xref: /freebsd/contrib/libpcap/pcap-linux.c (revision 145992504973bd16cf3518af9ba5ce185fefa82a)
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/socket.h>
131 #include <sys/ioctl.h>
132 #include <sys/utsname.h>
133 #include <sys/mman.h>
134 #include <linux/if.h>
135 #include <netinet/in.h>
136 #include <linux/if_ether.h>
137 #include <net/if_arp.h>
138 #include <poll.h>
139 #include <dirent.h>
140 
141 #include "pcap-int.h"
142 #include "pcap/sll.h"
143 #include "pcap/vlan.h"
144 
145 #ifdef HAVE_DAG_API
146 #include "pcap-dag.h"
147 #endif /* HAVE_DAG_API */
148 
149 #ifdef HAVE_SEPTEL_API
150 #include "pcap-septel.h"
151 #endif /* HAVE_SEPTEL_API */
152 
153 #ifdef HAVE_SNF_API
154 #include "pcap-snf.h"
155 #endif /* HAVE_SNF_API */
156 
157 #ifdef PCAP_SUPPORT_USB
158 #include "pcap-usb-linux.h"
159 #endif
160 
161 #ifdef PCAP_SUPPORT_BT
162 #include "pcap-bt-linux.h"
163 #endif
164 
165 #ifdef PCAP_SUPPORT_CAN
166 #include "pcap-can-linux.h"
167 #endif
168 
169 #if PCAP_SUPPORT_CANUSB
170 #include "pcap-canusb-linux.h"
171 #endif
172 
173 #ifdef PCAP_SUPPORT_NETFILTER
174 #include "pcap-netfilter-linux.h"
175 #endif
176 
177 /*
178  * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
179  * sockets rather than SOCK_PACKET sockets.
180  *
181  * To use them, we include <linux/if_packet.h> rather than
182  * <netpacket/packet.h>; we do so because
183  *
184  *	some Linux distributions (e.g., Slackware 4.0) have 2.2 or
185  *	later kernels and libc5, and don't provide a <netpacket/packet.h>
186  *	file;
187  *
188  *	not all versions of glibc2 have a <netpacket/packet.h> file
189  *	that defines stuff needed for some of the 2.4-or-later-kernel
190  *	features, so if the system has a 2.4 or later kernel, we
191  *	still can't use those features.
192  *
193  * We're already including a number of other <linux/XXX.h> headers, and
194  * this code is Linux-specific (no other OS has PF_PACKET sockets as
195  * a raw packet capture mechanism), so it's not as if you gain any
196  * useful portability by using <netpacket/packet.h>
197  *
198  * XXX - should we just include <linux/if_packet.h> even if PF_PACKET
199  * isn't defined?  It only defines one data structure in 2.0.x, so
200  * it shouldn't cause any problems.
201  */
202 #ifdef PF_PACKET
203 # include <linux/if_packet.h>
204 
205  /*
206   * On at least some Linux distributions (for example, Red Hat 5.2),
207   * there's no <netpacket/packet.h> file, but PF_PACKET is defined if
208   * you include <sys/socket.h>, but <linux/if_packet.h> doesn't define
209   * any of the PF_PACKET stuff such as "struct sockaddr_ll" or any of
210   * the PACKET_xxx stuff.
211   *
212   * So we check whether PACKET_HOST is defined, and assume that we have
213   * PF_PACKET sockets only if it is defined.
214   */
215 # ifdef PACKET_HOST
216 #  define HAVE_PF_PACKET_SOCKETS
217 #  ifdef PACKET_AUXDATA
218 #   define HAVE_PACKET_AUXDATA
219 #  endif /* PACKET_AUXDATA */
220 # endif /* PACKET_HOST */
221 
222 
223  /* check for memory mapped access avaibility. We assume every needed
224   * struct is defined if the macro TPACKET_HDRLEN is defined, because it
225   * uses many ring related structs and macros */
226 # ifdef TPACKET_HDRLEN
227 #  define HAVE_PACKET_RING
228 #  ifdef TPACKET2_HDRLEN
229 #   define HAVE_TPACKET2
230 #  else
231 #   define TPACKET_V1	0
232 #  endif /* TPACKET2_HDRLEN */
233 # endif /* TPACKET_HDRLEN */
234 #endif /* PF_PACKET */
235 
236 #ifdef SO_ATTACH_FILTER
237 #include <linux/types.h>
238 #include <linux/filter.h>
239 #endif
240 
241 /*
242  * We need linux/sockios.h if we have linux/net_tstamp.h (for time stamp
243  * specification) or linux/ethtool.h (for ethtool ioctls to get offloading
244  * information).
245  */
246 #if defined(HAVE_LINUX_NET_TSTAMP_H) || defined(HAVE_LINUX_ETHTOOL_H)
247 #include <linux/sockios.h>
248 #endif
249 
250 #ifdef HAVE_LINUX_NET_TSTAMP_H
251 #include <linux/net_tstamp.h>
252 #endif
253 
254 /*
255  * Got Wireless Extensions?
256  */
257 #ifdef HAVE_LINUX_WIRELESS_H
258 #include <linux/wireless.h>
259 #endif /* HAVE_LINUX_WIRELESS_H */
260 
261 /*
262  * Got libnl?
263  */
264 #ifdef HAVE_LIBNL
265 #include <linux/nl80211.h>
266 
267 #include <netlink/genl/genl.h>
268 #include <netlink/genl/family.h>
269 #include <netlink/genl/ctrl.h>
270 #include <netlink/msg.h>
271 #include <netlink/attr.h>
272 #endif /* HAVE_LIBNL */
273 
274 /*
275  * Got ethtool support?
276  */
277 #ifdef HAVE_LINUX_ETHTOOL_H
278 #include <linux/ethtool.h>
279 #endif
280 
281 #ifndef HAVE_SOCKLEN_T
282 typedef int		socklen_t;
283 #endif
284 
285 #ifndef MSG_TRUNC
286 /*
287  * This is being compiled on a system that lacks MSG_TRUNC; define it
288  * with the value it has in the 2.2 and later kernels, so that, on
289  * those kernels, when we pass it in the flags argument to "recvfrom()"
290  * we're passing the right value and thus get the MSG_TRUNC behavior
291  * we want.  (We don't get that behavior on 2.0[.x] kernels, because
292  * they didn't support MSG_TRUNC.)
293  */
294 #define MSG_TRUNC	0x20
295 #endif
296 
297 #ifndef SOL_PACKET
298 /*
299  * This is being compiled on a system that lacks SOL_PACKET; define it
300  * with the value it has in the 2.2 and later kernels, so that we can
301  * set promiscuous mode in the good modern way rather than the old
302  * 2.0-kernel crappy way.
303  */
304 #define SOL_PACKET	263
305 #endif
306 
307 #define MAX_LINKHEADER_SIZE	256
308 
309 /*
310  * When capturing on all interfaces we use this as the buffer size.
311  * Should be bigger then all MTUs that occur in real life.
312  * 64kB should be enough for now.
313  */
314 #define BIGGER_THAN_ALL_MTUS	(64*1024)
315 
316 /*
317  * Prototypes for internal functions and methods.
318  */
319 static void map_arphrd_to_dlt(pcap_t *, int, int);
320 #ifdef HAVE_PF_PACKET_SOCKETS
321 static short int map_packet_type_to_sll_type(short int);
322 #endif
323 static int pcap_activate_linux(pcap_t *);
324 static int activate_old(pcap_t *);
325 static int activate_new(pcap_t *);
326 static int activate_mmap(pcap_t *, int *);
327 static int pcap_can_set_rfmon_linux(pcap_t *);
328 static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *);
329 static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);
330 static int pcap_inject_linux(pcap_t *, const void *, size_t);
331 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
332 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
333 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
334 static void pcap_cleanup_linux(pcap_t *);
335 
336 union thdr {
337 	struct tpacket_hdr	*h1;
338 	struct tpacket2_hdr	*h2;
339 	void			*raw;
340 };
341 
342 #ifdef HAVE_PACKET_RING
343 #define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset])
344 
345 static void destroy_ring(pcap_t *handle);
346 static int create_ring(pcap_t *handle, int *status);
347 static int prepare_tpacket_socket(pcap_t *handle);
348 static void pcap_cleanup_linux_mmap(pcap_t *);
349 static int pcap_read_linux_mmap(pcap_t *, int, pcap_handler , u_char *);
350 static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *);
351 static int pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf);
352 static int pcap_getnonblock_mmap(pcap_t *p, char *errbuf);
353 static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
354     const u_char *bytes);
355 #endif
356 
357 /*
358  * Wrap some ioctl calls
359  */
360 #ifdef HAVE_PF_PACKET_SOCKETS
361 static int	iface_get_id(int fd, const char *device, char *ebuf);
362 #endif /* HAVE_PF_PACKET_SOCKETS */
363 static int	iface_get_mtu(int fd, const char *device, char *ebuf);
364 static int 	iface_get_arptype(int fd, const char *device, char *ebuf);
365 #ifdef HAVE_PF_PACKET_SOCKETS
366 static int 	iface_bind(int fd, int ifindex, char *ebuf);
367 #ifdef IW_MODE_MONITOR
368 static int	has_wext(int sock_fd, const char *device, char *ebuf);
369 #endif /* IW_MODE_MONITOR */
370 static int	enter_rfmon_mode(pcap_t *handle, int sock_fd,
371     const char *device);
372 #endif /* HAVE_PF_PACKET_SOCKETS */
373 static int	iface_get_offload(pcap_t *handle);
374 static int 	iface_bind_old(int fd, const char *device, char *ebuf);
375 
376 #ifdef SO_ATTACH_FILTER
377 static int	fix_program(pcap_t *handle, struct sock_fprog *fcode,
378     int is_mapped);
379 static int	fix_offset(struct bpf_insn *p);
380 static int	set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
381 static int	reset_kernel_filter(pcap_t *handle);
382 
383 static struct sock_filter	total_insn
384 	= BPF_STMT(BPF_RET | BPF_K, 0);
385 static struct sock_fprog	total_fcode
386 	= { 1, &total_insn };
387 #endif /* SO_ATTACH_FILTER */
388 
389 pcap_t *
390 pcap_create(const char *device, char *ebuf)
391 {
392 	pcap_t *handle;
393 
394 	/*
395 	 * A null device name is equivalent to the "any" device.
396 	 */
397 	if (device == NULL)
398 		device = "any";
399 
400 #ifdef HAVE_DAG_API
401 	if (strstr(device, "dag")) {
402 		return dag_create(device, ebuf);
403 	}
404 #endif /* HAVE_DAG_API */
405 
406 #ifdef HAVE_SEPTEL_API
407 	if (strstr(device, "septel")) {
408 		return septel_create(device, ebuf);
409 	}
410 #endif /* HAVE_SEPTEL_API */
411 
412 #ifdef HAVE_SNF_API
413         handle = snf_create(device, ebuf);
414         if (strstr(device, "snf") || handle != NULL)
415 		return handle;
416 
417 #endif /* HAVE_SNF_API */
418 
419 #ifdef PCAP_SUPPORT_BT
420 	if (strstr(device, "bluetooth")) {
421 		return bt_create(device, ebuf);
422 	}
423 #endif
424 
425 #if PCAP_SUPPORT_CANUSB
426   if (strstr(device, "canusb")) {
427     return canusb_create(device, ebuf);
428   }
429 #endif
430 
431 #ifdef PCAP_SUPPORT_CAN
432 	if ((strncmp(device, "can", 3) == 0 && isdigit(device[3])) ||
433 	    (strncmp(device, "vcan", 4) == 0 && isdigit(device[4]))) {
434 		return can_create(device, ebuf);
435 	}
436 #endif
437 
438 #ifdef PCAP_SUPPORT_USB
439 	if (strstr(device, "usbmon")) {
440 		return usb_create(device, ebuf);
441 	}
442 #endif
443 
444 #ifdef PCAP_SUPPORT_NETFILTER
445 	if (strncmp(device, "nflog", strlen("nflog")) == 0) {
446 		return nflog_create(device, ebuf);
447 	}
448 #endif
449 
450 	handle = pcap_create_common(device, ebuf);
451 	if (handle == NULL)
452 		return NULL;
453 
454 	handle->activate_op = pcap_activate_linux;
455 	handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
456 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
457 	/*
458 	 * We claim that we support:
459 	 *
460 	 *	software time stamps, with no details about their precision;
461 	 *	hardware time stamps, synced to the host time;
462 	 *	hardware time stamps, not synced to the host time.
