xref: /freebsd/contrib/libpcap/pcap-linux.c (revision 38f0b757fd84d17d0fc24739a7cda160c4516d81)
1 /*
2  *  pcap-linux.c: Packet capture interface to the Linux kernel
3  *
4  *  Copyright (c) 2000 Torsten Landschoff <torsten@debian.org>
5  *  		       Sebastian Krahmer  <krahmer@cs.uni-potsdam.de>
6  *
7  *  License: BSD
8  *
9  *  Redistribution and use in source and binary forms, with or without
10  *  modification, are permitted provided that the following conditions
11  *  are met:
12  *
13  *  1. Redistributions of source code must retain the above copyright
14  *     notice, this list of conditions and the following disclaimer.
15  *  2. Redistributions in binary form must reproduce the above copyright
16  *     notice, this list of conditions and the following disclaimer in
17  *     the documentation and/or other materials provided with the
18  *     distribution.
19  *  3. The names of the authors may not be used to endorse or promote
20  *     products derived from this software without specific prior
21  *     written permission.
22  *
23  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
24  *  IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
25  *  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
26  *
27  *  Modifications:     Added PACKET_MMAP support
28  *                     Paolo Abeni <paolo.abeni@email.it>
29  *
30  *                     based on previous works of:
31  *                     Simon Patarin <patarin@cs.unibo.it>
32  *                     Phil Wood <cpw@lanl.gov>
33  *
34  * Monitor-mode support for mac80211 includes code taken from the iw
35  * command; the copyright notice for that code is
36  *
37  * Copyright (c) 2007, 2008	Johannes Berg
38  * Copyright (c) 2007		Andy Lutomirski
39  * Copyright (c) 2007		Mike Kershaw
40  * Copyright (c) 2008		Gábor Stefanik
41  *
42  * All rights reserved.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. The name of the author may not be used to endorse or promote products
53  *    derived from this software without specific prior written permission.
54  *
55  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
56  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
57  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
58  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
59  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
60  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
61  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
62  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
63  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65  * SUCH DAMAGE.
66  */
67 
68 #ifndef lint
69 static const char rcsid[] _U_ =
70     "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.164 2008-12-14 22:00:57 guy Exp $ (LBL)";
71 #endif
72 
73 /*
74  * Known problems with 2.0[.x] kernels:
75  *
76  *   - The loopback device gives every packet twice; on 2.2[.x] kernels,
77  *     if we use PF_PACKET, we can filter out the transmitted version
78  *     of the packet by using data in the "sockaddr_ll" returned by
79  *     "recvfrom()", but, on 2.0[.x] kernels, we have to use
80  *     PF_INET/SOCK_PACKET, which means "recvfrom()" supplies a
81  *     "sockaddr_pkt" which doesn't give us enough information to let
82  *     us do that.
83  *
84  *   - We have to set the interface's IFF_PROMISC flag ourselves, if
85  *     we're to run in promiscuous mode, which means we have to turn
86  *     it off ourselves when we're done; the kernel doesn't keep track
87  *     of how many sockets are listening promiscuously, which means
88  *     it won't get turned off automatically when no sockets are
89  *     listening promiscuously.  We catch "pcap_close()" and, for
90  *     interfaces we put into promiscuous mode, take them out of
91  *     promiscuous mode - which isn't necessarily the right thing to
92  *     do, if another socket also requested promiscuous mode between
93  *     the time when we opened the socket and the time when we close
94  *     the socket.
95  *
96  *   - MSG_TRUNC isn't supported, so you can't specify that "recvfrom()"
97  *     return the amount of data that you could have read, rather than
98  *     the amount that was returned, so we can't just allocate a buffer
99  *     whose size is the snapshot length and pass the snapshot length
100  *     as the byte count, and also pass MSG_TRUNC, so that the return
101  *     value tells us how long the packet was on the wire.
102  *
103  *     This means that, if we want to get the actual size of the packet,
104  *     so we can return it in the "len" field of the packet header,
105  *     we have to read the entire packet, not just the part that fits
106  *     within the snapshot length, and thus waste CPU time copying data
107  *     from the kernel that our caller won't see.
108  *
109  *     We have to get the actual size, and supply it in "len", because
110  *     otherwise, the IP dissector in tcpdump, for example, will complain
111  *     about "truncated-ip", as the packet will appear to have been
112  *     shorter, on the wire, than the IP header said it should have been.
113  */
114 
115 
116 #define _GNU_SOURCE
117 
118 #ifdef HAVE_CONFIG_H
119 #include "config.h"
120 #endif
121 
122 #include <errno.h>
123 #include <stdio.h>
124 #include <stdlib.h>
125 #include <ctype.h>
126 #include <unistd.h>
127 #include <fcntl.h>
128 #include <string.h>
129 #include <limits.h>
130 #include <sys/stat.h>
131 #include <sys/socket.h>
132 #include <sys/ioctl.h>
133 #include <sys/utsname.h>
134 #include <sys/mman.h>
135 #include <linux/if.h>
136 #include <linux/if_packet.h>
137 #include <netinet/in.h>
138 #include <linux/if_ether.h>
139 #include <net/if_arp.h>
140 #include <poll.h>
141 #include <dirent.h>
142 
143 #include "pcap-int.h"
144 #include "pcap/sll.h"
145 #include "pcap/vlan.h"
146 
147 /*
148  * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
149  * sockets rather than SOCK_PACKET sockets.
150  *
151  * To use them, we include <linux/if_packet.h> rather than
152  * <netpacket/packet.h>; we do so because
153  *
154  *	some Linux distributions (e.g., Slackware 4.0) have 2.2 or
155  *	later kernels and libc5, and don't provide a <netpacket/packet.h>
156  *	file;
157  *
158  *	not all versions of glibc2 have a <netpacket/packet.h> file
159  *	that defines stuff needed for some of the 2.4-or-later-kernel
160  *	features, so if the system has a 2.4 or later kernel, we
161  *	still can't use those features.
162  *
163  * We're already including a number of other <linux/XXX.h> headers, and
164  * this code is Linux-specific (no other OS has PF_PACKET sockets as
165  * a raw packet capture mechanism), so it's not as if you gain any
166  * useful portability by using <netpacket/packet.h>
167  *
168  * XXX - should we just include <linux/if_packet.h> even if PF_PACKET
169  * isn't defined?  It only defines one data structure in 2.0.x, so
170  * it shouldn't cause any problems.
171  */
172 #ifdef PF_PACKET
173 # include <linux/if_packet.h>
174 
175  /*
176   * On at least some Linux distributions (for example, Red Hat 5.2),
177   * there's no <netpacket/packet.h> file, but PF_PACKET is defined if
178   * you include <sys/socket.h>, but <linux/if_packet.h> doesn't define
179   * any of the PF_PACKET stuff such as "struct sockaddr_ll" or any of
180   * the PACKET_xxx stuff.
181   *
182   * So we check whether PACKET_HOST is defined, and assume that we have
183   * PF_PACKET sockets only if it is defined.
184   */
185 # ifdef PACKET_HOST
186 #  define HAVE_PF_PACKET_SOCKETS
187 #  ifdef PACKET_AUXDATA
188 #   define HAVE_PACKET_AUXDATA
189 #  endif /* PACKET_AUXDATA */
190 # endif /* PACKET_HOST */
191 
192 
193  /* check for memory mapped access avaibility. We assume every needed
194   * struct is defined if the macro TPACKET_HDRLEN is defined, because it
195   * uses many ring related structs and macros */
196 # ifdef TPACKET_HDRLEN
197 #  define HAVE_PACKET_RING
198 #  ifdef TPACKET2_HDRLEN
199 #   define HAVE_TPACKET2
200 #  else
201 #   define TPACKET_V1	0
202 #  endif /* TPACKET2_HDRLEN */
203 # endif /* TPACKET_HDRLEN */
204 #endif /* PF_PACKET */
205 
206 #ifdef SO_ATTACH_FILTER
207 #include <linux/types.h>
208 #include <linux/filter.h>
209 #endif
210 
211 /*
212  * We need linux/sockios.h if we have linux/net_tstamp.h (for time stamp
213  * specification) or linux/ethtool.h (for ethtool ioctls to get offloading
214  * information).
215  */
216 #if defined(HAVE_LINUX_NET_TSTAMP_H) || defined(HAVE_LINUX_ETHTOOL_H)
217 #include <linux/sockios.h>
218 #endif
219 
220 #ifdef HAVE_LINUX_NET_TSTAMP_H
221 #include <linux/net_tstamp.h>
222 #endif
223 
224 /*
225  * Got Wireless Extensions?
226  */
227 #ifdef HAVE_LINUX_WIRELESS_H
228 #include <linux/wireless.h>
229 #endif /* HAVE_LINUX_WIRELESS_H */
230 
231 /*
232  * Got libnl?
233  */
234 #ifdef HAVE_LIBNL
235 #include <linux/nl80211.h>
236 
237 #include <netlink/genl/genl.h>
238 #include <netlink/genl/family.h>
239 #include <netlink/genl/ctrl.h>
240 #include <netlink/msg.h>
241 #include <netlink/attr.h>
242 #endif /* HAVE_LIBNL */
243 
244 /*
245  * Got ethtool support?
246  */
247 #ifdef HAVE_LINUX_ETHTOOL_H
248 #include <linux/ethtool.h>
249 #endif
250 
251 #ifndef HAVE_SOCKLEN_T
252 typedef int		socklen_t;
253 #endif
254 
255 #ifndef MSG_TRUNC
256 /*
257  * This is being compiled on a system that lacks MSG_TRUNC; define it
258  * with the value it has in the 2.2 and later kernels, so that, on
259  * those kernels, when we pass it in the flags argument to "recvfrom()"
260  * we're passing the right value and thus get the MSG_TRUNC behavior
261  * we want.  (We don't get that behavior on 2.0[.x] kernels, because
262  * they didn't support MSG_TRUNC.)
263  */
264 #define MSG_TRUNC	0x20
265 #endif
266 
267 #ifndef SOL_PACKET
268 /*
269  * This is being compiled on a system that lacks SOL_PACKET; define it
270  * with the value it has in the 2.2 and later kernels, so that we can
271  * set promiscuous mode in the good modern way rather than the old
272  * 2.0-kernel crappy way.
273  */
274 #define SOL_PACKET	263
275 #endif
276 
277 #define MAX_LINKHEADER_SIZE	256
278 
279 /*
280  * When capturing on all interfaces we use this as the buffer size.
281  * Should be bigger then all MTUs that occur in real life.
282  * 64kB should be enough for now.
283  */
284 #define BIGGER_THAN_ALL_MTUS	(64*1024)
285 
286 /*
287  * Prototypes for internal functions and methods.
288  */
289 static void map_arphrd_to_dlt(pcap_t *, int, int);
290 #ifdef HAVE_PF_PACKET_SOCKETS
291 static short int map_packet_type_to_sll_type(short int);
292 #endif
293 static int pcap_activate_linux(pcap_t *);
294 static int activate_old(pcap_t *);
295 static int activate_new(pcap_t *);
296 static int activate_mmap(pcap_t *, int *);
297 static int pcap_can_set_rfmon_linux(pcap_t *);
298 static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *);
299 static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);
300 static int pcap_inject_linux(pcap_t *, const void *, size_t);
301 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
302 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
303 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
304 static void pcap_cleanup_linux(pcap_t *);
305 
306 union thdr {
307 	struct tpacket_hdr	*h1;
308 	struct tpacket2_hdr	*h2;
309 	void			*raw;
310 };
311 
312 #ifdef HAVE_PACKET_RING
313 #define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset])
314 
315 static void destroy_ring(pcap_t *handle);
316 static int create_ring(pcap_t *handle, int *status);
317 static int prepare_tpacket_socket(pcap_t *handle);
318 static void pcap_cleanup_linux_mmap(pcap_t *);
319 static int pcap_read_linux_mmap(pcap_t *, int, pcap_handler , u_char *);
320 static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *);
321 static int pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf);
322 static int pcap_getnonblock_mmap(pcap_t *p, char *errbuf);
323 static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
324     const u_char *bytes);
325 #endif
326 
327 /*
328  * Wrap some ioctl calls
329  */
330 #ifdef HAVE_PF_PACKET_SOCKETS
331 static int	iface_get_id(int fd, const char *device, char *ebuf);
332 #endif /* HAVE_PF_PACKET_SOCKETS */
333 static int	iface_get_mtu(int fd, const char *device, char *ebuf);
334 static int 	iface_get_arptype(int fd, const char *device, char *ebuf);
335 #ifdef HAVE_PF_PACKET_SOCKETS
336 static int 	iface_bind(int fd, int ifindex, char *ebuf);
337 #ifdef IW_MODE_MONITOR
338 static int	has_wext(int sock_fd, const char *device, char *ebuf);
339 #endif /* IW_MODE_MONITOR */
340 static int	enter_rfmon_mode(pcap_t *handle, int sock_fd,
341     const char *device);
342 #endif /* HAVE_PF_PACKET_SOCKETS */
343 static int	iface_get_offload(pcap_t *handle);
344 static int 	iface_bind_old(int fd, const char *device, char *ebuf);
345 
346 #ifdef SO_ATTACH_FILTER
347 static int	fix_program(pcap_t *handle, struct sock_fprog *fcode,
348     int is_mapped);
349 static int	fix_offset(struct bpf_insn *p);
350 static int	set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
351 static int	reset_kernel_filter(pcap_t *handle);
352 
353 static struct sock_filter	total_insn
354 	= BPF_STMT(BPF_RET | BPF_K, 0);
355 static struct sock_fprog	total_fcode
356 	= { 1, &total_insn };
357 #endif /* SO_ATTACH_FILTER */
358 
359 pcap_t *
360 pcap_create_interface(const char *device, char *ebuf)
361 {
362 	pcap_t *handle;
363 
364 	handle = pcap_create_common(device, ebuf);
365 	if (handle == NULL)
366 		return NULL;
367 
368 	handle->activate_op = pcap_activate_linux;
369 	handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
370 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
371 	/*
372 	 * We claim that we support:
373 	 *
374 	 *	software time stamps, with no details about their precision;
375 	 *	hardware time stamps, synced to the host time;
376 	 *	hardware time stamps, not synced to the host time.
377 	 *
378 	 * XXX - we can't ask a device whether it supports
379 	 * hardware time stamps, so we just claim all devices do.
380 	 */
381 	handle->tstamp_type_count = 3;
382 	handle->tstamp_type_list = malloc(3 * sizeof(u_int));
383 	if (handle->tstamp_type_list == NULL) {
384 		free(handle);
385 		return NULL;
386 	}
387 	handle->tstamp_type_list[0] = PCAP_TSTAMP_HOST;
388 	handle->tstamp_type_list[1] = PCAP_TSTAMP_ADAPTER;
389 	handle->tstamp_type_list[2] = PCAP_TSTAMP_ADAPTER_UNSYNCED;
390 #endif
391 
392 	return handle;
393 }
394 
395 #ifdef HAVE_LIBNL
396 /*
397  * If interface {if} is a mac80211 driver, the file
398  * /sys/class/net/{if}/phy80211 is a symlink to
399  * /sys/class/ieee80211/{phydev}, for some {phydev}.
400  *
401  * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
402  * least, has a "wmaster0" device and a "wlan0" device; the
403  * latter is the one with the IP address.  Both show up in
404  * "tcpdump -D" output.  Capturing on the wmaster0 device
405  * captures with 802.11 headers.
406  *
407  * airmon-ng searches through /sys/class/net for devices named
408  * monN, starting with mon0; as soon as one *doesn't* exist,
409  * it chooses that as the monitor device name.  If the "iw"
410  * command exists, it does "iw dev {if} interface add {monif}
411  * type monitor", where {monif} is the monitor device.  It
412  * then (sigh) sleeps .1 second, and then configures the
413  * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
414  * is a file, it writes {mondev}, without a newline, to that file,
415  * and again (sigh) sleeps .1 second, and then iwconfig's that
416  * device into monitor mode and configures it up.  Otherwise,
417  * you can't do monitor mode.
418  *
419  * All these devices are "glued" together by having the
420  * /sys/class/net/{device}/phy80211 links pointing to the same
421  * place, so, given a wmaster, wlan, or mon device, you can
422  * find the other devices by looking for devices with
423  * the same phy80211 link.
424  *
425  * To turn monitor mode off, delete the monitor interface,
426  * either with "iw dev {monif} interface del" or by sending
427  * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
428  *
429  * Note: if you try to create a monitor device named "monN", and
430  * there's already a "monN" device, it fails, as least with
431  * the netlink interface (which is what iw uses), with a return
432  * value of -ENFILE.  (Return values are negative errnos.)  We
433  * could probably use that to find an unused device.
434  *
435  * Yes, you can have multiple monitor devices for a given
436  * physical device.
437 */
438 
439 /*
440  * Is this a mac80211 device?  If so, fill in the physical device path and
441  * return 1; if not, return 0.  On an error, fill in handle->errbuf and
442  * return PCAP_ERROR.
443  */
444 static int
445 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
446     size_t phydev_max_pathlen)
447 {
448 	char *pathstr;
449 	ssize_t bytes_read;
450 
451 	/*
452 	 * Generate the path string for the symlink to the physical device.
453 	 */
454 	if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
455 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
456 		    "%s: Can't generate path name string for /sys/class/net device",
457 		    device);
458 		return PCAP_ERROR;
459 	}
460 	bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
461 	if (bytes_read == -1) {
462 		if (errno == ENOENT || errno == EINVAL) {
463 			/*
464 			 * Doesn't exist, or not a symlink; assume that
465 			 * means it's not a mac80211 device.
466 			 */
467 			free(pathstr);
468 			return 0;
469 		}
470 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
471 		    "%s: Can't readlink %s: %s", device, pathstr,
472 		    strerror(errno));
473 		free(pathstr);
474 		return PCAP_ERROR;
475 	}
476 	free(pathstr);
477 	phydev_path[bytes_read] = '\0';
478 	return 1;
479 }
480 
481 #ifdef HAVE_LIBNL_SOCKETS
482 #define get_nl_errmsg	nl_geterror
483 #else
484 /* libnl 2.x compatibility code */
485 
486 #define nl_sock nl_handle
487 
488 static inline struct nl_handle *
489 nl_socket_alloc(void)
490 {
491 	return nl_handle_alloc();
492 }
493 
494 static inline void
495 nl_socket_free(struct nl_handle *h)
496 {
497 	nl_handle_destroy(h);
498 }
499 
500 #define get_nl_errmsg	strerror
501 
502 static inline int
503 __genl_ctrl_alloc_cache(struct nl_handle *h, struct nl_cache **cache)
504 {
505 	struct nl_cache *tmp = genl_ctrl_alloc_cache(h);
506 	if (!tmp)
507 		return -ENOMEM;
508 	*cache = tmp;
509 	return 0;
510 }
511 #define genl_ctrl_alloc_cache __genl_ctrl_alloc_cache
512 #endif /* !HAVE_LIBNL_SOCKETS */
513 
514 struct nl80211_state {
515 	struct nl_sock *nl_sock;
516 	struct nl_cache *nl_cache;
517 	struct genl_family *nl80211;
518 };
519 
520 static int
521 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
522 {
523 	int err;
524 
525 	state->nl_sock = nl_socket_alloc();
526 	if (!state->nl_sock) {
527 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
528 		    "%s: failed to allocate netlink handle", device);
529 		return PCAP_ERROR;
530 	}
531 
532 	if (genl_connect(state->nl_sock)) {
533 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
534 		    "%s: failed to connect to generic netlink", device);
535 		goto out_handle_destroy;
536 	}
537 
538 	err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
539 	if (err < 0) {
540 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
541 		    "%s: failed to allocate generic netlink cache: %s",
542 		    device, get_nl_errmsg(-err));
543 		goto out_handle_destroy;
544 	}
545 
546 	state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
547 	if (!state->nl80211) {
548 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
549 		    "%s: nl80211 not found", device);
550 		goto out_cache_free;
551 	}
552 
553 	return 0;
554 
555 out_cache_free:
556 	nl_cache_free(state->nl_cache);
557 out_handle_destroy:
558 	nl_socket_free(state->nl_sock);
559 	return PCAP_ERROR;
560 }
561 
562 static void
563 nl80211_cleanup(struct nl80211_state *state)
564 {
565 	genl_family_put(state->nl80211);
566 	nl_cache_free(state->nl_cache);
567 	nl_socket_free(state->nl_sock);
568 }
569 
570 static int
571 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
572     const char *device, const char *mondevice)
573 {
574 	int ifindex;
575 	struct nl_msg *msg;
576 	int err;
577 
578 	ifindex = iface_get_id(sock_fd, device, handle->errbuf);
579 	if (ifindex == -1)
580 		return PCAP_ERROR;
581 
582 	msg = nlmsg_alloc();
583 	if (!msg) {
584 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
585 		    "%s: failed to allocate netlink msg", device);
586 		return PCAP_ERROR;
587 	}
588 
589 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
590 		    0, NL80211_CMD_NEW_INTERFACE, 0);
591 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
592 	NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
593 	NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
594 
595 	err = nl_send_auto_complete(state->nl_sock, msg);
596 	if (err < 0) {
597 #if defined HAVE_LIBNL_NLE
598 		if (err == -NLE_FAILURE) {
599 #else
600 		if (err == -ENFILE) {
601 #endif
602 			/*
603 			 * Device not available; our caller should just
604 			 * keep trying.  (libnl 2.x maps ENFILE to
605 			 * NLE_FAILURE; it can also map other errors
606 			 * to that, but there's not much we can do
607 			 * about that.)
608 			 */
609 			nlmsg_free(msg);
610 			return 0;
611 		} else {
612 			/*
613 			 * Real failure, not just "that device is not
614 			 * available.
615 			 */
616 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
617 			    "%s: nl_send_auto_complete failed adding %s interface: %s",
618 			    device, mondevice, get_nl_errmsg(-err));
619 			nlmsg_free(msg);
620 			return PCAP_ERROR;
621 		}
622 	}
623 	err = nl_wait_for_ack(state->nl_sock);
624 	if (err < 0) {
625 #if defined HAVE_LIBNL_NLE
626 		if (err == -NLE_FAILURE) {
627 #else
628 		if (err == -ENFILE) {
629 #endif
630 			/*
631 			 * Device not available; our caller should just
632 			 * keep trying.  (libnl 2.x maps ENFILE to
633 			 * NLE_FAILURE; it can also map other errors
634 			 * to that, but there's not much we can do
635 			 * about that.)
636 			 */
637 			nlmsg_free(msg);
638 			return 0;
639 		} else {
640 			/*
641 			 * Real failure, not just "that device is not
642 			 * available.
643 			 */
644 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
645 			    "%s: nl_wait_for_ack failed adding %s interface: %s",
646 			    device, mondevice, get_nl_errmsg(-err));
647 			nlmsg_free(msg);
648 			return PCAP_ERROR;
649 		}
650 	}
651 
652 	/*
653 	 * Success.
654 	 */
655 	nlmsg_free(msg);
656 	return 1;
657 
658 nla_put_failure:
659 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
660 	    "%s: nl_put failed adding %s interface",
661 	    device, mondevice);
662 	nlmsg_free(msg);
663 	return PCAP_ERROR;
664 }
665 
666 static int
667 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
668     const char *device, const char *mondevice)
669 {
670 	int ifindex;
671 	struct nl_msg *msg;
672 	int err;
673 
674 	ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
675 	if (ifindex == -1)
676 		return PCAP_ERROR;
677 
678 	msg = nlmsg_alloc();
679 	if (!msg) {
680 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
681 		    "%s: failed to allocate netlink msg", device);
682 		return PCAP_ERROR;
683 	}
684 
685 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
686 		    0, NL80211_CMD_DEL_INTERFACE, 0);
687 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
688 
689 	err = nl_send_auto_complete(state->nl_sock, msg);
690 	if (err < 0) {
691 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
692 		    "%s: nl_send_auto_complete failed deleting %s interface: %s",
693 		    device, mondevice, get_nl_errmsg(-err));
694 		nlmsg_free(msg);
695 		return PCAP_ERROR;
696 	}
697 	err = nl_wait_for_ack(state->nl_sock);
698 	if (err < 0) {
699 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
700 		    "%s: nl_wait_for_ack failed adding %s interface: %s",
701 		    device, mondevice, get_nl_errmsg(-err));
702 		nlmsg_free(msg);
703 		return PCAP_ERROR;
704 	}
705 
706 	/*
707 	 * Success.
