savefile.c revision 75107
1/*
2 * Copyright (c) 1993, 1994, 1995, 1996, 1997
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 * savefile.c - supports offline use of tcpdump
22 *	Extraction/creation by Jeffrey Mogul, DECWRL
23 *	Modified by Steve McCanne, LBL.
24 *
25 * Used to save the received packet headers, after filtering, to
26 * a file, and then read them later.
27 * The first record in the file contains saved values for the machine
28 * dependent values so we can print the dump file on any architecture.
29 */
30
31#ifndef lint
32static const char rcsid[] =
33    "@(#) $Header: /tcpdump/master/libpcap/savefile.c,v 1.49 2000/12/21 10:29:23 guy Exp $ (LBL)";
34#endif
35
36#ifdef HAVE_CONFIG_H
37#include "config.h"
38#endif
39
40#include <sys/types.h>
41#include <sys/time.h>
42
43#include <errno.h>
44#include <memory.h>
45#include <stdio.h>
46#include <stdlib.h>
47#include <string.h>
48#include <unistd.h>
49
50#include "pcap-int.h"
51
52#ifdef HAVE_OS_PROTO_H
53#include "os-proto.h"
54#endif
55
56#define TCPDUMP_MAGIC 0xa1b2c3d4
57#define PATCHED_TCPDUMP_MAGIC 0xa1b2cd34
58
59/*
60 * We use the "receiver-makes-right" approach to byte order,
61 * because time is at a premium when we are writing the file.
62 * In other words, the pcap_file_header and pcap_pkthdr,
63 * records are written in host byte order.
64 * Note that the packets are always written in network byte order.
65 *
66 * ntoh[ls] aren't sufficient because we might need to swap on a big-endian
67 * machine (if the file was written in little-end order).
68 */
69#define	SWAPLONG(y) \
70((((y)&0xff)<<24) | (((y)&0xff00)<<8) | (((y)&0xff0000)>>8) | (((y)>>24)&0xff))
71#define	SWAPSHORT(y) \
72	( (((y)&0xff)<<8) | ((u_short)((y)&0xff00)>>8) )
73
74#define SFERR_TRUNC		1
75#define SFERR_BADVERSION	2
76#define SFERR_BADF		3
77#define SFERR_EOF		4 /* not really an error, just a status */
78
79/*
80 * We don't write DLT_* values to the capture file header, because
81 * they're not the same on all platforms.
82 *
83 * Unfortunately, the various flavors of BSD have not always used the same
84 * numerical values for the same data types, and various patches to
85 * libpcap for non-BSD OSes have added their own DLT_* codes for link
86 * layer encapsulation types seen on those OSes, and those codes have had,
87 * in some cases, values that were also used, on other platforms, for other
88 * link layer encapsulation types.
89 *
90 * This means that capture files of a type whose numerical DLT_* code
91 * means different things on different BSDs, or with different versions
92 * of libpcap, can't always be read on systems other than those like
93 * the one running on the machine on which the capture was made.
94 *
95 * Instead, we define here a set of LINKTYPE_* codes, and map DLT_* codes
96 * to LINKTYPE_* codes when writing a savefile header, and map LINKTYPE_*
97 * codes to DLT_* codes when reading a savefile header.
98 *
99 * For those DLT_* codes that have, as far as we know, the same values on
100 * all platforms (DLT_NULL through DLT_FDDI), we define LINKTYPE_xxx as
101 * DLT_xxx; that way, captures of those types can still be read by
102 * versions of libpcap that map LINKTYPE_* values to DLT_* values, and
103 * captures of those types written by versions of libpcap that map DLT_
104 * values to LINKTYPE_ values can still be read by older versions
105 * of libpcap.
106 *
107 * The other LINKTYPE_* codes are given values starting at 100, in the
108 * hopes that no DLT_* code will be given one of those values.
109 *
110 * In order to ensure that a given LINKTYPE_* code's value will refer to
111 * the same encapsulation type on all platforms, you should not allocate
112 * a new LINKTYPE_* value without consulting "tcpdump-workers@tcpdump.org".