463 	 *
464 	 * XXX - we can't ask a device whether it supports
465 	 * hardware time stamps, so we just claim all devices do.
466 	 */
467 	handle->tstamp_type_count = 3;
468 	handle->tstamp_type_list = malloc(3 * sizeof(u_int));
469 	if (handle->tstamp_type_list == NULL) {
470 		free(handle);
471 		return NULL;
472 	}
473 	handle->tstamp_type_list[0] = PCAP_TSTAMP_HOST;
474 	handle->tstamp_type_list[1] = PCAP_TSTAMP_ADAPTER;
475 	handle->tstamp_type_list[2] = PCAP_TSTAMP_ADAPTER_UNSYNCED;
476 #endif
477 
478 	return handle;
479 }
480 
481 #ifdef HAVE_LIBNL
482 /*
483  * If interface {if} is a mac80211 driver, the file
484  * /sys/class/net/{if}/phy80211 is a symlink to
485  * /sys/class/ieee80211/{phydev}, for some {phydev}.
486  *
487  * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
488  * least, has a "wmaster0" device and a "wlan0" device; the
489  * latter is the one with the IP address.  Both show up in
490  * "tcpdump -D" output.  Capturing on the wmaster0 device
491  * captures with 802.11 headers.
492  *
493  * airmon-ng searches through /sys/class/net for devices named
494  * monN, starting with mon0; as soon as one *doesn't* exist,
495  * it chooses that as the monitor device name.  If the "iw"
496  * command exists, it does "iw dev {if} interface add {monif}
497  * type monitor", where {monif} is the monitor device.  It
498  * then (sigh) sleeps .1 second, and then configures the
499  * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
500  * is a file, it writes {mondev}, without a newline, to that file,
501  * and again (sigh) sleeps .1 second, and then iwconfig's that
502  * device into monitor mode and configures it up.  Otherwise,
503  * you can't do monitor mode.
504  *
505  * All these devices are "glued" together by having the
506  * /sys/class/net/{device}/phy80211 links pointing to the same
507  * place, so, given a wmaster, wlan, or mon device, you can
508  * find the other devices by looking for devices with
509  * the same phy80211 link.
510  *
511  * To turn monitor mode off, delete the monitor interface,
512  * either with "iw dev {monif} interface del" or by sending
513  * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
514  *
515  * Note: if you try to create a monitor device named "monN", and
516  * there's already a "monN" device, it fails, as least with
517  * the netlink interface (which is what iw uses), with a return
518  * value of -ENFILE.  (Return values are negative errnos.)  We
519  * could probably use that to find an unused device.
520  *
521  * Yes, you can have multiple monitor devices for a given
522  * physical device.
523 */
524 
525 /*
526  * Is this a mac80211 device?  If so, fill in the physical device path and
527  * return 1; if not, return 0.  On an error, fill in handle->errbuf and
528  * return PCAP_ERROR.
529  */
530 static int
531 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
532     size_t phydev_max_pathlen)
533 {
534 	char *pathstr;
535 	ssize_t bytes_read;
536 
537 	/*
538 	 * Generate the path string for the symlink to the physical device.
539 	 */
540 	if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
541 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
542 		    "%s: Can't generate path name string for /sys/class/net device",
543 		    device);
544 		return PCAP_ERROR;
545 	}
546 	bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
547 	if (bytes_read == -1) {
548 		if (errno == ENOENT || errno == EINVAL) {
549 			/*
550 			 * Doesn't exist, or not a symlink; assume that
551 			 * means it's not a mac80211 device.
552 			 */
553 			free(pathstr);
554 			return 0;
555 		}
556 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
557 		    "%s: Can't readlink %s: %s", device, pathstr,
558 		    strerror(errno));
559 		free(pathstr);
560 		return PCAP_ERROR;
561 	}
562 	free(pathstr);
563 	phydev_path[bytes_read] = '\0';
564 	return 1;
565 }
566 
567 #ifdef HAVE_LIBNL_2_x
568 #define get_nl_errmsg	nl_geterror
569 #else
570 /* libnl 2.x compatibility code */
571 
572 #define nl_sock nl_handle
573 
574 static inline struct nl_handle *
575 nl_socket_alloc(void)
576 {
577 	return nl_handle_alloc();
578 }
579 
580 static inline void
581 nl_socket_free(struct nl_handle *h)
582 {
583 	nl_handle_destroy(h);
584 }
585 
586 #define get_nl_errmsg	strerror
587 
588 static inline int
589 __genl_ctrl_alloc_cache(struct nl_handle *h, struct nl_cache **cache)
590 {
591 	struct nl_cache *tmp = genl_ctrl_alloc_cache(h);
592 	if (!tmp)
593 		return -ENOMEM;
594 	*cache = tmp;
595 	return 0;
596 }
597 #define genl_ctrl_alloc_cache __genl_ctrl_alloc_cache
598 #endif /* !HAVE_LIBNL_2_x */
599 
600 struct nl80211_state {
601 	struct nl_sock *nl_sock;
602 	struct nl_cache *nl_cache;
603 	struct genl_family *nl80211;
604 };
605 
606 static int
607 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
608 {
609 	int err;
610 
611 	state->nl_sock = nl_socket_alloc();
612 	if (!state->nl_sock) {
613 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
614 		    "%s: failed to allocate netlink handle", device);
615 		return PCAP_ERROR;
616 	}
617 
618 	if (genl_connect(state->nl_sock)) {
619 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
620 		    "%s: failed to connect to generic netlink", device);
621 		goto out_handle_destroy;
622 	}
623 
624 	err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
625 	if (err < 0) {
626 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
627 		    "%s: failed to allocate generic netlink cache: %s",
628 		    device, get_nl_errmsg(-err));
629 		goto out_handle_destroy;
630 	}
631 
632 	state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
633 	if (!state->nl80211) {
634 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
635 		    "%s: nl80211 not found", device);
636 		goto out_cache_free;
637 	}
638 
639 	return 0;
640 
641 out_cache_free:
642 	nl_cache_free(state->nl_cache);
643 out_handle_destroy:
644 	nl_socket_free(state->nl_sock);
645 	return PCAP_ERROR;
646 }
647 
648 static void
649 nl80211_cleanup(struct nl80211_state *state)
650 {
651 	genl_family_put(state->nl80211);
652 	nl_cache_free(state->nl_cache);
653 	nl_socket_free(state->nl_sock);
654 }
655 
656 static int
657 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
658     const char *device, const char *mondevice)
659 {
660 	int ifindex;
661 	struct nl_msg *msg;
662 	int err;
663 
664 	ifindex = iface_get_id(sock_fd, device, handle->errbuf);
665 	if (ifindex == -1)
666 		return PCAP_ERROR;
667 
668 	msg = nlmsg_alloc();
669 	if (!msg) {
670 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
671 		    "%s: failed to allocate netlink msg", device);
672 		return PCAP_ERROR;
673 	}
674 
675 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
676 		    0, NL80211_CMD_NEW_INTERFACE, 0);
677 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
678 	NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
679 	NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
680 
681 	err = nl_send_auto_complete(state->nl_sock, msg);
682 	if (err < 0) {
683 #ifdef HAVE_LIBNL_2_x
684 		if (err == -NLE_FAILURE) {
685 #else
686 		if (err == -ENFILE) {
687 #endif
688 			/*
689 			 * Device not available; our caller should just
690 			 * keep trying.  (libnl 2.x maps ENFILE to
691 			 * NLE_FAILURE; it can also map other errors
692 			 * to that, but there's not much we can do
693 			 * about that.)
694 			 */
695 			nlmsg_free(msg);
696 			return 0;
697 		} else {
698 			/*
699 			 * Real failure, not just "that device is not
700 			 * available.
701 			 */
702 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
703 			    "%s: nl_send_auto_complete failed adding %s interface: %s",
704 			    device, mondevice, get_nl_errmsg(-err));
705 			nlmsg_free(msg);
706 			return PCAP_ERROR;
707 		}
708 	}
709 	err = nl_wait_for_ack(state->nl_sock);
710 	if (err < 0) {
711 #ifdef HAVE_LIBNL_2_x
712 		if (err == -NLE_FAILURE) {
713 #else
714 		if (err == -ENFILE) {
715 #endif
716 			/*
717 			 * Device not available; our caller should just
718 			 * keep trying.  (libnl 2.x maps ENFILE to
719 			 * NLE_FAILURE; it can also map other errors
720 			 * to that, but there's not much we can do
721 			 * about that.)
722 			 */
723 			nlmsg_free(msg);
724 			return 0;
725 		} else {
726 			/*
727 			 * Real failure, not just "that device is not
728 			 * available.
729 			 */
730 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
731 			    "%s: nl_wait_for_ack failed adding %s interface: %s",
732 			    device, mondevice, get_nl_errmsg(-err));
733 			nlmsg_free(msg);
734 			return PCAP_ERROR;
735 		}
736 	}
737 
738 	/*
739 	 * Success.
740 	 */
741 	nlmsg_free(msg);
742 	return 1;
743 
744 nla_put_failure:
745 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
746 	    "%s: nl_put failed adding %s interface",
747 	    device, mondevice);
748 	nlmsg_free(msg);
749 	return PCAP_ERROR;
750 }
751 
752 static int
753 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
754     const char *device, const char *mondevice)
755 {
756 	int ifindex;
757 	struct nl_msg *msg;
758 	int err;
759 
760 	ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
761 	if (ifindex == -1)
762 		return PCAP_ERROR;
763 
764 	msg = nlmsg_alloc();
765 	if (!msg) {
766 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
767 		    "%s: failed to allocate netlink msg", device);
768 		return PCAP_ERROR;
769 	}
770 
771 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
772 		    0, NL80211_CMD_DEL_INTERFACE, 0);
773 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
774 
775 	err = nl_send_auto_complete(state->nl_sock, msg);
776 	if (err < 0) {
777 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
778 		    "%s: nl_send_auto_complete failed deleting %s interface: %s",
779 		    device, mondevice, get_nl_errmsg(-err));
780 		nlmsg_free(msg);
781 		return PCAP_ERROR;
782 	}
783 	err = nl_wait_for_ack(state->nl_sock);
784 	if (err < 0) {
785 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
786 		    "%s: nl_wait_for_ack failed adding %s interface: %s",
787 		    device, mondevice, get_nl_errmsg(-err));
788 		nlmsg_free(msg);
789 		return PCAP_ERROR;
790 	}
791 
792 	/*
793 	 * Success.
794 	 */
795 	nlmsg_free(msg);
796 	return 1;
797 
798 nla_put_failure:
799 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
800 	    "%s: nl_put failed deleting %s interface",
801 	    device, mondevice);
802 	nlmsg_free(msg);
803 	return PCAP_ERROR;
804 }
805 
806 static int
807 enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
808 {
809 	int ret;
810 	char phydev_path[PATH_MAX+1];
811 	struct nl80211_state nlstate;
812 	struct ifreq ifr;
813 	u_int n;
814 
815 	/*
816 	 * Is this a mac80211 device?
817 	 */
818 	ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
819 	if (ret < 0)
820 		return ret;	/* error */
821 	if (ret == 0)
822 		return 0;	/* no error, but not mac80211 device */
823 
824 	/*
825 	 * XXX - is this already a monN device?
826 	 * If so, we're done.
827 	 * Is that determined by old Wireless Extensions ioctls?
828 	 */
829 
830 	/*
831 	 * OK, it's apparently a mac80211 device.
832 	 * Try to find an unused monN device for it.
833 	 */
834 	ret = nl80211_init(handle, &nlstate, device);
835 	if (ret != 0)
836 		return ret;
837 	for (n = 0; n < UINT_MAX; n++) {
838 		/*
839 		 * Try mon{n}.
840 		 */
841 		char mondevice[3+10+1];	/* mon{UINT_MAX}\0 */
842 
843 		snprintf(mondevice, sizeof mondevice, "mon%u", n);
844 		ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
845 		if (ret == 1) {
846 			handle->md.mondevice = strdup(mondevice);
847 			goto added;
848 		}
849 		if (ret < 0) {
850 			/*
851 			 * Hard failure.  Just return ret; handle->errbuf
852 			 * has already been set.
