pcap-pf.c revision 127664
1/*
2 * Copyright (c) 1990, 1991, 1992, 1993, 1994, 1995, 1996
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that: (1) source code distributions
7 * retain the above copyright notice and this paragraph in its entirety, (2)
8 * distributions including binary code include the above copyright notice and
9 * this paragraph in its entirety in the documentation or other materials
10 * provided with the distribution, and (3) all advertising materials mentioning
11 * features or use of this software display the following acknowledgement:
12 * ``This product includes software developed by the University of California,
13 * Lawrence Berkeley Laboratory and its contributors.'' Neither the name of
14 * the University nor the names of its contributors may be used to endorse
15 * or promote products derived from this software without specific prior
16 * written permission.
17 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF
19 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
20 *
21 * packet filter subroutines for tcpdump
22 *	Extraction/creation by Jeffrey Mogul, DECWRL
23 */
24
25#ifndef lint
26static const char rcsid[] _U_ =
27    "@(#) $Header: /tcpdump/master/libpcap/pcap-pf.c,v 1.79.2.5 2003/11/22 00:32:55 guy Exp $ (LBL)";
28#endif
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
42struct mbuf;
43struct 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
77static int pcap_setfilter_pf(pcap_t *, struct bpf_program *);
78
79/*
80 * BUFSPACE is the size in bytes of the packet read buffer.  Most tcpdump
81 * applications aren't going to need more than 200 bytes of packet header
82 * and the read shouldn't return more packets than packetfilter's internal
83 * queue limit (bounded at 256).
84 */
85#define BUFSPACE (200 * 256)
86
87static int
88pcap_read_pf(pcap_t *pc, int cnt, pcap_handler callback, u_char *user)
89{
90	register u_char *p, *bp;
91	struct bpf_insn *fcode;
92	register int cc, n, buflen, inc;
93	register struct enstamp *sp;
94#ifdef LBL_ALIGN
95	struct enstamp stamp;
96#endif
97#ifdef PCAP_FDDIPAD
98	register int pad;
99#endif
100
101	fcode = pc->md.use_bpf ? NULL : pc->fcode.bf_insns;
102 again:
103	cc = pc->cc;
104	if (cc == 0) {
105		cc = read(pc->fd, (char *)pc->buffer + pc->offset, pc->bufsize);
106		if (cc < 0) {
107			if (errno == EWOULDBLOCK)
108				return (0);
109			if (errno == EINVAL &&
110			    lseek(pc->fd, 0L, SEEK_CUR) + pc->bufsize < 0) {
111				/*
112				 * Due to a kernel bug, after 2^31 bytes,
113				 * the kernel file offset overflows and
114				 * read fails with EINVAL. The lseek()
115				 * to 0 will fix things.
116				 */
117				(void)lseek(pc->fd, 0L, SEEK_SET);
118				goto again;
119			}
120			snprintf(pc->errbuf, sizeof(pc->errbuf), "pf read: %s",
121				pcap_strerror(errno));
122			return (-1);
123		}
124		bp = pc->buffer + pc->offset;
125	} else
126		bp = pc->bp;
127	/*
128	 * Loop through each packet.
129	 */
130	n = 0;
131#ifdef PCAP_FDDIPAD
132	if (pc->linktype == DLT_FDDI)
133		pad = pcap_fddipad;
134	else
135		pad = 0;
136#endif
137	while (cc > 0) {
138		/*
139		 * Has "pcap_breakloop()" been called?
140		 * If so, return immediately - if we haven't read any
141		 * packets, clear the flag and return -2 to indicate
142		 * that we were told to break out of the loop, otherwise
143		 * leave the flag set, so that the *next* call will break
144		 * out of the loop without having read any packets, and
145		 * return the number of packets we've processed so far.