113 * The tcpdump developers will allocate a value for you, and will not
114 * subsequently allocate it to anybody else; that value will be added to
115 * the "pcap.h" in the tcpdump.org CVS repository, so that a future
116 * libpcap release will include it.
117 *
118 * You should, if possible, also contribute patches to libpcap and tcpdump
119 * to handle the new encapsulation type, so that they can also be checked
120 * into the tcpdump.org CVS repository and so that they will appear in
121 * future libpcap and tcpdump releases.
122 */
123#define LINKTYPE_NULL		DLT_NULL
124#define LINKTYPE_ETHERNET	DLT_EN10MB	/* also for 100Mb and up */
125#define LINKTYPE_EXP_ETHERNET	DLT_EN3MB	/* 3Mb experimental Ethernet */
126#define LINKTYPE_AX25		DLT_AX25
127#define LINKTYPE_PRONET		DLT_PRONET
128#define LINKTYPE_CHAOS		DLT_CHAOS
129#define LINKTYPE_TOKEN_RING	DLT_IEEE802	/* DLT_IEEE802 is used for Token Ring */
130#define LINKTYPE_ARCNET		DLT_ARCNET
131#define LINKTYPE_SLIP		DLT_SLIP
132#define LINKTYPE_PPP		DLT_PPP
133#define LINKTYPE_FDDI		DLT_FDDI
134
135/*
136 * LINKTYPE_PPP is for use when there might, or might not, be an RFC 1662
137 * PPP in HDLC-like framing header (with 0xff 0x03 before the PPP protocol
138 * field) at the beginning of the packet.
139 *
140 * This is for use when there is always such a header; the address field
141 * might be 0xff, for regular PPP, or it might be an address field for Cisco
142 * point-to-point with HDLC framing as per section 4.3.1 of RFC 1547 ("Cisco
143 * HDLC").  This is, for example, what you get with NetBSD's DLT_PPP_SERIAL.
144 *
145 * We give it the same value as NetBSD's DLT_PPP_SERIAL, in the hopes that
146 * nobody else will choose a DLT_ value of 50, and so that DLT_PPP_SERIAL
147 * captures will be written out with a link type that NetBSD's tcpdump
148 * can read.
149 */
150#define LINKTYPE_PPP_HDLC	50		/* PPP in HDLC-like framing */
151
152#define LINKTYPE_ATM_RFC1483	100		/* LLC/SNAP-encapsulated ATM */
153#define LINKTYPE_RAW		101		/* raw IP */
154#define LINKTYPE_SLIP_BSDOS	102		/* BSD/OS SLIP BPF header */
155#define LINKTYPE_PPP_BSDOS	103		/* BSD/OS PPP BPF header */
156#define LINKTYPE_C_HDLC		104		/* Cisco HDLC */
157#define LINKTYPE_ATM_CLIP	106		/* Linux Classical IP over ATM */
158
159/*
160 * Reserved for future use.
161 */
162#define LINKTYPE_IEEE802_11	105		/* IEEE 802.11 (wireless) */
163#define LINKTYPE_FR		107		/* BSD/OS Frame Relay */
164#define LINKTYPE_LOOP		108		/* OpenBSD loopback */
165#define LINKTYPE_ENC		109		/* OpenBSD IPSEC enc */
166#define LINKTYPE_LANE8023	110		/* ATM LANE + 802.3 */
167#define LINKTYPE_HIPPI		111		/* NetBSD HIPPI */
168#define LINKTYPE_HDLC		112		/* NetBSD HDLC framing */
169
170#define LINKTYPE_LINUX_SLL	113		/* Linux cooked socket capture */
171
172static struct linktype_map {
173	int	dlt;
174	int	linktype;
175} map[] = {
176	/*
177	 * These DLT_* codes have LINKTYPE_* codes with values identical
178	 * to the values of the corresponding DLT_* code.
