfl_len = list32->bfl_len;
1552 dltlist.bfl_list = PTRIN(list32->bfl_list);
1553 BPF_LOCK();
1554 if (d->bd_bif == NULL)
1555 error = EINVAL;
1556 else {
1557 error = bpf_getdltlist(d, &dltlist);
1558 if (error == 0)
1559 list32->bfl_len = dltlist.bfl_len;
1560 }
1561 BPF_UNLOCK();
1562 break;
1563 }
1564 #endif
1565
1566 case BIOCGDLTLIST:
1567 BPF_LOCK();
1568 if (d->bd_bif == NULL)
1569 error = EINVAL;
1570 else
1571 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1572 BPF_UNLOCK();
1573 break;
1574
1575 /*
1576 * Set data link type.
1577 */
1578 case BIOCSDLT:
1579 BPF_LOCK();
1580 if (d->bd_bif == NULL)
1581 error = EINVAL;
1582 else
1583 error = bpf_setdlt(d, *(u_int *)addr);
1584 BPF_UNLOCK();
1585 break;
1586
1587 /*
1588 * Get interface name.
1589 */
1590 case BIOCGETIF:
1591 BPF_LOCK();
1592 if (d->bd_bif == NULL)
1593 error = EINVAL;
1594 else {
1595 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1596 struct ifreq *const ifr = (struct ifreq *)addr;
1597
1598 strlcpy(ifr->ifr_name, ifp->if_xname,
1599 sizeof(ifr->ifr_name));
1600 }
1601 BPF_UNLOCK();
1602 break;
1603
1604 /*
1605 * Set interface.
1606 */
1607 case BIOCSETIF:
1608 {
1609 int alloc_buf, size;
1610
1611 /*
1612 * Behavior here depends on the buffering model. If
1613 * we're using kernel memory buffers, then we can
1614 * allocate them here. If we're using zero-copy,
1615 * then the user process must have registered buffers
1616 * by the time we get here.
1617 */
1618 alloc_buf = 0;
1619 BPFD_LOCK(d);
1620 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1621 d->bd_sbuf == NULL)
1622 alloc_buf = 1;
1623 BPFD_UNLOCK(d);
1624 if (alloc_buf) {
1625 size = d->bd_bufsize;
1626 error = bpf_buffer_ioctl_sblen(d, &size);
1627 if (error != 0)
1628 break;
1629 }
1630 BPF_LOCK();
1631 error = bpf_setif(d, (struct ifreq *)addr);
1632 BPF_UNLOCK();
1633 break;
1634 }
1635
1636 /*
1637 * Set read timeout.
1638 */
1639 case BIOCSRTIMEOUT:
1640 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1641 case BIOCSRTIMEOUT32:
1642 #endif
1643 {
1644 struct timeval *tv = (struct timeval *)addr;
1645 #if defined(COMPAT_FREEBSD32)
1646 struct timeval32 *tv32;
1647 struct timeval tv64;
1648
1649 if (cmd == BIOCSRTIMEOUT32) {
1650 tv32 = (struct timeval32 *)addr;
1651 tv = &tv64;
1652 tv->tv_sec = tv32->tv_sec;
1653 tv->tv_usec = tv32->tv_usec;
1654 } else
1655 #endif
1656 tv = (struct timeval *)addr;
1657
1658 /*
1659 * Subtract 1 tick from tvtohz() since this isn't
1660 * a one-shot timer.
1661 */
1662 if ((error = itimerfix(tv)) == 0)
1663 d->bd_rtout = tvtohz(tv) - 1;
1664 break;
1665 }
1666
1667 /*
1668 * Get read timeout.
1669 */
1670 case BIOCGRTIMEOUT:
1671 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1672 case BIOCGRTIMEOUT32:
1673 #endif
1674 {
1675 struct timeval *tv;
1676 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1677 struct timeval32 *tv32;
1678 struct timeval tv64;
1679
1680 if (cmd == BIOCGRTIMEOUT32)
1681 tv = &tv64;
1682 else
1683 #endif
1684 tv = (struct timeval *)addr;
1685
1686 tv->tv_sec = d->bd_rtout / hz;
1687 tv->tv_usec = (d->bd_rtout % hz) * tick;
1688 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1689 if (cmd == BIOCGRTIMEOUT32) {
1690 tv32 = (struct timeval32 *)addr;
1691 tv32->tv_sec = tv->tv_sec;
1692 tv32->tv_usec = tv->tv_usec;
1693 }
1694 #endif
1695
1696 break;
1697 }
1698
1699 /*
1700 * Get packet stats.
1701 */
1702 case BIOCGSTATS:
1703 {
1704 struct bpf_stat *bs = (struct bpf_stat *)addr;
1705
1706 /* XXXCSJP overflow */
1707 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1708 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1709 break;
1710 }
1711
1712 /*
1713 * Set immediate mode.
1714 */
1715 case BIOCIMMEDIATE:
1716 BPFD_LOCK(d);
1717 d->bd_immediate = *(u_int *)addr;
1718 BPFD_UNLOCK(d);
1719 break;
1720
1721 case BIOCVERSION:
1722 {
1723 struct bpf_version *bv = (struct bpf_version *)addr;
1724
1725 bv->bv_major = BPF_MAJOR_VERSION;
1726 bv->bv_minor = BPF_MINOR_VERSION;
1727 break;
1728 }
1729
1730 /*
1731 * Get "header already complete" flag
1732 */
1733 case BIOCGHDRCMPLT:
1734 BPFD_LOCK(d);
1735 *(u_int *)addr = d->bd_hdrcmplt;
1736 BPFD_UNLOCK(d);
1737 break;
1738
1739 /*
1740 * Set "header already complete" flag
1741 */
1742 case BIOCSHDRCMPLT:
1743 BPFD_LOCK(d);
1744 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1745 BPFD_UNLOCK(d);
1746 break;
1747
1748 /*
1749 * Get packet direction flag
1750 */
1751 case BIOCGDIRECTION:
1752 BPFD_LOCK(d);
1753 *(u_int *)addr = d->bd_direction;
1754 BPFD_UNLOCK(d);
1755 break;
1756
1757 /*
1758 * Set packet direction flag
1759 */
1760 case BIOCSDIRECTION:
1761 {
1762 u_int direction;
1763
1764 direction = *(u_int *)addr;
1765 switch (direction) {
1766 case BPF_D_IN:
1767 case BPF_D_INOUT:
1768 case BPF_D_OUT:
1769 BPFD_LOCK(d);
1770 d->bd_direction = direction;
1771 BPFD_UNLOCK(d);
1772 break;
1773 default:
1774 error = EINVAL;
1775 }
1776 }
1777 break;
1778
1779 /*
1780 * Get packet timestamp format and resolution.
