xref: /netbsd-src/external/bsd/libpcap/dist/pcap-linux.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: pcap-linux.c,v 1.1.1.4 2013/12/31 16:57:20 christos Exp $	*/
2 
3 /*
4  *  pcap-linux.c: Packet capture interface to the Linux kernel
5  *
6  *  Copyright (c) 2000 Torsten Landschoff <torsten@debian.org>
7  *  		       Sebastian Krahmer  <krahmer@cs.uni-potsdam.de>
8  *
9  *  License: BSD
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in
19  *     the documentation and/or other materials provided with the
20  *     distribution.
21  *  3. The names of the authors may not be used to endorse or promote
22  *     products derived from this software without specific prior
23  *     written permission.
24  *
25  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
26  *  IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
27  *  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
28  *
29  *  Modifications:     Added PACKET_MMAP support
30  *                     Paolo Abeni <paolo.abeni@email.it>
31  *                     Added TPACKET_V3 support
32  *                     Gabor Tatarka <gabor.tatarka@ericsson.com>
33  *
34  *                     based on previous works of:
35  *                     Simon Patarin <patarin@cs.unibo.it>
36  *                     Phil Wood <cpw@lanl.gov>
37  *
38  * Monitor-mode support for mac80211 includes code taken from the iw
39  * command; the copyright notice for that code is
40  *
41  * Copyright (c) 2007, 2008	Johannes Berg
42  * Copyright (c) 2007		Andy Lutomirski
43  * Copyright (c) 2007		Mike Kershaw
44  * Copyright (c) 2008		Gábor Stefanik
45  *
46  * All rights reserved.
47  *
48  * Redistribution and use in source and binary forms, with or without
49  * modification, are permitted provided that the following conditions
50  * are met:
51  * 1. Redistributions of source code must retain the above copyright
52  *    notice, this list of conditions and the following disclaimer.
53  * 2. Redistributions in binary form must reproduce the above copyright
54  *    notice, this list of conditions and the following disclaimer in the
55  *    documentation and/or other materials provided with the distribution.
56  * 3. The name of the author may not be used to endorse or promote products
57  *    derived from this software without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
60  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
61  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
62  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
63  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
64  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
65  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
66  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
67  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  */
71 
72 #ifndef lint
73 static const char rcsid[] _U_ =
74     "@(#) Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.164 2008-12-14 22:00:57 guy Exp  (LBL)";
75 #endif
76 
77 /*
78  * Known problems with 2.0[.x] kernels:
79  *
80  *   - The loopback device gives every packet twice; on 2.2[.x] kernels,
81  *     if we use PF_PACKET, we can filter out the transmitted version
82  *     of the packet by using data in the "sockaddr_ll" returned by
83  *     "recvfrom()", but, on 2.0[.x] kernels, we have to use
84  *     PF_INET/SOCK_PACKET, which means "recvfrom()" supplies a
85  *     "sockaddr_pkt" which doesn't give us enough information to let
86  *     us do that.
87  *
88  *   - We have to set the interface's IFF_PROMISC flag ourselves, if
89  *     we're to run in promiscuous mode, which means we have to turn
90  *     it off ourselves when we're done; the kernel doesn't keep track
91  *     of how many sockets are listening promiscuously, which means
92  *     it won't get turned off automatically when no sockets are
93  *     listening promiscuously.  We catch "pcap_close()" and, for
94  *     interfaces we put into promiscuous mode, take them out of
95  *     promiscuous mode - which isn't necessarily the right thing to
96  *     do, if another socket also requested promiscuous mode between
97  *     the time when we opened the socket and the time when we close
98  *     the socket.
99  *
100  *   - MSG_TRUNC isn't supported, so you can't specify that "recvfrom()"
101  *     return the amount of data that you could have read, rather than
102  *     the amount that was returned, so we can't just allocate a buffer
103  *     whose size is the snapshot length and pass the snapshot length
104  *     as the byte count, and also pass MSG_TRUNC, so that the return
105  *     value tells us how long the packet was on the wire.
106  *
107  *     This means that, if we want to get the actual size of the packet,
108  *     so we can return it in the "len" field of the packet header,
109  *     we have to read the entire packet, not just the part that fits
110  *     within the snapshot length, and thus waste CPU time copying data
111  *     from the kernel that our caller won't see.
112  *
113  *     We have to get the actual size, and supply it in "len", because
114  *     otherwise, the IP dissector in tcpdump, for example, will complain
115  *     about "truncated-ip", as the packet will appear to have been
116  *     shorter, on the wire, than the IP header said it should have been.
117  */
118 
119 
120 #define _GNU_SOURCE
121 
122 #ifdef HAVE_CONFIG_H
123 #include "config.h"
124 #endif
125 
126 #include <errno.h>
127 #include <stdio.h>
128 #include <stdlib.h>
129 #include <ctype.h>
130 #include <unistd.h>
131 #include <fcntl.h>
132 #include <string.h>
133 #include <limits.h>
134 #include <sys/stat.h>
135 #include <sys/socket.h>
136 #include <sys/ioctl.h>
137 #include <sys/utsname.h>
138 #include <sys/mman.h>
139 #include <linux/if.h>
140 #include <linux/if_packet.h>
141 #include <linux/sockios.h>
142 #include <netinet/in.h>
143 #include <linux/if_ether.h>
144 #include <net/if_arp.h>
145 #include <poll.h>
146 #include <dirent.h>
147 
148 #include "pcap-int.h"
149 #include "pcap/sll.h"
150 #include "pcap/vlan.h"
151 
152 /*
153  * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
154  * sockets rather than SOCK_PACKET sockets.
155  *
156  * To use them, we include <linux/if_packet.h> rather than
157  * <netpacket/packet.h>; we do so because
158  *
159  *	some Linux distributions (e.g., Slackware 4.0) have 2.2 or
160  *	later kernels and libc5, and don't provide a <netpacket/packet.h>
161  *	file;
162  *
163  *	not all versions of glibc2 have a <netpacket/packet.h> file
164  *	that defines stuff needed for some of the 2.4-or-later-kernel
165  *	features, so if the system has a 2.4 or later kernel, we
166  *	still can't use those features.
167  *
168  * We're already including a number of other <linux/XXX.h> headers, and
169  * this code is Linux-specific (no other OS has PF_PACKET sockets as
170  * a raw packet capture mechanism), so it's not as if you gain any
171  * useful portability by using <netpacket/packet.h>
172  *
173  * XXX - should we just include <linux/if_packet.h> even if PF_PACKET
174  * isn't defined?  It only defines one data structure in 2.0.x, so
175  * it shouldn't cause any problems.
176  */
177 #ifdef PF_PACKET
178 # include <linux/if_packet.h>
179 
180  /*
181   * On at least some Linux distributions (for example, Red Hat 5.2),
182   * there's no <netpacket/packet.h> file, but PF_PACKET is defined if
183   * you include <sys/socket.h>, but <linux/if_packet.h> doesn't define
184   * any of the PF_PACKET stuff such as "struct sockaddr_ll" or any of
185   * the PACKET_xxx stuff.
186   *
187   * So we check whether PACKET_HOST is defined, and assume that we have
188   * PF_PACKET sockets only if it is defined.
189   */
190 # ifdef PACKET_HOST
191 #  define HAVE_PF_PACKET_SOCKETS
192 #  ifdef PACKET_AUXDATA
193 #   define HAVE_PACKET_AUXDATA
194 #  endif /* PACKET_AUXDATA */
195 # endif /* PACKET_HOST */
196 
197 
198  /* check for memory mapped access avaibility. We assume every needed
199   * struct is defined if the macro TPACKET_HDRLEN is defined, because it
200   * uses many ring related structs and macros */
201 # ifdef TPACKET_HDRLEN
202 #  define HAVE_PACKET_RING
203 #  ifdef TPACKET3_HDRLEN
204 #   define HAVE_TPACKET3
205 #  endif /* TPACKET3_HDRLEN */
206 #  ifdef TPACKET2_HDRLEN
207 #   define HAVE_TPACKET2
208 #  else  /* TPACKET2_HDRLEN */
209 #   define TPACKET_V1	0    /* Old kernel with only V1, so no TPACKET_Vn defined */
210 #  endif /* TPACKET2_HDRLEN */
211 # endif /* TPACKET_HDRLEN */
212 #endif /* PF_PACKET */
213 
214 #ifdef SO_ATTACH_FILTER
215 #include <linux/types.h>
216 #include <linux/filter.h>
217 #endif
218 
219 #ifdef HAVE_LINUX_NET_TSTAMP_H
220 #include <linux/net_tstamp.h>
221 #endif
222 
223 /*
224  * Got Wireless Extensions?
225  */
226 #ifdef HAVE_LINUX_WIRELESS_H
227 #include <linux/wireless.h>
228 #endif /* HAVE_LINUX_WIRELESS_H */
229 
230 /*
231  * Got libnl?
232  */
233 #ifdef HAVE_LIBNL
234 #include <linux/nl80211.h>
235 
236 #include <netlink/genl/genl.h>
237 #include <netlink/genl/family.h>
238 #include <netlink/genl/ctrl.h>
239 #include <netlink/msg.h>
240 #include <netlink/attr.h>
241 #endif /* HAVE_LIBNL */
242 
243 /*
244  * Got ethtool support?
245  */
246 #ifdef HAVE_LINUX_ETHTOOL_H
247 #include <linux/ethtool.h>
248 #endif
249 
250 #ifndef HAVE_SOCKLEN_T
251 typedef int		socklen_t;
252 #endif
253 
254 #ifndef MSG_TRUNC
255 /*
256  * This is being compiled on a system that lacks MSG_TRUNC; define it
257  * with the value it has in the 2.2 and later kernels, so that, on
258  * those kernels, when we pass it in the flags argument to "recvfrom()"
259  * we're passing the right value and thus get the MSG_TRUNC behavior
260  * we want.  (We don't get that behavior on 2.0[.x] kernels, because
261  * they didn't support MSG_TRUNC.)
262  */
263 #define MSG_TRUNC	0x20
264 #endif
265 
266 #ifndef SOL_PACKET
267 /*
268  * This is being compiled on a system that lacks SOL_PACKET; define it
269  * with the value it has in the 2.2 and later kernels, so that we can
270  * set promiscuous mode in the good modern way rather than the old
271  * 2.0-kernel crappy way.
272  */
273 #define SOL_PACKET	263
274 #endif
275 
276 #define MAX_LINKHEADER_SIZE	256
277 
278 /*
279  * When capturing on all interfaces we use this as the buffer size.
280  * Should be bigger then all MTUs that occur in real life.
281  * 64kB should be enough for now.
282  */
283 #define BIGGER_THAN_ALL_MTUS	(64*1024)
284 
285 /*
286  * Private data for capturing on Linux SOCK_PACKET or PF_PACKET sockets.
287  */
288 struct pcap_linux {
289 	u_int	packets_read;	/* count of packets read with recvfrom() */
290 	long	proc_dropped;	/* packets reported dropped by /proc/net/dev */
291 	struct pcap_stat stat;
292 
293 	char	*device;	/* device name */
294 	int	filtering_in_kernel; /* using kernel filter */
295 	int	must_do_on_close; /* stuff we must do when we close */
296 	int	timeout;	/* timeout for buffering */
297 	int	sock_packet;	/* using Linux 2.0 compatible interface */
298 	int	cooked;		/* using SOCK_DGRAM rather than SOCK_RAW */
299 	int	ifindex;	/* interface index of device we're bound to */
300 	int	lo_ifindex;	/* interface index of the loopback device */
301 	bpf_u_int32 oldmode;	/* mode to restore when turning monitor mode off */
302 	char	*mondevice;	/* mac80211 monitor device we created */
303 	u_char	*mmapbuf;	/* memory-mapped region pointer */
304 	size_t	mmapbuflen;	/* size of region */
305 	int	vlan_offset;	/* offset at which to insert vlan tags; if -1, don't insert */
306 	u_int	tp_version;	/* version of tpacket_hdr for mmaped ring */
307 	u_int	tp_hdrlen;	/* hdrlen of tpacket_hdr for mmaped ring */
308 	u_char	*oneshot_buffer; /* buffer for copy of packet */
309 #ifdef HAVE_TPACKET3
310 	unsigned char *current_packet; /* Current packet within the TPACKET_V3 block. Move to next block if NULL. */
311 	int packets_left; /* Unhandled packets left within the block from previous call to pcap_read_linux_mmap_v3 in case of TPACKET_V3. */
312 #endif
313 };
314 
315 /*
316  * Stuff to do when we close.
317  */
318 #define MUST_CLEAR_PROMISC	0x00000001	/* clear promiscuous mode */
319 #define MUST_CLEAR_RFMON	0x00000002	/* clear rfmon (monitor) mode */
320 #define MUST_DELETE_MONIF	0x00000004	/* delete monitor-mode interface */
321 
322 /*
323  * Prototypes for internal functions and methods.
324  */
325 static void map_arphrd_to_dlt(pcap_t *, int, int);
326 #ifdef HAVE_PF_PACKET_SOCKETS
327 static short int map_packet_type_to_sll_type(short int);
328 #endif
329 static int pcap_activate_linux(pcap_t *);
330 static int activate_old(pcap_t *);
331 static int activate_new(pcap_t *);
332 static int activate_mmap(pcap_t *, int *);
333 static int pcap_can_set_rfmon_linux(pcap_t *);
334 static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *);
335 static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);
336 static int pcap_inject_linux(pcap_t *, const void *, size_t);
337 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
338 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
339 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
340 static int pcap_set_datalink_linux(pcap_t *, int);
341 static void pcap_cleanup_linux(pcap_t *);
342 
343 union thdr {
344 	struct tpacket_hdr		*h1;
345 #ifdef HAVE_TPACKET2
346 	struct tpacket2_hdr		*h2;
347 #endif
348 #ifdef HAVE_TPACKET3
349 	struct tpacket_block_desc	*h3;
350 #endif
351 	void				*raw;
352 };
353 
354 #ifdef HAVE_PACKET_RING
355 #define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset])
356 
357 static void destroy_ring(pcap_t *handle);
358 static int create_ring(pcap_t *handle, int *status);
359 static int prepare_tpacket_socket(pcap_t *handle);
360 static void pcap_cleanup_linux_mmap(pcap_t *);
361 static int pcap_read_linux_mmap_v1(pcap_t *, int, pcap_handler , u_char *);
362 #ifdef HAVE_TPACKET2
363 static int pcap_read_linux_mmap_v2(pcap_t *, int, pcap_handler , u_char *);
364 #endif
365 #ifdef HAVE_TPACKET3
366 static int pcap_read_linux_mmap_v3(pcap_t *, int, pcap_handler , u_char *);
367 #endif
368 static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *);
369 static int pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf);
370 static int pcap_getnonblock_mmap(pcap_t *p, char *errbuf);
371 static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
372     const u_char *bytes);
373 #endif
374 
375 /*
376  * Wrap some ioctl calls
377  */
378 #ifdef HAVE_PF_PACKET_SOCKETS
379 static int	iface_get_id(int fd, const char *device, char *ebuf);
380 #endif /* HAVE_PF_PACKET_SOCKETS */
381 static int	iface_get_mtu(int fd, const char *device, char *ebuf);
382 static int 	iface_get_arptype(int fd, const char *device, char *ebuf);
383 #ifdef HAVE_PF_PACKET_SOCKETS
384 static int 	iface_bind(int fd, int ifindex, char *ebuf);
385 #ifdef IW_MODE_MONITOR
386 static int	has_wext(int sock_fd, const char *device, char *ebuf);
387 #endif /* IW_MODE_MONITOR */
388 static int	enter_rfmon_mode(pcap_t *handle, int sock_fd,
389     const char *device);
390 #endif /* HAVE_PF_PACKET_SOCKETS */
391 static int	iface_get_offload(pcap_t *handle);
392 static int 	iface_bind_old(int fd, const char *device, char *ebuf);
393 
394 #ifdef SO_ATTACH_FILTER
395 static int	fix_program(pcap_t *handle, struct sock_fprog *fcode,
396     int is_mapped);
397 static int	fix_offset(struct bpf_insn *p);
398 static int	set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
399 static int	reset_kernel_filter(pcap_t *handle);
400 
401 static struct sock_filter	total_insn
402 	= BPF_STMT(BPF_RET | BPF_K, 0);
403 static struct sock_fprog	total_fcode
404 	= { 1, &total_insn };
405 #endif /* SO_ATTACH_FILTER */
406 
407 pcap_t *
408 pcap_create_interface(const char *device, char *ebuf)
409 {
410 	pcap_t *handle;
411 
412 	handle = pcap_create_common(device, ebuf, sizeof (struct pcap_linux));
413 	if (handle == NULL)
414 		return NULL;
415 
416 	handle->activate_op = pcap_activate_linux;
417 	handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
418 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
419 	/*
420 	 * We claim that we support:
421 	 *
422 	 *	software time stamps, with no details about their precision;
423 	 *	hardware time stamps, synced to the host time;
424 	 *	hardware time stamps, not synced to the host time.
425 	 *
426 	 * XXX - we can't ask a device whether it supports
427 	 * hardware time stamps, so we just claim all devices do.
428 	 */
429 	handle->tstamp_type_count = 3;
430 	handle->tstamp_type_list = malloc(3 * sizeof(u_int));
431 	if (handle->tstamp_type_list == NULL) {
432 		free(handle);
433 		return NULL;
434 	}
435 	handle->tstamp_type_list[0] = PCAP_TSTAMP_HOST;
436 	handle->tstamp_type_list[1] = PCAP_TSTAMP_ADAPTER;
437 	handle->tstamp_type_list[2] = PCAP_TSTAMP_ADAPTER_UNSYNCED;
438 #endif
439 
440 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
441 	/*
442 	 * We claim that we support microsecond and nanosecond time
443 	 * stamps.
444 	 *
445 	 * XXX - with adapter-supplied time stamps, can we choose
446 	 * microsecond or nanosecond time stamps on arbitrary
447 	 * adapters?
448 	 */
449 	handle->tstamp_precision_count = 2;
450 	handle->tstamp_precision_list = malloc(2 * sizeof(u_int));
451 	if (handle->tstamp_precision_list == NULL) {
452 		if (handle->tstamp_type_list != NULL)
453 			free(handle->tstamp_type_list);
454 		free(handle);
455 		return NULL;
456 	}
457 	handle->tstamp_precision_list[0] = PCAP_TSTAMP_PRECISION_MICRO;
458 	handle->tstamp_precision_list[1] = PCAP_TSTAMP_PRECISION_NANO;
459 #endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */
460 
461 	return handle;
462 }
463 
464 #ifdef HAVE_LIBNL
465 /*
466  * If interface {if} is a mac80211 driver, the file
467  * /sys/class/net/{if}/phy80211 is a symlink to
468  * /sys/class/ieee80211/{phydev}, for some {phydev}.
469  *
470  * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
471  * least, has a "wmaster0" device and a "wlan0" device; the
472  * latter is the one with the IP address.  Both show up in
473  * "tcpdump -D" output.  Capturing on the wmaster0 device
474  * captures with 802.11 headers.
475  *
476  * airmon-ng searches through /sys/class/net for devices named
477  * monN, starting with mon0; as soon as one *doesn't* exist,
478  * it chooses that as the monitor device name.  If the "iw"
479  * command exists, it does "iw dev {if} interface add {monif}
480  * type monitor", where {monif} is the monitor device.  It
481  * then (sigh) sleeps .1 second, and then configures the
482  * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
483  * is a file, it writes {mondev}, without a newline, to that file,
484  * and again (sigh) sleeps .1 second, and then iwconfig's that
485  * device into monitor mode and configures it up.  Otherwise,
486  * you can't do monitor mode.
487  *
488  * All these devices are "glued" together by having the
489  * /sys/class/net/{device}/phy80211 links pointing to the same
490  * place, so, given a wmaster, wlan, or mon device, you can
491  * find the other devices by looking for devices with
492  * the same phy80211 link.
493  *
494  * To turn monitor mode off, delete the monitor interface,
495  * either with "iw dev {monif} interface del" or by sending
496  * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
497  *
498  * Note: if you try to create a monitor device named "monN", and
499  * there's already a "monN" device, it fails, as least with
500  * the netlink interface (which is what iw uses), with a return
501  * value of -ENFILE.  (Return values are negative errnos.)  We
502  * could probably use that to find an unused device.
503  *
504  * Yes, you can have multiple monitor devices for a given
505  * physical device.
506 */
507 
508 /*
509  * Is this a mac80211 device?  If so, fill in the physical device path and
510  * return 1; if not, return 0.  On an error, fill in handle->errbuf and
511  * return PCAP_ERROR.
512  */
513 static int
514 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
515     size_t phydev_max_pathlen)
516 {
517 	char *pathstr;
518 	ssize_t bytes_read;
519 
520 	/*
521 	 * Generate the path string for the symlink to the physical device.
522 	 */
523 	if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
524 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
525 		    "%s: Can't generate path name string for /sys/class/net device",
526 		    device);
527 		return PCAP_ERROR;
528 	}
529 	bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
530 	if (bytes_read == -1) {
531 		if (errno == ENOENT || errno == EINVAL) {
532 			/*
533 			 * Doesn't exist, or not a symlink; assume that
534 			 * means it's not a mac80211 device.
535 			 */
536 			free(pathstr);
537 			return 0;
538 		}
539 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
540 		    "%s: Can't readlink %s: %s", device, pathstr,
541 		    strerror(errno));
542 		free(pathstr);
543 		return PCAP_ERROR;
544 	}
545 	free(pathstr);
546 	phydev_path[bytes_read] = '\0';
547 	return 1;
548 }
549 
550 #ifdef HAVE_LIBNL_SOCKETS
551 #define get_nl_errmsg	nl_geterror
552 #else
553 /* libnl 2.x compatibility code */
554 
555 #define nl_sock nl_handle
556 
557 static inline struct nl_handle *
558 nl_socket_alloc(void)
559 {
560 	return nl_handle_alloc();
561 }
562 
563 static inline void
564 nl_socket_free(struct nl_handle *h)
565 {
566 	nl_handle_destroy(h);
567 }
568 
569 #define get_nl_errmsg	strerror
570 
571 static inline int
572 __genl_ctrl_alloc_cache(struct nl_handle *h, struct nl_cache **cache)
573 {
574 	struct nl_cache *tmp = genl_ctrl_alloc_cache(h);
575 	if (!tmp)
576 		return -ENOMEM;
577 	*cache = tmp;
578 	return 0;
579 }
580 #define genl_ctrl_alloc_cache __genl_ctrl_alloc_cache
581 #endif /* !HAVE_LIBNL_SOCKETS */
582 
583 struct nl80211_state {
584 	struct nl_sock *nl_sock;
585 	struct nl_cache *nl_cache;
586 	struct genl_family *nl80211;
587 };
588 
589 static int
590 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
591 {
592 	int err;
593 
594 	state->nl_sock = nl_socket_alloc();
595 	if (!state->nl_sock) {
596 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
597 		    "%s: failed to allocate netlink handle", device);
598 		return PCAP_ERROR;
599 	}
600 
601 	if (genl_connect(state->nl_sock)) {
602 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
603 		    "%s: failed to connect to generic netlink", device);
604 		goto out_handle_destroy;
605 	}
606 
607 	err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
608 	if (err < 0) {
609 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
610 		    "%s: failed to allocate generic netlink cache: %s",
611 		    device, get_nl_errmsg(-err));
612 		goto out_handle_destroy;
613 	}
614 
615 	state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
616 	if (!state->nl80211) {
617 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
618 		    "%s: nl80211 not found", device);
619 		goto out_cache_free;
620 	}
621 
622 	return 0;
623 
624 out_cache_free:
625 	nl_cache_free(state->nl_cache);
626 out_handle_destroy:
627 	nl_socket_free(state->nl_sock);
628 	return PCAP_ERROR;
629 }
630 
631 static void
632 nl80211_cleanup(struct nl80211_state *state)
633 {
634 	genl_family_put(state->nl80211);
635 	nl_cache_free(state->nl_cache);
636 	nl_socket_free(state->nl_sock);
637 }
638 
639 static int
640 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
641     const char *device, const char *mondevice)
642 {
643 	int ifindex;
644 	struct nl_msg *msg;
645 	int err;
646 
647 	ifindex = iface_get_id(sock_fd, device, handle->errbuf);
648 	if (ifindex == -1)
649 		return PCAP_ERROR;
650 
651 	msg = nlmsg_alloc();
652 	if (!msg) {
653 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
654 		    "%s: failed to allocate netlink msg", device);
655 		return PCAP_ERROR;
656 	}
657 
658 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
659 		    0, NL80211_CMD_NEW_INTERFACE, 0);
660 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
661 	NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
662 	NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
663 
664 	err = nl_send_auto_complete(state->nl_sock, msg);
665 	if (err < 0) {
666 #if defined HAVE_LIBNL_NLE
667 		if (err == -NLE_FAILURE) {
668 #else
669 		if (err == -ENFILE) {
670 #endif
671 			/*
672 			 * Device not available; our caller should just
673 			 * keep trying.  (libnl 2.x maps ENFILE to
674 			 * NLE_FAILURE; it can also map other errors
675 			 * to that, but there's not much we can do
676 			 * about that.)
677 			 */
678 			nlmsg_free(msg);
679 			return 0;
680 		} else {
681 			/*
682 			 * Real failure, not just "that device is not
683 			 * available.
684 			 */
685 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
686 			    "%s: nl_send_auto_complete failed adding %s interface: %s",
687 			    device, mondevice, get_nl_errmsg(-err));
688 			nlmsg_free(msg);
689 			return PCAP_ERROR;
690 		}
691 	}
692 	err = nl_wait_for_ack(state->nl_sock);
693 	if (err < 0) {
694 #if defined HAVE_LIBNL_NLE
695 		if (err == -NLE_FAILURE) {
696 #else
697 		if (err == -ENFILE) {
698 #endif
699 			/*
700 			 * Device not available; our caller should just
701 			 * keep trying.  (libnl 2.x maps ENFILE to
702 			 * NLE_FAILURE; it can also map other errors
703 			 * to that, but there's not much we can do
704 			 * about that.)
705 			 */
706 			nlmsg_free(msg);
707 			return 0;
708 		} else {
709 			/*
710 			 * Real failure, not just "that device is not
711 			 * available.
712 			 */
713 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
714 			    "%s: nl_wait_for_ack failed adding %s interface: %s",
715 			    device, mondevice, get_nl_errmsg(-err));
716 			nlmsg_free(msg);
717 			return PCAP_ERROR;
718 		}
719 	}
720 
721 	/*
722 	 * Success.