179	 */
180	{ DLT_NULL,		LINKTYPE_NULL },
181	{ DLT_EN10MB,		LINKTYPE_ETHERNET },
182	{ DLT_EN3MB,		LINKTYPE_EXP_ETHERNET },
183	{ DLT_AX25,		LINKTYPE_AX25 },
184	{ DLT_PRONET,		LINKTYPE_PRONET },
185	{ DLT_CHAOS,		LINKTYPE_CHAOS },
186	{ DLT_IEEE802,		LINKTYPE_TOKEN_RING },
187	{ DLT_ARCNET,		LINKTYPE_ARCNET },
188	{ DLT_SLIP,		LINKTYPE_SLIP },
189	{ DLT_PPP,		LINKTYPE_PPP },
190	{ DLT_FDDI,	 	LINKTYPE_FDDI },
191
192	/*
193	 * These DLT_* codes have different values on different
194	 * platforms; we map them to LINKTYPE_* codes that
195	 * have values that should never be equal to any DLT_*
196	 * code.
197	 */
198	{ DLT_ATM_RFC1483, 	LINKTYPE_ATM_RFC1483 },
199	{ DLT_RAW,		LINKTYPE_RAW },
200	{ DLT_SLIP_BSDOS,	LINKTYPE_SLIP_BSDOS },
201	{ DLT_PPP_BSDOS,	LINKTYPE_PPP_BSDOS },
202
203	/* BSD/OS Cisco HDLC */
204	{ DLT_C_HDLC,		LINKTYPE_C_HDLC },
205
206	/*
207	 * These DLT_* codes are not on all platforms, but, so far,
208	 * there don't appear to be any platforms that define
209	 * other codes with those values; we map them to
210	 * different LINKTYPE_* values anyway, just in case.
211	 */
212
213	/* Linux ATM Classical IP */
214	{ DLT_ATM_CLIP,		LINKTYPE_ATM_CLIP },
215
216	/* NetBSD sync/async serial PPP (or Cisco HDLC) */
217	{ DLT_PPP_SERIAL,	LINKTYPE_PPP_HDLC },
218
219	/* IEEE 802.11 wireless */
220	{ DLT_IEEE802_11,	LINKTYPE_IEEE802_11 },
221
222	/* OpenBSD loopback */
223	{ DLT_LOOP,		LINKTYPE_LOOP },
224
225	/* Linux cooked socket capture */
226	{ DLT_LINUX_SLL,	LINKTYPE_LINUX_SLL },
227
228	/*
229	 * Any platform that defines additional DLT_* codes should:
230	 *
231	 *	request a LINKTYPE_* code and value from tcpdump.org,
232	 *	as per the above;
233	 *
234	 *	add, in their version of libpcap, an entry to map
235	 *	those DLT_* codes to the corresponding LINKTYPE_*
236	 *	code;
237	 *
238	 *	redefine, in their "net/bpf.h", any DLT_* values
239	 *	that collide with the values used by their additional
240	 *	DLT_* codes, to remove those collisions (but without
241	 *	making them collide with any of the LINKTYPE_*
242	 *	values equal to 50 or above; they should also avoid
243	 *	defining DLT_* values that collide with those
244	 *	LINKTYPE_* values, either).
245	 */
246	{ -1,			-1 }
247};
248
249static int
250dlt_to_linktype(int dlt)
251{
252	int i;
253
254	for (i = 0; map[i].dlt != -1; i++) {
255		if (map[i].dlt == dlt)
256			return (map[i].linktype);
257	}
258
259	/*
260	 * If we don't have a mapping for this DLT_ code, return an
261	 * error; that means that the table above needs to have an
262	 * entry added.
263	 */
264	return (-1);
265}
266
267static int
268linktype_to_dlt(int linktype)
269{
270	int i;
271
272	for (i = 0; map[i].linktype != -1; i++) {
273		if (map[i].linktype == linktype)
274			return (map[i].dlt);
275	}
276
277	/*
278	 * If we don't have an entry for this link type, return
279	 * the link type value; it may be a DLT_ value from an
280	 * older version of libpcap.
