Home | History | Annotate | Download | only in libpcap
      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
     22 static const char rcsid[] _U_ =
     23     "@(#) $Header: /tcpdump/master/libpcap/pcap-bpf.c,v 1.86.2.12 2007/06/15 17:57:27 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 <sys/cfgodm.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"
     78 static int bpfloadedflag = 0;
     79 static 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 
    107 static int pcap_setfilter_bpf(pcap_t *p, struct bpf_program *fp);
    108 static int pcap_setdirection_bpf(pcap_t *, pcap_direction_t);
    109 static int pcap_set_datalink_bpf(pcap_t *p, int dlt);
    110 
    111 static int
    112 pcap_stats_bpf(pcap_t *p, struct pcap_stat *ps)
    113 {
    114 	struct bpf_stat s;
    115 
    116 	/*
    117 	 * "ps_recv" counts packets handed to the filter, not packets
    118 	 * that passed the filter.  This includes packets later dropped
    119 	 * because we ran out of buffer space.
    120 	 *
    121 	 * "ps_drop" counts packets dropped inside the BPF device
    122 	 * because we ran out of buffer space.  It doesn't count
    123 	 * packets dropped by the interface driver.  It counts
    124 	 * only packets that passed the filter.
    125 	 *
    126 	 * Both statistics include packets not yet read from the kernel
    127 	 * by libpcap, and thus not yet seen by the application.
    128 	 */
    129 	if (ioctl(p->fd, BIOCGSTATS, (caddr_t)&s) < 0) {
    130 		snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "BIOCGSTATS: %s",
    131 		    pcap_strerror(errno));
    132 		return (-1);
    133 	}
    134 
    135 	ps->ps_recv = s.bs_recv;
    136 	ps->ps_drop = s.bs_drop;
    137 	return (0);
    138 }
    139 
    140 static int
    141 pcap_read_bpf(pcap_t *p, int cnt, pcap_handler callback, u_char *user)
    142 {
    143 	int cc;
    144 	int n = 0;
    145 	register u_char *bp, *ep;
    146 	u_char *datap;
    147 	struct bpf_insn *fcode;
    148 #ifdef PCAP_FDDIPAD
    149 	register int pad;
    150 #endif
    151 
    152 	fcode = p->md.use_bpf ? NULL : p->fcode.bf_insns;
    153  again:
    154 	/*
    155 	 * Has "pcap_breakloop()" been called?
    156 	 */
    157 	if (p->break_loop) {
    158 		/*
    159 		 * Yes - clear the flag that indicates that it
    160 		 * has, and return -2 to indicate that we were
    161 		 * told to break out of the loop.
    162 		 */
    163 		p->break_loop = 0;
    164 		return (-2);
    165 	}
    166 	cc = p->cc;
    167 	if (p->cc == 0) {
    168 		cc = read(p->fd, (char *)p->buffer, p->bufsize);
    169 		if (cc < 0) {
    170 			/* Don't choke when we get ptraced */
    171 			switch (errno) {
    172 
    173 			case EINTR:
    174 				goto again;
    175 
    176 #ifdef _AIX
    177 			case EFAULT:
    178 				/*
    179 				 * Sigh.  More AIX wonderfulness.
    180 				 *
    181 				 * For some unknown reason the uiomove()
    182 				 * operation in the bpf kernel extension
    183 				 * used to copy the buffer into user
    184 				 * space sometimes returns EFAULT. I have
    185 				 * no idea why this is the case given that
    186 				 * a kernel debugger shows the user buffer
    187 				 * is correct. This problem appears to
    188 				 * be mostly mitigated by the memset of
    189 				 * the buffer before it is first used.
    190 				 * Very strange.... Shaun Clowes
    191 				 *
    192 				 * In any case this means that we shouldn't
    193 				 * treat EFAULT as a fatal error; as we
    194 				 * don't have an API for returning
    195 				 * a "some packets were dropped since
    196 				 * the last packet you saw" indication,
    197 				 * we just ignore EFAULT and keep reading.
    198 				 */
    199 				goto again;
    200 #endif
    201 
    202 			case EWOULDBLOCK:
    203 				return (0);
    204 #if defined(sun) && !defined(BSD)
    205 			/*
    206 			 * Due to a SunOS bug, after 2^31 bytes, the kernel
    207 			 * file offset overflows and read fails with EINVAL.
    208 			 * The lseek() to 0 will fix things.
    209 			 */
    210 			case EINVAL:
    211 				if (lseek(p->fd, 0L, SEEK_CUR) +
    212 				    p->bufsize < 0) {
    213 					(void)lseek(p->fd, 0L, SEEK_SET);
    214 					goto again;
    215 				}
    216 				/* fall through */
    217 #endif
    218 			}
    219 			snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "read: %s",
    220 			    pcap_strerror(errno));
    221 			return (-1);
    222 		}
    223 		bp = p->buffer;
    224 	} else
    225 		bp = p->bp;
    226 
    227 	/*
    228 	 * Loop through each packet.
    229 	 */
    230 #define bhp ((struct bpf_hdr *)bp)
    231 	ep = bp + cc;
    232 #ifdef PCAP_FDDIPAD
    233 	pad = p->fddipad;
    234 #endif
    235 	while (bp < ep) {
    236 		register int caplen, hdrlen;
    237 
    238 		/*
    239 		 * Has "pcap_breakloop()" been called?