1781 */
1782 case BIOCGTSTAMP:
1783 BPFD_LOCK(d);
1784 *(u_int *)addr = d->bd_tstamp;
1785 BPFD_UNLOCK(d);
1786 break;
1787
1788 /*
1789 * Set packet timestamp format and resolution.
1790 */
1791 case BIOCSTSTAMP:
1792 {
1793 u_int func;
1794
1795 func = *(u_int *)addr;
1796 if (BPF_T_VALID(func))
1797 d->bd_tstamp = func;
1798 else
1799 error = EINVAL;
1800 }
1801 break;
1802
1803 case BIOCFEEDBACK:
1804 BPFD_LOCK(d);
1805 d->bd_feedback = *(u_int *)addr;
1806 BPFD_UNLOCK(d);
1807 break;
1808
1809 case BIOCLOCK:
1810 BPFD_LOCK(d);
1811 d->bd_locked = 1;
1812 BPFD_UNLOCK(d);
1813 break;
1814
1815 case FIONBIO: /* Non-blocking I/O */
1816 break;
1817
1818 case FIOASYNC: /* Send signal on receive packets */
1819 BPFD_LOCK(d);
1820 d->bd_async = *(int *)addr;
1821 BPFD_UNLOCK(d);
1822 break;
1823
1824 case FIOSETOWN:
1825 /*
1826 * XXX: Add some sort of locking here?
1827 * fsetown() can sleep.
1828 */
1829 error = fsetown(*(int *)addr, &d->bd_sigio);
1830 break;
1831
1832 case FIOGETOWN:
1833 BPFD_LOCK(d);
1834 *(int *)addr = fgetown(&d->bd_sigio);
1835 BPFD_UNLOCK(d);
1836 break;
1837
1838 /* This is deprecated, FIOSETOWN should be used instead. */
1839 case TIOCSPGRP:
1840 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1841 break;
1842
1843 /* This is deprecated, FIOGETOWN should be used instead. */
1844 case TIOCGPGRP:
1845 *(int *)addr = -fgetown(&d->bd_sigio);
1846 break;
1847
1848 case BIOCSRSIG: /* Set receive signal */
1849 {
1850 u_int sig;
1851
1852 sig = *(u_int *)addr;
1853
1854 if (sig >= NSIG)
1855 error = EINVAL;
1856 else {
1857 BPFD_LOCK(d);
1858 d->bd_sig = sig;
1859 BPFD_UNLOCK(d);
1860 }
1861 break;
1862 }
1863 case BIOCGRSIG:
1864 BPFD_LOCK(d);
1865 *(u_int *)addr = d->bd_sig;
1866 BPFD_UNLOCK(d);
1867 break;
1868
1869 case BIOCGETBUFMODE:
1870 BPFD_LOCK(d);
1871 *(u_int *)addr = d->bd_bufmode;
1872 BPFD_UNLOCK(d);
1873 break;
1874
1875 case BIOCSETBUFMODE:
1876 /*
1877 * Allow the buffering mode to be changed as long as we
1878 * haven't yet committed to a particular mode. Our
1879 * definition of commitment, for now, is whether or not a
1880 * buffer has been allocated or an interface attached, since
1881 * that's the point where things get tricky.
1882 */
1883 switch (*(u_int *)addr) {
1884 case BPF_BUFMODE_BUFFER:
1885 break;
1886
1887 case BPF_BUFMODE_ZBUF:
1888 if (bpf_zerocopy_enable)
1889 break;
1890 /* FALLSTHROUGH */
1891
1892 default:
1893 CURVNET_RESTORE();
1894 return (EINVAL);
1895 }
1896
1897 BPFD_LOCK(d);
1898 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1899 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1900 BPFD_UNLOCK(d);
1901 CURVNET_RESTORE();
1902 return (EBUSY);
1903 }
1904 d->bd_bufmode = *(u_int *)addr;
1905 BPFD_UNLOCK(d);
1906 break;
1907
1908 case BIOCGETZMAX:
1909 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1910 break;
1911
1912 case BIOCSETZBUF:
1913 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1914 break;
1915
1916 case BIOCROTZBUF:
1917 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1918 break;
1919
1920 case BIOCSETVLANPCP:
1921 {
1922 u_int pcp;
1923
1924 pcp = *(u_int *)addr;
1925 if (pcp > BPF_PRIO_MAX || pcp < 0) {
1926 error = EINVAL;
1927 break;
1928 }
1929 d->bd_pcp = pcp;
1930 break;
1931 }
1932 }
1933 CURVNET_RESTORE();
1934 return (error);
1935 }
1936
1937 /*
1938 * Set d's packet filter program to fp. If this file already has a filter,
1939 * free it and replace it. Returns EINVAL for bogus requests.
1940 *
1941 * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1942 * calls.
1943 */
1944 static int
bpf_setf(struct bpf_d * d,struct bpf_program * fp,u_long cmd)1945 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1946 {
1947 #ifdef COMPAT_FREEBSD32
1948 struct bpf_program fp_swab;
1949 struct bpf_program32 *fp32;
1950 #endif
1951 struct bpf_program_buffer *fcode;
1952 struct bpf_insn *filter;
1953 #ifdef BPF_JITTER
1954 bpf_jit_filter *jfunc;
1955 #endif
1956 size_t size;
1957 u_int flen;
1958 bool track_event;
1959
1960 #ifdef COMPAT_FREEBSD32
1961 switch (cmd) {
1962 case BIOCSETF32:
1963 case BIOCSETWF32:
1964 case BIOCSETFNR32:
1965 fp32 = (struct bpf_program32 *)fp;
1966 fp_swab.bf_len = fp32->bf_len;
1967 fp_swab.bf_insns =
1968 (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1969 fp = &fp_swab;
1970 switch (cmd) {
1971 case BIOCSETF32:
1972 cmd = BIOCSETF;
1973 break;
1974 case BIOCSETWF32:
1975 cmd = BIOCSETWF;
1976 break;
1977 }
1978 break;
1979 }
1980 #endif
1981
1982 filter = NULL;
1983 #ifdef BPF_JITTER
1984 jfunc = NULL;
1985 #endif
1986 /*
1987 * Check new filter validness before acquiring any locks.
