pcap-bpf.c revision 127664
1/*
2 * Copyright (c) 1993, 1994, 1995, 1996, 1998
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#ifndef lint
22static const char rcsid[] _U_ =
23    "@(#) $Header: /tcpdump/master/libpcap/pcap-bpf.c,v 1.67.2.4 2003/11/22 00:06:28 guy Exp $ (LBL)";
24#endif
25
26#ifdef HAVE_CONFIG_H
27#include "config.h"
28#endif
29
30#include <sys/param.h>			/* optionally get BSD define */
31#include <sys/time.h>
32#include <sys/timeb.h>
33#include <sys/socket.h>
34#include <sys/file.h>
35#include <sys/ioctl.h>
36#include <sys/utsname.h>
37
38#include <net/if.h>
39
40#ifdef _AIX
41
42/*
43 * Make "pcap.h" not include "pcap-bpf.h"; we are going to include the
44 * native OS version, as we need "struct bpf_config" from it.
45 */
46#define PCAP_DONT_INCLUDE_PCAP_BPF_H
47
48#include <sys/types.h>
49
50/*
51 * Prevent bpf.h from redefining the DLT_ values to their
52 * IFT_ values, as we're going to return the standard libpcap
53 * values, not IBM's non-standard IFT_ values.
54 */
55#undef _AIX
56#include <net/bpf.h>
57#define _AIX
58
59#include <net/if_types.h>		/* for IFT_ values */
60#include <sys/sysconfig.h>
61#include <sys/device.h>
62#include <odmi.h>
63#include <cf.h>
64
65#ifdef __64BIT__
66#define domakedev makedev64
67#define getmajor major64
68#define bpf_hdr bpf_hdr32
69#else /* __64BIT__ */
70#define domakedev makedev
71#define getmajor major
72#endif /* __64BIT__ */
73
74#define BPF_NAME "bpf"
75#define BPF_MINORS 4
76#define DRIVER_PATH "/usr/lib/drivers"
77#define BPF_NODE "/dev/bpf"
78static int bpfloadedflag = 0;
79static int odmlockid = 0;
80
81#else /* _AIX */
82
83#include <net/bpf.h>
84
85#endif /* _AIX */
86
87#include <ctype.h>
88#include <errno.h>
89#include <netdb.h>
90#include <stdio.h>
91#include <stdlib.h>
92#include <string.h>
93#include <unistd.h>
94
95#include "pcap-int.h"
96
97#ifdef HAVE_DAG_API
98#include "pcap-dag.h"
99#endif /* HAVE_DAG_API */
100
101#ifdef HAVE_OS_PROTO_H
102#include "os-proto.h"
103#endif
104
105#include "gencode.h"	/* for "no_optimize" */
106
107static int pcap_setfilter_bpf(pcap_t *p, struct bpf_program *fp);
108static int pcap_set_datalink_bpf(pcap_t *p, int dlt);
109
110static int
111pcap_stats_bpf(pcap_t *p, struct pcap_stat *ps)
112{
113	struct bpf_stat s;
114
115	/*
116	 * "ps_recv" counts packets handed to the filter, not packets
117	 * that passed the filter.  This includes packets later dropped
118	 * because we ran out of buffer space.
119	 *
120	 * "ps_drop" counts packets dropped inside the BPF device
121	 * because we ran out of buffer space.  It doesn't count
122	 * packets dropped by the interface driver.  It counts
123	 * only packets that passed the filter.
124	 *
125	 * Both statistics include packets not yet read from the kernel
126	 * by libpcap, and thus not yet seen by the application.