723 	 */
724 	nlmsg_free(msg);
725 	return 1;
726 
727 nla_put_failure:
728 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
729 	    "%s: nl_put failed adding %s interface",
730 	    device, mondevice);
731 	nlmsg_free(msg);
732 	return PCAP_ERROR;
733 }
734 
735 static int
736 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
737     const char *device, const char *mondevice)
738 {
739 	int ifindex;
740 	struct nl_msg *msg;
741 	int err;
742 
743 	ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
744 	if (ifindex == -1)
745 		return PCAP_ERROR;
746 
747 	msg = nlmsg_alloc();
748 	if (!msg) {
749 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
750 		    "%s: failed to allocate netlink msg", device);
751 		return PCAP_ERROR;
752 	}
753 
754 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
755 		    0, NL80211_CMD_DEL_INTERFACE, 0);
756 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
757 
758 	err = nl_send_auto_complete(state->nl_sock, msg);
759 	if (err < 0) {
760 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
761 		    "%s: nl_send_auto_complete failed deleting %s interface: %s",
762 		    device, mondevice, get_nl_errmsg(-err));
763 		nlmsg_free(msg);
764 		return PCAP_ERROR;
765 	}
766 	err = nl_wait_for_ack(state->nl_sock);
767 	if (err < 0) {
768 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
769 		    "%s: nl_wait_for_ack failed adding %s interface: %s",
770 		    device, mondevice, get_nl_errmsg(-err));
771 		nlmsg_free(msg);
772 		return PCAP_ERROR;
773 	}
774 
775 	/*
776 	 * Success.
777 	 */
778 	nlmsg_free(msg);
779 	return 1;
780 
781 nla_put_failure:
782 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
783 	    "%s: nl_put failed deleting %s interface",
784 	    device, mondevice);
785 	nlmsg_free(msg);
786 	return PCAP_ERROR;
787 }
788 
789 static int
790 enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
791 {
792 	struct pcap_linux *handlep = handle->priv;
793 	int ret;
794 	char phydev_path[PATH_MAX+1];
795 	struct nl80211_state nlstate;
796 	struct ifreq ifr;
797 	u_int n;
798 
799 	/*
800 	 * Is this a mac80211 device?
801 	 */
802 	ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
803 	if (ret < 0)
804 		return ret;	/* error */
805 	if (ret == 0)
806 		return 0;	/* no error, but not mac80211 device */
807 
808 	/*
809 	 * XXX - is this already a monN device?
810 	 * If so, we're done.
811 	 * Is that determined by old Wireless Extensions ioctls?
812 	 */
813 
814 	/*
815 	 * OK, it's apparently a mac80211 device.
816 	 * Try to find an unused monN device for it.
817 	 */
818 	ret = nl80211_init(handle, &nlstate, device);
819 	if (ret != 0)
820 		return ret;
821 	for (n = 0; n < UINT_MAX; n++) {
822 		/*
823 		 * Try mon{n}.
824 		 */
825 		char mondevice[3+10+1];	/* mon{UINT_MAX}\0 */
826 
827 		snprintf(mondevice, sizeof mondevice, "mon%u", n);
828 		ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
829 		if (ret == 1) {
830 			handlep->mondevice = strdup(mondevice);
831 			goto added;
832 		}
833 		if (ret < 0) {
834 			/*
835 			 * Hard failure.  Just return ret; handle->errbuf
836 			 * has already been set.
837 			 */
838 			nl80211_cleanup(&nlstate);
839 			return ret;
840 		}
841 	}
842 
843 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
844 	    "%s: No free monN interfaces", device);
845 	nl80211_cleanup(&nlstate);
846 	return PCAP_ERROR;
847 
848 added:
849 
850 #if 0
851 	/*
852 	 * Sleep for .1 seconds.
853 	 */
854 	delay.tv_sec = 0;
855 	delay.tv_nsec = 500000000;
856 	nanosleep(&delay, NULL);
857 #endif
858 
859 	/*
860 	 * If we haven't already done so, arrange to have
861 	 * "pcap_close_all()" called when we exit.
862 	 */
863 	if (!pcap_do_addexit(handle)) {
864 		/*
865 		 * "atexit()" failed; don't put the interface
866 		 * in rfmon mode, just give up.
867 		 */
868 		return PCAP_ERROR_RFMON_NOTSUP;
869 	}
870 
871 	/*
872 	 * Now configure the monitor interface up.
873 	 */
874 	memset(&ifr, 0, sizeof(ifr));
875 	strncpy(ifr.ifr_name, handlep->mondevice, sizeof(ifr.ifr_name));
876 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
877 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
878 		    "%s: Can't get flags for %s: %s", device,
879 		    handlep->mondevice, strerror(errno));
880 		del_mon_if(handle, sock_fd, &nlstate, device,
881 		    handlep->mondevice);
882 		nl80211_cleanup(&nlstate);
883 		return PCAP_ERROR;
884 	}
885 	ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
886 	if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
887 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
888 		    "%s: Can't set flags for %s: %s", device,
889 		    handlep->mondevice, strerror(errno));
890 		del_mon_if(handle, sock_fd, &nlstate, device,
891 		    handlep->mondevice);
892 		nl80211_cleanup(&nlstate);
893 		return PCAP_ERROR;
894 	}
895 
896 	/*
897 	 * Success.  Clean up the libnl state.
898 	 */
899 	nl80211_cleanup(&nlstate);
900 
901 	/*
902 	 * Note that we have to delete the monitor device when we close
903 	 * the handle.
904 	 */
905 	handlep->must_do_on_close |= MUST_DELETE_MONIF;
906 
907 	/*
908 	 * Add this to the list of pcaps to close when we exit.
909 	 */
910 	pcap_add_to_pcaps_to_close(handle);
911 
912 	return 1;
913 }
914 #endif /* HAVE_LIBNL */
915 
916 static int
917 pcap_can_set_rfmon_linux(pcap_t *handle)
918 {
919 #ifdef HAVE_LIBNL
920 	char phydev_path[PATH_MAX+1];
921 	int ret;
922 #endif
923 #ifdef IW_MODE_MONITOR
924 	int sock_fd;
925 	struct iwreq ireq;
926 #endif
927 
928 	if (strcmp(handle->opt.source, "any") == 0) {
929 		/*
930 		 * Monitor mode makes no sense on the "any" device.
931 		 */
932 		return 0;
933 	}
934 
935 #ifdef HAVE_LIBNL
936 	/*
937 	 * Bleah.  There doesn't seem to be a way to ask a mac80211
938 	 * device, through libnl, whether it supports monitor mode;
939 	 * we'll just check whether the device appears to be a
940 	 * mac80211 device and, if so, assume the device supports
941 	 * monitor mode.
942 	 *
943 	 * wmaster devices don't appear to support the Wireless
944 	 * Extensions, but we can create a mon device for a
945 	 * wmaster device, so we don't bother checking whether
946 	 * a mac80211 device supports the Wireless Extensions.
947 	 */
948 	ret = get_mac80211_phydev(handle, handle->opt.source, phydev_path,
949 	    PATH_MAX);
950 	if (ret < 0)
951 		return ret;	/* error */
952 	if (ret == 1)
953 		return 1;	/* mac80211 device */
954 #endif
955 
956 #ifdef IW_MODE_MONITOR
957 	/*
958 	 * Bleah.  There doesn't appear to be an ioctl to use to ask
959 	 * whether a device supports monitor mode; we'll just do
960 	 * SIOCGIWMODE and, if it succeeds, assume the device supports
961 	 * monitor mode.
962 	 *
963 	 * Open a socket on which to attempt to get the mode.
964 	 * (We assume that if we have Wireless Extensions support
965 	 * we also have PF_PACKET support.)
966 	 */
967 	sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
968 	if (sock_fd == -1) {
969 		(void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
970 		    "socket: %s", pcap_strerror(errno));
971 		return PCAP_ERROR;
972 	}
973 
974 	/*
975 	 * Attempt to get the current mode.
976 	 */
977 	strncpy(ireq.ifr_ifrn.ifrn_name, handle->opt.source,
978 	    sizeof ireq.ifr_ifrn.ifrn_name);
979 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
980 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
981 		/*
982 		 * Well, we got the mode; assume we can set it.
983 		 */
984 		close(sock_fd);
985 		return 1;
986 	}
987 	if (errno == ENODEV) {
988 		/* The device doesn't even exist. */
989 		(void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
990 		    "SIOCGIWMODE failed: %s", pcap_strerror(errno));
991 		close(sock_fd);
992 		return PCAP_ERROR_NO_SUCH_DEVICE;
993 	}
994 	close(sock_fd);
995 #endif
996 	return 0;
997 }
998 
999 /*
1000  * Grabs the number of dropped packets by the interface from /proc/net/dev.
1001  *
1002  * XXX - what about /sys/class/net/{interface name}/rx_*?  There are
1003  * individual devices giving, in ASCII, various rx_ and tx_ statistics.
1004  *
1005  * Or can we get them in binary form from netlink?
1006  */
1007 static long int
1008 linux_if_drops(const char * if_name)
1009 {
1010 	char buffer[512];
1011 	char * bufptr;
1012 	FILE * file;
1013 	int field_to_convert = 3, if_name_sz = strlen(if_name);
1014 	long int dropped_pkts = 0;
1015 
1016 	file = fopen("/proc/net/dev", "r");
1017 	if (!file)
1018 		return 0;
1019 
1020 	while (!dropped_pkts && fgets( buffer, sizeof(buffer), file ))
1021 	{
1022 		/* 	search for 'bytes' -- if its in there, then
1023 			that means we need to grab the fourth field. otherwise
1024 			grab the third field. */
1025 		if (field_to_convert != 4 && strstr(buffer, "bytes"))
1026 		{
1027 			field_to_convert = 4;
1028 			continue;
1029 		}
1030 
1031 		/* find iface and make sure it actually matches -- space before the name and : after it */
1032 		if ((bufptr = strstr(buffer, if_name)) &&
1033 			(bufptr == buffer || *(bufptr-1) == ' ') &&
1034 			*(bufptr + if_name_sz) == ':')
1035 		{
1036 			bufptr = bufptr + if_name_sz + 1;
1037 
1038 			/* grab the nth field from it */
1039 			while( --field_to_convert && *bufptr != '\0')
1040 			{
1041 				while (*bufptr != '\0' && *(bufptr++) == ' ');
1042 				while (*bufptr != '\0' && *(bufptr++) != ' ');
1043 			}
1044 
1045 			/* get rid of any final spaces */
1046 			while (*bufptr != '\0' && *bufptr == ' ') bufptr++;
1047 
1048 			if (*bufptr != '\0')
1049 				dropped_pkts = strtol(bufptr, NULL, 10);
1050 
1051 			break;
1052 		}
1053 	}
1054 
1055 	fclose(file);
1056 	return dropped_pkts;
1057 }
1058 
1059 
1060 /*
1061  * With older kernels promiscuous mode is kind of interesting because we
1062  * have to reset the interface before exiting. The problem can't really
1063  * be solved without some daemon taking care of managing usage counts.
1064  * If we put the interface into promiscuous mode, we set a flag indicating
1065  * that we must take it out of that mode when the interface is closed,
1066  * and, when closing the interface, if that flag is set we take it out
1067  * of promiscuous mode.
1068  *
1069  * Even with newer kernels, we have the same issue with rfmon mode.
1070  */
1071 
1072 static void	pcap_cleanup_linux( pcap_t *handle )
1073 {
1074 	struct pcap_linux *handlep = handle->priv;
1075 	struct ifreq	ifr;
1076 #ifdef HAVE_LIBNL
1077 	struct nl80211_state nlstate;
1078 	int ret;
1079 #endif /* HAVE_LIBNL */
1080 #ifdef IW_MODE_MONITOR
1081 	int oldflags;
1082 	struct iwreq ireq;
1083 #endif /* IW_MODE_MONITOR */
1084 
1085 	if (handlep->must_do_on_close != 0) {
1086 		/*
1087 		 * There's something we have to do when closing this
1088 		 * pcap_t.
1089 		 */
1090 		if (handlep->must_do_on_close & MUST_CLEAR_PROMISC) {
1091 			/*
1092 			 * We put the interface into promiscuous mode;
1093 			 * take it out of promiscuous mode.
1094 			 *
1095 			 * XXX - if somebody else wants it in promiscuous
1096 			 * mode, this code cannot know that, so it'll take
1097 			 * it out of promiscuous mode.  That's not fixable
1098 			 * in 2.0[.x] kernels.
1099 			 */
1100 			memset(&ifr, 0, sizeof(ifr));
1101 			strncpy(ifr.ifr_name, handlep->device,
1102 			    sizeof(ifr.ifr_name));
1103 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
1104 				fprintf(stderr,
1105 				    "Can't restore interface %s flags (SIOCGIFFLAGS failed: %s).\n"
1106 				    "Please adjust manually.\n"
1107 				    "Hint: This can't happen with Linux >= 2.2.0.\n",
1108 				    handlep->device, strerror(errno));
1109 			} else {
1110 				if (ifr.ifr_flags & IFF_PROMISC) {
1111 					/*
1112 					 * Promiscuous mode is currently on;
1113 					 * turn it off.
1114 					 */
1115 					ifr.ifr_flags &= ~IFF_PROMISC;
1116 					if (ioctl(handle->fd, SIOCSIFFLAGS,
1117 					    &ifr) == -1) {
1118 						fprintf(stderr,
1119 						    "Can't restore interface %s flags (SIOCSIFFLAGS failed: %s).\n"
1120 						    "Please adjust manually.\n"
1121 						    "Hint: This can't happen with Linux >= 2.2.0.\n",
1122 						    handlep->device,
1123 						    strerror(errno));
1124 					}
1125 				}
1126 			}
1127 		}
1128 
1129 #ifdef HAVE_LIBNL
1130 		if (handlep->must_do_on_close & MUST_DELETE_MONIF) {
1131 			ret = nl80211_init(handle, &nlstate, handlep->device);
1132 			if (ret >= 0) {
1133 				ret = del_mon_if(handle, handle->fd, &nlstate,
1134 				    handlep->device, handlep->mondevice);
1135 				nl80211_cleanup(&nlstate);
1136 			}
1137 			if (ret < 0) {
1138 				fprintf(stderr,
1139 				    "Can't delete monitor interface %s (%s).\n"
1140 				    "Please delete manually.\n",
1141 				    handlep->mondevice, handle->errbuf);
1142 			}
1143 		}
1144 #endif /* HAVE_LIBNL */
1145 
1146 #ifdef IW_MODE_MONITOR
1147 		if (handlep->must_do_on_close & MUST_CLEAR_RFMON) {
1148 			/*
1149 			 * We put the interface into rfmon mode;
1150 			 * take it out of rfmon mode.
1151 			 *
1152 			 * XXX - if somebody else wants it in rfmon
1153 			 * mode, this code cannot know that, so it'll take
1154 			 * it out of rfmon mode.
1155 			 */
1156 
1157 			/*
1158 			 * First, take the interface down if it's up;
1159 			 * otherwise, we might get EBUSY.
1160 			 * If we get errors, just drive on and print
1161 			 * a warning if we can't restore the mode.
1162 			 */
1163 			oldflags = 0;
1164 			memset(&ifr, 0, sizeof(ifr));
1165 			strncpy(ifr.ifr_name, handlep->device,
1166 			    sizeof(ifr.ifr_name));
1167 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) != -1) {
1168 				if (ifr.ifr_flags & IFF_UP) {
1169 					oldflags = ifr.ifr_flags;
1170 					ifr.ifr_flags &= ~IFF_UP;
1171 					if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1)
1172 						oldflags = 0;	/* didn't set, don't restore */
1173 				}
1174 			}
1175 
1176 			/*
1177 			 * Now restore the mode.
1178 			 */
1179 			strncpy(ireq.ifr_ifrn.ifrn_name, handlep->device,
1180 			    sizeof ireq.ifr_ifrn.ifrn_name);
1181 			ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1]
1182 			    = 0;
1183 			ireq.u.mode = handlep->oldmode;
1184 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
1185 				/*
1186 				 * Scientist, you've failed.
1187 				 */
1188 				fprintf(stderr,
1189 				    "Can't restore interface %s wireless mode (SIOCSIWMODE failed: %s).\n"
1190 				    "Please adjust manually.\n",
1191 				    handlep->device, strerror(errno));
1192 			}
1193 
1194 			/*
1195 			 * Now bring the interface back up if we brought
1196 			 * it down.
1197 			 */
1198 			if (oldflags != 0) {
1199 				ifr.ifr_flags = oldflags;
1200 				if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
1201 					fprintf(stderr,
1202 					    "Can't bring interface %s back up (SIOCSIFFLAGS failed: %s).\n"
1203 					    "Please adjust manually.\n",
1204 					    handlep->device, strerror(errno));
1205 				}
1206 			}
1207 		}
1208 #endif /* IW_MODE_MONITOR */
1209 
1210 		/*
1211 		 * Take this pcap out of the list of pcaps for which we
1212 		 * have to take the interface out of some mode.
1213 		 */
1214 		pcap_remove_from_pcaps_to_close(handle);
1215 	}
1216 
1217 	if (handlep->mondevice != NULL) {
1218 		free(handlep->mondevice);
1219 		handlep->mondevice = NULL;
1220 	}
1221 	if (handlep->device != NULL) {
1222 		free(handlep->device);
1223 		handlep->device = NULL;
1224 	}
1225 	pcap_cleanup_live_common(handle);
1226 }
1227 
1228 /*
1229  *  Get a handle for a live capture from the given device. You can
1230  *  pass NULL as device to get all packages (without link level
1231  *  information of course). If you pass 1 as promisc the interface
1232  *  will be set to promiscous mode (XXX: I think this usage should
1233  *  be deprecated and functions be added to select that later allow
1234  *  modification of that values -- Torsten).
1235  */
1236 static int
1237 pcap_activate_linux(pcap_t *handle)
1238 {
1239 	struct pcap_linux *handlep = handle->priv;
1240 	const char	*device;
1241 	int		status = 0;
1242 
1243 	device = handle->opt.source;
1244 
1245 	handle->inject_op = pcap_inject_linux;
1246 	handle->setfilter_op = pcap_setfilter_linux;
1247 	handle->setdirection_op = pcap_setdirection_linux;
1248 	handle->set_datalink_op = pcap_set_datalink_linux;
1249 	handle->getnonblock_op = pcap_getnonblock_fd;
1250 	handle->setnonblock_op = pcap_setnonblock_fd;
1251 	handle->cleanup_op = pcap_cleanup_linux;
1252 	handle->read_op = pcap_read_linux;
1253 	handle->stats_op = pcap_stats_linux;
1254 
1255 	/*
1256 	 * The "any" device is a special device which causes us not
1257 	 * to bind to a particular device and thus to look at all
1258 	 * devices.
1259 	 */
1260 	if (strcmp(device, "any") == 0) {
1261 		if (handle->opt.promisc) {
1262 			handle->opt.promisc = 0;
1263 			/* Just a warning. */
1264 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1265 			    "Promiscuous mode not supported on the \"any\" device");
1266 			status = PCAP_WARNING_PROMISC_NOTSUP;
1267 		}
1268 	}
1269 
1270 	handlep->device	= strdup(device);
1271 	if (handlep->device == NULL) {
1272 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
1273 			 pcap_strerror(errno) );
1274 		return PCAP_ERROR;
1275 	}
1276 
1277 	/* copy timeout value */
1278 	handlep->timeout = handle->opt.timeout;
1279 
1280 	/*
1281 	 * If we're in promiscuous mode, then we probably want
1282 	 * to see when the interface drops packets too, so get an
1283 	 * initial count from /proc/net/dev
1284 	 */
1285 	if (handle->opt.promisc)
1286 		handlep->proc_dropped = linux_if_drops(handlep->device);
1287 
1288 	/*
1289 	 * Current Linux kernels use the protocol family PF_PACKET to
1290 	 * allow direct access to all packets on the network while
1291 	 * older kernels had a special socket type SOCK_PACKET to
1292 	 * implement this feature.
1293 	 * While this old implementation is kind of obsolete we need
1294 	 * to be compatible with older kernels for a while so we are
1295 	 * trying both methods with the newer method preferred.
1296 	 */
1297 	status = activate_new(handle);
1298 	if (status < 0) {
1299 		/*
1300 		 * Fatal error with the new way; just fail.
1301 		 * status has the error return; if it's PCAP_ERROR,
1302 		 * handle->errbuf has been set appropriately.
1303 		 */
1304 		goto fail;
1305 	}
1306 	if (status == 1) {
1307 		/*
1308 		 * Success.
1309 		 * Try to use memory-mapped access.
1310 		 */
1311 		switch (activate_mmap(handle, &status)) {
1312 
1313 		case 1:
1314 			/*
1315 			 * We succeeded.  status has been
1316 			 * set to the status to return,
1317 			 * which might be 0, or might be
1318 			 * a PCAP_WARNING_ value.
1319 			 */
1320 			return status;
1321 
1322 		case 0:
1323 			/*
1324 			 * Kernel doesn't support it - just continue
1325 			 * with non-memory-mapped access.
1326 			 */
1327 			break;
1328 
1329 		case -1:
1330 			/*
1331 			 * We failed to set up to use it, or the kernel
1332 			 * supports it, but we failed to enable it.
1333 			 * status has been set to the error status to
1334 			 * return and, if it's PCAP_ERROR, handle->errbuf
1335 			 * contains the error message.
1336 			 */
1337 			goto fail;
1338 		}
1339 	}
1340 	else if (status == 0) {
1341 		/* Non-fatal error; try old way */
1342 		if ((status = activate_old(handle)) != 1) {
1343 			/*
1344 			 * Both methods to open the packet socket failed.
1345 			 * Tidy up and report our failure (handle->errbuf
1346 			 * is expected to be set by the functions above).
1347 			 */
1348 			goto fail;
1349 		}
1350 	}
1351 
1352 	/*
1353 	 * We set up the socket, but not with memory-mapped access.
1354 	 */
1355 	status = 0;
1356 	if (handle->opt.buffer_size != 0) {
1357 		/*
1358 		 * Set the socket buffer size to the specified value.
1359 		 */
1360 		if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF,
1361 		    &handle->opt.buffer_size,
1362 		    sizeof(handle->opt.buffer_size)) == -1) {
1363 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1364 				 "SO_RCVBUF: %s", pcap_strerror(errno));
1365 			status = PCAP_ERROR;
1366 			goto fail;
1367 		}
1368 	}
1369 
1370 	/* Allocate the buffer */
1371 
1372 	handle->buffer	 = malloc(handle->bufsize + handle->offset);
1373 	if (!handle->buffer) {
1374 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1375 			 "malloc: %s", pcap_strerror(errno));
1376 		status = PCAP_ERROR;
1377 		goto fail;
1378 	}
1379 
1380 	/*
1381 	 * "handle->fd" is a socket, so "select()" and "poll()"
1382 	 * should work on it.
1383 	 */
1384 	handle->selectable_fd = handle->fd;
1385 
1386 	return status;
1387 
1388 fail:
1389 	pcap_cleanup_linux(handle);
1390 	return status;
1391 }
1392 
1393 /*
1394  *  Read at most max_packets from the capture stream and call the callback
1395  *  for each of them. Returns the number of packets handled or -1 if an
1396  *  error occured.
1397  */
1398 static int
1399 pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
1400 {
1401 	/*
1402 	 * Currently, on Linux only one packet is delivered per read,
1403 	 * so we don't loop.
1404 	 */
1405 	return pcap_read_packet(handle, callback, user);
1406 }
1407 
1408 static int
1409 pcap_set_datalink_linux(pcap_t *handle, int dlt)
1410 {
1411 	handle->linktype = dlt;
1412 	return 0;
1413 }
1414 
1415 /*
1416  * linux_check_direction()
1417  *
1418  * Do checks based on packet direction.
1419  */
1420 static inline int
1421 linux_check_direction(const pcap_t *handle, const struct sockaddr_ll *sll)
1422 {
1423 	struct pcap_linux	*handlep = handle->priv;
1424 
1425 	if (sll->sll_pkttype == PACKET_OUTGOING) {
1426 		/*
1427 		 * Outgoing packet.
1428 		 * If this is from the loopback device, reject it;
1429 		 * we'll see the packet as an incoming packet as well,
1430 		 * and we don't want to see it twice.
1431 		 */
1432 		if (sll->sll_ifindex == handlep->lo_ifindex)
1433 			return 0;
1434 
1435 		/*
1436 		 * If the user only wants incoming packets, reject it.
1437 		 */
1438 		if (handle->direction == PCAP_D_IN)
1439 			return 0;
1440 	} else {
1441 		/*
1442 		 * Incoming packet.
1443 		 * If the user only wants outgoing packets, reject it.
1444 		 */
1445 		if (handle->direction == PCAP_D_OUT)
1446 			return 0;
1447 	}
1448 	return 1;
1449 }
1450 
1451 /*
1452  *  Read a packet from the socket calling the handler provided by
1453  *  the user. Returns the number of packets received or -1 if an
1454  *  error occured.
1455  */
1456 static int
1457 pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
1458 {
1459 	struct pcap_linux	*handlep = handle->priv;
1460 	u_char			*bp;
1461 	int			offset;
1462 #ifdef HAVE_PF_PACKET_SOCKETS
1463 	struct sockaddr_ll	from;
1464 	struct sll_header	*hdrp;
1465 #else
1466 	struct sockaddr		from;
1467 #endif
1468 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1469 	struct iovec		iov;
1470 	struct msghdr		msg;
1471 	struct cmsghdr		*cmsg;
1472 	union {
1473 		struct cmsghdr	cmsg;
1474 		char		buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
1475 	} cmsg_buf;
1476 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1477 	socklen_t		fromlen;
1478 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1479 	int			packet_len, caplen;
1480 	struct pcap_pkthdr	pcap_header;
1481 
1482 #ifdef HAVE_PF_PACKET_SOCKETS
1483 	/*
1484 	 * If this is a cooked device, leave extra room for a
1485 	 * fake packet header.
1486 	 */
1487 	if (handlep->cooked)
1488 		offset = SLL_HDR_LEN;
1489 	else
1490 		offset = 0;
1491 #else
1492 	/*
1493 	 * This system doesn't have PF_PACKET sockets, so it doesn't
1494 	 * support cooked devices.
1495 	 */
1496 	offset = 0;
1497 #endif
1498 
1499 	/*
1500 	 * Receive a single packet from the kernel.
1501 	 * We ignore EINTR, as that might just be due to a signal
1502 	 * being delivered - if the signal should interrupt the
1503 	 * loop, the signal handler should call pcap_breakloop()
1504 	 * to set handle->break_loop (we ignore it on other
1505 	 * platforms as well).
1506 	 * We also ignore ENETDOWN, so that we can continue to
1507 	 * capture traffic if the interface goes down and comes
1508 	 * back up again; comments in the kernel indicate that
1509 	 * we'll just block waiting for packets if we try to
1510 	 * receive from a socket that delivered ENETDOWN, and,
1511 	 * if we're using a memory-mapped buffer, we won't even
1512 	 * get notified of "network down" events.