708 	 */
709 	nlmsg_free(msg);
710 	return 1;
711 
712 nla_put_failure:
713 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
714 	    "%s: nl_put failed deleting %s interface",
715 	    device, mondevice);
716 	nlmsg_free(msg);
717 	return PCAP_ERROR;
718 }
719 
720 static int
721 enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
722 {
723 	int ret;
724 	char phydev_path[PATH_MAX+1];
725 	struct nl80211_state nlstate;
726 	struct ifreq ifr;
727 	u_int n;
728 
729 	/*
730 	 * Is this a mac80211 device?
731 	 */
732 	ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
733 	if (ret < 0)
734 		return ret;	/* error */
735 	if (ret == 0)
736 		return 0;	/* no error, but not mac80211 device */
737 
738 	/*
739 	 * XXX - is this already a monN device?
740 	 * If so, we're done.
741 	 * Is that determined by old Wireless Extensions ioctls?
742 	 */
743 
744 	/*
745 	 * OK, it's apparently a mac80211 device.
746 	 * Try to find an unused monN device for it.
747 	 */
748 	ret = nl80211_init(handle, &nlstate, device);
749 	if (ret != 0)
750 		return ret;
751 	for (n = 0; n < UINT_MAX; n++) {
752 		/*
753 		 * Try mon{n}.
754 		 */
755 		char mondevice[3+10+1];	/* mon{UINT_MAX}\0 */
756 
757 		snprintf(mondevice, sizeof mondevice, "mon%u", n);
758 		ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
759 		if (ret == 1) {
760 			handle->md.mondevice = strdup(mondevice);
761 			goto added;
762 		}
763 		if (ret < 0) {
764 			/*
765 			 * Hard failure.  Just return ret; handle->errbuf
766 			 * has already been set.
767 			 */
768 			nl80211_cleanup(&nlstate);
769 			return ret;
770 		}
771 	}
772 
773 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
774 	    "%s: No free monN interfaces", device);
775 	nl80211_cleanup(&nlstate);
776 	return PCAP_ERROR;
777 
778 added:
779 
780 #if 0
781 	/*
782 	 * Sleep for .1 seconds.
783 	 */
784 	delay.tv_sec = 0;
785 	delay.tv_nsec = 500000000;
786 	nanosleep(&delay, NULL);
787 #endif
788 
789 	/*
790 	 * If we haven't already done so, arrange to have
791 	 * "pcap_close_all()" called when we exit.
792 	 */
793 	if (!pcap_do_addexit(handle)) {
794 		/*
795 		 * "atexit()" failed; don't put the interface
796 		 * in rfmon mode, just give up.
797 		 */
798 		return PCAP_ERROR_RFMON_NOTSUP;
799 	}
800 
801 	/*
802 	 * Now configure the monitor interface up.
803 	 */
804 	memset(&ifr, 0, sizeof(ifr));
805 	strncpy(ifr.ifr_name, handle->md.mondevice, sizeof(ifr.ifr_name));
806 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
807 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
808 		    "%s: Can't get flags for %s: %s", device,
809 		    handle->md.mondevice, strerror(errno));
810 		del_mon_if(handle, sock_fd, &nlstate, device,
811 		    handle->md.mondevice);
812 		nl80211_cleanup(&nlstate);
813 		return PCAP_ERROR;
814 	}
815 	ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
816 	if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
817 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
818 		    "%s: Can't set flags for %s: %s", device,
819 		    handle->md.mondevice, strerror(errno));
820 		del_mon_if(handle, sock_fd, &nlstate, device,
821 		    handle->md.mondevice);
822 		nl80211_cleanup(&nlstate);
823 		return PCAP_ERROR;
824 	}
825 
826 	/*
827 	 * Success.  Clean up the libnl state.
828 	 */
829 	nl80211_cleanup(&nlstate);
830 
831 	/*
832 	 * Note that we have to delete the monitor device when we close
833 	 * the handle.
834 	 */
835 	handle->md.must_do_on_close |= MUST_DELETE_MONIF;
836 
837 	/*
838 	 * Add this to the list of pcaps to close when we exit.
839 	 */
840 	pcap_add_to_pcaps_to_close(handle);
841 
842 	return 1;
843 }
844 #endif /* HAVE_LIBNL */
845 
846 static int
847 pcap_can_set_rfmon_linux(pcap_t *handle)
848 {
849 #ifdef HAVE_LIBNL
850 	char phydev_path[PATH_MAX+1];
851 	int ret;
852 #endif
853 #ifdef IW_MODE_MONITOR
854 	int sock_fd;
855 	struct iwreq ireq;
856 #endif
857 
858 	if (strcmp(handle->opt.source, "any") == 0) {
859 		/*
860 		 * Monitor mode makes no sense on the "any" device.
861 		 */
862 		return 0;
863 	}
864 
865 #ifdef HAVE_LIBNL
866 	/*
867 	 * Bleah.  There doesn't seem to be a way to ask a mac80211
868 	 * device, through libnl, whether it supports monitor mode;
869 	 * we'll just check whether the device appears to be a
870 	 * mac80211 device and, if so, assume the device supports
871 	 * monitor mode.
872 	 *
873 	 * wmaster devices don't appear to support the Wireless
874 	 * Extensions, but we can create a mon device for a
875 	 * wmaster device, so we don't bother checking whether
876 	 * a mac80211 device supports the Wireless Extensions.
877 	 */
878 	ret = get_mac80211_phydev(handle, handle->opt.source, phydev_path,
879 	    PATH_MAX);
880 	if (ret < 0)
881 		return ret;	/* error */
882 	if (ret == 1)
883 		return 1;	/* mac80211 device */
884 #endif
885 
886 #ifdef IW_MODE_MONITOR
887 	/*
888 	 * Bleah.  There doesn't appear to be an ioctl to use to ask
889 	 * whether a device supports monitor mode; we'll just do
890 	 * SIOCGIWMODE and, if it succeeds, assume the device supports
891 	 * monitor mode.
892 	 *
893 	 * Open a socket on which to attempt to get the mode.
894 	 * (We assume that if we have Wireless Extensions support
895 	 * we also have PF_PACKET support.)
896 	 */
897 	sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
898 	if (sock_fd == -1) {
899 		(void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
900 		    "socket: %s", pcap_strerror(errno));
901 		return PCAP_ERROR;
902 	}
903 
904 	/*
905 	 * Attempt to get the current mode.
906 	 */
907 	strncpy(ireq.ifr_ifrn.ifrn_name, handle->opt.source,
908 	    sizeof ireq.ifr_ifrn.ifrn_name);
909 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
910 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
911 		/*
912 		 * Well, we got the mode; assume we can set it.
913 		 */
914 		close(sock_fd);
915 		return 1;
916 	}
917 	if (errno == ENODEV) {
918 		/* The device doesn't even exist. */
919 		(void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
920 		    "SIOCGIWMODE failed: %s", pcap_strerror(errno));
921 		close(sock_fd);
922 		return PCAP_ERROR_NO_SUCH_DEVICE;
923 	}
924 	close(sock_fd);
925 #endif
926 	return 0;
927 }
928 
929 /*
930  * Grabs the number of dropped packets by the interface from /proc/net/dev.
931  *
932  * XXX - what about /sys/class/net/{interface name}/rx_*?  There are
933  * individual devices giving, in ASCII, various rx_ and tx_ statistics.
934  *
935  * Or can we get them in binary form from netlink?
936  */
937 static long int
938 linux_if_drops(const char * if_name)
939 {
940 	char buffer[512];
941 	char * bufptr;
942 	FILE * file;
943 	int field_to_convert = 3, if_name_sz = strlen(if_name);
944 	long int dropped_pkts = 0;
945 
946 	file = fopen("/proc/net/dev", "r");
947 	if (!file)
948 		return 0;
949 
950 	while (!dropped_pkts && fgets( buffer, sizeof(buffer), file ))
951 	{
952 		/* 	search for 'bytes' -- if its in there, then
953 			that means we need to grab the fourth field. otherwise
954 			grab the third field. */
955 		if (field_to_convert != 4 && strstr(buffer, "bytes"))
956 		{
957 			field_to_convert = 4;
958 			continue;
959 		}
960 
961 		/* find iface and make sure it actually matches -- space before the name and : after it */
962 		if ((bufptr = strstr(buffer, if_name)) &&
963 			(bufptr == buffer || *(bufptr-1) == ' ') &&
964 			*(bufptr + if_name_sz) == ':')
965 		{
966 			bufptr = bufptr + if_name_sz + 1;
967 
968 			/* grab the nth field from it */
969 			while( --field_to_convert && *bufptr != '\0')
970 			{
971 				while (*bufptr != '\0' && *(bufptr++) == ' ');
972 				while (*bufptr != '\0' && *(bufptr++) != ' ');
973 			}
974 
975 			/* get rid of any final spaces */
976 			while (*bufptr != '\0' && *bufptr == ' ') bufptr++;
977 
978 			if (*bufptr != '\0')
979 				dropped_pkts = strtol(bufptr, NULL, 10);
980 
981 			break;
982 		}
983 	}
984 
985 	fclose(file);
986 	return dropped_pkts;
987 }
988 
989 
990 /*
991  * With older kernels promiscuous mode is kind of interesting because we
992  * have to reset the interface before exiting. The problem can't really
993  * be solved without some daemon taking care of managing usage counts.
994  * If we put the interface into promiscuous mode, we set a flag indicating
995  * that we must take it out of that mode when the interface is closed,
996  * and, when closing the interface, if that flag is set we take it out
997  * of promiscuous mode.
998  *
999  * Even with newer kernels, we have the same issue with rfmon mode.
1000  */
1001 
1002 static void	pcap_cleanup_linux( pcap_t *handle )
1003 {
1004 	struct ifreq	ifr;
1005 #ifdef HAVE_LIBNL
1006 	struct nl80211_state nlstate;
1007 	int ret;
1008 #endif /* HAVE_LIBNL */
1009 #ifdef IW_MODE_MONITOR
1010 	int oldflags;
1011 	struct iwreq ireq;
1012 #endif /* IW_MODE_MONITOR */
1013 
1014 	if (handle->md.must_do_on_close != 0) {
1015 		/*
1016 		 * There's something we have to do when closing this
1017 		 * pcap_t.
1018 		 */
1019 		if (handle->md.must_do_on_close & MUST_CLEAR_PROMISC) {
1020 			/*
1021 			 * We put the interface into promiscuous mode;
1022 			 * take it out of promiscuous mode.
1023 			 *
1024 			 * XXX - if somebody else wants it in promiscuous
1025 			 * mode, this code cannot know that, so it'll take
1026 			 * it out of promiscuous mode.  That's not fixable
1027 			 * in 2.0[.x] kernels.
1028 			 */
1029 			memset(&ifr, 0, sizeof(ifr));
1030 			strncpy(ifr.ifr_name, handle->md.device,
1031 			    sizeof(ifr.ifr_name));
1032 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
1033 				fprintf(stderr,
1034 				    "Can't restore interface %s flags (SIOCGIFFLAGS failed: %s).\n"
1035 				    "Please adjust manually.\n"
1036 				    "Hint: This can't happen with Linux >= 2.2.0.\n",
1037 				    handle->md.device, strerror(errno));
1038 			} else {
1039 				if (ifr.ifr_flags & IFF_PROMISC) {
1040 					/*
1041 					 * Promiscuous mode is currently on;
1042 					 * turn it off.
1043 					 */
1044 					ifr.ifr_flags &= ~IFF_PROMISC;
1045 					if (ioctl(handle->fd, SIOCSIFFLAGS,
1046 					    &ifr) == -1) {
1047 						fprintf(stderr,
1048 						    "Can't restore interface %s flags (SIOCSIFFLAGS failed: %s).\n"
1049 						    "Please adjust manually.\n"
1050 						    "Hint: This can't happen with Linux >= 2.2.0.\n",
1051 						    handle->md.device,
1052 						    strerror(errno));
1053 					}
1054 				}
1055 			}
1056 		}
1057 
1058 #ifdef HAVE_LIBNL
1059 		if (handle->md.must_do_on_close & MUST_DELETE_MONIF) {
1060 			ret = nl80211_init(handle, &nlstate, handle->md.device);
1061 			if (ret >= 0) {
1062 				ret = del_mon_if(handle, handle->fd, &nlstate,
1063 				    handle->md.device, handle->md.mondevice);
1064 				nl80211_cleanup(&nlstate);
1065 			}
1066 			if (ret < 0) {
1067 				fprintf(stderr,
1068 				    "Can't delete monitor interface %s (%s).\n"
1069 				    "Please delete manually.\n",
1070 				    handle->md.mondevice, handle->errbuf);
1071 			}
1072 		}
1073 #endif /* HAVE_LIBNL */
1074 
1075 #ifdef IW_MODE_MONITOR
1076 		if (handle->md.must_do_on_close & MUST_CLEAR_RFMON) {
1077 			/*
1078 			 * We put the interface into rfmon mode;
1079 			 * take it out of rfmon mode.
1080 			 *
1081 			 * XXX - if somebody else wants it in rfmon
1082 			 * mode, this code cannot know that, so it'll take
1083 			 * it out of rfmon mode.
1084 			 */
1085 
1086 			/*
1087 			 * First, take the interface down if it's up;
1088 			 * otherwise, we might get EBUSY.
1089 			 * If we get errors, just drive on and print
1090 			 * a warning if we can't restore the mode.
1091 			 */
1092 			oldflags = 0;
1093 			memset(&ifr, 0, sizeof(ifr));
1094 			strncpy(ifr.ifr_name, handle->md.device,
1095 			    sizeof(ifr.ifr_name));
1096 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) != -1) {
1097 				if (ifr.ifr_flags & IFF_UP) {
1098 					oldflags = ifr.ifr_flags;
1099 					ifr.ifr_flags &= ~IFF_UP;
1100 					if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1)
1101 						oldflags = 0;	/* didn't set, don't restore */
1102 				}
1103 			}
1104 
1105 			/*
1106 			 * Now restore the mode.
1107 			 */
1108 			strncpy(ireq.ifr_ifrn.ifrn_name, handle->md.device,
1109 			    sizeof ireq.ifr_ifrn.ifrn_name);
1110 			ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1]
1111 			    = 0;
1112 			ireq.u.mode = handle->md.oldmode;
1113 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
1114 				/*
1115 				 * Scientist, you've failed.
1116 				 */
1117 				fprintf(stderr,
1118 				    "Can't restore interface %s wireless mode (SIOCSIWMODE failed: %s).\n"
1119 				    "Please adjust manually.\n",
1120 				    handle->md.device, strerror(errno));
1121 			}
1122 
1123 			/*
1124 			 * Now bring the interface back up if we brought
1125 			 * it down.
1126 			 */
1127 			if (oldflags != 0) {
1128 				ifr.ifr_flags = oldflags;
1129 				if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
1130 					fprintf(stderr,
1131 					    "Can't bring interface %s back up (SIOCSIFFLAGS failed: %s).\n"
1132 					    "Please adjust manually.\n",
1133 					    handle->md.device, strerror(errno));
1134 				}
1135 			}
1136 		}
1137 #endif /* IW_MODE_MONITOR */
1138 
1139 		/*
1140 		 * Take this pcap out of the list of pcaps for which we
1141 		 * have to take the interface out of some mode.
1142 		 */
1143 		pcap_remove_from_pcaps_to_close(handle);
1144 	}
1145 
1146 	if (handle->md.mondevice != NULL) {
1147 		free(handle->md.mondevice);
1148 		handle->md.mondevice = NULL;
1149 	}
1150 	if (handle->md.device != NULL) {
1151 		free(handle->md.device);
1152 		handle->md.device = NULL;
1153 	}
1154 	pcap_cleanup_live_common(handle);
1155 }
1156 
1157 /*
1158  *  Get a handle for a live capture from the given device. You can
1159  *  pass NULL as device to get all packages (without link level
1160  *  information of course). If you pass 1 as promisc the interface
1161  *  will be set to promiscous mode (XXX: I think this usage should
1162  *  be deprecated and functions be added to select that later allow
1163  *  modification of that values -- Torsten).
1164  */
1165 static int
1166 pcap_activate_linux(pcap_t *handle)
1167 {
1168 	const char	*device;
1169 	int		status = 0;
1170 
1171 	device = handle->opt.source;
1172 
1173 	handle->inject_op = pcap_inject_linux;
1174 	handle->setfilter_op = pcap_setfilter_linux;
1175 	handle->setdirection_op = pcap_setdirection_linux;
1176 	handle->set_datalink_op = NULL;	/* can't change data link type */
1177 	handle->getnonblock_op = pcap_getnonblock_fd;
1178 	handle->setnonblock_op = pcap_setnonblock_fd;
1179 	handle->cleanup_op = pcap_cleanup_linux;
1180 	handle->read_op = pcap_read_linux;
1181 	handle->stats_op = pcap_stats_linux;
1182 
1183 	/*
1184 	 * The "any" device is a special device which causes us not
1185 	 * to bind to a particular device and thus to look at all
1186 	 * devices.
1187 	 */
1188 	if (strcmp(device, "any") == 0) {
1189 		if (handle->opt.promisc) {
1190 			handle->opt.promisc = 0;
1191 			/* Just a warning. */
1192 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1193 			    "Promiscuous mode not supported on the \"any\" device");
1194 			status = PCAP_WARNING_PROMISC_NOTSUP;
1195 		}
1196 	}
1197 
1198 	handle->md.device	= strdup(device);
1199 	if (handle->md.device == NULL) {
1200 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
1201 			 pcap_strerror(errno) );
1202 		return PCAP_ERROR;
1203 	}
1204 
1205 	/*
1206 	 * If we're in promiscuous mode, then we probably want
1207 	 * to see when the interface drops packets too, so get an
1208 	 * initial count from /proc/net/dev
1209 	 */
1210 	if (handle->opt.promisc)
1211 		handle->md.proc_dropped = linux_if_drops(handle->md.device);
1212 
1213 	/*
1214 	 * Current Linux kernels use the protocol family PF_PACKET to
1215 	 * allow direct access to all packets on the network while
1216 	 * older kernels had a special socket type SOCK_PACKET to
1217 	 * implement this feature.
1218 	 * While this old implementation is kind of obsolete we need
1219 	 * to be compatible with older kernels for a while so we are
1220 	 * trying both methods with the newer method preferred.
1221 	 */
1222 	status = activate_new(handle);
1223 	if (status < 0) {
1224 		/*
1225 		 * Fatal error with the new way; just fail.
1226 		 * status has the error return; if it's PCAP_ERROR,
1227 		 * handle->errbuf has been set appropriately.
1228 		 */
1229 		goto fail;
1230 	}
1231 	if (status == 1) {
1232 		/*
1233 		 * Success.
1234 		 * Try to use memory-mapped access.
1235 		 */
1236 		switch (activate_mmap(handle, &status)) {
1237 
1238 		case 1:
1239 			/*
1240 			 * We succeeded.  status has been
1241 			 * set to the status to return,
1242 			 * which might be 0, or might be
1243 			 * a PCAP_WARNING_ value.
1244 			 */
1245 			return status;
1246 
1247 		case 0:
1248 			/*
1249 			 * Kernel doesn't support it - just continue
1250 			 * with non-memory-mapped access.
1251 			 */
1252 			break;
1253 
1254 		case -1:
1255 			/*
1256 			 * We failed to set up to use it, or the kernel
1257 			 * supports it, but we failed to enable it.
1258 			 * status has been set to the error status to
1259 			 * return and, if it's PCAP_ERROR, handle->errbuf
1260 			 * contains the error message.
1261 			 */
1262 			goto fail;
1263 		}
1264 	}
1265 	else if (status == 0) {
1266 		/* Non-fatal error; try old way */
1267 		if ((status = activate_old(handle)) != 1) {
1268 			/*
1269 			 * Both methods to open the packet socket failed.
1270 			 * Tidy up and report our failure (handle->errbuf
1271 			 * is expected to be set by the functions above).
1272 			 */
1273 			goto fail;
1274 		}
1275 	}
1276 
1277 	/*
1278 	 * We set up the socket, but not with memory-mapped access.
1279 	 */
1280 	status = 0;
1281 	if (handle->opt.buffer_size != 0) {
1282 		/*
1283 		 * Set the socket buffer size to the specified value.
1284 		 */
1285 		if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF,
1286 		    &handle->opt.buffer_size,
1287 		    sizeof(handle->opt.buffer_size)) == -1) {
1288 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1289 				 "SO_RCVBUF: %s", pcap_strerror(errno));
1290 			status = PCAP_ERROR;
1291 			goto fail;
1292 		}
1293 	}
1294 
1295 	/* Allocate the buffer */
1296 
1297 	handle->buffer	 = malloc(handle->bufsize + handle->offset);
1298 	if (!handle->buffer) {
1299 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1300 			 "malloc: %s", pcap_strerror(errno));
1301 		status = PCAP_ERROR;
1302 		goto fail;
1303 	}
1304 
1305 	/*
1306 	 * "handle->fd" is a socket, so "select()" and "poll()"
1307 	 * should work on it.
1308 	 */
1309 	handle->selectable_fd = handle->fd;
1310 
1311 	return status;
1312 
1313 fail:
1314 	pcap_cleanup_linux(handle);
1315 	return status;
1316 }
1317 
1318 /*
1319  *  Read at most max_packets from the capture stream and call the callback
1320  *  for each of them. Returns the number of packets handled or -1 if an
1321  *  error occured.
1322  */
1323 static int
1324 pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
1325 {
1326 	/*
1327 	 * Currently, on Linux only one packet is delivered per read,
1328 	 * so we don't loop.
1329 	 */
1330 	return pcap_read_packet(handle, callback, user);
1331 }
1332 
1333 /*
1334  *  Read a packet from the socket calling the handler provided by
1335  *  the user. Returns the number of packets received or -1 if an
1336  *  error occured.
1337  */
1338 static int
1339 pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
1340 {
1341 	u_char			*bp;
1342 	int			offset;
1343 #ifdef HAVE_PF_PACKET_SOCKETS
1344 	struct sockaddr_ll	from;
1345 	struct sll_header	*hdrp;
1346 #else
1347 	struct sockaddr		from;
1348 #endif
1349 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1350 	struct iovec		iov;
1351 	struct msghdr		msg;
1352 	struct cmsghdr		*cmsg;
1353 	union {
1354 		struct cmsghdr	cmsg;
1355 		char		buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
1356 	} cmsg_buf;
1357 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1358 	socklen_t		fromlen;
1359 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1360 	int			packet_len, caplen;
1361 	struct pcap_pkthdr	pcap_header;
1362 
1363 #ifdef HAVE_PF_PACKET_SOCKETS
1364 	/*
1365 	 * If this is a cooked device, leave extra room for a
1366 	 * fake packet header.
1367 	 */
1368 	if (handle->md.cooked)
1369 		offset = SLL_HDR_LEN;
1370 	else
1371 		offset = 0;
1372 #else
1373 	/*
1374 	 * This system doesn't have PF_PACKET sockets, so it doesn't
1375 	 * support cooked devices.
1376 	 */
1377 	offset = 0;
1378 #endif
1379 
1380 	/*
1381 	 * Receive a single packet from the kernel.
1382 	 * We ignore EINTR, as that might just be due to a signal
1383 	 * being delivered - if the signal should interrupt the
1384 	 * loop, the signal handler should call pcap_breakloop()
1385 	 * to set handle->break_loop (we ignore it on other
1386 	 * platforms as well).
1387 	 * We also ignore ENETDOWN, so that we can continue to
1388 	 * capture traffic if the interface goes down and comes
1389 	 * back up again; comments in the kernel indicate that
1390 	 * we'll just block waiting for packets if we try to
1391 	 * receive from a socket that delivered ENETDOWN, and,
1392 	 * if we're using a memory-mapped buffer, we won't even
1393 	 * get notified of "network down" events.
1394 	 */
1395 	bp = handle->buffer + handle->offset;
1396 
1397 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1398 	msg.msg_name		= &from;
1399 	msg.msg_namelen		= sizeof(from);
1400 	msg.msg_iov		= &iov;
1401 	msg.msg_iovlen		= 1;
1402 	msg.msg_control		= &cmsg_buf;
1403 	msg.msg_controllen	= sizeof(cmsg_buf);
1404 	msg.msg_flags		= 0;
1405 
1406 	iov.iov_len		= handle->bufsize - offset;
1407 	iov.iov_base		= bp + offset;
1408 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1409 
1410 	do {
1411 		/*
1412 		 * Has "pcap_breakloop()" been called?