127	 */
128	if (ioctl(p->fd, BIOCGSTATS, (caddr_t)&s) < 0) {
129		snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "BIOCGSTATS: %s",
130		    pcap_strerror(errno));
131		return (-1);
132	}
133
134	ps->ps_recv = s.bs_recv;
135	ps->ps_drop = s.bs_drop;
136	return (0);
137}
138
139static int
140pcap_read_bpf(pcap_t *p, int cnt, pcap_handler callback, u_char *user)
141{
142	int cc;
143	int n = 0;
144	register u_char *bp, *ep;
145	struct bpf_insn *fcode;
146
147	fcode = p->md.use_bpf ? NULL : p->fcode.bf_insns;
148 again:
149	/*
150	 * Has "pcap_breakloop()" been called?
151	 */
152	if (p->break_loop) {
153		/*
154		 * Yes - clear the flag that indicates that it
155		 * has, and return -2 to indicate that we were
156		 * told to break out of the loop.
157		 */
158		p->break_loop = 0;
159		return (-2);
160	}
161	cc = p->cc;
162	if (p->cc == 0) {
163		cc = read(p->fd, (char *)p->buffer, p->bufsize);
164		if (cc < 0) {
165			/* Don't choke when we get ptraced */
166			switch (errno) {
167
168			case EINTR:
169				goto again;
170
171#ifdef _AIX
172			case EFAULT:
173				/*
174				 * Sigh.  More AIX wonderfulness.
175				 *
176				 * For some unknown reason the uiomove()
177				 * operation in the bpf kernel extension
178				 * used to copy the buffer into user
179				 * space sometimes returns EFAULT. I have
180				 * no idea why this is the case given that
181				 * a kernel debugger shows the user buffer
182				 * is correct. This problem appears to
183				 * be mostly mitigated by the memset of
184				 * the buffer before it is first used.
185				 * Very strange.... Shaun Clowes
186				 *
187				 * In any case this means that we shouldn't
188				 * treat EFAULT as a fatal error; as we
189				 * don't have an API for returning
190				 * a "some packets were dropped since
191				 * the last packet you saw" indication,
192				 * we just ignore EFAULT and keep reading.
193				 */
194				goto again;
195#endif
196
197			case EWOULDBLOCK:
198				return (0);
199#if defined(sun) && !defined(BSD)
200			/*
201			 * Due to a SunOS bug, after 2^31 bytes, the kernel
202			 * file offset overflows and read fails with EINVAL.
203			 * The lseek() to 0 will fix things.
204			 */
205			case EINVAL:
206				if (lseek(p->fd, 0L, SEEK_CUR) +
207				    p->bufsize < 0) {
208					(void)lseek(p->fd, 0L, SEEK_SET);
209					goto again;
210				}
211				/* fall through */
212#endif
213			}
214			snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "read: %s",
215			    pcap_strerror(errno));
216			return (-1);
217		}
218		bp = p->buffer;
219	} else
220		bp = p->bp;
221
222	/*
223	 * Loop through each packet.
224	 */
225#define bhp ((struct bpf_hdr *)bp)
226	ep = bp + cc;
227	while (bp < ep) {
228		register int caplen, hdrlen;
229
230		/*
231		 * Has "pcap_breakloop()" been called?
232		 * If so, return immediately - if we haven't read any
233		 * packets, clear the flag and return -2 to indicate
234		 * that we were told to break out of the loop, otherwise
235		 * leave the flag set, so that the *next* call will break
236		 * out of the loop without having read any packets, and
237		 * return the number of packets we've processed so far.
238		 */
239		if (p->break_loop) {
240			if (n == 0) {
241				p->break_loop = 0;
242				return (-2);
243			} else {
244				p->bp = bp;
245				p->cc = ep - bp;
246				return (n);
247			}
248		}
249
250		caplen = bhp->bh_caplen;
251		hdrlen = bhp->bh_hdrlen;
252		/*
253		 * Short-circuit evaluation: if using BPF filter
254		 * in kernel, no need to do it now.
255		 */
256		if (fcode == NULL ||
257		    bpf_filter(fcode, bp + hdrlen, bhp->bh_datalen, caplen)) {
258#ifdef _AIX
259			/*
260			 * AIX's BPF returns seconds/nanoseconds time
261			 * stamps, not seconds/microseconds time stamps.