1513 	 */
1514 	bp = handle->buffer + handle->offset;
1515 
1516 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1517 	msg.msg_name		= &from;
1518 	msg.msg_namelen		= sizeof(from);
1519 	msg.msg_iov		= &iov;
1520 	msg.msg_iovlen		= 1;
1521 	msg.msg_control		= &cmsg_buf;
1522 	msg.msg_controllen	= sizeof(cmsg_buf);
1523 	msg.msg_flags		= 0;
1524 
1525 	iov.iov_len		= handle->bufsize - offset;
1526 	iov.iov_base		= bp + offset;
1527 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1528 
1529 	do {
1530 		/*
1531 		 * Has "pcap_breakloop()" been called?
1532 		 */
1533 		if (handle->break_loop) {
1534 			/*
1535 			 * Yes - clear the flag that indicates that it has,
1536 			 * and return PCAP_ERROR_BREAK as an indication that
1537 			 * we were told to break out of the loop.
1538 			 */
1539 			handle->break_loop = 0;
1540 			return PCAP_ERROR_BREAK;
1541 		}
1542 
1543 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1544 		packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC);
1545 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1546 		fromlen = sizeof(from);
1547 		packet_len = recvfrom(
1548 			handle->fd, bp + offset,
1549 			handle->bufsize - offset, MSG_TRUNC,
1550 			(struct sockaddr *) &from, &fromlen);
1551 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1552 	} while (packet_len == -1 && errno == EINTR);
1553 
1554 	/* Check if an error occured */
1555 
1556 	if (packet_len == -1) {
1557 		switch (errno) {
1558 
1559 		case EAGAIN:
1560 			return 0;	/* no packet there */
1561 
1562 		case ENETDOWN:
1563 			/*
1564 			 * The device on which we're capturing went away.
1565 			 *
1566 			 * XXX - we should really return
1567 			 * PCAP_ERROR_IFACE_NOT_UP, but pcap_dispatch()
1568 			 * etc. aren't defined to return that.
1569 			 */
1570 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1571 				"The interface went down");
1572 			return PCAP_ERROR;
1573 
1574 		default:
1575 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1576 				 "recvfrom: %s", pcap_strerror(errno));
1577 			return PCAP_ERROR;
1578 		}
1579 	}
1580 
1581 #ifdef HAVE_PF_PACKET_SOCKETS
1582 	if (!handlep->sock_packet) {
1583 		/*
1584 		 * Unfortunately, there is a window between socket() and
1585 		 * bind() where the kernel may queue packets from any
1586 		 * interface.  If we're bound to a particular interface,
1587 		 * discard packets not from that interface.
1588 		 *
1589 		 * (If socket filters are supported, we could do the
1590 		 * same thing we do when changing the filter; however,
1591 		 * that won't handle packet sockets without socket
1592 		 * filter support, and it's a bit more complicated.
1593 		 * It would save some instructions per packet, however.)
1594 		 */
1595 		if (handlep->ifindex != -1 &&
1596 		    from.sll_ifindex != handlep->ifindex)
1597 			return 0;
1598 
1599 		/*
1600 		 * Do checks based on packet direction.
1601 		 * We can only do this if we're using PF_PACKET; the
1602 		 * address returned for SOCK_PACKET is a "sockaddr_pkt"
1603 		 * which lacks the relevant packet type information.
1604 		 */
1605 		if (!linux_check_direction(handle, &from))
1606 			return 0;
1607 	}
1608 #endif
1609 
1610 #ifdef HAVE_PF_PACKET_SOCKETS
1611 	/*
1612 	 * If this is a cooked device, fill in the fake packet header.
1613 	 */
1614 	if (handlep->cooked) {
1615 		/*
1616 		 * Add the length of the fake header to the length
1617 		 * of packet data we read.
1618 		 */
1619 		packet_len += SLL_HDR_LEN;
1620 
1621 		hdrp = (struct sll_header *)bp;
1622 		hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype);
1623 		hdrp->sll_hatype = htons(from.sll_hatype);
1624 		hdrp->sll_halen = htons(from.sll_halen);
1625 		memcpy(hdrp->sll_addr, from.sll_addr,
1626 		    (from.sll_halen > SLL_ADDRLEN) ?
1627 		      SLL_ADDRLEN :
1628 		      from.sll_halen);
1629 		hdrp->sll_protocol = from.sll_protocol;
1630 	}
1631 
1632 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1633 	if (handlep->vlan_offset != -1) {
1634 		for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
1635 			struct tpacket_auxdata *aux;
1636 			unsigned int len;
1637 			struct vlan_tag *tag;
1638 
1639 			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
1640 			    cmsg->cmsg_level != SOL_PACKET ||
1641 			    cmsg->cmsg_type != PACKET_AUXDATA)
1642 				continue;
1643 
1644 			aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
1645 #if defined(TP_STATUS_VLAN_VALID)
1646 			if ((aux->tp_vlan_tci == 0) && !(aux->tp_status & TP_STATUS_VLAN_VALID))
1647 #else
1648 			if (aux->tp_vlan_tci == 0) /* this is ambigious but without the
1649 						TP_STATUS_VLAN_VALID flag, there is
1650 						nothing that we can do */
1651 #endif
1652 				continue;
1653 
1654 			len = packet_len > iov.iov_len ? iov.iov_len : packet_len;
1655 			if (len < (unsigned int) handlep->vlan_offset)
1656 				break;
1657 
1658 			bp -= VLAN_TAG_LEN;
1659 			memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
1660 
1661 			tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
1662 			tag->vlan_tpid = htons(ETH_P_8021Q);
1663 			tag->vlan_tci = htons(aux->tp_vlan_tci);
1664 
1665 			packet_len += VLAN_TAG_LEN;
1666 		}
1667 	}
1668 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1669 #endif /* HAVE_PF_PACKET_SOCKETS */
1670 
1671 	/*
1672 	 * XXX: According to the kernel source we should get the real
1673 	 * packet len if calling recvfrom with MSG_TRUNC set. It does
1674 	 * not seem to work here :(, but it is supported by this code
1675 	 * anyway.
1676 	 * To be honest the code RELIES on that feature so this is really
1677 	 * broken with 2.2.x kernels.
1678 	 * I spend a day to figure out what's going on and I found out
1679 	 * that the following is happening:
1680 	 *
1681 	 * The packet comes from a random interface and the packet_rcv
1682 	 * hook is called with a clone of the packet. That code inserts
1683 	 * the packet into the receive queue of the packet socket.
1684 	 * If a filter is attached to that socket that filter is run
1685 	 * first - and there lies the problem. The default filter always
1686 	 * cuts the packet at the snaplen:
1687 	 *
1688 	 * # tcpdump -d
1689 	 * (000) ret      #68
1690 	 *
1691 	 * So the packet filter cuts down the packet. The recvfrom call
1692 	 * says "hey, it's only 68 bytes, it fits into the buffer" with
1693 	 * the result that we don't get the real packet length. This
1694 	 * is valid at least until kernel 2.2.17pre6.
1695 	 *
1696 	 * We currently handle this by making a copy of the filter
1697 	 * program, fixing all "ret" instructions with non-zero
1698 	 * operands to have an operand of 65535 so that the filter
1699 	 * doesn't truncate the packet, and supplying that modified
1700 	 * filter to the kernel.
1701 	 */
1702 
1703 	caplen = packet_len;
1704 	if (caplen > handle->snapshot)
1705 		caplen = handle->snapshot;
1706 
1707 	/* Run the packet filter if not using kernel filter */
1708 	if (!handlep->filtering_in_kernel && handle->fcode.bf_insns) {
1709 		if (bpf_filter(handle->fcode.bf_insns, bp,
1710 		                packet_len, caplen) == 0)
1711 		{
1712 			/* rejected by filter */
1713 			return 0;
1714 		}
1715 	}
1716 
1717 	/* Fill in our own header data */
1718 
1719 	/* get timestamp for this packet */
1720 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
1721 	if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
1722 		if (ioctl(handle->fd, SIOCGSTAMPNS, &pcap_header.ts) == -1) {
1723 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1724 					"SIOCGSTAMPNS: %s", pcap_strerror(errno));
1725 			return PCAP_ERROR;
1726 		}
1727         } else
1728 #endif
1729 	{
1730 		if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
1731 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1732 					"SIOCGSTAMP: %s", pcap_strerror(errno));
1733 			return PCAP_ERROR;
1734 		}
1735         }
1736 
1737 	pcap_header.caplen	= caplen;
1738 	pcap_header.len		= packet_len;
1739 
1740 	/*
1741 	 * Count the packet.
1742 	 *
1743 	 * Arguably, we should count them before we check the filter,
1744 	 * as on many other platforms "ps_recv" counts packets
1745 	 * handed to the filter rather than packets that passed
1746 	 * the filter, but if filtering is done in the kernel, we
1747 	 * can't get a count of packets that passed the filter,
1748 	 * and that would mean the meaning of "ps_recv" wouldn't
1749 	 * be the same on all Linux systems.
1750 	 *
1751 	 * XXX - it's not the same on all systems in any case;
1752 	 * ideally, we should have a "get the statistics" call
1753 	 * that supplies more counts and indicates which of them
1754 	 * it supplies, so that we supply a count of packets
1755 	 * handed to the filter only on platforms where that
1756 	 * information is available.
1757 	 *
1758 	 * We count them here even if we can get the packet count
1759 	 * from the kernel, as we can only determine at run time
1760 	 * whether we'll be able to get it from the kernel (if
1761 	 * HAVE_TPACKET_STATS isn't defined, we can't get it from
1762 	 * the kernel, but if it is defined, the library might
1763 	 * have been built with a 2.4 or later kernel, but we
1764 	 * might be running on a 2.2[.x] kernel without Alexey
1765 	 * Kuznetzov's turbopacket patches, and thus the kernel
1766 	 * might not be able to supply those statistics).  We
1767 	 * could, I guess, try, when opening the socket, to get
1768 	 * the statistics, and if we can not increment the count
1769 	 * here, but it's not clear that always incrementing
1770 	 * the count is more expensive than always testing a flag
1771 	 * in memory.
1772 	 *
1773 	 * We keep the count in "handlep->packets_read", and use that
1774 	 * for "ps_recv" if we can't get the statistics from the kernel.
1775 	 * We do that because, if we *can* get the statistics from
1776 	 * the kernel, we use "handlep->stat.ps_recv" and
1777 	 * "handlep->stat.ps_drop" as running counts, as reading the
1778 	 * statistics from the kernel resets the kernel statistics,
1779 	 * and if we directly increment "handlep->stat.ps_recv" here,
1780 	 * that means it will count packets *twice* on systems where
1781 	 * we can get kernel statistics - once here, and once in
1782 	 * pcap_stats_linux().
1783 	 */
1784 	handlep->packets_read++;
1785 
1786 	/* Call the user supplied callback function */
1787 	callback(userdata, &pcap_header, bp);
1788 
1789 	return 1;
1790 }
1791 
1792 static int
1793 pcap_inject_linux(pcap_t *handle, const void *buf, size_t size)
1794 {
1795 	struct pcap_linux *handlep = handle->priv;
1796 	int ret;
1797 
1798 #ifdef HAVE_PF_PACKET_SOCKETS
1799 	if (!handlep->sock_packet) {
1800 		/* PF_PACKET socket */
1801 		if (handlep->ifindex == -1) {
1802 			/*
1803 			 * We don't support sending on the "any" device.
1804 			 */
1805 			strlcpy(handle->errbuf,
1806 			    "Sending packets isn't supported on the \"any\" device",
1807 			    PCAP_ERRBUF_SIZE);
1808 			return (-1);
1809 		}
1810 
1811 		if (handlep->cooked) {
1812 			/*
1813 			 * We don't support sending on the "any" device.
1814 			 *
1815 			 * XXX - how do you send on a bound cooked-mode
1816 			 * socket?
1817 			 * Is a "sendto()" required there?
1818 			 */
1819 			strlcpy(handle->errbuf,
1820 			    "Sending packets isn't supported in cooked mode",
1821 			    PCAP_ERRBUF_SIZE);
1822 			return (-1);
1823 		}
1824 	}
1825 #endif
1826 
1827 	ret = send(handle->fd, buf, size, 0);
1828 	if (ret == -1) {
1829 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
1830 		    pcap_strerror(errno));
1831 		return (-1);
1832 	}
1833 	return (ret);
1834 }
1835 
1836 /*
1837  *  Get the statistics for the given packet capture handle.
1838  *  Reports the number of dropped packets iff the kernel supports
1839  *  the PACKET_STATISTICS "getsockopt()" argument (2.4 and later
1840  *  kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket
1841  *  patches); otherwise, that information isn't available, and we lie
1842  *  and report 0 as the count of dropped packets.
1843  */
1844 static int
1845 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
1846 {
1847 	struct pcap_linux *handlep = handle->priv;
1848 #ifdef HAVE_TPACKET_STATS
1849 #ifdef HAVE_TPACKET3
1850 	/*
1851 	 * For sockets using TPACKET_V1 or TPACKET_V2, the extra
1852 	 * stuff at the end of a struct tpacket_stats_v3 will not
1853 	 * be filled in, and we don't look at it so this is OK even
1854 	 * for those sockets.  In addition, the PF_PACKET socket
1855 	 * code in the kernel only uses the length parameter to
1856 	 * compute how much data to copy out and to indicate how
1857 	 * much data was copied out, so it's OK to base it on the
1858 	 * size of a struct tpacket_stats.
1859 	 *
1860 	 * XXX - it's probably OK, in fact, to just use a
1861 	 * struct tpacket_stats for V3 sockets, as we don't
1862 	 * care about the tp_freeze_q_cnt stat.
1863 	 */
1864 	struct tpacket_stats_v3 kstats;
1865 #else /* HAVE_TPACKET3 */
1866 	struct tpacket_stats kstats;
1867 #endif /* HAVE_TPACKET3 */
1868 	socklen_t len = sizeof (struct tpacket_stats);
1869 #endif /* HAVE_TPACKET_STATS */
1870 
1871 	long if_dropped = 0;
1872 
1873 	/*
1874 	 *	To fill in ps_ifdrop, we parse /proc/net/dev for the number
1875 	 */
1876 	if (handle->opt.promisc)
1877 	{
1878 		if_dropped = handlep->proc_dropped;
1879 		handlep->proc_dropped = linux_if_drops(handlep->device);
1880 		handlep->stat.ps_ifdrop += (handlep->proc_dropped - if_dropped);
1881 	}
1882 
1883 #ifdef HAVE_TPACKET_STATS
1884 	/*
1885 	 * Try to get the packet counts from the kernel.
1886 	 */
1887 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
1888 			&kstats, &len) > -1) {
1889 		/*
1890 		 * On systems where the PACKET_STATISTICS "getsockopt()"
1891 		 * argument is supported on PF_PACKET sockets:
1892 		 *
1893 		 *	"ps_recv" counts only packets that *passed* the
1894 		 *	filter, not packets that didn't pass the filter.
1895 		 *	This includes packets later dropped because we
1896 		 *	ran out of buffer space.
1897 		 *
1898 		 *	"ps_drop" counts packets dropped because we ran
1899 		 *	out of buffer space.  It doesn't count packets
1900 		 *	dropped by the interface driver.  It counts only
1901 		 *	packets that passed the filter.
1902 		 *
1903 		 *	See above for ps_ifdrop.
1904 		 *
1905 		 *	Both statistics include packets not yet read from
1906 		 *	the kernel by libpcap, and thus not yet seen by
1907 		 *	the application.
1908 		 *
1909 		 * In "linux/net/packet/af_packet.c", at least in the
1910 		 * 2.4.9 kernel, "tp_packets" is incremented for every
1911 		 * packet that passes the packet filter *and* is
1912 		 * successfully queued on the socket; "tp_drops" is
1913 		 * incremented for every packet dropped because there's
1914 		 * not enough free space in the socket buffer.
1915 		 *
1916 		 * When the statistics are returned for a PACKET_STATISTICS
1917 		 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
1918 		 * so that "tp_packets" counts all packets handed to
1919 		 * the PF_PACKET socket, including packets dropped because
1920 		 * there wasn't room on the socket buffer - but not
1921 		 * including packets that didn't pass the filter.
1922 		 *
1923 		 * In the BSD BPF, the count of received packets is
1924 		 * incremented for every packet handed to BPF, regardless
1925 		 * of whether it passed the filter.
1926 		 *
1927 		 * We can't make "pcap_stats()" work the same on both
1928 		 * platforms, but the best approximation is to return
1929 		 * "tp_packets" as the count of packets and "tp_drops"
1930 		 * as the count of drops.
1931 		 *
1932 		 * Keep a running total because each call to
1933 		 *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
1934 		 * resets the counters to zero.
1935 		 */
1936 		handlep->stat.ps_recv += kstats.tp_packets;
1937 		handlep->stat.ps_drop += kstats.tp_drops;
1938 		*stats = handlep->stat;
1939 		return 0;
1940 	}
1941 	else
1942 	{
1943 		/*
1944 		 * If the error was EOPNOTSUPP, fall through, so that
1945 		 * if you build the library on a system with
1946 		 * "struct tpacket_stats" and run it on a system
1947 		 * that doesn't, it works as it does if the library
1948 		 * is built on a system without "struct tpacket_stats".
1949 		 */
1950 		if (errno != EOPNOTSUPP) {
1951 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1952 			    "pcap_stats: %s", pcap_strerror(errno));
1953 			return -1;
1954 		}
1955 	}
1956 #endif
1957 	/*
1958 	 * On systems where the PACKET_STATISTICS "getsockopt()" argument
1959 	 * is not supported on PF_PACKET sockets:
1960 	 *
1961 	 *	"ps_recv" counts only packets that *passed* the filter,
1962 	 *	not packets that didn't pass the filter.  It does not
1963 	 *	count packets dropped because we ran out of buffer
1964 	 *	space.
1965 	 *
1966 	 *	"ps_drop" is not supported.
1967 	 *
1968 	 *	"ps_ifdrop" is supported. It will return the number
1969 	 *	of drops the interface reports in /proc/net/dev,
1970 	 *	if that is available.
1971 	 *
1972 	 *	"ps_recv" doesn't include packets not yet read from
1973 	 *	the kernel by libpcap.
1974 	 *
1975 	 * We maintain the count of packets processed by libpcap in
1976 	 * "handlep->packets_read", for reasons described in the comment
1977 	 * at the end of pcap_read_packet().  We have no idea how many
1978 	 * packets were dropped by the kernel buffers -- but we know
1979 	 * how many the interface dropped, so we can return that.
1980 	 */
1981 
1982 	stats->ps_recv = handlep->packets_read;
1983 	stats->ps_drop = 0;
1984 	stats->ps_ifdrop = handlep->stat.ps_ifdrop;
1985 	return 0;
1986 }
1987 
1988 /*
1989  * Get from "/sys/class/net" all interfaces listed there; if they're
1990  * already in the list of interfaces we have, that won't add another
1991  * instance, but if they're not, that'll add them.
1992  *
1993  * We don't bother getting any addresses for them; it appears you can't
1994  * use SIOCGIFADDR on Linux to get IPv6 addresses for interfaces, and,
1995  * although some other types of addresses can be fetched with SIOCGIFADDR,
1996  * we don't bother with them for now.
1997  *
1998  * We also don't fail if we couldn't open "/sys/class/net"; we just leave
1999  * the list of interfaces as is, and return 0, so that we can try
2000  * scanning /proc/net/dev.
2001  */
2002 static int
2003 scan_sys_class_net(pcap_if_t **devlistp, char *errbuf)
2004 {
2005 	DIR *sys_class_net_d;
2006 	int fd;
2007 	struct dirent *ent;
2008 	char subsystem_path[PATH_MAX+1];
2009 	struct stat statb;
2010 	char *p;
2011 	char name[512];	/* XXX - pick a size */
2012 	char *q, *saveq;
2013 	struct ifreq ifrflags;
2014 	int ret = 1;
2015 
2016 	sys_class_net_d = opendir("/sys/class/net");
2017 	if (sys_class_net_d == NULL) {
2018 		/*
2019 		 * Don't fail if it doesn't exist at all.
2020 		 */
2021 		if (errno == ENOENT)
2022 			return (0);
2023 
2024 		/*
2025 		 * Fail if we got some other error.
2026 		 */
2027 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2028 		    "Can't open /sys/class/net: %s", pcap_strerror(errno));
2029 		return (-1);
2030 	}
2031 
2032 	/*
2033 	 * Create a socket from which to fetch interface information.
2034 	 */
2035 	fd = socket(AF_INET, SOCK_DGRAM, 0);
2036 	if (fd < 0) {
2037 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2038 		    "socket: %s", pcap_strerror(errno));
2039 		(void)closedir(sys_class_net_d);
2040 		return (-1);
2041 	}
2042 
2043 	for (;;) {
2044 		errno = 0;
2045 		ent = readdir(sys_class_net_d);
2046 		if (ent == NULL) {
2047 			/*
2048 			 * Error or EOF; if errno != 0, it's an error.
2049 			 */
2050 			break;
2051 		}
2052 
2053 		/*
2054 		 * Ignore "." and "..".
2055 		 */
2056 		if (strcmp(ent->d_name, ".") == 0 ||
2057 		    strcmp(ent->d_name, "..") == 0)
2058 			continue;
2059 
2060 		/*
2061 		 * Ignore plain files; they do not have subdirectories
2062 		 * and thus have no attributes.
2063 		 */
2064 		if (ent->d_type == DT_REG)
2065 			continue;
2066 
2067 		/*
2068 		 * Is there an "ifindex" file under that name?
2069 		 * (We don't care whether it's a directory or
2070 		 * a symlink; older kernels have directories
2071 		 * for devices, newer kernels have symlinks to
2072 		 * directories.)
2073 		 */
2074 		snprintf(subsystem_path, sizeof subsystem_path,
2075 		    "/sys/class/net/%s/ifindex", ent->d_name);
2076 		if (lstat(subsystem_path, &statb) != 0) {
2077 			/*
2078 			 * Stat failed.  Either there was an error
2079 			 * other than ENOENT, and we don't know if
2080 			 * this is an interface, or it's ENOENT,
2081 			 * and either some part of "/sys/class/net/{if}"
2082 			 * disappeared, in which case it probably means
2083 			 * the interface disappeared, or there's no
2084 			 * "ifindex" file, which means it's not a
2085 			 * network interface.
2086 			 */
2087 			continue;
2088 		}
2089 
2090 		/*
2091 		 * Get the interface name.
2092 		 */
2093 		p = &ent->d_name[0];
2094 		q = &name[0];
2095 		while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2096 			if (*p == ':') {
2097 				/*
2098 				 * This could be the separator between a
2099 				 * name and an alias number, or it could be
2100 				 * the separator between a name with no
2101 				 * alias number and the next field.
2102 				 *
2103 				 * If there's a colon after digits, it
2104 				 * separates the name and the alias number,
2105 				 * otherwise it separates the name and the
2106 				 * next field.
2107 				 */
2108 				saveq = q;
2109 				while (isascii(*p) && isdigit(*p))
2110 					*q++ = *p++;
2111 				if (*p != ':') {
2112 					/*
2113 					 * That was the next field,
2114 					 * not the alias number.
2115 					 */
2116 					q = saveq;
2117 				}
2118 				break;
2119 			} else
2120 				*q++ = *p++;
2121 		}
2122 		*q = '\0';
2123 
2124 		/*
2125 		 * Get the flags for this interface, and skip it if
2126 		 * it's not up.
2127 		 */
2128 		strncpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2129 		if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2130 			if (errno == ENXIO || errno == ENODEV)
2131 				continue;
2132 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2133 			    "SIOCGIFFLAGS: %.*s: %s",
2134 			    (int)sizeof(ifrflags.ifr_name),
2135 			    ifrflags.ifr_name,
2136 			    pcap_strerror(errno));
2137 			ret = -1;
2138 			break;
2139 		}
2140 		if (!(ifrflags.ifr_flags & IFF_UP))
2141 			continue;
2142 
2143 		/*
2144 		 * Add an entry for this interface, with no addresses.
2145 		 */
2146 		if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2147 		    errbuf) == -1) {
2148 			/*
2149 			 * Failure.
2150 			 */
2151 			ret = -1;
2152 			break;
2153 		}
2154 	}
2155 	if (ret != -1) {
2156 		/*
2157 		 * Well, we didn't fail for any other reason; did we
2158 		 * fail due to an error reading the directory?
2159 		 */
2160 		if (errno != 0) {
2161 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2162 			    "Error reading /sys/class/net: %s",
2163 			    pcap_strerror(errno));
2164 			ret = -1;
2165 		}
2166 	}
2167 
2168 	(void)close(fd);
2169 	(void)closedir(sys_class_net_d);
2170 	return (ret);
2171 }
2172 
2173 /*
2174  * Get from "/proc/net/dev" all interfaces listed there; if they're
2175  * already in the list of interfaces we have, that won't add another
2176  * instance, but if they're not, that'll add them.
2177  *
2178  * See comments from scan_sys_class_net().
2179  */
2180 static int
2181 scan_proc_net_dev(pcap_if_t **devlistp, char *errbuf)
2182 {
2183 	FILE *proc_net_f;
2184 	int fd;
2185 	char linebuf[512];
2186 	int linenum;
2187 	char *p;
2188 	char name[512];	/* XXX - pick a size */
2189 	char *q, *saveq;
2190 	struct ifreq ifrflags;
2191 	int ret = 0;
2192 
2193 	proc_net_f = fopen("/proc/net/dev", "r");
2194 	if (proc_net_f == NULL) {
2195 		/*
2196 		 * Don't fail if it doesn't exist at all.
2197 		 */
2198 		if (errno == ENOENT)
2199 			return (0);
2200 
2201 		/*
2202 		 * Fail if we got some other error.
2203 		 */
2204 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2205 		    "Can't open /proc/net/dev: %s", pcap_strerror(errno));
2206 		return (-1);
2207 	}
2208 
2209 	/*
2210 	 * Create a socket from which to fetch interface information.
2211 	 */
2212 	fd = socket(AF_INET, SOCK_DGRAM, 0);
2213 	if (fd < 0) {
2214 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2215 		    "socket: %s", pcap_strerror(errno));
2216 		(void)fclose(proc_net_f);
2217 		return (-1);
2218 	}
2219 
2220 	for (linenum = 1;
2221 	    fgets(linebuf, sizeof linebuf, proc_net_f) != NULL; linenum++) {
2222 		/*
2223 		 * Skip the first two lines - they're headers.
2224 		 */
2225 		if (linenum <= 2)
2226 			continue;
2227 
2228 		p = &linebuf[0];
2229 
2230 		/*
2231 		 * Skip leading white space.