    240 		 * If so, return immediately - if we haven't read any
    241 		 * packets, clear the flag and return -2 to indicate
    242 		 * that we were told to break out of the loop, otherwise
    243 		 * leave the flag set, so that the *next* call will break
    244 		 * out of the loop without having read any packets, and
    245 		 * return the number of packets we've processed so far.
    246 		 */
    247 		if (p->break_loop) {
    248 			if (n == 0) {
    249 				p->break_loop = 0;
    250 				return (-2);
    251 			} else {
    252 				p->bp = bp;
    253 				p->cc = ep - bp;
    254 				return (n);
    255 			}
    256 		}
    257 
    258 		caplen = bhp->bh_caplen;
    259 		hdrlen = bhp->bh_hdrlen;
    260 		datap = bp + hdrlen;
    261 		/*
    262 		 * Short-circuit evaluation: if using BPF filter
    263 		 * in kernel, no need to do it now.
    264 		 *
    265 #ifdef PCAP_FDDIPAD
    266 		 * Note: the filter code was generated assuming
    267 		 * that p->fddipad was the amount of padding
    268 		 * before the header, as that's what's required
    269 		 * in the kernel, so we run the filter before
    270 		 * skipping that padding.
    271 #endif
    272 		 */
    273 		if (fcode == NULL ||
    274 		    bpf_filter(fcode, datap, bhp->bh_datalen, caplen)) {
    275 			struct pcap_pkthdr pkthdr;
    276 
    277 			pkthdr.ts.tv_sec = bhp->bh_tstamp.tv_sec;
    278 #ifdef _AIX
    279 			/*
    280 			 * AIX's BPF returns seconds/nanoseconds time
    281 			 * stamps, not seconds/microseconds time stamps.
    282 			 */
    283 			pkthdr.ts.tv_usec = bhp->bh_tstamp.tv_usec/1000;
    284 #else
    285 			pkthdr.ts.tv_usec = bhp->bh_tstamp.tv_usec;
    286 #endif
    287 #ifdef PCAP_FDDIPAD
    288 			if (caplen > pad)
    289 				pkthdr.caplen = caplen - pad;
    290 			else
    291 				pkthdr.caplen = 0;
    292 			if (bhp->bh_datalen > pad)
    293 				pkthdr.len = bhp->bh_datalen - pad;
    294 			else
    295 				pkthdr.len = 0;
    296 			datap += pad;
    297 #else
    298 			pkthdr.caplen = caplen;
    299 			pkthdr.len = bhp->bh_datalen;
    300 #endif
    301 			(*callback)(user, &pkthdr, datap);
    302 			bp += BPF_WORDALIGN(caplen + hdrlen);
    303 			if (++n >= cnt && cnt > 0) {
    304 				p->bp = bp;
    305 				p->cc = ep - bp;
    306 				return (n);
    307 			}
    308 		} else {
    309 			/*
    310 			 * Skip this packet.
    311 			 */
    312 			bp += BPF_WORDALIGN(caplen + hdrlen);
    313 		}
    314 	}
    315 #undef bhp
    316 	p->cc = 0;
    317 	return (n);
    318 }
    319 
    320 static int
    321 pcap_inject_bpf(pcap_t *p, const void *buf, size_t size)
    322 {
    323 	int ret;
    324 
    325 	ret = write(p->fd, buf, size);
    326 #ifdef __APPLE__
    327 	if (ret == -1 && errno == EAFNOSUPPORT) {
    328 		/*
    329 		 * In Mac OS X, there's a bug wherein setting the
    330 		 * BIOCSHDRCMPLT flag causes writes to fail; see,
    331 		 * for example:
    332 		 *
    333 		 *	http://cerberus.sourcefire.com/~jeff/archives/patches/macosx/BIOCSHDRCMPLT-10.3.3.patch
    334 		 *
    335 		 * So, if, on OS X, we get EAFNOSUPPORT from the write, we
    336 		 * assume it's due to that bug, and turn off that flag
    337 		 * and try again.  If we succeed, it either means that
    338 		 * somebody applied the fix from that URL, or other patches
    339 		 * for that bug from
    340 		 *
    341 		 *	http://cerberus.sourcefire.com/~jeff/archives/patches/macosx/
    342 		 *
    343 		 * and are running a Darwin kernel with those fixes, or
    344 		 * that Apple fixed the problem in some OS X release.
    345 		 */
    346 		u_int spoof_eth_src = 0;
    347 
    348 		if (ioctl(p->fd, BIOCSHDRCMPLT, &spoof_eth_src) == -1) {
    349 			(void)snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
    350 			    "send: can't turn off BIOCSHDRCMPLT: %s",
    351 			    pcap_strerror(errno));
    352 			return (-1);
    353 		}
    354 
    355 		/*
    356 		 * Now try the write again.