262			 *
263			 * XXX - I'm guessing here that it's a "struct
264			 * timestamp"; if not, this code won't compile,
265			 * but, if not, you want to send us a bug report
266			 * and fall back on using DLPI.  It's not as if
267			 * BPF used to work right on AIX before this
268			 * change; this change attempts to fix the fact
269			 * that it didn't....
270			 */
271			bhp->bh_tstamp.tv_usec = bhp->bh_tstamp.tv_usec/1000;
272#endif
273			/*
274			 * XXX A bpf_hdr matches a pcap_pkthdr.
275			 */
276			(*callback)(user, (struct pcap_pkthdr*)bp, bp + hdrlen);
277			bp += BPF_WORDALIGN(caplen + hdrlen);
278			if (++n >= cnt && cnt > 0) {
279				p->bp = bp;
280				p->cc = ep - bp;
281				return (n);
282			}
283		} else {
284			/*
285			 * Skip this packet.
286			 */
287			bp += BPF_WORDALIGN(caplen + hdrlen);
288		}
289	}
290#undef bhp
291	p->cc = 0;
292	return (n);
293}
294
295#ifdef _AIX
296static int
297bpf_odminit(char *errbuf)
298{
299	char *errstr;
300
301	if (odm_initialize() == -1) {
302		if (odm_err_msg(odmerrno, &errstr) == -1)
303			errstr = "Unknown error";
304		snprintf(errbuf, PCAP_ERRBUF_SIZE,
305		    "bpf_load: odm_initialize failed: %s",
306		    errstr);
307		return (-1);
308	}
309
310	if ((odmlockid = odm_lock("/etc/objrepos/config_lock", ODM_WAIT)) == -1) {
311		if (odm_err_msg(odmerrno, &errstr) == -1)
312			errstr = "Unknown error";
313		snprintf(errbuf, PCAP_ERRBUF_SIZE,
314		    "bpf_load: odm_lock of /etc/objrepos/config_lock failed: %s",
315		    errstr);
316		return (-1);
317	}
318
319	return (0);
320}
321
322static int
323bpf_odmcleanup(char *errbuf)
324{
325	char *errstr;
326
327	if (odm_unlock(odmlockid) == -1) {
328		if (odm_err_msg(odmerrno, &errstr) == -1)
329			errstr = "Unknown error";
330		snprintf(errbuf, PCAP_ERRBUF_SIZE,
331		    "bpf_load: odm_unlock failed: %s",
332		    errstr);
333		return (-1);
334	}
335
336	if (odm_terminate() == -1) {
337		if (odm_err_msg(odmerrno, &errstr) == -1)
338			errstr = "Unknown error";
339		snprintf(errbuf, PCAP_ERRBUF_SIZE,
340		    "bpf_load: odm_terminate failed: %s",
341		    errstr);
342		return (-1);
343	}
344
345	return (0);
346}
347
348static int
349bpf_load(char *errbuf)
350{
351	long major;
352	int *minors;
353	int numminors, i, rc;
354	char buf[1024];
355	struct stat sbuf;
356	struct bpf_config cfg_bpf;
357	struct cfg_load cfg_ld;
358	struct cfg_kmod cfg_km;
359
360	/*
361	 * This is very very close to what happens in the real implementation
362	 * but I've fixed some (unlikely) bug situations.