2232 		 */
2233 		while (*p != '\0' && isascii(*p) && isspace(*p))
2234 			p++;
2235 		if (*p == '\0' || *p == '\n')
2236 			continue;	/* blank line */
2237 
2238 		/*
2239 		 * Get the interface name.
2240 		 */
2241 		q = &name[0];
2242 		while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2243 			if (*p == ':') {
2244 				/*
2245 				 * This could be the separator between a
2246 				 * name and an alias number, or it could be
2247 				 * the separator between a name with no
2248 				 * alias number and the next field.
2249 				 *
2250 				 * If there's a colon after digits, it
2251 				 * separates the name and the alias number,
2252 				 * otherwise it separates the name and the
2253 				 * next field.
2254 				 */
2255 				saveq = q;
2256 				while (isascii(*p) && isdigit(*p))
2257 					*q++ = *p++;
2258 				if (*p != ':') {
2259 					/*
2260 					 * That was the next field,
2261 					 * not the alias number.
2262 					 */
2263 					q = saveq;
2264 				}
2265 				break;
2266 			} else
2267 				*q++ = *p++;
2268 		}
2269 		*q = '\0';
2270 
2271 		/*
2272 		 * Get the flags for this interface, and skip it if
2273 		 * it's not up.
2274 		 */
2275 		strncpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2276 		if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2277 			if (errno == ENXIO)
2278 				continue;
2279 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2280 			    "SIOCGIFFLAGS: %.*s: %s",
2281 			    (int)sizeof(ifrflags.ifr_name),
2282 			    ifrflags.ifr_name,
2283 			    pcap_strerror(errno));
2284 			ret = -1;
2285 			break;
2286 		}
2287 		if (!(ifrflags.ifr_flags & IFF_UP))
2288 			continue;
2289 
2290 		/*
2291 		 * Add an entry for this interface, with no addresses.
2292 		 */
2293 		if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2294 		    errbuf) == -1) {
2295 			/*
2296 			 * Failure.
2297 			 */
2298 			ret = -1;
2299 			break;
2300 		}
2301 	}
2302 	if (ret != -1) {
2303 		/*
2304 		 * Well, we didn't fail for any other reason; did we
2305 		 * fail due to an error reading the file?
2306 		 */
2307 		if (ferror(proc_net_f)) {
2308 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2309 			    "Error reading /proc/net/dev: %s",
2310 			    pcap_strerror(errno));
2311 			ret = -1;
2312 		}
2313 	}
2314 
2315 	(void)close(fd);
2316 	(void)fclose(proc_net_f);
2317 	return (ret);
2318 }
2319 
2320 /*
2321  * Description string for the "any" device.
2322  */
2323 static const char any_descr[] = "Pseudo-device that captures on all interfaces";
2324 
2325 int
2326 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
2327 {
2328 	int ret;
2329 
2330 	/*
2331 	 * Read "/sys/class/net", and add to the list of interfaces all
2332 	 * interfaces listed there that we don't already have, because,
2333 	 * on Linux, SIOCGIFCONF reports only interfaces with IPv4 addresses,
2334 	 * and even getifaddrs() won't return information about
2335 	 * interfaces with no addresses, so you need to read "/sys/class/net"
2336 	 * to get the names of the rest of the interfaces.
2337 	 */
2338 	ret = scan_sys_class_net(alldevsp, errbuf);
2339 	if (ret == -1)
2340 		return (-1);	/* failed */
2341 	if (ret == 0) {
2342 		/*
2343 		 * No /sys/class/net; try reading /proc/net/dev instead.
2344 		 */
2345 		if (scan_proc_net_dev(alldevsp, errbuf) == -1)
2346 			return (-1);
2347 	}
2348 
2349 	/*
2350 	 * Add the "any" device.
2351 	 */
2352 	if (pcap_add_if(alldevsp, "any", 0, any_descr, errbuf) < 0)
2353 		return (-1);
2354 
2355 	return (0);
2356 }
2357 
2358 /*
2359  *  Attach the given BPF code to the packet capture device.
2360  */
2361 static int
2362 pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
2363     int is_mmapped)
2364 {
2365 	struct pcap_linux *handlep;
2366 #ifdef SO_ATTACH_FILTER
2367 	struct sock_fprog	fcode;
2368 	int			can_filter_in_kernel;
2369 	int			err = 0;
2370 #endif
2371 
2372 	if (!handle)
2373 		return -1;
2374 	if (!filter) {
2375 	        strncpy(handle->errbuf, "setfilter: No filter specified",
2376 			PCAP_ERRBUF_SIZE);
2377 		return -1;
2378 	}
2379 
2380 	handlep = handle->priv;
2381 
2382 	/* Make our private copy of the filter */
2383 
2384 	if (install_bpf_program(handle, filter) < 0)
2385 		/* install_bpf_program() filled in errbuf */
2386 		return -1;
2387 
2388 	/*
2389 	 * Run user level packet filter by default. Will be overriden if
2390 	 * installing a kernel filter succeeds.
2391 	 */
2392 	handlep->filtering_in_kernel = 0;
2393 
2394 	/* Install kernel level filter if possible */
2395 
2396 #ifdef SO_ATTACH_FILTER
2397 #ifdef USHRT_MAX
2398 	if (handle->fcode.bf_len > USHRT_MAX) {
2399 		/*
2400 		 * fcode.len is an unsigned short for current kernel.
2401 		 * I have yet to see BPF-Code with that much
2402 		 * instructions but still it is possible. So for the
2403 		 * sake of correctness I added this check.
2404 		 */
2405 		fprintf(stderr, "Warning: Filter too complex for kernel\n");
2406 		fcode.len = 0;
2407 		fcode.filter = NULL;
2408 		can_filter_in_kernel = 0;
2409 	} else
2410 #endif /* USHRT_MAX */
2411 	{
2412 		/*
2413 		 * Oh joy, the Linux kernel uses struct sock_fprog instead
2414 		 * of struct bpf_program and of course the length field is
2415 		 * of different size. Pointed out by Sebastian
2416 		 *
2417 		 * Oh, and we also need to fix it up so that all "ret"
2418 		 * instructions with non-zero operands have 65535 as the
2419 		 * operand if we're not capturing in memory-mapped modee,
2420 		 * and so that, if we're in cooked mode, all memory-reference
2421 		 * instructions use special magic offsets in references to
2422 		 * the link-layer header and assume that the link-layer
2423 		 * payload begins at 0; "fix_program()" will do that.
2424 		 */
2425 		switch (fix_program(handle, &fcode, is_mmapped)) {
2426 
2427 		case -1:
2428 		default:
2429 			/*
2430 			 * Fatal error; just quit.
2431 			 * (The "default" case shouldn't happen; we
2432 			 * return -1 for that reason.)
2433 			 */
2434 			return -1;
2435 
2436 		case 0:
2437 			/*
2438 			 * The program performed checks that we can't make
2439 			 * work in the kernel.
2440 			 */
2441 			can_filter_in_kernel = 0;
2442 			break;
2443 
2444 		case 1:
2445 			/*
2446 			 * We have a filter that'll work in the kernel.
2447 			 */
2448 			can_filter_in_kernel = 1;
2449 			break;
2450 		}
2451 	}
2452 
2453 	/*
2454 	 * NOTE: at this point, we've set both the "len" and "filter"
2455 	 * fields of "fcode".  As of the 2.6.32.4 kernel, at least,
2456 	 * those are the only members of the "sock_fprog" structure,
2457 	 * so we initialize every member of that structure.
2458 	 *
2459 	 * If there is anything in "fcode" that is not initialized,
2460 	 * it is either a field added in a later kernel, or it's
2461 	 * padding.
2462 	 *
2463 	 * If a new field is added, this code needs to be updated
2464 	 * to set it correctly.
2465 	 *
2466 	 * If there are no other fields, then:
2467 	 *
2468 	 *	if the Linux kernel looks at the padding, it's
2469 	 *	buggy;
2470 	 *
2471 	 *	if the Linux kernel doesn't look at the padding,
2472 	 *	then if some tool complains that we're passing
2473 	 *	uninitialized data to the kernel, then the tool
2474 	 *	is buggy and needs to understand that it's just
2475 	 *	padding.
2476 	 */
2477 	if (can_filter_in_kernel) {
2478 		if ((err = set_kernel_filter(handle, &fcode)) == 0)
2479 		{
2480 			/* Installation succeded - using kernel filter. */
2481 			handlep->filtering_in_kernel = 1;
2482 		}
2483 		else if (err == -1)	/* Non-fatal error */
2484 		{
2485 			/*
2486 			 * Print a warning if we weren't able to install
2487 			 * the filter for a reason other than "this kernel
2488 			 * isn't configured to support socket filters.
2489 			 */
2490 			if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
2491 				fprintf(stderr,
2492 				    "Warning: Kernel filter failed: %s\n",
2493 					pcap_strerror(errno));
2494 			}
2495 		}
2496 	}
2497 
2498 	/*
2499 	 * If we're not using the kernel filter, get rid of any kernel
2500 	 * filter that might've been there before, e.g. because the
2501 	 * previous filter could work in the kernel, or because some other
2502 	 * code attached a filter to the socket by some means other than
2503 	 * calling "pcap_setfilter()".  Otherwise, the kernel filter may
2504 	 * filter out packets that would pass the new userland filter.
2505 	 */
2506 	if (!handlep->filtering_in_kernel)
2507 		reset_kernel_filter(handle);
2508 
2509 	/*
2510 	 * Free up the copy of the filter that was made by "fix_program()".
2511 	 */
2512 	if (fcode.filter != NULL)
2513 		free(fcode.filter);
2514 
2515 	if (err == -2)
2516 		/* Fatal error */
2517 		return -1;
2518 #endif /* SO_ATTACH_FILTER */
2519 
2520 	return 0;
2521 }
2522 
2523 static int
2524 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
2525 {
2526 	return pcap_setfilter_linux_common(handle, filter, 0);
2527 }
2528 
2529 
2530 /*
2531  * Set direction flag: Which packets do we accept on a forwarding
2532  * single device? IN, OUT or both?
2533  */
2534 static int
2535 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
2536 {
2537 #ifdef HAVE_PF_PACKET_SOCKETS
2538 	struct pcap_linux *handlep = handle->priv;
2539 
2540 	if (!handlep->sock_packet) {
2541 		handle->direction = d;
2542 		return 0;
2543 	}
2544 #endif
2545 	/*
2546 	 * We're not using PF_PACKET sockets, so we can't determine
2547 	 * the direction of the packet.
2548 	 */
2549 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2550 	    "Setting direction is not supported on SOCK_PACKET sockets");
2551 	return -1;
2552 }
2553 
2554 #ifdef HAVE_PF_PACKET_SOCKETS
2555 /*
2556  * Map the PACKET_ value to a LINUX_SLL_ value; we
2557  * want the same numerical value to be used in
2558  * the link-layer header even if the numerical values
2559  * for the PACKET_ #defines change, so that programs
2560  * that look at the packet type field will always be
2561  * able to handle DLT_LINUX_SLL captures.
2562  */
2563 static short int
2564 map_packet_type_to_sll_type(short int sll_pkttype)
2565 {
2566 	switch (sll_pkttype) {
2567 
2568 	case PACKET_HOST:
2569 		return htons(LINUX_SLL_HOST);
2570 
2571 	case PACKET_BROADCAST:
2572 		return htons(LINUX_SLL_BROADCAST);
2573 
2574 	case PACKET_MULTICAST:
2575 		return  htons(LINUX_SLL_MULTICAST);
2576 
2577 	case PACKET_OTHERHOST:
2578 		return htons(LINUX_SLL_OTHERHOST);
2579 
2580 	case PACKET_OUTGOING:
2581 		return htons(LINUX_SLL_OUTGOING);
2582 
2583 	default:
2584 		return -1;
2585 	}
2586 }
2587 #endif
2588 
2589 /*
2590  *  Linux uses the ARP hardware type to identify the type of an
2591  *  interface. pcap uses the DLT_xxx constants for this. This
2592  *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
2593  *  constant, as arguments, and sets "handle->linktype" to the
2594  *  appropriate DLT_XXX constant and sets "handle->offset" to
2595  *  the appropriate value (to make "handle->offset" plus link-layer
2596  *  header length be a multiple of 4, so that the link-layer payload
2597  *  will be aligned on a 4-byte boundary when capturing packets).
2598  *  (If the offset isn't set here, it'll be 0; add code as appropriate
2599  *  for cases where it shouldn't be 0.)
2600  *
2601  *  If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
2602  *  in cooked mode; otherwise, we can't use cooked mode, so we have
2603  *  to pick some type that works in raw mode, or fail.
2604  *
2605  *  Sets the link type to -1 if unable to map the type.
2606  */
2607 static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
2608 {
2609 	switch (arptype) {
2610 
2611 	case ARPHRD_ETHER:
2612 		/*
2613 		 * This is (presumably) a real Ethernet capture; give it a
2614 		 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
2615 		 * that an application can let you choose it, in case you're
2616 		 * capturing DOCSIS traffic that a Cisco Cable Modem
2617 		 * Termination System is putting out onto an Ethernet (it
2618 		 * doesn't put an Ethernet header onto the wire, it puts raw
2619 		 * DOCSIS frames out on the wire inside the low-level
2620 		 * Ethernet framing).
2621 		 *
2622 		 * XXX - are there any sorts of "fake Ethernet" that have
2623 		 * ARPHRD_ETHER but that *shouldn't offer DLT_DOCSIS as
2624 		 * a Cisco CMTS won't put traffic onto it or get traffic
2625 		 * bridged onto it?  ISDN is handled in "activate_new()",
2626 		 * as we fall back on cooked mode there; are there any
2627 		 * others?
2628 		 */
2629 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2630 		/*
2631 		 * If that fails, just leave the list empty.
2632 		 */
2633 		if (handle->dlt_list != NULL) {
2634 			handle->dlt_list[0] = DLT_EN10MB;
2635 			handle->dlt_list[1] = DLT_DOCSIS;
2636 			handle->dlt_count = 2;
2637 		}
2638 		/* FALLTHROUGH */
2639 
2640 	case ARPHRD_METRICOM:
2641 	case ARPHRD_LOOPBACK:
2642 		handle->linktype = DLT_EN10MB;
2643 		handle->offset = 2;
2644 		break;
2645 
2646 	case ARPHRD_EETHER:
2647 		handle->linktype = DLT_EN3MB;
2648 		break;
2649 
2650 	case ARPHRD_AX25:
2651 		handle->linktype = DLT_AX25_KISS;
2652 		break;
2653 
2654 	case ARPHRD_PRONET:
2655 		handle->linktype = DLT_PRONET;
2656 		break;
2657 
2658 	case ARPHRD_CHAOS:
2659 		handle->linktype = DLT_CHAOS;
2660 		break;
2661 #ifndef ARPHRD_CAN
2662 #define ARPHRD_CAN 280
2663 #endif
2664 	case ARPHRD_CAN:
2665 		handle->linktype = DLT_CAN_SOCKETCAN;
2666 		break;
2667 
2668 #ifndef ARPHRD_IEEE802_TR
2669 #define ARPHRD_IEEE802_TR 800	/* From Linux 2.4 */
2670 #endif
2671 	case ARPHRD_IEEE802_TR:
2672 	case ARPHRD_IEEE802:
2673 		handle->linktype = DLT_IEEE802;
2674 		handle->offset = 2;
2675 		break;
2676 
2677 	case ARPHRD_ARCNET:
2678 		handle->linktype = DLT_ARCNET_LINUX;
2679 		break;
2680 
2681 #ifndef ARPHRD_FDDI	/* From Linux 2.2.13 */
2682 #define ARPHRD_FDDI	774
2683 #endif
2684 	case ARPHRD_FDDI:
2685 		handle->linktype = DLT_FDDI;
2686 		handle->offset = 3;
2687 		break;
2688 
2689 #ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
2690 #define ARPHRD_ATM 19
2691 #endif
2692 	case ARPHRD_ATM:
2693 		/*
2694 		 * The Classical IP implementation in ATM for Linux
2695 		 * supports both what RFC 1483 calls "LLC Encapsulation",
2696 		 * in which each packet has an LLC header, possibly
2697 		 * with a SNAP header as well, prepended to it, and
2698 		 * what RFC 1483 calls "VC Based Multiplexing", in which
2699 		 * different virtual circuits carry different network
2700 		 * layer protocols, and no header is prepended to packets.
2701 		 *
2702 		 * They both have an ARPHRD_ type of ARPHRD_ATM, so
2703 		 * you can't use the ARPHRD_ type to find out whether
2704 		 * captured packets will have an LLC header, and,
2705 		 * while there's a socket ioctl to *set* the encapsulation
2706 		 * type, there's no ioctl to *get* the encapsulation type.
2707 		 *
2708 		 * This means that
2709 		 *
2710 		 *	programs that dissect Linux Classical IP frames
2711 		 *	would have to check for an LLC header and,
2712 		 *	depending on whether they see one or not, dissect
2713 		 *	the frame as LLC-encapsulated or as raw IP (I
2714 		 *	don't know whether there's any traffic other than
2715 		 *	IP that would show up on the socket, or whether
2716 		 *	there's any support for IPv6 in the Linux
2717 		 *	Classical IP code);
2718 		 *
2719 		 *	filter expressions would have to compile into
2720 		 *	code that checks for an LLC header and does
2721 		 *	the right thing.
2722 		 *
2723 		 * Both of those are a nuisance - and, at least on systems
2724 		 * that support PF_PACKET sockets, we don't have to put
2725 		 * up with those nuisances; instead, we can just capture
2726 		 * in cooked mode.  That's what we'll do, if we can.
2727 		 * Otherwise, we'll just fail.
2728 		 */
2729 		if (cooked_ok)
2730 			handle->linktype = DLT_LINUX_SLL;
2731 		else
2732 			handle->linktype = -1;
2733 		break;
2734 
2735 #ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
2736 #define ARPHRD_IEEE80211 801
2737 #endif
2738 	case ARPHRD_IEEE80211:
2739 		handle->linktype = DLT_IEEE802_11;
2740 		break;
2741 
2742 #ifndef ARPHRD_IEEE80211_PRISM  /* From Linux 2.4.18 */
2743 #define ARPHRD_IEEE80211_PRISM 802
2744 #endif
2745 	case ARPHRD_IEEE80211_PRISM:
2746 		handle->linktype = DLT_PRISM_HEADER;
2747 		break;
2748 
2749 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
2750 #define ARPHRD_IEEE80211_RADIOTAP 803
2751 #endif
2752 	case ARPHRD_IEEE80211_RADIOTAP:
2753 		handle->linktype = DLT_IEEE802_11_RADIO;
2754 		break;
2755 
2756 	case ARPHRD_PPP:
2757 		/*
2758 		 * Some PPP code in the kernel supplies no link-layer
2759 		 * header whatsoever to PF_PACKET sockets; other PPP
2760 		 * code supplies PPP link-layer headers ("syncppp.c");
2761 		 * some PPP code might supply random link-layer
2762 		 * headers (PPP over ISDN - there's code in Ethereal,
2763 		 * for example, to cope with PPP-over-ISDN captures
2764 		 * with which the Ethereal developers have had to cope,
2765 		 * heuristically trying to determine which of the
2766 		 * oddball link-layer headers particular packets have).
2767 		 *
2768 		 * As such, we just punt, and run all PPP interfaces
2769 		 * in cooked mode, if we can; otherwise, we just treat
2770 		 * it as DLT_RAW, for now - if somebody needs to capture,
2771 		 * on a 2.0[.x] kernel, on PPP devices that supply a
2772 		 * link-layer header, they'll have to add code here to
2773 		 * map to the appropriate DLT_ type (possibly adding a
2774 		 * new DLT_ type, if necessary).
2775 		 */
2776 		if (cooked_ok)
2777 			handle->linktype = DLT_LINUX_SLL;
2778 		else {
2779 			/*
2780 			 * XXX - handle ISDN types here?  We can't fall
2781 			 * back on cooked sockets, so we'd have to
2782 			 * figure out from the device name what type of
2783 			 * link-layer encapsulation it's using, and map
2784 			 * that to an appropriate DLT_ value, meaning
2785 			 * we'd map "isdnN" devices to DLT_RAW (they
2786 			 * supply raw IP packets with no link-layer
2787 			 * header) and "isdY" devices to a new DLT_I4L_IP
2788 			 * type that has only an Ethernet packet type as
2789 			 * a link-layer header.
2790 			 *
2791 			 * But sometimes we seem to get random crap
2792 			 * in the link-layer header when capturing on
2793 			 * ISDN devices....
2794 			 */
2795 			handle->linktype = DLT_RAW;
2796 		}
2797 		break;
2798 
2799 #ifndef ARPHRD_CISCO
2800 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
2801 #endif
2802 	case ARPHRD_CISCO:
2803 		handle->linktype = DLT_C_HDLC;
2804 		break;
2805 
2806 	/* Not sure if this is correct for all tunnels, but it
2807 	 * works for CIPE */
2808 	case ARPHRD_TUNNEL:
2809 #ifndef ARPHRD_SIT
2810 #define ARPHRD_SIT 776	/* From Linux 2.2.13 */
2811 #endif
2812 	case ARPHRD_SIT:
2813 	case ARPHRD_CSLIP:
2814 	case ARPHRD_SLIP6:
2815 	case ARPHRD_CSLIP6:
2816 	case ARPHRD_ADAPT:
2817 	case ARPHRD_SLIP:
2818 #ifndef ARPHRD_RAWHDLC
2819 #define ARPHRD_RAWHDLC 518
2820 #endif
2821 	case ARPHRD_RAWHDLC:
2822 #ifndef ARPHRD_DLCI
2823 #define ARPHRD_DLCI 15
2824 #endif
2825 	case ARPHRD_DLCI:
2826 		/*
2827 		 * XXX - should some of those be mapped to DLT_LINUX_SLL
2828 		 * instead?  Should we just map all of them to DLT_LINUX_SLL?
2829 		 */
2830 		handle->linktype = DLT_RAW;
2831 		break;
2832 
2833 #ifndef ARPHRD_FRAD
2834 #define ARPHRD_FRAD 770
2835 #endif
2836 	case ARPHRD_FRAD:
2837 		handle->linktype = DLT_FRELAY;
2838 		break;
2839 
2840 	case ARPHRD_LOCALTLK:
2841 		handle->linktype = DLT_LTALK;
2842 		break;
2843 
2844 	case 18:
2845 		/*
2846 		 * RFC 4338 defines an encapsulation for IP and ARP
2847 		 * packets that's compatible with the RFC 2625
2848 		 * encapsulation, but that uses a different ARP
2849 		 * hardware type and hardware addresses.  That
2850 		 * ARP hardware type is 18; Linux doesn't define
2851 		 * any ARPHRD_ value as 18, but if it ever officially
2852 		 * supports RFC 4338-style IP-over-FC, it should define
2853 		 * one.
2854 		 *
2855 		 * For now, we map it to DLT_IP_OVER_FC, in the hopes
2856 		 * that this will encourage its use in the future,
2857 		 * should Linux ever officially support RFC 4338-style
2858 		 * IP-over-FC.
2859 		 */
2860 		handle->linktype = DLT_IP_OVER_FC;
2861 		break;
2862 
2863 #ifndef ARPHRD_FCPP
2864 #define ARPHRD_FCPP	784
2865 #endif
2866 	case ARPHRD_FCPP:
2867 #ifndef ARPHRD_FCAL
2868 #define ARPHRD_FCAL	785
2869 #endif
2870 	case ARPHRD_FCAL:
2871 #ifndef ARPHRD_FCPL
2872 #define ARPHRD_FCPL	786
2873 #endif
2874 	case ARPHRD_FCPL:
2875 #ifndef ARPHRD_FCFABRIC
2876 #define ARPHRD_FCFABRIC	787
2877 #endif
2878 	case ARPHRD_FCFABRIC:
2879 		/*
2880 		 * Back in 2002, Donald Lee at Cray wanted a DLT_ for
2881 		 * IP-over-FC:
2882 		 *
2883 		 *	http://www.mail-archive.com/tcpdump-workers@sandelman.ottawa.on.ca/msg01043.html
2884 		 *
2885 		 * and one was assigned.
2886 		 *
2887 		 * In a later private discussion (spun off from a message
2888 		 * on the ethereal-users list) on how to get that DLT_
2889 		 * value in libpcap on Linux, I ended up deciding that
2890 		 * the best thing to do would be to have him tweak the
2891 		 * driver to set the ARPHRD_ value to some ARPHRD_FCxx
2892 		 * type, and map all those types to DLT_IP_OVER_FC:
2893 		 *
2894 		 *	I've checked into the libpcap and tcpdump CVS tree
2895 		 *	support for DLT_IP_OVER_FC.  In order to use that,
2896 		 *	you'd have to modify your modified driver to return
2897 		 *	one of the ARPHRD_FCxxx types, in "fcLINUXfcp.c" -
2898 		 *	change it to set "dev->type" to ARPHRD_FCFABRIC, for
2899 		 *	example (the exact value doesn't matter, it can be
2900 		 *	any of ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, or
2901 		 *	ARPHRD_FCFABRIC).
2902 		 *
2903 		 * 11 years later, Christian Svensson wanted to map
2904 		 * various ARPHRD_ values to DLT_FC_2 and
2905 		 * DLT_FC_2_WITH_FRAME_DELIMS for raw Fibre Channel
2906 		 * frames:
2907 		 *
2908 		 *	https://github.com/mcr/libpcap/pull/29
2909 		 *
2910 		 * There doesn't seem to be any network drivers that uses
2911 		 * any of the ARPHRD_FC* values for IP-over-FC, and
2912 		 * it's not exactly clear what the "Dummy types for non
2913 		 * ARP hardware" are supposed to mean (link-layer
2914 		 * header type?  Physical network type?), so it's
2915 		 * not exactly clear why the ARPHRD_FC* types exist
2916 		 * in the first place.
2917 		 *
2918 		 * For now, we map them to DLT_FC_2, and provide an
2919 		 * option of DLT_FC_2_WITH_FRAME_DELIMS, as well as
2920 		 * DLT_IP_OVER_FC just in case there's some old
2921 		 * driver out there that uses one of those types for
2922 		 * IP-over-FC on which somebody wants to capture
2923 		 * packets.
2924 		 */
2925 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2926 		/*
2927 		 * If that fails, just leave the list empty.