146		 */
147		if (pc->break_loop) {
148			if (n == 0) {
149				pc->break_loop = 0;
150				return (-2);
151			} else {
152				pc->cc = cc;
153				pc->bp = bp;
154				return (n);
155			}
156		}
157		if (cc < sizeof(*sp)) {
158			snprintf(pc->errbuf, sizeof(pc->errbuf),
159			    "pf short read (%d)", cc);
160			return (-1);
161		}
162#ifdef LBL_ALIGN
163		if ((long)bp & 3) {
164			sp = &stamp;
165			memcpy((char *)sp, (char *)bp, sizeof(*sp));
166		} else
167#endif
168			sp = (struct enstamp *)bp;
169		if (sp->ens_stamplen != sizeof(*sp)) {
170			snprintf(pc->errbuf, sizeof(pc->errbuf),
171			    "pf short stamplen (%d)",
172			    sp->ens_stamplen);
173			return (-1);
174		}
175
176		p = bp + sp->ens_stamplen;
177		buflen = sp->ens_count;
178		if (buflen > pc->snapshot)
179			buflen = pc->snapshot;
180
181		/* Calculate inc before possible pad update */
182		inc = ENALIGN(buflen + sp->ens_stamplen);
183		cc -= inc;
184		bp += inc;
185#ifdef PCAP_FDDIPAD
186		p += pad;
187		buflen -= pad;
188#endif
189		pc->md.TotPkts++;
190		pc->md.TotDrops += sp->ens_dropped;
191		pc->md.TotMissed = sp->ens_ifoverflows;
192		if (pc->md.OrigMissed < 0)
193			pc->md.OrigMissed = pc->md.TotMissed;
194
195		/*
196		 * Short-circuit evaluation: if using BPF filter
197		 * in kernel, no need to do it now.
198		 */
199		if (fcode == NULL ||
200		    bpf_filter(fcode, p, sp->ens_count, buflen)) {
201			struct pcap_pkthdr h;
202			pc->md.TotAccepted++;
203			h.ts = sp->ens_tstamp;
204#ifdef PCAP_FDDIPAD
205			h.len = sp->ens_count - pad;
206#else
207			h.len = sp->ens_count;
208#endif
209			h.caplen = buflen;
210			(*callback)(user, &h, p);
211			if (++n >= cnt && cnt > 0) {
212				pc->cc = cc;
213				pc->bp = bp;
214				return (n);
215			}
216		}
217	}
218	pc->cc = 0;
219	return (n);
220}
221
222static int
223pcap_stats_pf(pcap_t *p, struct pcap_stat *ps)
224{
225
226	/*
227	 * If packet filtering is being done in the kernel:
228	 *
229	 *	"ps_recv" counts only packets that passed the filter.
230	 *	This does not include packets dropped because we
231	 *	ran out of buffer space.  (XXX - perhaps it should,
232	 *	by adding "ps_drop" to "ps_recv", for compatibility
233	 *	with some other platforms.  On the other hand, on
234	 *	some platforms "ps_recv" counts only packets that
235	 *	passed the filter, and on others it counts packets
236	 *	that didn't pass the filter....)
237	 *
238	 *	"ps_drop" counts packets that passed the kernel filter
239	 *	(if any) but were dropped because the input queue was
240	 *	full.
241	 *
242	 *	"ps_ifdrop" counts packets dropped by the network
243	 *	inteface (regardless of whether they would have passed
244	 *	the input filter, of course).
245	 *
246	 * If packet filtering is not being done in the kernel:
247	 *
248	 *	"ps_recv" counts only packets that passed the filter.
249	 *
250	 *	"ps_drop" counts packets that were dropped because the
251	 *	input queue was full, regardless of whether they passed
252	 *	the userland filter.
253	 *
254	 *	"ps_ifdrop" counts packets dropped by the network
255	 *	inteface (regardless of whether they would have passed
256	 *	the input filter, of course).
257	 *
258	 * These statistics don't include packets not yet read from
259	 * the kernel by libpcap, but they may include packets not
260	 * yet read from libpcap by the application.
261	 */
262	ps->ps_recv = p->md.TotAccepted;
263	ps->ps_drop = p->md.TotDrops;
264	ps->ps_ifdrop = p->md.TotMissed - p->md.OrigMissed;
265	return (0);
266}
267
268static void
269pcap_close_pf(pcap_t *p)
270{
271	if (p->buffer != NULL)
272		free(p->buffer);
273	if (p->fd >= 0)
274		close(p->fd);
275}
276
277pcap_t *
278pcap_open_live(const char *device, int snaplen, int promisc, int to_ms,
279    char *ebuf)
280{
281	pcap_t *p;
282	short enmode;
283	int backlog = -1;	/* request the most */
284	struct enfilter Filter;
285	struct endevp devparams;
286
287	p = (pcap_t *)malloc(sizeof(*p));
288	if (p == NULL) {
289		snprintf(ebuf, PCAP_ERRBUF_SIZE,
290		    "pcap_open_live: %s", pcap_strerror(errno));
291		return (0);
292	}
293	memset(p, 0, sizeof(*p));
294
295	/*
296	 * XXX - we assume here that "pfopen()" does not, in fact, modify
297	 * its argument, even though it takes a "char *" rather than a
298	 * "const char *" as its first argument.  That appears to be
299	 * the case, at least on Digital UNIX 4.0.