    357 		 */
    358 		ret = write(p->fd, buf, size);
    359 	}
    360 #endif /* __APPLE__ */
    361 	if (ret == -1) {
    362 		snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
    363 		    pcap_strerror(errno));
    364 		return (-1);
    365 	}
    366 	return (ret);
    367 }
    368 
    369 #ifdef _AIX
    370 static int
    371 bpf_odminit(char *errbuf)
    372 {
    373 	char *errstr;
    374 
    375 	if (odm_initialize() == -1) {
    376 		if (odm_err_msg(odmerrno, &errstr) == -1)
    377 			errstr = "Unknown error";
    378 		snprintf(errbuf, PCAP_ERRBUF_SIZE,
    379 		    "bpf_load: odm_initialize failed: %s",
    380 		    errstr);
    381 		return (-1);
    382 	}
    383 
    384 	if ((odmlockid = odm_lock("/etc/objrepos/config_lock", ODM_WAIT)) == -1) {
    385 		if (odm_err_msg(odmerrno, &errstr) == -1)
    386 			errstr = "Unknown error";
    387 		snprintf(errbuf, PCAP_ERRBUF_SIZE,
    388 		    "bpf_load: odm_lock of /etc/objrepos/config_lock failed: %s",
    389 		    errstr);
    390 		return (-1);
    391 	}
    392 
    393 	return (0);
    394 }
    395 
    396 static int
    397 bpf_odmcleanup(char *errbuf)
    398 {
    399 	char *errstr;
    400 
    401 	if (odm_unlock(odmlockid) == -1) {
    402 		if (odm_err_msg(odmerrno, &errstr) == -1)
    403 			errstr = "Unknown error";
    404 		snprintf(errbuf, PCAP_ERRBUF_SIZE,
    405 		    "bpf_load: odm_unlock failed: %s",
    406 		    errstr);
    407 		return (-1);
    408 	}
    409 
    410 	if (odm_terminate() == -1) {
    411 		if (odm_err_msg(odmerrno, &errstr) == -1)
    412 			errstr = "Unknown error";
    413 		snprintf(errbuf, PCAP_ERRBUF_SIZE,
    414 		    "bpf_load: odm_terminate failed: %s",
    415 		    errstr);
    416 		return (-1);
    417 	}
    418 
    419 	return (0);
    420 }
    421 
    422 static int
    423 bpf_load(char *errbuf)
    424 {
    425 	long major;
    426 	int *minors;
    427 	int numminors, i, rc;
    428 	char buf[1024];
    429 	struct stat sbuf;
    430 	struct bpf_config cfg_bpf;
    431 	struct cfg_load cfg_ld;
    432 	struct cfg_kmod cfg_km;
    433 
    434 	/*
    435 	 * This is very very close to what happens in the real implementation
    436 	 * but I've fixed some (unlikely) bug situations.
    437 	 */
    438 	if (bpfloadedflag)
    439 		return (0);
    440 
    441 	if (bpf_odminit(errbuf) != 0)
    442 		return (-1);
    443 
    444 	major = genmajor(BPF_NAME);
    445 	if (major == -1) {
    446 		snprintf(errbuf, PCAP_ERRBUF_SIZE,
    447 		    "bpf_load: genmajor failed: %s", pcap_strerror(errno));
    448 		return (-1);
    449 	}
    450 
    451 	minors = getminor(major, &numminors, BPF_NAME);
    452 	if (!minors) {
    453 		minors = genminor("bpf", major, 0, BPF_MINORS, 1, 1);
    454 		if (!minors) {
    455 			snprintf(errbuf, PCAP_ERRBUF_SIZE,
    456 			    "bpf_load: genminor failed: %s",
    457 			    pcap_strerror(errno));
    458 			return (-1);
    459 		}
    460 	}
    461 
    462 	if (bpf_odmcleanup(errbuf))
    463 		return (-1);
    464 
    465 	rc = stat(BPF_NODE "0", &sbuf);
    466 	if (rc == -1 && errno != ENOENT) {
    467 		snprintf(errbuf, PCAP_ERRBUF_SIZE,
    468 		    "bpf_load: can't stat %s: %s",
    469 		    BPF_NODE "0", pcap_strerror(errno));
    470 		return (-1);
    471 	}
    472 
    473 	if (rc == -1 || getmajor(sbuf.st_rdev) != major) {
    474 		for (i = 0; i < BPF_MINORS; i++) {
    475 			sprintf(buf, "%s%d", BPF_NODE, i);
    476 			unlink(buf);
    477 			if (mknod(buf, S_IRUSR | S_IFCHR, domakedev(major, i)) == -1) {
    478 				snprintf(errbuf, PCAP_ERRBUF_SIZE,
    479 				    "bpf_load: can't mknod %s: %s",
    480 				    buf, pcap_strerror(errno));
    481 				return (-1);
    482 			}
    483 		}
    484 	}
    485 
    486 	/* Check if the driver is loaded */
    487 	memset(&cfg_ld, 0x0, sizeof(cfg_ld));
    488 	cfg_ld.path = buf;
    489 	sprintf(cfg_ld.path, "%s/%s", DRIVER_PATH, BPF_NAME);
    490 	if ((sysconfig(SYS_QUERYLOAD, (void *)&cfg_ld, sizeof(cfg_ld)) == -1) ||
    491 	    (cfg_ld.kmid == 0)) {
    492 		/* Driver isn't loaded, load it now */
    493 		if (sysconfig(SYS_SINGLELOAD, (void *)&cfg_ld, sizeof(cfg_ld)) == -1) {
    494 			snprintf(errbuf, PCAP_ERRBUF_SIZE,
    495 			    "bpf_load: could not load driver: %s",
    496 			    strerror(errno));
    497 			return (-1);
    498 		}
    499 	}
    500 
    501 	/* Configure the driver */
    502 	cfg_km.cmd = CFG_INIT;
    503 	cfg_km.kmid = cfg_ld.kmid;
    504 	cfg_km.mdilen = sizeof(cfg_bpf);
    505 	cfg_km.mdiptr = (void *)&cfg_bpf;
    506 	for (i = 0; i < BPF_MINORS; i++) {
    507 		cfg_bpf.devno = domakedev(major, i);
    508 		if (sysconfig(SYS_CFGKMOD, (void *)&cfg_km, sizeof(cfg_km)) == -1) {
    509 			snprintf(errbuf, PCAP_ERRBUF_SIZE,
    510 			    "bpf_load: could not configure driver: %s",
    511 			    strerror(errno));
    512 			return (-1);
    513 		}
    514 	}
    515 
    516 	bpfloadedflag = 1;
    517 
    518 	return (0);
    519 }
    520 #endif
    521 
    522 static inline int
    523 bpf_open(pcap_t *p, char *errbuf)
    524 {
    525 	int fd;
    526 #ifdef HAVE_CLONING_BPF
    527 	static const char device[] = "/dev/bpf";
    528 #else
    529 	int n = 0;
    530 	char device[sizeof "/dev/bpf0000000000"];
    531 #endif
    532 
    533 #ifdef _AIX
    534 	/*
    535 	 * Load the bpf driver, if it isn't already loaded,
    536 	 * and create the BPF device entries, if they don't
    537 	 * already exist.