1988 * Allocate memory for new filter, if needed.
1989 */
1990 flen = fp->bf_len;
1991 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1992 return (EINVAL);
1993 size = flen * sizeof(*fp->bf_insns);
1994 if (size > 0) {
1995 /* We're setting up new filter. Copy and check actual data. */
1996 fcode = bpf_program_buffer_alloc(size, M_WAITOK);
1997 filter = (struct bpf_insn *)fcode->buffer;
1998 if (copyin(fp->bf_insns, filter, size) != 0 ||
1999 !bpf_validate(filter, flen)) {
2000 free(fcode, M_BPF);
2001 return (EINVAL);
2002 }
2003 #ifdef BPF_JITTER
2004 if (cmd != BIOCSETWF) {
2005 /*
2006 * Filter is copied inside fcode and is
2007 * perfectly valid.
2008 */
2009 jfunc = bpf_jitter(filter, flen);
2010 }
2011 #endif
2012 }
2013
2014 track_event = false;
2015 fcode = NULL;
2016
2017 BPF_LOCK();
2018 BPFD_LOCK(d);
2019 /* Set up new filter. */
2020 if (cmd == BIOCSETWF) {
2021 if (d->bd_wfilter != NULL) {
2022 fcode = __containerof((void *)d->bd_wfilter,
2023 struct bpf_program_buffer, buffer);
2024 #ifdef BPF_JITTER
2025 fcode->func = NULL;
2026 #endif
2027 }
2028 d->bd_wfilter = filter;
2029 } else {
2030 if (d->bd_rfilter != NULL) {
2031 fcode = __containerof((void *)d->bd_rfilter,
2032 struct bpf_program_buffer, buffer);
2033 #ifdef BPF_JITTER
2034 fcode->func = d->bd_bfilter;
2035 #endif
2036 }
2037 d->bd_rfilter = filter;
2038 #ifdef BPF_JITTER
2039 d->bd_bfilter = jfunc;
2040 #endif
2041 if (cmd == BIOCSETF)
2042 reset_d(d);
2043
2044 if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2045 /*
2046 * Filter can be set several times without
2047 * specifying interface. In this case just mark d
2048 * as reader.
2049 */
2050 d->bd_writer = 0;
2051 if (d->bd_bif != NULL) {
2052 /*
2053 * Remove descriptor from writers-only list
2054 * and add it to active readers list.
2055 */
2056 CK_LIST_REMOVE(d, bd_next);
2057 CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2058 d, bd_next);
2059 CTR2(KTR_NET,
2060 "%s: upgrade required by pid %d",
2061 __func__, d->bd_pid);
2062 track_event = true;
2063 }
2064 }
2065 }
2066 BPFD_UNLOCK(d);
2067
2068 if (fcode != NULL)
2069 NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2070
2071 if (track_event)
2072 EVENTHANDLER_INVOKE(bpf_track,
2073 d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2074
2075 BPF_UNLOCK();
2076 return (0);
2077 }
2078
2079 /*
2080 * Detach a file from its current interface (if attached at all) and attach
2081 * to the interface indicated by the name stored in ifr.
2082 * Return an errno or 0.
2083 */
2084 static int
bpf_setif(struct bpf_d * d,struct ifreq * ifr)2085 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2086 {
2087 struct bpf_if *bp;
2088 struct ifnet *theywant;
2089
2090 BPF_LOCK_ASSERT();
2091
2092 theywant = ifunit(ifr->ifr_name);
2093 if (theywant == NULL || theywant->if_bpf == NULL)
2094 return (ENXIO);
2095
2096 bp = theywant->if_bpf;
2097 /*
2098 * At this point, we expect the buffer is already allocated. If not,
2099 * return an error.
2100 */
2101 switch (d->bd_bufmode) {
2102 case BPF_BUFMODE_BUFFER:
2103 case BPF_BUFMODE_ZBUF:
2104 if (d->bd_sbuf == NULL)
2105 return (EINVAL);
2106 break;
2107
2108 default:
2109 panic("bpf_setif: bufmode %d", d->bd_bufmode);
2110 }
2111 if (bp != d->bd_bif)
2112 bpf_attachd(d, bp);
2113 else {
2114 BPFD_LOCK(d);
2115 reset_d(d);
2116 BPFD_UNLOCK(d);
2117 }
2118 return (0);
2119 }
2120
2121 /*
2122 * Support for select() and poll() system calls
2123 *
2124 * Return true iff the specific operation will not block indefinitely.
2125 * Otherwise, return false but make a note that a selwakeup() must be done.
2126 */
2127 static int
bpfpoll(struct cdev * dev,int events,struct thread * td)2128 bpfpoll(struct cdev *dev, int events, struct thread *td)
2129 {
2130 struct bpf_d *d;
2131 int revents;
2132
2133 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2134 return (events &
2135 (POLLHUP | POLLIN | POLLRDNORM | POLLOUT | POLLWRNORM));
2136
2137 /*
2138 * Refresh PID associated with this descriptor.
2139 */
2140 revents = events & (POLLOUT | POLLWRNORM);
2141 BPFD_LOCK(d);
2142 BPF_PID_REFRESH(d, td);
2143 if (events & (POLLIN | POLLRDNORM)) {
2144 if (bpf_ready(d))
2145 revents |= events & (POLLIN | POLLRDNORM);
2146 else {
2147 selrecord(td, &d->bd_sel);
2148 /* Start the read timeout if necessary. */
2149 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2150 callout_reset(&d->bd_callout, d->bd_rtout,
2151 bpf_timed_out, d);
2152 d->bd_state = BPF_WAITING;
2153 }
2154 }
2155 }
2156 BPFD_UNLOCK(d);
2157 return (revents);
2158 }
2159
2160 /*
2161 * Support for kevent() system call. Register EVFILT_READ filters and
2162 * reject all others.