363	 */
364	if (bpfloadedflag)
365		return (0);
366
367	if (bpf_odminit(errbuf) != 0)
368		return (-1);
369
370	major = genmajor(BPF_NAME);
371	if (major == -1) {
372		snprintf(errbuf, PCAP_ERRBUF_SIZE,
373		    "bpf_load: genmajor failed: %s", pcap_strerror(errno));
374		return (-1);
375	}
376
377	minors = getminor(major, &numminors, BPF_NAME);
378	if (!minors) {
379		minors = genminor("bpf", major, 0, BPF_MINORS, 1, 1);
380		if (!minors) {
381			snprintf(errbuf, PCAP_ERRBUF_SIZE,
382			    "bpf_load: genminor failed: %s",
383			    pcap_strerror(errno));
384			return (-1);
385		}
386	}
387
388	if (bpf_odmcleanup(errbuf))
389		return (-1);
390
391	rc = stat(BPF_NODE "0", &sbuf);
392	if (rc == -1 && errno != ENOENT) {
393		snprintf(errbuf, PCAP_ERRBUF_SIZE,
394		    "bpf_load: can't stat %s: %s",
395		    BPF_NODE "0", pcap_strerror(errno));
396		return (-1);
397	}
398
399	if (rc == -1 || getmajor(sbuf.st_rdev) != major) {
400		for (i = 0; i < BPF_MINORS; i++) {
401			sprintf(buf, "%s%d", BPF_NODE, i);
402			unlink(buf);
403			if (mknod(buf, S_IRUSR | S_IFCHR, domakedev(major, i)) == -1) {
404				snprintf(errbuf, PCAP_ERRBUF_SIZE,
405				    "bpf_load: can't mknod %s: %s",
406				    buf, pcap_strerror(errno));
407				return (-1);
408			}
409		}
410	}
411
412	/* Check if the driver is loaded */
413	memset(&cfg_ld, 0x0, sizeof(cfg_ld));
414	cfg_ld.path = buf;
415	sprintf(cfg_ld.path, "%s/%s", DRIVER_PATH, BPF_NAME);
416	if ((sysconfig(SYS_QUERYLOAD, (void *)&cfg_ld, sizeof(cfg_ld)) == -1) ||
417	    (cfg_ld.kmid == 0)) {
418		/* Driver isn't loaded, load it now */
419		if (sysconfig(SYS_SINGLELOAD, (void *)&cfg_ld, sizeof(cfg_ld)) == -1) {
420			snprintf(errbuf, PCAP_ERRBUF_SIZE,
421			    "bpf_load: could not load driver: %s",
422			    strerror(errno));
423			return (-1);
424		}
425	}
426
427	/* Configure the driver */
428	cfg_km.cmd = CFG_INIT;
429	cfg_km.kmid = cfg_ld.kmid;
430	cfg_km.mdilen = sizeof(cfg_bpf);
431	cfg_km.mdiptr = (void *)&cfg_bpf;
432	for (i = 0; i < BPF_MINORS; i++) {
433		cfg_bpf.devno = domakedev(major, i);
434		if (sysconfig(SYS_CFGKMOD, (void *)&cfg_km, sizeof(cfg_km)) == -1) {
435			snprintf(errbuf, PCAP_ERRBUF_SIZE,
436			    "bpf_load: could not configure driver: %s",
437			    strerror(errno));
438			return (-1);
439		}
440	}
441
442	bpfloadedflag = 1;
443
444	return (0);
445}
446#endif
447
448static inline int
449bpf_open(pcap_t *p, char *errbuf)
450{
451	int fd;
452	int n = 0;
453	char device[sizeof "/dev/bpf0000000000"];
454
455#ifdef _AIX
456	/*
457	 * Load the bpf driver, if it isn't already loaded,
458	 * and create the BPF device entries, if they don't
459	 * already exist.
460	 */
461	if (bpf_load(errbuf) == -1)
462		return (-1);
463#endif
464
465	/*
466	 * Go through all the minors and find one that isn't in use.