281	 */
282	return linktype;
283}
284
285static int
286sf_write_header(FILE *fp, int linktype, int thiszone, int snaplen)
287{
288	struct pcap_file_header hdr;
289
290	hdr.magic = TCPDUMP_MAGIC;
291	hdr.version_major = PCAP_VERSION_MAJOR;
292	hdr.version_minor = PCAP_VERSION_MINOR;
293
294	hdr.thiszone = thiszone;
295	hdr.snaplen = snaplen;
296	hdr.sigfigs = 0;
297	hdr.linktype = linktype;
298
299	if (fwrite((char *)&hdr, sizeof(hdr), 1, fp) != 1)
300		return (-1);
301
302	return (0);
303}
304
305static void
306swap_hdr(struct pcap_file_header *hp)
307{
308	hp->version_major = SWAPSHORT(hp->version_major);
309	hp->version_minor = SWAPSHORT(hp->version_minor);
310	hp->thiszone = SWAPLONG(hp->thiszone);
311	hp->sigfigs = SWAPLONG(hp->sigfigs);
312	hp->snaplen = SWAPLONG(hp->snaplen);
313	hp->linktype = SWAPLONG(hp->linktype);
314}
315
316pcap_t *
317pcap_open_offline(const char *fname, char *errbuf)
318{
319	register pcap_t *p;
320	register FILE *fp;
321	struct pcap_file_header hdr;
322	bpf_u_int32 magic;
323	int linklen;
324
325	p = (pcap_t *)malloc(sizeof(*p));
326	if (p == NULL) {
327		strlcpy(errbuf, "out of swap", PCAP_ERRBUF_SIZE);
328		return (NULL);
329	}
330
331	memset((char *)p, 0, sizeof(*p));
332	/*
333	 * Set this field so we don't close stdin in pcap_close!
334	 */
335	p->fd = -1;
336
337	if (fname[0] == '-' && fname[1] == '\0')
338		fp = stdin;
339	else {
340		fp = fopen(fname, "r");
341		if (fp == NULL) {
342			snprintf(errbuf, PCAP_ERRBUF_SIZE, "%s: %s", fname,
343			    pcap_strerror(errno));
344			goto bad;
345		}
346	}
347	if (fread((char *)&hdr, sizeof(hdr), 1, fp) != 1) {
348		snprintf(errbuf, PCAP_ERRBUF_SIZE, "fread: %s",
349		    pcap_strerror(errno));
350		goto bad;
351	}
352	magic = hdr.magic;
353	if (magic != TCPDUMP_MAGIC && magic != PATCHED_TCPDUMP_MAGIC) {
354		magic = SWAPLONG(magic);
355		if (magic != TCPDUMP_MAGIC && magic != PATCHED_TCPDUMP_MAGIC) {
356			snprintf(errbuf, PCAP_ERRBUF_SIZE,
357			    "bad dump file format");
358			goto bad;
359		}
360		p->sf.swapped = 1;
361		swap_hdr(&hdr);
362	}
363	if (magic == PATCHED_TCPDUMP_MAGIC) {
364		/*
365		 * XXX - the patch that's in some versions of libpcap
366		 * changes the packet header but not the magic number;
367		 * we'd have to use some hacks^H^H^H^H^Hheuristics to
368		 * detect that.