2163 */
2164 int
bpfkqfilter(struct cdev * dev,struct knote * kn)2165 bpfkqfilter(struct cdev *dev, struct knote *kn)
2166 {
2167 struct bpf_d *d;
2168
2169 if (devfs_get_cdevpriv((void **)&d) != 0)
2170 return (1);
2171
2172 switch (kn->kn_filter) {
2173 case EVFILT_READ:
2174 kn->kn_fop = &bpfread_filtops;
2175 break;
2176
2177 case EVFILT_WRITE:
2178 kn->kn_fop = &bpfwrite_filtops;
2179 break;
2180
2181 default:
2182 return (1);
2183 }
2184
2185 /*
2186 * Refresh PID associated with this descriptor.
2187 */
2188 BPFD_LOCK(d);
2189 BPF_PID_REFRESH_CUR(d);
2190 kn->kn_hook = d;
2191 knlist_add(&d->bd_sel.si_note, kn, 1);
2192 BPFD_UNLOCK(d);
2193
2194 return (0);
2195 }
2196
2197 static void
filt_bpfdetach(struct knote * kn)2198 filt_bpfdetach(struct knote *kn)
2199 {
2200 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2201
2202 knlist_remove(&d->bd_sel.si_note, kn, 0);
2203 }
2204
2205 static int
filt_bpfread(struct knote * kn,long hint)2206 filt_bpfread(struct knote *kn, long hint)
2207 {
2208 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2209 int ready;
2210
2211 BPFD_LOCK_ASSERT(d);
2212 ready = bpf_ready(d);
2213 if (ready) {
2214 kn->kn_data = d->bd_slen;
2215 /*
2216 * Ignore the hold buffer if it is being copied to user space.
2217 */
2218 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2219 kn->kn_data += d->bd_hlen;
2220 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2221 callout_reset(&d->bd_callout, d->bd_rtout,
2222 bpf_timed_out, d);
2223 d->bd_state = BPF_WAITING;
2224 }
2225
2226 return (ready);
2227 }
2228
2229 static int
filt_bpfwrite(struct knote * kn,long hint)2230 filt_bpfwrite(struct knote *kn, long hint)
2231 {
2232 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2233
2234 BPFD_LOCK_ASSERT(d);
2235
2236 if (d->bd_bif == NULL) {
2237 kn->kn_data = 0;
2238 return (0);
2239 } else {
2240 kn->kn_data = d->bd_bif->bif_ifp->if_mtu;
2241 return (1);
2242 }
2243 }
2244
2245 #define BPF_TSTAMP_NONE 0
2246 #define BPF_TSTAMP_FAST 1
2247 #define BPF_TSTAMP_NORMAL 2
2248 #define BPF_TSTAMP_EXTERN 3
2249
2250 static int
bpf_ts_quality(int tstype)2251 bpf_ts_quality(int tstype)
2252 {
2253
2254 if (tstype == BPF_T_NONE)
2255 return (BPF_TSTAMP_NONE);
2256 if ((tstype & BPF_T_FAST) != 0)
2257 return (BPF_TSTAMP_FAST);
2258
2259 return (BPF_TSTAMP_NORMAL);
2260 }
2261
2262 static int
bpf_gettime(struct bintime * bt,int tstype,struct mbuf * m)2263 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2264 {
2265 struct timespec ts;
2266 struct m_tag *tag;
2267 int quality;
2268
2269 quality = bpf_ts_quality(tstype);
2270 if (quality == BPF_TSTAMP_NONE)
2271 return (quality);
2272
2273 if (m != NULL) {
2274 if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) {
2275 mbuf_tstmp2timespec(m, &ts);
2276 timespec2bintime(&ts, bt);
2277 return (BPF_TSTAMP_EXTERN);
2278 }
2279 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2280 if (tag != NULL) {
2281 *bt = *(struct bintime *)(tag + 1);
2282 return (BPF_TSTAMP_EXTERN);
2283 }
2284 }
2285 if (quality == BPF_TSTAMP_NORMAL)
2286 binuptime(bt);
2287 else
2288 getbinuptime(bt);
2289
2290 return (quality);
2291 }
2292
2293 /*
2294 * Incoming linkage from device drivers. Process the packet pkt, of length
2295 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2296 * by each process' filter, and if accepted, stashed into the corresponding
2297 * buffer.
2298 */
2299 void
bpf_tap(struct bpf_if * bp,u_char * pkt,u_int pktlen)2300 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2301 {
2302 struct epoch_tracker et;
2303 struct bintime bt;
2304 struct bpf_d *d;
2305 #ifdef BPF_JITTER
2306 bpf_jit_filter *bf;
2307 #endif
2308 u_int slen;
2309 int gottime;
2310
2311 gottime = BPF_TSTAMP_NONE;
2312 NET_EPOCH_ENTER(et);
2313 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2314 counter_u64_add(d->bd_rcount, 1);
2315 /*
2316 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2317 * is no way for the caller to indiciate to us whether this
2318 * packet is inbound or outbound. In the bpf_mtap() routines,
2319 * we use the interface pointers on the mbuf to figure it out.
2320 */
2321 #ifdef BPF_JITTER
2322 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2323 if (bf != NULL)
2324 slen = (*(bf->func))(pkt, pktlen, pktlen);
2325 else
2326 #endif
2327 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2328 if (slen != 0) {
2329 /*
2330 * Filter matches. Let's to acquire write lock.
2331 */
2332 BPFD_LOCK(d);
2333 counter_u64_add(d->bd_fcount, 1);
2334 if (gottime < bpf_ts_quality(d->bd_tstamp))
2335 gottime = bpf_gettime(&bt, d->bd_tstamp,
2336 NULL);
2337 #ifdef MAC
2338 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2339 #endif
2340 catchpacket(d, pkt, pktlen, slen,
2341 bpf_append_bytes, &bt);
2342 BPFD_UNLOCK(d);
2343 }
2344 }
2345 NET_EPOCH_EXIT(et);
2346 }
2347
2348 void
bpf_tap_if(if_t ifp,u_char * pkt,u_int pktlen)2349 bpf_tap_if(if_t ifp, u_char *pkt, u_int pktlen)
2350 {
2351 if (bpf_peers_present(ifp->if_bpf))
2352 bpf_tap(ifp->if_bpf, pkt, pktlen);
2353 }
2354
2355 #define BPF_CHECK_DIRECTION(d, r, i) \
2356 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2357 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2358
2359 /*
2360 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2361 * Locking model is explained in bpf_tap().