300	 */
301	p->fd = pfopen(device, O_RDONLY);
302	if (p->fd < 0) {
303		snprintf(ebuf, PCAP_ERRBUF_SIZE, "pf open: %s: %s\n\
304your system may not be properly configured; see the packetfilter(4) man page\n",
305			device, pcap_strerror(errno));
306		goto bad;
307	}
308	p->md.OrigMissed = -1;
309	enmode = ENTSTAMP|ENBATCH|ENNONEXCL;
310	if (promisc)
311		enmode |= ENPROMISC;
312	if (ioctl(p->fd, EIOCMBIS, (caddr_t)&enmode) < 0) {
313		snprintf(ebuf, PCAP_ERRBUF_SIZE, "EIOCMBIS: %s",
314		    pcap_strerror(errno));
315		goto bad;
316	}
317#ifdef	ENCOPYALL
318	/* Try to set COPYALL mode so that we see packets to ourself */
319	enmode = ENCOPYALL;
320	(void)ioctl(p->fd, EIOCMBIS, (caddr_t)&enmode);/* OK if this fails */
321#endif
322	/* set the backlog */
323	if (ioctl(p->fd, EIOCSETW, (caddr_t)&backlog) < 0) {
324		snprintf(ebuf, PCAP_ERRBUF_SIZE, "EIOCSETW: %s",
325		    pcap_strerror(errno));
326		goto bad;
327	}
328	/* discover interface type */
329	if (ioctl(p->fd, EIOCDEVP, (caddr_t)&devparams) < 0) {
330		snprintf(ebuf, PCAP_ERRBUF_SIZE, "EIOCDEVP: %s",
331		    pcap_strerror(errno));
332		goto bad;
333	}
334	/* HACK: to compile prior to Ultrix 4.2 */
335#ifndef	ENDT_FDDI
336#define	ENDT_FDDI	4
337#endif
338	switch (devparams.end_dev_type) {
339
340	case ENDT_10MB:
341		p->linktype = DLT_EN10MB;
342		p->offset = 2;
343		break;
344
345	case ENDT_FDDI:
346		p->linktype = DLT_FDDI;
347		break;
348
349#ifdef ENDT_SLIP
350	case ENDT_SLIP:
351		p->linktype = DLT_SLIP;
352		break;
353#endif
354
355#ifdef ENDT_PPP
356	case ENDT_PPP:
357		p->linktype = DLT_PPP;
358		break;
359#endif
360
361#ifdef ENDT_LOOPBACK
362	case ENDT_LOOPBACK:
363		/*
364		 * It appears to use Ethernet framing, at least on
365		 * Digital UNIX 4.0.
366		 */
367		p->linktype = DLT_EN10MB;
368		p->offset = 2;
369		break;
370#endif
371
372#ifdef ENDT_TRN
373	case ENDT_TRN:
374		p->linktype = DLT_IEEE802;
375		break;
376#endif
377
378	default:
379		/*
380		 * XXX - what about ENDT_IEEE802?  The pfilt.h header
381		 * file calls this "IEEE 802 networks (non-Ethernet)",
382		 * but that doesn't specify a specific link layer type;
383		 * it could be 802.4, or 802.5 (except that 802.5 is
384		 * ENDT_TRN), or 802.6, or 802.11, or....  That's why
385		 * DLT_IEEE802 was hijacked to mean Token Ring in various
386		 * BSDs, and why we went along with that hijacking.
387		 *
388		 * XXX - what about ENDT_HDLC and ENDT_NULL?
389		 * Presumably, as ENDT_OTHER is just "Miscellaneous
390		 * framing", there's not much we can do, as that
391		 * doesn't specify a particular type of header.
