1 /* $NetBSD: pcap-pf.c,v 1.3 2015/03/31 21:39:42 christos Exp $ */
2
3 /*
4 * Copyright (c) 1990, 1991, 1992, 1993, 1994, 1995, 1996
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that: (1) source code distributions
9 * retain the above copyright notice and this paragraph in its entirety, (2)
10 * distributions including binary code include the above copyright notice and
11 * this paragraph in its entirety in the documentation or other materials
12 * provided with the distribution, and (3) all advertising materials mentioning
13 * features or use of this software display the following acknowledgement:
14 * ``This product includes software developed by the University of California,
15 * Lawrence Berkeley Laboratory and its contributors.'' Neither the name of
16 * the University nor the names of its contributors may be used to endorse
17 * or promote products derived from this software without specific prior
18 * written permission.
19 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED
20 * WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF
21 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
22 *
23 * packet filter subroutines for tcpdump
24 * Extraction/creation by Jeffrey Mogul, DECWRL
25 */
26
27 #include <sys/cdefs.h>
28 __RCSID("$NetBSD: pcap-pf.c,v 1.3 2015/03/31 21:39:42 christos Exp $");
29
30 #ifdef HAVE_CONFIG_H
31 #include "config.h"
32 #endif
33
34 #include <sys/types.h>
35 #include <sys/time.h>
36 #include <sys/timeb.h>
37 #include <sys/socket.h>
38 #include <sys/file.h>
39 #include <sys/ioctl.h>
40 #include <net/pfilt.h>
41
42 struct mbuf;
43 struct rtentry;
44 #include <net/if.h>
45
46 #include <netinet/in.h>
47 #include <netinet/in_systm.h>
48 #include <netinet/ip.h>
49 #include <netinet/if_ether.h>
50 #include <netinet/ip_var.h>
51 #include <netinet/udp.h>
52 #include <netinet/udp_var.h>
53 #include <netinet/tcp.h>
54 #include <netinet/tcpip.h>
55
56 #include <ctype.h>
57 #include <errno.h>
58 #include <netdb.h>
59 #include <stdio.h>
60 #include <stdlib.h>
61 #include <string.h>
62 #include <unistd.h>
63
64 /*
65 * Make "pcap.h" not include "pcap/bpf.h"; we are going to include the
66 * native OS version, as we need various BPF ioctls from it.
67 */
68 #define PCAP_DONT_INCLUDE_PCAP_BPF_H
69 #include <net/bpf.h>
70
71 #include "pcap-int.h"
72
73 #ifdef HAVE_OS_PROTO_H
74 #include "os-proto.h"
75 #endif
76
77 /*
78 * FDDI packets are padded to make everything line up on a nice boundary.
79 */
80 #define PCAP_FDDIPAD 3
81
82 /*
83 * Private data for capturing on Ultrix and DEC OSF/1^WDigital UNIX^W^W
84 * Tru64 UNIX packetfilter devices.
85 */
86 struct pcap_pf {
87 int filtering_in_kernel; /* using kernel filter */
88 u_long TotPkts; /* can't oflow for 79 hrs on ether */
89 u_long TotAccepted; /* count accepted by filter */
90 u_long TotDrops; /* count of dropped packets */
91 long TotMissed; /* missed by i/f during this run */
92 long OrigMissed; /* missed by i/f before this run */
93 };
94
95 static int pcap_setfilter_pf(pcap_t *, struct bpf_program *);
96
97 /*
98 * BUFSPACE is the size in bytes of the packet read buffer. Most tcpdump
99 * applications aren't going to need more than 200 bytes of packet header
100 * and the read shouldn't return more packets than packetfilter's internal
101 * queue limit (bounded at 256).
102 */
103 #define BUFSPACE (200 * 256)
104
105 static int
pcap_read_pf(pcap_t * pc,int cnt,pcap_handler callback,u_char * user)106 pcap_read_pf(pcap_t *pc, int cnt, pcap_handler callback, u_char *user)
107 {
108 struct pcap_pf *pf = pc->priv;
109 register u_char *p, *bp;
110 register int cc, n, buflen, inc;
111 register struct enstamp *sp;
112 #ifdef LBL_ALIGN
113 struct enstamp stamp;
114 #endif
115 register int pad;
116
117 again:
118 cc = pc->cc;
119 if (cc == 0) {
120 cc = read(pc->fd, (char *)pc->buffer + pc->offset, pc->bufsize);
121 if (cc < 0) {
122 if (errno == EWOULDBLOCK)
123 return (0);
124 if (errno == EINVAL &&
125 lseek(pc->fd, 0L, SEEK_CUR) + pc->bufsize < 0) {
126 /*
127 * Due to a kernel bug, after 2^31 bytes,
128 * the kernel file offset overflows and
129 * read fails with EINVAL. The lseek()
130 * to 0 will fix things.