2362 */
2363 void
bpf_mtap(struct bpf_if * bp,struct mbuf * m)2364 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2365 {
2366 struct epoch_tracker et;
2367 struct bintime bt;
2368 struct bpf_d *d;
2369 #ifdef BPF_JITTER
2370 bpf_jit_filter *bf;
2371 #endif
2372 u_int pktlen, slen;
2373 int gottime;
2374
2375 /* Skip outgoing duplicate packets. */
2376 if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2377 m->m_flags &= ~M_PROMISC;
2378 return;
2379 }
2380
2381 pktlen = m_length(m, NULL);
2382 gottime = BPF_TSTAMP_NONE;
2383
2384 NET_EPOCH_ENTER(et);
2385 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2386 if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2387 continue;
2388 counter_u64_add(d->bd_rcount, 1);
2389 #ifdef BPF_JITTER
2390 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2391 /* XXX We cannot handle multiple mbufs. */
2392 if (bf != NULL && m->m_next == NULL)
2393 slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2394 pktlen);
2395 else
2396 #endif
2397 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2398 if (slen != 0) {
2399 BPFD_LOCK(d);
2400
2401 counter_u64_add(d->bd_fcount, 1);
2402 if (gottime < bpf_ts_quality(d->bd_tstamp))
2403 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2404 #ifdef MAC
2405 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2406 #endif
2407 catchpacket(d, (u_char *)m, pktlen, slen,
2408 bpf_append_mbuf, &bt);
2409 BPFD_UNLOCK(d);
2410 }
2411 }
2412 NET_EPOCH_EXIT(et);
2413 }
2414
2415 void
bpf_mtap_if(if_t ifp,struct mbuf * m)2416 bpf_mtap_if(if_t ifp, struct mbuf *m)
2417 {
2418 if (bpf_peers_present(ifp->if_bpf)) {
2419 M_ASSERTVALID(m);
2420 bpf_mtap(ifp->if_bpf, m);
2421 }
2422 }
2423
2424 /*
2425 * Incoming linkage from device drivers, when packet is in
2426 * an mbuf chain and to be prepended by a contiguous header.
2427 */
2428 void
bpf_mtap2(struct bpf_if * bp,void * data,u_int dlen,struct mbuf * m)2429 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2430 {
2431 struct epoch_tracker et;
2432 struct bintime bt;
2433 struct mbuf mb;
2434 struct bpf_d *d;
2435 u_int pktlen, slen;
2436 int gottime;
2437
2438 /* Skip outgoing duplicate packets. */
2439 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2440 m->m_flags &= ~M_PROMISC;
2441 return;
2442 }
2443
2444 pktlen = m_length(m, NULL);
2445 /*
2446 * Craft on-stack mbuf suitable for passing to bpf_filter.
2447 * Note that we cut corners here; we only setup what's
2448 * absolutely needed--this mbuf should never go anywhere else.
2449 */
2450 mb.m_flags = 0;
2451 mb.m_next = m;
2452 mb.m_data = data;
2453 mb.m_len = dlen;
2454 pktlen += dlen;
2455
2456 gottime = BPF_TSTAMP_NONE;
2457
2458 NET_EPOCH_ENTER(et);
2459 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2460 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2461 continue;
2462 counter_u64_add(d->bd_rcount, 1);
2463 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2464 if (slen != 0) {
2465 BPFD_LOCK(d);
2466
2467 counter_u64_add(d->bd_fcount, 1);
2468 if (gottime < bpf_ts_quality(d->bd_tstamp))
2469 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2470 #ifdef MAC
2471 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2472 #endif
2473 catchpacket(d, (u_char *)&mb, pktlen, slen,
2474 bpf_append_mbuf, &bt);
2475 BPFD_UNLOCK(d);
2476 }
2477 }
2478 NET_EPOCH_EXIT(et);
2479 }
2480
2481 void
bpf_mtap2_if(if_t ifp,void * data,u_int dlen,struct mbuf * m)2482 bpf_mtap2_if(if_t ifp, void *data, u_int dlen, struct mbuf *m)
2483 {
2484 if (bpf_peers_present(ifp->if_bpf)) {
2485 M_ASSERTVALID(m);
2486 bpf_mtap2(ifp->if_bpf, data, dlen, m);
2487 }
2488 }
2489
2490 #undef BPF_CHECK_DIRECTION
2491 #undef BPF_TSTAMP_NONE
2492 #undef BPF_TSTAMP_FAST
2493 #undef BPF_TSTAMP_NORMAL
2494 #undef BPF_TSTAMP_EXTERN
2495
2496 static int
bpf_hdrlen(struct bpf_d * d)2497 bpf_hdrlen(struct bpf_d *d)
2498 {
2499 int hdrlen;
2500
2501 hdrlen = d->bd_bif->bif_hdrlen;
2502 #ifndef BURN_BRIDGES
2503 if (d->bd_tstamp == BPF_T_NONE ||
2504 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2505 #ifdef COMPAT_FREEBSD32
2506 if (d->bd_compat32)
2507 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2508 else
2509 #endif
2510 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2511 else
2512 #endif
2513 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2514 #ifdef COMPAT_FREEBSD32
2515 if (d->bd_compat32)
2516 hdrlen = BPF_WORDALIGN32(hdrlen);
2517 else
2518 #endif
2519 hdrlen = BPF_WORDALIGN(hdrlen);
2520
2521 return (hdrlen - d->bd_bif->bif_hdrlen);
2522 }
2523
2524 static void
bpf_bintime2ts(struct bintime * bt,struct bpf_ts * ts,int tstype)2525 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2526 {
2527 struct bintime bt2, boottimebin;
2528 struct timeval tsm;
2529 struct timespec tsn;
2530
2531 if ((tstype & BPF_T_MONOTONIC) == 0) {
2532 bt2 = *bt;
2533 getboottimebin(&boottimebin);
2534 bintime_add(&bt2, &boottimebin);
2535 bt = &bt2;
2536 }
2537 switch (BPF_T_FORMAT(tstype)) {
2538 case BPF_T_MICROTIME:
2539 bintime2timeval(bt, &tsm);
2540 ts->bt_sec = tsm.tv_sec;
2541 ts->bt_frac = tsm.tv_usec;
2542 break;
2543 case BPF_T_NANOTIME:
2544 bintime2timespec(bt, &tsn);
2545 ts->bt_sec = tsn.tv_sec;
2546 ts->bt_frac = tsn.tv_nsec;
2547 break;
2548 case BPF_T_BINTIME:
2549 ts->bt_sec = bt->sec;
2550 ts->bt_frac = bt->frac;
2551 break;
2552 }
2553 }
2554
2555 /*
2556 * Move the packet data from interface memory (pkt) into the
2557 * store buffer. "cpfn" is the routine called to do the actual data
2558 * transfer. bcopy is passed in to copy contiguous chunks, while
2559 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2560 * pkt is really an mbuf.