467	 */
468	do {
469		(void)snprintf(device, sizeof(device), "/dev/bpf%d", n++);
470		fd = open(device, O_RDONLY);
471	} while (fd < 0 && errno == EBUSY);
472
473	/*
474	 * XXX better message for all minors used
475	 */
476	if (fd < 0)
477		snprintf(errbuf, PCAP_ERRBUF_SIZE, "(no devices found) %s: %s",
478		    device, pcap_strerror(errno));
479
480	return (fd);
481}
482
483static void
484pcap_close_bpf(pcap_t *p)
485{
486	if (p->buffer != NULL)
487		free(p->buffer);
488	if (p->fd >= 0)
489		close(p->fd);
490}
491
492/*
493 * XXX - on AIX, IBM's tcpdump (and perhaps the incompatible-with-everybody-
494 * else's libpcap in AIX 5.1) appears to forcibly load the BPF driver
495 * if it's not already loaded, and to create the BPF devices if they
496 * don't exist.
497 *
498 * It'd be nice if we could do the same, although the code to do so
499 * might be version-dependent, alas (the way to do it isn't necessarily
500 * documented).
501 */
502pcap_t *
503pcap_open_live(const char *device, int snaplen, int promisc, int to_ms,
504    char *ebuf)
505{
506	int fd;
507	struct ifreq ifr;
508	struct bpf_version bv;
509#ifdef BIOCGDLTLIST
510	struct bpf_dltlist bdl;
511#endif
512	u_int v;
513	pcap_t *p;
514	struct utsname osinfo;
515
516#ifdef HAVE_DAG_API
517	if (strstr(device, "dag")) {
518		return dag_open_live(device, snaplen, promisc, to_ms, ebuf);
519	}
520#endif /* HAVE_DAG_API */
521
522#ifdef BIOCGDLTLIST
523	memset(&bdl, 0, sizeof(bdl));
524#endif
525
526	p = (pcap_t *)malloc(sizeof(*p));
527	if (p == NULL) {
528		snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
529		    pcap_strerror(errno));
530		return (NULL);
531	}
532	memset(p, 0, sizeof(*p));
533	fd = bpf_open(p, ebuf);
534	if (fd < 0)
535		goto bad;
536
537	p->fd = fd;
538	p->snapshot = snaplen;
539
540	if (ioctl(fd, BIOCVERSION, (caddr_t)&bv) < 0) {
541		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCVERSION: %s",
542		    pcap_strerror(errno));
543		goto bad;
544	}
545	if (bv.bv_major != BPF_MAJOR_VERSION ||
546	    bv.bv_minor < BPF_MINOR_VERSION) {
547		snprintf(ebuf, PCAP_ERRBUF_SIZE,
548		    "kernel bpf filter out of date");
549		goto bad;
550	}
551
552	/*
553	 * Try finding a good size for the buffer; 32768 may be too
554	 * big, so keep cutting it in half until we find a size
555	 * that works, or run out of sizes to try.  If the default
556	 * is larger, don't make it smaller.
557	 *
558	 * XXX - there should be a user-accessible hook to set the
559	 * initial buffer size.
560	 */
561	if ((ioctl(fd, BIOCGBLEN, (caddr_t)&v) < 0) || v < 32768)
562		v = 32768;
563	for ( ; v != 0; v >>= 1) {
564		/* Ignore the return value - this is because the call fails
565		 * on BPF systems that don't have kernel malloc.  And if
566		 * the call fails, it's no big deal, we just continue to
567		 * use the standard buffer size.
568		 */
569		(void) ioctl(fd, BIOCSBLEN, (caddr_t)&v);
570
571		(void)strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
572		if (ioctl(fd, BIOCSETIF, (caddr_t)&ifr) >= 0)
573			break;	/* that size worked; we're done */
574
575		if (errno != ENOBUFS) {
576			snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCSETIF: %s: %s",
577			    device, pcap_strerror(errno));
578			goto bad;
579		}
580	}
581
582	if (v == 0) {
583		snprintf(ebuf, PCAP_ERRBUF_SIZE,
584			 "BIOCSBLEN: %s: No buffer size worked", device);
585		goto bad;
586	}
587
588	/* Get the data link layer type. */
589	if (ioctl(fd, BIOCGDLT, (caddr_t)&v) < 0) {
590		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCGDLT: %s",
591		    pcap_strerror(errno));
592		goto bad;
593	}
594#ifdef _AIX
595	/*
596	 * AIX's BPF returns IFF_ types, not DLT_ types, in BIOCGDLT.