2928 		 */
2929 		if (handle->dlt_list != NULL) {
2930 			handle->dlt_list[0] = DLT_FC_2;
2931 			handle->dlt_list[1] = DLT_FC_2_WITH_FRAME_DELIMS;
2932 			handle->dlt_list[2] = DLT_IP_OVER_FC;
2933 			handle->dlt_count = 3;
2934 		}
2935 		handle->linktype = DLT_FC_2;
2936 		break;
2937 
2938 #ifndef ARPHRD_IRDA
2939 #define ARPHRD_IRDA	783
2940 #endif
2941 	case ARPHRD_IRDA:
2942 		/* Don't expect IP packet out of this interfaces... */
2943 		handle->linktype = DLT_LINUX_IRDA;
2944 		/* We need to save packet direction for IrDA decoding,
2945 		 * so let's use "Linux-cooked" mode. Jean II */
2946 		//handlep->cooked = 1;
2947 		break;
2948 
2949 	/* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
2950 	 * is needed, please report it to <daniele@orlandi.com> */
2951 #ifndef ARPHRD_LAPD
2952 #define ARPHRD_LAPD	8445
2953 #endif
2954 	case ARPHRD_LAPD:
2955 		/* Don't expect IP packet out of this interfaces... */
2956 		handle->linktype = DLT_LINUX_LAPD;
2957 		break;
2958 
2959 #ifndef ARPHRD_NONE
2960 #define ARPHRD_NONE	0xFFFE
2961 #endif
2962 	case ARPHRD_NONE:
2963 		/*
2964 		 * No link-layer header; packets are just IP
2965 		 * packets, so use DLT_RAW.
2966 		 */
2967 		handle->linktype = DLT_RAW;
2968 		break;
2969 
2970 #ifndef ARPHRD_IEEE802154
2971 #define ARPHRD_IEEE802154      804
2972 #endif
2973        case ARPHRD_IEEE802154:
2974                handle->linktype =  DLT_IEEE802_15_4_NOFCS;
2975                break;
2976 
2977 	default:
2978 		handle->linktype = -1;
2979 		break;
2980 	}
2981 }
2982 
2983 /* ===== Functions to interface to the newer kernels ================== */
2984 
2985 /*
2986  * Try to open a packet socket using the new kernel PF_PACKET interface.
2987  * Returns 1 on success, 0 on an error that means the new interface isn't
2988  * present (so the old SOCK_PACKET interface should be tried), and a
2989  * PCAP_ERROR_ value on an error that means that the old mechanism won't
2990  * work either (so it shouldn't be tried).
2991  */
2992 static int
2993 activate_new(pcap_t *handle)
2994 {
2995 #ifdef HAVE_PF_PACKET_SOCKETS
2996 	struct pcap_linux *handlep = handle->priv;
2997 	const char		*device = handle->opt.source;
2998 	int			is_any_device = (strcmp(device, "any") == 0);
2999 	int			sock_fd = -1, arptype;
3000 #ifdef HAVE_PACKET_AUXDATA
3001 	int			val;
3002 #endif
3003 	int			err = 0;
3004 	struct packet_mreq	mr;
3005 
3006 	/*
3007 	 * Open a socket with protocol family packet. If the
3008 	 * "any" device was specified, we open a SOCK_DGRAM
3009 	 * socket for the cooked interface, otherwise we first
3010 	 * try a SOCK_RAW socket for the raw interface.
3011 	 */
3012 	sock_fd = is_any_device ?
3013 		socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) :
3014 		socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
3015 
3016 	if (sock_fd == -1) {
3017 		if (errno == EINVAL || errno == EAFNOSUPPORT) {
3018 			/*
3019 			 * We don't support PF_PACKET/SOCK_whatever
3020 			 * sockets; try the old mechanism.
3021 			 */
3022 			return 0;
3023 		}
3024 
3025 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
3026 			 pcap_strerror(errno) );
3027 		if (errno == EPERM || errno == EACCES) {
3028 			/*
3029 			 * You don't have permission to open the
3030 			 * socket.
3031 			 */
3032 			return PCAP_ERROR_PERM_DENIED;
3033 		} else {
3034 			/*
3035 			 * Other error.
3036 			 */
3037 			return PCAP_ERROR;
3038 		}
3039 	}
3040 
3041 	/* It seems the kernel supports the new interface. */
3042 	handlep->sock_packet = 0;
3043 
3044 	/*
3045 	 * Get the interface index of the loopback device.
3046 	 * If the attempt fails, don't fail, just set the
3047 	 * "handlep->lo_ifindex" to -1.
3048 	 *
3049 	 * XXX - can there be more than one device that loops
3050 	 * packets back, i.e. devices other than "lo"?  If so,
3051 	 * we'd need to find them all, and have an array of
3052 	 * indices for them, and check all of them in
3053 	 * "pcap_read_packet()".
3054 	 */
3055 	handlep->lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
3056 
3057 	/*
3058 	 * Default value for offset to align link-layer payload
3059 	 * on a 4-byte boundary.
3060 	 */
3061 	handle->offset	 = 0;
3062 
3063 	/*
3064 	 * What kind of frames do we have to deal with? Fall back
3065 	 * to cooked mode if we have an unknown interface type
3066 	 * or a type we know doesn't work well in raw mode.
3067 	 */
3068 	if (!is_any_device) {
3069 		/* Assume for now we don't need cooked mode. */
3070 		handlep->cooked = 0;
3071 
3072 		if (handle->opt.rfmon) {
3073 			/*
3074 			 * We were asked to turn on monitor mode.
3075 			 * Do so before we get the link-layer type,
3076 			 * because entering monitor mode could change
3077 			 * the link-layer type.
3078 			 */
3079 			err = enter_rfmon_mode(handle, sock_fd, device);
3080 			if (err < 0) {
3081 				/* Hard failure */
3082 				close(sock_fd);
3083 				return err;
3084 			}
3085 			if (err == 0) {
3086 				/*
3087 				 * Nothing worked for turning monitor mode
3088 				 * on.
3089 				 */
3090 				close(sock_fd);
3091 				return PCAP_ERROR_RFMON_NOTSUP;
3092 			}
3093 
3094 			/*
3095 			 * Either monitor mode has been turned on for
3096 			 * the device, or we've been given a different
3097 			 * device to open for monitor mode.  If we've
3098 			 * been given a different device, use it.
3099 			 */
3100 			if (handlep->mondevice != NULL)
3101 				device = handlep->mondevice;
3102 		}
3103 		arptype	= iface_get_arptype(sock_fd, device, handle->errbuf);
3104 		if (arptype < 0) {
3105 			close(sock_fd);
3106 			return arptype;
3107 		}
3108 		map_arphrd_to_dlt(handle, arptype, 1);
3109 		if (handle->linktype == -1 ||
3110 		    handle->linktype == DLT_LINUX_SLL ||
3111 		    handle->linktype == DLT_LINUX_IRDA ||
3112 		    handle->linktype == DLT_LINUX_LAPD ||
3113 		    (handle->linktype == DLT_EN10MB &&
3114 		     (strncmp("isdn", device, 4) == 0 ||
3115 		      strncmp("isdY", device, 4) == 0))) {
3116 			/*
3117 			 * Unknown interface type (-1), or a
3118 			 * device we explicitly chose to run
3119 			 * in cooked mode (e.g., PPP devices),
3120 			 * or an ISDN device (whose link-layer
3121 			 * type we can only determine by using
3122 			 * APIs that may be different on different
3123 			 * kernels) - reopen in cooked mode.
3124 			 */
3125 			if (close(sock_fd) == -1) {
3126 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3127 					 "close: %s", pcap_strerror(errno));
3128 				return PCAP_ERROR;
3129 			}
3130 			sock_fd = socket(PF_PACKET, SOCK_DGRAM,
3131 			    htons(ETH_P_ALL));
3132 			if (sock_fd == -1) {
3133 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3134 				    "socket: %s", pcap_strerror(errno));
3135 				if (errno == EPERM || errno == EACCES) {
3136 					/*
3137 					 * You don't have permission to
3138 					 * open the socket.
3139 					 */
3140 					return PCAP_ERROR_PERM_DENIED;
3141 				} else {
3142 					/*
3143 					 * Other error.
3144 					 */
3145 					return PCAP_ERROR;
3146 				}
3147 			}
3148 			handlep->cooked = 1;
3149 
3150 			/*
3151 			 * Get rid of any link-layer type list
3152 			 * we allocated - this only supports cooked
3153 			 * capture.
3154 			 */
3155 			if (handle->dlt_list != NULL) {
3156 				free(handle->dlt_list);
3157 				handle->dlt_list = NULL;
3158 				handle->dlt_count = 0;
3159 			}
3160 
3161 			if (handle->linktype == -1) {
3162 				/*
3163 				 * Warn that we're falling back on
3164 				 * cooked mode; we may want to
3165 				 * update "map_arphrd_to_dlt()"
3166 				 * to handle the new type.
3167 				 */
3168 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3169 					"arptype %d not "
3170 					"supported by libpcap - "
3171 					"falling back to cooked "
3172 					"socket",
3173 					arptype);
3174 			}
3175 
3176 			/*
3177 			 * IrDA capture is not a real "cooked" capture,
3178 			 * it's IrLAP frames, not IP packets.  The
3179 			 * same applies to LAPD capture.
3180 			 */
3181 			if (handle->linktype != DLT_LINUX_IRDA &&
3182 			    handle->linktype != DLT_LINUX_LAPD)
3183 				handle->linktype = DLT_LINUX_SLL;
3184 		}
3185 
3186 		handlep->ifindex = iface_get_id(sock_fd, device,
3187 		    handle->errbuf);
3188 		if (handlep->ifindex == -1) {
3189 			close(sock_fd);
3190 			return PCAP_ERROR;
3191 		}
3192 
3193 		if ((err = iface_bind(sock_fd, handlep->ifindex,
3194 		    handle->errbuf)) != 1) {
3195 		    	close(sock_fd);
3196 			if (err < 0)
3197 				return err;
3198 			else
3199 				return 0;	/* try old mechanism */
3200 		}
3201 	} else {
3202 		/*
3203 		 * The "any" device.
3204 		 */
3205 		if (handle->opt.rfmon) {
3206 			/*
3207 			 * It doesn't support monitor mode.
3208 			 */
3209 			return PCAP_ERROR_RFMON_NOTSUP;
3210 		}
3211 
3212 		/*
3213 		 * It uses cooked mode.
3214 		 */
3215 		handlep->cooked = 1;
3216 		handle->linktype = DLT_LINUX_SLL;
3217 
3218 		/*
3219 		 * We're not bound to a device.
3220 		 * For now, we're using this as an indication
3221 		 * that we can't transmit; stop doing that only
3222 		 * if we figure out how to transmit in cooked
3223 		 * mode.
3224 		 */
3225 		handlep->ifindex = -1;
3226 	}
3227 
3228 	/*
3229 	 * Select promiscuous mode on if "promisc" is set.
3230 	 *
3231 	 * Do not turn allmulti mode on if we don't select
3232 	 * promiscuous mode - on some devices (e.g., Orinoco
3233 	 * wireless interfaces), allmulti mode isn't supported
3234 	 * and the driver implements it by turning promiscuous
3235 	 * mode on, and that screws up the operation of the
3236 	 * card as a normal networking interface, and on no
3237 	 * other platform I know of does starting a non-
3238 	 * promiscuous capture affect which multicast packets
3239 	 * are received by the interface.
3240 	 */
3241 
3242 	/*
3243 	 * Hmm, how can we set promiscuous mode on all interfaces?
3244 	 * I am not sure if that is possible at all.  For now, we
3245 	 * silently ignore attempts to turn promiscuous mode on
3246 	 * for the "any" device (so you don't have to explicitly
3247 	 * disable it in programs such as tcpdump).
3248 	 */
3249 
3250 	if (!is_any_device && handle->opt.promisc) {
3251 		memset(&mr, 0, sizeof(mr));
3252 		mr.mr_ifindex = handlep->ifindex;
3253 		mr.mr_type    = PACKET_MR_PROMISC;
3254 		if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
3255 		    &mr, sizeof(mr)) == -1) {
3256 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3257 				"setsockopt: %s", pcap_strerror(errno));
3258 			close(sock_fd);
3259 			return PCAP_ERROR;
3260 		}
3261 	}
3262 
3263 	/* Enable auxillary data if supported and reserve room for
3264 	 * reconstructing VLAN headers. */
3265 #ifdef HAVE_PACKET_AUXDATA
3266 	val = 1;
3267 	if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
3268 		       sizeof(val)) == -1 && errno != ENOPROTOOPT) {
3269 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3270 			 "setsockopt: %s", pcap_strerror(errno));
3271 		close(sock_fd);
3272 		return PCAP_ERROR;
3273 	}
3274 	handle->offset += VLAN_TAG_LEN;
3275 #endif /* HAVE_PACKET_AUXDATA */
3276 
3277 	/*
3278 	 * This is a 2.2[.x] or later kernel (we know that
3279 	 * because we're not using a SOCK_PACKET socket -
3280 	 * PF_PACKET is supported only in 2.2 and later
3281 	 * kernels).
3282 	 *
3283 	 * We can safely pass "recvfrom()" a byte count
3284 	 * based on the snapshot length.
3285 	 *
3286 	 * If we're in cooked mode, make the snapshot length
3287 	 * large enough to hold a "cooked mode" header plus
3288 	 * 1 byte of packet data (so we don't pass a byte
3289 	 * count of 0 to "recvfrom()").
3290 	 */
3291 	if (handlep->cooked) {
3292 		if (handle->snapshot < SLL_HDR_LEN + 1)
3293 			handle->snapshot = SLL_HDR_LEN + 1;
3294 	}
3295 	handle->bufsize = handle->snapshot;
3296 
3297 	/*
3298 	 * Set the offset at which to insert VLAN tags.
3299 	 */
3300 	switch (handle->linktype) {
3301 
3302 	case DLT_EN10MB:
3303 		handlep->vlan_offset = 2 * ETH_ALEN;
3304 		break;
3305 
3306 	case DLT_LINUX_SLL:
3307 		handlep->vlan_offset = 14;
3308 		break;
3309 
3310 	default:
3311 		handlep->vlan_offset = -1; /* unknown */
3312 		break;
3313 	}
3314 
3315 	/* Save the socket FD in the pcap structure */
3316 	handle->fd = sock_fd;
3317 
3318 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
3319 	if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
3320 		int nsec_tstamps = 1;
3321 
3322 		if (setsockopt(handle->fd, SOL_SOCKET, SO_TIMESTAMPNS, &nsec_tstamps, sizeof(nsec_tstamps)) < 0) {
3323 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "setsockopt: unable to set SO_TIMESTAMPNS");
3324 			return PCAP_ERROR;
3325 		}
3326 	}
3327 #endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */
3328 
3329 	return 1;
3330 #else /* HAVE_PF_PACKET_SOCKETS */
3331 	strncpy(ebuf,
3332 		"New packet capturing interface not supported by build "
3333 		"environment", PCAP_ERRBUF_SIZE);
3334 	return 0;
3335 #endif /* HAVE_PF_PACKET_SOCKETS */
3336 }
3337 
3338 #ifdef HAVE_PACKET_RING
3339 /*
3340  * Attempt to activate with memory-mapped access.
3341  *
3342  * On success, returns 1, and sets *status to 0 if there are no warnings
3343  * or to a PCAP_WARNING_ code if there is a warning.
3344  *
3345  * On failure due to lack of support for memory-mapped capture, returns
3346  * 0.
3347  *
3348  * On error, returns -1, and sets *status to the appropriate error code;
3349  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3350  */
3351 static int
3352 activate_mmap(pcap_t *handle, int *status)
3353 {
3354 	struct pcap_linux *handlep = handle->priv;
3355 	int ret;
3356 
3357 	/*
3358 	 * Attempt to allocate a buffer to hold the contents of one
3359 	 * packet, for use by the oneshot callback.
3360 	 */
3361 	handlep->oneshot_buffer = malloc(handle->snapshot);
3362 	if (handlep->oneshot_buffer == NULL) {
3363 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3364 			 "can't allocate oneshot buffer: %s",
3365 			 pcap_strerror(errno));
3366 		*status = PCAP_ERROR;
3367 		return -1;
3368 	}
3369 
3370 	if (handle->opt.buffer_size == 0) {
3371 		/* by default request 2M for the ring buffer */
3372 		handle->opt.buffer_size = 2*1024*1024;
3373 	}
3374 	ret = prepare_tpacket_socket(handle);
3375 	if (ret == -1) {
3376 		free(handlep->oneshot_buffer);
3377 		*status = PCAP_ERROR;
3378 		return ret;
3379 	}
3380 	ret = create_ring(handle, status);
3381 	if (ret == 0) {
3382 		/*
3383 		 * We don't support memory-mapped capture; our caller
3384 		 * will fall back on reading from the socket.
3385 		 */
3386 		free(handlep->oneshot_buffer);
3387 		return 0;
3388 	}
3389 	if (ret == -1) {
3390 		/*
3391 		 * Error attempting to enable memory-mapped capture;
3392 		 * fail.  create_ring() has set *status.
3393 		 */
3394 		free(handlep->oneshot_buffer);
3395 		return -1;
3396 	}
3397 
3398 	/*
3399 	 * Success.  *status has been set either to 0 if there are no
3400 	 * warnings or to a PCAP_WARNING_ value if there is a warning.
3401 	 *
3402 	 * Override some defaults and inherit the other fields from
3403 	 * activate_new.
3404 	 * handle->offset is used to get the current position into the rx ring.
3405 	 * handle->cc is used to store the ring size.
3406 	 */
3407 
3408 	switch (handlep->tp_version) {
3409 	case TPACKET_V1:
3410 		handle->read_op = pcap_read_linux_mmap_v1;
3411 		break;
3412 #ifdef HAVE_TPACKET2
3413 	case TPACKET_V2:
3414 		handle->read_op = pcap_read_linux_mmap_v2;
3415 		break;
3416 #endif
3417 #ifdef HAVE_TPACKET3
3418 	case TPACKET_V3:
3419 		handle->read_op = pcap_read_linux_mmap_v3;
3420 		break;
3421 #endif
3422 	}
3423 	handle->cleanup_op = pcap_cleanup_linux_mmap;
3424 	handle->setfilter_op = pcap_setfilter_linux_mmap;
3425 	handle->setnonblock_op = pcap_setnonblock_mmap;
3426 	handle->getnonblock_op = pcap_getnonblock_mmap;
3427 	handle->oneshot_callback = pcap_oneshot_mmap;
3428 	handle->selectable_fd = handle->fd;
3429 	return 1;
3430 }
3431 #else /* HAVE_PACKET_RING */
3432 static int
3433 activate_mmap(pcap_t *handle _U_, int *status _U_)
3434 {
3435 	return 0;
3436 }
3437 #endif /* HAVE_PACKET_RING */
3438 
3439 #ifdef HAVE_PACKET_RING
3440 
3441 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
3442 /*
3443  * Attempt to set the socket to the specified version of the memory-mapped
3444  * header.
3445  *
3446  * Return 0 if we succeed; return 1 if we fail because that version isn't
3447  * supported; return -1 on any other error, and set handle->errbuf.
3448  */
3449 static int
3450 init_tpacket(pcap_t *handle, int version, const char *version_str)
3451 {
3452 	struct pcap_linux *handlep = handle->priv;
3453 	int val = version;
3454 	socklen_t len = sizeof(val);
3455 
3456 	/* Probe whether kernel supports the specified TPACKET version */
3457 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
3458 		if (errno == ENOPROTOOPT || errno == EINVAL)
3459 			return 1;	/* no */
3460 
3461 		/* Failed to even find out; this is a fatal error. */
3462 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3463 			"can't get %s header len on packet socket: %s",
3464 			version_str,
3465 			pcap_strerror(errno));
3466 		return -1;
3467 	}
3468 	handlep->tp_hdrlen = val;
3469 
3470 	val = version;
3471 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
3472 			   sizeof(val)) < 0) {
3473 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3474 			"can't activate %s on packet socket: %s",
3475 			version_str,
3476 			pcap_strerror(errno));
3477 		return -1;
3478 	}
3479 	handlep->tp_version = version;
3480 
3481 	/* Reserve space for VLAN tag reconstruction */
3482 	val = VLAN_TAG_LEN;
3483 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val,
3484 			   sizeof(val)) < 0) {
3485 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3486 			"can't set up reserve on packet socket: %s",
3487 			pcap_strerror(errno));
3488 		return -1;
3489 	}
3490 
3491 	return 0;
3492 }
3493 #endif /* defined HAVE_TPACKET2 || defined HAVE_TPACKET3 */
3494 
3495 /*
3496  * Attempt to set the socket to version 3 of the memory-mapped header and,
3497  * if that fails because version 3 isn't supported, attempt to fall
3498  * back to version 2.  If version 2 isn't supported, just leave it at
3499  * version 1.
3500  *
3501  * Return 1 if we succeed or if we fail because neither version 2 nor 3 is
3502  * supported; return -1 on any other error, and set handle->errbuf.
3503  */
3504 static int
3505 prepare_tpacket_socket(pcap_t *handle)
3506 {
3507 	struct pcap_linux *handlep = handle->priv;
3508 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
3509 	int ret;
3510 #endif
3511 
3512 	handlep->tp_version = TPACKET_V1;
3513 	handlep->tp_hdrlen = sizeof(struct tpacket_hdr);
3514 
3515 #ifdef HAVE_TPACKET3
3516 	ret = init_tpacket(handle, TPACKET_V3, "TPACKET_V3");
3517 	if (-1 == ret) {
3518 		/* Error during setting up TPACKET_V3. */
3519 		return -1;
3520 	} else if (1 == ret) {
3521 		/* TPACKET_V3 not supported - fall back to TPACKET_V2. */
3522 #endif /* HAVE_TPACKET3 */
3523 
3524 #ifdef HAVE_TPACKET2
3525 		ret = init_tpacket(handle, TPACKET_V2, "TPACKET_V2");
3526 		if (-1 == ret) {
3527 			/* Error during setting up TPACKET_V2. */
3528 			return -1;
3529 		}
3530 #endif /* HAVE_TPACKET2 */
3531 
3532 #ifdef HAVE_TPACKET3
3533 	}
3534 #endif /* HAVE_TPACKET3 */
3535 
3536 	return 1;
3537 }
3538 
3539 /*
3540  * Attempt to set up memory-mapped access.
3541  *
3542  * On success, returns 1, and sets *status to 0 if there are no warnings
3543  * or to a PCAP_WARNING_ code if there is a warning.
3544  *
3545  * On failure due to lack of support for memory-mapped capture, returns
3546  * 0.
3547  *
3548  * On error, returns -1, and sets *status to the appropriate error code;
3549  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3550  */
3551 static int
3552 create_ring(pcap_t *handle, int *status)
3553 {
3554 	struct pcap_linux *handlep = handle->priv;
3555 	unsigned i, j, frames_per_block;
3556 #ifdef HAVE_TPACKET3
3557 	/*
3558 	 * For sockets using TPACKET_V1 or TPACKET_V2, the extra
3559 	 * stuff at the end of a struct tpacket_req3 will be
3560 	 * ignored, so this is OK even for those sockets.
3561 	 */
3562 	struct tpacket_req3 req;
3563 #else
3564 	struct tpacket_req req;
3565 #endif
3566 	socklen_t len;
3567 	unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
3568 	unsigned int frame_size;
3569 
3570 	/*
3571 	 * Start out assuming no warnings or errors.
3572 	 */
3573 	*status = 0;
3574 
3575 	switch (handlep->tp_version) {
3576 
3577 	case TPACKET_V1:
3578 #ifdef HAVE_TPACKET2
3579 	case TPACKET_V2:
3580 #endif
3581 		/* Note that with large snapshot length (say 64K, which is
3582 		 * the default for recent versions of tcpdump, the value that
3583 		 * "-s 0" has given for a long time with tcpdump, and the
3584 		 * default in Wireshark/TShark/dumpcap), if we use the snapshot
3585 		 * length to calculate the frame length, only a few frames
3586 		 * will be available in the ring even with pretty
3587 		 * large ring size (and a lot of memory will be unused).
3588 		 *
3589 		 * Ideally, we should choose a frame length based on the
3590 		 * minimum of the specified snapshot length and the maximum
3591 		 * packet size.  That's not as easy as it sounds; consider,
3592 		 * for example, an 802.11 interface in monitor mode, where
3593 		 * the frame would include a radiotap header, where the
3594 		 * maximum radiotap header length is device-dependent.
3595 		 *
3596 		 * So, for now, we just do this for Ethernet devices, where
3597 		 * there's no metadata header, and the link-layer header is
3598 		 * fixed length.  We can get the maximum packet size by
3599 		 * adding 18, the Ethernet header length plus the CRC length
3600 		 * (just in case we happen to get the CRC in the packet), to
3601 		 * the MTU of the interface; we fetch the MTU in the hopes
3602 		 * that it reflects support for jumbo frames.  (Even if the
3603 		 * interface is just being used for passive snooping, the
3604 		 * driver might set the size of buffers in the receive ring
3605 		 * based on the MTU, so that the MTU limits the maximum size
3606 		 * of packets that we can receive.)
3607 		 *
3608 		 * We don't do that if segmentation/fragmentation or receive
3609 		 * offload are enabled, so we don't get rudely surprised by
3610 		 * "packets" bigger than the MTU. */
3611 		frame_size = handle->snapshot;
3612 		if (handle->linktype == DLT_EN10MB) {
3613 			int mtu;
3614 			int offload;
3615 
3616 			offload = iface_get_offload(handle);
3617 			if (offload == -1) {
3618 				*status = PCAP_ERROR;
3619 				return -1;
3620 			}
3621 			if (!offload) {
3622 				mtu = iface_get_mtu(handle->fd, handle->opt.source,
3623 				    handle->errbuf);
3624 				if (mtu == -1) {
3625 					*status = PCAP_ERROR;
3626 					return -1;
3627 				}
3628 				if (frame_size > mtu + 18)
3629 					frame_size = mtu + 18;
3630 			}
3631 		}
3632 
3633 		/* NOTE: calculus matching those in tpacket_rcv()
3634 		 * in linux-2.6/net/packet/af_packet.c
3635 		 */
3636 		len = sizeof(sk_type);
3637 		if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type,
3638 		    &len) < 0) {
3639 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3640 			    "getsockopt: %s", pcap_strerror(errno));
3641 			*status = PCAP_ERROR;
3642 			return -1;
3643 		}
3644 #ifdef PACKET_RESERVE
3645 		len = sizeof(tp_reserve);
3646 		if (getsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE,
3647 		    &tp_reserve, &len) < 0) {
3648 			if (errno != ENOPROTOOPT) {
3649 				/*
3650 				 * ENOPROTOOPT means "kernel doesn't support
3651 				 * PACKET_RESERVE", in which case we fall back
3652 				 * as best we can.