853 			 */
854 			nl80211_cleanup(&nlstate);
855 			return ret;
856 		}
857 	}
858 
859 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
860 	    "%s: No free monN interfaces", device);
861 	nl80211_cleanup(&nlstate);
862 	return PCAP_ERROR;
863 
864 added:
865 
866 #if 0
867 	/*
868 	 * Sleep for .1 seconds.
869 	 */
870 	delay.tv_sec = 0;
871 	delay.tv_nsec = 500000000;
872 	nanosleep(&delay, NULL);
873 #endif
874 
875 	/*
876 	 * If we haven't already done so, arrange to have
877 	 * "pcap_close_all()" called when we exit.
878 	 */
879 	if (!pcap_do_addexit(handle)) {
880 		/*
881 		 * "atexit()" failed; don't put the interface
882 		 * in rfmon mode, just give up.
883 		 */
884 		return PCAP_ERROR_RFMON_NOTSUP;
885 	}
886 
887 	/*
888 	 * Now configure the monitor interface up.
889 	 */
890 	memset(&ifr, 0, sizeof(ifr));
891 	strncpy(ifr.ifr_name, handle->md.mondevice, sizeof(ifr.ifr_name));
892 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
893 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
894 		    "%s: Can't get flags for %s: %s", device,
895 		    handle->md.mondevice, strerror(errno));
896 		del_mon_if(handle, sock_fd, &nlstate, device,
897 		    handle->md.mondevice);
898 		nl80211_cleanup(&nlstate);
899 		return PCAP_ERROR;
900 	}
901 	ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
902 	if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
903 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
904 		    "%s: Can't set flags for %s: %s", device,
905 		    handle->md.mondevice, strerror(errno));
906 		del_mon_if(handle, sock_fd, &nlstate, device,
907 		    handle->md.mondevice);
908 		nl80211_cleanup(&nlstate);
909 		return PCAP_ERROR;
910 	}
911 
912 	/*
913 	 * Success.  Clean up the libnl state.
914 	 */
915 	nl80211_cleanup(&nlstate);
916 
917 	/*
918 	 * Note that we have to delete the monitor device when we close
919 	 * the handle.
920 	 */
921 	handle->md.must_do_on_close |= MUST_DELETE_MONIF;
922 
923 	/*
924 	 * Add this to the list of pcaps to close when we exit.
925 	 */
926 	pcap_add_to_pcaps_to_close(handle);
927 
928 	return 1;
929 }
930 #endif /* HAVE_LIBNL */
931 
932 static int
933 pcap_can_set_rfmon_linux(pcap_t *handle)
934 {
935 #ifdef HAVE_LIBNL
936 	char phydev_path[PATH_MAX+1];
937 	int ret;
938 #endif
939 #ifdef IW_MODE_MONITOR
940 	int sock_fd;
941 	struct iwreq ireq;
942 #endif
943 
944 	if (strcmp(handle->opt.source, "any") == 0) {
945 		/*
946 		 * Monitor mode makes no sense on the "any" device.
947 		 */
948 		return 0;
949 	}
950 
951 #ifdef HAVE_LIBNL
952 	/*
953 	 * Bleah.  There doesn't seem to be a way to ask a mac80211
954 	 * device, through libnl, whether it supports monitor mode;
955 	 * we'll just check whether the device appears to be a
956 	 * mac80211 device and, if so, assume the device supports
957 	 * monitor mode.
958 	 *
959 	 * wmaster devices don't appear to support the Wireless
960 	 * Extensions, but we can create a mon device for a
961 	 * wmaster device, so we don't bother checking whether
962 	 * a mac80211 device supports the Wireless Extensions.
963 	 */
964 	ret = get_mac80211_phydev(handle, handle->opt.source, phydev_path,
965 	    PATH_MAX);
966 	if (ret < 0)
967 		return ret;	/* error */
968 	if (ret == 1)
969 		return 1;	/* mac80211 device */
970 #endif
971 
972 #ifdef IW_MODE_MONITOR
973 	/*
974 	 * Bleah.  There doesn't appear to be an ioctl to use to ask
975 	 * whether a device supports monitor mode; we'll just do
976 	 * SIOCGIWMODE and, if it succeeds, assume the device supports
977 	 * monitor mode.
978 	 *
979 	 * Open a socket on which to attempt to get the mode.
980 	 * (We assume that if we have Wireless Extensions support
981 	 * we also have PF_PACKET support.)
982 	 */
983 	sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
984 	if (sock_fd == -1) {
985 		(void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
986 		    "socket: %s", pcap_strerror(errno));
987 		return PCAP_ERROR;
988 	}
989 
990 	/*
991 	 * Attempt to get the current mode.
992 	 */
993 	strncpy(ireq.ifr_ifrn.ifrn_name, handle->opt.source,
994 	    sizeof ireq.ifr_ifrn.ifrn_name);
995 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
996 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
997 		/*
998 		 * Well, we got the mode; assume we can set it.
999 		 */
1000 		close(sock_fd);
1001 		return 1;
1002 	}
1003 	if (errno == ENODEV) {
1004 		/* The device doesn't even exist. */
1005 		(void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1006 		    "SIOCGIWMODE failed: %s", pcap_strerror(errno));
1007 		close(sock_fd);
1008 		return PCAP_ERROR_NO_SUCH_DEVICE;
1009 	}
1010 	close(sock_fd);
1011 #endif
1012 	return 0;
1013 }
1014 
1015 /*
1016  * Grabs the number of dropped packets by the interface from /proc/net/dev.
1017  *
1018  * XXX - what about /sys/class/net/{interface name}/rx_*?  There are
1019  * individual devices giving, in ASCII, various rx_ and tx_ statistics.
1020  *
1021  * Or can we get them in binary form from netlink?
1022  */
1023 static long int
1024 linux_if_drops(const char * if_name)
1025 {
1026 	char buffer[512];
1027 	char * bufptr;
1028 	FILE * file;
1029 	int field_to_convert = 3, if_name_sz = strlen(if_name);
1030 	long int dropped_pkts = 0;
1031 
1032 	file = fopen("/proc/net/dev", "r");
1033 	if (!file)
1034 		return 0;
1035 
1036 	while (!dropped_pkts && fgets( buffer, sizeof(buffer), file ))
1037 	{
1038 		/* 	search for 'bytes' -- if its in there, then
1039 			that means we need to grab the fourth field. otherwise
1040 			grab the third field. */
1041 		if (field_to_convert != 4 && strstr(buffer, "bytes"))
1042 		{
1043 			field_to_convert = 4;
1044 			continue;
1045 		}
1046 
1047 		/* find iface and make sure it actually matches -- space before the name and : after it */
1048 		if ((bufptr = strstr(buffer, if_name)) &&
1049 			(bufptr == buffer || *(bufptr-1) == ' ') &&
1050 			*(bufptr + if_name_sz) == ':')
1051 		{
1052 			bufptr = bufptr + if_name_sz + 1;
1053 
1054 			/* grab the nth field from it */
1055 			while( --field_to_convert && *bufptr != '\0')
1056 			{
1057 				while (*bufptr != '\0' && *(bufptr++) == ' ');
1058 				while (*bufptr != '\0' && *(bufptr++) != ' ');
1059 			}
1060 
1061 			/* get rid of any final spaces */
1062 			while (*bufptr != '\0' && *bufptr == ' ') bufptr++;
1063 
1064 			if (*bufptr != '\0')
1065 				dropped_pkts = strtol(bufptr, NULL, 10);
1066 
1067 			break;
1068 		}
1069 	}
1070 
1071 	fclose(file);
1072 	return dropped_pkts;
1073 }
1074 
1075 
1076 /*
1077  * With older kernels promiscuous mode is kind of interesting because we
1078  * have to reset the interface before exiting. The problem can't really
1079  * be solved without some daemon taking care of managing usage counts.
1080  * If we put the interface into promiscuous mode, we set a flag indicating
1081  * that we must take it out of that mode when the interface is closed,
1082  * and, when closing the interface, if that flag is set we take it out
1083  * of promiscuous mode.
1084  *
1085  * Even with newer kernels, we have the same issue with rfmon mode.
1086  */
1087 
1088 static void	pcap_cleanup_linux( pcap_t *handle )
1089 {
1090 	struct ifreq	ifr;
1091 #ifdef HAVE_LIBNL
1092 	struct nl80211_state nlstate;
1093 	int ret;
1094 #endif /* HAVE_LIBNL */
1095 #ifdef IW_MODE_MONITOR
1096 	int oldflags;
1097 	struct iwreq ireq;
1098 #endif /* IW_MODE_MONITOR */
1099 
1100 	if (handle->md.must_do_on_close != 0) {
1101 		/*
1102 		 * There's something we have to do when closing this
1103 		 * pcap_t.
1104 		 */
1105 		if (handle->md.must_do_on_close & MUST_CLEAR_PROMISC) {
1106 			/*
1107 			 * We put the interface into promiscuous mode;
1108 			 * take it out of promiscuous mode.
1109 			 *
1110 			 * XXX - if somebody else wants it in promiscuous
1111 			 * mode, this code cannot know that, so it'll take
1112 			 * it out of promiscuous mode.  That's not fixable
1113 			 * in 2.0[.x] kernels.
1114 			 */
1115 			memset(&ifr, 0, sizeof(ifr));
1116 			strncpy(ifr.ifr_name, handle->md.device,
1117 			    sizeof(ifr.ifr_name));
1118 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
1119 				fprintf(stderr,
1120 				    "Can't restore interface %s flags (SIOCGIFFLAGS failed: %s).\n"
1121 				    "Please adjust manually.\n"
1122 				    "Hint: This can't happen with Linux >= 2.2.0.\n",
1123 				    handle->md.device, strerror(errno));
1124 			} else {
1125 				if (ifr.ifr_flags & IFF_PROMISC) {
1126 					/*
1127 					 * Promiscuous mode is currently on;
1128 					 * turn it off.
1129 					 */
1130 					ifr.ifr_flags &= ~IFF_PROMISC;
1131 					if (ioctl(handle->fd, SIOCSIFFLAGS,
1132 					    &ifr) == -1) {
1133 						fprintf(stderr,
1134 						    "Can't restore interface %s flags (SIOCSIFFLAGS failed: %s).\n"
1135 						    "Please adjust manually.\n"
1136 						    "Hint: This can't happen with Linux >= 2.2.0.\n",
1137 						    handle->md.device,
1138 						    strerror(errno));
1139 					}
1140 				}
1141 			}
1142 		}
1143 
1144 #ifdef HAVE_LIBNL
1145 		if (handle->md.must_do_on_close & MUST_DELETE_MONIF) {
1146 			ret = nl80211_init(handle, &nlstate, handle->md.device);
1147 			if (ret >= 0) {
1148 				ret = del_mon_if(handle, handle->fd, &nlstate,
1149 				    handle->md.device, handle->md.mondevice);
1150 				nl80211_cleanup(&nlstate);
1151 			}
1152 			if (ret < 0) {
1153 				fprintf(stderr,
1154 				    "Can't delete monitor interface %s (%s).\n"
1155 				    "Please delete manually.\n",
1156 				    handle->md.mondevice, handle->errbuf);
1157 			}
1158 		}
1159 #endif /* HAVE_LIBNL */
1160 
1161 #ifdef IW_MODE_MONITOR
1162 		if (handle->md.must_do_on_close & MUST_CLEAR_RFMON) {
1163 			/*
1164 			 * We put the interface into rfmon mode;
1165 			 * take it out of rfmon mode.
1166 			 *
1167 			 * XXX - if somebody else wants it in rfmon
1168 			 * mode, this code cannot know that, so it'll take
1169 			 * it out of rfmon mode.
1170 			 */
1171 
1172 			/*
1173 			 * First, take the interface down if it's up;
1174 			 * otherwise, we might get EBUSY.
1175 			 * If we get errors, just drive on and print
1176 			 * a warning if we can't restore the mode.