597	 */
598	switch (v) {
599
600	case IFT_ETHER:
601	case IFT_ISO88023:
602		v = DLT_EN10MB;
603		break;
604
605	case IFT_FDDI:
606		v = DLT_FDDI;
607		break;
608
609	case IFT_ISO88025:
610		v = DLT_IEEE802;
611		break;
612
613	case IFT_LOOP:
614		v = DLT_NULL;
615		break;
616
617	default:
618		/*
619		 * We don't know what to map this to yet.
620		 */
621		snprintf(ebuf, PCAP_ERRBUF_SIZE, "unknown interface type %u",
622		    v);
623		goto bad;
624	}
625#endif
626#if _BSDI_VERSION - 0 >= 199510
627	/* The SLIP and PPP link layer header changed in BSD/OS 2.1 */
628	switch (v) {
629
630	case DLT_SLIP:
631		v = DLT_SLIP_BSDOS;
632		break;
633
634	case DLT_PPP:
635		v = DLT_PPP_BSDOS;
636		break;
637
638	case 11:	/*DLT_FR*/
639		v = DLT_FRELAY;
640		break;
641
642	case 12:	/*DLT_C_HDLC*/
643		v = DLT_CHDLC;
644		break;
645	}
646#endif
647	p->linktype = v;
648
649#ifdef BIOCGDLTLIST
650	/*
651	 * We know the default link type -- now determine all the DLTs
652	 * this interface supports.  If this fails with EINVAL, it's
653	 * not fatal; we just don't get to use the feature later.
654	 */
655	if (ioctl(fd, BIOCGDLTLIST, (caddr_t)&bdl) == 0) {
656		bdl.bfl_list = (u_int *) malloc(sizeof(u_int) * bdl.bfl_len);
657		if (bdl.bfl_list == NULL) {
658			(void)snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
659			    pcap_strerror(errno));
660			goto bad;
661		}
662
663		if (ioctl(fd, BIOCGDLTLIST, (caddr_t)&bdl) < 0) {
664			(void)snprintf(ebuf, PCAP_ERRBUF_SIZE,
665			    "BIOCGDLTLIST: %s", pcap_strerror(errno));
666			goto bad;
667		}
668
669		p->dlt_count = bdl.bfl_len;
670		p->dlt_list = bdl.bfl_list;
671	} else {
672		if (errno != EINVAL) {
673			(void)snprintf(ebuf, PCAP_ERRBUF_SIZE,
674			    "BIOCGDLTLIST: %s", pcap_strerror(errno));
675			goto bad;
676		}
677	}
678#endif
679
680	/* set timeout */
681	if (to_ms != 0) {
682		/*
683		 * XXX - is this seconds/nanoseconds in AIX?
684		 * (Treating it as such doesn't fix the timeout
685		 * problem described below.)
686		 */
687		struct timeval to;
688		to.tv_sec = to_ms / 1000;
689		to.tv_usec = (to_ms * 1000) % 1000000;
690		if (ioctl(p->fd, BIOCSRTIMEOUT, (caddr_t)&to) < 0) {
691			snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCSRTIMEOUT: %s",
692			    pcap_strerror(errno));
693			goto bad;
694		}
695	}
696
697#ifdef _AIX
698#ifdef	BIOCIMMEDIATE
699	/*
700	 * Darren Reed notes that
701	 *
702	 *	On AIX (4.2 at least), if BIOCIMMEDIATE is not set, the
703	 *	timeout appears to be ignored and it waits until the buffer
704	 *	is filled before returning.  The result of not having it
705	 *	set is almost worse than useless if your BPF filter
706	 *	is reducing things to only a few packets (i.e. one every
707	 *	second or so).
708	 *
709	 * so we turn BIOCIMMEDIATE mode on if this is AIX.