131 */
132 (void)lseek(pc->fd, 0L, SEEK_SET);
133 goto again;
134 }
135 snprintf(pc->errbuf, sizeof(pc->errbuf), "pf read: %s",
136 pcap_strerror(errno));
137 return (-1);
138 }
139 bp = pc->buffer + pc->offset;
140 } else
141 bp = pc->bp;
142 /*
143 * Loop through each packet.
144 */
145 n = 0;
146 pad = pc->fddipad;
147 while (cc > 0) {
148 /*
149 * Has "pcap_breakloop()" been called?
150 * If so, return immediately - if we haven't read any
151 * packets, clear the flag and return -2 to indicate
152 * that we were told to break out of the loop, otherwise
153 * leave the flag set, so that the *next* call will break
154 * out of the loop without having read any packets, and
155 * return the number of packets we've processed so far.
156 */
157 if (pc->break_loop) {
158 if (n == 0) {
159 pc->break_loop = 0;
160 return (-2);
161 } else {
162 pc->cc = cc;
163 pc->bp = bp;
164 return (n);
165 }
166 }
167 if (cc < sizeof(*sp)) {
168 snprintf(pc->errbuf, sizeof(pc->errbuf),
169 "pf short read (%d)", cc);
170 return (-1);
171 }
172 #ifdef LBL_ALIGN
173 if ((long)bp & 3) {
174 sp = &stamp;
175 memcpy((char *)sp, (char *)bp, sizeof(*sp));
176 } else
177 #endif
178 sp = (struct enstamp *)bp;
179 if (sp->ens_stamplen != sizeof(*sp)) {
180 snprintf(pc->errbuf, sizeof(pc->errbuf),
181 "pf short stamplen (%d)",
182 sp->ens_stamplen);
183 return (-1);
184 }
185
186 p = bp + sp->ens_stamplen;
187 buflen = sp->ens_count;
188 if (buflen > pc->snapshot)
189 buflen = pc->snapshot;
190
191 /* Calculate inc before possible pad update */
192 inc = ENALIGN(buflen + sp->ens_stamplen);
193 cc -= inc;
194 bp += inc;
195 pf->TotPkts++;
196 pf->TotDrops += sp->ens_dropped;
197 pf->TotMissed = sp->ens_ifoverflows;
198 if (pf->OrigMissed < 0)
199 pf->OrigMissed = pf->TotMissed;
200
201 /*
202 * Short-circuit evaluation: if using BPF filter
203 * in kernel, no need to do it now - we already know
204 * the packet passed the filter.
205 *
206 * Note: the filter code was generated assuming
207 * that pc->fddipad was the amount of padding
208 * before the header, as that's what's required
209 * in the kernel, so we run the filter before
210 * skipping that padding.
211 */
212 if (pf->filtering_in_kernel ||
213 bpf_filter(pc->fcode.bf_insns, p, sp->ens_count, buflen)) {
214 struct pcap_pkthdr h;
215 pf->TotAccepted++;
216 h.ts = sp->ens_tstamp;
217 h.len = sp->ens_count - pad;
218 p += pad;
219 buflen -= pad;
220 h.caplen = buflen;
221 (*callback)(user, &h, p);
222 if (++n >= cnt && !PACKET_COUNT_IS_UNLIMITED(cnt)) {
223 pc->cc = cc;
224 pc->bp = bp;
225 return (n);
226 }
227 }
228 }
229 pc->cc = 0;
230 return (n);
231 }
232
233 static int
pcap_inject_pf(pcap_t * p,const void * buf,size_t size)234 pcap_inject_pf(pcap_t *p, const void *buf, size_t size)
235 {
236 int ret;
237
238 ret = write(p->fd, buf, size);
239 if (ret == -1) {
240 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
241 pcap_strerror(errno));
242 return (-1);
243 }
244 return (ret);
245 }
246
247 static int
pcap_stats_pf(pcap_t * p,struct pcap_stat * ps)248 pcap_stats_pf(pcap_t *p, struct pcap_stat *ps)
249 {
250 struct pcap_pf *pf = p->priv;
251
252 /*
253 * If packet filtering is being done in the kernel:
254 *
255 * "ps_recv" counts only packets that passed the filter.