    538 	 */
    539 	if (bpf_load(errbuf) == -1)
    540 		return (-1);
    541 #endif
    542 
    543 #ifdef HAVE_CLONING_BPF
    544 	if ((fd = open(device, O_RDWR)) == -1 &&
    545 	    (errno != EACCES || (fd = open(device, O_RDONLY)) == -1))
    546 		snprintf(errbuf, PCAP_ERRBUF_SIZE,
    547 		  "(cannot open device) %s: %s", device, pcap_strerror(errno));
    548 #else
    549 	/*
    550 	 * Go through all the minors and find one that isn't in use.
    551 	 */
    552 	do {
    553 		(void)snprintf(device, sizeof(device), "/dev/bpf%d", n++);
    554 		/*
    555 		 * Initially try a read/write open (to allow the inject
    556 		 * method to work).  If that fails due to permission
    557 		 * issues, fall back to read-only.  This allows a
    558 		 * non-root user to be granted specific access to pcap
    559 		 * capabilities via file permissions.
    560 		 *
    561 		 * XXX - we should have an API that has a flag that
    562 		 * controls whether to open read-only or read-write,
    563 		 * so that denial of permission to send (or inability
    564 		 * to send, if sending packets isn't supported on
    565 		 * the device in question) can be indicated at open
    566 		 * time.
    567 		 */
    568 		fd = open(device, O_RDWR);
    569 		if (fd == -1 && errno == EACCES)
    570 			fd = open(device, O_RDONLY);
    571 	} while (fd < 0 && errno == EBUSY);
    572 
    573 	/*
    574 	 * XXX better message for all minors used
    575 	 */
    576 	if (fd < 0)
    577 		snprintf(errbuf, PCAP_ERRBUF_SIZE, "(no devices found) %s: %s",
    578 		    device, pcap_strerror(errno));
    579 #endif
    580 
    581 	return (fd);
    582 }
    583 
    584 /*
    585  * We include the OS's <net/bpf.h>, not our "pcap-bpf.h", so we probably
    586  * don't get DLT_DOCSIS defined.
    587  */
    588 #ifndef DLT_DOCSIS
    589 #define DLT_DOCSIS	143
    590 #endif
    591 
    592 pcap_t *
    593 pcap_open_live(const char *device, int snaplen, int promisc, int to_ms,
    594     char *ebuf)
    595 {
    596 	int fd;
    597 	struct ifreq ifr;
    598 	struct bpf_version bv;
    599 #ifdef BIOCGDLTLIST
    600 	struct bpf_dltlist bdl;
    601 #endif
    602 #if defined(BIOCGHDRCMPLT) && defined(BIOCSHDRCMPLT)
    603 	u_int spoof_eth_src = 1;
    604 #endif
    605 	u_int v;
    606 	pcap_t *p;
    607 	struct bpf_insn total_insn;
    608 	struct bpf_program total_prog;
    609 	struct utsname osinfo;
    610 
    611 #ifdef HAVE_DAG_API
    612 	if (strstr(device, "dag")) {
    613 		return dag_open_live(device, snaplen, promisc, to_ms, ebuf);
    614 	}
    615 #endif /* HAVE_DAG_API */
    616 
    617 #ifdef BIOCGDLTLIST
    618 	memset(&bdl, 0, sizeof(bdl));
    619 #endif
    620 
    621 	p = (pcap_t *)malloc(sizeof(*p));
    622 	if (p == NULL) {
    623 		snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
    624 		    pcap_strerror(errno));
    625 		return (NULL);
    626 	}
    627 	memset(p, 0, sizeof(*p));
    628 	fd = bpf_open(p, ebuf);
    629 	if (fd < 0)
    630 		goto bad;
    631 
    632 	p->fd = fd;
    633 	p->snapshot = snaplen;
    634 
    635 	if (ioctl(fd, BIOCVERSION, (caddr_t)&bv) < 0) {
    636 		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCVERSION: %s",
    637 		    pcap_strerror(errno));
    638 		goto bad;
    639 	}
    640 	if (bv.bv_major != BPF_MAJOR_VERSION ||
    641 	    bv.bv_minor < BPF_MINOR_VERSION) {
    642 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
    643 		    "kernel bpf filter out of date");
    644 		goto bad;
    645 	}
    646 
    647 	/*
    648 	 * Try finding a good size for the buffer; 32768 may be too
    649 	 * big, so keep cutting it in half until we find a size
    650 	 * that works, or run out of sizes to try.  If the default
    651 	 * is larger, don't make it smaller.
    652 	 *
    653 	 * XXX - there should be a user-accessible hook to set the
    654 	 * initial buffer size.