3653 				 */
3654 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3655 				    "getsockopt: %s", pcap_strerror(errno));
3656 				*status = PCAP_ERROR;
3657 				return -1;
3658 			}
3659 			tp_reserve = 0;	/* older kernel, reserve not supported */
3660 		}
3661 #else
3662 		tp_reserve = 0;	/* older kernel, reserve not supported */
3663 #endif
3664 		maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
3665 			/* XXX: in the kernel maclen is calculated from
3666 			 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
3667 			 * in:  packet_snd()           in linux-2.6/net/packet/af_packet.c
3668 			 * then packet_alloc_skb()     in linux-2.6/net/packet/af_packet.c
3669 			 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
3670 			 * but I see no way to get those sizes in userspace,
3671 			 * like for instance with an ifreq ioctl();
3672 			 * the best thing I've found so far is MAX_HEADER in
3673 			 * the kernel part of linux-2.6/include/linux/netdevice.h
3674 			 * which goes up to 128+48=176; since pcap-linux.c
3675 			 * defines a MAX_LINKHEADER_SIZE of 256 which is
3676 			 * greater than that, let's use it.. maybe is it even
3677 			 * large enough to directly replace macoff..
3678 			 */
3679 		tp_hdrlen = TPACKET_ALIGN(handlep->tp_hdrlen) + sizeof(struct sockaddr_ll) ;
3680 		netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
3681 			/* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN
3682 			 * of netoff, which contradicts
3683 			 * linux-2.6/Documentation/networking/packet_mmap.txt
3684 			 * documenting that:
3685 			 * "- Gap, chosen so that packet data (Start+tp_net)
3686 			 * aligns to TPACKET_ALIGNMENT=16"
3687 			 */
3688 			/* NOTE: in linux-2.6/include/linux/skbuff.h:
3689 			 * "CPUs often take a performance hit
3690 			 *  when accessing unaligned memory locations"
3691 			 */
3692 		macoff = netoff - maclen;
3693 		req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
3694 		req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
3695 		break;
3696 
3697 #ifdef HAVE_TPACKET3
3698 	case TPACKET_V3:
3699 		/* The "frames" for this are actually buffers that
3700 		 * contain multiple variable-sized frames.
3701 		 *
3702 		 * We pick a "frame" size of 128K to leave enough
3703 		 * room for at least one reasonably-sized packet
3704 		 * in the "frame". */
3705 		req.tp_frame_size = 131072;
3706 		req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
3707 		break;
3708 #endif
3709 	}
3710 
3711 	/* compute the minumum block size that will handle this frame.
3712 	 * The block has to be page size aligned.
3713 	 * The max block size allowed by the kernel is arch-dependent and
3714 	 * it's not explicitly checked here. */
3715 	req.tp_block_size = getpagesize();
3716 	while (req.tp_block_size < req.tp_frame_size)
3717 		req.tp_block_size <<= 1;
3718 
3719 	frames_per_block = req.tp_block_size/req.tp_frame_size;
3720 
3721 	/*
3722 	 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
3723 	 * so we check for PACKET_TIMESTAMP.  We check for
3724 	 * linux/net_tstamp.h just in case a system somehow has
3725 	 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
3726 	 * be unnecessary.
3727 	 *
3728 	 * SIOCSHWTSTAMP was introduced in the patch that introduced
3729 	 * linux/net_tstamp.h, so we don't bother checking whether
3730 	 * SIOCSHWTSTAMP is defined (if your Linux system has
3731 	 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
3732 	 * Linux system is badly broken).
3733 	 */
3734 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
3735 	/*
3736 	 * If we were told to do so, ask the kernel and the driver
3737 	 * to use hardware timestamps.
3738 	 *
3739 	 * Hardware timestamps are only supported with mmapped
3740 	 * captures.
3741 	 */
3742 	if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
3743 	    handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
3744 		struct hwtstamp_config hwconfig;
3745 		struct ifreq ifr;
3746 		int timesource;
3747 
3748 		/*
3749 		 * Ask for hardware time stamps on all packets,
3750 		 * including transmitted packets.
3751 		 */
3752 		memset(&hwconfig, 0, sizeof(hwconfig));
3753 		hwconfig.tx_type = HWTSTAMP_TX_ON;
3754 		hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
3755 
3756 		memset(&ifr, 0, sizeof(ifr));
3757 		strcpy(ifr.ifr_name, handle->opt.source);
3758 		ifr.ifr_data = (void *)&hwconfig;
3759 
3760 		if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
3761 			switch (errno) {
3762 
3763 			case EPERM:
3764 				/*
3765 				 * Treat this as an error, as the
3766 				 * user should try to run this
3767 				 * with the appropriate privileges -
3768 				 * and, if they can't, shouldn't
3769 				 * try requesting hardware time stamps.
3770 				 */
3771 				*status = PCAP_ERROR_PERM_DENIED;
3772 				return -1;
3773 
3774 			case EOPNOTSUPP:
3775 				/*
3776 				 * Treat this as a warning, as the
3777 				 * only way to fix the warning is to
3778 				 * get an adapter that supports hardware
3779 				 * time stamps.  We'll just fall back
3780 				 * on the standard host time stamps.
3781 				 */
3782 				*status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
3783 				break;
3784 
3785 			default:
3786 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3787 					"SIOCSHWTSTAMP failed: %s",
3788 					pcap_strerror(errno));
3789 				*status = PCAP_ERROR;
3790 				return -1;
3791 			}
3792 		} else {
3793 			/*
3794 			 * Well, that worked.  Now specify the type of
3795 			 * hardware time stamp we want for this
3796 			 * socket.
3797 			 */
3798 			if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
3799 				/*
3800 				 * Hardware timestamp, synchronized
3801 				 * with the system clock.
3802 				 */
3803 				timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
3804 			} else {
3805 				/*
3806 				 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
3807 				 * timestamp, not synchronized with the
3808 				 * system clock.
3809 				 */
3810 				timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
3811 			}
3812 			if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
3813 				(void *)&timesource, sizeof(timesource))) {
3814 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3815 					"can't set PACKET_TIMESTAMP: %s",
3816 					pcap_strerror(errno));
3817 				*status = PCAP_ERROR;
3818 				return -1;
3819 			}
3820 		}
3821 	}
3822 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
3823 
3824 	/* ask the kernel to create the ring */
3825 retry:
3826 	req.tp_block_nr = req.tp_frame_nr / frames_per_block;
3827 
3828 	/* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
3829 	req.tp_frame_nr = req.tp_block_nr * frames_per_block;
3830 
3831 #ifdef HAVE_TPACKET3
3832 	/* timeout value to retire block - use the configured buffering timeout, or default if <0. */
3833 	req.tp_retire_blk_tov = (handlep->timeout>=0)?handlep->timeout:0;
3834 	/* private data not used */
3835 	req.tp_sizeof_priv = 0;
3836 	/* Rx ring - feature request bits - none (rxhash will not be filled) */
3837 	req.tp_feature_req_word = 0;
3838 #endif
3839 
3840 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3841 					(void *) &req, sizeof(req))) {
3842 		if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
3843 			/*
3844 			 * Memory failure; try to reduce the requested ring
3845 			 * size.
3846 			 *
3847 			 * We used to reduce this by half -- do 5% instead.
3848 			 * That may result in more iterations and a longer
3849 			 * startup, but the user will be much happier with
3850 			 * the resulting buffer size.
3851 			 */
3852 			if (req.tp_frame_nr < 20)
3853 				req.tp_frame_nr -= 1;
3854 			else
3855 				req.tp_frame_nr -= req.tp_frame_nr/20;
3856 			goto retry;
3857 		}
3858 		if (errno == ENOPROTOOPT) {
3859 			/*
3860 			 * We don't have ring buffer support in this kernel.
3861 			 */
3862 			return 0;
3863 		}
3864 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3865 		    "can't create rx ring on packet socket: %s",
3866 		    pcap_strerror(errno));
3867 		*status = PCAP_ERROR;
3868 		return -1;
3869 	}
3870 
3871 	/* memory map the rx ring */
3872 	handlep->mmapbuflen = req.tp_block_nr * req.tp_block_size;
3873 	handlep->mmapbuf = mmap(0, handlep->mmapbuflen,
3874 	    PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
3875 	if (handlep->mmapbuf == MAP_FAILED) {
3876 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3877 		    "can't mmap rx ring: %s", pcap_strerror(errno));
3878 
3879 		/* clear the allocated ring on error*/
3880 		destroy_ring(handle);
3881 		*status = PCAP_ERROR;
3882 		return -1;
3883 	}
3884 
3885 	/* allocate a ring for each frame header pointer*/
3886 	handle->cc = req.tp_frame_nr;
3887 	handle->buffer = malloc(handle->cc * sizeof(union thdr *));
3888 	if (!handle->buffer) {
3889 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3890 		    "can't allocate ring of frame headers: %s",
3891 		    pcap_strerror(errno));
3892 
3893 		destroy_ring(handle);
3894 		*status = PCAP_ERROR;
3895 		return -1;
3896 	}
3897 
3898 	/* fill the header ring with proper frame ptr*/
3899 	handle->offset = 0;
3900 	for (i=0; i<req.tp_block_nr; ++i) {
3901 		void *base = &handlep->mmapbuf[i*req.tp_block_size];
3902 		for (j=0; j<frames_per_block; ++j, ++handle->offset) {
3903 			RING_GET_FRAME(handle) = base;
3904 			base += req.tp_frame_size;
3905 		}
3906 	}
3907 
3908 	handle->bufsize = req.tp_frame_size;
3909 	handle->offset = 0;
3910 	return 1;
3911 }
3912 
3913 /* free all ring related resources*/
3914 static void
3915 destroy_ring(pcap_t *handle)
3916 {
3917 	struct pcap_linux *handlep = handle->priv;
3918 
3919 	/* tell the kernel to destroy the ring*/
3920 	struct tpacket_req req;
3921 	memset(&req, 0, sizeof(req));
3922 	setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3923 				(void *) &req, sizeof(req));
3924 
3925 	/* if ring is mapped, unmap it*/
3926 	if (handlep->mmapbuf) {
3927 		/* do not test for mmap failure, as we can't recover from any error */
3928 		munmap(handlep->mmapbuf, handlep->mmapbuflen);
3929 		handlep->mmapbuf = NULL;
3930 	}
3931 }
3932 
3933 /*
3934  * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
3935  * for Linux mmapped capture.
3936  *
3937  * The problem is that pcap_next() and pcap_next_ex() expect the packet
3938  * data handed to the callback to be valid after the callback returns,
3939  * but pcap_read_linux_mmap() has to release that packet as soon as
3940  * the callback returns (otherwise, the kernel thinks there's still
3941  * at least one unprocessed packet available in the ring, so a select()
3942  * will immediately return indicating that there's data to process), so,
3943  * in the callback, we have to make a copy of the packet.
3944  *
3945  * Yes, this means that, if the capture is using the ring buffer, using
3946  * pcap_next() or pcap_next_ex() requires more copies than using
3947  * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
3948  * pcap_next() or pcap_next_ex().
3949  */
3950 static void
3951 pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
3952     const u_char *bytes)
3953 {
3954 	struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
3955 	pcap_t *handle = sp->pd;
3956 	struct pcap_linux *handlep = handle->priv;
3957 
3958 	*sp->hdr = *h;
3959 	memcpy(handlep->oneshot_buffer, bytes, h->caplen);
3960 	*sp->pkt = handlep->oneshot_buffer;
3961 }
3962 
3963 static void
3964 pcap_cleanup_linux_mmap( pcap_t *handle )
3965 {
3966 	struct pcap_linux *handlep = handle->priv;
3967 
3968 	destroy_ring(handle);
3969 	if (handlep->oneshot_buffer != NULL) {
3970 		free(handlep->oneshot_buffer);
3971 		handlep->oneshot_buffer = NULL;
3972 	}
3973 	pcap_cleanup_linux(handle);
3974 }
3975 
3976 
3977 static int
3978 pcap_getnonblock_mmap(pcap_t *p, char *errbuf)
3979 {
3980 	struct pcap_linux *handlep = p->priv;
3981 
3982 	/* use negative value of timeout to indicate non blocking ops */
3983 	return (handlep->timeout<0);
3984 }
3985 
3986 static int
3987 pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
3988 {
3989 	struct pcap_linux *handlep = p->priv;
3990 
3991 	/*
3992 	 * Map each value to their corresponding negation to
3993 	 * preserve the timeout value provided with pcap_set_timeout.
3994 	 */
3995 	if (nonblock) {
3996 		if (handlep->timeout >= 0) {
3997 			/*
3998 			 * Indicate that we're switching to
3999 			 * non-blocking mode.
4000 			 */
4001 			handlep->timeout = ~handlep->timeout;
4002 		}
4003 	} else {
4004 		if (handlep->timeout < 0) {
4005 			handlep->timeout = ~handlep->timeout;
4006 		}
4007 	}
4008 	return 0;
4009 }
4010 
4011 static inline union thdr *
4012 pcap_get_ring_frame(pcap_t *handle, int status)
4013 {
4014 	struct pcap_linux *handlep = handle->priv;
4015 	union thdr h;
4016 
4017 	h.raw = RING_GET_FRAME(handle);
4018 	switch (handlep->tp_version) {
4019 	case TPACKET_V1:
4020 		if (status != (h.h1->tp_status ? TP_STATUS_USER :
4021 						TP_STATUS_KERNEL))
4022 			return NULL;
4023 		break;
4024 #ifdef HAVE_TPACKET2
4025 	case TPACKET_V2:
4026 		if (status != (h.h2->tp_status ? TP_STATUS_USER :
4027 						TP_STATUS_KERNEL))
4028 			return NULL;
4029 		break;
4030 #endif
4031 #ifdef HAVE_TPACKET3
4032 	case TPACKET_V3:
4033 		if (status != (h.h3->hdr.bh1.block_status ? TP_STATUS_USER :
4034 						TP_STATUS_KERNEL))
4035 			return NULL;
4036 		break;
4037 #endif
4038 	}
4039 	return h.raw;
4040 }
4041 
4042 #ifndef POLLRDHUP
4043 #define POLLRDHUP 0
4044 #endif
4045 
4046 /* wait for frames availability.*/
4047 static int pcap_wait_for_frames_mmap(pcap_t *handle)
4048 {
4049 	if (!pcap_get_ring_frame(handle, TP_STATUS_USER)) {
4050 		struct pcap_linux *handlep = handle->priv;
4051 		int timeout;
4052 		char c;
4053 		struct pollfd pollinfo;
4054 		int ret;
4055 
4056 		pollinfo.fd = handle->fd;
4057 		pollinfo.events = POLLIN;
4058 
4059 		if (handlep->timeout == 0)
4060 			timeout = -1;	/* block forever */
4061 		else if (handlep->timeout > 0)
4062 			timeout = handlep->timeout;	/* block for that amount of time */
4063 		else
4064 			timeout = 0;	/* non-blocking mode - poll to pick up errors */
4065 		do {
4066 			ret = poll(&pollinfo, 1, timeout);
4067 			if (ret < 0 && errno != EINTR) {
4068 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4069 					"can't poll on packet socket: %s",
4070 					pcap_strerror(errno));
4071 				return PCAP_ERROR;
4072 			} else if (ret > 0 &&
4073 				(pollinfo.revents & (POLLHUP|POLLRDHUP|POLLERR|POLLNVAL))) {
4074 				/*
4075 				 * There's some indication other than
4076 				 * "you can read on this descriptor" on
4077 				 * the descriptor.
4078 				 */
4079 				if (pollinfo.revents & (POLLHUP | POLLRDHUP)) {
4080 					snprintf(handle->errbuf,
4081 						PCAP_ERRBUF_SIZE,
4082 						"Hangup on packet socket");
4083 					return PCAP_ERROR;
4084 				}
4085 				if (pollinfo.revents & POLLERR) {
4086 					/*
4087 					 * A recv() will give us the
4088 					 * actual error code.
4089 					 *
4090 					 * XXX - make the socket non-blocking?
4091 					 */
4092 					if (recv(handle->fd, &c, sizeof c,
4093 						MSG_PEEK) != -1)
4094 						continue;	/* what, no error? */
4095 					if (errno == ENETDOWN) {
4096 						/*
4097 						 * The device on which we're
4098 						 * capturing went away.
4099 						 *
4100 						 * XXX - we should really return
4101 						 * PCAP_ERROR_IFACE_NOT_UP,
4102 						 * but pcap_dispatch() etc.
4103 						 * aren't defined to return
4104 						 * that.
4105 						 */
4106 						snprintf(handle->errbuf,
4107 							PCAP_ERRBUF_SIZE,
4108 							"The interface went down");
4109 					} else {
4110 						snprintf(handle->errbuf,
4111 							PCAP_ERRBUF_SIZE,
4112 							"Error condition on packet socket: %s",
4113 							strerror(errno));
4114 					}
4115 					return PCAP_ERROR;
4116 				}
4117 				if (pollinfo.revents & POLLNVAL) {
4118 					snprintf(handle->errbuf,
4119 						PCAP_ERRBUF_SIZE,
4120 						"Invalid polling request on packet socket");
4121 					return PCAP_ERROR;
4122 				}
4123 			}
4124 			/* check for break loop condition on interrupted syscall*/
4125 			if (handle->break_loop) {
4126 				handle->break_loop = 0;
4127 				return PCAP_ERROR_BREAK;
4128 			}
4129 		} while (ret < 0);
4130 	}
4131 	return 0;
4132 }
4133 
4134 /* handle a single memory mapped packet */
4135 static int pcap_handle_packet_mmap(
4136 		pcap_t *handle,
4137 		pcap_handler callback,
4138 		u_char *user,
4139 		unsigned char *frame,
4140 		unsigned int tp_len,
4141 		unsigned int tp_mac,
4142 		unsigned int tp_snaplen,
4143 		unsigned int tp_sec,
4144 		unsigned int tp_usec,
4145 		int tp_vlan_tci_valid,
4146 		__u16 tp_vlan_tci)
4147 {
4148 	struct pcap_linux *handlep = handle->priv;
4149 	unsigned char *bp;
4150 	int run_bpf;
4151 	struct sockaddr_ll *sll;
4152 	struct pcap_pkthdr pcaphdr;
4153 
4154 	/* perform sanity check on internal offset. */
4155 	if (tp_mac + tp_snaplen > handle->bufsize) {
4156 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4157 			"corrupted frame on kernel ring mac "
4158 			"offset %d + caplen %d > frame len %d",
4159 			tp_mac, tp_snaplen, handle->bufsize);
4160 		return -1;
4161 	}
4162 
4163 	/* run filter on received packet
4164 	 * If the kernel filtering is enabled we need to run the
4165 	 * filter until all the frames present into the ring
4166 	 * at filter creation time are processed.
4167 	 * In such case filtering_in_kernel is used as a counter for the
4168 	 * packet we need to filter.
4169 	 * Note: alternatively it could be possible to stop applying
4170 	 * the filter when the ring became empty, but it can possibly
4171 	 * happen a lot later... */
4172 	bp = frame + tp_mac;
4173 	run_bpf = (!handlep->filtering_in_kernel) ||
4174 		((handlep->filtering_in_kernel>1) && handlep->filtering_in_kernel--);
4175 	if (run_bpf && handle->fcode.bf_insns &&
4176 			(bpf_filter(handle->fcode.bf_insns, bp,
4177 				tp_len, tp_snaplen) == 0))
4178 		return 0;
4179 
4180 	sll = (void *)frame + TPACKET_ALIGN(handlep->tp_hdrlen);
4181 	if (!linux_check_direction(handle, sll))
4182 		return 0;
4183 
4184 	/* get required packet info from ring header */
4185 	pcaphdr.ts.tv_sec = tp_sec;
4186 	pcaphdr.ts.tv_usec = tp_usec;
4187 	pcaphdr.caplen = tp_snaplen;
4188 	pcaphdr.len = tp_len;
4189 
4190 	/* if required build in place the sll header*/
4191 	if (handlep->cooked) {
4192 		struct sll_header *hdrp;
4193 
4194 		/*
4195 		 * The kernel should have left us with enough
4196 		 * space for an sll header; back up the packet
4197 		 * data pointer into that space, as that'll be
4198 		 * the beginning of the packet we pass to the
4199 		 * callback.
4200 		 */
4201 		bp -= SLL_HDR_LEN;
4202 
4203 		/*/*
4204 		 * Let's make sure that's past the end of
4205 		 * the tpacket header, i.e. >=
4206 		 * ((u_char *)thdr + TPACKET_HDRLEN), so we
4207 		 * don't step on the header when we construct
4208 		 * the sll header.
4209 		 */
4210 		if (bp < (u_char *)frame +
4211 				   TPACKET_ALIGN(handlep->tp_hdrlen) +
4212 				   sizeof(struct sockaddr_ll)) {
4213 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4214 				"cooked-mode frame doesn't have room for sll header");
4215 			return -1;
4216 		}
4217 
4218 		/*
4219 		 * OK, that worked; construct the sll header.
4220 		 */
4221 		hdrp = (struct sll_header *)bp;
4222 		hdrp->sll_pkttype = map_packet_type_to_sll_type(
4223 						sll->sll_pkttype);
4224 		hdrp->sll_hatype = htons(sll->sll_hatype);
4225 		hdrp->sll_halen = htons(sll->sll_halen);
4226 		memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
4227 		hdrp->sll_protocol = sll->sll_protocol;
4228 
4229 		/* update packet len */
4230 		pcaphdr.caplen += SLL_HDR_LEN;
4231 		pcaphdr.len += SLL_HDR_LEN;
4232 	}
4233 
4234 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
4235 	if (tp_vlan_tci_valid &&
4236 		handlep->vlan_offset != -1 &&
4237 		tp_snaplen >= (unsigned int) handlep->vlan_offset)
4238 	{
4239 		struct vlan_tag *tag;
4240 
4241 		bp -= VLAN_TAG_LEN;
4242 		memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
4243 
4244 		tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
4245 		tag->vlan_tpid = htons(ETH_P_8021Q);
4246 		tag->vlan_tci = htons(tp_vlan_tci);
4247 
4248 		pcaphdr.caplen += VLAN_TAG_LEN;
4249 		pcaphdr.len += VLAN_TAG_LEN;
4250 	}
4251 #endif
4252 
4253 	/*
4254 	 * The only way to tell the kernel to cut off the
4255 	 * packet at a snapshot length is with a filter program;
4256 	 * if there's no filter program, the kernel won't cut
4257 	 * the packet off.
4258 	 *
4259 	 * Trim the snapshot length to be no longer than the
4260 	 * specified snapshot length.
4261 	 */
4262 	if (pcaphdr.caplen > handle->snapshot)
4263 		pcaphdr.caplen = handle->snapshot;
4264 
4265 	/* pass the packet to the user */
4266 	callback(user, &pcaphdr, bp);
4267 
4268 	return 1;
4269 }
4270 
4271 static int
4272 pcap_read_linux_mmap_v1(pcap_t *handle, int max_packets, pcap_handler callback,
4273 		u_char *user)
4274 {
4275 	struct pcap_linux *handlep = handle->priv;
4276 	int pkts = 0;
4277 	int ret;
4278 
4279 	/* wait for frames availability.*/
4280 	ret = pcap_wait_for_frames_mmap(handle);
4281 	if (ret) {
4282 		return ret;
4283 	}
4284 
4285 	/* non-positive values of max_packets are used to require all
4286 	 * packets currently available in the ring */
4287 	while ((pkts < max_packets) || (max_packets <= 0)) {
4288 		union thdr h;
4289 
4290 		h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
4291 		if (!h.raw)
4292 			break;
4293 
4294 		ret = pcap_handle_packet_mmap(
4295 				handle,
4296 				callback,
4297 				user,
4298 				h.raw,
4299 				h.h1->tp_len,
4300 				h.h1->tp_mac,
4301 				h.h1->tp_snaplen,
4302 				h.h1->tp_sec,
4303 				h.h1->tp_usec,
4304 				0,
4305 				0);
4306 		if (ret == 1) {
4307 			pkts++;
4308 			handlep->packets_read++;
4309 		} else if (ret < 0) {
4310 			return ret;
4311 		}
4312 
4313 		/* next packet */
4314 		h.h1->tp_status = TP_STATUS_KERNEL;
4315 
4316 		if (++handle->offset >= handle->cc)
4317 			handle->offset = 0;
4318 
4319 		/* check for break loop condition*/
4320 		if (handle->break_loop) {
4321 			handle->break_loop = 0;
4322 			return PCAP_ERROR_BREAK;
4323 		}
4324 	}
4325 	return pkts;
4326 }
4327 
4328 #ifdef HAVE_TPACKET2
4329 static int
4330 pcap_read_linux_mmap_v2(pcap_t *handle, int max_packets, pcap_handler callback,
4331 		u_char *user)
4332 {
4333 	struct pcap_linux *handlep = handle->priv;
4334 	int pkts = 0;
4335 	int ret;
4336 
4337 	/* wait for frames availability.*/
4338 	ret = pcap_wait_for_frames_mmap(handle);
4339 	if (ret) {
4340 		return ret;
4341 	}
4342 
4343 	/* non-positive values of max_packets are used to require all
4344 	 * packets currently available in the ring */
4345 	while ((pkts < max_packets) || (max_packets <= 0)) {
4346 		union thdr h;
4347 
4348 		h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
4349 		if (!h.raw)
4350 			break;
4351 
4352 		ret = pcap_handle_packet_mmap(
4353 				handle,
4354 				callback,
4355 				user,
4356 				h.raw,
4357 				h.h2->tp_len,
4358 				h.h2->tp_mac,
4359 				h.h2->tp_snaplen,
4360 				h.h2->tp_sec,
4361 				handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? h.h2->tp_nsec : h.h2->tp_nsec / 1000,
4362 #if defined(TP_STATUS_VLAN_VALID)
4363 				(h.h2->tp_vlan_tci || (h.h2->tp_status & TP_STATUS_VLAN_VALID)),
4364 #else
4365 				h.h2->tp_vlan_tci != 0,
4366 #endif
4367 				h.h2->tp_vlan_tci);
4368 		if (ret == 1) {
4369 			pkts++;
4370 			handlep->packets_read++;
4371 		} else if (ret < 0) {
4372 			return ret;
4373 		}
4374 
4375 		/* next packet */
4376 		h.h2->tp_status = TP_STATUS_KERNEL;
4377 
4378 		if (++handle->offset >= handle->cc)
4379 			handle->offset = 0;
4380 
4381 		/* check for break loop condition*/
4382 		if (handle->break_loop) {
4383 			handle->break_loop = 0;
4384 			return PCAP_ERROR_BREAK;
4385 		}
4386 	}
4387 	return pkts;
4388 }
4389 #endif /* HAVE_TPACKET2 */
4390 
4391 #ifdef HAVE_TPACKET3
4392 static int
4393 pcap_read_linux_mmap_v3(pcap_t *handle, int max_packets, pcap_handler callback,
4394 		u_char *user)
4395 {
4396 	struct pcap_linux *handlep = handle->priv;
4397 	union thdr h;
4398 	int pkts = 0;
4399 	int ret;
4400 
4401 	if (handlep->current_packet == NULL) {
4402 		/* wait for frames availability.*/
4403 		ret = pcap_wait_for_frames_mmap(handle);
4404 		if (ret) {
4405 			return ret;
4406 		}
4407 	}
4408 	h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
4409 	if (!h.raw)
4410 		return pkts;
4411 
4412 	/* non-positive values of max_packets are used to require all
4413 	 * packets currently available in the ring */
4414 	while ((pkts < max_packets) || (max_packets <= 0)) {
4415 		if (handlep->current_packet == NULL) {
4416 			h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
4417 			if (!h.raw)
4418 				break;
4419 
4420 			handlep->current_packet = h.raw + h.h3->hdr.bh1.offset_to_first_pkt;
4421 			handlep->packets_left = h.h3->hdr.bh1.num_pkts;
4422 		}
4423 		int packets_to_read = handlep->packets_left;
4424 
4425 		if (max_packets >= 0 && packets_to_read > max_packets) {
4426 			packets_to_read = max_packets;
4427 		}
4428 
4429 		while(packets_to_read--) {
4430 			struct tpacket3_hdr* tp3_hdr = (struct tpacket3_hdr*) handlep->current_packet;
4431 			ret = pcap_handle_packet_mmap(
4432 					handle,
4433 					callback,
4434 					user,
4435 					handlep->current_packet,
4436 					tp3_hdr->tp_len,
4437 					tp3_hdr->tp_mac,
4438 					tp3_hdr->tp_snaplen,
4439 					tp3_hdr->tp_sec,
4440 					handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? tp3_hdr->tp_nsec : tp3_hdr->tp_nsec / 1000,
4441 #if defined(TP_STATUS_VLAN_VALID)
4442 					(tp3_hdr->hv1.tp_vlan_tci || (tp3_hdr->tp_status & TP_STATUS_VLAN_VALID)),
4443 #else
4444 					tp3_hdr->hv1.tp_vlan_tci != 0,
4445 #endif
4446 					tp3_hdr->hv1.tp_vlan_tci);
4447 			if (ret == 1) {
4448 				pkts++;
4449 				handlep->packets_read++;
4450 			} else if (ret < 0) {
4451 				handlep->current_packet = NULL;
4452 				return ret;
4453 			}
4454 			handlep->current_packet += tp3_hdr->tp_next_offset;
4455 			handlep->packets_left--;
4456 		}
4457 
4458 		if (handlep->packets_left <= 0) {
4459 			/* next block */
4460 			h.h3->hdr.bh1.block_status = TP_STATUS_KERNEL;
4461 
4462 			if (++handle->offset >= handle->cc)
4463 				handle->offset = 0;
4464 
4465 			handlep->current_packet = NULL;
4466 		}
4467 
4468 		/* check for break loop condition*/
4469 		if (handle->break_loop) {
4470 			handle->break_loop = 0;
4471 			return PCAP_ERROR_BREAK;
4472 		}
4473 	}
4474 	return pkts;
4475 }
4476 #endif /* HAVE_TPACKET3 */
4477 
4478 static int
4479 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
4480 {
4481 	struct pcap_linux *handlep = handle->priv;
4482 	int n, offset;
4483 	int ret;
4484 
4485 	/*
4486 	 * Don't rewrite "ret" instructions; we don't need to, as
4487 	 * we're not reading packets with recvmsg(), and we don't
4488 	 * want to, as, by not rewriting them, the kernel can avoid
4489 	 * copying extra data.