1413 		 */
1414 		if (handle->break_loop) {
1415 			/*
1416 			 * Yes - clear the flag that indicates that it has,
1417 			 * and return PCAP_ERROR_BREAK as an indication that
1418 			 * we were told to break out of the loop.
1419 			 */
1420 			handle->break_loop = 0;
1421 			return PCAP_ERROR_BREAK;
1422 		}
1423 
1424 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1425 		packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC);
1426 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1427 		fromlen = sizeof(from);
1428 		packet_len = recvfrom(
1429 			handle->fd, bp + offset,
1430 			handle->bufsize - offset, MSG_TRUNC,
1431 			(struct sockaddr *) &from, &fromlen);
1432 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1433 	} while (packet_len == -1 && errno == EINTR);
1434 
1435 	/* Check if an error occured */
1436 
1437 	if (packet_len == -1) {
1438 		switch (errno) {
1439 
1440 		case EAGAIN:
1441 			return 0;	/* no packet there */
1442 
1443 		case ENETDOWN:
1444 			/*
1445 			 * The device on which we're capturing went away.
1446 			 *
1447 			 * XXX - we should really return
1448 			 * PCAP_ERROR_IFACE_NOT_UP, but pcap_dispatch()
1449 			 * etc. aren't defined to return that.
1450 			 */
1451 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1452 				"The interface went down");
1453 			return PCAP_ERROR;
1454 
1455 		default:
1456 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1457 				 "recvfrom: %s", pcap_strerror(errno));
1458 			return PCAP_ERROR;
1459 		}
1460 	}
1461 
1462 #ifdef HAVE_PF_PACKET_SOCKETS
1463 	if (!handle->md.sock_packet) {
1464 		/*
1465 		 * Unfortunately, there is a window between socket() and
1466 		 * bind() where the kernel may queue packets from any
1467 		 * interface.  If we're bound to a particular interface,
1468 		 * discard packets not from that interface.
1469 		 *
1470 		 * (If socket filters are supported, we could do the
1471 		 * same thing we do when changing the filter; however,
1472 		 * that won't handle packet sockets without socket
1473 		 * filter support, and it's a bit more complicated.
1474 		 * It would save some instructions per packet, however.)
1475 		 */
1476 		if (handle->md.ifindex != -1 &&
1477 		    from.sll_ifindex != handle->md.ifindex)
1478 			return 0;
1479 
1480 		/*
1481 		 * Do checks based on packet direction.
1482 		 * We can only do this if we're using PF_PACKET; the
1483 		 * address returned for SOCK_PACKET is a "sockaddr_pkt"
1484 		 * which lacks the relevant packet type information.
1485 		 */
1486 		if (from.sll_pkttype == PACKET_OUTGOING) {
1487 			/*
1488 			 * Outgoing packet.
1489 			 * If this is from the loopback device, reject it;
1490 			 * we'll see the packet as an incoming packet as well,
1491 			 * and we don't want to see it twice.
1492 			 */
1493 			if (from.sll_ifindex == handle->md.lo_ifindex)
1494 				return 0;
1495 
1496 			/*
1497 			 * If the user only wants incoming packets, reject it.
1498 			 */
1499 			if (handle->direction == PCAP_D_IN)
1500 				return 0;
1501 		} else {
1502 			/*
1503 			 * Incoming packet.
1504 			 * If the user only wants outgoing packets, reject it.
1505 			 */
1506 			if (handle->direction == PCAP_D_OUT)
1507 				return 0;
1508 		}
1509 	}
1510 #endif
1511 
1512 #ifdef HAVE_PF_PACKET_SOCKETS
1513 	/*
1514 	 * If this is a cooked device, fill in the fake packet header.
1515 	 */
1516 	if (handle->md.cooked) {
1517 		/*
1518 		 * Add the length of the fake header to the length
1519 		 * of packet data we read.
1520 		 */
1521 		packet_len += SLL_HDR_LEN;
1522 
1523 		hdrp = (struct sll_header *)bp;
1524 		hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype);
1525 		hdrp->sll_hatype = htons(from.sll_hatype);
1526 		hdrp->sll_halen = htons(from.sll_halen);
1527 		memcpy(hdrp->sll_addr, from.sll_addr,
1528 		    (from.sll_halen > SLL_ADDRLEN) ?
1529 		      SLL_ADDRLEN :
1530 		      from.sll_halen);
1531 		hdrp->sll_protocol = from.sll_protocol;
1532 	}
1533 
1534 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1535 	if (handle->md.vlan_offset != -1) {
1536 		for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
1537 			struct tpacket_auxdata *aux;
1538 			unsigned int len;
1539 			struct vlan_tag *tag;
1540 
1541 			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
1542 			    cmsg->cmsg_level != SOL_PACKET ||
1543 			    cmsg->cmsg_type != PACKET_AUXDATA)
1544 				continue;
1545 
1546 			aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
1547 #if defined(TP_STATUS_VLAN_VALID)
1548 			if ((aux->tp_vlan_tci == 0) && !(aux->tp_status & TP_STATUS_VLAN_VALID))
1549 #else
1550 			if (aux->tp_vlan_tci == 0) /* this is ambigious but without the
1551 						TP_STATUS_VLAN_VALID flag, there is
1552 						nothing that we can do */
1553 #endif
1554 				continue;
1555 
1556 			len = packet_len > iov.iov_len ? iov.iov_len : packet_len;
1557 			if (len < (unsigned int) handle->md.vlan_offset)
1558 				break;
1559 
1560 			bp -= VLAN_TAG_LEN;
1561 			memmove(bp, bp + VLAN_TAG_LEN, handle->md.vlan_offset);
1562 
1563 			tag = (struct vlan_tag *)(bp + handle->md.vlan_offset);
1564 			tag->vlan_tpid = htons(ETH_P_8021Q);
1565 			tag->vlan_tci = htons(aux->tp_vlan_tci);
1566 
1567 			packet_len += VLAN_TAG_LEN;
1568 		}
1569 	}
1570 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1571 #endif /* HAVE_PF_PACKET_SOCKETS */
1572 
1573 	/*
1574 	 * XXX: According to the kernel source we should get the real
1575 	 * packet len if calling recvfrom with MSG_TRUNC set. It does
1576 	 * not seem to work here :(, but it is supported by this code
1577 	 * anyway.
1578 	 * To be honest the code RELIES on that feature so this is really
1579 	 * broken with 2.2.x kernels.
1580 	 * I spend a day to figure out what's going on and I found out
1581 	 * that the following is happening:
1582 	 *
1583 	 * The packet comes from a random interface and the packet_rcv
1584 	 * hook is called with a clone of the packet. That code inserts
1585 	 * the packet into the receive queue of the packet socket.
1586 	 * If a filter is attached to that socket that filter is run
1587 	 * first - and there lies the problem. The default filter always
1588 	 * cuts the packet at the snaplen:
1589 	 *
1590 	 * # tcpdump -d
1591 	 * (000) ret      #68
1592 	 *
1593 	 * So the packet filter cuts down the packet. The recvfrom call
1594 	 * says "hey, it's only 68 bytes, it fits into the buffer" with
1595 	 * the result that we don't get the real packet length. This
1596 	 * is valid at least until kernel 2.2.17pre6.
1597 	 *
1598 	 * We currently handle this by making a copy of the filter
1599 	 * program, fixing all "ret" instructions with non-zero
1600 	 * operands to have an operand of 65535 so that the filter
1601 	 * doesn't truncate the packet, and supplying that modified
1602 	 * filter to the kernel.
1603 	 */
1604 
1605 	caplen = packet_len;
1606 	if (caplen > handle->snapshot)
1607 		caplen = handle->snapshot;
1608 
1609 	/* Run the packet filter if not using kernel filter */
1610 	if (!handle->md.use_bpf && handle->fcode.bf_insns) {
1611 		if (bpf_filter(handle->fcode.bf_insns, bp,
1612 		                packet_len, caplen) == 0)
1613 		{
1614 			/* rejected by filter */
1615 			return 0;
1616 		}
1617 	}
1618 
1619 	/* Fill in our own header data */
1620 
1621 	if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
1622 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1623 			 "SIOCGSTAMP: %s", pcap_strerror(errno));
1624 		return PCAP_ERROR;
1625 	}
1626 	pcap_header.caplen	= caplen;
1627 	pcap_header.len		= packet_len;
1628 
1629 	/*
1630 	 * Count the packet.
1631 	 *
1632 	 * Arguably, we should count them before we check the filter,
1633 	 * as on many other platforms "ps_recv" counts packets
1634 	 * handed to the filter rather than packets that passed
1635 	 * the filter, but if filtering is done in the kernel, we
1636 	 * can't get a count of packets that passed the filter,
1637 	 * and that would mean the meaning of "ps_recv" wouldn't
1638 	 * be the same on all Linux systems.
1639 	 *
1640 	 * XXX - it's not the same on all systems in any case;
1641 	 * ideally, we should have a "get the statistics" call
1642 	 * that supplies more counts and indicates which of them
1643 	 * it supplies, so that we supply a count of packets
1644 	 * handed to the filter only on platforms where that
1645 	 * information is available.
1646 	 *
1647 	 * We count them here even if we can get the packet count
1648 	 * from the kernel, as we can only determine at run time
1649 	 * whether we'll be able to get it from the kernel (if
1650 	 * HAVE_TPACKET_STATS isn't defined, we can't get it from
1651 	 * the kernel, but if it is defined, the library might
1652 	 * have been built with a 2.4 or later kernel, but we
1653 	 * might be running on a 2.2[.x] kernel without Alexey
1654 	 * Kuznetzov's turbopacket patches, and thus the kernel
1655 	 * might not be able to supply those statistics).  We
1656 	 * could, I guess, try, when opening the socket, to get
1657 	 * the statistics, and if we can not increment the count
1658 	 * here, but it's not clear that always incrementing
1659 	 * the count is more expensive than always testing a flag
1660 	 * in memory.
1661 	 *
1662 	 * We keep the count in "md.packets_read", and use that for
1663 	 * "ps_recv" if we can't get the statistics from the kernel.
1664 	 * We do that because, if we *can* get the statistics from
1665 	 * the kernel, we use "md.stat.ps_recv" and "md.stat.ps_drop"
1666 	 * as running counts, as reading the statistics from the
1667 	 * kernel resets the kernel statistics, and if we directly
1668 	 * increment "md.stat.ps_recv" here, that means it will
1669 	 * count packets *twice* on systems where we can get kernel
1670 	 * statistics - once here, and once in pcap_stats_linux().
1671 	 */
1672 	handle->md.packets_read++;
1673 
1674 	/* Call the user supplied callback function */
1675 	callback(userdata, &pcap_header, bp);
1676 
1677 	return 1;
1678 }
1679 
1680 static int
1681 pcap_inject_linux(pcap_t *handle, const void *buf, size_t size)
1682 {
1683 	int ret;
1684 
1685 #ifdef HAVE_PF_PACKET_SOCKETS
1686 	if (!handle->md.sock_packet) {
1687 		/* PF_PACKET socket */
1688 		if (handle->md.ifindex == -1) {
1689 			/*
1690 			 * We don't support sending on the "any" device.
1691 			 */
1692 			strlcpy(handle->errbuf,
1693 			    "Sending packets isn't supported on the \"any\" device",
1694 			    PCAP_ERRBUF_SIZE);
1695 			return (-1);
1696 		}
1697 
1698 		if (handle->md.cooked) {
1699 			/*
1700 			 * We don't support sending on the "any" device.
1701 			 *
1702 			 * XXX - how do you send on a bound cooked-mode
1703 			 * socket?
1704 			 * Is a "sendto()" required there?
1705 			 */
1706 			strlcpy(handle->errbuf,
1707 			    "Sending packets isn't supported in cooked mode",
1708 			    PCAP_ERRBUF_SIZE);
1709 			return (-1);
1710 		}
1711 	}
1712 #endif
1713 
1714 	ret = send(handle->fd, buf, size, 0);
1715 	if (ret == -1) {
1716 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
1717 		    pcap_strerror(errno));
1718 		return (-1);
1719 	}
1720 	return (ret);
1721 }
1722 
1723 /*
1724  *  Get the statistics for the given packet capture handle.
1725  *  Reports the number of dropped packets iff the kernel supports
1726  *  the PACKET_STATISTICS "getsockopt()" argument (2.4 and later
1727  *  kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket
1728  *  patches); otherwise, that information isn't available, and we lie
1729  *  and report 0 as the count of dropped packets.
1730  */
1731 static int
1732 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
1733 {
1734 #ifdef HAVE_TPACKET_STATS
1735 	struct tpacket_stats kstats;
1736 	socklen_t len = sizeof (struct tpacket_stats);
1737 #endif
1738 
1739 	long if_dropped = 0;
1740 
1741 	/*
1742 	 *	To fill in ps_ifdrop, we parse /proc/net/dev for the number
1743 	 */
1744 	if (handle->opt.promisc)
1745 	{
1746 		if_dropped = handle->md.proc_dropped;
1747 		handle->md.proc_dropped = linux_if_drops(handle->md.device);
1748 		handle->md.stat.ps_ifdrop += (handle->md.proc_dropped - if_dropped);
1749 	}
1750 
1751 #ifdef HAVE_TPACKET_STATS
1752 	/*
1753 	 * Try to get the packet counts from the kernel.
1754 	 */
1755 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
1756 			&kstats, &len) > -1) {
1757 		/*
1758 		 * On systems where the PACKET_STATISTICS "getsockopt()"
1759 		 * argument is supported on PF_PACKET sockets:
1760 		 *
1761 		 *	"ps_recv" counts only packets that *passed* the
1762 		 *	filter, not packets that didn't pass the filter.
1763 		 *	This includes packets later dropped because we
1764 		 *	ran out of buffer space.
1765 		 *
1766 		 *	"ps_drop" counts packets dropped because we ran
1767 		 *	out of buffer space.  It doesn't count packets
1768 		 *	dropped by the interface driver.  It counts only
1769 		 *	packets that passed the filter.
1770 		 *
1771 		 *	See above for ps_ifdrop.
1772 		 *
1773 		 *	Both statistics include packets not yet read from
1774 		 *	the kernel by libpcap, and thus not yet seen by
1775 		 *	the application.
1776 		 *
1777 		 * In "linux/net/packet/af_packet.c", at least in the
1778 		 * 2.4.9 kernel, "tp_packets" is incremented for every
1779 		 * packet that passes the packet filter *and* is
1780 		 * successfully queued on the socket; "tp_drops" is
1781 		 * incremented for every packet dropped because there's
1782 		 * not enough free space in the socket buffer.
1783 		 *
1784 		 * When the statistics are returned for a PACKET_STATISTICS
1785 		 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
1786 		 * so that "tp_packets" counts all packets handed to
1787 		 * the PF_PACKET socket, including packets dropped because
1788 		 * there wasn't room on the socket buffer - but not
1789 		 * including packets that didn't pass the filter.
1790 		 *
1791 		 * In the BSD BPF, the count of received packets is
1792 		 * incremented for every packet handed to BPF, regardless
1793 		 * of whether it passed the filter.
1794 		 *
1795 		 * We can't make "pcap_stats()" work the same on both
1796 		 * platforms, but the best approximation is to return
1797 		 * "tp_packets" as the count of packets and "tp_drops"
1798 		 * as the count of drops.
1799 		 *
1800 		 * Keep a running total because each call to
1801 		 *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
1802 		 * resets the counters to zero.
1803 		 */
1804 		handle->md.stat.ps_recv += kstats.tp_packets;
1805 		handle->md.stat.ps_drop += kstats.tp_drops;
1806 		*stats = handle->md.stat;
1807 		return 0;
1808 	}
1809 	else
1810 	{
1811 		/*
1812 		 * If the error was EOPNOTSUPP, fall through, so that
1813 		 * if you build the library on a system with
1814 		 * "struct tpacket_stats" and run it on a system
1815 		 * that doesn't, it works as it does if the library
1816 		 * is built on a system without "struct tpacket_stats".
1817 		 */
1818 		if (errno != EOPNOTSUPP) {
1819 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1820 			    "pcap_stats: %s", pcap_strerror(errno));
1821 			return -1;
1822 		}
1823 	}
1824 #endif
1825 	/*
1826 	 * On systems where the PACKET_STATISTICS "getsockopt()" argument
1827 	 * is not supported on PF_PACKET sockets:
1828 	 *
1829 	 *	"ps_recv" counts only packets that *passed* the filter,
1830 	 *	not packets that didn't pass the filter.  It does not
1831 	 *	count packets dropped because we ran out of buffer
1832 	 *	space.
1833 	 *
1834 	 *	"ps_drop" is not supported.
1835 	 *
1836 	 *	"ps_ifdrop" is supported. It will return the number
1837 	 *	of drops the interface reports in /proc/net/dev,
1838 	 *	if that is available.
1839 	 *
1840 	 *	"ps_recv" doesn't include packets not yet read from
1841 	 *	the kernel by libpcap.
1842 	 *
1843 	 * We maintain the count of packets processed by libpcap in
1844 	 * "md.packets_read", for reasons described in the comment
1845 	 * at the end of pcap_read_packet().  We have no idea how many
1846 	 * packets were dropped by the kernel buffers -- but we know
1847 	 * how many the interface dropped, so we can return that.
1848 	 */
1849 
1850 	stats->ps_recv = handle->md.packets_read;
1851 	stats->ps_drop = 0;
1852 	stats->ps_ifdrop = handle->md.stat.ps_ifdrop;
1853 	return 0;
1854 }
1855 
1856 /*
1857  * Get from "/sys/class/net" all interfaces listed there; if they're
1858  * already in the list of interfaces we have, that won't add another
1859  * instance, but if they're not, that'll add them.
1860  *
1861  * We don't bother getting any addresses for them; it appears you can't
1862  * use SIOCGIFADDR on Linux to get IPv6 addresses for interfaces, and,
1863  * although some other types of addresses can be fetched with SIOCGIFADDR,
1864  * we don't bother with them for now.
1865  *
1866  * We also don't fail if we couldn't open "/sys/class/net"; we just leave
1867  * the list of interfaces as is, and return 0, so that we can try
1868  * scanning /proc/net/dev.
1869  */
1870 static int
1871 scan_sys_class_net(pcap_if_t **devlistp, char *errbuf)
1872 {
1873 	DIR *sys_class_net_d;
1874 	int fd;
1875 	struct dirent *ent;
1876 	char subsystem_path[PATH_MAX+1];
1877 	struct stat statb;
1878 	char *p;
1879 	char name[512];	/* XXX - pick a size */
1880 	char *q, *saveq;
1881 	struct ifreq ifrflags;
1882 	int ret = 1;
1883 
1884 	sys_class_net_d = opendir("/sys/class/net");
1885 	if (sys_class_net_d == NULL) {
1886 		/*
1887 		 * Don't fail if it doesn't exist at all.
1888 		 */
1889 		if (errno == ENOENT)
1890 			return (0);
1891 
1892 		/*
1893 		 * Fail if we got some other error.
1894 		 */
1895 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
1896 		    "Can't open /sys/class/net: %s", pcap_strerror(errno));
1897 		return (-1);
1898 	}
1899 
1900 	/*
1901 	 * Create a socket from which to fetch interface information.
1902 	 */
1903 	fd = socket(AF_INET, SOCK_DGRAM, 0);
1904 	if (fd < 0) {
1905 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
1906 		    "socket: %s", pcap_strerror(errno));
1907 		(void)closedir(sys_class_net_d);
1908 		return (-1);
1909 	}
1910 
1911 	for (;;) {
1912 		errno = 0;
1913 		ent = readdir(sys_class_net_d);
1914 		if (ent == NULL) {
1915 			/*
1916 			 * Error or EOF; if errno != 0, it's an error.
1917 			 */
1918 			break;
1919 		}
1920 
1921 		/*
1922 		 * Ignore "." and "..".
1923 		 */
1924 		if (strcmp(ent->d_name, ".") == 0 ||
1925 		    strcmp(ent->d_name, "..") == 0)
1926 			continue;
1927 
1928 		/*
1929 		 * Ignore plain files; they do not have subdirectories
1930 		 * and thus have no attributes.
1931 		 */
1932 		if (ent->d_type == DT_REG)
1933 			continue;
1934 
1935 		/*
1936 		 * Is there an "ifindex" file under that name?
1937 		 * (We don't care whether it's a directory or
1938 		 * a symlink; older kernels have directories
1939 		 * for devices, newer kernels have symlinks to
1940 		 * directories.)
1941 		 */
1942 		snprintf(subsystem_path, sizeof subsystem_path,
1943 		    "/sys/class/net/%s/ifindex", ent->d_name);
1944 		if (lstat(subsystem_path, &statb) != 0) {
1945 			/*
1946 			 * Stat failed.  Either there was an error
1947 			 * other than ENOENT, and we don't know if
1948 			 * this is an interface, or it's ENOENT,
1949 			 * and either some part of "/sys/class/net/{if}"
1950 			 * disappeared, in which case it probably means
1951 			 * the interface disappeared, or there's no
1952 			 * "ifindex" file, which means it's not a
1953 			 * network interface.
1954 			 */
1955 			continue;
1956 		}
1957 
1958 		/*
1959 		 * Get the interface name.
1960 		 */
1961 		p = &ent->d_name[0];
1962 		q = &name[0];
1963 		while (*p != '\0' && isascii(*p) && !isspace(*p)) {
1964 			if (*p == ':') {
1965 				/*
1966 				 * This could be the separator between a
1967 				 * name and an alias number, or it could be
1968 				 * the separator between a name with no
1969 				 * alias number and the next field.
1970 				 *
1971 				 * If there's a colon after digits, it
1972 				 * separates the name and the alias number,
1973 				 * otherwise it separates the name and the
1974 				 * next field.
1975 				 */
1976 				saveq = q;
1977 				while (isascii(*p) && isdigit(*p))
1978 					*q++ = *p++;
1979 				if (*p != ':') {
1980 					/*
1981 					 * That was the next field,
1982 					 * not the alias number.
1983 					 */
1984 					q = saveq;
1985 				}
1986 				break;
1987 			} else
1988 				*q++ = *p++;
1989 		}
1990 		*q = '\0';
1991 
1992 		/*
1993 		 * Get the flags for this interface, and skip it if
1994 		 * it's not up.
1995 		 */
1996 		strncpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
1997 		if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
1998 			if (errno == ENXIO || errno == ENODEV)
1999 				continue;
2000 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2001 			    "SIOCGIFFLAGS: %.*s: %s",
2002 			    (int)sizeof(ifrflags.ifr_name),
2003 			    ifrflags.ifr_name,
2004 			    pcap_strerror(errno));
2005 			ret = -1;
2006 			break;
2007 		}
2008 		if (!(ifrflags.ifr_flags & IFF_UP))
2009 			continue;
2010 
2011 		/*
2012 		 * Add an entry for this interface, with no addresses.
2013 		 */
2014 		if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2015 		    errbuf) == -1) {
2016 			/*
2017 			 * Failure.
2018 			 */
2019 			ret = -1;
2020 			break;
2021 		}
2022 	}
2023 	if (ret != -1) {
2024 		/*
2025 		 * Well, we didn't fail for any other reason; did we
2026 		 * fail due to an error reading the directory?
2027 		 */
2028 		if (errno != 0) {
2029 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2030 			    "Error reading /sys/class/net: %s",
2031 			    pcap_strerror(errno));
2032 			ret = -1;
2033 		}
2034 	}
2035 
2036 	(void)close(fd);
2037 	(void)closedir(sys_class_net_d);
2038 	return (ret);
2039 }
2040 
2041 /*
2042  * Get from "/proc/net/dev" all interfaces listed there; if they're
2043  * already in the list of interfaces we have, that won't add another
2044  * instance, but if they're not, that'll add them.
2045  *
2046  * See comments from scan_sys_class_net().
2047  */
2048 static int
2049 scan_proc_net_dev(pcap_if_t **devlistp, char *errbuf)
2050 {
2051 	FILE *proc_net_f;
2052 	int fd;
2053 	char linebuf[512];
2054 	int linenum;
2055 	char *p;
2056 	char name[512];	/* XXX - pick a size */
2057 	char *q, *saveq;
2058 	struct ifreq ifrflags;
2059 	int ret = 0;
2060 
2061 	proc_net_f = fopen("/proc/net/dev", "r");
2062 	if (proc_net_f == NULL) {
2063 		/*
2064 		 * Don't fail if it doesn't exist at all.