369		 */
370		p->sf.hdrsize = sizeof(struct pcap_sf_patched_pkthdr);
371	} else
372		p->sf.hdrsize = sizeof(struct pcap_sf_pkthdr);
373	if (hdr.version_major < PCAP_VERSION_MAJOR) {
374		snprintf(errbuf, PCAP_ERRBUF_SIZE, "archaic file format");
375		goto bad;
376	}
377	p->tzoff = hdr.thiszone;
378	p->snapshot = hdr.snaplen;
379	p->linktype = linktype_to_dlt(hdr.linktype);
380	p->sf.rfile = fp;
381	p->bufsize = hdr.snaplen;
382
383	/* Align link header as required for proper data alignment */
384	/* XXX should handle all types */
385	switch (p->linktype) {
386
387	case DLT_EN10MB:
388		linklen = 14;
389		break;
390
391	case DLT_FDDI:
392		linklen = 13 + 8;	/* fddi_header + llc */
393		break;
394
395	case DLT_NULL:
396	default:
397		linklen = 0;
398		break;
399	}
400
401	if (p->bufsize < 0)
402		p->bufsize = BPF_MAXBUFSIZE;
403	p->sf.base = (u_char *)malloc(p->bufsize + BPF_ALIGNMENT);
404	if (p->sf.base == NULL) {
405		strlcpy(errbuf, "out of swap", PCAP_ERRBUF_SIZE);
406		goto bad;
407	}
408	p->buffer = p->sf.base + BPF_ALIGNMENT - (linklen % BPF_ALIGNMENT);
409	p->sf.version_major = hdr.version_major;
410	p->sf.version_minor = hdr.version_minor;
411#ifdef PCAP_FDDIPAD
412	/* XXX padding only needed for kernel fcode */
413	pcap_fddipad = 0;
414#endif
415
416	return (p);
417 bad:
418	free(p);
419	return (NULL);
420}
421
422/*
423 * Read sf_readfile and return the next packet.  Return the header in hdr
424 * and the contents in buf.  Return 0 on success, SFERR_EOF if there were
425 * no more packets, and SFERR_TRUNC if a partial packet was encountered.
426 */
427static int
428sf_next_packet(pcap_t *p, struct pcap_pkthdr *hdr, u_char *buf, int buflen)
429{
430	struct pcap_sf_patched_pkthdr sf_hdr;
431	FILE *fp = p->sf.rfile;
432
433	/*
434	 * Read the packet header; the structure we use as a buffer
435	 * is the longer structure for files generated by the patched
436	 * libpcap, but if the file has the magic number for an
437	 * unpatched libpcap we only read as many bytes as the regular
438	 * header has.
439	 */
440	if (fread(&sf_hdr, p->sf.hdrsize, 1, fp) != 1) {
441		/* probably an EOF, though could be a truncated packet */
442		return (1);
443	}
444
445	if (p->sf.swapped) {
446		/* these were written in opposite byte order */
447		hdr->caplen = SWAPLONG(sf_hdr.caplen);
448		hdr->len = SWAPLONG(sf_hdr.len);
449		hdr->ts.tv_sec = SWAPLONG(sf_hdr.ts.tv_sec);
450		hdr->ts.tv_usec = SWAPLONG(sf_hdr.ts.tv_usec);
451	} else {
452		hdr->caplen = sf_hdr.caplen;
453		hdr->len = sf_hdr.len;
454		hdr->ts.tv_sec = sf_hdr.ts.tv_sec;
455		hdr->ts.tv_usec = sf_hdr.ts.tv_usec;
456	}
457	/*
458	 * We interchanged the caplen and len fields at version 2.3,
459	 * in order to match the bpf header layout.  But unfortunately
460	 * some files were written with version 2.3 in their headers
461	 * but without the interchanged fields.
462	 */
463	if (p->sf.version_minor < 3 ||
464	    (p->sf.version_minor == 3 && hdr->caplen > hdr->len)) {
465		int t = hdr->caplen;
466		hdr->caplen = hdr->len;
467		hdr->len = t;
468	}
469
470	if (hdr->caplen > buflen) {
471		/*
472		 * This can happen due to Solaris 2.3 systems tripping
473		 * over the BUFMOD problem and not setting the snapshot
474		 * correctly in the savefile header.  If the caplen isn't
475		 * grossly wrong, try to salvage.