2561 */
2562 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)2563 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2564 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2565 struct bintime *bt)
2566 {
2567 static char zeroes[BPF_ALIGNMENT];
2568 struct bpf_xhdr hdr;
2569 #ifndef BURN_BRIDGES
2570 struct bpf_hdr hdr_old;
2571 #ifdef COMPAT_FREEBSD32
2572 struct bpf_hdr32 hdr32_old;
2573 #endif
2574 #endif
2575 int caplen, curlen, hdrlen, pad, totlen;
2576 int do_wakeup = 0;
2577 int do_timestamp;
2578 int tstype;
2579
2580 BPFD_LOCK_ASSERT(d);
2581 if (d->bd_bif == NULL) {
2582 /* Descriptor was detached in concurrent thread */
2583 counter_u64_add(d->bd_dcount, 1);
2584 return;
2585 }
2586
2587 /*
2588 * Detect whether user space has released a buffer back to us, and if
2589 * so, move it from being a hold buffer to a free buffer. This may
2590 * not be the best place to do it (for example, we might only want to
2591 * run this check if we need the space), but for now it's a reliable
2592 * spot to do it.
2593 */
2594 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2595 d->bd_fbuf = d->bd_hbuf;
2596 d->bd_hbuf = NULL;
2597 d->bd_hlen = 0;
2598 bpf_buf_reclaimed(d);
2599 }
2600
2601 /*
2602 * Figure out how many bytes to move. If the packet is
2603 * greater or equal to the snapshot length, transfer that
2604 * much. Otherwise, transfer the whole packet (unless
2605 * we hit the buffer size limit).
2606 */
2607 hdrlen = bpf_hdrlen(d);
2608 totlen = hdrlen + min(snaplen, pktlen);
2609 if (totlen > d->bd_bufsize)
2610 totlen = d->bd_bufsize;
2611
2612 /*
2613 * Round up the end of the previous packet to the next longword.
2614 *
2615 * Drop the packet if there's no room and no hope of room
2616 * If the packet would overflow the storage buffer or the storage
2617 * buffer is considered immutable by the buffer model, try to rotate
2618 * the buffer and wakeup pending processes.
2619 */
2620 #ifdef COMPAT_FREEBSD32
2621 if (d->bd_compat32)
2622 curlen = BPF_WORDALIGN32(d->bd_slen);
2623 else
2624 #endif
2625 curlen = BPF_WORDALIGN(d->bd_slen);
2626 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2627 if (d->bd_fbuf == NULL) {
2628 /*
2629 * There's no room in the store buffer, and no
2630 * prospect of room, so drop the packet. Notify the
2631 * buffer model.
2632 */
2633 bpf_buffull(d);
2634 counter_u64_add(d->bd_dcount, 1);
2635 return;
2636 }
2637 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2638 ROTATE_BUFFERS(d);
2639 do_wakeup = 1;
2640 curlen = 0;
2641 } else {
2642 if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
2643 /*
2644 * Immediate mode is set, or the read timeout has
2645 * already expired during a select call. A packet
2646 * arrived, so the reader should be woken up.
2647 */
2648 do_wakeup = 1;
2649 }
2650 pad = curlen - d->bd_slen;
2651 KASSERT(pad >= 0 && pad <= sizeof(zeroes),
2652 ("%s: invalid pad byte count %d", __func__, pad));
2653 if (pad > 0) {
2654 /* Zero pad bytes. */
2655 bpf_append_bytes(d, d->bd_sbuf, d->bd_slen, zeroes,
2656 pad);
2657 }
2658 }
2659
2660 caplen = totlen - hdrlen;
2661 tstype = d->bd_tstamp;
2662 do_timestamp = tstype != BPF_T_NONE;
2663 #ifndef BURN_BRIDGES
2664 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2665 struct bpf_ts ts;
2666 if (do_timestamp)
2667 bpf_bintime2ts(bt, &ts, tstype);
2668 #ifdef COMPAT_FREEBSD32
2669 if (d->bd_compat32) {
2670 bzero(&hdr32_old, sizeof(hdr32_old));
2671 if (do_timestamp) {
2672 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2673 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2674 }
2675 hdr32_old.bh_datalen = pktlen;
2676 hdr32_old.bh_hdrlen = hdrlen;
2677 hdr32_old.bh_caplen = caplen;
2678 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2679 sizeof(hdr32_old));
2680 goto copy;
2681 }
2682 #endif
2683 bzero(&hdr_old, sizeof(hdr_old));
2684 if (do_timestamp) {
2685 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2686 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2687 }
2688 hdr_old.bh_datalen = pktlen;
2689 hdr_old.bh_hdrlen = hdrlen;
2690 hdr_old.bh_caplen = caplen;
2691 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2692 sizeof(hdr_old));
2693 goto copy;
2694 }
2695 #endif
2696
2697 /*
2698 * Append the bpf header. Note we append the actual header size, but
2699 * move forward the length of the header plus padding.
2700 */
2701 bzero(&hdr, sizeof(hdr));
2702 if (do_timestamp)
2703 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2704 hdr.bh_datalen = pktlen;
2705 hdr.bh_hdrlen = hdrlen;
2706 hdr.bh_caplen = caplen;
2707 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2708
2709 /*
2710 * Copy the packet data into the store buffer and update its length.
2711 */
2712 #ifndef BURN_BRIDGES
2713 copy:
2714 #endif
2715 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2716 d->bd_slen = curlen + totlen;
2717
2718 if (do_wakeup)
2719 bpf_wakeup(d);
2720 }
2721
2722 /*
2723 * Free buffers currently in use by a descriptor.
2724 * Called on close.