1177 			 */
1178 			oldflags = 0;
1179 			memset(&ifr, 0, sizeof(ifr));
1180 			strncpy(ifr.ifr_name, handle->md.device,
1181 			    sizeof(ifr.ifr_name));
1182 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) != -1) {
1183 				if (ifr.ifr_flags & IFF_UP) {
1184 					oldflags = ifr.ifr_flags;
1185 					ifr.ifr_flags &= ~IFF_UP;
1186 					if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1)
1187 						oldflags = 0;	/* didn't set, don't restore */
1188 				}
1189 			}
1190 
1191 			/*
1192 			 * Now restore the mode.
1193 			 */
1194 			strncpy(ireq.ifr_ifrn.ifrn_name, handle->md.device,
1195 			    sizeof ireq.ifr_ifrn.ifrn_name);
1196 			ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1]
1197 			    = 0;
1198 			ireq.u.mode = handle->md.oldmode;
1199 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
1200 				/*
1201 				 * Scientist, you've failed.
1202 				 */
1203 				fprintf(stderr,
1204 				    "Can't restore interface %s wireless mode (SIOCSIWMODE failed: %s).\n"
1205 				    "Please adjust manually.\n",
1206 				    handle->md.device, strerror(errno));
1207 			}
1208 
1209 			/*
1210 			 * Now bring the interface back up if we brought
1211 			 * it down.
1212 			 */
1213 			if (oldflags != 0) {
1214 				ifr.ifr_flags = oldflags;
1215 				if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
1216 					fprintf(stderr,
1217 					    "Can't bring interface %s back up (SIOCSIFFLAGS failed: %s).\n"
1218 					    "Please adjust manually.\n",
1219 					    handle->md.device, strerror(errno));
1220 				}
1221 			}
1222 		}
1223 #endif /* IW_MODE_MONITOR */
1224 
1225 		/*
1226 		 * Take this pcap out of the list of pcaps for which we
1227 		 * have to take the interface out of some mode.
1228 		 */
1229 		pcap_remove_from_pcaps_to_close(handle);
1230 	}
1231 
1232 	if (handle->md.mondevice != NULL) {
1233 		free(handle->md.mondevice);
1234 		handle->md.mondevice = NULL;
1235 	}
1236 	if (handle->md.device != NULL) {
1237 		free(handle->md.device);
1238 		handle->md.device = NULL;
1239 	}
1240 	pcap_cleanup_live_common(handle);
1241 }
1242 
1243 /*
1244  *  Get a handle for a live capture from the given device. You can
1245  *  pass NULL as device to get all packages (without link level
1246  *  information of course). If you pass 1 as promisc the interface
1247  *  will be set to promiscous mode (XXX: I think this usage should
1248  *  be deprecated and functions be added to select that later allow
1249  *  modification of that values -- Torsten).
1250  */
1251 static int
1252 pcap_activate_linux(pcap_t *handle)
1253 {
1254 	const char	*device;
1255 	int		status = 0;
1256 
1257 	device = handle->opt.source;
1258 
1259 	handle->inject_op = pcap_inject_linux;
1260 	handle->setfilter_op = pcap_setfilter_linux;
1261 	handle->setdirection_op = pcap_setdirection_linux;
1262 	handle->set_datalink_op = NULL;	/* can't change data link type */
1263 	handle->getnonblock_op = pcap_getnonblock_fd;
1264 	handle->setnonblock_op = pcap_setnonblock_fd;
1265 	handle->cleanup_op = pcap_cleanup_linux;
1266 	handle->read_op = pcap_read_linux;
1267 	handle->stats_op = pcap_stats_linux;
1268 
1269 	/*
1270 	 * The "any" device is a special device which causes us not
1271 	 * to bind to a particular device and thus to look at all
1272 	 * devices.
1273 	 */
1274 	if (strcmp(device, "any") == 0) {
1275 		if (handle->opt.promisc) {
1276 			handle->opt.promisc = 0;
1277 			/* Just a warning. */
1278 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1279 			    "Promiscuous mode not supported on the \"any\" device");
1280 			status = PCAP_WARNING_PROMISC_NOTSUP;
1281 		}
1282 	}
1283 
1284 	handle->md.device	= strdup(device);
1285 	if (handle->md.device == NULL) {
1286 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
1287 			 pcap_strerror(errno) );
1288 		return PCAP_ERROR;
1289 	}
1290 
1291 	/*
1292 	 * If we're in promiscuous mode, then we probably want
1293 	 * to see when the interface drops packets too, so get an
1294 	 * initial count from /proc/net/dev
1295 	 */
1296 	if (handle->opt.promisc)
1297 		handle->md.proc_dropped = linux_if_drops(handle->md.device);
1298 
1299 	/*
1300 	 * Current Linux kernels use the protocol family PF_PACKET to
1301 	 * allow direct access to all packets on the network while
1302 	 * older kernels had a special socket type SOCK_PACKET to
1303 	 * implement this feature.
1304 	 * While this old implementation is kind of obsolete we need
1305 	 * to be compatible with older kernels for a while so we are
1306 	 * trying both methods with the newer method preferred.
1307 	 */
1308 	status = activate_new(handle);
1309 	if (status < 0) {
1310 		/*
1311 		 * Fatal error with the new way; just fail.
1312 		 * status has the error return; if it's PCAP_ERROR,
1313 		 * handle->errbuf has been set appropriately.
1314 		 */
1315 		goto fail;
1316 	}
1317 	if (status == 1) {
1318 		/*
1319 		 * Success.
1320 		 * Try to use memory-mapped access.
1321 		 */
1322 		switch (activate_mmap(handle, &status)) {
1323 
1324 		case 1:
1325 			/*
1326 			 * We succeeded.  status has been
1327 			 * set to the status to return,
1328 			 * which might be 0, or might be
1329 			 * a PCAP_WARNING_ value.
1330 			 */
1331 			return status;
1332 
1333 		case 0:
1334 			/*
1335 			 * Kernel doesn't support it - just continue
1336 			 * with non-memory-mapped access.
1337 			 */
1338 			break;
1339 
1340 		case -1:
1341 			/*
1342 			 * We failed to set up to use it, or the kernel
1343 			 * supports it, but we failed to enable it.
1344 			 * status has been set to the error status to
1345 			 * return and, if it's PCAP_ERROR, handle->errbuf
1346 			 * contains the error message.
1347 			 */
1348 			goto fail;
1349 		}
1350 	}
1351 	else if (status == 0) {
1352 		/* Non-fatal error; try old way */
1353 		if ((status = activate_old(handle)) != 1) {
1354 			/*
1355 			 * Both methods to open the packet socket failed.
1356 			 * Tidy up and report our failure (handle->errbuf
1357 			 * is expected to be set by the functions above).
1358 			 */
1359 			goto fail;
1360 		}
1361 	}
1362 
1363 	/*
1364 	 * We set up the socket, but not with memory-mapped access.
1365 	 */
1366 	status = 0;
1367 	if (handle->opt.buffer_size != 0) {
1368 		/*
1369 		 * Set the socket buffer size to the specified value.
1370 		 */
1371 		if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF,
1372 		    &handle->opt.buffer_size,
1373 		    sizeof(handle->opt.buffer_size)) == -1) {
1374 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1375 				 "SO_RCVBUF: %s", pcap_strerror(errno));
1376 			status = PCAP_ERROR;
1377 			goto fail;
1378 		}
1379 	}
1380 
1381 	/* Allocate the buffer */
1382 
1383 	handle->buffer	 = malloc(handle->bufsize + handle->offset);
1384 	if (!handle->buffer) {
1385 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1386 			 "malloc: %s", pcap_strerror(errno));
1387 		status = PCAP_ERROR;
1388 		goto fail;
1389 	}
1390 
1391 	/*
1392 	 * "handle->fd" is a socket, so "select()" and "poll()"
1393 	 * should work on it.
1394 	 */
1395 	handle->selectable_fd = handle->fd;
1396 
1397 	return status;
1398 
1399 fail:
1400 	pcap_cleanup_linux(handle);
1401 	return status;
1402 }
1403 
1404 /*
1405  *  Read at most max_packets from the capture stream and call the callback
1406  *  for each of them. Returns the number of packets handled or -1 if an
1407  *  error occured.
1408  */
1409 static int
1410 pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
1411 {
1412 	/*
1413 	 * Currently, on Linux only one packet is delivered per read,
1414 	 * so we don't loop.
1415 	 */
1416 	return pcap_read_packet(handle, callback, user);
1417 }
1418 
1419 /*
1420  *  Read a packet from the socket calling the handler provided by
1421  *  the user. Returns the number of packets received or -1 if an
1422  *  error occured.
1423  */
1424 static int
1425 pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
1426 {
1427 	u_char			*bp;
1428 	int			offset;
1429 #ifdef HAVE_PF_PACKET_SOCKETS
1430 	struct sockaddr_ll	from;
1431 	struct sll_header	*hdrp;
1432 #else
1433 	struct sockaddr		from;
1434 #endif
1435 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1436 	struct iovec		iov;
1437 	struct msghdr		msg;
1438 	struct cmsghdr		*cmsg;
1439 	union {
1440 		struct cmsghdr	cmsg;
1441 		char		buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
1442 	} cmsg_buf;
1443 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1444 	socklen_t		fromlen;
1445 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1446 	int			packet_len, caplen;
1447 	struct pcap_pkthdr	pcap_header;
1448 
1449 #ifdef HAVE_PF_PACKET_SOCKETS
1450 	/*
1451 	 * If this is a cooked device, leave extra room for a
1452 	 * fake packet header.
1453 	 */
1454 	if (handle->md.cooked)
1455 		offset = SLL_HDR_LEN;
1456 	else
1457 		offset = 0;
1458 #else
1459 	/*
1460 	 * This system doesn't have PF_PACKET sockets, so it doesn't
1461 	 * support cooked devices.
1462 	 */
1463 	offset = 0;
1464 #endif
1465 
1466 	/*
1467 	 * Receive a single packet from the kernel.
1468 	 * We ignore EINTR, as that might just be due to a signal
1469 	 * being delivered - if the signal should interrupt the
1470 	 * loop, the signal handler should call pcap_breakloop()
1471 	 * to set handle->break_loop (we ignore it on other
1472 	 * platforms as well).
1473 	 * We also ignore ENETDOWN, so that we can continue to
1474 	 * capture traffic if the interface goes down and comes
1475 	 * back up again; comments in the kernel indicate that
1476 	 * we'll just block waiting for packets if we try to
1477 	 * receive from a socket that delivered ENETDOWN, and,
1478 	 * if we're using a memory-mapped buffer, we won't even
1479 	 * get notified of "network down" events.
1480 	 */
1481 	bp = handle->buffer + handle->offset;
1482 
1483 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1484 	msg.msg_name		= &from;
1485 	msg.msg_namelen		= sizeof(from);
1486 	msg.msg_iov		= &iov;
1487 	msg.msg_iovlen		= 1;
1488 	msg.msg_control		= &cmsg_buf;
1489 	msg.msg_controllen	= sizeof(cmsg_buf);
1490 	msg.msg_flags		= 0;
1491 
1492 	iov.iov_len		= handle->bufsize - offset;
1493 	iov.iov_base		= bp + offset;
1494 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1495 
1496 	do {
1497 		/*
1498 		 * Has "pcap_breakloop()" been called?
1499 		 */
1500 		if (handle->break_loop) {
1501 			/*
1502 			 * Yes - clear the flag that indicates that it has,
1503 			 * and return PCAP_ERROR_BREAK as an indication that
1504 			 * we were told to break out of the loop.
1505 			 */
1506 			handle->break_loop = 0;
1507 			return PCAP_ERROR_BREAK;
1508 		}
1509 
1510 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1511 		packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC);
1512 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1513 		fromlen = sizeof(from);
1514 		packet_len = recvfrom(
1515 			handle->fd, bp + offset,
1516 			handle->bufsize - offset, MSG_TRUNC,
1517 			(struct sockaddr *) &from, &fromlen);
1518 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1519 	} while (packet_len == -1 && errno == EINTR);
1520 
1521 	/* Check if an error occured */
1522 
1523 	if (packet_len == -1) {
1524 		switch (errno) {
1525 
1526 		case EAGAIN:
1527 			return 0;	/* no packet there */
1528 
1529 		case ENETDOWN:
1530 			/*
1531 			 * The device on which we're capturing went away.