    655 	 */
    656 	if ((ioctl(fd, BIOCGBLEN, (caddr_t)&v) < 0) || v < 32768)
    657 		v = 32768;
    658 	for ( ; v != 0; v >>= 1) {
    659 		/* Ignore the return value - this is because the call fails
    660 		 * on BPF systems that don't have kernel malloc.  And if
    661 		 * the call fails, it's no big deal, we just continue to
    662 		 * use the standard buffer size.
    663 		 */
    664 		(void) ioctl(fd, BIOCSBLEN, (caddr_t)&v);
    665 
    666 		(void)strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
    667 		if (ioctl(fd, BIOCSETIF, (caddr_t)&ifr) >= 0)
    668 			break;	/* that size worked; we're done */
    669 
    670 		if (errno != ENOBUFS) {
    671 			snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCSETIF: %s: %s",
    672 			    device, pcap_strerror(errno));
    673 			goto bad;
    674 		}
    675 	}
    676 
    677 	if (v == 0) {
    678 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
    679 			 "BIOCSBLEN: %s: No buffer size worked", device);
    680 		goto bad;
    681 	}
    682 
    683 	/* Get the data link layer type. */
    684 	if (ioctl(fd, BIOCGDLT, (caddr_t)&v) < 0) {
    685 		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCGDLT: %s",
    686 		    pcap_strerror(errno));
    687 		goto bad;
    688 	}
    689 #ifdef _AIX
    690 	/*
    691 	 * AIX's BPF returns IFF_ types, not DLT_ types, in BIOCGDLT.
    692 	 */
    693 	switch (v) {
    694 
    695 	case IFT_ETHER:
    696 	case IFT_ISO88023:
    697 		v = DLT_EN10MB;
    698 		break;
    699 
    700 	case IFT_FDDI:
    701 		v = DLT_FDDI;
    702 		break;
    703 
    704 	case IFT_ISO88025:
    705 		v = DLT_IEEE802;
    706 		break;
    707 
    708 	case IFT_LOOP:
    709 		v = DLT_NULL;
    710 		break;
    711 
    712 	default:
    713 		/*
    714 		 * We don't know what to map this to yet.
    715 		 */
    716 		snprintf(ebuf, PCAP_ERRBUF_SIZE, "unknown interface type %u",
    717 		    v);
    718 		goto bad;
    719 	}
    720 #endif
    721 #if _BSDI_VERSION - 0 >= 199510
    722 	/* The SLIP and PPP link layer header changed in BSD/OS 2.1 */
    723 	switch (v) {
    724 
    725 	case DLT_SLIP:
    726 		v = DLT_SLIP_BSDOS;
    727 		break;
    728 
    729 	case DLT_PPP:
    730 		v = DLT_PPP_BSDOS;
    731 		break;
    732 
    733 	case 11:	/*DLT_FR*/
    734 		v = DLT_FRELAY;
    735 		break;
    736 
    737 	case 12:	/*DLT_C_HDLC*/
    738 		v = DLT_CHDLC;
    739 		break;
    740 	}
    741 #endif
    742 #ifdef PCAP_FDDIPAD
    743 	if (v == DLT_FDDI)
    744 		p->fddipad = PCAP_FDDIPAD;
    745 	else
    746 		p->fddipad = 0;
    747 #endif
    748 	p->linktype = v;
    749 
    750 #ifdef BIOCGDLTLIST
    751 	/*
    752 	 * We know the default link type -- now determine all the DLTs
    753 	 * this interface supports.  If this fails with EINVAL, it's
    754 	 * not fatal; we just don't get to use the feature later.
    755 	 */
    756 	if (ioctl(fd, BIOCGDLTLIST, (caddr_t)&bdl) == 0) {
    757 		u_int i;
    758 		int is_ethernet;
    759 
    760 		bdl.bfl_list = (u_int *) malloc(sizeof(u_int) * (bdl.bfl_len + 1));
    761 		if (bdl.bfl_list == NULL) {
    762 			(void)snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
    763 			    pcap_strerror(errno));
    764 			goto bad;
    765 		}
    766 
    767 		if (ioctl(fd, BIOCGDLTLIST, (caddr_t)&bdl) < 0) {
    768 			(void)snprintf(ebuf, PCAP_ERRBUF_SIZE,
    769 			    "BIOCGDLTLIST: %s", pcap_strerror(errno));
    770 			free(bdl.bfl_list);
    771 			goto bad;
    772 		}
    773 
    774 		/*
    775 		 * OK, for real Ethernet devices, add DLT_DOCSIS to the
    776 		 * list, so that an application can let you choose it,
    777 		 * in case you're capturing DOCSIS traffic that a Cisco
    778 		 * Cable Modem Termination System is putting out onto
    779 		 * an Ethernet (it doesn't put an Ethernet header onto
    780 		 * the wire, it puts raw DOCSIS frames out on the wire
    781 		 * inside the low-level Ethernet framing).
    782 		 *
    783 		 * A "real Ethernet device" is defined here as a device
    784 		 * that has a link-layer type of DLT_EN10MB and that has
    785 		 * no alternate link-layer types; that's done to exclude
    786 		 * 802.11 interfaces (which might or might not be the
    787 		 * right thing to do, but I suspect it is - Ethernet <->
    788 		 * 802.11 bridges would probably badly mishandle frames
    789 		 * that don't have Ethernet headers).