2725 */
2726 static void
bpfd_free(epoch_context_t ctx)2727 bpfd_free(epoch_context_t ctx)
2728 {
2729 struct bpf_d *d;
2730 struct bpf_program_buffer *p;
2731
2732 /*
2733 * We don't need to lock out interrupts since this descriptor has
2734 * been detached from its interface and it yet hasn't been marked
2735 * free.
2736 */
2737 d = __containerof(ctx, struct bpf_d, epoch_ctx);
2738 bpf_free(d);
2739 if (d->bd_rfilter != NULL) {
2740 p = __containerof((void *)d->bd_rfilter,
2741 struct bpf_program_buffer, buffer);
2742 #ifdef BPF_JITTER
2743 p->func = d->bd_bfilter;
2744 #endif
2745 bpf_program_buffer_free(&p->epoch_ctx);
2746 }
2747 if (d->bd_wfilter != NULL) {
2748 p = __containerof((void *)d->bd_wfilter,
2749 struct bpf_program_buffer, buffer);
2750 #ifdef BPF_JITTER
2751 p->func = NULL;
2752 #endif
2753 bpf_program_buffer_free(&p->epoch_ctx);
2754 }
2755
2756 mtx_destroy(&d->bd_lock);
2757 counter_u64_free(d->bd_rcount);
2758 counter_u64_free(d->bd_dcount);
2759 counter_u64_free(d->bd_fcount);
2760 counter_u64_free(d->bd_wcount);
2761 counter_u64_free(d->bd_wfcount);
2762 counter_u64_free(d->bd_wdcount);
2763 counter_u64_free(d->bd_zcopy);
2764 free(d, M_BPF);
2765 }
2766
2767 /*
2768 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2769 * fixed size of the link header (variable length headers not yet supported).
2770 */
2771 void
bpfattach(struct ifnet * ifp,u_int dlt,u_int hdrlen)2772 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2773 {
2774
2775 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2776 }
2777
2778 /*
2779 * Attach an interface to bpf. ifp is a pointer to the structure
2780 * defining the interface to be attached, dlt is the link layer type,
2781 * and hdrlen is the fixed size of the link header (variable length
2782 * headers are not yet supporrted).
2783 */
2784 void
bpfattach2(struct ifnet * ifp,u_int dlt,u_int hdrlen,struct bpf_if ** driverp)2785 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2786 struct bpf_if **driverp)
2787 {
2788 struct bpf_if *bp;
2789
2790 KASSERT(*driverp == NULL,
2791 ("bpfattach2: driverp already initialized"));
2792
2793 bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2794
2795 CK_LIST_INIT(&bp->bif_dlist);
2796 CK_LIST_INIT(&bp->bif_wlist);
2797 bp->bif_ifp = ifp;
2798 bp->bif_dlt = dlt;
2799 bp->bif_hdrlen = hdrlen;
2800 bp->bif_bpf = driverp;
2801 refcount_init(&bp->bif_refcnt, 1);
2802 *driverp = bp;
2803 /*
2804 * Reference ifnet pointer, so it won't freed until
2805 * we release it.
2806 */
2807 if_ref(ifp);
2808 BPF_LOCK();
2809 CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2810 BPF_UNLOCK();
2811
2812 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2813 if_printf(ifp, "bpf attached\n");
2814 }
2815
2816 #ifdef VIMAGE
2817 /*
2818 * When moving interfaces between vnet instances we need a way to
2819 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2820 * after the vmove. We unfortunately have no device driver infrastructure
2821 * to query the interface for these values after creation/attach, thus
2822 * add this as a workaround.
2823 */
2824 int
bpf_get_bp_params(struct bpf_if * bp,u_int * bif_dlt,u_int * bif_hdrlen)2825 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2826 {
2827
2828 if (bp == NULL)
2829 return (ENXIO);
2830 if (bif_dlt == NULL && bif_hdrlen == NULL)
2831 return (0);
2832
2833 if (bif_dlt != NULL)
2834 *bif_dlt = bp->bif_dlt;
2835 if (bif_hdrlen != NULL)
2836 *bif_hdrlen = bp->bif_hdrlen;
2837
2838 return (0);
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_locked(d, true);
2869 }
2870
2871 /* Detach writer-only descriptors */
2872 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2873 bpf_detachd_locked(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 CK_LIST_INIT(&bpf_iflist);
2973
2974 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2975 /* For compatibility */
2976 make_dev_alias(dev, "bpf0");
2977 }
2978
2979 /*
2980 * Zero out the various packet counters associated with all of the bpf
2981 * descriptors. At some point, we will probably want to get a bit more
2982 * granular and allow the user to specify descriptors to be zeroed.
2983 */
2984 static void
bpf_zero_counters(void)2985 bpf_zero_counters(void)
2986 {
2987 struct bpf_if *bp;
2988 struct bpf_d *bd;
2989
2990 BPF_LOCK();
2991 /*
2992 * We are protected by global lock here, interfaces and
2993 * descriptors can not be deleted while we hold it.
2994 */
2995 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2996 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2997 counter_u64_zero(bd->bd_rcount);
2998 counter_u64_zero(bd->bd_dcount);
2999 counter_u64_zero(bd->bd_fcount);
3000 counter_u64_zero(bd->bd_wcount);
3001 counter_u64_zero(bd->bd_wfcount);
3002 counter_u64_zero(bd->bd_zcopy);
3003 }
3004 }
3005 BPF_UNLOCK();
3006 }
3007
3008 /*
3009 * Fill filter statistics
3010 */
3011 static void
bpfstats_fill_xbpf(struct xbpf_d * d,struct bpf_d * bd)3012 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
3013 {
3014
3015 BPF_LOCK_ASSERT();
3016 bzero(d, sizeof(*d));
3017 d->bd_structsize = sizeof(*d);
3018 d->bd_immediate = bd->bd_immediate;
3019 d->bd_promisc = bd->bd_promisc;
3020 d->bd_hdrcmplt = bd->bd_hdrcmplt;
3021 d->bd_direction = bd->bd_direction;
3022 d->bd_feedback = bd->bd_feedback;
3023 d->bd_async = bd->bd_async;
3024 d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
3025 d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
3026 d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
3027 d->bd_sig = bd->bd_sig;
3028 d->bd_slen = bd->bd_slen;
3029 d->bd_hlen = bd->bd_hlen;
3030 d->bd_bufsize = bd->bd_bufsize;
3031 d->bd_pid = bd->bd_pid;
3032 strlcpy(d->bd_ifname,
3033 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
3034 d->bd_locked = bd->bd_locked;
3035 d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
3036 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
3037 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
3038 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
3039 d->bd_bufmode = bd->bd_bufmode;
3040 }
3041
3042 /*
3043 * Handle `netstat -B' stats request
3044 */
3045 static int
bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)3046 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
3047 {
3048 static const struct xbpf_d zerostats;
3049 struct xbpf_d *xbdbuf, *xbd, tempstats;
3050 int index, error;
3051 struct bpf_if *bp;
3052 struct bpf_d *bd;
3053
3054 /*
3055 * XXX This is not technically correct. It is possible for non
3056 * privileged users to open bpf devices. It would make sense
3057 * if the users who opened the devices were able to retrieve
3058 * the statistics for them, too.