1532 			 *
1533 			 * XXX - we should really return
1534 			 * PCAP_ERROR_IFACE_NOT_UP, but pcap_dispatch()
1535 			 * etc. aren't defined to return that.
1536 			 */
1537 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1538 				"The interface went down");
1539 			return PCAP_ERROR;
1540 
1541 		default:
1542 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1543 				 "recvfrom: %s", pcap_strerror(errno));
1544 			return PCAP_ERROR;
1545 		}
1546 	}
1547 
1548 #ifdef HAVE_PF_PACKET_SOCKETS
1549 	if (!handle->md.sock_packet) {
1550 		/*
1551 		 * Unfortunately, there is a window between socket() and
1552 		 * bind() where the kernel may queue packets from any
1553 		 * interface.  If we're bound to a particular interface,
1554 		 * discard packets not from that interface.
1555 		 *
1556 		 * (If socket filters are supported, we could do the
1557 		 * same thing we do when changing the filter; however,
1558 		 * that won't handle packet sockets without socket
1559 		 * filter support, and it's a bit more complicated.
1560 		 * It would save some instructions per packet, however.)
1561 		 */
1562 		if (handle->md.ifindex != -1 &&
1563 		    from.sll_ifindex != handle->md.ifindex)
1564 			return 0;
1565 
1566 		/*
1567 		 * Do checks based on packet direction.
1568 		 * We can only do this if we're using PF_PACKET; the
1569 		 * address returned for SOCK_PACKET is a "sockaddr_pkt"
1570 		 * which lacks the relevant packet type information.
1571 		 */
1572 		if (from.sll_pkttype == PACKET_OUTGOING) {
1573 			/*
1574 			 * Outgoing packet.
1575 			 * If this is from the loopback device, reject it;
1576 			 * we'll see the packet as an incoming packet as well,
1577 			 * and we don't want to see it twice.
1578 			 */
1579 			if (from.sll_ifindex == handle->md.lo_ifindex)
1580 				return 0;
1581 
1582 			/*
1583 			 * If the user only wants incoming packets, reject it.
1584 			 */
1585 			if (handle->direction == PCAP_D_IN)
1586 				return 0;
1587 		} else {
1588 			/*
1589 			 * Incoming packet.
1590 			 * If the user only wants outgoing packets, reject it.
1591 			 */
1592 			if (handle->direction == PCAP_D_OUT)
1593 				return 0;
1594 		}
1595 	}
1596 #endif
1597 
1598 #ifdef HAVE_PF_PACKET_SOCKETS
1599 	/*
1600 	 * If this is a cooked device, fill in the fake packet header.
1601 	 */
1602 	if (handle->md.cooked) {
1603 		/*
1604 		 * Add the length of the fake header to the length
1605 		 * of packet data we read.
1606 		 */
1607 		packet_len += SLL_HDR_LEN;
1608 
1609 		hdrp = (struct sll_header *)bp;
1610 		hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype);
1611 		hdrp->sll_hatype = htons(from.sll_hatype);
1612 		hdrp->sll_halen = htons(from.sll_halen);
1613 		memcpy(hdrp->sll_addr, from.sll_addr,
1614 		    (from.sll_halen > SLL_ADDRLEN) ?
1615 		      SLL_ADDRLEN :
1616 		      from.sll_halen);
1617 		hdrp->sll_protocol = from.sll_protocol;
1618 	}
1619 
1620 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1621 	for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
1622 		struct tpacket_auxdata *aux;
1623 		unsigned int len;
1624 		struct vlan_tag *tag;
1625 
1626 		if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
1627 		    cmsg->cmsg_level != SOL_PACKET ||
1628 		    cmsg->cmsg_type != PACKET_AUXDATA)
1629 			continue;
1630 
1631 		aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
1632 		if (aux->tp_vlan_tci == 0)
1633 			continue;
1634 
1635 		len = packet_len > iov.iov_len ? iov.iov_len : packet_len;
1636 		if (len < 2 * ETH_ALEN)
1637 			break;
1638 
1639 		bp -= VLAN_TAG_LEN;
1640 		memmove(bp, bp + VLAN_TAG_LEN, 2 * ETH_ALEN);
1641 
1642 		tag = (struct vlan_tag *)(bp + 2 * ETH_ALEN);
1643 		tag->vlan_tpid = htons(ETH_P_8021Q);
1644 		tag->vlan_tci = htons(aux->tp_vlan_tci);
1645 
1646 		packet_len += VLAN_TAG_LEN;
1647 	}
1648 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1649 #endif /* HAVE_PF_PACKET_SOCKETS */
1650 
1651 	/*
1652 	 * XXX: According to the kernel source we should get the real
1653 	 * packet len if calling recvfrom with MSG_TRUNC set. It does
1654 	 * not seem to work here :(, but it is supported by this code
1655 	 * anyway.
1656 	 * To be honest the code RELIES on that feature so this is really
1657 	 * broken with 2.2.x kernels.
1658 	 * I spend a day to figure out what's going on and I found out
1659 	 * that the following is happening:
1660 	 *
1661 	 * The packet comes from a random interface and the packet_rcv
1662 	 * hook is called with a clone of the packet. That code inserts
1663 	 * the packet into the receive queue of the packet socket.
1664 	 * If a filter is attached to that socket that filter is run
1665 	 * first - and there lies the problem. The default filter always
1666 	 * cuts the packet at the snaplen:
1667 	 *
1668 	 * # tcpdump -d
1669 	 * (000) ret      #68
1670 	 *
1671 	 * So the packet filter cuts down the packet. The recvfrom call
1672 	 * says "hey, it's only 68 bytes, it fits into the buffer" with
1673 	 * the result that we don't get the real packet length. This
1674 	 * is valid at least until kernel 2.2.17pre6.
1675 	 *
1676 	 * We currently handle this by making a copy of the filter
1677 	 * program, fixing all "ret" instructions with non-zero
1678 	 * operands to have an operand of 65535 so that the filter
1679 	 * doesn't truncate the packet, and supplying that modified
1680 	 * filter to the kernel.
1681 	 */
1682 
1683 	caplen = packet_len;
1684 	if (caplen > handle->snapshot)
1685 		caplen = handle->snapshot;
1686 
1687 	/* Run the packet filter if not using kernel filter */
1688 	if (!handle->md.use_bpf && handle->fcode.bf_insns) {
1689 		if (bpf_filter(handle->fcode.bf_insns, bp,
1690 		                packet_len, caplen) == 0)
1691 		{
1692 			/* rejected by filter */
1693 			return 0;
1694 		}
1695 	}
1696 
1697 	/* Fill in our own header data */
1698 
1699 	if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
1700 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1701 			 "SIOCGSTAMP: %s", pcap_strerror(errno));
1702 		return PCAP_ERROR;
1703 	}
1704 	pcap_header.caplen	= caplen;
1705 	pcap_header.len		= packet_len;
1706 
1707 	/*
1708 	 * Count the packet.
1709 	 *
1710 	 * Arguably, we should count them before we check the filter,
1711 	 * as on many other platforms "ps_recv" counts packets
1712 	 * handed to the filter rather than packets that passed
1713 	 * the filter, but if filtering is done in the kernel, we
1714 	 * can't get a count of packets that passed the filter,
1715 	 * and that would mean the meaning of "ps_recv" wouldn't
1716 	 * be the same on all Linux systems.
1717 	 *
1718 	 * XXX - it's not the same on all systems in any case;
1719 	 * ideally, we should have a "get the statistics" call
1720 	 * that supplies more counts and indicates which of them
1721 	 * it supplies, so that we supply a count of packets
1722 	 * handed to the filter only on platforms where that
1723 	 * information is available.
1724 	 *
1725 	 * We count them here even if we can get the packet count
1726 	 * from the kernel, as we can only determine at run time
1727 	 * whether we'll be able to get it from the kernel (if
1728 	 * HAVE_TPACKET_STATS isn't defined, we can't get it from
1729 	 * the kernel, but if it is defined, the library might
1730 	 * have been built with a 2.4 or later kernel, but we
1731 	 * might be running on a 2.2[.x] kernel without Alexey
1732 	 * Kuznetzov's turbopacket patches, and thus the kernel
1733 	 * might not be able to supply those statistics).  We
1734 	 * could, I guess, try, when opening the socket, to get
1735 	 * the statistics, and if we can not increment the count
1736 	 * here, but it's not clear that always incrementing
1737 	 * the count is more expensive than always testing a flag
1738 	 * in memory.
1739 	 *
1740 	 * We keep the count in "md.packets_read", and use that for
1741 	 * "ps_recv" if we can't get the statistics from the kernel.
1742 	 * We do that because, if we *can* get the statistics from
1743 	 * the kernel, we use "md.stat.ps_recv" and "md.stat.ps_drop"
1744 	 * as running counts, as reading the statistics from the
1745 	 * kernel resets the kernel statistics, and if we directly
1746 	 * increment "md.stat.ps_recv" here, that means it will
1747 	 * count packets *twice* on systems where we can get kernel
1748 	 * statistics - once here, and once in pcap_stats_linux().
1749 	 */
1750 	handle->md.packets_read++;
1751 
1752 	/* Call the user supplied callback function */
1753 	callback(userdata, &pcap_header, bp);
1754 
1755 	return 1;
1756 }
1757 
1758 static int
1759 pcap_inject_linux(pcap_t *handle, const void *buf, size_t size)
1760 {
1761 	int ret;
1762 
1763 #ifdef HAVE_PF_PACKET_SOCKETS
1764 	if (!handle->md.sock_packet) {
1765 		/* PF_PACKET socket */
1766 		if (handle->md.ifindex == -1) {
1767 			/*
1768 			 * We don't support sending on the "any" device.
1769 			 */
1770 			strlcpy(handle->errbuf,
1771 			    "Sending packets isn't supported on the \"any\" device",
1772 			    PCAP_ERRBUF_SIZE);
1773 			return (-1);
1774 		}
1775 
1776 		if (handle->md.cooked) {
1777 			/*
1778 			 * We don't support sending on the "any" device.
1779 			 *
1780 			 * XXX - how do you send on a bound cooked-mode
1781 			 * socket?
1782 			 * Is a "sendto()" required there?
1783 			 */
1784 			strlcpy(handle->errbuf,
1785 			    "Sending packets isn't supported in cooked mode",
1786 			    PCAP_ERRBUF_SIZE);
1787 			return (-1);
1788 		}
1789 	}
1790 #endif
1791 
1792 	ret = send(handle->fd, buf, size, 0);
1793 	if (ret == -1) {
1794 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
1795 		    pcap_strerror(errno));
1796 		return (-1);
1797 	}
1798 	return (ret);
1799 }
1800 
1801 /*
1802  *  Get the statistics for the given packet capture handle.
1803  *  Reports the number of dropped packets iff the kernel supports
1804  *  the PACKET_STATISTICS "getsockopt()" argument (2.4 and later
1805  *  kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket
1806  *  patches); otherwise, that information isn't available, and we lie
1807  *  and report 0 as the count of dropped packets.
1808  */
1809 static int
1810 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
1811 {
1812 #ifdef HAVE_TPACKET_STATS
1813 	struct tpacket_stats kstats;
1814 	socklen_t len = sizeof (struct tpacket_stats);
1815 #endif
1816 
1817 	long if_dropped = 0;
1818 
1819 	/*
1820 	 *	To fill in ps_ifdrop, we parse /proc/net/dev for the number
1821 	 */
1822 	if (handle->opt.promisc)
1823 	{
1824 		if_dropped = handle->md.proc_dropped;
1825 		handle->md.proc_dropped = linux_if_drops(handle->md.device);
1826 		handle->md.stat.ps_ifdrop += (handle->md.proc_dropped - if_dropped);
1827 	}
1828 
1829 #ifdef HAVE_TPACKET_STATS
1830 	/*
1831 	 * Try to get the packet counts from the kernel.
1832 	 */
1833 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
1834 			&kstats, &len) > -1) {
1835 		/*
1836 		 * On systems where the PACKET_STATISTICS "getsockopt()"
1837 		 * argument is supported on PF_PACKET sockets:
1838 		 *
1839 		 *	"ps_recv" counts only packets that *passed* the
1840 		 *	filter, not packets that didn't pass the filter.