    790 		 */
    791 		if (p->linktype == DLT_EN10MB) {
    792 			is_ethernet = 1;
    793 			for (i = 0; i < bdl.bfl_len; i++) {
    794 				if (bdl.bfl_list[i] != DLT_EN10MB) {
    795 					is_ethernet = 0;
    796 					break;
    797 				}
    798 			}
    799 			if (is_ethernet) {
    800 				/*
    801 				 * We reserved one more slot at the end of
    802 				 * the list.
    803 				 */
    804 				bdl.bfl_list[bdl.bfl_len] = DLT_DOCSIS;
    805 				bdl.bfl_len++;
    806 			}
    807 		}
    808 		p->dlt_count = bdl.bfl_len;
    809 		p->dlt_list = bdl.bfl_list;
    810 	} else {
    811 		if (errno != EINVAL) {
    812 			(void)snprintf(ebuf, PCAP_ERRBUF_SIZE,
    813 			    "BIOCGDLTLIST: %s", pcap_strerror(errno));
    814 			goto bad;
    815 		}
    816 	}
    817 #endif
    818 
    819 	/*
    820 	 * If this is an Ethernet device, and we don't have a DLT_ list,
    821 	 * give it a list with DLT_EN10MB and DLT_DOCSIS.  (That'd give
    822 	 * 802.11 interfaces DLT_DOCSIS, which isn't the right thing to
    823 	 * do, but there's not much we can do about that without finding
    824 	 * some other way of determining whether it's an Ethernet or 802.11
    825 	 * device.)
    826 	 */
    827 	if (p->linktype == DLT_EN10MB && p->dlt_count == 0) {
    828 		p->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
    829 		/*
    830 		 * If that fails, just leave the list empty.
    831 		 */
    832 		if (p->dlt_list != NULL) {
    833 			p->dlt_list[0] = DLT_EN10MB;
    834 			p->dlt_list[1] = DLT_DOCSIS;
    835 			p->dlt_count = 2;
    836 		}
    837 	}
    838 
    839 #if defined(BIOCGHDRCMPLT) && defined(BIOCSHDRCMPLT)
    840 	/*
    841 	 * Do a BIOCSHDRCMPLT, if defined, to turn that flag on, so
    842 	 * the link-layer source address isn't forcibly overwritten.
    843 	 * (Should we ignore errors?  Should we do this only if
    844 	 * we're open for writing?)
    845 	 *
    846 	 * XXX - I seem to remember some packet-sending bug in some
    847 	 * BSDs - check CVS log for "bpf.c"?
    848 	 */
    849 	if (ioctl(fd, BIOCSHDRCMPLT, &spoof_eth_src) == -1) {
    850 		(void)snprintf(ebuf, PCAP_ERRBUF_SIZE,
    851 		    "BIOCSHDRCMPLT: %s", pcap_strerror(errno));
    852 		goto bad;
    853 	}
    854 #endif
    855 	/* set timeout */
    856 	if (to_ms != 0) {
    857 		/*
    858 		 * XXX - is this seconds/nanoseconds in AIX?
    859 		 * (Treating it as such doesn't fix the timeout
    860 		 * problem described below.)
    861 		 */
    862 		struct timeval to;
    863 		to.tv_sec = to_ms / 1000;
    864 		to.tv_usec = (to_ms * 1000) % 1000000;
    865 		if (ioctl(p->fd, BIOCSRTIMEOUT, (caddr_t)&to) < 0) {
    866 			snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCSRTIMEOUT: %s",
    867 			    pcap_strerror(errno));
    868 			goto bad;
    869 		}
    870 	}
    871 
    872 #ifdef _AIX
    873 #ifdef	BIOCIMMEDIATE
    874 	/*
    875 	 * Darren Reed notes that
    876 	 *
    877 	 *	On AIX (4.2 at least), if BIOCIMMEDIATE is not set, the
    878 	 *	timeout appears to be ignored and it waits until the buffer
    879 	 *	is filled before returning.  The result of not having it
    880 	 *	set is almost worse than useless if your BPF filter
    881 	 *	is reducing things to only a few packets (i.e. one every
    882 	 *	second or so).
    883 	 *
    884 	 * so we turn BIOCIMMEDIATE mode on if this is AIX.
    885 	 *
    886 	 * We don't turn it on for other platforms, as that means we
    887 	 * get woken up for every packet, which may not be what we want;
    888 	 * in the Winter 1993 USENIX paper on BPF, they say:
    889 	 *
    890 	 *	Since a process might want to look at every packet on a
    891 	 *	network and the time between packets can be only a few
    892 	 *	microseconds, it is not possible to do a read system call
    893 	 *	per packet and BPF must collect the data from several
    894 	 *	packets and return it as a unit when the monitoring
    895 	 *	application does a read.
    896 	 *
    897 	 * which I infer is the reason for the timeout - it means we
    898 	 * wait that amount of time, in the hopes that more packets
    899 	 * will arrive and we'll get them all with one read.
    900 	 *
    901 	 * Setting BIOCIMMEDIATE mode on FreeBSD (and probably other
    902 	 * BSDs) causes the timeout to be ignored.
    903 	 *
    904 	 * On the other hand, some platforms (e.g., Linux) don't support
    905 	 * timeouts, they just hand stuff to you as soon as it arrives;
    906 	 * if that doesn't cause a problem on those platforms, it may
    907 	 * be OK to have BIOCIMMEDIATE mode on BSD as well.