256 * This does not include packets dropped because we
257 * ran out of buffer space. (XXX - perhaps it should,
258 * by adding "ps_drop" to "ps_recv", for compatibility
259 * with some other platforms. On the other hand, on
260 * some platforms "ps_recv" counts only packets that
261 * passed the filter, and on others it counts packets
262 * that didn't pass the filter....)
263 *
264 * "ps_drop" counts packets that passed the kernel filter
265 * (if any) but were dropped because the input queue was
266 * full.
267 *
268 * "ps_ifdrop" counts packets dropped by the network
269 * inteface (regardless of whether they would have passed
270 * the input filter, of course).
271 *
272 * If packet filtering is not being done in the kernel:
273 *
274 * "ps_recv" counts only packets that passed the filter.
275 *
276 * "ps_drop" counts packets that were dropped because the
277 * input queue was full, regardless of whether they passed
278 * the userland filter.
279 *
280 * "ps_ifdrop" counts packets dropped by the network
281 * inteface (regardless of whether they would have passed
282 * the input filter, of course).
283 *
284 * These statistics don't include packets not yet read from
285 * the kernel by libpcap, but they may include packets not
286 * yet read from libpcap by the application.
287 */
288 ps->ps_recv = pf->TotAccepted;
289 ps->ps_drop = pf->TotDrops;
290 ps->ps_ifdrop = pf->TotMissed - pf->OrigMissed;
291 return (0);
292 }
293
294 /*
295 * We include the OS's <net/bpf.h>, not our "pcap/bpf.h", so we probably
296 * don't get DLT_DOCSIS defined.
297 */
298 #ifndef DLT_DOCSIS
299 #define DLT_DOCSIS 143
300 #endif
301
302 static int
pcap_activate_pf(pcap_t * p)303 pcap_activate_pf(pcap_t *p)
304 {
305 struct pcap_pf *pf = p->priv;
306 short enmode;
307 int backlog = -1; /* request the most */
308 struct enfilter Filter;
309 struct endevp devparams;
310
311 /*
312 * Initially try a read/write open (to allow the inject
313 * method to work). If that fails due to permission
314 * issues, fall back to read-only. This allows a
315 * non-root user to be granted specific access to pcap
316 * capabilities via file permissions.
317 *
318 * XXX - we should have an API that has a flag that
319 * controls whether to open read-only or read-write,
320 * so that denial of permission to send (or inability
321 * to send, if sending packets isn't supported on
322 * the device in question) can be indicated at open
323 * time.
324 *
325 * XXX - we assume here that "pfopen()" does not, in fact, modify
326 * its argument, even though it takes a "char *" rather than a
327 * "const char *" as its first argument. That appears to be
328 * the case, at least on Digital UNIX 4.0.
329 */
330 p->fd = pfopen(p->opt.source, O_RDWR);
331 if (p->fd == -1 && errno == EACCES)
332 p->fd = pfopen(p->opt.source, O_RDONLY);
333 if (p->fd < 0) {
334 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "pf open: %s: %s\n\
335 your system may not be properly configured; see the packetfilter(4) man page\n",
336 p->opt.source, pcap_strerror(errno));
337 goto bad;
338 }
339 pf->OrigMissed = -1;
340 enmode = ENTSTAMP|ENNONEXCL;
341 if (!p->opt.immediate)
342 enmode |= ENBATCH;
343 if (p->opt.promisc)
344 enmode |= ENPROMISC;
345 if (ioctl(p->fd, EIOCMBIS, (caddr_t)&enmode) < 0) {
346 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCMBIS: %s",
347 pcap_strerror(errno));
348 goto bad;
349 }
350 #ifdef ENCOPYALL
351 /* Try to set COPYALL mode so that we see packets to ourself */
352 enmode = ENCOPYALL;
353 (void)ioctl(p->fd, EIOCMBIS, (caddr_t)&enmode);/* OK if this fails */
354 #endif
355 /* set the backlog */
356 if (ioctl(p->fd, EIOCSETW, (caddr_t)&backlog) < 0) {
357 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSETW: %s",
358 pcap_strerror(errno));
359 goto bad;
360 }
361 /* discover interface type */
362 if (ioctl(p->fd, EIOCDEVP, (caddr_t)&devparams) < 0) {
363 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCDEVP: %s",
364 pcap_strerror(errno));
365 goto bad;
366 }
367 /* HACK: to compile prior to Ultrix 4.2 */
368 #ifndef ENDT_FDDI
369 #define ENDT_FDDI 4
370 #endif
371 switch (devparams.end_dev_type) {
372
373 case ENDT_10MB:
374 p->linktype = DLT_EN10MB;
375 p->offset = 2;
376 /*
377 * This is (presumably) a real Ethernet capture; give it a
378 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
379 * that an application can let you choose it, in case you're
380 * capturing DOCSIS traffic that a Cisco Cable Modem
381 * Termination System is putting out onto an Ethernet (it
382 * doesn't put an Ethernet header onto the wire, it puts raw
383 * DOCSIS frames out on the wire inside the low-level
384 * Ethernet framing).