1841 		 *	This includes packets later dropped because we
1842 		 *	ran out of buffer space.
1843 		 *
1844 		 *	"ps_drop" counts packets dropped because we ran
1845 		 *	out of buffer space.  It doesn't count packets
1846 		 *	dropped by the interface driver.  It counts only
1847 		 *	packets that passed the filter.
1848 		 *
1849 		 *	See above for ps_ifdrop.
1850 		 *
1851 		 *	Both statistics include packets not yet read from
1852 		 *	the kernel by libpcap, and thus not yet seen by
1853 		 *	the application.
1854 		 *
1855 		 * In "linux/net/packet/af_packet.c", at least in the
1856 		 * 2.4.9 kernel, "tp_packets" is incremented for every
1857 		 * packet that passes the packet filter *and* is
1858 		 * successfully queued on the socket; "tp_drops" is
1859 		 * incremented for every packet dropped because there's
1860 		 * not enough free space in the socket buffer.
1861 		 *
1862 		 * When the statistics are returned for a PACKET_STATISTICS
1863 		 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
1864 		 * so that "tp_packets" counts all packets handed to
1865 		 * the PF_PACKET socket, including packets dropped because
1866 		 * there wasn't room on the socket buffer - but not
1867 		 * including packets that didn't pass the filter.
1868 		 *
1869 		 * In the BSD BPF, the count of received packets is
1870 		 * incremented for every packet handed to BPF, regardless
1871 		 * of whether it passed the filter.
1872 		 *
1873 		 * We can't make "pcap_stats()" work the same on both
1874 		 * platforms, but the best approximation is to return
1875 		 * "tp_packets" as the count of packets and "tp_drops"
1876 		 * as the count of drops.
1877 		 *
1878 		 * Keep a running total because each call to
1879 		 *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
1880 		 * resets the counters to zero.
1881 		 */
1882 		handle->md.stat.ps_recv += kstats.tp_packets;
1883 		handle->md.stat.ps_drop += kstats.tp_drops;
1884 		*stats = handle->md.stat;
1885 		return 0;
1886 	}
1887 	else
1888 	{
1889 		/*
1890 		 * If the error was EOPNOTSUPP, fall through, so that
1891 		 * if you build the library on a system with
1892 		 * "struct tpacket_stats" and run it on a system
1893 		 * that doesn't, it works as it does if the library
1894 		 * is built on a system without "struct tpacket_stats".
1895 		 */
1896 		if (errno != EOPNOTSUPP) {
1897 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1898 			    "pcap_stats: %s", pcap_strerror(errno));
1899 			return -1;
1900 		}
1901 	}
1902 #endif
1903 	/*
1904 	 * On systems where the PACKET_STATISTICS "getsockopt()" argument
1905 	 * is not supported on PF_PACKET sockets:
1906 	 *
1907 	 *	"ps_recv" counts only packets that *passed* the filter,
1908 	 *	not packets that didn't pass the filter.  It does not
1909 	 *	count packets dropped because we ran out of buffer
1910 	 *	space.
1911 	 *
1912 	 *	"ps_drop" is not supported.
1913 	 *
1914 	 *	"ps_ifdrop" is supported. It will return the number
1915 	 *	of drops the interface reports in /proc/net/dev,
1916 	 *	if that is available.
1917 	 *
1918 	 *	"ps_recv" doesn't include packets not yet read from
1919 	 *	the kernel by libpcap.
1920 	 *
1921 	 * We maintain the count of packets processed by libpcap in
1922 	 * "md.packets_read", for reasons described in the comment
1923 	 * at the end of pcap_read_packet().  We have no idea how many
1924 	 * packets were dropped by the kernel buffers -- but we know
1925 	 * how many the interface dropped, so we can return that.
1926 	 */
1927 
1928 	stats->ps_recv = handle->md.packets_read;
1929 	stats->ps_drop = 0;
1930 	stats->ps_ifdrop = handle->md.stat.ps_ifdrop;
1931 	return 0;
1932 }
1933 
1934 /*
1935  * Get from "/sys/class/net" all interfaces listed there; if they're
1936  * already in the list of interfaces we have, that won't add another
1937  * instance, but if they're not, that'll add them.
1938  *
1939  * We don't bother getting any addresses for them; it appears you can't
1940  * use SIOCGIFADDR on Linux to get IPv6 addresses for interfaces, and,
1941  * although some other types of addresses can be fetched with SIOCGIFADDR,
1942  * we don't bother with them for now.
1943  *
1944  * We also don't fail if we couldn't open "/sys/class/net"; we just leave
1945  * the list of interfaces as is, and return 0, so that we can try
1946  * scanning /proc/net/dev.
1947  */
1948 static int
1949 scan_sys_class_net(pcap_if_t **devlistp, char *errbuf)
1950 {
1951 	DIR *sys_class_net_d;
1952 	int fd;
1953 	struct dirent *ent;
1954 	char *p;
1955 	char name[512];	/* XXX - pick a size */
1956 	char *q, *saveq;
1957 	struct ifreq ifrflags;
1958 	int ret = 1;
1959 
1960 	sys_class_net_d = opendir("/sys/class/net");
1961 	if (sys_class_net_d == NULL) {
1962 		/*
1963 		 * Don't fail if it doesn't exist at all.
1964 		 */
1965 		if (errno == ENOENT)
1966 			return (0);
1967 
1968 		/*
1969 		 * Fail if we got some other error.
1970 		 */
1971 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
1972 		    "Can't open /sys/class/net: %s", pcap_strerror(errno));
1973 		return (-1);
1974 	}
1975 
1976 	/*
1977 	 * Create a socket from which to fetch interface information.
1978 	 */
1979 	fd = socket(AF_INET, SOCK_DGRAM, 0);
1980 	if (fd < 0) {
1981 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
1982 		    "socket: %s", pcap_strerror(errno));
1983 		(void)closedir(sys_class_net_d);
1984 		return (-1);
1985 	}
1986 
1987 	for (;;) {
1988 		errno = 0;
1989 		ent = readdir(sys_class_net_d);
1990 		if (ent == NULL) {
1991 			/*
1992 			 * Error or EOF; if errno != 0, it's an error.
1993 			 */
1994 			break;
1995 		}
1996 
1997 		/*
1998 		 * Ignore directories (".", "..", and any subdirectories).
1999 		 */
2000 		if (ent->d_type == DT_DIR)
2001 			continue;
2002 
2003 		/*
2004 		 * Get the interface name.
2005 		 */
2006 		p = &ent->d_name[0];
2007 		q = &name[0];
2008 		while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2009 			if (*p == ':') {
2010 				/*
2011 				 * This could be the separator between a
2012 				 * name and an alias number, or it could be
2013 				 * the separator between a name with no
2014 				 * alias number and the next field.
2015 				 *
2016 				 * If there's a colon after digits, it
2017 				 * separates the name and the alias number,
2018 				 * otherwise it separates the name and the
2019 				 * next field.
2020 				 */
2021 				saveq = q;
2022 				while (isascii(*p) && isdigit(*p))
2023 					*q++ = *p++;
2024 				if (*p != ':') {
2025 					/*
2026 					 * That was the next field,
2027 					 * not the alias number.
2028 					 */
2029 					q = saveq;
2030 				}
2031 				break;
2032 			} else
2033 				*q++ = *p++;
2034 		}
2035 		*q = '\0';
2036 
2037 		/*
2038 		 * Get the flags for this interface, and skip it if
2039 		 * it's not up.
2040 		 */
2041 		strncpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2042 		if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2043 			if (errno == ENXIO || errno == ENODEV)
2044 				continue;
2045 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2046 			    "SIOCGIFFLAGS: %.*s: %s",
2047 			    (int)sizeof(ifrflags.ifr_name),
2048 			    ifrflags.ifr_name,
2049 			    pcap_strerror(errno));
2050 			ret = -1;
2051 			break;
2052 		}
2053 		if (!(ifrflags.ifr_flags & IFF_UP))
2054 			continue;
2055 
2056 		/*
2057 		 * Add an entry for this interface, with no addresses.
2058 		 */
2059 		if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2060 		    errbuf) == -1) {
2061 			/*
2062 			 * Failure.
2063 			 */
2064 			ret = -1;
2065 			break;
2066 		}
2067 	}
2068 	if (ret != -1) {
2069 		/*
2070 		 * Well, we didn't fail for any other reason; did we
2071 		 * fail due to an error reading the directory?
2072 		 */
2073 		if (errno != 0) {
2074 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2075 			    "Error reading /sys/class/net: %s",
2076 			    pcap_strerror(errno));
2077 			ret = -1;
2078 		}
2079 	}
2080 
2081 	(void)close(fd);
2082 	(void)closedir(sys_class_net_d);
2083 	return (ret);
2084 }
2085 
2086 /*
2087  * Get from "/proc/net/dev" all interfaces listed there; if they're
2088  * already in the list of interfaces we have, that won't add another
2089  * instance, but if they're not, that'll add them.
2090  *
2091  * See comments from scan_sys_class_net().
2092  */
2093 static int
2094 scan_proc_net_dev(pcap_if_t **devlistp, char *errbuf)
2095 {
2096 	FILE *proc_net_f;
2097 	int fd;
2098 	char linebuf[512];
2099 	int linenum;
2100 	char *p;
2101 	char name[512];	/* XXX - pick a size */
2102 	char *q, *saveq;
2103 	struct ifreq ifrflags;
2104 	int ret = 0;
2105 
2106 	proc_net_f = fopen("/proc/net/dev", "r");
2107 	if (proc_net_f == NULL) {
2108 		/*
2109 		 * Don't fail if it doesn't exist at all.
2110 		 */
2111 		if (errno == ENOENT)
2112 			return (0);
2113 
2114 		/*
2115 		 * Fail if we got some other error.
2116 		 */
2117 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2118 		    "Can't open /proc/net/dev: %s", pcap_strerror(errno));
2119 		return (-1);
2120 	}
2121 
2122 	/*
2123 	 * Create a socket from which to fetch interface information.
2124 	 */
2125 	fd = socket(AF_INET, SOCK_DGRAM, 0);
2126 	if (fd < 0) {
2127 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2128 		    "socket: %s", pcap_strerror(errno));
2129 		(void)fclose(proc_net_f);
2130 		return (-1);
2131 	}
2132 
2133 	for (linenum = 1;
2134 	    fgets(linebuf, sizeof linebuf, proc_net_f) != NULL; linenum++) {
2135 		/*
2136 		 * Skip the first two lines - they're headers.
2137 		 */
2138 		if (linenum <= 2)
2139 			continue;
2140 
2141 		p = &linebuf[0];
2142 
2143 		/*
2144 		 * Skip leading white space.
2145 		 */
2146 		while (*p != '\0' && isascii(*p) && isspace(*p))
2147 			p++;
2148 		if (*p == '\0' || *p == '\n')
2149 			continue;	/* blank line */
2150 
2151 		/*
2152 		 * Get the interface name.
2153 		 */
2154 		q = &name[0];
2155 		while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2156 			if (*p == ':') {
2157 				/*
2158 				 * This could be the separator between a
2159 				 * name and an alias number, or it could be
2160 				 * the separator between a name with no
2161 				 * alias number and the next field.
2162 				 *
2163 				 * If there's a colon after digits, it
2164 				 * separates the name and the alias number,
2165 				 * otherwise it separates the name and the
2166 				 * next field.
2167 				 */
2168 				saveq = q;
2169 				while (isascii(*p) && isdigit(*p))
2170 					*q++ = *p++;
2171 				if (*p != ':') {
2172 					/*
2173 					 * That was the next field,
2174 					 * not the alias number.
2175 					 */
2176 					q = saveq;
2177 				}
2178 				break;
2179 			} else
2180 				*q++ = *p++;
2181 		}
2182 		*q = '\0';
2183 
2184 		/*
2185 		 * Get the flags for this interface, and skip it if
2186 		 * it's not up.
2187 		 */
2188 		strncpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2189 		if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2190 			if (errno == ENXIO)
2191 				continue;
2192 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2193 			    "SIOCGIFFLAGS: %.*s: %s",
2194 			    (int)sizeof(ifrflags.ifr_name),
2195 			    ifrflags.ifr_name,
2196 			    pcap_strerror(errno));
2197 			ret = -1;
2198 			break;
2199 		}
2200 		if (!(ifrflags.ifr_flags & IFF_UP))
2201 			continue;
2202 
2203 		/*
2204 		 * Add an entry for this interface, with no addresses.