4490 	 */
4491 	ret = pcap_setfilter_linux_common(handle, filter, 1);
4492 	if (ret < 0)
4493 		return ret;
4494 
4495 	/* if the kernel filter is enabled, we need to apply the filter on
4496 	 * all packets present into the ring. Get an upper bound of their number
4497 	 */
4498 	if (!handlep->filtering_in_kernel)
4499 		return ret;
4500 
4501 	/* walk the ring backward and count the free slot */
4502 	offset = handle->offset;
4503 	if (--handle->offset < 0)
4504 		handle->offset = handle->cc - 1;
4505 	for (n=0; n < handle->cc; ++n) {
4506 		if (--handle->offset < 0)
4507 			handle->offset = handle->cc - 1;
4508 		if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL))
4509 			break;
4510 	}
4511 
4512 	/* be careful to not change current ring position */
4513 	handle->offset = offset;
4514 
4515 	/* store the number of packets currently present in the ring */
4516 	handlep->filtering_in_kernel = 1 + (handle->cc - n);
4517 	return ret;
4518 }
4519 
4520 #endif /* HAVE_PACKET_RING */
4521 
4522 
4523 #ifdef HAVE_PF_PACKET_SOCKETS
4524 /*
4525  *  Return the index of the given device name. Fill ebuf and return
4526  *  -1 on failure.
4527  */
4528 static int
4529 iface_get_id(int fd, const char *device, char *ebuf)
4530 {
4531 	struct ifreq	ifr;
4532 
4533 	memset(&ifr, 0, sizeof(ifr));
4534 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4535 
4536 	if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
4537 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4538 			 "SIOCGIFINDEX: %s", pcap_strerror(errno));
4539 		return -1;
4540 	}
4541 
4542 	return ifr.ifr_ifindex;
4543 }
4544 
4545 /*
4546  *  Bind the socket associated with FD to the given device.
4547  *  Return 1 on success, 0 if we should try a SOCK_PACKET socket,
4548  *  or a PCAP_ERROR_ value on a hard error.
4549  */
4550 static int
4551 iface_bind(int fd, int ifindex, char *ebuf)
4552 {
4553 	struct sockaddr_ll	sll;
4554 	int			err;
4555 	socklen_t		errlen = sizeof(err);
4556 
4557 	memset(&sll, 0, sizeof(sll));
4558 	sll.sll_family		= AF_PACKET;
4559 	sll.sll_ifindex		= ifindex;
4560 	sll.sll_protocol	= htons(ETH_P_ALL);
4561 
4562 	if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
4563 		if (errno == ENETDOWN) {
4564 			/*
4565 			 * Return a "network down" indication, so that
4566 			 * the application can report that rather than
4567 			 * saying we had a mysterious failure and
4568 			 * suggest that they report a problem to the
4569 			 * libpcap developers.
4570 			 */
4571 			return PCAP_ERROR_IFACE_NOT_UP;
4572 		} else {
4573 			snprintf(ebuf, PCAP_ERRBUF_SIZE,
4574 				 "bind: %s", pcap_strerror(errno));
4575 			return PCAP_ERROR;
4576 		}
4577 	}
4578 
4579 	/* Any pending errors, e.g., network is down? */
4580 
4581 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
4582 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4583 			"getsockopt: %s", pcap_strerror(errno));
4584 		return 0;
4585 	}
4586 
4587 	if (err == ENETDOWN) {
4588 		/*
4589 		 * Return a "network down" indication, so that
4590 		 * the application can report that rather than
4591 		 * saying we had a mysterious failure and
4592 		 * suggest that they report a problem to the
4593 		 * libpcap developers.
4594 		 */
4595 		return PCAP_ERROR_IFACE_NOT_UP;
4596 	} else if (err > 0) {
4597 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4598 			"bind: %s", pcap_strerror(err));
4599 		return 0;
4600 	}
4601 
4602 	return 1;
4603 }
4604 
4605 #ifdef IW_MODE_MONITOR
4606 /*
4607  * Check whether the device supports the Wireless Extensions.
4608  * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE
4609  * if the device doesn't even exist.
4610  */
4611 static int
4612 has_wext(int sock_fd, const char *device, char *ebuf)
4613 {
4614 	struct iwreq ireq;
4615 
4616 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4617 	    sizeof ireq.ifr_ifrn.ifrn_name);
4618 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4619 	if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
4620 		return 1;	/* yes */
4621 	snprintf(ebuf, PCAP_ERRBUF_SIZE,
4622 	    "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
4623 	if (errno == ENODEV)
4624 		return PCAP_ERROR_NO_SUCH_DEVICE;
4625 	return 0;
4626 }
4627 
4628 /*
4629  * Per me si va ne la citta dolente,
4630  * Per me si va ne l'etterno dolore,
4631  *	...
4632  * Lasciate ogne speranza, voi ch'intrate.
4633  *
4634  * XXX - airmon-ng does special stuff with the Orinoco driver and the
4635  * wlan-ng driver.
4636  */
4637 typedef enum {
4638 	MONITOR_WEXT,
4639 	MONITOR_HOSTAP,
4640 	MONITOR_PRISM,
4641 	MONITOR_PRISM54,
4642 	MONITOR_ACX100,
4643 	MONITOR_RT2500,
4644 	MONITOR_RT2570,
4645 	MONITOR_RT73,
4646 	MONITOR_RTL8XXX
4647 } monitor_type;
4648 
4649 /*
4650  * Use the Wireless Extensions, if we have them, to try to turn monitor mode
4651  * on if it's not already on.
4652  *
4653  * Returns 1 on success, 0 if we don't support the Wireless Extensions
4654  * on this device, or a PCAP_ERROR_ value if we do support them but
4655  * we weren't able to turn monitor mode on.
4656  */
4657 static int
4658 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
4659 {
4660 	/*
4661 	 * XXX - at least some adapters require non-Wireless Extensions
4662 	 * mechanisms to turn monitor mode on.
4663 	 *
4664 	 * Atheros cards might require that a separate "monitor virtual access
4665 	 * point" be created, with later versions of the madwifi driver.
4666 	 * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode
4667 	 * monitor -bssid", which apparently spits out a line "athN"
4668 	 * where "athN" is the monitor mode device.  To leave monitor
4669 	 * mode, it destroys the monitor mode device.
4670 	 *
4671 	 * Some Intel Centrino adapters might require private ioctls to get
4672 	 * radio headers; the ipw2200 and ipw3945 drivers allow you to
4673 	 * configure a separate "rtapN" interface to capture in monitor
4674 	 * mode without preventing the adapter from operating normally.
4675 	 * (airmon-ng doesn't appear to use that, though.)
4676 	 *
4677 	 * It would be Truly Wonderful if mac80211 and nl80211 cleaned this
4678 	 * up, and if all drivers were converted to mac80211 drivers.
4679 	 *
4680 	 * If interface {if} is a mac80211 driver, the file
4681 	 * /sys/class/net/{if}/phy80211 is a symlink to
4682 	 * /sys/class/ieee80211/{phydev}, for some {phydev}.
4683 	 *
4684 	 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
4685 	 * least, has a "wmaster0" device and a "wlan0" device; the
4686 	 * latter is the one with the IP address.  Both show up in
4687 	 * "tcpdump -D" output.  Capturing on the wmaster0 device
4688 	 * captures with 802.11 headers.
4689 	 *
4690 	 * airmon-ng searches through /sys/class/net for devices named
4691 	 * monN, starting with mon0; as soon as one *doesn't* exist,
4692 	 * it chooses that as the monitor device name.  If the "iw"
4693 	 * command exists, it does "iw dev {if} interface add {monif}
4694 	 * type monitor", where {monif} is the monitor device.  It
4695 	 * then (sigh) sleeps .1 second, and then configures the
4696 	 * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
4697 	 * is a file, it writes {mondev}, without a newline, to that file,
4698 	 * and again (sigh) sleeps .1 second, and then iwconfig's that
4699 	 * device into monitor mode and configures it up.  Otherwise,
4700 	 * you can't do monitor mode.
4701 	 *
4702 	 * All these devices are "glued" together by having the
4703 	 * /sys/class/net/{device}/phy80211 links pointing to the same
4704 	 * place, so, given a wmaster, wlan, or mon device, you can
4705 	 * find the other devices by looking for devices with
4706 	 * the same phy80211 link.
4707 	 *
4708 	 * To turn monitor mode off, delete the monitor interface,
4709 	 * either with "iw dev {monif} interface del" or by sending
4710 	 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
4711 	 *
4712 	 * Note: if you try to create a monitor device named "monN", and
4713 	 * there's already a "monN" device, it fails, as least with
4714 	 * the netlink interface (which is what iw uses), with a return
4715 	 * value of -ENFILE.  (Return values are negative errnos.)  We
4716 	 * could probably use that to find an unused device.
4717 	 */
4718 	struct pcap_linux *handlep = handle->priv;
4719 	int err;
4720 	struct iwreq ireq;
4721 	struct iw_priv_args *priv;
4722 	monitor_type montype;
4723 	int i;
4724 	__u32 cmd;
4725 	struct ifreq ifr;
4726 	int oldflags;
4727 	int args[2];
4728 	int channel;
4729 
4730 	/*
4731 	 * Does this device *support* the Wireless Extensions?
4732 	 */
4733 	err = has_wext(sock_fd, device, handle->errbuf);
4734 	if (err <= 0)
4735 		return err;	/* either it doesn't or the device doesn't even exist */
4736 	/*
4737 	 * Start out assuming we have no private extensions to control
4738 	 * radio metadata.
4739 	 */
4740 	montype = MONITOR_WEXT;
4741 	cmd = 0;
4742 
4743 	/*
4744 	 * Try to get all the Wireless Extensions private ioctls
4745 	 * supported by this device.
4746 	 *
4747 	 * First, get the size of the buffer we need, by supplying no
4748 	 * buffer and a length of 0.  If the device supports private
4749 	 * ioctls, it should return E2BIG, with ireq.u.data.length set
4750 	 * to the length we need.  If it doesn't support them, it should
4751 	 * return EOPNOTSUPP.
4752 	 */
4753 	memset(&ireq, 0, sizeof ireq);
4754 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4755 	    sizeof ireq.ifr_ifrn.ifrn_name);
4756 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4757 	ireq.u.data.pointer = (void *)args;
4758 	ireq.u.data.length = 0;
4759 	ireq.u.data.flags = 0;
4760 	if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
4761 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4762 		    "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!",
4763 		    device);
4764 		return PCAP_ERROR;
4765 	}
4766 	if (errno != EOPNOTSUPP) {
4767 		/*
4768 		 * OK, it's not as if there are no private ioctls.
4769 		 */
4770 		if (errno != E2BIG) {
4771 			/*
4772 			 * Failed.
4773 			 */
4774 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4775 			    "%s: SIOCGIWPRIV: %s", device,
4776 			    pcap_strerror(errno));
4777 			return PCAP_ERROR;
4778 		}
4779 
4780 		/*
4781 		 * OK, try to get the list of private ioctls.
4782 		 */
4783 		priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args));
4784 		if (priv == NULL) {
4785 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4786 			    "malloc: %s", pcap_strerror(errno));
4787 			return PCAP_ERROR;
4788 		}
4789 		ireq.u.data.pointer = (void *)priv;
4790 		if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) {
4791 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4792 			    "%s: SIOCGIWPRIV: %s", device,
4793 			    pcap_strerror(errno));
4794 			free(priv);
4795 			return PCAP_ERROR;
4796 		}
4797 
4798 		/*
4799 		 * Look for private ioctls to turn monitor mode on or, if
4800 		 * monitor mode is on, to set the header type.
4801 		 */
4802 		for (i = 0; i < ireq.u.data.length; i++) {
4803 			if (strcmp(priv[i].name, "monitor_type") == 0) {
4804 				/*
4805 				 * Hostap driver, use this one.
4806 				 * Set monitor mode first.
4807 				 * You can set it to 0 to get DLT_IEEE80211,
4808 				 * 1 to get DLT_PRISM, 2 to get
4809 				 * DLT_IEEE80211_RADIO_AVS, and, with more
4810 				 * recent versions of the driver, 3 to get
4811 				 * DLT_IEEE80211_RADIO.
4812 				 */
4813 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4814 					break;
4815 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4816 					break;
4817 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4818 					break;
4819 				montype = MONITOR_HOSTAP;
4820 				cmd = priv[i].cmd;
4821 				break;
4822 			}
4823 			if (strcmp(priv[i].name, "set_prismhdr") == 0) {
4824 				/*
4825 				 * Prism54 driver, use this one.
4826 				 * Set monitor mode first.
4827 				 * You can set it to 2 to get DLT_IEEE80211
4828 				 * or 3 or get DLT_PRISM.
4829 				 */
4830 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4831 					break;
4832 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4833 					break;
4834 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4835 					break;
4836 				montype = MONITOR_PRISM54;
4837 				cmd = priv[i].cmd;
4838 				break;
4839 			}
4840 			if (strcmp(priv[i].name, "forceprismheader") == 0) {
4841 				/*
4842 				 * RT2570 driver, use this one.
4843 				 * Do this after turning monitor mode on.
4844 				 * You can set it to 1 to get DLT_PRISM or 2
4845 				 * to get DLT_IEEE80211.
4846 				 */
4847 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4848 					break;
4849 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4850 					break;
4851 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4852 					break;
4853 				montype = MONITOR_RT2570;
4854 				cmd = priv[i].cmd;
4855 				break;
4856 			}
4857 			if (strcmp(priv[i].name, "forceprism") == 0) {
4858 				/*
4859 				 * RT73 driver, use this one.
4860 				 * Do this after turning monitor mode on.
4861 				 * Its argument is a *string*; you can
4862 				 * set it to "1" to get DLT_PRISM or "2"
4863 				 * to get DLT_IEEE80211.
4864 				 */
4865 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR)
4866 					break;
4867 				if (priv[i].set_args & IW_PRIV_SIZE_FIXED)
4868 					break;
4869 				montype = MONITOR_RT73;
4870 				cmd = priv[i].cmd;
4871 				break;
4872 			}
4873 			if (strcmp(priv[i].name, "prismhdr") == 0) {
4874 				/*
4875 				 * One of the RTL8xxx drivers, use this one.
4876 				 * It can only be done after monitor mode
4877 				 * has been turned on.  You can set it to 1
4878 				 * to get DLT_PRISM or 0 to get DLT_IEEE80211.
4879 				 */
4880 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4881 					break;
4882 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4883 					break;
4884 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4885 					break;
4886 				montype = MONITOR_RTL8XXX;
4887 				cmd = priv[i].cmd;
4888 				break;
4889 			}
4890 			if (strcmp(priv[i].name, "rfmontx") == 0) {
4891 				/*
4892 				 * RT2500 or RT61 driver, use this one.
4893 				 * It has one one-byte parameter; set
4894 				 * u.data.length to 1 and u.data.pointer to
4895 				 * point to the parameter.
4896 				 * It doesn't itself turn monitor mode on.
4897 				 * You can set it to 1 to allow transmitting
4898 				 * in monitor mode(?) and get DLT_IEEE80211,
4899 				 * or set it to 0 to disallow transmitting in
4900 				 * monitor mode(?) and get DLT_PRISM.
4901 				 */
4902 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4903 					break;
4904 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2)
4905 					break;
4906 				montype = MONITOR_RT2500;
4907 				cmd = priv[i].cmd;
4908 				break;
4909 			}
4910 			if (strcmp(priv[i].name, "monitor") == 0) {
4911 				/*
4912 				 * Either ACX100 or hostap, use this one.
4913 				 * It turns monitor mode on.
4914 				 * If it takes two arguments, it's ACX100;
4915 				 * the first argument is 1 for DLT_PRISM
4916 				 * or 2 for DLT_IEEE80211, and the second
4917 				 * argument is the channel on which to
4918 				 * run.  If it takes one argument, it's
4919 				 * HostAP, and the argument is 2 for
4920 				 * DLT_IEEE80211 and 3 for DLT_PRISM.
4921 				 *
4922 				 * If we see this, we don't quit, as this
4923 				 * might be a version of the hostap driver
4924 				 * that also supports "monitor_type".
4925 				 */
4926 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4927 					break;
4928 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4929 					break;
4930 				switch (priv[i].set_args & IW_PRIV_SIZE_MASK) {
4931 
4932 				case 1:
4933 					montype = MONITOR_PRISM;
4934 					cmd = priv[i].cmd;
4935 					break;
4936 
4937 				case 2:
4938 					montype = MONITOR_ACX100;
4939 					cmd = priv[i].cmd;
4940 					break;
4941 
4942 				default:
4943 					break;
4944 				}
4945 			}
4946 		}
4947 		free(priv);
4948 	}
4949 
4950 	/*
4951 	 * XXX - ipw3945?  islism?
4952 	 */
4953 
4954 	/*
4955 	 * Get the old mode.
4956 	 */
4957 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4958 	    sizeof ireq.ifr_ifrn.ifrn_name);
4959 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4960 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) {
4961 		/*
4962 		 * We probably won't be able to set the mode, either.
4963 		 */
4964 		return PCAP_ERROR_RFMON_NOTSUP;
4965 	}
4966 
4967 	/*
4968 	 * Is it currently in monitor mode?
4969 	 */
4970 	if (ireq.u.mode == IW_MODE_MONITOR) {
4971 		/*
4972 		 * Yes.  Just leave things as they are.
4973 		 * We don't offer multiple link-layer types, as
4974 		 * changing the link-layer type out from under
4975 		 * somebody else capturing in monitor mode would
4976 		 * be considered rude.
4977 		 */
4978 		return 1;
4979 	}
4980 	/*
4981 	 * No.  We have to put the adapter into rfmon mode.
4982 	 */
4983 
4984 	/*
4985 	 * If we haven't already done so, arrange to have
4986 	 * "pcap_close_all()" called when we exit.
4987 	 */
4988 	if (!pcap_do_addexit(handle)) {
4989 		/*
4990 		 * "atexit()" failed; don't put the interface
4991 		 * in rfmon mode, just give up.
4992 		 */
4993 		return PCAP_ERROR_RFMON_NOTSUP;
4994 	}
4995 
4996 	/*
4997 	 * Save the old mode.
4998 	 */
4999 	handlep->oldmode = ireq.u.mode;
5000 
5001 	/*
5002 	 * Put the adapter in rfmon mode.  How we do this depends
5003 	 * on whether we have a special private ioctl or not.
5004 	 */
5005 	if (montype == MONITOR_PRISM) {
5006 		/*
5007 		 * We have the "monitor" private ioctl, but none of
5008 		 * the other private ioctls.  Use this, and select
5009 		 * the Prism header.
5010 		 *
5011 		 * If it fails, just fall back on SIOCSIWMODE.
5012 		 */
5013 		memset(&ireq, 0, sizeof ireq);
5014 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5015 		    sizeof ireq.ifr_ifrn.ifrn_name);
5016 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5017 		ireq.u.data.length = 1;	/* 1 argument */
5018 		args[0] = 3;	/* request Prism header */
5019 		memcpy(ireq.u.name, args, IFNAMSIZ);
5020 		if (ioctl(sock_fd, cmd, &ireq) != -1) {
5021 			/*
5022 			 * Success.
5023 			 * Note that we have to put the old mode back
5024 			 * when we close the device.
5025 			 */
5026 			handlep->must_do_on_close |= MUST_CLEAR_RFMON;
5027 
5028 			/*
5029 			 * Add this to the list of pcaps to close
5030 			 * when we exit.
5031 			 */
5032 			pcap_add_to_pcaps_to_close(handle);
5033 
5034 			return 1;
5035 		}
5036 
5037 		/*
5038 		 * Failure.  Fall back on SIOCSIWMODE.
5039 		 */
5040 	}
5041 
5042 	/*
5043 	 * First, take the interface down if it's up; otherwise, we
5044 	 * might get EBUSY.
5045 	 */
5046 	memset(&ifr, 0, sizeof(ifr));
5047 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5048 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
5049 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5050 		    "%s: Can't get flags: %s", device, strerror(errno));
5051 		return PCAP_ERROR;
5052 	}
5053 	oldflags = 0;
5054 	if (ifr.ifr_flags & IFF_UP) {
5055 		oldflags = ifr.ifr_flags;
5056 		ifr.ifr_flags &= ~IFF_UP;
5057 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5058 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5059 			    "%s: Can't set flags: %s", device, strerror(errno));
5060 			return PCAP_ERROR;
5061 		}
5062 	}
5063 
5064 	/*
5065 	 * Then turn monitor mode on.
5066 	 */
5067 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
5068 	    sizeof ireq.ifr_ifrn.ifrn_name);
5069 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5070 	ireq.u.mode = IW_MODE_MONITOR;
5071 	if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) {
5072 		/*
5073 		 * Scientist, you've failed.
5074 		 * Bring the interface back up if we shut it down.
5075 		 */
5076 		ifr.ifr_flags = oldflags;
5077 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5078 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5079 			    "%s: Can't set flags: %s", device, strerror(errno));
5080 			return PCAP_ERROR;
5081 		}
5082 		return PCAP_ERROR_RFMON_NOTSUP;
5083 	}
5084 
5085 	/*
5086 	 * XXX - airmon-ng does "iwconfig {if} key off" after setting
5087 	 * monitor mode and setting the channel, and then does
5088 	 * "iwconfig up".
5089 	 */
5090 
5091 	/*
5092 	 * Now select the appropriate radio header.
5093 	 */
5094 	switch (montype) {
5095 
5096 	case MONITOR_WEXT:
5097 		/*
5098 		 * We don't have any private ioctl to set the header.
5099 		 */
5100 		break;
5101 
5102 	case MONITOR_HOSTAP:
5103 		/*
5104 		 * Try to select the radiotap header.
5105 		 */
5106 		memset(&ireq, 0, sizeof ireq);
5107 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5108 		    sizeof ireq.ifr_ifrn.ifrn_name);
5109 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5110 		args[0] = 3;	/* request radiotap header */
5111 		memcpy(ireq.u.name, args, sizeof (int));
5112 		if (ioctl(sock_fd, cmd, &ireq) != -1)
5113 			break;	/* success */
5114 
5115 		/*
5116 		 * That failed.  Try to select the AVS header.
5117 		 */
5118 		memset(&ireq, 0, sizeof ireq);
5119 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5120 		    sizeof ireq.ifr_ifrn.ifrn_name);
5121 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5122 		args[0] = 2;	/* request AVS header */
5123 		memcpy(ireq.u.name, args, sizeof (int));
5124 		if (ioctl(sock_fd, cmd, &ireq) != -1)
5125 			break;	/* success */
5126 
5127 		/*
5128 		 * That failed.  Try to select the Prism header.
5129 		 */
5130 		memset(&ireq, 0, sizeof ireq);
5131 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5132 		    sizeof ireq.ifr_ifrn.ifrn_name);
5133 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5134 		args[0] = 1;	/* request Prism header */
5135 		memcpy(ireq.u.name, args, sizeof (int));
5136 		ioctl(sock_fd, cmd, &ireq);
5137 		break;
5138 
5139 	case MONITOR_PRISM:
5140 		/*
5141 		 * The private ioctl failed.
5142 		 */
5143 		break;
5144 
5145 	case MONITOR_PRISM54:
5146 		/*
5147 		 * Select the Prism header.
5148 		 */
5149 		memset(&ireq, 0, sizeof ireq);
5150 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5151 		    sizeof ireq.ifr_ifrn.ifrn_name);
5152 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5153 		args[0] = 3;	/* request Prism header */
5154 		memcpy(ireq.u.name, args, sizeof (int));
5155 		ioctl(sock_fd, cmd, &ireq);
5156 		break;
5157 
5158 	case MONITOR_ACX100:
5159 		/*
5160 		 * Get the current channel.