385 */
386 p->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
387 /*
388 * If that fails, just leave the list empty.
389 */
390 if (p->dlt_list != NULL) {
391 p->dlt_list[0] = DLT_EN10MB;
392 p->dlt_list[1] = DLT_DOCSIS;
393 p->dlt_count = 2;
394 }
395 break;
396
397 case ENDT_FDDI:
398 p->linktype = DLT_FDDI;
399 break;
400
401 #ifdef ENDT_SLIP
402 case ENDT_SLIP:
403 p->linktype = DLT_SLIP;
404 break;
405 #endif
406
407 #ifdef ENDT_PPP
408 case ENDT_PPP:
409 p->linktype = DLT_PPP;
410 break;
411 #endif
412
413 #ifdef ENDT_LOOPBACK
414 case ENDT_LOOPBACK:
415 /*
416 * It appears to use Ethernet framing, at least on
417 * Digital UNIX 4.0.
418 */
419 p->linktype = DLT_EN10MB;
420 p->offset = 2;
421 break;
422 #endif
423
424 #ifdef ENDT_TRN
425 case ENDT_TRN:
426 p->linktype = DLT_IEEE802;
427 break;
428 #endif
429
430 default:
431 /*
432 * XXX - what about ENDT_IEEE802? The pfilt.h header
433 * file calls this "IEEE 802 networks (non-Ethernet)",
434 * but that doesn't specify a specific link layer type;
435 * it could be 802.4, or 802.5 (except that 802.5 is
436 * ENDT_TRN), or 802.6, or 802.11, or.... That's why
437 * DLT_IEEE802 was hijacked to mean Token Ring in various
438 * BSDs, and why we went along with that hijacking.
439 *
440 * XXX - what about ENDT_HDLC and ENDT_NULL?
441 * Presumably, as ENDT_OTHER is just "Miscellaneous
442 * framing", there's not much we can do, as that
443 * doesn't specify a particular type of header.
444 */
445 snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
446 "unknown data-link type %u", devparams.end_dev_type);
447 goto bad;
448 }
449 /* set truncation */
450 if (p->linktype == DLT_FDDI) {
451 p->fddipad = PCAP_FDDIPAD;
452
453 /* packetfilter includes the padding in the snapshot */
454 p->snapshot += PCAP_FDDIPAD;
455 } else
456 p->fddipad = 0;
457 if (ioctl(p->fd, EIOCTRUNCATE, (caddr_t)&p->snapshot) < 0) {
458 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCTRUNCATE: %s",
459 pcap_strerror(errno));
460 goto bad;
461 }
462 /* accept all packets */
463 memset(&Filter, 0, sizeof(Filter));
464 Filter.enf_Priority = 37; /* anything > 2 */
465 Filter.enf_FilterLen = 0; /* means "always true" */
466 if (ioctl(p->fd, EIOCSETF, (caddr_t)&Filter) < 0) {
467 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSETF: %s",
468 pcap_strerror(errno));
469 goto bad;
470 }
471
472 if (p->opt.timeout != 0) {
473 struct timeval timeout;
474 timeout.tv_sec = p->opt.timeout / 1000;
475 timeout.tv_usec = (p->opt.timeout * 1000) % 1000000;
476 if (ioctl(p->fd, EIOCSRTIMEOUT, (caddr_t)&timeout) < 0) {
477 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSRTIMEOUT: %s",
478 pcap_strerror(errno));
479 goto bad;
480 }
481 }
482
483 p->bufsize = BUFSPACE;
484 p->buffer = (u_char*)malloc(p->bufsize + p->offset);
485 if (p->buffer == NULL) {
486 strlcpy(p->errbuf, pcap_strerror(errno), PCAP_ERRBUF_SIZE);
487 goto bad;
488 }
489
490 /*
491 * "select()" and "poll()" work on packetfilter devices.