392		 */
393		snprintf(ebuf, PCAP_ERRBUF_SIZE, "unknown data-link type %u",
394		    devparams.end_dev_type);
395		goto bad;
396	}
397	/* set truncation */
398#ifdef PCAP_FDDIPAD
399	if (p->linktype == DLT_FDDI)
400		/* packetfilter includes the padding in the snapshot */
401		snaplen += pcap_fddipad;
402#endif
403	if (ioctl(p->fd, EIOCTRUNCATE, (caddr_t)&snaplen) < 0) {
404		snprintf(ebuf, PCAP_ERRBUF_SIZE, "EIOCTRUNCATE: %s",
405		    pcap_strerror(errno));
406		goto bad;
407	}
408	p->snapshot = snaplen;
409	/* accept all packets */
410	memset(&Filter, 0, sizeof(Filter));
411	Filter.enf_Priority = 37;	/* anything > 2 */
412	Filter.enf_FilterLen = 0;	/* means "always true" */
413	if (ioctl(p->fd, EIOCSETF, (caddr_t)&Filter) < 0) {
414		snprintf(ebuf, PCAP_ERRBUF_SIZE, "EIOCSETF: %s",
415		    pcap_strerror(errno));
416		goto bad;
417	}
418
419	if (to_ms != 0) {
420		struct timeval timeout;
421		timeout.tv_sec = to_ms / 1000;
422		timeout.tv_usec = (to_ms * 1000) % 1000000;
423		if (ioctl(p->fd, EIOCSRTIMEOUT, (caddr_t)&timeout) < 0) {
424			snprintf(ebuf, PCAP_ERRBUF_SIZE, "EIOCSRTIMEOUT: %s",
425				pcap_strerror(errno));
426			goto bad;
427		}
428	}
429
430	p->bufsize = BUFSPACE;
431	p->buffer = (u_char*)malloc(p->bufsize + p->offset);
432	if (p->buffer == NULL) {
433		strlcpy(ebuf, pcap_strerror(errno), PCAP_ERRBUF_SIZE);
434		goto bad;
435	}
436
437	/*
438	 * "select()" and "poll()" work on packetfilter devices.
439	 */
440	p->selectable_fd = p->fd;
441
442	p->read_op = pcap_read_pf;
443	p->setfilter_op = pcap_setfilter_pf;
444	p->set_datalink_op = NULL;	/* can't change data link type */
445	p->getnonblock_op = pcap_getnonblock_fd;
446	p->setnonblock_op = pcap_setnonblock_fd;
447	p->stats_op = pcap_stats_pf;
448	p->close_op = pcap_close_pf;
449
450	return (p);
451 bad:
452	if (p->fd >= 0)
453		close(p->fd);
454	free(p);
455	return (NULL);
456}
457
458int
459pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
460{
461	return (0);
462}
463
464static int
465pcap_setfilter_pf(pcap_t *p, struct bpf_program *fp)
466{
467	struct bpf_version bv;
468
469	/*
470	 * See if BIOCVERSION works.  If not, we assume the kernel doesn't
471	 * support BPF-style filters (it's not documented in the bpf(7)
472	 * or packetfiler(7) man pages, but the code used to fail if
473	 * BIOCSETF worked but BIOCVERSION didn't, and I've seen it do
474	 * kernel filtering in DU 4.0, so presumably BIOCVERSION works
475	 * there, at least).
476	 */
477	if (ioctl(p->fd, BIOCVERSION, (caddr_t)&bv) >= 0) {
478		/*
479		 * OK, we have the version of the BPF interpreter;
480		 * is it the same major version as us, and the same
481		 * or better minor version?
482		 */
483		if (bv.bv_major == BPF_MAJOR_VERSION &&
484		    bv.bv_minor >= BPF_MINOR_VERSION) {
485			/*
486			 * Yes.  Try to install the filter.
487			 */
488			if (ioctl(p->fd, BIOCSETF, (caddr_t)fp) < 0) {
489				snprintf(p->errbuf, sizeof(p->errbuf),
490				    "BIOCSETF: %s", pcap_strerror(errno));
491				return (-1);
492			}
493
494			/*
495			 * OK, that succeeded.  We're doing filtering in
496			 * the kernel.  (We assume we don't have a
497			 * userland filter installed - that'd require
498			 * a previous version check to have failed but
499			 * this one to succeed.)
500			 *
501			 * XXX - this message should be supplied to the
502			 * application as a warning of some sort,
503			 * except that if it's a GUI application, it's
504			 * not clear that it should be displayed in
505			 * a window to annoy the user.
506			 */
507			fprintf(stderr, "tcpdump: Using kernel BPF filter\n");
508			p->md.use_bpf = 1;
509			return (0);
510		}
511
512		/*
513		 * We can't use the kernel's BPF interpreter; don't give
514		 * up, just log a message and be inefficient.
515		 *
516		 * XXX - this should really be supplied to the application
517		 * as a warning of some sort.
518		 */
519		fprintf(stderr,
520	    "tcpdump: Requires BPF language %d.%d or higher; kernel is %d.%d\n",
521		    BPF_MAJOR_VERSION, BPF_MINOR_VERSION,
522		    bv.bv_major, bv.bv_minor);
523	}
524
525	/*
526	 * We couldn't do filtering in the kernel; do it in userland.
527	 */
528	if (install_bpf_program(p, fp) < 0)
529		return (-1);
530
531	/*
532	 * XXX - this message should be supplied by the application as
533	 * a warning of some sort.
534	 */
535	fprintf(stderr, "tcpdump: Filtering in user process\n");
536	p->md.use_bpf = 0;
537	return (0);
538}
539