476		 */
477		static u_char *tp = NULL;
478		static int tsize = 0;
479
480		if (hdr->caplen > 65535) {
481			snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
482			    "bogus savefile header");
483			return (-1);
484		}
485
486		if (tsize < hdr->caplen) {
487			tsize = ((hdr->caplen + 1023) / 1024) * 1024;
488			if (tp != NULL)
489				free((u_char *)tp);
490			tp = (u_char *)malloc(tsize);
491			if (tp == NULL) {
492				tsize = 0;
493				snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
494				    "BUFMOD hack malloc");
495				return (-1);
496			}
497		}
498		if (fread((char *)tp, hdr->caplen, 1, fp) != 1) {
499			snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
500			    "truncated dump file");
501			return (-1);
502		}
503		/*
504		 * We can only keep up to buflen bytes.  Since caplen > buflen
505		 * is exactly how we got here, we know we can only keep the
506		 * first buflen bytes and must drop the remainder.  Adjust
507		 * caplen accordingly, so we don't get confused later as
508		 * to how many bytes we have to play with.
509		 */
510		hdr->caplen = buflen;
511		memcpy((char *)buf, (char *)tp, buflen);
512
513	} else {
514		/* read the packet itself */
515
516		if (fread((char *)buf, hdr->caplen, 1, fp) != 1) {
517			snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
518			    "truncated dump file");
519			return (-1);
520		}
521	}
522	return (0);
523}
524
525/*
526 * Print out packets stored in the file initialized by sf_read_init().
527 * If cnt > 0, return after 'cnt' packets, otherwise continue until eof.
528 */
529int
530pcap_offline_read(pcap_t *p, int cnt, pcap_handler callback, u_char *user)
531{
532	struct bpf_insn *fcode = p->fcode.bf_insns;
533	int status = 0;
534	int n = 0;
535
536	while (status == 0) {
537		struct pcap_pkthdr h;
538
539		status = sf_next_packet(p, &h, p->buffer, p->bufsize);
540		if (status) {
541			if (status == 1)
542				return (0);
543			return (status);
544		}
545
546		if (fcode == NULL ||
547		    bpf_filter(fcode, p->buffer, h.len, h.caplen)) {
548			(*callback)(user, &h, p->buffer);
549			if (++n >= cnt && cnt > 0)
550				break;
551		}
552	}
553	/*XXX this breaks semantics tcpslice expects */
554	return (n);
555}
556
557/*
558 * Output a packet to the initialized dump file.
559 */
560void
561pcap_dump(u_char *user, const struct pcap_pkthdr *h, const u_char *sp)
562{
563	register FILE *f;
564	struct pcap_sf_pkthdr sf_hdr;
565
566	f = (FILE *)user;
567	sf_hdr.ts.tv_sec  = h->ts.tv_sec;
568	sf_hdr.ts.tv_usec = h->ts.tv_usec;
569	sf_hdr.caplen     = h->caplen;
570	sf_hdr.len        = h->len;
571	/* XXX we should check the return status */
572	(void)fwrite(&sf_hdr, sizeof(sf_hdr), 1, f);
573	(void)fwrite((char *)sp, h->caplen, 1, f);
574}
575
576/*
577 * Initialize so that sf_write() will output to the file named 'fname'.
578 */
579pcap_dumper_t *
580pcap_dump_open(pcap_t *p, const char *fname)
581{
582	FILE *f;
583	int linktype;
584
585	linktype = dlt_to_linktype(p->linktype);
586	if (linktype == -1) {
587		snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
588		    "%s: link-layer type %d isn't supported in savefiles",
589		    fname, linktype);
590		return (NULL);
591	}
592
593	if (fname[0] == '-' && fname[1] == '\0')
594		f = stdout;
595	else {
596		f = fopen(fname, "w");
597		if (f == NULL) {
598			snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "%s: %s",
599			    fname, pcap_strerror(errno));
600			return (NULL);
601		}
602	}
603	(void)sf_write_header(f, linktype, p->tzoff, p->snapshot);
604	return ((pcap_dumper_t *)f);
605}
606
607void
608pcap_dump_close(pcap_dumper_t *p)
609{
610
611#ifdef notyet
612	if (ferror((FILE *)p))
613		return-an-error;
614	/* XXX should check return from fclose() too */
615#endif
616	(void)fclose((FILE *)p);
617}
618