    908 	 *
    909 	 * (Note, though, that applications may depend on the read
    910 	 * completing, even if no packets have arrived, when the timeout
    911 	 * expires, e.g. GUI applications that have to check for input
    912 	 * while waiting for packets to arrive; a non-zero timeout
    913 	 * prevents "select()" from working right on FreeBSD and
    914 	 * possibly other BSDs, as the timer doesn't start until a
    915 	 * "read()" is done, so the timer isn't in effect if the
    916 	 * application is blocked on a "select()", and the "select()"
    917 	 * doesn't get woken up for a BPF device until the buffer
    918 	 * fills up.)
    919 	 */
    920 	v = 1;
    921 	if (ioctl(p->fd, BIOCIMMEDIATE, &v) < 0) {
    922 		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCIMMEDIATE: %s",
    923 		    pcap_strerror(errno));
    924 		goto bad;
    925 	}
    926 #endif	/* BIOCIMMEDIATE */
    927 #endif	/* _AIX */
    928 
    929 	if (promisc) {
    930 		/* set promiscuous mode, okay if it fails */
    931 		if (ioctl(p->fd, BIOCPROMISC, NULL) < 0) {
    932 			snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCPROMISC: %s",
    933 			    pcap_strerror(errno));
    934 		}
    935 	}
    936 
    937 	if (ioctl(fd, BIOCGBLEN, (caddr_t)&v) < 0) {
    938 		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCGBLEN: %s",
    939 		    pcap_strerror(errno));
    940 		goto bad;
    941 	}
    942 	p->bufsize = v;
    943 	p->buffer = (u_char *)malloc(p->bufsize);
    944 	if (p->buffer == NULL) {
    945 		snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
    946 		    pcap_strerror(errno));
    947 		goto bad;
    948 	}
    949 #ifdef _AIX
    950 	/* For some strange reason this seems to prevent the EFAULT
    951 	 * problems we have experienced from AIX BPF. */
    952 	memset(p->buffer, 0x0, p->bufsize);
    953 #endif
    954 
    955 	/*
    956 	 * If there's no filter program installed, there's
    957 	 * no indication to the kernel of what the snapshot
    958 	 * length should be, so no snapshotting is done.
    959 	 *
    960 	 * Therefore, when we open the device, we install
    961 	 * an "accept everything" filter with the specified
    962 	 * snapshot length.
    963 	 */
    964 	total_insn.code = (u_short)(BPF_RET | BPF_K);
    965 	total_insn.jt = 0;
    966 	total_insn.jf = 0;
    967 	total_insn.k = snaplen;
    968 
    969 	total_prog.bf_len = 1;
    970 	total_prog.bf_insns = &total_insn;
    971 	if (ioctl(p->fd, BIOCSETF, (caddr_t)&total_prog) < 0) {
    972 		snprintf(ebuf, PCAP_ERRBUF_SIZE, "BIOCSETF: %s",
    973 		    pcap_strerror(errno));
    974 		goto bad;
    975 	}
    976 
    977 	/*
    978 	 * On most BPF platforms, either you can do a "select()" or
    979 	 * "poll()" on a BPF file descriptor and it works correctly,
    980 	 * or you can do it and it will return "readable" if the
    981 	 * hold buffer is full but not if the timeout expires *and*
    982 	 * a non-blocking read will, if the hold buffer is empty
    983 	 * but the store buffer isn't empty, rotate the buffers
    984 	 * and return what packets are available.
    985 	 *
    986 	 * In the latter case, the fact that a non-blocking read
    987 	 * will give you the available packets means you can work
    988 	 * around the failure of "select()" and "poll()" to wake up
    989 	 * and return "readable" when the timeout expires by using
    990 	 * the timeout as the "select()" or "poll()" timeout, putting
    991 	 * the BPF descriptor into non-blocking mode, and read from
    992 	 * it regardless of whether "select()" reports it as readable
    993 	 * or not.
    994 	 *
    995 	 * However, in FreeBSD 4.3 and 4.4, "select()" and "poll()"
    996 	 * won't wake up and return "readable" if the timer expires
    997 	 * and non-blocking reads return EWOULDBLOCK if the hold
    998 	 * buffer is empty, even if the store buffer is non-empty.
    999 	 *
   1000 	 * This means the workaround in question won't work.
   1001 	 *
   1002 	 * Therefore, on FreeBSD 4.3 and 4.4, we set "p->selectable_fd"
   1003 	 * to -1, which means "sorry, you can't use 'select()' or 'poll()'
   1004 	 * here".  On all other BPF platforms, we set it to the FD for
   1005 	 * the BPF device; in NetBSD, OpenBSD, and Darwin, a non-blocking
   1006 	 * read will, if the hold buffer is empty and the store buffer
   1007 	 * isn't empty, rotate the buffers and return what packets are
   1008 	 * there (and in sufficiently recent versions of OpenBSD
   1009 	 * "select()" and "poll()" should work correctly).
   1010 	 *
   1011 	 * XXX - what about AIX?
   1012 	 */
   1013 	p->selectable_fd = p->fd;	/* assume select() works until we know otherwise */
   1014 	if (uname(&osinfo) == 0) {
   1015 		/*
   1016 		 * We can check what OS this is.