710	 *
711	 * We don't turn it on for other platforms, as that means we
712	 * get woken up for every packet, which may not be what we want;
713	 * in the Winter 1993 USENIX paper on BPF, they say:
714	 *
715	 *	Since a process might want to look at every packet on a
716	 *	network and the time between packets can be only a few
717	 *	microseconds, it is not possible to do a read system call
718	 *	per packet and BPF must collect the data from several
719	 *	packets and return it as a unit when the monitoring
720	 *	application does a read.
721	 *
722	 * which I infer is the reason for the timeout - it means we
723	 * wait that amount of time, in the hopes that more packets
724	 * will arrive and we'll get them all with one read.
725	 *
726	 * Setting BIOCIMMEDIATE mode on FreeBSD (and probably other
727	 * BSDs) causes the timeout to be ignored.
728	 *
729	 * On the other hand, some platforms (e.g., Linux) don't support
730	 * timeouts, they just hand stuff to you as soon as it arrives;
731	 * if that doesn't cause a problem on those platforms, it may
732	 * be OK to have BIOCIMMEDIATE mode on BSD as well.
733	 *
734	 * (Note, though, that applications may depend on the read
735	 * completing, even if no packets have arrived, when the timeout
736	 * expires, e.g. GUI applications that have to check for input
737	 * while waiting for packets to arrive; a non-zero timeout
738	 * prevents "select()" from working right on FreeBSD and
739	 * possibly other BSDs, as the timer doesn't start until a
740	 * "read()" is done, so the timer isn't in effect if the
741	 * application is blocked on a "select()", and the "select()"
742	 * doesn't get woken up for a BPF device until the buffer
743	 * fills up.)
744	 */
745	v = 1;
746	if (ioctl(p->fd, BIOCIMMEDIATE, &v) < 0) {
747		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCIMMEDIATE: %s",
748		    pcap_strerror(errno));
749		goto bad;
750	}
751#endif	/* BIOCIMMEDIATE */
752#endif	/* _AIX */
753
754	if (promisc) {
755		/* set promiscuous mode, okay if it fails */
756		if (ioctl(p->fd, BIOCPROMISC, NULL) < 0) {
757			snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCPROMISC: %s",
758			    pcap_strerror(errno));
759		}
760	}
761
762	if (ioctl(fd, BIOCGBLEN, (caddr_t)&v) < 0) {
763		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCGBLEN: %s",
764		    pcap_strerror(errno));
765		goto bad;
766	}
767	p->bufsize = v;
768	p->buffer = (u_char *)malloc(p->bufsize);
769	if (p->buffer == NULL) {
770		snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
771		    pcap_strerror(errno));
772		goto bad;
773	}
774#ifdef _AIX
775	/* For some strange reason this seems to prevent the EFAULT
776	 * problems we have experienced from AIX BPF. */
777	memset(p->buffer, 0x0, p->bufsize);
778#endif
779
780	/*
781	 * On most BPF platforms, either you can do a "select()" or
782	 * "poll()" on a BPF file descriptor and it works correctly,
783	 * or you can do it and it will return "readable" if the
784	 * hold buffer is full but not if the timeout expires *and*
785	 * a non-blocking read will, if the hold buffer is empty
786	 * but the store buffer isn't empty, rotate the buffers
787	 * and return what packets are available.
788	 *
789	 * In the latter case, the fact that a non-blocking read
790	 * will give you the available packets means you can work
791	 * around the failure of "select()" and "poll()" to wake up
792	 * and return "readable" when the timeout expires by using
793	 * the timeout as the "select()" or "poll()" timeout, putting
794	 * the BPF descriptor into non-blocking mode, and read from
795	 * it regardless of whether "select()" reports it as readable
796	 * or not.
797	 *
798	 * However, in FreeBSD 4.3 and 4.4, "select()" and "poll()"
799	 * won't wake up and return "readable" if the timer expires
800	 * and non-blocking reads return EWOULDBLOCK if the hold
801	 * buffer is empty, even if the store buffer is non-empty.