2205 		 */
2206 		if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2207 		    errbuf) == -1) {
2208 			/*
2209 			 * Failure.
2210 			 */
2211 			ret = -1;
2212 			break;
2213 		}
2214 	}
2215 	if (ret != -1) {
2216 		/*
2217 		 * Well, we didn't fail for any other reason; did we
2218 		 * fail due to an error reading the file?
2219 		 */
2220 		if (ferror(proc_net_f)) {
2221 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2222 			    "Error reading /proc/net/dev: %s",
2223 			    pcap_strerror(errno));
2224 			ret = -1;
2225 		}
2226 	}
2227 
2228 	(void)close(fd);
2229 	(void)fclose(proc_net_f);
2230 	return (ret);
2231 }
2232 
2233 /*
2234  * Description string for the "any" device.
2235  */
2236 static const char any_descr[] = "Pseudo-device that captures on all interfaces";
2237 
2238 int
2239 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
2240 {
2241 	int ret;
2242 
2243 	/*
2244 	 * Read "/sys/class/net", and add to the list of interfaces all
2245 	 * interfaces listed there that we don't already have, because,
2246 	 * on Linux, SIOCGIFCONF reports only interfaces with IPv4 addresses,
2247 	 * and even getifaddrs() won't return information about
2248 	 * interfaces with no addresses, so you need to read "/sys/class/net"
2249 	 * to get the names of the rest of the interfaces.
2250 	 */
2251 	ret = scan_sys_class_net(alldevsp, errbuf);
2252 	if (ret == -1)
2253 		return (-1);	/* failed */
2254 	if (ret == 0) {
2255 		/*
2256 		 * No /sys/class/net; try reading /proc/net/dev instead.
2257 		 */
2258 		if (scan_proc_net_dev(alldevsp, errbuf) == -1)
2259 			return (-1);
2260 	}
2261 
2262 	/*
2263 	 * Add the "any" device.
2264 	 */
2265 	if (pcap_add_if(alldevsp, "any", 0, any_descr, errbuf) < 0)
2266 		return (-1);
2267 
2268 #ifdef HAVE_DAG_API
2269 	/*
2270 	 * Add DAG devices.
2271 	 */
2272 	if (dag_platform_finddevs(alldevsp, errbuf) < 0)
2273 		return (-1);
2274 #endif /* HAVE_DAG_API */
2275 
2276 #ifdef HAVE_SEPTEL_API
2277 	/*
2278 	 * Add Septel devices.
2279 	 */
2280 	if (septel_platform_finddevs(alldevsp, errbuf) < 0)
2281 		return (-1);
2282 #endif /* HAVE_SEPTEL_API */
2283 
2284 #ifdef HAVE_SNF_API
2285 	if (snf_platform_finddevs(alldevsp, errbuf) < 0)
2286 		return (-1);
2287 #endif /* HAVE_SNF_API */
2288 
2289 #ifdef PCAP_SUPPORT_BT
2290 	/*
2291 	 * Add Bluetooth devices.
2292 	 */
2293 	if (bt_platform_finddevs(alldevsp, errbuf) < 0)
2294 		return (-1);
2295 #endif
2296 
2297 #ifdef PCAP_SUPPORT_USB
2298 	/*
2299 	 * Add USB devices.
2300 	 */
2301 	if (usb_platform_finddevs(alldevsp, errbuf) < 0)
2302 		return (-1);
2303 #endif
2304 
2305 #ifdef PCAP_SUPPORT_NETFILTER
2306 	/*
2307 	 * Add netfilter devices.
2308 	 */
2309 	if (netfilter_platform_finddevs(alldevsp, errbuf) < 0)
2310 		return (-1);
2311 #endif
2312 
2313 #if PCAP_SUPPORT_CANUSB
2314   if (canusb_platform_finddevs(alldevsp, errbuf) < 0)
2315     return (-1);
2316 #endif
2317 
2318 	return (0);
2319 }
2320 
2321 /*
2322  *  Attach the given BPF code to the packet capture device.
2323  */
2324 static int
2325 pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
2326     int is_mmapped)
2327 {
2328 #ifdef SO_ATTACH_FILTER
2329 	struct sock_fprog	fcode;
2330 	int			can_filter_in_kernel;
2331 	int			err = 0;
2332 #endif
2333 
2334 	if (!handle)
2335 		return -1;
2336 	if (!filter) {
2337 	        strncpy(handle->errbuf, "setfilter: No filter specified",
2338 			PCAP_ERRBUF_SIZE);
2339 		return -1;
2340 	}
2341 
2342 	/* Make our private copy of the filter */
2343 
2344 	if (install_bpf_program(handle, filter) < 0)
2345 		/* install_bpf_program() filled in errbuf */
2346 		return -1;
2347 
2348 	/*
2349 	 * Run user level packet filter by default. Will be overriden if
2350 	 * installing a kernel filter succeeds.
2351 	 */
2352 	handle->md.use_bpf = 0;
2353 
2354 	/* Install kernel level filter if possible */
2355 
2356 #ifdef SO_ATTACH_FILTER
2357 #ifdef USHRT_MAX
2358 	if (handle->fcode.bf_len > USHRT_MAX) {
2359 		/*
2360 		 * fcode.len is an unsigned short for current kernel.
2361 		 * I have yet to see BPF-Code with that much
2362 		 * instructions but still it is possible. So for the
2363 		 * sake of correctness I added this check.
2364 		 */
2365 		fprintf(stderr, "Warning: Filter too complex for kernel\n");
2366 		fcode.len = 0;
2367 		fcode.filter = NULL;
2368 		can_filter_in_kernel = 0;
2369 	} else
2370 #endif /* USHRT_MAX */
2371 	{
2372 		/*
2373 		 * Oh joy, the Linux kernel uses struct sock_fprog instead
2374 		 * of struct bpf_program and of course the length field is
2375 		 * of different size. Pointed out by Sebastian
2376 		 *
2377 		 * Oh, and we also need to fix it up so that all "ret"
2378 		 * instructions with non-zero operands have 65535 as the
2379 		 * operand if we're not capturing in memory-mapped modee,
2380 		 * and so that, if we're in cooked mode, all memory-reference
2381 		 * instructions use special magic offsets in references to
2382 		 * the link-layer header and assume that the link-layer
2383 		 * payload begins at 0; "fix_program()" will do that.
2384 		 */
2385 		switch (fix_program(handle, &fcode, is_mmapped)) {
2386 
2387 		case -1:
2388 		default:
2389 			/*
2390 			 * Fatal error; just quit.
2391 			 * (The "default" case shouldn't happen; we
2392 			 * return -1 for that reason.)
2393 			 */
2394 			return -1;
2395 
2396 		case 0:
2397 			/*
2398 			 * The program performed checks that we can't make
2399 			 * work in the kernel.
2400 			 */
2401 			can_filter_in_kernel = 0;
2402 			break;
2403 
2404 		case 1:
2405 			/*
2406 			 * We have a filter that'll work in the kernel.
2407 			 */
2408 			can_filter_in_kernel = 1;
2409 			break;
2410 		}
2411 	}
2412 
2413 	/*
2414 	 * NOTE: at this point, we've set both the "len" and "filter"
2415 	 * fields of "fcode".  As of the 2.6.32.4 kernel, at least,
2416 	 * those are the only members of the "sock_fprog" structure,
2417 	 * so we initialize every member of that structure.
2418 	 *
2419 	 * If there is anything in "fcode" that is not initialized,
2420 	 * it is either a field added in a later kernel, or it's
2421 	 * padding.
2422 	 *
2423 	 * If a new field is added, this code needs to be updated
2424 	 * to set it correctly.
2425 	 *
2426 	 * If there are no other fields, then:
2427 	 *
2428 	 *	if the Linux kernel looks at the padding, it's
2429 	 *	buggy;
2430 	 *
2431 	 *	if the Linux kernel doesn't look at the padding,
2432 	 *	then if some tool complains that we're passing
2433 	 *	uninitialized data to the kernel, then the tool
2434 	 *	is buggy and needs to understand that it's just
2435 	 *	padding.
2436 	 */
2437 	if (can_filter_in_kernel) {
2438 		if ((err = set_kernel_filter(handle, &fcode)) == 0)
2439 		{
2440 			/* Installation succeded - using kernel filter. */
2441 			handle->md.use_bpf = 1;
2442 		}
2443 		else if (err == -1)	/* Non-fatal error */
2444 		{
2445 			/*
2446 			 * Print a warning if we weren't able to install
2447 			 * the filter for a reason other than "this kernel
2448 			 * isn't configured to support socket filters.
2449 			 */
2450 			if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
2451 				fprintf(stderr,
2452 				    "Warning: Kernel filter failed: %s\n",
2453 					pcap_strerror(errno));
2454 			}
2455 		}
2456 	}
2457 
2458 	/*
2459 	 * If we're not using the kernel filter, get rid of any kernel
2460 	 * filter that might've been there before, e.g. because the
2461 	 * previous filter could work in the kernel, or because some other
2462 	 * code attached a filter to the socket by some means other than
2463 	 * calling "pcap_setfilter()".  Otherwise, the kernel filter may
2464 	 * filter out packets that would pass the new userland filter.
2465 	 */
2466 	if (!handle->md.use_bpf)
2467 		reset_kernel_filter(handle);
2468 
2469 	/*
2470 	 * Free up the copy of the filter that was made by "fix_program()".
2471 	 */
2472 	if (fcode.filter != NULL)
2473 		free(fcode.filter);
2474 
2475 	if (err == -2)
2476 		/* Fatal error */
2477 		return -1;
2478 #endif /* SO_ATTACH_FILTER */
2479 
2480 	return 0;
2481 }
2482 
2483 static int
2484 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
2485 {
2486 	return pcap_setfilter_linux_common(handle, filter, 0);
2487 }
2488 
2489 
2490 /*
2491  * Set direction flag: Which packets do we accept on a forwarding
2492  * single device? IN, OUT or both?
2493  */
2494 static int
2495 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
2496 {
2497 #ifdef HAVE_PF_PACKET_SOCKETS
2498 	if (!handle->md.sock_packet) {
2499 		handle->direction = d;
2500 		return 0;
2501 	}
2502 #endif
2503 	/*
2504 	 * We're not using PF_PACKET sockets, so we can't determine
2505 	 * the direction of the packet.
2506 	 */
2507 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2508 	    "Setting direction is not supported on SOCK_PACKET sockets");
2509 	return -1;
2510 }
2511 
2512 #ifdef HAVE_PF_PACKET_SOCKETS
2513 /*
2514  * Map the PACKET_ value to a LINUX_SLL_ value; we
2515  * want the same numerical value to be used in
2516  * the link-layer header even if the numerical values
2517  * for the PACKET_ #defines change, so that programs
2518  * that look at the packet type field will always be
2519  * able to handle DLT_LINUX_SLL captures.
2520  */
2521 static short int
2522 map_packet_type_to_sll_type(short int sll_pkttype)
2523 {
2524 	switch (sll_pkttype) {
2525 
2526 	case PACKET_HOST:
2527 		return htons(LINUX_SLL_HOST);
2528 
2529 	case PACKET_BROADCAST:
2530 		return htons(LINUX_SLL_BROADCAST);
2531 
2532 	case PACKET_MULTICAST:
2533 		return  htons(LINUX_SLL_MULTICAST);
2534 
2535 	case PACKET_OTHERHOST:
2536 		return htons(LINUX_SLL_OTHERHOST);
2537 
2538 	case PACKET_OUTGOING:
2539 		return htons(LINUX_SLL_OUTGOING);
2540 
2541 	default:
2542 		return -1;
2543 	}
2544 }
2545 #endif
2546 
2547 /*
2548  *  Linux uses the ARP hardware type to identify the type of an
2549  *  interface. pcap uses the DLT_xxx constants for this. This
2550  *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
2551  *  constant, as arguments, and sets "handle->linktype" to the
2552  *  appropriate DLT_XXX constant and sets "handle->offset" to
2553  *  the appropriate value (to make "handle->offset" plus link-layer
2554  *  header length be a multiple of 4, so that the link-layer payload
2555  *  will be aligned on a 4-byte boundary when capturing packets).