   1017 		 */
   1018 		if (strcmp(osinfo.sysname, "FreeBSD") == 0) {
   1019 			if (strncmp(osinfo.release, "4.3-", 4) == 0 ||
   1020 			     strncmp(osinfo.release, "4.4-", 4) == 0)
   1021 				p->selectable_fd = -1;
   1022 		}
   1023 	}
   1024 
   1025 	p->read_op = pcap_read_bpf;
   1026 	p->inject_op = pcap_inject_bpf;
   1027 	p->setfilter_op = pcap_setfilter_bpf;
   1028 	p->setdirection_op = pcap_setdirection_bpf;
   1029 	p->set_datalink_op = pcap_set_datalink_bpf;
   1030 	p->getnonblock_op = pcap_getnonblock_fd;
   1031 	p->setnonblock_op = pcap_setnonblock_fd;
   1032 	p->stats_op = pcap_stats_bpf;
   1033 	p->close_op = pcap_close_common;
   1034 
   1035 	return (p);
   1036  bad:
   1037 	(void)close(fd);
   1038 	if (p->dlt_list != NULL)
   1039 		free(p->dlt_list);
   1040 	free(p);
   1041 	return (NULL);
   1042 }
   1043 
   1044 int
   1045 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
   1046 {
   1047 #ifdef HAVE_DAG_API
   1048 	if (dag_platform_finddevs(alldevsp, errbuf) < 0)
   1049 		return (-1);
   1050 #endif /* HAVE_DAG_API */
   1051 
   1052 	return (0);
   1053 }
   1054 
   1055 static int
   1056 pcap_setfilter_bpf(pcap_t *p, struct bpf_program *fp)
   1057 {
   1058 	/*
   1059 	 * It looks that BPF code generated by gen_protochain() is not
   1060 	 * compatible with some of kernel BPF code (for example BSD/OS 3.1).
   1061 	 * Take a safer side for now.
   1062 	 */
   1063 	if (no_optimize) {
   1064 		/*
   1065 		 * XXX - what if we already have a filter in the kernel?
   1066 		 */
   1067 		if (install_bpf_program(p, fp) < 0)
   1068 			return (-1);
   1069 		p->md.use_bpf = 0;	/* filtering in userland */
   1070 		return (0);
   1071 	}
   1072 
   1073 	/*
   1074 	 * Free any user-mode filter we might happen to have installed.
   1075 	 */
   1076 	pcap_freecode(&p->fcode);
   1077 
   1078 	/*
   1079 	 * Try to install the kernel filter.
   1080 	 */
   1081 	if (ioctl(p->fd, BIOCSETF, (caddr_t)fp) < 0) {
   1082 		snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "BIOCSETF: %s",
   1083 		    pcap_strerror(errno));
   1084 		return (-1);
   1085 	}
   1086 	p->md.use_bpf = 1;	/* filtering in the kernel */
   1087 
   1088 	/*
   1089 	 * Discard any previously-received packets, as they might have
   1090 	 * passed whatever filter was formerly in effect, but might
   1091 	 * not pass this filter (BIOCSETF discards packets buffered
   1092 	 * in the kernel, so you can lose packets in any case).
   1093 	 */
   1094 	p->cc = 0;
   1095 	return (0);
   1096 }
   1097 
   1098 /*
   1099  * Set direction flag: Which packets do we accept on a forwarding
   1100  * single device? IN, OUT or both?
   1101  */
   1102 static int
   1103 pcap_setdirection_bpf(pcap_t *p, pcap_direction_t d)
   1104 {
   1105 #if defined(BIOCSDIRECTION)
   1106 	u_int direction;
   1107 
   1108 	direction = (d == PCAP_D_IN) ? BPF_D_IN :
   1109 	    ((d == PCAP_D_OUT) ? BPF_D_OUT : BPF_D_INOUT);
   1110 	if (ioctl(p->fd, BIOCSDIRECTION, &direction) == -1) {
   1111 		(void) snprintf(p->errbuf, sizeof(p->errbuf),
   1112 		    "Cannot set direction to %s: %s",
   1113 		        (d == PCAP_D_IN) ? "PCAP_D_IN" :
   1114 			((d == PCAP_D_OUT) ? "PCAP_D_OUT" : "PCAP_D_INOUT"),
   1115 			strerror(errno));
   1116 		return (-1);
   1117 	}
   1118 	return (0);
   1119 #elif defined(BIOCSSEESENT)
   1120 	u_int seesent;
   1121 
   1122 	/*
   1123 	 * We don't support PCAP_D_OUT.
   1124 	 */
   1125 	if (d == PCAP_D_OUT) {
   1126 		snprintf(p->errbuf, sizeof(p->errbuf),
   1127 		    "Setting direction to PCAP_D_OUT is not supported on BPF");
   1128 		return -1;
   1129 	}
   1130 
   1131 	seesent = (d == PCAP_D_INOUT);
   1132 	if (ioctl(p->fd, BIOCSSEESENT, &seesent) == -1) {
   1133 		(void) snprintf(p->errbuf, sizeof(p->errbuf),
   1134 		    "Cannot set direction to %s: %s",
   1135 		        (d == PCAP_D_INOUT) ? "PCAP_D_INOUT" : "PCAP_D_IN",
   1136 			strerror(errno));
   1137 		return (-1);
   1138 	}
   1139 	return (0);
   1140 #else
   1141 	(void) snprintf(p->errbuf, sizeof(p->errbuf),
   1142 	    "This system doesn't support BIOCSSEESENT, so the direction can't be set");
   1143 	return (-1);
   1144 #endif
   1145 }
   1146 
   1147 static int
   1148 pcap_set_datalink_bpf(pcap_t *p, int dlt)
   1149 {
   1150 #ifdef BIOCSDLT
   1151 	if (ioctl(p->fd, BIOCSDLT, &dlt) == -1) {
   1152 		(void) snprintf(p->errbuf, sizeof(p->errbuf),
   1153 		    "Cannot set DLT %d: %s", dlt, strerror(errno));
   1154 		return (-1);
   1155 	}
   1156 #endif
   1157 	return (0);
   1158 }
   1159