492 */
493 p->selectable_fd = p->fd;
494
495 p->read_op = pcap_read_pf;
496 p->inject_op = pcap_inject_pf;
497 p->setfilter_op = pcap_setfilter_pf;
498 p->setdirection_op = NULL; /* Not implemented. */
499 p->set_datalink_op = NULL; /* can't change data link type */
500 p->getnonblock_op = pcap_getnonblock_fd;
501 p->setnonblock_op = pcap_setnonblock_fd;
502 p->stats_op = pcap_stats_pf;
503
504 return (0);
505 bad:
506 pcap_cleanup_live_common(p);
507 return (PCAP_ERROR);
508 }
509
510 pcap_t *
pcap_create_interface(const char * device,char * ebuf)511 pcap_create_interface(const char *device, char *ebuf)
512 {
513 pcap_t *p;
514
515 p = pcap_create_common(device, ebuf, sizeof (struct pcap_pf));
516 if (p == NULL)
517 return (NULL);
518
519 p->activate_op = pcap_activate_pf;
520 return (p);
521 }
522
523 int
pcap_platform_finddevs(pcap_if_t ** alldevsp,char * errbuf)524 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
525 {
526 return (0);
527 }
528
529 static int
pcap_setfilter_pf(pcap_t * p,struct bpf_program * fp)530 pcap_setfilter_pf(pcap_t *p, struct bpf_program *fp)
531 {
532 struct pcap_pf *pf = p->priv;
533 struct bpf_version bv;
534
535 /*
536 * See if BIOCVERSION works. If not, we assume the kernel doesn't
537 * support BPF-style filters (it's not documented in the bpf(7)
538 * or packetfiler(7) man pages, but the code used to fail if
539 * BIOCSETF worked but BIOCVERSION didn't, and I've seen it do
540 * kernel filtering in DU 4.0, so presumably BIOCVERSION works
541 * there, at least).
542 */
543 if (ioctl(p->fd, BIOCVERSION, (caddr_t)&bv) >= 0) {
544 /*
545 * OK, we have the version of the BPF interpreter;
546 * is it the same major version as us, and the same
547 * or better minor version?
548 */
549 if (bv.bv_major == BPF_MAJOR_VERSION &&
550 bv.bv_minor >= BPF_MINOR_VERSION) {
551 /*
552 * Yes. Try to install the filter.
553 */
554 if (ioctl(p->fd, BIOCSETF, (caddr_t)fp) < 0) {
555 snprintf(p->errbuf, sizeof(p->errbuf),
556 "BIOCSETF: %s", pcap_strerror(errno));
557 return (-1);
558 }
559
560 /*
561 * OK, that succeeded. We're doing filtering in
562 * the kernel. (We assume we don't have a
563 * userland filter installed - that'd require
564 * a previous version check to have failed but
565 * this one to succeed.)
566 *
567 * XXX - this message should be supplied to the
568 * application as a warning of some sort,
569 * except that if it's a GUI application, it's
570 * not clear that it should be displayed in
571 * a window to annoy the user.
572 */
573 fprintf(stderr, "tcpdump: Using kernel BPF filter\n");
574 pf->filtering_in_kernel = 1;
575
576 /*
577 * Discard any previously-received packets,
578 * as they might have passed whatever filter
579 * was formerly in effect, but might not pass
580 * this filter (BIOCSETF discards packets buffered
581 * in the kernel, so you can lose packets in any
582 * case).
583 */
584 p->cc = 0;
585 return (0);
586 }
587
588 /*
589 * We can't use the kernel's BPF interpreter; don't give
590 * up, just log a message and be inefficient.
591 *
592 * XXX - this should really be supplied to the application
593 * as a warning of some sort.
594 */
595 fprintf(stderr,
596 "tcpdump: Requires BPF language %d.%d or higher; kernel is %d.%d\n",
597 BPF_MAJOR_VERSION, BPF_MINOR_VERSION,
598 bv.bv_major, bv.bv_minor);
599 }
600
601 /*
602 * We couldn't do filtering in the kernel; do it in userland.
603 */
604 if (install_bpf_program(p, fp) < 0)
605 return (-1);
606
607 /*
608 * XXX - this message should be supplied by the application as
609 * a warning of some sort.
610 */
611 fprintf(stderr, "tcpdump: Filtering in user process\n");
612 pf->filtering_in_kernel = 0;
613 return (0);
614 }
615