2065 		 */
2066 		if (errno == ENOENT)
2067 			return (0);
2068 
2069 		/*
2070 		 * Fail if we got some other error.
2071 		 */
2072 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2073 		    "Can't open /proc/net/dev: %s", pcap_strerror(errno));
2074 		return (-1);
2075 	}
2076 
2077 	/*
2078 	 * Create a socket from which to fetch interface information.
2079 	 */
2080 	fd = socket(AF_INET, SOCK_DGRAM, 0);
2081 	if (fd < 0) {
2082 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2083 		    "socket: %s", pcap_strerror(errno));
2084 		(void)fclose(proc_net_f);
2085 		return (-1);
2086 	}
2087 
2088 	for (linenum = 1;
2089 	    fgets(linebuf, sizeof linebuf, proc_net_f) != NULL; linenum++) {
2090 		/*
2091 		 * Skip the first two lines - they're headers.
2092 		 */
2093 		if (linenum <= 2)
2094 			continue;
2095 
2096 		p = &linebuf[0];
2097 
2098 		/*
2099 		 * Skip leading white space.
2100 		 */
2101 		while (*p != '\0' && isascii(*p) && isspace(*p))
2102 			p++;
2103 		if (*p == '\0' || *p == '\n')
2104 			continue;	/* blank line */
2105 
2106 		/*
2107 		 * Get the interface name.
2108 		 */
2109 		q = &name[0];
2110 		while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2111 			if (*p == ':') {
2112 				/*
2113 				 * This could be the separator between a
2114 				 * name and an alias number, or it could be
2115 				 * the separator between a name with no
2116 				 * alias number and the next field.
2117 				 *
2118 				 * If there's a colon after digits, it
2119 				 * separates the name and the alias number,
2120 				 * otherwise it separates the name and the
2121 				 * next field.
2122 				 */
2123 				saveq = q;
2124 				while (isascii(*p) && isdigit(*p))
2125 					*q++ = *p++;
2126 				if (*p != ':') {
2127 					/*
2128 					 * That was the next field,
2129 					 * not the alias number.
2130 					 */
2131 					q = saveq;
2132 				}
2133 				break;
2134 			} else
2135 				*q++ = *p++;
2136 		}
2137 		*q = '\0';
2138 
2139 		/*
2140 		 * Get the flags for this interface, and skip it if
2141 		 * it's not up.
2142 		 */
2143 		strncpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2144 		if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2145 			if (errno == ENXIO)
2146 				continue;
2147 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2148 			    "SIOCGIFFLAGS: %.*s: %s",
2149 			    (int)sizeof(ifrflags.ifr_name),
2150 			    ifrflags.ifr_name,
2151 			    pcap_strerror(errno));
2152 			ret = -1;
2153 			break;
2154 		}
2155 		if (!(ifrflags.ifr_flags & IFF_UP))
2156 			continue;
2157 
2158 		/*
2159 		 * Add an entry for this interface, with no addresses.
2160 		 */
2161 		if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2162 		    errbuf) == -1) {
2163 			/*
2164 			 * Failure.
2165 			 */
2166 			ret = -1;
2167 			break;
2168 		}
2169 	}
2170 	if (ret != -1) {
2171 		/*
2172 		 * Well, we didn't fail for any other reason; did we
2173 		 * fail due to an error reading the file?
2174 		 */
2175 		if (ferror(proc_net_f)) {
2176 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2177 			    "Error reading /proc/net/dev: %s",
2178 			    pcap_strerror(errno));
2179 			ret = -1;
2180 		}
2181 	}
2182 
2183 	(void)close(fd);
2184 	(void)fclose(proc_net_f);
2185 	return (ret);
2186 }
2187 
2188 /*
2189  * Description string for the "any" device.
2190  */
2191 static const char any_descr[] = "Pseudo-device that captures on all interfaces";
2192 
2193 int
2194 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
2195 {
2196 	int ret;
2197 
2198 	/*
2199 	 * Read "/sys/class/net", and add to the list of interfaces all
2200 	 * interfaces listed there that we don't already have, because,
2201 	 * on Linux, SIOCGIFCONF reports only interfaces with IPv4 addresses,
2202 	 * and even getifaddrs() won't return information about
2203 	 * interfaces with no addresses, so you need to read "/sys/class/net"
2204 	 * to get the names of the rest of the interfaces.
2205 	 */
2206 	ret = scan_sys_class_net(alldevsp, errbuf);
2207 	if (ret == -1)
2208 		return (-1);	/* failed */
2209 	if (ret == 0) {
2210 		/*
2211 		 * No /sys/class/net; try reading /proc/net/dev instead.
2212 		 */
2213 		if (scan_proc_net_dev(alldevsp, errbuf) == -1)
2214 			return (-1);
2215 	}
2216 
2217 	/*
2218 	 * Add the "any" device.
2219 	 */
2220 	if (pcap_add_if(alldevsp, "any", 0, any_descr, errbuf) < 0)
2221 		return (-1);
2222 
2223 	return (0);
2224 }
2225 
2226 /*
2227  *  Attach the given BPF code to the packet capture device.
2228  */
2229 static int
2230 pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
2231     int is_mmapped)
2232 {
2233 #ifdef SO_ATTACH_FILTER
2234 	struct sock_fprog	fcode;
2235 	int			can_filter_in_kernel;
2236 	int			err = 0;
2237 #endif
2238 
2239 	if (!handle)
2240 		return -1;
2241 	if (!filter) {
2242 	        strncpy(handle->errbuf, "setfilter: No filter specified",
2243 			PCAP_ERRBUF_SIZE);
2244 		return -1;
2245 	}
2246 
2247 	/* Make our private copy of the filter */
2248 
2249 	if (install_bpf_program(handle, filter) < 0)
2250 		/* install_bpf_program() filled in errbuf */
2251 		return -1;
2252 
2253 	/*
2254 	 * Run user level packet filter by default. Will be overriden if
2255 	 * installing a kernel filter succeeds.
2256 	 */
2257 	handle->md.use_bpf = 0;
2258 
2259 	/* Install kernel level filter if possible */
2260 
2261 #ifdef SO_ATTACH_FILTER
2262 #ifdef USHRT_MAX
2263 	if (handle->fcode.bf_len > USHRT_MAX) {
2264 		/*
2265 		 * fcode.len is an unsigned short for current kernel.
2266 		 * I have yet to see BPF-Code with that much
2267 		 * instructions but still it is possible. So for the
2268 		 * sake of correctness I added this check.
2269 		 */
2270 		fprintf(stderr, "Warning: Filter too complex for kernel\n");
2271 		fcode.len = 0;
2272 		fcode.filter = NULL;
2273 		can_filter_in_kernel = 0;
2274 	} else
2275 #endif /* USHRT_MAX */
2276 	{
2277 		/*
2278 		 * Oh joy, the Linux kernel uses struct sock_fprog instead
2279 		 * of struct bpf_program and of course the length field is
2280 		 * of different size. Pointed out by Sebastian
2281 		 *
2282 		 * Oh, and we also need to fix it up so that all "ret"
2283 		 * instructions with non-zero operands have 65535 as the
2284 		 * operand if we're not capturing in memory-mapped modee,
2285 		 * and so that, if we're in cooked mode, all memory-reference
2286 		 * instructions use special magic offsets in references to
2287 		 * the link-layer header and assume that the link-layer
2288 		 * payload begins at 0; "fix_program()" will do that.
2289 		 */
2290 		switch (fix_program(handle, &fcode, is_mmapped)) {
2291 
2292 		case -1:
2293 		default:
2294 			/*
2295 			 * Fatal error; just quit.
2296 			 * (The "default" case shouldn't happen; we
2297 			 * return -1 for that reason.)
2298 			 */
2299 			return -1;
2300 
2301 		case 0:
2302 			/*
2303 			 * The program performed checks that we can't make
2304 			 * work in the kernel.
2305 			 */
2306 			can_filter_in_kernel = 0;
2307 			break;
2308 
2309 		case 1:
2310 			/*
2311 			 * We have a filter that'll work in the kernel.
2312 			 */
2313 			can_filter_in_kernel = 1;
2314 			break;
2315 		}
2316 	}
2317 
2318 	/*
2319 	 * NOTE: at this point, we've set both the "len" and "filter"
2320 	 * fields of "fcode".  As of the 2.6.32.4 kernel, at least,
2321 	 * those are the only members of the "sock_fprog" structure,
2322 	 * so we initialize every member of that structure.
2323 	 *
2324 	 * If there is anything in "fcode" that is not initialized,
2325 	 * it is either a field added in a later kernel, or it's
2326 	 * padding.
2327 	 *
2328 	 * If a new field is added, this code needs to be updated
2329 	 * to set it correctly.
2330 	 *
2331 	 * If there are no other fields, then:
2332 	 *
2333 	 *	if the Linux kernel looks at the padding, it's
2334 	 *	buggy;
2335 	 *
2336 	 *	if the Linux kernel doesn't look at the padding,
2337 	 *	then if some tool complains that we're passing
2338 	 *	uninitialized data to the kernel, then the tool
2339 	 *	is buggy and needs to understand that it's just
2340 	 *	padding.
2341 	 */
2342 	if (can_filter_in_kernel) {
2343 		if ((err = set_kernel_filter(handle, &fcode)) == 0)
2344 		{
2345 			/* Installation succeded - using kernel filter. */
2346 			handle->md.use_bpf = 1;
2347 		}
2348 		else if (err == -1)	/* Non-fatal error */
2349 		{
2350 			/*
2351 			 * Print a warning if we weren't able to install
2352 			 * the filter for a reason other than "this kernel
2353 			 * isn't configured to support socket filters.
2354 			 */
2355 			if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
2356 				fprintf(stderr,
2357 				    "Warning: Kernel filter failed: %s\n",
2358 					pcap_strerror(errno));
2359 			}
2360 		}
2361 	}
2362 
2363 	/*
2364 	 * If we're not using the kernel filter, get rid of any kernel
2365 	 * filter that might've been there before, e.g. because the
2366 	 * previous filter could work in the kernel, or because some other
2367 	 * code attached a filter to the socket by some means other than
2368 	 * calling "pcap_setfilter()".  Otherwise, the kernel filter may
2369 	 * filter out packets that would pass the new userland filter.
2370 	 */
2371 	if (!handle->md.use_bpf)
2372 		reset_kernel_filter(handle);
2373 
2374 	/*
2375 	 * Free up the copy of the filter that was made by "fix_program()".
2376 	 */
2377 	if (fcode.filter != NULL)
2378 		free(fcode.filter);
2379 
2380 	if (err == -2)
2381 		/* Fatal error */
2382 		return -1;
2383 #endif /* SO_ATTACH_FILTER */
2384 
2385 	return 0;
2386 }
2387 
2388 static int
2389 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
2390 {
2391 	return pcap_setfilter_linux_common(handle, filter, 0);
2392 }
2393 
2394 
2395 /*
2396  * Set direction flag: Which packets do we accept on a forwarding
2397  * single device? IN, OUT or both?
2398  */
2399 static int
2400 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
2401 {
2402 #ifdef HAVE_PF_PACKET_SOCKETS
2403 	if (!handle->md.sock_packet) {
2404 		handle->direction = d;
2405 		return 0;
2406 	}
2407 #endif
2408 	/*
2409 	 * We're not using PF_PACKET sockets, so we can't determine
2410 	 * the direction of the packet.
2411 	 */
2412 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2413 	    "Setting direction is not supported on SOCK_PACKET sockets");
2414 	return -1;
2415 }
2416 
2417 #ifdef HAVE_PF_PACKET_SOCKETS
2418 /*
2419  * Map the PACKET_ value to a LINUX_SLL_ value; we
2420  * want the same numerical value to be used in
2421  * the link-layer header even if the numerical values
2422  * for the PACKET_ #defines change, so that programs
2423  * that look at the packet type field will always be
2424  * able to handle DLT_LINUX_SLL captures.
2425  */
2426 static short int
2427 map_packet_type_to_sll_type(short int sll_pkttype)
2428 {
2429 	switch (sll_pkttype) {
2430 
2431 	case PACKET_HOST:
2432 		return htons(LINUX_SLL_HOST);
2433 
2434 	case PACKET_BROADCAST:
2435 		return htons(LINUX_SLL_BROADCAST);
2436 
2437 	case PACKET_MULTICAST:
2438 		return  htons(LINUX_SLL_MULTICAST);
2439 
2440 	case PACKET_OTHERHOST:
2441 		return htons(LINUX_SLL_OTHERHOST);
2442 
2443 	case PACKET_OUTGOING:
2444 		return htons(LINUX_SLL_OUTGOING);
2445 
2446 	default:
2447 		return -1;
2448 	}
2449 }
2450 #endif
2451 
2452 /*
2453  *  Linux uses the ARP hardware type to identify the type of an
2454  *  interface. pcap uses the DLT_xxx constants for this. This
2455  *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
2456  *  constant, as arguments, and sets "handle->linktype" to the
2457  *  appropriate DLT_XXX constant and sets "handle->offset" to
2458  *  the appropriate value (to make "handle->offset" plus link-layer
2459  *  header length be a multiple of 4, so that the link-layer payload
2460  *  will be aligned on a 4-byte boundary when capturing packets).
2461  *  (If the offset isn't set here, it'll be 0; add code as appropriate
2462  *  for cases where it shouldn't be 0.)
2463  *
2464  *  If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
2465  *  in cooked mode; otherwise, we can't use cooked mode, so we have
2466  *  to pick some type that works in raw mode, or fail.
2467  *
2468  *  Sets the link type to -1 if unable to map the type.
2469  */
2470 static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
2471 {
2472 	switch (arptype) {
2473 
2474 	case ARPHRD_ETHER:
2475 		/*
2476 		 * This is (presumably) a real Ethernet capture; give it a
2477 		 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
2478 		 * that an application can let you choose it, in case you're
2479 		 * capturing DOCSIS traffic that a Cisco Cable Modem
2480 		 * Termination System is putting out onto an Ethernet (it
2481 		 * doesn't put an Ethernet header onto the wire, it puts raw
2482 		 * DOCSIS frames out on the wire inside the low-level
2483 		 * Ethernet framing).
2484 		 *
2485 		 * XXX - are there any sorts of "fake Ethernet" that have
2486 		 * ARPHRD_ETHER but that *shouldn't offer DLT_DOCSIS as
2487 		 * a Cisco CMTS won't put traffic onto it or get traffic
2488 		 * bridged onto it?  ISDN is handled in "activate_new()",
2489 		 * as we fall back on cooked mode there; are there any
2490 		 * others?
2491 		 */
2492 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2493 		/*
2494 		 * If that fails, just leave the list empty.
2495 		 */
2496 		if (handle->dlt_list != NULL) {
2497 			handle->dlt_list[0] = DLT_EN10MB;
2498 			handle->dlt_list[1] = DLT_DOCSIS;
2499 			handle->dlt_count = 2;
2500 		}
2501 		/* FALLTHROUGH */
2502 
2503 	case ARPHRD_METRICOM:
2504 	case ARPHRD_LOOPBACK:
2505 		handle->linktype = DLT_EN10MB;
2506 		handle->offset = 2;
2507 		break;
2508 
2509 	case ARPHRD_EETHER:
2510 		handle->linktype = DLT_EN3MB;
2511 		break;
2512 
2513 	case ARPHRD_AX25:
2514 		handle->linktype = DLT_AX25_KISS;
2515 		break;
2516 
2517 	case ARPHRD_PRONET:
2518 		handle->linktype = DLT_PRONET;
2519 		break;
2520 
2521 	case ARPHRD_CHAOS:
2522 		handle->linktype = DLT_CHAOS;
2523 		break;
2524 #ifndef ARPHRD_CAN
2525 #define ARPHRD_CAN 280
2526 #endif
2527 	case ARPHRD_CAN:
2528 		handle->linktype = DLT_CAN_SOCKETCAN;
2529 		break;
2530 
2531 #ifndef ARPHRD_IEEE802_TR
2532 #define ARPHRD_IEEE802_TR 800	/* From Linux 2.4 */
2533 #endif
2534 	case ARPHRD_IEEE802_TR:
2535 	case ARPHRD_IEEE802:
2536 		handle->linktype = DLT_IEEE802;
2537 		handle->offset = 2;
2538 		break;
2539 
2540 	case ARPHRD_ARCNET:
2541 		handle->linktype = DLT_ARCNET_LINUX;
2542 		break;
2543 
2544 #ifndef ARPHRD_FDDI	/* From Linux 2.2.13 */
2545 #define ARPHRD_FDDI	774
2546 #endif
2547 	case ARPHRD_FDDI:
2548 		handle->linktype = DLT_FDDI;
2549 		handle->offset = 3;
2550 		break;
2551 
2552 #ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
2553 #define ARPHRD_ATM 19
2554 #endif
2555 	case ARPHRD_ATM:
2556 		/*
2557 		 * The Classical IP implementation in ATM for Linux
2558 		 * supports both what RFC 1483 calls "LLC Encapsulation",
2559 		 * in which each packet has an LLC header, possibly
2560 		 * with a SNAP header as well, prepended to it, and
2561 		 * what RFC 1483 calls "VC Based Multiplexing", in which
2562 		 * different virtual circuits carry different network
2563 		 * layer protocols, and no header is prepended to packets.
2564 		 *
2565 		 * They both have an ARPHRD_ type of ARPHRD_ATM, so
2566 		 * you can't use the ARPHRD_ type to find out whether
2567 		 * captured packets will have an LLC header, and,
2568 		 * while there's a socket ioctl to *set* the encapsulation
2569 		 * type, there's no ioctl to *get* the encapsulation type.
2570 		 *
2571 		 * This means that
2572 		 *
2573 		 *	programs that dissect Linux Classical IP frames
2574 		 *	would have to check for an LLC header and,
2575 		 *	depending on whether they see one or not, dissect
2576 		 *	the frame as LLC-encapsulated or as raw IP (I
2577 		 *	don't know whether there's any traffic other than
2578 		 *	IP that would show up on the socket, or whether
2579 		 *	there's any support for IPv6 in the Linux
2580 		 *	Classical IP code);
2581 		 *
2582 		 *	filter expressions would have to compile into
2583 		 *	code that checks for an LLC header and does
2584 		 *	the right thing.
2585 		 *
2586 		 * Both of those are a nuisance - and, at least on systems
2587 		 * that support PF_PACKET sockets, we don't have to put
2588 		 * up with those nuisances; instead, we can just capture
2589 		 * in cooked mode.  That's what we'll do, if we can.
2590 		 * Otherwise, we'll just fail.
2591 		 */
2592 		if (cooked_ok)
2593 			handle->linktype = DLT_LINUX_SLL;
2594 		else
2595 			handle->linktype = -1;
2596 		break;
2597 
2598 #ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
2599 #define ARPHRD_IEEE80211 801
2600 #endif
2601 	case ARPHRD_IEEE80211:
2602 		handle->linktype = DLT_IEEE802_11;
2603 		break;
2604 
2605 #ifndef ARPHRD_IEEE80211_PRISM  /* From Linux 2.4.18 */
2606 #define ARPHRD_IEEE80211_PRISM 802
2607 #endif
2608 	case ARPHRD_IEEE80211_PRISM:
2609 		handle->linktype = DLT_PRISM_HEADER;
2610 		break;
2611 
2612 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
2613 #define ARPHRD_IEEE80211_RADIOTAP 803
2614 #endif
2615 	case ARPHRD_IEEE80211_RADIOTAP:
2616 		handle->linktype = DLT_IEEE802_11_RADIO;
2617 		break;
2618 
2619 	case ARPHRD_PPP:
2620 		/*
2621 		 * Some PPP code in the kernel supplies no link-layer
2622 		 * header whatsoever to PF_PACKET sockets; other PPP
2623 		 * code supplies PPP link-layer headers ("syncppp.c");
2624 		 * some PPP code might supply random link-layer
2625 		 * headers (PPP over ISDN - there's code in Ethereal,
2626 		 * for example, to cope with PPP-over-ISDN captures
2627 		 * with which the Ethereal developers have had to cope,
2628 		 * heuristically trying to determine which of the
2629 		 * oddball link-layer headers particular packets have).
2630 		 *
2631 		 * As such, we just punt, and run all PPP interfaces
2632 		 * in cooked mode, if we can; otherwise, we just treat
2633 		 * it as DLT_RAW, for now - if somebody needs to capture,
2634 		 * on a 2.0[.x] kernel, on PPP devices that supply a
2635 		 * link-layer header, they'll have to add code here to
2636 		 * map to the appropriate DLT_ type (possibly adding a
2637 		 * new DLT_ type, if necessary).
2638 		 */
2639 		if (cooked_ok)
2640 			handle->linktype = DLT_LINUX_SLL;
2641 		else {
2642 			/*
2643 			 * XXX - handle ISDN types here?  We can't fall
2644 			 * back on cooked sockets, so we'd have to
2645 			 * figure out from the device name what type of
2646 			 * link-layer encapsulation it's using, and map
2647 			 * that to an appropriate DLT_ value, meaning
2648 			 * we'd map "isdnN" devices to DLT_RAW (they
2649 			 * supply raw IP packets with no link-layer
2650 			 * header) and "isdY" devices to a new DLT_I4L_IP
2651 			 * type that has only an Ethernet packet type as
2652 			 * a link-layer header.
2653 			 *
2654 			 * But sometimes we seem to get random crap
2655 			 * in the link-layer header when capturing on
2656 			 * ISDN devices....
2657 			 */
2658 			handle->linktype = DLT_RAW;
2659 		}
2660 		break;
2661 
2662 #ifndef ARPHRD_CISCO
2663 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
2664 #endif
2665 	case ARPHRD_CISCO:
2666 		handle->linktype = DLT_C_HDLC;
2667 		break;
2668 
2669 	/* Not sure if this is correct for all tunnels, but it
2670 	 * works for CIPE */
2671 	case ARPHRD_TUNNEL:
2672 #ifndef ARPHRD_SIT
2673 #define ARPHRD_SIT 776	/* From Linux 2.2.13 */
2674 #endif
2675 	case ARPHRD_SIT:
2676 	case ARPHRD_CSLIP:
2677 	case ARPHRD_SLIP6:
2678 	case ARPHRD_CSLIP6:
2679 	case ARPHRD_ADAPT:
2680 	case ARPHRD_SLIP:
2681 #ifndef ARPHRD_RAWHDLC
2682 #define ARPHRD_RAWHDLC 518
2683 #endif
2684 	case ARPHRD_RAWHDLC:
2685 #ifndef ARPHRD_DLCI
2686 #define ARPHRD_DLCI 15
2687 #endif
2688 	case ARPHRD_DLCI:
2689 		/*
2690 		 * XXX - should some of those be mapped to DLT_LINUX_SLL
2691 		 * instead?  Should we just map all of them to DLT_LINUX_SLL?
2692 		 */
2693 		handle->linktype = DLT_RAW;
2694 		break;
2695 
2696 #ifndef ARPHRD_FRAD
2697 #define ARPHRD_FRAD 770
2698 #endif
2699 	case ARPHRD_FRAD:
2700 		handle->linktype = DLT_FRELAY;
2701 		break;
2702 
2703 	case ARPHRD_LOCALTLK:
2704 		handle->linktype = DLT_LTALK;
2705 		break;
2706 
2707 #ifndef ARPHRD_FCPP
2708 #define ARPHRD_FCPP	784
2709 #endif
2710 	case ARPHRD_FCPP:
2711 #ifndef ARPHRD_FCAL
2712 #define ARPHRD_FCAL	785
2713 #endif
2714 	case ARPHRD_FCAL:
2715 #ifndef ARPHRD_FCPL
2716 #define ARPHRD_FCPL	786
2717 #endif
2718 	case ARPHRD_FCPL:
2719 #ifndef ARPHRD_FCFABRIC
2720 #define ARPHRD_FCFABRIC	787
2721 #endif
2722 	case ARPHRD_FCFABRIC:
2723 		/*
2724 		 * We assume that those all mean RFC 2625 IP-over-
2725 		 * Fibre Channel, with the RFC 2625 header at
2726 		 * the beginning of the packet.