3059 */
3060 error = priv_check(req->td, PRIV_NET_BPF);
3061 if (error)
3062 return (error);
3063 /*
3064 * Check to see if the user is requesting that the counters be
3065 * zeroed out. Explicitly check that the supplied data is zeroed,
3066 * as we aren't allowing the user to set the counters currently.
3067 */
3068 if (req->newptr != NULL) {
3069 if (req->newlen != sizeof(tempstats))
3070 return (EINVAL);
3071 memset(&tempstats, 0, sizeof(tempstats));
3072 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
3073 if (error)
3074 return (error);
3075 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
3076 return (EINVAL);
3077 bpf_zero_counters();
3078 return (0);
3079 }
3080 if (req->oldptr == NULL)
3081 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
3082 if (bpf_bpfd_cnt == 0)
3083 return (SYSCTL_OUT(req, 0, 0));
3084 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
3085 BPF_LOCK();
3086 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
3087 BPF_UNLOCK();
3088 free(xbdbuf, M_BPF);
3089 return (ENOMEM);
3090 }
3091 index = 0;
3092 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
3093 /* Send writers-only first */
3094 CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
3095 xbd = &xbdbuf[index++];
3096 bpfstats_fill_xbpf(xbd, bd);
3097 }
3098 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
3099 xbd = &xbdbuf[index++];
3100 bpfstats_fill_xbpf(xbd, bd);
3101 }
3102 }
3103 BPF_UNLOCK();
3104 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
3105 free(xbdbuf, M_BPF);
3106 return (error);
3107 }
3108
3109 SYSINIT(bpfdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, bpf_drvinit, NULL);
3110
3111 #else /* !DEV_BPF && !NETGRAPH_BPF */
3112
3113 /*
3114 * NOP stubs to allow bpf-using drivers to load and function.
3115 *
3116 * A 'better' implementation would allow the core bpf functionality
3117 * to be loaded at runtime.
3118 */
3119
3120 void
bpf_tap(struct bpf_if * bp,u_char * pkt,u_int pktlen)3121 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3122 {
3123 }
3124
3125 void
bpf_tap_if(if_t ifp,u_char * pkt,u_int pktlen)3126 bpf_tap_if(if_t ifp, u_char *pkt, u_int pktlen)
3127 {
3128 }
3129
3130 void
bpf_mtap(struct bpf_if * bp,struct mbuf * m)3131 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3132 {
3133 }
3134
3135 void
bpf_mtap_if(if_t ifp,struct mbuf * m)3136 bpf_mtap_if(if_t ifp, struct mbuf *m)
3137 {
3138 }
3139
3140 void
bpf_mtap2(struct bpf_if * bp,void * d,u_int l,struct mbuf * m)3141 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3142 {
3143 }
3144
3145 void
bpf_mtap2_if(if_t ifp,void * data,u_int dlen,struct mbuf * m)3146 bpf_mtap2_if(if_t ifp, void *data, u_int dlen, struct mbuf *m)
3147 {
3148 }
3149
3150 void
bpfattach(struct ifnet * ifp,u_int dlt,u_int hdrlen)3151 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3152 {
3153
3154 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3155 }
3156
3157 void
bpfattach2(struct ifnet * ifp,u_int dlt,u_int hdrlen,struct bpf_if ** driverp)3158 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3159 {
3160
3161 *driverp = __DECONST(struct bpf_if *, &dead_bpf_if);
3162 }
3163
3164 void
bpfdetach(struct ifnet * ifp)3165 bpfdetach(struct ifnet *ifp)
3166 {
3167 }
3168
3169 bool
bpf_peers_present_if(struct ifnet * ifp)3170 bpf_peers_present_if(struct ifnet *ifp)
3171 {
3172 return (false);
3173 }
3174
3175 u_int
bpf_filter(const struct bpf_insn * pc,u_char * p,u_int wirelen,u_int buflen)3176 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3177 {
3178 return (-1); /* "no filter" behaviour */
3179 }
3180
3181 int
bpf_validate(const struct bpf_insn * f,int len)3182 bpf_validate(const struct bpf_insn *f, int len)
3183 {
3184 return (0); /* false */
3185 }
3186
3187 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3188
3189 #ifdef DDB
3190 static void
bpf_show_bpf_if(struct bpf_if * bpf_if)3191 bpf_show_bpf_if(struct bpf_if *bpf_if)
3192 {
3193
3194 if (bpf_if == NULL)
3195 return;
3196 db_printf("%p:\n", bpf_if);
3197 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3198 #define BPF_DB_PRINTF_RAW(f, e) db_printf(" %s = " f "\n", #e, e);
3199 /* bif_ext.bif_next */
3200 /* bif_ext.bif_dlist */
3201 BPF_DB_PRINTF("%#x", bif_dlt);
3202 BPF_DB_PRINTF("%u", bif_hdrlen);
3203 /* bif_wlist */
3204 BPF_DB_PRINTF("%p", bif_ifp);
3205 BPF_DB_PRINTF("%p", bif_bpf);
3206 BPF_DB_PRINTF_RAW("%u", refcount_load(&bpf_if->bif_refcnt));
3207 }
3208
DB_SHOW_COMMAND(bpf_if,db_show_bpf_if)3209 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3210 {
3211
3212 if (!have_addr) {
3213 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3214 return;
3215 }
3216
3217 bpf_show_bpf_if((struct bpf_if *)addr);
3218 }
3219 #endif
3220