5161 		 */
5162 		memset(&ireq, 0, sizeof ireq);
5163 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5164 		    sizeof ireq.ifr_ifrn.ifrn_name);
5165 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5166 		if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) {
5167 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5168 			    "%s: SIOCGIWFREQ: %s", device,
5169 			    pcap_strerror(errno));
5170 			return PCAP_ERROR;
5171 		}
5172 		channel = ireq.u.freq.m;
5173 
5174 		/*
5175 		 * Select the Prism header, and set the channel to the
5176 		 * current value.
5177 		 */
5178 		memset(&ireq, 0, sizeof ireq);
5179 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5180 		    sizeof ireq.ifr_ifrn.ifrn_name);
5181 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5182 		args[0] = 1;		/* request Prism header */
5183 		args[1] = channel;	/* set channel */
5184 		memcpy(ireq.u.name, args, 2*sizeof (int));
5185 		ioctl(sock_fd, cmd, &ireq);
5186 		break;
5187 
5188 	case MONITOR_RT2500:
5189 		/*
5190 		 * Disallow transmission - that turns on the
5191 		 * Prism header.
5192 		 */
5193 		memset(&ireq, 0, sizeof ireq);
5194 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5195 		    sizeof ireq.ifr_ifrn.ifrn_name);
5196 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5197 		args[0] = 0;	/* disallow transmitting */
5198 		memcpy(ireq.u.name, args, sizeof (int));
5199 		ioctl(sock_fd, cmd, &ireq);
5200 		break;
5201 
5202 	case MONITOR_RT2570:
5203 		/*
5204 		 * Force the Prism header.
5205 		 */
5206 		memset(&ireq, 0, sizeof ireq);
5207 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5208 		    sizeof ireq.ifr_ifrn.ifrn_name);
5209 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5210 		args[0] = 1;	/* request Prism header */
5211 		memcpy(ireq.u.name, args, sizeof (int));
5212 		ioctl(sock_fd, cmd, &ireq);
5213 		break;
5214 
5215 	case MONITOR_RT73:
5216 		/*
5217 		 * Force the Prism header.
5218 		 */
5219 		memset(&ireq, 0, sizeof ireq);
5220 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5221 		    sizeof ireq.ifr_ifrn.ifrn_name);
5222 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5223 		ireq.u.data.length = 1;	/* 1 argument */
5224 		ireq.u.data.pointer = "1";
5225 		ireq.u.data.flags = 0;
5226 		ioctl(sock_fd, cmd, &ireq);
5227 		break;
5228 
5229 	case MONITOR_RTL8XXX:
5230 		/*
5231 		 * Force the Prism header.
5232 		 */
5233 		memset(&ireq, 0, sizeof ireq);
5234 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
5235 		    sizeof ireq.ifr_ifrn.ifrn_name);
5236 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
5237 		args[0] = 1;	/* request Prism header */
5238 		memcpy(ireq.u.name, args, sizeof (int));
5239 		ioctl(sock_fd, cmd, &ireq);
5240 		break;
5241 	}
5242 
5243 	/*
5244 	 * Now bring the interface back up if we brought it down.
5245 	 */
5246 	if (oldflags != 0) {
5247 		ifr.ifr_flags = oldflags;
5248 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5249 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5250 			    "%s: Can't set flags: %s", device, strerror(errno));
5251 
5252 			/*
5253 			 * At least try to restore the old mode on the
5254 			 * interface.
5255 			 */
5256 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
5257 				/*
5258 				 * Scientist, you've failed.
5259 				 */
5260 				fprintf(stderr,
5261 				    "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n"
5262 				    "Please adjust manually.\n",
5263 				    strerror(errno));
5264 			}
5265 			return PCAP_ERROR;
5266 		}
5267 	}
5268 
5269 	/*
5270 	 * Note that we have to put the old mode back when we
5271 	 * close the device.
5272 	 */
5273 	handlep->must_do_on_close |= MUST_CLEAR_RFMON;
5274 
5275 	/*
5276 	 * Add this to the list of pcaps to close when we exit.
5277 	 */
5278 	pcap_add_to_pcaps_to_close(handle);
5279 
5280 	return 1;
5281 }
5282 #endif /* IW_MODE_MONITOR */
5283 
5284 /*
5285  * Try various mechanisms to enter monitor mode.
5286  */
5287 static int
5288 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
5289 {
5290 #if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
5291 	int ret;
5292 #endif
5293 
5294 #ifdef HAVE_LIBNL
5295 	ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
5296 	if (ret < 0)
5297 		return ret;	/* error attempting to do so */
5298 	if (ret == 1)
5299 		return 1;	/* success */
5300 #endif /* HAVE_LIBNL */
5301 
5302 #ifdef IW_MODE_MONITOR
5303 	ret = enter_rfmon_mode_wext(handle, sock_fd, device);
5304 	if (ret < 0)
5305 		return ret;	/* error attempting to do so */
5306 	if (ret == 1)
5307 		return 1;	/* success */
5308 #endif /* IW_MODE_MONITOR */
5309 
5310 	/*
5311 	 * Either none of the mechanisms we know about work or none
5312 	 * of those mechanisms are available, so we can't do monitor
5313 	 * mode.
5314 	 */
5315 	return 0;
5316 }
5317 
5318 /*
5319  * Find out if we have any form of fragmentation/reassembly offloading.
5320  *
5321  * We do so using SIOCETHTOOL checking for various types of offloading;
5322  * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
5323  * of the types of offloading, there's nothing we can do to check, so
5324  * we just say "no, we don't".
5325  */
5326 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
5327 static int
5328 iface_ethtool_ioctl(pcap_t *handle, int cmd, const char *cmdname)
5329 {
5330 	struct ifreq	ifr;
5331 	struct ethtool_value eval;
5332 
5333 	memset(&ifr, 0, sizeof(ifr));
5334 	strncpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name));
5335 	eval.cmd = cmd;
5336 	eval.data = 0;
5337 	ifr.ifr_data = (caddr_t)&eval;
5338 	if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
5339 		if (errno == EOPNOTSUPP || errno == EINVAL) {
5340 			/*
5341 			 * OK, let's just return 0, which, in our
5342 			 * case, either means "no, what we're asking
5343 			 * about is not enabled" or "all the flags
5344 			 * are clear (i.e., nothing is enabled)".
5345 			 */
5346 			return 0;
5347 		}
5348 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5349 		    "%s: SIOETHTOOL(%s) ioctl failed: %s", handle->opt.source,
5350 		    cmdname, strerror(errno));
5351 		return -1;
5352 	}
5353 	return eval.data;
5354 }
5355 
5356 static int
5357 iface_get_offload(pcap_t *handle)
5358 {
5359 	int ret;
5360 
5361 #ifdef ETHTOOL_GTSO
5362 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO");
5363 	if (ret == -1)
5364 		return -1;
5365 	if (ret)
5366 		return 1;	/* TCP segmentation offloading on */
5367 #endif
5368 
5369 #ifdef ETHTOOL_GUFO
5370 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO");
5371 	if (ret == -1)
5372 		return -1;
5373 	if (ret)
5374 		return 1;	/* UDP fragmentation offloading on */
5375 #endif
5376 
5377 #ifdef ETHTOOL_GGSO
5378 	/*
5379 	 * XXX - will this cause large unsegmented packets to be
5380 	 * handed to PF_PACKET sockets on transmission?  If not,
5381 	 * this need not be checked.
5382 	 */
5383 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO");
5384 	if (ret == -1)
5385 		return -1;
5386 	if (ret)
5387 		return 1;	/* generic segmentation offloading on */
5388 #endif
5389 
5390 #ifdef ETHTOOL_GFLAGS
5391 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
5392 	if (ret == -1)
5393 		return -1;
5394 	if (ret & ETH_FLAG_LRO)
5395 		return 1;	/* large receive offloading on */
5396 #endif
5397 
5398 #ifdef ETHTOOL_GGRO
5399 	/*
5400 	 * XXX - will this cause large reassembled packets to be
5401 	 * handed to PF_PACKET sockets on receipt?  If not,
5402 	 * this need not be checked.
5403 	 */
5404 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO");
5405 	if (ret == -1)
5406 		return -1;
5407 	if (ret)
5408 		return 1;	/* generic (large) receive offloading on */
5409 #endif
5410 
5411 	return 0;
5412 }
5413 #else /* SIOCETHTOOL */
5414 static int
5415 iface_get_offload(pcap_t *handle _U_)
5416 {
5417 	/*
5418 	 * XXX - do we need to get this information if we don't
5419 	 * have the ethtool ioctls?  If so, how do we do that?
5420 	 */
5421 	return 0;
5422 }
5423 #endif /* SIOCETHTOOL */
5424 
5425 #endif /* HAVE_PF_PACKET_SOCKETS */
5426 
5427 /* ===== Functions to interface to the older kernels ================== */
5428 
5429 /*
5430  * Try to open a packet socket using the old kernel interface.
5431  * Returns 1 on success and a PCAP_ERROR_ value on an error.
5432  */
5433 static int
5434 activate_old(pcap_t *handle)
5435 {
5436 	struct pcap_linux *handlep = handle->priv;
5437 	int		arptype;
5438 	struct ifreq	ifr;
5439 	const char	*device = handle->opt.source;
5440 	struct utsname	utsname;
5441 	int		mtu;
5442 
5443 	/* Open the socket */
5444 
5445 	handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
5446 	if (handle->fd == -1) {
5447 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5448 			 "socket: %s", pcap_strerror(errno));
5449 		if (errno == EPERM || errno == EACCES) {
5450 			/*
5451 			 * You don't have permission to open the
5452 			 * socket.
5453 			 */
5454 			return PCAP_ERROR_PERM_DENIED;
5455 		} else {
5456 			/*
5457 			 * Other error.
5458 			 */
5459 			return PCAP_ERROR;
5460 		}
5461 	}
5462 
5463 	/* It worked - we are using the old interface */
5464 	handlep->sock_packet = 1;
5465 
5466 	/* ...which means we get the link-layer header. */
5467 	handlep->cooked = 0;
5468 
5469 	/* Bind to the given device */
5470 
5471 	if (strcmp(device, "any") == 0) {
5472 		strncpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
5473 			PCAP_ERRBUF_SIZE);
5474 		return PCAP_ERROR;
5475 	}
5476 	if (iface_bind_old(handle->fd, device, handle->errbuf) == -1)
5477 		return PCAP_ERROR;
5478 
5479 	/*
5480 	 * Try to get the link-layer type.
5481 	 */
5482 	arptype = iface_get_arptype(handle->fd, device, handle->errbuf);
5483 	if (arptype < 0)
5484 		return PCAP_ERROR;
5485 
5486 	/*
5487 	 * Try to find the DLT_ type corresponding to that
5488 	 * link-layer type.
5489 	 */
5490 	map_arphrd_to_dlt(handle, arptype, 0);
5491 	if (handle->linktype == -1) {
5492 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5493 			 "unknown arptype %d", arptype);
5494 		return PCAP_ERROR;
5495 	}
5496 
5497 	/* Go to promisc mode if requested */
5498 
5499 	if (handle->opt.promisc) {
5500 		memset(&ifr, 0, sizeof(ifr));
5501 		strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5502 		if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
5503 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5504 				 "SIOCGIFFLAGS: %s", pcap_strerror(errno));
5505 			return PCAP_ERROR;
5506 		}
5507 		if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
5508 			/*
5509 			 * Promiscuous mode isn't currently on,
5510 			 * so turn it on, and remember that
5511 			 * we should turn it off when the
5512 			 * pcap_t is closed.
5513 			 */
5514 
5515 			/*
5516 			 * If we haven't already done so, arrange
5517 			 * to have "pcap_close_all()" called when
5518 			 * we exit.
5519 			 */
5520 			if (!pcap_do_addexit(handle)) {
5521 				/*
5522 				 * "atexit()" failed; don't put
5523 				 * the interface in promiscuous
5524 				 * mode, just give up.
5525 				 */
5526 				return PCAP_ERROR;
5527 			}
5528 
5529 			ifr.ifr_flags |= IFF_PROMISC;
5530 			if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
5531 			        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5532 					 "SIOCSIFFLAGS: %s",
5533 					 pcap_strerror(errno));
5534 				return PCAP_ERROR;
5535 			}
5536 			handlep->must_do_on_close |= MUST_CLEAR_PROMISC;
5537 
5538 			/*
5539 			 * Add this to the list of pcaps
5540 			 * to close when we exit.
5541 			 */
5542 			pcap_add_to_pcaps_to_close(handle);
5543 		}
5544 	}
5545 
5546 	/*
5547 	 * Compute the buffer size.
5548 	 *
5549 	 * We're using SOCK_PACKET, so this might be a 2.0[.x]
5550 	 * kernel, and might require special handling - check.
5551 	 */
5552 	if (uname(&utsname) < 0 ||
5553 	    strncmp(utsname.release, "2.0", 3) == 0) {
5554 		/*
5555 		 * Either we couldn't find out what kernel release
5556 		 * this is, or it's a 2.0[.x] kernel.
5557 		 *
5558 		 * In the 2.0[.x] kernel, a "recvfrom()" on
5559 		 * a SOCK_PACKET socket, with MSG_TRUNC set, will
5560 		 * return the number of bytes read, so if we pass
5561 		 * a length based on the snapshot length, it'll
5562 		 * return the number of bytes from the packet
5563 		 * copied to userland, not the actual length
5564 		 * of the packet.
5565 		 *
5566 		 * This means that, for example, the IP dissector
5567 		 * in tcpdump will get handed a packet length less
5568 		 * than the length in the IP header, and will
5569 		 * complain about "truncated-ip".
5570 		 *
5571 		 * So we don't bother trying to copy from the
5572 		 * kernel only the bytes in which we're interested,
5573 		 * but instead copy them all, just as the older
5574 		 * versions of libpcap for Linux did.
5575 		 *
5576 		 * The buffer therefore needs to be big enough to
5577 		 * hold the largest packet we can get from this
5578 		 * device.  Unfortunately, we can't get the MRU
5579 		 * of the network; we can only get the MTU.  The
5580 		 * MTU may be too small, in which case a packet larger
5581 		 * than the buffer size will be truncated *and* we
5582 		 * won't get the actual packet size.
5583 		 *
5584 		 * However, if the snapshot length is larger than
5585 		 * the buffer size based on the MTU, we use the
5586 		 * snapshot length as the buffer size, instead;
5587 		 * this means that with a sufficiently large snapshot
5588 		 * length we won't artificially truncate packets
5589 		 * to the MTU-based size.
5590 		 *
5591 		 * This mess just one of many problems with packet
5592 		 * capture on 2.0[.x] kernels; you really want a
5593 		 * 2.2[.x] or later kernel if you want packet capture
5594 		 * to work well.
5595 		 */
5596 		mtu = iface_get_mtu(handle->fd, device, handle->errbuf);
5597 		if (mtu == -1)
5598 			return PCAP_ERROR;
5599 		handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
5600 		if (handle->bufsize < handle->snapshot)
5601 			handle->bufsize = handle->snapshot;
5602 	} else {
5603 		/*
5604 		 * This is a 2.2[.x] or later kernel.
5605 		 *
5606 		 * We can safely pass "recvfrom()" a byte count
5607 		 * based on the snapshot length.
5608 		 */
5609 		handle->bufsize = handle->snapshot;
5610 	}
5611 
5612 	/*
5613 	 * Default value for offset to align link-layer payload
5614 	 * on a 4-byte boundary.
5615 	 */
5616 	handle->offset	 = 0;
5617 
5618 	/*
5619 	 * SOCK_PACKET sockets don't supply information from
5620 	 * stripped VLAN tags.
5621 	 */
5622 	handlep->vlan_offset = -1; /* unknown */
5623 
5624 	return 1;
5625 }
5626 
5627 /*
5628  *  Bind the socket associated with FD to the given device using the
5629  *  interface of the old kernels.
5630  */
5631 static int
5632 iface_bind_old(int fd, const char *device, char *ebuf)
5633 {
5634 	struct sockaddr	saddr;
5635 	int		err;
5636 	socklen_t	errlen = sizeof(err);
5637 
5638 	memset(&saddr, 0, sizeof(saddr));
5639 	strncpy(saddr.sa_data, device, sizeof(saddr.sa_data));
5640 	if (bind(fd, &saddr, sizeof(saddr)) == -1) {
5641 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5642 			 "bind: %s", pcap_strerror(errno));
5643 		return -1;
5644 	}
5645 
5646 	/* Any pending errors, e.g., network is down? */
5647 
5648 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
5649 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5650 			"getsockopt: %s", pcap_strerror(errno));
5651 		return -1;
5652 	}
5653 
5654 	if (err > 0) {
5655 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5656 			"bind: %s", pcap_strerror(err));
5657 		return -1;
5658 	}
5659 
5660 	return 0;
5661 }
5662 
5663 
5664 /* ===== System calls available on all supported kernels ============== */
5665 
5666 /*
5667  *  Query the kernel for the MTU of the given interface.
5668  */
5669 static int
5670 iface_get_mtu(int fd, const char *device, char *ebuf)
5671 {
5672 	struct ifreq	ifr;
5673 
5674 	if (!device)
5675 		return BIGGER_THAN_ALL_MTUS;
5676 
5677 	memset(&ifr, 0, sizeof(ifr));
5678 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5679 
5680 	if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
5681 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5682 			 "SIOCGIFMTU: %s", pcap_strerror(errno));
5683 		return -1;
5684 	}
5685 
5686 	return ifr.ifr_mtu;
5687 }
5688 
5689 /*
5690  *  Get the hardware type of the given interface as ARPHRD_xxx constant.
5691  */
5692 static int
5693 iface_get_arptype(int fd, const char *device, char *ebuf)
5694 {
5695 	struct ifreq	ifr;
5696 
5697 	memset(&ifr, 0, sizeof(ifr));
5698 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5699 
5700 	if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
5701 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5702 			 "SIOCGIFHWADDR: %s", pcap_strerror(errno));
5703 		if (errno == ENODEV) {
5704 			/*
5705 			 * No such device.
5706 			 */
5707 			return PCAP_ERROR_NO_SUCH_DEVICE;
5708 		}
5709 		return PCAP_ERROR;
5710 	}
5711 
5712 	return ifr.ifr_hwaddr.sa_family;
5713 }
5714 
5715 #ifdef SO_ATTACH_FILTER
5716 static int
5717 fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
5718 {
5719 	struct pcap_linux *handlep = handle->priv;
5720 	size_t prog_size;
5721 	register int i;
5722 	register struct bpf_insn *p;
5723 	struct bpf_insn *f;
5724 	int len;
5725 
5726 	/*
5727 	 * Make a copy of the filter, and modify that copy if
5728 	 * necessary.
5729 	 */
5730 	prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
5731 	len = handle->fcode.bf_len;
5732 	f = (struct bpf_insn *)malloc(prog_size);
5733 	if (f == NULL) {
5734 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5735 			 "malloc: %s", pcap_strerror(errno));
5736 		return -1;
5737 	}
5738 	memcpy(f, handle->fcode.bf_insns, prog_size);
5739 	fcode->len = len;
5740 	fcode->filter = (struct sock_filter *) f;
5741 
5742 	for (i = 0; i < len; ++i) {
5743 		p = &f[i];
5744 		/*
5745 		 * What type of instruction is this?
5746 		 */
5747 		switch (BPF_CLASS(p->code)) {
5748 
5749 		case BPF_RET:
5750 			/*
5751 			 * It's a return instruction; are we capturing
5752 			 * in memory-mapped mode?
5753 			 */
5754 			if (!is_mmapped) {
5755 				/*
5756 				 * No; is the snapshot length a constant,
5757 				 * rather than the contents of the
5758 				 * accumulator?
5759 				 */
5760 				if (BPF_MODE(p->code) == BPF_K) {
5761 					/*
5762 					 * Yes - if the value to be returned,
5763 					 * i.e. the snapshot length, is
5764 					 * anything other than 0, make it
5765 					 * 65535, so that the packet is
5766 					 * truncated by "recvfrom()",
5767 					 * not by the filter.
5768 					 *
5769 					 * XXX - there's nothing we can
5770 					 * easily do if it's getting the
5771 					 * value from the accumulator; we'd
5772 					 * have to insert code to force
5773 					 * non-zero values to be 65535.
5774 					 */
5775 					if (p->k != 0)
5776 						p->k = 65535;
5777 				}
5778 			}
5779 			break;
5780 
5781 		case BPF_LD:
5782 		case BPF_LDX:
5783 			/*
5784 			 * It's a load instruction; is it loading
5785 			 * from the packet?
5786 			 */
5787 			switch (BPF_MODE(p->code)) {
5788 
5789 			case BPF_ABS:
5790 			case BPF_IND:
5791 			case BPF_MSH:
5792 				/*
5793 				 * Yes; are we in cooked mode?
5794 				 */
5795 				if (handlep->cooked) {
5796 					/*
5797 					 * Yes, so we need to fix this
5798 					 * instruction.
5799 					 */
5800 					if (fix_offset(p) < 0) {
5801 						/*
5802 						 * We failed to do so.
5803 						 * Return 0, so our caller
5804 						 * knows to punt to userland.
5805 						 */
5806 						return 0;
5807 					}
5808 				}
5809 				break;
5810 			}
5811 			break;
5812 		}
5813 	}
5814 	return 1;	/* we succeeded */
5815 }
5816 
5817 static int
5818 fix_offset(struct bpf_insn *p)
5819 {
5820 	/*
5821 	 * What's the offset?
5822 	 */
5823 	if (p->k >= SLL_HDR_LEN) {
5824 		/*
5825 		 * It's within the link-layer payload; that starts at an
5826 		 * offset of 0, as far as the kernel packet filter is
5827 		 * concerned, so subtract the length of the link-layer
5828 		 * header.
5829 		 */
5830 		p->k -= SLL_HDR_LEN;
5831 	} else if (p->k == 0) {
5832 		/*
5833 		 * It's the packet type field; map it to the special magic
5834 		 * kernel offset for that field.
5835 		 */
5836 		p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
5837 	} else if (p->k == 14) {
5838 		/*
5839 		 * It's the protocol field; map it to the special magic
5840 		 * kernel offset for that field.
5841 		 */
5842 		p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5843 	} else if ((bpf_int32)(p->k) > 0) {
5844 		/*
5845 		 * It's within the header, but it's not one of those
5846 		 * fields; we can't do that in the kernel, so punt
5847 		 * to userland.
5848 		 */
5849 		return -1;
5850 	}
5851 	return 0;
5852 }
5853 
5854 static int
5855 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
5856 {
5857 	int total_filter_on = 0;
5858 	int save_mode;
5859 	int ret;
5860 	int save_errno;
5861 
5862 	/*
5863 	 * The socket filter code doesn't discard all packets queued
5864 	 * up on the socket when the filter is changed; this means
5865 	 * that packets that don't match the new filter may show up
5866 	 * after the new filter is put onto the socket, if those
5867 	 * packets haven't yet been read.
5868 	 *
5869 	 * This means, for example, that if you do a tcpdump capture
5870 	 * with a filter, the first few packets in the capture might
5871 	 * be packets that wouldn't have passed the filter.
5872 	 *
5873 	 * We therefore discard all packets queued up on the socket
5874 	 * when setting a kernel filter.  (This isn't an issue for
5875 	 * userland filters, as the userland filtering is done after
5876 	 * packets are queued up.)
5877 	 *
5878 	 * To flush those packets, we put the socket in read-only mode,
5879 	 * and read packets from the socket until there are no more to
5880 	 * read.
5881 	 *
5882 	 * In order to keep that from being an infinite loop - i.e.,
5883 	 * to keep more packets from arriving while we're draining
5884 	 * the queue - we put the "total filter", which is a filter
5885 	 * that rejects all packets, onto the socket before draining
5886 	 * the queue.
5887 	 *
5888 	 * This code deliberately ignores any errors, so that you may
5889 	 * get bogus packets if an error occurs, rather than having
5890 	 * the filtering done in userland even if it could have been
5891 	 * done in the kernel.
5892 	 */
5893 	if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5894 		       &total_fcode, sizeof(total_fcode)) == 0) {
5895 		char drain[1];
5896 
5897 		/*
5898 		 * Note that we've put the total filter onto the socket.
5899 		 */
5900 		total_filter_on = 1;
5901 
5902 		/*
5903 		 * Save the socket's current mode, and put it in
5904 		 * non-blocking mode; we drain it by reading packets
5905 		 * until we get an error (which is normally a
5906 		 * "nothing more to be read" error).
5907 		 */
5908 		save_mode = fcntl(handle->fd, F_GETFL, 0);
5909 		if (save_mode != -1 &&
5910 		    fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) >= 0) {
5911 			while (recv(handle->fd, &drain, sizeof drain,
5912 			       MSG_TRUNC) >= 0)
5913 				;
5914 			save_errno = errno;
5915 			fcntl(handle->fd, F_SETFL, save_mode);
5916 			if (save_errno != EAGAIN) {
5917 				/* Fatal error */
5918 				reset_kernel_filter(handle);
5919 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5920 				 "recv: %s", pcap_strerror(save_errno));
5921 				return -2;
5922 			}
5923 		}
5924 	}
5925 
5926 	/*
5927 	 * Now attach the new filter.
5928 	 */
5929 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5930 			 fcode, sizeof(*fcode));
5931 	if (ret == -1 && total_filter_on) {
5932 		/*
5933 		 * Well, we couldn't set that filter on the socket,
5934 		 * but we could set the total filter on the socket.
5935 		 *
5936 		 * This could, for example, mean that the filter was
5937 		 * too big to put into the kernel, so we'll have to
5938 		 * filter in userland; in any case, we'll be doing
5939 		 * filtering in userland, so we need to remove the
5940 		 * total filter so we see packets.
5941 		 */
5942 		save_errno = errno;
5943 
5944 		/*
5945 		 * XXX - if this fails, we're really screwed;
5946 		 * we have the total filter on the socket,
5947 		 * and it won't come off.  What do we do then?
5948 		 */
5949 		reset_kernel_filter(handle);
5950 
5951 		errno = save_errno;
5952 	}
5953 	return ret;
5954 }
5955 
5956 static int
5957 reset_kernel_filter(pcap_t *handle)
5958 {
5959 	/*
5960 	 * setsockopt() barfs unless it get a dummy parameter.
5961 	 * valgrind whines unless the value is initialized,
5962 	 * as it has no idea that setsockopt() ignores its
5963 	 * parameter.
5964 	 */
5965 	int dummy = 0;
5966 
5967 	return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
5968 				   &dummy, sizeof(dummy));
5969 }
5970 #endif
5971