2727 		 */
2728 		handle->linktype = DLT_IP_OVER_FC;
2729 		break;
2730 
2731 #ifndef ARPHRD_IRDA
2732 #define ARPHRD_IRDA	783
2733 #endif
2734 	case ARPHRD_IRDA:
2735 		/* Don't expect IP packet out of this interfaces... */
2736 		handle->linktype = DLT_LINUX_IRDA;
2737 		/* We need to save packet direction for IrDA decoding,
2738 		 * so let's use "Linux-cooked" mode. Jean II */
2739 		//handle->md.cooked = 1;
2740 		break;
2741 
2742 	/* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
2743 	 * is needed, please report it to <daniele@orlandi.com> */
2744 #ifndef ARPHRD_LAPD
2745 #define ARPHRD_LAPD	8445
2746 #endif
2747 	case ARPHRD_LAPD:
2748 		/* Don't expect IP packet out of this interfaces... */
2749 		handle->linktype = DLT_LINUX_LAPD;
2750 		break;
2751 
2752 #ifndef ARPHRD_NONE
2753 #define ARPHRD_NONE	0xFFFE
2754 #endif
2755 	case ARPHRD_NONE:
2756 		/*
2757 		 * No link-layer header; packets are just IP
2758 		 * packets, so use DLT_RAW.
2759 		 */
2760 		handle->linktype = DLT_RAW;
2761 		break;
2762 
2763 #ifndef ARPHRD_IEEE802154
2764 #define ARPHRD_IEEE802154      804
2765 #endif
2766        case ARPHRD_IEEE802154:
2767                handle->linktype =  DLT_IEEE802_15_4_NOFCS;
2768                break;
2769 
2770 	default:
2771 		handle->linktype = -1;
2772 		break;
2773 	}
2774 }
2775 
2776 /* ===== Functions to interface to the newer kernels ================== */
2777 
2778 /*
2779  * Try to open a packet socket using the new kernel PF_PACKET interface.
2780  * Returns 1 on success, 0 on an error that means the new interface isn't
2781  * present (so the old SOCK_PACKET interface should be tried), and a
2782  * PCAP_ERROR_ value on an error that means that the old mechanism won't
2783  * work either (so it shouldn't be tried).
2784  */
2785 static int
2786 activate_new(pcap_t *handle)
2787 {
2788 #ifdef HAVE_PF_PACKET_SOCKETS
2789 	const char		*device = handle->opt.source;
2790 	int			is_any_device = (strcmp(device, "any") == 0);
2791 	int			sock_fd = -1, arptype;
2792 #ifdef HAVE_PACKET_AUXDATA
2793 	int			val;
2794 #endif
2795 	int			err = 0;
2796 	struct packet_mreq	mr;
2797 
2798 	/*
2799 	 * Open a socket with protocol family packet. If the
2800 	 * "any" device was specified, we open a SOCK_DGRAM
2801 	 * socket for the cooked interface, otherwise we first
2802 	 * try a SOCK_RAW socket for the raw interface.
2803 	 */
2804 	sock_fd = is_any_device ?
2805 		socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) :
2806 		socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
2807 
2808 	if (sock_fd == -1) {
2809 		if (errno == EINVAL || errno == EAFNOSUPPORT) {
2810 			/*
2811 			 * We don't support PF_PACKET/SOCK_whatever
2812 			 * sockets; try the old mechanism.
2813 			 */
2814 			return 0;
2815 		}
2816 
2817 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
2818 			 pcap_strerror(errno) );
2819 		if (errno == EPERM || errno == EACCES) {
2820 			/*
2821 			 * You don't have permission to open the
2822 			 * socket.
2823 			 */
2824 			return PCAP_ERROR_PERM_DENIED;
2825 		} else {
2826 			/*
2827 			 * Other error.
2828 			 */
2829 			return PCAP_ERROR;
2830 		}
2831 	}
2832 
2833 	/* It seems the kernel supports the new interface. */
2834 	handle->md.sock_packet = 0;
2835 
2836 	/*
2837 	 * Get the interface index of the loopback device.
2838 	 * If the attempt fails, don't fail, just set the
2839 	 * "md.lo_ifindex" to -1.
2840 	 *
2841 	 * XXX - can there be more than one device that loops
2842 	 * packets back, i.e. devices other than "lo"?  If so,
2843 	 * we'd need to find them all, and have an array of
2844 	 * indices for them, and check all of them in
2845 	 * "pcap_read_packet()".
2846 	 */
2847 	handle->md.lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
2848 
2849 	/*
2850 	 * Default value for offset to align link-layer payload
2851 	 * on a 4-byte boundary.
2852 	 */
2853 	handle->offset	 = 0;
2854 
2855 	/*
2856 	 * What kind of frames do we have to deal with? Fall back
2857 	 * to cooked mode if we have an unknown interface type
2858 	 * or a type we know doesn't work well in raw mode.
2859 	 */
2860 	if (!is_any_device) {
2861 		/* Assume for now we don't need cooked mode. */
2862 		handle->md.cooked = 0;
2863 
2864 		if (handle->opt.rfmon) {
2865 			/*
2866 			 * We were asked to turn on monitor mode.
2867 			 * Do so before we get the link-layer type,
2868 			 * because entering monitor mode could change
2869 			 * the link-layer type.
2870 			 */
2871 			err = enter_rfmon_mode(handle, sock_fd, device);
2872 			if (err < 0) {
2873 				/* Hard failure */
2874 				close(sock_fd);
2875 				return err;
2876 			}
2877 			if (err == 0) {
2878 				/*
2879 				 * Nothing worked for turning monitor mode
2880 				 * on.
2881 				 */
2882 				close(sock_fd);
2883 				return PCAP_ERROR_RFMON_NOTSUP;
2884 			}
2885 
2886 			/*
2887 			 * Either monitor mode has been turned on for
2888 			 * the device, or we've been given a different
2889 			 * device to open for monitor mode.  If we've
2890 			 * been given a different device, use it.
2891 			 */
2892 			if (handle->md.mondevice != NULL)
2893 				device = handle->md.mondevice;
2894 		}
2895 		arptype	= iface_get_arptype(sock_fd, device, handle->errbuf);
2896 		if (arptype < 0) {
2897 			close(sock_fd);
2898 			return arptype;
2899 		}
2900 		map_arphrd_to_dlt(handle, arptype, 1);
2901 		if (handle->linktype == -1 ||
2902 		    handle->linktype == DLT_LINUX_SLL ||
2903 		    handle->linktype == DLT_LINUX_IRDA ||
2904 		    handle->linktype == DLT_LINUX_LAPD ||
2905 		    (handle->linktype == DLT_EN10MB &&
2906 		     (strncmp("isdn", device, 4) == 0 ||
2907 		      strncmp("isdY", device, 4) == 0))) {
2908 			/*
2909 			 * Unknown interface type (-1), or a
2910 			 * device we explicitly chose to run
2911 			 * in cooked mode (e.g., PPP devices),
2912 			 * or an ISDN device (whose link-layer
2913 			 * type we can only determine by using
2914 			 * APIs that may be different on different
2915 			 * kernels) - reopen in cooked mode.
2916 			 */
2917 			if (close(sock_fd) == -1) {
2918 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2919 					 "close: %s", pcap_strerror(errno));
2920 				return PCAP_ERROR;
2921 			}
2922 			sock_fd = socket(PF_PACKET, SOCK_DGRAM,
2923 			    htons(ETH_P_ALL));
2924 			if (sock_fd == -1) {
2925 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2926 				    "socket: %s", pcap_strerror(errno));
2927 				if (errno == EPERM || errno == EACCES) {
2928 					/*
2929 					 * You don't have permission to
2930 					 * open the socket.
2931 					 */
2932 					return PCAP_ERROR_PERM_DENIED;
2933 				} else {
2934 					/*
2935 					 * Other error.
2936 					 */
2937 					return PCAP_ERROR;
2938 				}
2939 			}
2940 			handle->md.cooked = 1;
2941 
2942 			/*
2943 			 * Get rid of any link-layer type list
2944 			 * we allocated - this only supports cooked
2945 			 * capture.
2946 			 */
2947 			if (handle->dlt_list != NULL) {
2948 				free(handle->dlt_list);
2949 				handle->dlt_list = NULL;
2950 				handle->dlt_count = 0;
2951 			}
2952 
2953 			if (handle->linktype == -1) {
2954 				/*
2955 				 * Warn that we're falling back on
2956 				 * cooked mode; we may want to
2957 				 * update "map_arphrd_to_dlt()"
2958 				 * to handle the new type.
2959 				 */
2960 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2961 					"arptype %d not "
2962 					"supported by libpcap - "
2963 					"falling back to cooked "
2964 					"socket",
2965 					arptype);
2966 			}
2967 
2968 			/*
2969 			 * IrDA capture is not a real "cooked" capture,
2970 			 * it's IrLAP frames, not IP packets.  The
2971 			 * same applies to LAPD capture.
2972 			 */
2973 			if (handle->linktype != DLT_LINUX_IRDA &&
2974 			    handle->linktype != DLT_LINUX_LAPD)
2975 				handle->linktype = DLT_LINUX_SLL;
2976 		}
2977 
2978 		handle->md.ifindex = iface_get_id(sock_fd, device,
2979 		    handle->errbuf);
2980 		if (handle->md.ifindex == -1) {
2981 			close(sock_fd);
2982 			return PCAP_ERROR;
2983 		}
2984 
2985 		if ((err = iface_bind(sock_fd, handle->md.ifindex,
2986 		    handle->errbuf)) != 1) {
2987 		    	close(sock_fd);
2988 			if (err < 0)
2989 				return err;
2990 			else
2991 				return 0;	/* try old mechanism */
2992 		}
2993 	} else {
2994 		/*
2995 		 * The "any" device.
2996 		 */
2997 		if (handle->opt.rfmon) {
2998 			/*
2999 			 * It doesn't support monitor mode.
3000 			 */
3001 			return PCAP_ERROR_RFMON_NOTSUP;
3002 		}
3003 
3004 		/*
3005 		 * It uses cooked mode.
3006 		 */
3007 		handle->md.cooked = 1;
3008 		handle->linktype = DLT_LINUX_SLL;
3009 
3010 		/*
3011 		 * We're not bound to a device.
3012 		 * For now, we're using this as an indication
3013 		 * that we can't transmit; stop doing that only
3014 		 * if we figure out how to transmit in cooked
3015 		 * mode.
3016 		 */
3017 		handle->md.ifindex = -1;
3018 	}
3019 
3020 	/*
3021 	 * Select promiscuous mode on if "promisc" is set.
3022 	 *
3023 	 * Do not turn allmulti mode on if we don't select
3024 	 * promiscuous mode - on some devices (e.g., Orinoco
3025 	 * wireless interfaces), allmulti mode isn't supported
3026 	 * and the driver implements it by turning promiscuous
3027 	 * mode on, and that screws up the operation of the
3028 	 * card as a normal networking interface, and on no
3029 	 * other platform I know of does starting a non-
3030 	 * promiscuous capture affect which multicast packets
3031 	 * are received by the interface.
3032 	 */
3033 
3034 	/*
3035 	 * Hmm, how can we set promiscuous mode on all interfaces?
3036 	 * I am not sure if that is possible at all.  For now, we
3037 	 * silently ignore attempts to turn promiscuous mode on
3038 	 * for the "any" device (so you don't have to explicitly
3039 	 * disable it in programs such as tcpdump).
3040 	 */
3041 
3042 	if (!is_any_device && handle->opt.promisc) {
3043 		memset(&mr, 0, sizeof(mr));
3044 		mr.mr_ifindex = handle->md.ifindex;
3045 		mr.mr_type    = PACKET_MR_PROMISC;
3046 		if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
3047 		    &mr, sizeof(mr)) == -1) {
3048 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3049 				"setsockopt: %s", pcap_strerror(errno));
3050 			close(sock_fd);
3051 			return PCAP_ERROR;
3052 		}
3053 	}
3054 
3055 	/* Enable auxillary data if supported and reserve room for
3056 	 * reconstructing VLAN headers. */
3057 #ifdef HAVE_PACKET_AUXDATA
3058 	val = 1;
3059 	if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
3060 		       sizeof(val)) == -1 && errno != ENOPROTOOPT) {
3061 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3062 			 "setsockopt: %s", pcap_strerror(errno));
3063 		close(sock_fd);
3064 		return PCAP_ERROR;
3065 	}
3066 	handle->offset += VLAN_TAG_LEN;
3067 #endif /* HAVE_PACKET_AUXDATA */
3068 
3069 	/*
3070 	 * This is a 2.2[.x] or later kernel (we know that
3071 	 * because we're not using a SOCK_PACKET socket -
3072 	 * PF_PACKET is supported only in 2.2 and later
3073 	 * kernels).
3074 	 *
3075 	 * We can safely pass "recvfrom()" a byte count
3076 	 * based on the snapshot length.
3077 	 *
3078 	 * If we're in cooked mode, make the snapshot length
3079 	 * large enough to hold a "cooked mode" header plus
3080 	 * 1 byte of packet data (so we don't pass a byte
3081 	 * count of 0 to "recvfrom()").
3082 	 */
3083 	if (handle->md.cooked) {
3084 		if (handle->snapshot < SLL_HDR_LEN + 1)
3085 			handle->snapshot = SLL_HDR_LEN + 1;
3086 	}
3087 	handle->bufsize = handle->snapshot;
3088 
3089 	/*
3090 	 * Set the offset at which to insert VLAN tags.
3091 	 */
3092 	switch (handle->linktype) {
3093 
3094 	case DLT_EN10MB:
3095 		handle->md.vlan_offset = 2 * ETH_ALEN;
3096 		break;
3097 
3098 	case DLT_LINUX_SLL:
3099 		handle->md.vlan_offset = 14;
3100 		break;
3101 
3102 	default:
3103 		handle->md.vlan_offset = -1; /* unknown */
3104 		break;
3105 	}
3106 
3107 	/* Save the socket FD in the pcap structure */
3108 	handle->fd = sock_fd;
3109 
3110 	return 1;
3111 #else
3112 	strncpy(ebuf,
3113 		"New packet capturing interface not supported by build "
3114 		"environment", PCAP_ERRBUF_SIZE);
3115 	return 0;
3116 #endif
3117 }
3118 
3119 #ifdef HAVE_PACKET_RING
3120 /*
3121  * Attempt to activate with memory-mapped access.
3122  *
3123  * On success, returns 1, and sets *status to 0 if there are no warnings
3124  * or to a PCAP_WARNING_ code if there is a warning.
3125  *
3126  * On failure due to lack of support for memory-mapped capture, returns
3127  * 0.
3128  *
3129  * On error, returns -1, and sets *status to the appropriate error code;
3130  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3131  */
3132 static int
3133 activate_mmap(pcap_t *handle, int *status)
3134 {
3135 	int ret;
3136 
3137 	/*
3138 	 * Attempt to allocate a buffer to hold the contents of one
3139 	 * packet, for use by the oneshot callback.
3140 	 */
3141 	handle->md.oneshot_buffer = malloc(handle->snapshot);
3142 	if (handle->md.oneshot_buffer == NULL) {
3143 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3144 			 "can't allocate oneshot buffer: %s",
3145 			 pcap_strerror(errno));
3146 		*status = PCAP_ERROR;
3147 		return -1;
3148 	}
3149 
3150 	if (handle->opt.buffer_size == 0) {
3151 		/* by default request 2M for the ring buffer */
3152 		handle->opt.buffer_size = 2*1024*1024;
3153 	}
3154 	ret = prepare_tpacket_socket(handle);
3155 	if (ret == -1) {
3156 		free(handle->md.oneshot_buffer);
3157 		*status = PCAP_ERROR;
3158 		return ret;
3159 	}
3160 	ret = create_ring(handle, status);
3161 	if (ret == 0) {
3162 		/*
3163 		 * We don't support memory-mapped capture; our caller
3164 		 * will fall back on reading from the socket.
3165 		 */
3166 		free(handle->md.oneshot_buffer);
3167 		return 0;
3168 	}
3169 	if (ret == -1) {
3170 		/*
3171 		 * Error attempting to enable memory-mapped capture;
3172 		 * fail.  create_ring() has set *status.
3173 		 */
3174 		free(handle->md.oneshot_buffer);
3175 		return -1;
3176 	}
3177 
3178 	/*
3179 	 * Success.  *status has been set either to 0 if there are no
3180 	 * warnings or to a PCAP_WARNING_ value if there is a warning.
3181 	 *
3182 	 * Override some defaults and inherit the other fields from
3183 	 * activate_new.
3184 	 * handle->offset is used to get the current position into the rx ring.
3185 	 * handle->cc is used to store the ring size.
3186 	 */
3187 	handle->read_op = pcap_read_linux_mmap;
3188 	handle->cleanup_op = pcap_cleanup_linux_mmap;
3189 	handle->setfilter_op = pcap_setfilter_linux_mmap;
3190 	handle->setnonblock_op = pcap_setnonblock_mmap;
3191 	handle->getnonblock_op = pcap_getnonblock_mmap;
3192 	handle->oneshot_callback = pcap_oneshot_mmap;
3193 	handle->selectable_fd = handle->fd;
3194 	return 1;
3195 }
3196 #else /* HAVE_PACKET_RING */
3197 static int
3198 activate_mmap(pcap_t *handle _U_, int *status _U_)
3199 {
3200 	return 0;
3201 }
3202 #endif /* HAVE_PACKET_RING */
3203 
3204 #ifdef HAVE_PACKET_RING
3205 /*
3206  * Attempt to set the socket to version 2 of the memory-mapped header.
3207  * Return 1 if we succeed or if we fail because version 2 isn't
3208  * supported; return -1 on any other error, and set handle->errbuf.
3209  */
3210 static int
3211 prepare_tpacket_socket(pcap_t *handle)
3212 {
3213 #ifdef HAVE_TPACKET2
3214 	socklen_t len;
3215 	int val;
3216 #endif
3217 
3218 	handle->md.tp_version = TPACKET_V1;
3219 	handle->md.tp_hdrlen = sizeof(struct tpacket_hdr);
3220 
3221 #ifdef HAVE_TPACKET2
3222 	/* Probe whether kernel supports TPACKET_V2 */
3223 	val = TPACKET_V2;
3224 	len = sizeof(val);
3225 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
3226 		if (errno == ENOPROTOOPT)
3227 			return 1;	/* no - just drive on */
3228 
3229 		/* Yes - treat as a failure. */
3230 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3231 		    "can't get TPACKET_V2 header len on packet socket: %s",
3232 		    pcap_strerror(errno));
3233 		return -1;
3234 	}
3235 	handle->md.tp_hdrlen = val;
3236 
3237 	val = TPACKET_V2;
3238 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
3239 		       sizeof(val)) < 0) {
3240 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3241 		    "can't activate TPACKET_V2 on packet socket: %s",
3242 		    pcap_strerror(errno));
3243 		return -1;
3244 	}
3245 	handle->md.tp_version = TPACKET_V2;
3246 
3247 	/* Reserve space for VLAN tag reconstruction */
3248 	val = VLAN_TAG_LEN;
3249 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val,
3250 		       sizeof(val)) < 0) {
3251 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3252 		    "can't set up reserve on packet socket: %s",
3253 		    pcap_strerror(errno));
3254 		return -1;
3255 	}
3256 
3257 #endif /* HAVE_TPACKET2 */
3258 	return 1;
3259 }
3260 
3261 /*
3262  * Attempt to set up memory-mapped access.
3263  *
3264  * On success, returns 1, and sets *status to 0 if there are no warnings
3265  * or to a PCAP_WARNING_ code if there is a warning.
3266  *
3267  * On failure due to lack of support for memory-mapped capture, returns
3268  * 0.
3269  *
3270  * On error, returns -1, and sets *status to the appropriate error code;
3271  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3272  */
3273 static int
3274 create_ring(pcap_t *handle, int *status)
3275 {
3276 	unsigned i, j, frames_per_block;
3277 	struct tpacket_req req;
3278 	socklen_t len;
3279 	unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
3280 	unsigned int frame_size;
3281 
3282 	/*
3283 	 * Start out assuming no warnings or errors.
3284 	 */
3285 	*status = 0;
3286 
3287 	/* Note that with large snapshot length (say 64K, which is the default
3288 	 * for recent versions of tcpdump, the value that "-s 0" has given
3289 	 * for a long time with tcpdump, and the default in Wireshark/TShark),
3290 	 * if we use the snapshot length to calculate the frame length,
3291 	 * only a few frames will be available in the ring even with pretty
3292 	 * large ring size (and a lot of memory will be unused).
3293 	 *
3294 	 * Ideally, we should choose a frame length based on the
3295 	 * minimum of the specified snapshot length and the maximum
3296 	 * packet size.  That's not as easy as it sounds; consider, for
3297 	 * example, an 802.11 interface in monitor mode, where the
3298 	 * frame would include a radiotap header, where the maximum
3299 	 * radiotap header length is device-dependent.
3300 	 *
3301 	 * So, for now, we just do this for Ethernet devices, where
3302 	 * there's no metadata header, and the link-layer header is
3303 	 * fixed length.  We can get the maximum packet size by
3304 	 * adding 18, the Ethernet header length plus the CRC length
3305 	 * (just in case we happen to get the CRC in the packet), to
3306 	 * the MTU of the interface; we fetch the MTU in the hopes
3307 	 * that it reflects support for jumbo frames.  (Even if the
3308 	 * interface is just being used for passive snooping, the driver
3309 	 * might set the size of buffers in the receive ring based on
3310 	 * the MTU, so that the MTU limits the maximum size of packets
3311 	 * that we can receive.)
3312 	 *
3313 	 * We don't do that if segmentation/fragmentation or receive
3314 	 * offload are enabled, so we don't get rudely surprised by
3315 	 * "packets" bigger than the MTU. */
3316 	frame_size = handle->snapshot;
3317 	if (handle->linktype == DLT_EN10MB) {
3318 		int mtu;
3319 		int offload;
3320 
3321 		offload = iface_get_offload(handle);
3322 		if (offload == -1) {
3323 			*status = PCAP_ERROR;
3324 			return -1;
3325 		}
3326 		if (!offload) {
3327 			mtu = iface_get_mtu(handle->fd, handle->opt.source,
3328 			    handle->errbuf);
3329 			if (mtu == -1) {
3330 				*status = PCAP_ERROR;
3331 				return -1;
3332 			}
3333 			if (frame_size > mtu + 18)
3334 				frame_size = mtu + 18;
3335 		}
3336 	}
3337 
3338 	/* NOTE: calculus matching those in tpacket_rcv()
3339 	 * in linux-2.6/net/packet/af_packet.c
3340 	 */
3341 	len = sizeof(sk_type);
3342 	if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type, &len) < 0) {
3343 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "getsockopt: %s", pcap_strerror(errno));
3344 		*status = PCAP_ERROR;
3345 		return -1;
3346 	}
3347 #ifdef PACKET_RESERVE
3348 	len = sizeof(tp_reserve);
3349 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &tp_reserve, &len) < 0) {
3350 		if (errno != ENOPROTOOPT) {
3351 			/*
3352 			 * ENOPROTOOPT means "kernel doesn't support
3353 			 * PACKET_RESERVE", in which case we fall back
3354 			 * as best we can.
3355 			 */
3356 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "getsockopt: %s", pcap_strerror(errno));
3357 			*status = PCAP_ERROR;
3358 			return -1;
3359 		}
3360 		tp_reserve = 0;	/* older kernel, reserve not supported */
3361 	}
3362 #else
3363 	tp_reserve = 0;	/* older kernel, reserve not supported */
3364 #endif
3365 	maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
3366 		/* XXX: in the kernel maclen is calculated from
3367 		 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
3368 		 * in:  packet_snd()           in linux-2.6/net/packet/af_packet.c
3369 		 * then packet_alloc_skb()     in linux-2.6/net/packet/af_packet.c
3370 		 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
3371 		 * but I see no way to get those sizes in userspace,
3372 		 * like for instance with an ifreq ioctl();
3373 		 * the best thing I've found so far is MAX_HEADER in the kernel
3374 		 * part of linux-2.6/include/linux/netdevice.h
3375 		 * which goes up to 128+48=176; since pcap-linux.c defines
3376 		 * a MAX_LINKHEADER_SIZE of 256 which is greater than that,
3377 		 * let's use it.. maybe is it even large enough to directly
3378 		 * replace macoff..