802	 *
803	 * This means the workaround in question won't work.
804	 *
805	 * Therefore, on FreeBSD 4.3 and 4.4, we set "p->selectable_fd"
806	 * to -1, which means "sorry, you can't use 'select()' or 'poll()'
807	 * here".  On all other BPF platforms, we set it to the FD for
808	 * the BPF device; in NetBSD, OpenBSD, and Darwin, a non-blocking
809	 * read will, if the hold buffer is empty and the store buffer
810	 * isn't empty, rotate the buffers and return what packets are
811	 * there (and in sufficiently recent versions of OpenBSD
812	 * "select()" and "poll()" should work correctly).
813	 *
814	 * XXX - what about AIX?
815	 */
816	if (uname(&osinfo) == 0) {
817		/*
818		 * We can check what OS this is.
819		 */
820		if (strcmp(osinfo.sysname, "FreeBSD") == 0 &&
821		    (strcmp(osinfo.release, "4.3") == 0 ||
822		     strcmp(osinfo.release, "4.4") == 0))
823			p->selectable_fd = -1;
824		else
825			p->selectable_fd = p->fd;
826	} else {
827		/*
828		 * We can't find out what OS this is, so assume we can
829		 * do a "select()" or "poll()".
830		 */
831		p->selectable_fd = p->fd;
832	}
833
834	p->read_op = pcap_read_bpf;
835	p->setfilter_op = pcap_setfilter_bpf;
836	p->set_datalink_op = pcap_set_datalink_bpf;
837	p->getnonblock_op = pcap_getnonblock_fd;
838	p->setnonblock_op = pcap_setnonblock_fd;
839	p->stats_op = pcap_stats_bpf;
840	p->close_op = pcap_close_bpf;
841
842	return (p);
843 bad:
844	(void)close(fd);
845#ifdef BIOCGDLTLIST
846	if (bdl.bfl_list != NULL)
847		free(bdl.bfl_list);
848#endif
849	free(p);
850	return (NULL);
851}
852
853int
854pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
855{
856#ifdef HAVE_DAG_API
857	if (dag_platform_finddevs(alldevsp, errbuf) < 0)
858		return (-1);
859#endif /* HAVE_DAG_API */
860
861	return (0);
862}
863
864static int
865pcap_setfilter_bpf(pcap_t *p, struct bpf_program *fp)
866{
867	/*
868	 * It looks that BPF code generated by gen_protochain() is not
869	 * compatible with some of kernel BPF code (for example BSD/OS 3.1).
870	 * Take a safer side for now.
871	 */
872	if (no_optimize) {
873		/*
874		 * XXX - what if we already have a filter in the kernel?
875		 */
876		if (install_bpf_program(p, fp) < 0)
877			return (-1);
878		p->md.use_bpf = 0;	/* filtering in userland */
879		return (0);
880	}
881
882	/*
883	 * Free any user-mode filter we might happen to have installed.
884	 */
885	pcap_freecode(&p->fcode);
886
887	/*
888	 * Try to install the kernel filter.
889	 */
890	if (ioctl(p->fd, BIOCSETF, (caddr_t)fp) < 0) {
891		snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "BIOCSETF: %s",
892		    pcap_strerror(errno));
893		return (-1);
894	}
895	p->md.use_bpf = 1;	/* filtering in the kernel */
896	return (0);
897}
898
899static int
900pcap_set_datalink_bpf(pcap_t *p, int dlt)
901{
902#ifdef BIOCSDLT
903	if (ioctl(p->fd, BIOCSDLT, &dlt) == -1) {
904		(void) snprintf(p->errbuf, sizeof(p->errbuf),
905		    "Cannot set DLT %d: %s", dlt, strerror(errno));
906		return (-1);
907	}
908#endif
909	return (0);
910}
911