3379 		 */
3380 	tp_hdrlen = TPACKET_ALIGN(handle->md.tp_hdrlen) + sizeof(struct sockaddr_ll) ;
3381 	netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
3382 		/* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN of
3383 		 * netoff, which contradicts
3384 		 * linux-2.6/Documentation/networking/packet_mmap.txt
3385 		 * documenting that:
3386 		 * "- Gap, chosen so that packet data (Start+tp_net)
3387 		 * aligns to TPACKET_ALIGNMENT=16"
3388 		 */
3389 		/* NOTE: in linux-2.6/include/linux/skbuff.h:
3390 		 * "CPUs often take a performance hit
3391 		 *  when accessing unaligned memory locations"
3392 		 */
3393 	macoff = netoff - maclen;
3394 	req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
3395 	req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
3396 
3397 	/* compute the minumum block size that will handle this frame.
3398 	 * The block has to be page size aligned.
3399 	 * The max block size allowed by the kernel is arch-dependent and
3400 	 * it's not explicitly checked here. */
3401 	req.tp_block_size = getpagesize();
3402 	while (req.tp_block_size < req.tp_frame_size)
3403 		req.tp_block_size <<= 1;
3404 
3405 	frames_per_block = req.tp_block_size/req.tp_frame_size;
3406 
3407 	/*
3408 	 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
3409 	 * so we check for PACKET_TIMESTAMP.  We check for
3410 	 * linux/net_tstamp.h just in case a system somehow has
3411 	 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
3412 	 * be unnecessary.
3413 	 *
3414 	 * SIOCSHWTSTAMP was introduced in the patch that introduced
3415 	 * linux/net_tstamp.h, so we don't bother checking whether
3416 	 * SIOCSHWTSTAMP is defined (if your Linux system has
3417 	 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
3418 	 * Linux system is badly broken).
3419 	 */
3420 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
3421 	/*
3422 	 * If we were told to do so, ask the kernel and the driver
3423 	 * to use hardware timestamps.
3424 	 *
3425 	 * Hardware timestamps are only supported with mmapped
3426 	 * captures.
3427 	 */
3428 	if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
3429 	    handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
3430 		struct hwtstamp_config hwconfig;
3431 		struct ifreq ifr;
3432 		int timesource;
3433 
3434 		/*
3435 		 * Ask for hardware time stamps on all packets,
3436 		 * including transmitted packets.
3437 		 */
3438 		memset(&hwconfig, 0, sizeof(hwconfig));
3439 		hwconfig.tx_type = HWTSTAMP_TX_ON;
3440 		hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
3441 
3442 		memset(&ifr, 0, sizeof(ifr));
3443 		strcpy(ifr.ifr_name, handle->opt.source);
3444 		ifr.ifr_data = (void *)&hwconfig;
3445 
3446 		if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
3447 			switch (errno) {
3448 
3449 			case EPERM:
3450 				/*
3451 				 * Treat this as an error, as the
3452 				 * user should try to run this
3453 				 * with the appropriate privileges -
3454 				 * and, if they can't, shouldn't
3455 				 * try requesting hardware time stamps.
3456 				 */
3457 				*status = PCAP_ERROR_PERM_DENIED;
3458 				return -1;
3459 
3460 			case EOPNOTSUPP:
3461 				/*
3462 				 * Treat this as a warning, as the
3463 				 * only way to fix the warning is to
3464 				 * get an adapter that supports hardware
3465 				 * time stamps.  We'll just fall back
3466 				 * on the standard host time stamps.
3467 				 */
3468 				*status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
3469 				break;
3470 
3471 			default:
3472 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3473 					"SIOCSHWTSTAMP failed: %s",
3474 					pcap_strerror(errno));
3475 				*status = PCAP_ERROR;
3476 				return -1;
3477 			}
3478 		} else {
3479 			/*
3480 			 * Well, that worked.  Now specify the type of
3481 			 * hardware time stamp we want for this
3482 			 * socket.
3483 			 */
3484 			if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
3485 				/*
3486 				 * Hardware timestamp, synchronized
3487 				 * with the system clock.
3488 				 */
3489 				timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
3490 			} else {
3491 				/*
3492 				 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
3493 				 * timestamp, not synchronized with the
3494 				 * system clock.
3495 				 */
3496 				timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
3497 			}
3498 			if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
3499 				(void *)&timesource, sizeof(timesource))) {
3500 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3501 					"can't set PACKET_TIMESTAMP: %s",
3502 					pcap_strerror(errno));
3503 				*status = PCAP_ERROR;
3504 				return -1;
3505 			}
3506 		}
3507 	}
3508 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
3509 
3510 	/* ask the kernel to create the ring */
3511 retry:
3512 	req.tp_block_nr = req.tp_frame_nr / frames_per_block;
3513 
3514 	/* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
3515 	req.tp_frame_nr = req.tp_block_nr * frames_per_block;
3516 
3517 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3518 					(void *) &req, sizeof(req))) {
3519 		if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
3520 			/*
3521 			 * Memory failure; try to reduce the requested ring
3522 			 * size.
3523 			 *
3524 			 * We used to reduce this by half -- do 5% instead.
3525 			 * That may result in more iterations and a longer
3526 			 * startup, but the user will be much happier with
3527 			 * the resulting buffer size.
3528 			 */
3529 			if (req.tp_frame_nr < 20)
3530 				req.tp_frame_nr -= 1;
3531 			else
3532 				req.tp_frame_nr -= req.tp_frame_nr/20;
3533 			goto retry;
3534 		}
3535 		if (errno == ENOPROTOOPT) {
3536 			/*
3537 			 * We don't have ring buffer support in this kernel.
3538 			 */
3539 			return 0;
3540 		}
3541 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3542 		    "can't create rx ring on packet socket: %s",
3543 		    pcap_strerror(errno));
3544 		*status = PCAP_ERROR;
3545 		return -1;
3546 	}
3547 
3548 	/* memory map the rx ring */
3549 	handle->md.mmapbuflen = req.tp_block_nr * req.tp_block_size;
3550 	handle->md.mmapbuf = mmap(0, handle->md.mmapbuflen,
3551 	    PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
3552 	if (handle->md.mmapbuf == MAP_FAILED) {
3553 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3554 		    "can't mmap rx ring: %s", pcap_strerror(errno));
3555 
3556 		/* clear the allocated ring on error*/
3557 		destroy_ring(handle);
3558 		*status = PCAP_ERROR;
3559 		return -1;
3560 	}
3561 
3562 	/* allocate a ring for each frame header pointer*/
3563 	handle->cc = req.tp_frame_nr;
3564 	handle->buffer = malloc(handle->cc * sizeof(union thdr *));
3565 	if (!handle->buffer) {
3566 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3567 		    "can't allocate ring of frame headers: %s",
3568 		    pcap_strerror(errno));
3569 
3570 		destroy_ring(handle);
3571 		*status = PCAP_ERROR;
3572 		return -1;
3573 	}
3574 
3575 	/* fill the header ring with proper frame ptr*/
3576 	handle->offset = 0;
3577 	for (i=0; i<req.tp_block_nr; ++i) {
3578 		void *base = &handle->md.mmapbuf[i*req.tp_block_size];
3579 		for (j=0; j<frames_per_block; ++j, ++handle->offset) {
3580 			RING_GET_FRAME(handle) = base;
3581 			base += req.tp_frame_size;
3582 		}
3583 	}
3584 
3585 	handle->bufsize = req.tp_frame_size;
3586 	handle->offset = 0;
3587 	return 1;
3588 }
3589 
3590 /* free all ring related resources*/
3591 static void
3592 destroy_ring(pcap_t *handle)
3593 {
3594 	/* tell the kernel to destroy the ring*/
3595 	struct tpacket_req req;
3596 	memset(&req, 0, sizeof(req));
3597 	setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3598 				(void *) &req, sizeof(req));
3599 
3600 	/* if ring is mapped, unmap it*/
3601 	if (handle->md.mmapbuf) {
3602 		/* do not test for mmap failure, as we can't recover from any error */
3603 		munmap(handle->md.mmapbuf, handle->md.mmapbuflen);
3604 		handle->md.mmapbuf = NULL;
3605 	}
3606 }
3607 
3608 /*
3609  * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
3610  * for Linux mmapped capture.
3611  *
3612  * The problem is that pcap_next() and pcap_next_ex() expect the packet
3613  * data handed to the callback to be valid after the callback returns,
3614  * but pcap_read_linux_mmap() has to release that packet as soon as
3615  * the callback returns (otherwise, the kernel thinks there's still
3616  * at least one unprocessed packet available in the ring, so a select()
3617  * will immediately return indicating that there's data to process), so,
3618  * in the callback, we have to make a copy of the packet.
3619  *
3620  * Yes, this means that, if the capture is using the ring buffer, using
3621  * pcap_next() or pcap_next_ex() requires more copies than using
3622  * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
3623  * pcap_next() or pcap_next_ex().
3624  */
3625 static void
3626 pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
3627     const u_char *bytes)
3628 {
3629 	struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
3630 
3631 	*sp->hdr = *h;
3632 	memcpy(sp->pd->md.oneshot_buffer, bytes, h->caplen);
3633 	*sp->pkt = sp->pd->md.oneshot_buffer;
3634 }
3635 
3636 static void
3637 pcap_cleanup_linux_mmap( pcap_t *handle )
3638 {
3639 	destroy_ring(handle);
3640 	if (handle->md.oneshot_buffer != NULL) {
3641 		free(handle->md.oneshot_buffer);
3642 		handle->md.oneshot_buffer = NULL;
3643 	}
3644 	pcap_cleanup_linux(handle);
3645 }
3646 
3647 
3648 static int
3649 pcap_getnonblock_mmap(pcap_t *p, char *errbuf)
3650 {
3651 	/* use negative value of timeout to indicate non blocking ops */
3652 	return (p->md.timeout<0);
3653 }
3654 
3655 static int
3656 pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
3657 {
3658 	/*
3659 	 * Map each value to their corresponding negation to
3660 	 * preserve the timeout value provided with pcap_set_timeout.
3661 	 */
3662 	if (nonblock) {
3663 		if (p->md.timeout >= 0) {
3664 			/*
3665 			 * Indicate that we're switching to
3666 			 * non-blocking mode.
3667 			 */
3668 			p->md.timeout = ~p->md.timeout;
3669 		}
3670 	} else {
3671 		if (p->md.timeout < 0) {
3672 			p->md.timeout = ~p->md.timeout;
3673 		}
3674 	}
3675 	return 0;
3676 }
3677 
3678 static inline union thdr *
3679 pcap_get_ring_frame(pcap_t *handle, int status)
3680 {
3681 	union thdr h;
3682 
3683 	h.raw = RING_GET_FRAME(handle);
3684 	switch (handle->md.tp_version) {
3685 	case TPACKET_V1:
3686 		if (status != (h.h1->tp_status ? TP_STATUS_USER :
3687 						TP_STATUS_KERNEL))
3688 			return NULL;
3689 		break;
3690 #ifdef HAVE_TPACKET2
3691 	case TPACKET_V2:
3692 		if (status != (h.h2->tp_status ? TP_STATUS_USER :
3693 						TP_STATUS_KERNEL))
3694 			return NULL;
3695 		break;
3696 #endif
3697 	}
3698 	return h.raw;
3699 }
3700 
3701 #ifndef POLLRDHUP
3702 #define POLLRDHUP 0
3703 #endif
3704 
3705 static int
3706 pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
3707 		u_char *user)
3708 {
3709 	int timeout;
3710 	int pkts = 0;
3711 	char c;
3712 
3713 	/* wait for frames availability.*/
3714 	if (!pcap_get_ring_frame(handle, TP_STATUS_USER)) {
3715 		struct pollfd pollinfo;
3716 		int ret;
3717 
3718 		pollinfo.fd = handle->fd;
3719 		pollinfo.events = POLLIN;
3720 
3721 		if (handle->md.timeout == 0)
3722 			timeout = -1;	/* block forever */
3723 		else if (handle->md.timeout > 0)
3724 			timeout = handle->md.timeout;	/* block for that amount of time */
3725 		else
3726 			timeout = 0;	/* non-blocking mode - poll to pick up errors */
3727 		do {
3728 			ret = poll(&pollinfo, 1, timeout);
3729 			if (ret < 0 && errno != EINTR) {
3730 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3731 					"can't poll on packet socket: %s",
3732 					pcap_strerror(errno));
3733 				return PCAP_ERROR;
3734 			} else if (ret > 0 &&
3735 			    (pollinfo.revents & (POLLHUP|POLLRDHUP|POLLERR|POLLNVAL))) {
3736 				/*
3737 				 * There's some indication other than
3738 				 * "you can read on this descriptor" on
3739 				 * the descriptor.
3740 				 */
3741 				if (pollinfo.revents & (POLLHUP | POLLRDHUP)) {
3742 					snprintf(handle->errbuf,
3743 						PCAP_ERRBUF_SIZE,
3744 						"Hangup on packet socket");
3745 					return PCAP_ERROR;
3746 				}
3747 				if (pollinfo.revents & POLLERR) {
3748 					/*
3749 					 * A recv() will give us the
3750 					 * actual error code.
3751 					 *
3752 					 * XXX - make the socket non-blocking?
3753 					 */
3754 					if (recv(handle->fd, &c, sizeof c,
3755 					    MSG_PEEK) != -1)
3756 						continue;	/* what, no error? */
3757 					if (errno == ENETDOWN) {
3758 						/*
3759 						 * The device on which we're
3760 						 * capturing went away.
3761 						 *
3762 						 * XXX - we should really return
3763 						 * PCAP_ERROR_IFACE_NOT_UP,
3764 						 * but pcap_dispatch() etc.
3765 						 * aren't defined to return
3766 						 * that.
3767 						 */
3768 						snprintf(handle->errbuf,
3769 							PCAP_ERRBUF_SIZE,
3770 							"The interface went down");
3771 					} else {
3772 						snprintf(handle->errbuf,
3773 							PCAP_ERRBUF_SIZE,
3774 							"Error condition on packet socket: %s",
3775 							strerror(errno));
3776 					}
3777 					return PCAP_ERROR;
3778 				}
3779 				if (pollinfo.revents & POLLNVAL) {
3780 					snprintf(handle->errbuf,
3781 						PCAP_ERRBUF_SIZE,
3782 						"Invalid polling request on packet socket");
3783 					return PCAP_ERROR;
3784 				}
3785   			}
3786 			/* check for break loop condition on interrupted syscall*/
3787 			if (handle->break_loop) {
3788 				handle->break_loop = 0;
3789 				return PCAP_ERROR_BREAK;
3790 			}
3791 		} while (ret < 0);
3792 	}
3793 
3794 	/* non-positive values of max_packets are used to require all
3795 	 * packets currently available in the ring */
3796 	while ((pkts < max_packets) || (max_packets <= 0)) {
3797 		int run_bpf;
3798 		struct sockaddr_ll *sll;
3799 		struct pcap_pkthdr pcaphdr;
3800 		unsigned char *bp;
3801 		union thdr h;
3802 		unsigned int tp_len;
3803 		unsigned int tp_mac;
3804 		unsigned int tp_snaplen;
3805 		unsigned int tp_sec;
3806 		unsigned int tp_usec;
3807 
3808 		h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
3809 		if (!h.raw)
3810 			break;
3811 
3812 		switch (handle->md.tp_version) {
3813 		case TPACKET_V1:
3814 			tp_len	   = h.h1->tp_len;
3815 			tp_mac	   = h.h1->tp_mac;
3816 			tp_snaplen = h.h1->tp_snaplen;
3817 			tp_sec	   = h.h1->tp_sec;
3818 			tp_usec	   = h.h1->tp_usec;
3819 			break;
3820 #ifdef HAVE_TPACKET2
3821 		case TPACKET_V2:
3822 			tp_len	   = h.h2->tp_len;
3823 			tp_mac	   = h.h2->tp_mac;
3824 			tp_snaplen = h.h2->tp_snaplen;
3825 			tp_sec	   = h.h2->tp_sec;
3826 			tp_usec	   = h.h2->tp_nsec / 1000;
3827 			break;
3828 #endif
3829 		default:
3830 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3831 				"unsupported tpacket version %d",
3832 				handle->md.tp_version);
3833 			return -1;
3834 		}
3835 		/* perform sanity check on internal offset. */
3836 		if (tp_mac + tp_snaplen > handle->bufsize) {
3837 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3838 				"corrupted frame on kernel ring mac "
3839 				"offset %d + caplen %d > frame len %d",
3840 				tp_mac, tp_snaplen, handle->bufsize);
3841 			return -1;
3842 		}
3843 
3844 		/* run filter on received packet
3845 		 * If the kernel filtering is enabled we need to run the
3846 		 * filter until all the frames present into the ring
3847 		 * at filter creation time are processed.
3848 		 * In such case md.use_bpf is used as a counter for the
3849 		 * packet we need to filter.
3850 		 * Note: alternatively it could be possible to stop applying
3851 		 * the filter when the ring became empty, but it can possibly
3852 		 * happen a lot later... */
3853 		bp = (unsigned char*)h.raw + tp_mac;
3854 		run_bpf = (!handle->md.use_bpf) ||
3855 			((handle->md.use_bpf>1) && handle->md.use_bpf--);
3856 		if (run_bpf && handle->fcode.bf_insns &&
3857 				(bpf_filter(handle->fcode.bf_insns, bp,
3858 					tp_len, tp_snaplen) == 0))
3859 			goto skip;
3860 
3861 		/*
3862 		 * Do checks based on packet direction.
3863 		 */
3864 		sll = (void *)h.raw + TPACKET_ALIGN(handle->md.tp_hdrlen);
3865 		if (sll->sll_pkttype == PACKET_OUTGOING) {
3866 			/*
3867 			 * Outgoing packet.
3868 			 * If this is from the loopback device, reject it;
3869 			 * we'll see the packet as an incoming packet as well,
3870 			 * and we don't want to see it twice.
3871 			 */
3872 			if (sll->sll_ifindex == handle->md.lo_ifindex)
3873 				goto skip;
3874 
3875 			/*
3876 			 * If the user only wants incoming packets, reject it.
3877 			 */
3878 			if (handle->direction == PCAP_D_IN)
3879 				goto skip;
3880 		} else {
3881 			/*
3882 			 * Incoming packet.
3883 			 * If the user only wants outgoing packets, reject it.
3884 			 */
3885 			if (handle->direction == PCAP_D_OUT)
3886 				goto skip;
3887 		}
3888 
3889 		/* get required packet info from ring header */
3890 		pcaphdr.ts.tv_sec = tp_sec;
3891 		pcaphdr.ts.tv_usec = tp_usec;
3892 		pcaphdr.caplen = tp_snaplen;
3893 		pcaphdr.len = tp_len;
3894 
3895 		/* if required build in place the sll header*/
3896 		if (handle->md.cooked) {
3897 			struct sll_header *hdrp;
3898 
3899 			/*
3900 			 * The kernel should have left us with enough
3901 			 * space for an sll header; back up the packet
3902 			 * data pointer into that space, as that'll be
3903 			 * the beginning of the packet we pass to the
3904 			 * callback.
3905 			 */
3906 			bp -= SLL_HDR_LEN;
3907 
3908 			/*
3909 			 * Let's make sure that's past the end of
3910 			 * the tpacket header, i.e. >=
3911 			 * ((u_char *)thdr + TPACKET_HDRLEN), so we
3912 			 * don't step on the header when we construct
3913 			 * the sll header.
3914 			 */
3915 			if (bp < (u_char *)h.raw +
3916 					   TPACKET_ALIGN(handle->md.tp_hdrlen) +
3917 					   sizeof(struct sockaddr_ll)) {
3918 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3919 					"cooked-mode frame doesn't have room for sll header");
3920 				return -1;
3921 			}
3922 
3923 			/*
3924 			 * OK, that worked; construct the sll header.
3925 			 */
3926 			hdrp = (struct sll_header *)bp;
3927 			hdrp->sll_pkttype = map_packet_type_to_sll_type(
3928 							sll->sll_pkttype);
3929 			hdrp->sll_hatype = htons(sll->sll_hatype);
3930 			hdrp->sll_halen = htons(sll->sll_halen);
3931 			memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
3932 			hdrp->sll_protocol = sll->sll_protocol;
3933 
3934 			/* update packet len */
3935 			pcaphdr.caplen += SLL_HDR_LEN;
3936 			pcaphdr.len += SLL_HDR_LEN;
3937 		}
3938 
3939 #ifdef HAVE_TPACKET2
3940 		if ((handle->md.tp_version == TPACKET_V2) &&
3941 #if defined(TP_STATUS_VLAN_VALID)
3942 		(h.h2->tp_vlan_tci || (h.h2->tp_status & TP_STATUS_VLAN_VALID)) &&
3943 #else
3944 		h.h2->tp_vlan_tci &&
3945 #endif
3946 		    handle->md.vlan_offset != -1 &&
3947 		    tp_snaplen >= (unsigned int) handle->md.vlan_offset) {
3948 			struct vlan_tag *tag;
3949 
3950 			bp -= VLAN_TAG_LEN;
3951 			memmove(bp, bp + VLAN_TAG_LEN, handle->md.vlan_offset);
3952 
3953 			tag = (struct vlan_tag *)(bp + handle->md.vlan_offset);
3954 			tag->vlan_tpid = htons(ETH_P_8021Q);
3955 			tag->vlan_tci = htons(h.h2->tp_vlan_tci);
3956 
3957 			pcaphdr.caplen += VLAN_TAG_LEN;
3958 			pcaphdr.len += VLAN_TAG_LEN;
3959 		}
3960 #endif
3961 
3962 		/*
3963 		 * The only way to tell the kernel to cut off the
3964 		 * packet at a snapshot length is with a filter program;
3965 		 * if there's no filter program, the kernel won't cut
3966 		 * the packet off.
3967 		 *
3968 		 * Trim the snapshot length to be no longer than the
3969 		 * specified snapshot length.
3970 		 */
3971 		if (pcaphdr.caplen > handle->snapshot)
3972 			pcaphdr.caplen = handle->snapshot;
3973 
3974 		/* pass the packet to the user */
3975 		pkts++;
3976 		callback(user, &pcaphdr, bp);
3977 		handle->md.packets_read++;
3978 
3979 skip:
3980 		/* next packet */
3981 		switch (handle->md.tp_version) {
3982 		case TPACKET_V1:
3983 			h.h1->tp_status = TP_STATUS_KERNEL;
3984 			break;
3985 #ifdef HAVE_TPACKET2
3986 		case TPACKET_V2:
3987 			h.h2->tp_status = TP_STATUS_KERNEL;
3988 			break;
3989 #endif
3990 		}
3991 		if (++handle->offset >= handle->cc)
3992 			handle->offset = 0;
3993 
3994 		/* check for break loop condition*/
3995 		if (handle->break_loop) {
3996 			handle->break_loop = 0;
3997 			return PCAP_ERROR_BREAK;
3998 		}
3999 	}
4000 	return pkts;
4001 }
4002 
4003 static int
4004 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
4005 {
4006 	int n, offset;
4007 	int ret;
4008 
4009 	/*
4010 	 * Don't rewrite "ret" instructions; we don't need to, as
4011 	 * we're not reading packets with recvmsg(), and we don't
4012 	 * want to, as, by not rewriting them, the kernel can avoid
4013 	 * copying extra data.
4014 	 */
4015 	ret = pcap_setfilter_linux_common(handle, filter, 1);
4016 	if (ret < 0)
4017 		return ret;
4018 
4019 	/* if the kernel filter is enabled, we need to apply the filter on
4020 	 * all packets present into the ring. Get an upper bound of their number
4021 	 */
4022 	if (!handle->md.use_bpf)
4023 		return ret;
4024 
4025 	/* walk the ring backward and count the free slot */
4026 	offset = handle->offset;
4027 	if (--handle->offset < 0)
4028 		handle->offset = handle->cc - 1;
4029 	for (n=0; n < handle->cc; ++n) {
4030 		if (--handle->offset < 0)
4031 			handle->offset = handle->cc - 1;
4032 		if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL))
4033 			break;
4034 	}
4035 
4036 	/* be careful to not change current ring position */
4037 	handle->offset = offset;
4038 
4039 	/* store the number of packets currently present in the ring */
4040 	handle->md.use_bpf = 1 + (handle->cc - n);
4041 	return ret;
4042 }
4043 
4044 #endif /* HAVE_PACKET_RING */
4045 
4046 
4047 #ifdef HAVE_PF_PACKET_SOCKETS
4048 /*
4049  *  Return the index of the given device name. Fill ebuf and return
4050  *  -1 on failure.
4051  */
4052 static int
4053 iface_get_id(int fd, const char *device, char *ebuf)
4054 {
4055 	struct ifreq	ifr;
4056 
4057 	memset(&ifr, 0, sizeof(ifr));
4058 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4059 
4060 	if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
4061 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4062 			 "SIOCGIFINDEX: %s", pcap_strerror(errno));
4063 		return -1;
4064 	}
4065 
4066 	return ifr.ifr_ifindex;
4067 }
4068 
4069 /*
4070  *  Bind the socket associated with FD to the given device.
4071  *  Return 1 on success, 0 if we should try a SOCK_PACKET socket,
4072  *  or a PCAP_ERROR_ value on a hard error.
4073  */
4074 static int
4075 iface_bind(int fd, int ifindex, char *ebuf)
4076 {
4077 	struct sockaddr_ll	sll;
4078 	int			err;
4079 	socklen_t		errlen = sizeof(err);
4080 
4081 	memset(&sll, 0, sizeof(sll));
4082 	sll.sll_family		= AF_PACKET;
4083 	sll.sll_ifindex		= ifindex;
4084 	sll.sll_protocol	= htons(ETH_P_ALL);
4085 
4086 	if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
4087 		if (errno == ENETDOWN) {
4088 			/*
4089 			 * Return a "network down" indication, so that
4090 			 * the application can report that rather than
4091 			 * saying we had a mysterious failure and
4092 			 * suggest that they report a problem to the
4093 			 * libpcap developers.
4094 			 */
4095 			return PCAP_ERROR_IFACE_NOT_UP;
4096 		} else {
4097 			snprintf(ebuf, PCAP_ERRBUF_SIZE,
4098 				 "bind: %s", pcap_strerror(errno));
4099 			return PCAP_ERROR;
4100 		}
4101 	}
4102 
4103 	/* Any pending errors, e.g., network is down? */
4104 
4105 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
4106 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4107 			"getsockopt: %s", pcap_strerror(errno));
4108 		return 0;
4109 	}
4110 
4111 	if (err == ENETDOWN) {
4112 		/*
4113 		 * Return a "network down" indication, so that
4114 		 * the application can report that rather than
4115 		 * saying we had a mysterious failure and
4116 		 * suggest that they report a problem to the
4117 		 * libpcap developers.
4118 		 */
4119 		return PCAP_ERROR_IFACE_NOT_UP;
4120 	} else if (err > 0) {
4121 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4122 			"bind: %s", pcap_strerror(err));
4123 		return 0;
4124 	}
4125 
4126 	return 1;
4127 }
4128 
4129 #ifdef IW_MODE_MONITOR
4130 /*
4131  * Check whether the device supports the Wireless Extensions.
4132  * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE
4133  * if the device doesn't even exist.
4134  */
4135 static int
4136 has_wext(int sock_fd, const char *device, char *ebuf)
4137 {
4138 	struct iwreq ireq;
4139 
4140 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4141 	    sizeof ireq.ifr_ifrn.ifrn_name);
4142 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4143 	if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
4144 		return 1;	/* yes */
4145 	snprintf(ebuf, PCAP_ERRBUF_SIZE,
4146 	    "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
4147 	if (errno == ENODEV)
4148 		return PCAP_ERROR_NO_SUCH_DEVICE;
4149 	return 0;
4150 }
4151 
4152 /*
4153  * Per me si va ne la citta dolente,
4154  * Per me si va ne l'etterno dolore,
4155  *	...
4156  * Lasciate ogne speranza, voi ch'intrate.
4157  *
4158  * XXX - airmon-ng does special stuff with the Orinoco driver and the
4159  * wlan-ng driver.
4160  */
4161 typedef enum {
4162 	MONITOR_WEXT,
4163 	MONITOR_HOSTAP,
4164 	MONITOR_PRISM,
4165 	MONITOR_PRISM54,
4166 	MONITOR_ACX100,
4167 	MONITOR_RT2500,
4168 	MONITOR_RT2570,
4169 	MONITOR_RT73,
4170 	MONITOR_RTL8XXX
4171 } monitor_type;
4172 
4173 /*
4174  * Use the Wireless Extensions, if we have them, to try to turn monitor mode
4175  * on if it's not already on.
4176  *
4177  * Returns 1 on success, 0 if we don't support the Wireless Extensions
4178  * on this device, or a PCAP_ERROR_ value if we do support them but
4179  * we weren't able to turn monitor mode on.
4180  */
4181 static int
4182 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
4183 {
4184 	/*
4185 	 * XXX - at least some adapters require non-Wireless Extensions
4186 	 * mechanisms to turn monitor mode on.
4187 	 *
4188 	 * Atheros cards might require that a separate "monitor virtual access
4189 	 * point" be created, with later versions of the madwifi driver.
4190 	 * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode
4191 	 * monitor -bssid", which apparently spits out a line "athN"
4192 	 * where "athN" is the monitor mode device.  To leave monitor
4193 	 * mode, it destroys the monitor mode device.
4194 	 *
4195 	 * Some Intel Centrino adapters might require private ioctls to get
4196 	 * radio headers; the ipw2200 and ipw3945 drivers allow you to
4197 	 * configure a separate "rtapN" interface to capture in monitor
4198 	 * mode without preventing the adapter from operating normally.
4199 	 * (airmon-ng doesn't appear to use that, though.)
4200 	 *
4201 	 * It would be Truly Wonderful if mac80211 and nl80211 cleaned this
4202 	 * up, and if all drivers were converted to mac80211 drivers.
4203 	 *
4204 	 * If interface {if} is a mac80211 driver, the file
4205 	 * /sys/class/net/{if}/phy80211 is a symlink to
4206 	 * /sys/class/ieee80211/{phydev}, for some {phydev}.
4207 	 *
4208 	 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
4209 	 * least, has a "wmaster0" device and a "wlan0" device; the
4210 	 * latter is the one with the IP address.  Both show up in
4211 	 * "tcpdump -D" output.  Capturing on the wmaster0 device
4212 	 * captures with 802.11 headers.
4213 	 *
4214 	 * airmon-ng searches through /sys/class/net for devices named
4215 	 * monN, starting with mon0; as soon as one *doesn't* exist,
4216 	 * it chooses that as the monitor device name.  If the "iw"
4217 	 * command exists, it does "iw dev {if} interface add {monif}
4218 	 * type monitor", where {monif} is the monitor device.  It
4219 	 * then (sigh) sleeps .1 second, and then configures the
4220 	 * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
4221 	 * is a file, it writes {mondev}, without a newline, to that file,
4222 	 * and again (sigh) sleeps .1 second, and then iwconfig's that
4223 	 * device into monitor mode and configures it up.  Otherwise,
4224 	 * you can't do monitor mode.
4225 	 *
4226 	 * All these devices are "glued" together by having the
4227 	 * /sys/class/net/{device}/phy80211 links pointing to the same
4228 	 * place, so, given a wmaster, wlan, or mon device, you can
4229 	 * find the other devices by looking for devices with
4230 	 * the same phy80211 link.
4231 	 *
4232 	 * To turn monitor mode off, delete the monitor interface,
4233 	 * either with "iw dev {monif} interface del" or by sending
4234 	 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
4235 	 *
4236 	 * Note: if you try to create a monitor device named "monN", and
4237 	 * there's already a "monN" device, it fails, as least with
4238 	 * the netlink interface (which is what iw uses), with a return
4239 	 * value of -ENFILE.  (Return values are negative errnos.)  We
4240 	 * could probably use that to find an unused device.
4241 	 */
4242 	int err;
4243 	struct iwreq ireq;
4244 	struct iw_priv_args *priv;
4245 	monitor_type montype;
4246 	int i;
4247 	__u32 cmd;
4248 	struct ifreq ifr;
4249 	int oldflags;
4250 	int args[2];
4251 	int channel;
4252 
4253 	/*
4254 	 * Does this device *support* the Wireless Extensions?
4255 	 */
4256 	err = has_wext(sock_fd, device, handle->errbuf);
4257 	if (err <= 0)
4258 		return err;	/* either it doesn't or the device doesn't even exist */
4259 	/*
4260 	 * Start out assuming we have no private extensions to control
4261 	 * radio metadata.
4262 	 */
4263 	montype = MONITOR_WEXT;
4264 	cmd = 0;
4265 
4266 	/*
4267 	 * Try to get all the Wireless Extensions private ioctls
4268 	 * supported by this device.
4269 	 *
4270 	 * First, get the size of the buffer we need, by supplying no
4271 	 * buffer and a length of 0.  If the device supports private
4272 	 * ioctls, it should return E2BIG, with ireq.u.data.length set
4273 	 * to the length we need.  If it doesn't support them, it should
4274 	 * return EOPNOTSUPP.
4275 	 */
4276 	memset(&ireq, 0, sizeof ireq);
4277 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4278 	    sizeof ireq.ifr_ifrn.ifrn_name);
4279 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4280 	ireq.u.data.pointer = (void *)args;
4281 	ireq.u.data.length = 0;
4282 	ireq.u.data.flags = 0;
4283 	if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
4284 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4285 		    "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!",
4286 		    device);
4287 		return PCAP_ERROR;
4288 	}
4289 	if (errno != EOPNOTSUPP) {
4290 		/*
4291 		 * OK, it's not as if there are no private ioctls.
4292 		 */
4293 		if (errno != E2BIG) {
4294 			/*
4295 			 * Failed.
4296 			 */
4297 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4298 			    "%s: SIOCGIWPRIV: %s", device,
4299 			    pcap_strerror(errno));
4300 			return PCAP_ERROR;
4301 		}
4302 
4303 		/*
4304 		 * OK, try to get the list of private ioctls.
4305 		 */
4306 		priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args));
4307 		if (priv == NULL) {
4308 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4309 			    "malloc: %s", pcap_strerror(errno));
4310 			return PCAP_ERROR;
4311 		}
4312 		ireq.u.data.pointer = (void *)priv;
4313 		if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) {
4314 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4315 			    "%s: SIOCGIWPRIV: %s", device,
4316 			    pcap_strerror(errno));
4317 			free(priv);
4318 			return PCAP_ERROR;
4319 		}
4320 
4321 		/*
4322 		 * Look for private ioctls to turn monitor mode on or, if
4323 		 * monitor mode is on, to set the header type.
4324 		 */
4325 		for (i = 0; i < ireq.u.data.length; i++) {
4326 			if (strcmp(priv[i].name, "monitor_type") == 0) {
4327 				/*
4328 				 * Hostap driver, use this one.
4329 				 * Set monitor mode first.
4330 				 * You can set it to 0 to get DLT_IEEE80211,
4331 				 * 1 to get DLT_PRISM, 2 to get
4332 				 * DLT_IEEE80211_RADIO_AVS, and, with more
4333 				 * recent versions of the driver, 3 to get
4334 				 * DLT_IEEE80211_RADIO.
4335 				 */
4336 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4337 					break;
4338 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4339 					break;
4340 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4341 					break;
4342 				montype = MONITOR_HOSTAP;
4343 				cmd = priv[i].cmd;
4344 				break;
4345 			}
4346 			if (strcmp(priv[i].name, "set_prismhdr") == 0) {
4347 				/*
4348 				 * Prism54 driver, use this one.
4349 				 * Set monitor mode first.
4350 				 * You can set it to 2 to get DLT_IEEE80211
4351 				 * or 3 or get DLT_PRISM.
4352 				 */
4353 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4354 					break;
4355 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4356 					break;
4357 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4358 					break;
4359 				montype = MONITOR_PRISM54;
4360 				cmd = priv[i].cmd;
4361 				break;
4362 			}
4363 			if (strcmp(priv[i].name, "forceprismheader") == 0) {
4364 				/*
4365 				 * RT2570 driver, use this one.
4366 				 * Do this after turning monitor mode on.
4367 				 * You can set it to 1 to get DLT_PRISM or 2
4368 				 * to get DLT_IEEE80211.
4369 				 */
4370 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4371 					break;
4372 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4373 					break;
4374 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4375 					break;
4376 				montype = MONITOR_RT2570;
4377 				cmd = priv[i].cmd;
4378 				break;
4379 			}
4380 			if (strcmp(priv[i].name, "forceprism") == 0) {
4381 				/*
4382 				 * RT73 driver, use this one.
4383 				 * Do this after turning monitor mode on.
4384 				 * Its argument is a *string*; you can
4385 				 * set it to "1" to get DLT_PRISM or "2"
4386 				 * to get DLT_IEEE80211.
4387 				 */
4388 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR)
4389 					break;
4390 				if (priv[i].set_args & IW_PRIV_SIZE_FIXED)
4391 					break;
4392 				montype = MONITOR_RT73;
4393 				cmd = priv[i].cmd;
4394 				break;
4395 			}
4396 			if (strcmp(priv[i].name, "prismhdr") == 0) {
4397 				/*
4398 				 * One of the RTL8xxx drivers, use this one.
4399 				 * It can only be done after monitor mode
4400 				 * has been turned on.  You can set it to 1
4401 				 * to get DLT_PRISM or 0 to get DLT_IEEE80211.
4402 				 */
4403 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4404 					break;
4405 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4406 					break;
4407 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4408 					break;
4409 				montype = MONITOR_RTL8XXX;
4410 				cmd = priv[i].cmd;
4411 				break;
4412 			}
4413 			if (strcmp(priv[i].name, "rfmontx") == 0) {
4414 				/*
4415 				 * RT2500 or RT61 driver, use this one.
4416 				 * It has one one-byte parameter; set
4417 				 * u.data.length to 1 and u.data.pointer to
4418 				 * point to the parameter.
4419 				 * It doesn't itself turn monitor mode on.
4420 				 * You can set it to 1 to allow transmitting
4421 				 * in monitor mode(?) and get DLT_IEEE80211,
4422 				 * or set it to 0 to disallow transmitting in
4423 				 * monitor mode(?) and get DLT_PRISM.
4424 				 */
4425 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4426 					break;
4427 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2)
4428 					break;
4429 				montype = MONITOR_RT2500;
4430 				cmd = priv[i].cmd;
4431 				break;
4432 			}
4433 			if (strcmp(priv[i].name, "monitor") == 0) {
4434 				/*
4435 				 * Either ACX100 or hostap, use this one.
4436 				 * It turns monitor mode on.
4437 				 * If it takes two arguments, it's ACX100;
4438 				 * the first argument is 1 for DLT_PRISM
4439 				 * or 2 for DLT_IEEE80211, and the second
4440 				 * argument is the channel on which to
4441 				 * run.  If it takes one argument, it's
4442 				 * HostAP, and the argument is 2 for
4443 				 * DLT_IEEE80211 and 3 for DLT_PRISM.
4444 				 *
4445 				 * If we see this, we don't quit, as this
4446 				 * might be a version of the hostap driver
4447 				 * that also supports "monitor_type".
4448 				 */
4449 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4450 					break;
4451 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4452 					break;
4453 				switch (priv[i].set_args & IW_PRIV_SIZE_MASK) {
4454 
4455 				case 1:
4456 					montype = MONITOR_PRISM;
4457 					cmd = priv[i].cmd;
4458 					break;
4459 
4460 				case 2:
4461 					montype = MONITOR_ACX100;
4462 					cmd = priv[i].cmd;
4463 					break;
4464 
4465 				default:
4466 					break;
4467 				}
4468 			}
4469 		}
4470 		free(priv);
4471 	}
4472 
4473 	/*
4474 	 * XXX - ipw3945?  islism?
4475 	 */
4476 
4477 	/*
4478 	 * Get the old mode.
4479 	 */
4480 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4481 	    sizeof ireq.ifr_ifrn.ifrn_name);
4482 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4483 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) {
4484 		/*
4485 		 * We probably won't be able to set the mode, either.
4486 		 */
4487 		return PCAP_ERROR_RFMON_NOTSUP;
4488 	}
4489 
4490 	/*
4491 	 * Is it currently in monitor mode?
4492 	 */
4493 	if (ireq.u.mode == IW_MODE_MONITOR) {
4494 		/*
4495 		 * Yes.  Just leave things as they are.
4496 		 * We don't offer multiple link-layer types, as
4497 		 * changing the link-layer type out from under
4498 		 * somebody else capturing in monitor mode would
4499 		 * be considered rude.
4500 		 */
4501 		return 1;
4502 	}
4503 	/*
4504 	 * No.  We have to put the adapter into rfmon mode.
4505 	 */
4506 
4507 	/*
4508 	 * If we haven't already done so, arrange to have
4509 	 * "pcap_close_all()" called when we exit.
4510 	 */
4511 	if (!pcap_do_addexit(handle)) {
4512 		/*
4513 		 * "atexit()" failed; don't put the interface
4514 		 * in rfmon mode, just give up.
4515 		 */
4516 		return PCAP_ERROR_RFMON_NOTSUP;
4517 	}
4518 
4519 	/*
4520 	 * Save the old mode.
4521 	 */
4522 	handle->md.oldmode = ireq.u.mode;
4523 
4524 	/*
4525 	 * Put the adapter in rfmon mode.  How we do this depends
4526 	 * on whether we have a special private ioctl or not.
4527 	 */
4528 	if (montype == MONITOR_PRISM) {
4529 		/*
4530 		 * We have the "monitor" private ioctl, but none of
4531 		 * the other private ioctls.  Use this, and select
4532 		 * the Prism header.
4533 		 *
4534 		 * If it fails, just fall back on SIOCSIWMODE.
4535 		 */
4536 		memset(&ireq, 0, sizeof ireq);
4537 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4538 		    sizeof ireq.ifr_ifrn.ifrn_name);
4539 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4540 		ireq.u.data.length = 1;	/* 1 argument */
4541 		args[0] = 3;	/* request Prism header */
4542 		memcpy(ireq.u.name, args, IFNAMSIZ);
4543 		if (ioctl(sock_fd, cmd, &ireq) != -1) {
4544 			/*
4545 			 * Success.
4546 			 * Note that we have to put the old mode back
4547 			 * when we close the device.
4548 			 */
4549 			handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
4550 
4551 			/*
4552 			 * Add this to the list of pcaps to close
4553 			 * when we exit.
4554 			 */
4555 			pcap_add_to_pcaps_to_close(handle);
4556 
4557 			return 1;
4558 		}
4559 
4560 		/*
4561 		 * Failure.  Fall back on SIOCSIWMODE.
4562 		 */
4563 	}
4564 
4565 	/*
4566 	 * First, take the interface down if it's up; otherwise, we
4567 	 * might get EBUSY.
4568 	 */
4569 	memset(&ifr, 0, sizeof(ifr));
4570 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4571 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
4572 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4573 		    "%s: Can't get flags: %s", device, strerror(errno));
4574 		return PCAP_ERROR;
4575 	}
4576 	oldflags = 0;
4577 	if (ifr.ifr_flags & IFF_UP) {
4578 		oldflags = ifr.ifr_flags;
4579 		ifr.ifr_flags &= ~IFF_UP;
4580 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4581 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4582 			    "%s: Can't set flags: %s", device, strerror(errno));
4583 			return PCAP_ERROR;
4584 		}
4585 	}
4586 
4587 	/*
4588 	 * Then turn monitor mode on.
4589 	 */
4590 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4591 	    sizeof ireq.ifr_ifrn.ifrn_name);
4592 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4593 	ireq.u.mode = IW_MODE_MONITOR;
4594 	if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) {
4595 		/*
4596 		 * Scientist, you've failed.
4597 		 * Bring the interface back up if we shut it down.
4598 		 */
4599 		ifr.ifr_flags = oldflags;
4600 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4601 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4602 			    "%s: Can't set flags: %s", device, strerror(errno));
4603 			return PCAP_ERROR;
4604 		}
4605 		return PCAP_ERROR_RFMON_NOTSUP;
4606 	}
4607 
4608 	/*
4609 	 * XXX - airmon-ng does "iwconfig {if} key off" after setting
4610 	 * monitor mode and setting the channel, and then does
4611 	 * "iwconfig up".
4612 	 */
4613 
4614 	/*
4615 	 * Now select the appropriate radio header.
4616 	 */
4617 	switch (montype) {
4618 
4619 	case MONITOR_WEXT:
4620 		/*
4621 		 * We don't have any private ioctl to set the header.
4622 		 */
4623 		break;
4624 
4625 	case MONITOR_HOSTAP:
4626 		/*
4627 		 * Try to select the radiotap header.
4628 		 */
4629 		memset(&ireq, 0, sizeof ireq);
4630 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4631 		    sizeof ireq.ifr_ifrn.ifrn_name);
4632 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4633 		args[0] = 3;	/* request radiotap header */
4634 		memcpy(ireq.u.name, args, sizeof (int));
4635 		if (ioctl(sock_fd, cmd, &ireq) != -1)
4636 			break;	/* success */
4637 
4638 		/*
4639 		 * That failed.  Try to select the AVS header.
4640 		 */
4641 		memset(&ireq, 0, sizeof ireq);
4642 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4643 		    sizeof ireq.ifr_ifrn.ifrn_name);
4644 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4645 		args[0] = 2;	/* request AVS header */
4646 		memcpy(ireq.u.name, args, sizeof (int));
4647 		if (ioctl(sock_fd, cmd, &ireq) != -1)
4648 			break;	/* success */
4649 
4650 		/*
4651 		 * That failed.  Try to select the Prism header.
4652 		 */
4653 		memset(&ireq, 0, sizeof ireq);
4654 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4655 		    sizeof ireq.ifr_ifrn.ifrn_name);
4656 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4657 		args[0] = 1;	/* request Prism header */
4658 		memcpy(ireq.u.name, args, sizeof (int));
4659 		ioctl(sock_fd, cmd, &ireq);
4660 		break;
4661 
4662 	case MONITOR_PRISM:
4663 		/*
4664 		 * The private ioctl failed.
4665 		 */
4666 		break;
4667 
4668 	case MONITOR_PRISM54:
4669 		/*
4670 		 * Select the Prism header.
4671 		 */
4672 		memset(&ireq, 0, sizeof ireq);
4673 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4674 		    sizeof ireq.ifr_ifrn.ifrn_name);
4675 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4676 		args[0] = 3;	/* request Prism header */
4677 		memcpy(ireq.u.name, args, sizeof (int));
4678 		ioctl(sock_fd, cmd, &ireq);
4679 		break;
4680 
4681 	case MONITOR_ACX100:
4682 		/*
4683 		 * Get the current channel.
4684 		 */
4685 		memset(&ireq, 0, sizeof ireq);
4686 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4687 		    sizeof ireq.ifr_ifrn.ifrn_name);
4688 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4689 		if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) {
4690 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4691 			    "%s: SIOCGIWFREQ: %s", device,
4692 			    pcap_strerror(errno));
4693 			return PCAP_ERROR;
4694 		}
4695 		channel = ireq.u.freq.m;
4696 
4697 		/*
4698 		 * Select the Prism header, and set the channel to the
4699 		 * current value.
4700 		 */
4701 		memset(&ireq, 0, sizeof ireq);
4702 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4703 		    sizeof ireq.ifr_ifrn.ifrn_name);
4704 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4705 		args[0] = 1;		/* request Prism header */
4706 		args[1] = channel;	/* set channel */
4707 		memcpy(ireq.u.name, args, 2*sizeof (int));
4708 		ioctl(sock_fd, cmd, &ireq);
4709 		break;
4710 
4711 	case MONITOR_RT2500:
4712 		/*
4713 		 * Disallow transmission - that turns on the
4714 		 * Prism header.
4715 		 */
4716 		memset(&ireq, 0, sizeof ireq);
4717 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4718 		    sizeof ireq.ifr_ifrn.ifrn_name);
4719 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4720 		args[0] = 0;	/* disallow transmitting */
4721 		memcpy(ireq.u.name, args, sizeof (int));
4722 		ioctl(sock_fd, cmd, &ireq);
4723 		break;
4724 
4725 	case MONITOR_RT2570:
4726 		/*
4727 		 * Force the Prism header.
4728 		 */
4729 		memset(&ireq, 0, sizeof ireq);
4730 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4731 		    sizeof ireq.ifr_ifrn.ifrn_name);
4732 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4733 		args[0] = 1;	/* request Prism header */
4734 		memcpy(ireq.u.name, args, sizeof (int));
4735 		ioctl(sock_fd, cmd, &ireq);
4736 		break;
4737 
4738 	case MONITOR_RT73:
4739 		/*
4740 		 * Force the Prism header.
4741 		 */
4742 		memset(&ireq, 0, sizeof ireq);
4743 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4744 		    sizeof ireq.ifr_ifrn.ifrn_name);
4745 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4746 		ireq.u.data.length = 1;	/* 1 argument */
4747 		ireq.u.data.pointer = "1";
4748 		ireq.u.data.flags = 0;
4749 		ioctl(sock_fd, cmd, &ireq);
4750 		break;
4751 
4752 	case MONITOR_RTL8XXX:
4753 		/*
4754 		 * Force the Prism header.
4755 		 */
4756 		memset(&ireq, 0, sizeof ireq);
4757 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4758 		    sizeof ireq.ifr_ifrn.ifrn_name);
4759 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4760 		args[0] = 1;	/* request Prism header */
4761 		memcpy(ireq.u.name, args, sizeof (int));
4762 		ioctl(sock_fd, cmd, &ireq);
4763 		break;
4764 	}
4765 
4766 	/*
4767 	 * Now bring the interface back up if we brought it down.
4768 	 */
4769 	if (oldflags != 0) {
4770 		ifr.ifr_flags = oldflags;
4771 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4772 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4773 			    "%s: Can't set flags: %s", device, strerror(errno));
4774 
4775 			/*
4776 			 * At least try to restore the old mode on the
4777 			 * interface.
4778 			 */
4779 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
4780 				/*
4781 				 * Scientist, you've failed.
4782 				 */
4783 				fprintf(stderr,
4784 				    "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n"
4785 				    "Please adjust manually.\n",
4786 				    strerror(errno));
4787 			}
4788 			return PCAP_ERROR;
4789 		}
4790 	}
4791 
4792 	/*
4793 	 * Note that we have to put the old mode back when we
4794 	 * close the device.
4795 	 */
4796 	handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
4797 
4798 	/*
4799 	 * Add this to the list of pcaps to close when we exit.
4800 	 */
4801 	pcap_add_to_pcaps_to_close(handle);
4802 
4803 	return 1;
4804 }
4805 #endif /* IW_MODE_MONITOR */
4806 
4807 /*
4808  * Try various mechanisms to enter monitor mode.
4809  */
4810 static int
4811 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
4812 {
4813 #if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
4814 	int ret;
4815 #endif
4816 
4817 #ifdef HAVE_LIBNL
4818 	ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
4819 	if (ret < 0)
4820 		return ret;	/* error attempting to do so */
4821 	if (ret == 1)
4822 		return 1;	/* success */
4823 #endif /* HAVE_LIBNL */
4824 
4825 #ifdef IW_MODE_MONITOR
4826 	ret = enter_rfmon_mode_wext(handle, sock_fd, device);
4827 	if (ret < 0)
4828 		return ret;	/* error attempting to do so */
4829 	if (ret == 1)
4830 		return 1;	/* success */
4831 #endif /* IW_MODE_MONITOR */
4832 
4833 	/*
4834 	 * Either none of the mechanisms we know about work or none
4835 	 * of those mechanisms are available, so we can't do monitor
4836 	 * mode.
4837 	 */
4838 	return 0;
4839 }
4840 
4841 /*
4842  * Find out if we have any form of fragmentation/reassembly offloading.
4843  *
4844  * We do so using SIOCETHTOOL checking for various types of offloading;
4845  * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
4846  * of the types of offloading, there's nothing we can do to check, so
4847  * we just say "no, we don't".
4848  */
4849 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
4850 static int
4851 iface_ethtool_ioctl(pcap_t *handle, int cmd, const char *cmdname)
4852 {
4853 	struct ifreq	ifr;
4854 	struct ethtool_value eval;
4855 
4856 	memset(&ifr, 0, sizeof(ifr));
4857 	strncpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name));
4858 	eval.cmd = cmd;
4859 	ifr.ifr_data = (caddr_t)&eval;
4860 	if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
4861 		if (errno == EOPNOTSUPP || errno == EINVAL) {
4862 			/*
4863 			 * OK, let's just return 0, which, in our
4864 			 * case, either means "no, what we're asking
4865 			 * about is not enabled" or "all the flags
4866 			 * are clear (i.e., nothing is enabled)".
4867 			 */
4868 			return 0;
4869 		}
4870 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4871 		    "%s: SIOETHTOOL(%s) ioctl failed: %s", handle->opt.source,
4872 		    cmdname, strerror(errno));
4873 		return -1;
4874 	}
4875 	return eval.data;
4876 }
4877 
4878 static int
4879 iface_get_offload(pcap_t *handle)
4880 {
4881 	int ret;
4882 
4883 #ifdef ETHTOOL_GTSO
4884 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO");
4885 	if (ret == -1)
4886 		return -1;
4887 	if (ret)
4888 		return 1;	/* TCP segmentation offloading on */
4889 #endif
4890 
4891 #ifdef ETHTOOL_GUFO
4892 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO");
4893 	if (ret == -1)
4894 		return -1;
4895 	if (ret)
4896 		return 1;	/* UDP fragmentation offloading on */
4897 #endif
4898 
4899 #ifdef ETHTOOL_GGSO
4900 	/*
4901 	 * XXX - will this cause large unsegmented packets to be
4902 	 * handed to PF_PACKET sockets on transmission?  If not,
4903 	 * this need not be checked.
4904 	 */
4905 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO");
4906 	if (ret == -1)
4907 		return -1;
4908 	if (ret)
4909 		return 1;	/* generic segmentation offloading on */
4910 #endif
4911 
4912 #ifdef ETHTOOL_GFLAGS
4913 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4914 	if (ret == -1)
4915 		return -1;
4916 	if (ret & ETH_FLAG_LRO)
4917 		return 1;	/* large receive offloading on */
4918 #endif
4919 
4920 #ifdef ETHTOOL_GGRO
4921 	/*
4922 	 * XXX - will this cause large reassembled packets to be
4923 	 * handed to PF_PACKET sockets on receipt?  If not,
4924 	 * this need not be checked.
4925 	 */
4926 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO");
4927 	if (ret == -1)
4928 		return -1;
4929 	if (ret)
4930 		return 1;	/* generic (large) receive offloading on */
4931 #endif
4932 
4933 	return 0;
4934 }
4935 #else /* SIOCETHTOOL */
4936 static int
4937 iface_get_offload(pcap_t *handle _U_)
4938 {
4939 	/*
4940 	 * XXX - do we need to get this information if we don't
4941 	 * have the ethtool ioctls?  If so, how do we do that?
4942 	 */
4943 	return 0;
4944 }
4945 #endif /* SIOCETHTOOL */
4946 
4947 #endif /* HAVE_PF_PACKET_SOCKETS */
4948 
4949 /* ===== Functions to interface to the older kernels ================== */
4950 
4951 /*
4952  * Try to open a packet socket using the old kernel interface.
4953  * Returns 1 on success and a PCAP_ERROR_ value on an error.
4954  */
4955 static int
4956 activate_old(pcap_t *handle)
4957 {
4958 	int		arptype;
4959 	struct ifreq	ifr;
4960 	const char	*device = handle->opt.source;
4961 	struct utsname	utsname;
4962 	int		mtu;
4963 
4964 	/* Open the socket */
4965 
4966 	handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
4967 	if (handle->fd == -1) {
4968 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4969 			 "socket: %s", pcap_strerror(errno));
4970 		if (errno == EPERM || errno == EACCES) {
4971 			/*
4972 			 * You don't have permission to open the
4973 			 * socket.
4974 			 */
4975 			return PCAP_ERROR_PERM_DENIED;
4976 		} else {
4977 			/*
4978 			 * Other error.
4979 			 */
4980 			return PCAP_ERROR;
4981 		}
4982 	}
4983 
4984 	/* It worked - we are using the old interface */
4985 	handle->md.sock_packet = 1;
4986 
4987 	/* ...which means we get the link-layer header. */
4988 	handle->md.cooked = 0;
4989 
4990 	/* Bind to the given device */
4991 
4992 	if (strcmp(device, "any") == 0) {
4993 		strncpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
4994 			PCAP_ERRBUF_SIZE);
4995 		return PCAP_ERROR;
4996 	}
4997 	if (iface_bind_old(handle->fd, device, handle->errbuf) == -1)
4998 		return PCAP_ERROR;
4999 
5000 	/*
5001 	 * Try to get the link-layer type.
5002 	 */
5003 	arptype = iface_get_arptype(handle->fd, device, handle->errbuf);
5004 	if (arptype < 0)
5005 		return PCAP_ERROR;
5006 
5007 	/*
5008 	 * Try to find the DLT_ type corresponding to that
5009 	 * link-layer type.
5010 	 */
5011 	map_arphrd_to_dlt(handle, arptype, 0);
5012 	if (handle->linktype == -1) {
5013 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5014 			 "unknown arptype %d", arptype);
5015 		return PCAP_ERROR;
5016 	}
5017 
5018 	/* Go to promisc mode if requested */
5019 
5020 	if (handle->opt.promisc) {
5021 		memset(&ifr, 0, sizeof(ifr));
5022 		strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5023 		if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
5024 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5025 				 "SIOCGIFFLAGS: %s", pcap_strerror(errno));
5026 			return PCAP_ERROR;
5027 		}
5028 		if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
5029 			/*
5030 			 * Promiscuous mode isn't currently on,
5031 			 * so turn it on, and remember that
5032 			 * we should turn it off when the
5033 			 * pcap_t is closed.
5034 			 */
5035 
5036 			/*
5037 			 * If we haven't already done so, arrange
5038 			 * to have "pcap_close_all()" called when
5039 			 * we exit.
5040 			 */
5041 			if (!pcap_do_addexit(handle)) {
5042 				/*
5043 				 * "atexit()" failed; don't put
5044 				 * the interface in promiscuous
5045 				 * mode, just give up.
5046 				 */
5047 				return PCAP_ERROR;
5048 			}
5049 
5050 			ifr.ifr_flags |= IFF_PROMISC;
5051 			if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
5052 			        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5053 					 "SIOCSIFFLAGS: %s",
5054 					 pcap_strerror(errno));
5055 				return PCAP_ERROR;
5056 			}
5057 			handle->md.must_do_on_close |= MUST_CLEAR_PROMISC;
5058 
5059 			/*
5060 			 * Add this to the list of pcaps
5061 			 * to close when we exit.
5062 			 */
5063 			pcap_add_to_pcaps_to_close(handle);
5064 		}
5065 	}
5066 
5067 	/*
5068 	 * Compute the buffer size.
5069 	 *
5070 	 * We're using SOCK_PACKET, so this might be a 2.0[.x]
5071 	 * kernel, and might require special handling - check.
5072 	 */
5073 	if (uname(&utsname) < 0 ||
5074 	    strncmp(utsname.release, "2.0", 3) == 0) {
5075 		/*
5076 		 * Either we couldn't find out what kernel release
5077 		 * this is, or it's a 2.0[.x] kernel.
5078 		 *
5079 		 * In the 2.0[.x] kernel, a "recvfrom()" on
5080 		 * a SOCK_PACKET socket, with MSG_TRUNC set, will
5081 		 * return the number of bytes read, so if we pass
5082 		 * a length based on the snapshot length, it'll
5083 		 * return the number of bytes from the packet
5084 		 * copied to userland, not the actual length
5085 		 * of the packet.
5086 		 *
5087 		 * This means that, for example, the IP dissector
5088 		 * in tcpdump will get handed a packet length less
5089 		 * than the length in the IP header, and will
5090 		 * complain about "truncated-ip".
5091 		 *
5092 		 * So we don't bother trying to copy from the
5093 		 * kernel only the bytes in which we're interested,
5094 		 * but instead copy them all, just as the older
5095 		 * versions of libpcap for Linux did.
5096 		 *
5097 		 * The buffer therefore needs to be big enough to
5098 		 * hold the largest packet we can get from this
5099 		 * device.  Unfortunately, we can't get the MRU
5100 		 * of the network; we can only get the MTU.  The
5101 		 * MTU may be too small, in which case a packet larger
5102 		 * than the buffer size will be truncated *and* we
5103 		 * won't get the actual packet size.
5104 		 *
5105 		 * However, if the snapshot length is larger than
5106 		 * the buffer size based on the MTU, we use the
5107 		 * snapshot length as the buffer size, instead;
5108 		 * this means that with a sufficiently large snapshot
5109 		 * length we won't artificially truncate packets
5110 		 * to the MTU-based size.
5111 		 *
5112 		 * This mess just one of many problems with packet
5113 		 * capture on 2.0[.x] kernels; you really want a
5114 		 * 2.2[.x] or later kernel if you want packet capture
5115 		 * to work well.
5116 		 */
5117 		mtu = iface_get_mtu(handle->fd, device, handle->errbuf);
5118 		if (mtu == -1)
5119 			return PCAP_ERROR;
5120 		handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
5121 		if (handle->bufsize < handle->snapshot)
5122 			handle->bufsize = handle->snapshot;
5123 	} else {
5124 		/*
5125 		 * This is a 2.2[.x] or later kernel.
5126 		 *
5127 		 * We can safely pass "recvfrom()" a byte count
5128 		 * based on the snapshot length.
5129 		 */
5130 		handle->bufsize = handle->snapshot;
5131 	}
5132 
5133 	/*
5134 	 * Default value for offset to align link-layer payload
5135 	 * on a 4-byte boundary.
5136 	 */
5137 	handle->offset	 = 0;
5138 
5139 	/*
5140 	 * SOCK_PACKET sockets don't supply information from
5141 	 * stripped VLAN tags.
5142 	 */
5143 	handle->md.vlan_offset = -1; /* unknown */
5144 
5145 	return 1;
5146 }
5147 
5148 /*
5149  *  Bind the socket associated with FD to the given device using the
5150  *  interface of the old kernels.
5151  */
5152 static int
5153 iface_bind_old(int fd, const char *device, char *ebuf)
5154 {
5155 	struct sockaddr	saddr;
5156 	int		err;
5157 	socklen_t	errlen = sizeof(err);
5158 
5159 	memset(&saddr, 0, sizeof(saddr));
5160 	strncpy(saddr.sa_data, device, sizeof(saddr.sa_data));
5161 	if (bind(fd, &saddr, sizeof(saddr)) == -1) {
5162 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5163 			 "bind: %s", pcap_strerror(errno));
5164 		return -1;
5165 	}
5166 
5167 	/* Any pending errors, e.g., network is down? */
5168 
5169 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
5170 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5171 			"getsockopt: %s", pcap_strerror(errno));
5172 		return -1;
5173 	}
5174 
5175 	if (err > 0) {
5176 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5177 			"bind: %s", pcap_strerror(err));
5178 		return -1;
5179 	}
5180 
5181 	return 0;
5182 }
5183 
5184 
5185 /* ===== System calls available on all supported kernels ============== */
5186 
5187 /*
5188  *  Query the kernel for the MTU of the given interface.
5189  */
5190 static int
5191 iface_get_mtu(int fd, const char *device, char *ebuf)
5192 {
5193 	struct ifreq	ifr;
5194 
5195 	if (!device)
5196 		return BIGGER_THAN_ALL_MTUS;
5197 
5198 	memset(&ifr, 0, sizeof(ifr));
5199 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5200 
5201 	if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
5202 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5203 			 "SIOCGIFMTU: %s", pcap_strerror(errno));
5204 		return -1;
5205 	}
5206 
5207 	return ifr.ifr_mtu;
5208 }
5209 
5210 /*
5211  *  Get the hardware type of the given interface as ARPHRD_xxx constant.
5212  */
5213 static int
5214 iface_get_arptype(int fd, const char *device, char *ebuf)
5215 {
5216 	struct ifreq	ifr;
5217 
5218 	memset(&ifr, 0, sizeof(ifr));
5219 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5220 
5221 	if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
5222 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5223 			 "SIOCGIFHWADDR: %s", pcap_strerror(errno));
5224 		if (errno == ENODEV) {
5225 			/*
5226 			 * No such device.
5227 			 */
5228 			return PCAP_ERROR_NO_SUCH_DEVICE;
5229 		}
5230 		return PCAP_ERROR;
5231 	}
5232 
5233 	return ifr.ifr_hwaddr.sa_family;
5234 }
5235 
5236 #ifdef SO_ATTACH_FILTER
5237 static int
5238 fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
5239 {
5240 	size_t prog_size;
5241 	register int i;
5242 	register struct bpf_insn *p;
5243 	struct bpf_insn *f;
5244 	int len;
5245 
5246 	/*
5247 	 * Make a copy of the filter, and modify that copy if
5248 	 * necessary.
5249 	 */
5250 	prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
5251 	len = handle->fcode.bf_len;
5252 	f = (struct bpf_insn *)malloc(prog_size);
5253 	if (f == NULL) {
5254 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5255 			 "malloc: %s", pcap_strerror(errno));
5256 		return -1;
5257 	}
5258 	memcpy(f, handle->fcode.bf_insns, prog_size);
5259 	fcode->len = len;
5260 	fcode->filter = (struct sock_filter *) f;
5261 
5262 	for (i = 0; i < len; ++i) {
5263 		p = &f[i];
5264 		/*
5265 		 * What type of instruction is this?
5266 		 */
5267 		switch (BPF_CLASS(p->code)) {
5268 
5269 		case BPF_RET:
5270 			/*
5271 			 * It's a return instruction; are we capturing
5272 			 * in memory-mapped mode?
5273 			 */
5274 			if (!is_mmapped) {
5275 				/*
5276 				 * No; is the snapshot length a constant,
5277 				 * rather than the contents of the
5278 				 * accumulator?
5279 				 */
5280 				if (BPF_MODE(p->code) == BPF_K) {
5281 					/*
5282 					 * Yes - if the value to be returned,
5283 					 * i.e. the snapshot length, is
5284 					 * anything other than 0, make it
5285 					 * 65535, so that the packet is
5286 					 * truncated by "recvfrom()",
5287 					 * not by the filter.
5288 					 *
5289 					 * XXX - there's nothing we can
5290 					 * easily do if it's getting the
5291 					 * value from the accumulator; we'd
5292 					 * have to insert code to force
5293 					 * non-zero values to be 65535.
5294 					 */
5295 					if (p->k != 0)
5296 						p->k = 65535;
5297 				}
5298 			}
5299 			break;
5300 
5301 		case BPF_LD:
5302 		case BPF_LDX:
5303 			/*
5304 			 * It's a load instruction; is it loading
5305 			 * from the packet?
5306 			 */
5307 			switch (BPF_MODE(p->code)) {
5308 
5309 			case BPF_ABS:
5310 			case BPF_IND:
5311 			case BPF_MSH:
5312 				/*
5313 				 * Yes; are we in cooked mode?
5314 				 */
5315 				if (handle->md.cooked) {
5316 					/*
5317 					 * Yes, so we need to fix this
5318 					 * instruction.
5319 					 */
5320 					if (fix_offset(p) < 0) {
5321 						/*
5322 						 * We failed to do so.
5323 						 * Return 0, so our caller
5324 						 * knows to punt to userland.
5325 						 */
5326 						return 0;
5327 					}
5328 				}
5329 				break;
5330 			}
5331 			break;
5332 		}
5333 	}
5334 	return 1;	/* we succeeded */
5335 }
5336 
5337 static int
5338 fix_offset(struct bpf_insn *p)
5339 {
5340 	/*
5341 	 * What's the offset?
5342 	 */
5343 	if (p->k >= SLL_HDR_LEN) {
5344 		/*
5345 		 * It's within the link-layer payload; that starts at an
5346 		 * offset of 0, as far as the kernel packet filter is
5347 		 * concerned, so subtract the length of the link-layer
5348 		 * header.
5349 		 */
5350 		p->k -= SLL_HDR_LEN;
5351 	} else if (p->k == 0) {
5352 		/*
5353 		 * It's the packet type field; map it to the special magic
5354 		 * kernel offset for that field.
5355 		 */
5356 		p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
5357 	} else if (p->k == 14) {
5358 		/*
5359 		 * It's the protocol field; map it to the special magic
5360 		 * kernel offset for that field.
5361 		 */
5362 		p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5363 	} else if ((bpf_int32)(p->k) > 0) {
5364 		/*
5365 		 * It's within the header, but it's not one of those
5366 		 * fields; we can't do that in the kernel, so punt
5367 		 * to userland.
5368 		 */
5369 		return -1;
5370 	}
5371 	return 0;
5372 }
5373 
5374 static int
5375 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
5376 {
5377 	int total_filter_on = 0;
5378 	int save_mode;
5379 	int ret;
5380 	int save_errno;
5381 
5382 	/*
5383 	 * The socket filter code doesn't discard all packets queued
5384 	 * up on the socket when the filter is changed; this means
5385 	 * that packets that don't match the new filter may show up
5386 	 * after the new filter is put onto the socket, if those
5387 	 * packets haven't yet been read.
5388 	 *
5389 	 * This means, for example, that if you do a tcpdump capture
5390 	 * with a filter, the first few packets in the capture might
5391 	 * be packets that wouldn't have passed the filter.
5392 	 *
5393 	 * We therefore discard all packets queued up on the socket
5394 	 * when setting a kernel filter.  (This isn't an issue for
5395 	 * userland filters, as the userland filtering is done after
5396 	 * packets are queued up.)
5397 	 *
5398 	 * To flush those packets, we put the socket in read-only mode,
5399 	 * and read packets from the socket until there are no more to
5400 	 * read.
5401 	 *
5402 	 * In order to keep that from being an infinite loop - i.e.,
5403 	 * to keep more packets from arriving while we're draining
5404 	 * the queue - we put the "total filter", which is a filter
5405 	 * that rejects all packets, onto the socket before draining
5406 	 * the queue.
5407 	 *
5408 	 * This code deliberately ignores any errors, so that you may
5409 	 * get bogus packets if an error occurs, rather than having
5410 	 * the filtering done in userland even if it could have been
5411 	 * done in the kernel.
5412 	 */
5413 	if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5414 		       &total_fcode, sizeof(total_fcode)) == 0) {
5415 		char drain[1];
5416 
5417 		/*
5418 		 * Note that we've put the total filter onto the socket.
5419 		 */
5420 		total_filter_on = 1;
5421 
5422 		/*
5423 		 * Save the socket's current mode, and put it in
5424 		 * non-blocking mode; we drain it by reading packets
5425 		 * until we get an error (which is normally a
5426 		 * "nothing more to be read" error).
5427 		 */
5428 		save_mode = fcntl(handle->fd, F_GETFL, 0);
5429 		if (save_mode != -1 &&
5430 		    fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) >= 0) {
5431 			while (recv(handle->fd, &drain, sizeof drain,
5432 			       MSG_TRUNC) >= 0)
5433 				;
5434 			save_errno = errno;
5435 			fcntl(handle->fd, F_SETFL, save_mode);
5436 			if (save_errno != EAGAIN) {
5437 				/* Fatal error */
5438 				reset_kernel_filter(handle);
5439 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5440 				 "recv: %s", pcap_strerror(save_errno));
5441 				return -2;
5442 			}
5443 		}
5444 	}
5445 
5446 	/*
5447 	 * Now attach the new filter.
5448 	 */
5449 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5450 			 fcode, sizeof(*fcode));
5451 	if (ret == -1 && total_filter_on) {
5452 		/*
5453 		 * Well, we couldn't set that filter on the socket,
5454 		 * but we could set the total filter on the socket.
5455 		 *
5456 		 * This could, for example, mean that the filter was
5457 		 * too big to put into the kernel, so we'll have to
5458 		 * filter in userland; in any case, we'll be doing
5459 		 * filtering in userland, so we need to remove the
5460 		 * total filter so we see packets.
5461 		 */
5462 		save_errno = errno;
5463 
5464 		/*
5465 		 * XXX - if this fails, we're really screwed;
5466 		 * we have the total filter on the socket,
5467 		 * and it won't come off.  What do we do then?
5468 		 */
5469 		reset_kernel_filter(handle);
5470 
5471 		errno = save_errno;
5472 	}
5473 	return ret;
5474 }
5475 
5476 static int
5477 reset_kernel_filter(pcap_t *handle)
5478 {
5479 	/*
5480 	 * setsockopt() barfs unless it get a dummy parameter.
5481 	 * valgrind whines unless the value is initialized,
5482 	 * as it has no idea that setsockopt() ignores its
5483 	 * parameter.
5484 	 */
5485 	int dummy = 0;
5486 
5487 	return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
5488 				   &dummy, sizeof(dummy));
5489 }
5490 #endif
5491