2556  *  (If the offset isn't set here, it'll be 0; add code as appropriate
2557  *  for cases where it shouldn't be 0.)
2558  *
2559  *  If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
2560  *  in cooked mode; otherwise, we can't use cooked mode, so we have
2561  *  to pick some type that works in raw mode, or fail.
2562  *
2563  *  Sets the link type to -1 if unable to map the type.
2564  */
2565 static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
2566 {
2567 	switch (arptype) {
2568 
2569 	case ARPHRD_ETHER:
2570 		/*
2571 		 * This is (presumably) a real Ethernet capture; give it a
2572 		 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
2573 		 * that an application can let you choose it, in case you're
2574 		 * capturing DOCSIS traffic that a Cisco Cable Modem
2575 		 * Termination System is putting out onto an Ethernet (it
2576 		 * doesn't put an Ethernet header onto the wire, it puts raw
2577 		 * DOCSIS frames out on the wire inside the low-level
2578 		 * Ethernet framing).
2579 		 *
2580 		 * XXX - are there any sorts of "fake Ethernet" that have
2581 		 * ARPHRD_ETHER but that *shouldn't offer DLT_DOCSIS as
2582 		 * a Cisco CMTS won't put traffic onto it or get traffic
2583 		 * bridged onto it?  ISDN is handled in "activate_new()",
2584 		 * as we fall back on cooked mode there; are there any
2585 		 * others?
2586 		 */
2587 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2588 		/*
2589 		 * If that fails, just leave the list empty.
2590 		 */
2591 		if (handle->dlt_list != NULL) {
2592 			handle->dlt_list[0] = DLT_EN10MB;
2593 			handle->dlt_list[1] = DLT_DOCSIS;
2594 			handle->dlt_count = 2;
2595 		}
2596 		/* FALLTHROUGH */
2597 
2598 	case ARPHRD_METRICOM:
2599 	case ARPHRD_LOOPBACK:
2600 		handle->linktype = DLT_EN10MB;
2601 		handle->offset = 2;
2602 		break;
2603 
2604 	case ARPHRD_EETHER:
2605 		handle->linktype = DLT_EN3MB;
2606 		break;
2607 
2608 	case ARPHRD_AX25:
2609 		handle->linktype = DLT_AX25_KISS;
2610 		break;
2611 
2612 	case ARPHRD_PRONET:
2613 		handle->linktype = DLT_PRONET;
2614 		break;
2615 
2616 	case ARPHRD_CHAOS:
2617 		handle->linktype = DLT_CHAOS;
2618 		break;
2619 #ifndef ARPHRD_CAN
2620 #define ARPHRD_CAN 280
2621 #endif
2622 	case ARPHRD_CAN:
2623 		handle->linktype = DLT_CAN_SOCKETCAN;
2624 		break;
2625 
2626 #ifndef ARPHRD_IEEE802_TR
2627 #define ARPHRD_IEEE802_TR 800	/* From Linux 2.4 */
2628 #endif
2629 	case ARPHRD_IEEE802_TR:
2630 	case ARPHRD_IEEE802:
2631 		handle->linktype = DLT_IEEE802;
2632 		handle->offset = 2;
2633 		break;
2634 
2635 	case ARPHRD_ARCNET:
2636 		handle->linktype = DLT_ARCNET_LINUX;
2637 		break;
2638 
2639 #ifndef ARPHRD_FDDI	/* From Linux 2.2.13 */
2640 #define ARPHRD_FDDI	774
2641 #endif
2642 	case ARPHRD_FDDI:
2643 		handle->linktype = DLT_FDDI;
2644 		handle->offset = 3;
2645 		break;
2646 
2647 #ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
2648 #define ARPHRD_ATM 19
2649 #endif
2650 	case ARPHRD_ATM:
2651 		/*
2652 		 * The Classical IP implementation in ATM for Linux
2653 		 * supports both what RFC 1483 calls "LLC Encapsulation",
2654 		 * in which each packet has an LLC header, possibly
2655 		 * with a SNAP header as well, prepended to it, and
2656 		 * what RFC 1483 calls "VC Based Multiplexing", in which
2657 		 * different virtual circuits carry different network
2658 		 * layer protocols, and no header is prepended to packets.
2659 		 *
2660 		 * They both have an ARPHRD_ type of ARPHRD_ATM, so
2661 		 * you can't use the ARPHRD_ type to find out whether
2662 		 * captured packets will have an LLC header, and,
2663 		 * while there's a socket ioctl to *set* the encapsulation
2664 		 * type, there's no ioctl to *get* the encapsulation type.
2665 		 *
2666 		 * This means that
2667 		 *
2668 		 *	programs that dissect Linux Classical IP frames
2669 		 *	would have to check for an LLC header and,
2670 		 *	depending on whether they see one or not, dissect
2671 		 *	the frame as LLC-encapsulated or as raw IP (I
2672 		 *	don't know whether there's any traffic other than
2673 		 *	IP that would show up on the socket, or whether
2674 		 *	there's any support for IPv6 in the Linux
2675 		 *	Classical IP code);
2676 		 *
2677 		 *	filter expressions would have to compile into
2678 		 *	code that checks for an LLC header and does
2679 		 *	the right thing.
2680 		 *
2681 		 * Both of those are a nuisance - and, at least on systems
2682 		 * that support PF_PACKET sockets, we don't have to put
2683 		 * up with those nuisances; instead, we can just capture
2684 		 * in cooked mode.  That's what we'll do, if we can.
2685 		 * Otherwise, we'll just fail.
2686 		 */
2687 		if (cooked_ok)
2688 			handle->linktype = DLT_LINUX_SLL;
2689 		else
2690 			handle->linktype = -1;
2691 		break;
2692 
2693 #ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
2694 #define ARPHRD_IEEE80211 801
2695 #endif
2696 	case ARPHRD_IEEE80211:
2697 		handle->linktype = DLT_IEEE802_11;
2698 		break;
2699 
2700 #ifndef ARPHRD_IEEE80211_PRISM  /* From Linux 2.4.18 */
2701 #define ARPHRD_IEEE80211_PRISM 802
2702 #endif
2703 	case ARPHRD_IEEE80211_PRISM:
2704 		handle->linktype = DLT_PRISM_HEADER;
2705 		break;
2706 
2707 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
2708 #define ARPHRD_IEEE80211_RADIOTAP 803
2709 #endif
2710 	case ARPHRD_IEEE80211_RADIOTAP:
2711 		handle->linktype = DLT_IEEE802_11_RADIO;
2712 		break;
2713 
2714 	case ARPHRD_PPP:
2715 		/*
2716 		 * Some PPP code in the kernel supplies no link-layer
2717 		 * header whatsoever to PF_PACKET sockets; other PPP
2718 		 * code supplies PPP link-layer headers ("syncppp.c");
2719 		 * some PPP code might supply random link-layer
2720 		 * headers (PPP over ISDN - there's code in Ethereal,
2721 		 * for example, to cope with PPP-over-ISDN captures
2722 		 * with which the Ethereal developers have had to cope,
2723 		 * heuristically trying to determine which of the
2724 		 * oddball link-layer headers particular packets have).
2725 		 *
2726 		 * As such, we just punt, and run all PPP interfaces
2727 		 * in cooked mode, if we can; otherwise, we just treat
2728 		 * it as DLT_RAW, for now - if somebody needs to capture,
2729 		 * on a 2.0[.x] kernel, on PPP devices that supply a
2730 		 * link-layer header, they'll have to add code here to
2731 		 * map to the appropriate DLT_ type (possibly adding a
2732 		 * new DLT_ type, if necessary).
2733 		 */
2734 		if (cooked_ok)
2735 			handle->linktype = DLT_LINUX_SLL;
2736 		else {
2737 			/*
2738 			 * XXX - handle ISDN types here?  We can't fall
2739 			 * back on cooked sockets, so we'd have to
2740 			 * figure out from the device name what type of
2741 			 * link-layer encapsulation it's using, and map
2742 			 * that to an appropriate DLT_ value, meaning
2743 			 * we'd map "isdnN" devices to DLT_RAW (they
2744 			 * supply raw IP packets with no link-layer
2745 			 * header) and "isdY" devices to a new DLT_I4L_IP
2746 			 * type that has only an Ethernet packet type as
2747 			 * a link-layer header.
2748 			 *
2749 			 * But sometimes we seem to get random crap
2750 			 * in the link-layer header when capturing on
2751 			 * ISDN devices....
2752 			 */
2753 			handle->linktype = DLT_RAW;
2754 		}
2755 		break;
2756 
2757 #ifndef ARPHRD_CISCO
2758 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
2759 #endif
2760 	case ARPHRD_CISCO:
2761 		handle->linktype = DLT_C_HDLC;
2762 		break;
2763 
2764 	/* Not sure if this is correct for all tunnels, but it
2765 	 * works for CIPE */
2766 	case ARPHRD_TUNNEL:
2767 #ifndef ARPHRD_SIT
2768 #define ARPHRD_SIT 776	/* From Linux 2.2.13 */
2769 #endif
2770 	case ARPHRD_SIT:
2771 	case ARPHRD_CSLIP:
2772 	case ARPHRD_SLIP6:
2773 	case ARPHRD_CSLIP6:
2774 	case ARPHRD_ADAPT:
2775 	case ARPHRD_SLIP:
2776 #ifndef ARPHRD_RAWHDLC
2777 #define ARPHRD_RAWHDLC 518
2778 #endif
2779 	case ARPHRD_RAWHDLC:
2780 #ifndef ARPHRD_DLCI
2781 #define ARPHRD_DLCI 15
2782 #endif
2783 	case ARPHRD_DLCI:
2784 		/*
2785 		 * XXX - should some of those be mapped to DLT_LINUX_SLL
2786 		 * instead?  Should we just map all of them to DLT_LINUX_SLL?
2787 		 */
2788 		handle->linktype = DLT_RAW;
2789 		break;
2790 
2791 #ifndef ARPHRD_FRAD
2792 #define ARPHRD_FRAD 770
2793 #endif
2794 	case ARPHRD_FRAD:
2795 		handle->linktype = DLT_FRELAY;
2796 		break;
2797 
2798 	case ARPHRD_LOCALTLK:
2799 		handle->linktype = DLT_LTALK;
2800 		break;
2801 
2802 #ifndef ARPHRD_FCPP
2803 #define ARPHRD_FCPP	784
2804 #endif
2805 	case ARPHRD_FCPP:
2806 #ifndef ARPHRD_FCAL
2807 #define ARPHRD_FCAL	785
2808 #endif
2809 	case ARPHRD_FCAL:
2810 #ifndef ARPHRD_FCPL
2811 #define ARPHRD_FCPL	786
2812 #endif
2813 	case ARPHRD_FCPL:
2814 #ifndef ARPHRD_FCFABRIC
2815 #define ARPHRD_FCFABRIC	787
2816 #endif
2817 	case ARPHRD_FCFABRIC:
2818 		/*
2819 		 * We assume that those all mean RFC 2625 IP-over-
2820 		 * Fibre Channel, with the RFC 2625 header at
2821 		 * the beginning of the packet.
2822 		 */
2823 		handle->linktype = DLT_IP_OVER_FC;
2824 		break;
2825 
2826 #ifndef ARPHRD_IRDA
2827 #define ARPHRD_IRDA	783
2828 #endif
2829 	case ARPHRD_IRDA:
2830 		/* Don't expect IP packet out of this interfaces... */
2831 		handle->linktype = DLT_LINUX_IRDA;
2832 		/* We need to save packet direction for IrDA decoding,
2833 		 * so let's use "Linux-cooked" mode. Jean II */
2834 		//handle->md.cooked = 1;
2835 		break;
2836 
2837 	/* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
2838 	 * is needed, please report it to <daniele@orlandi.com> */
2839 #ifndef ARPHRD_LAPD
2840 #define ARPHRD_LAPD	8445
2841 #endif
2842 	case ARPHRD_LAPD:
2843 		/* Don't expect IP packet out of this interfaces... */
2844 		handle->linktype = DLT_LINUX_LAPD;
2845 		break;
2846 
2847 #ifndef ARPHRD_NONE
2848 #define ARPHRD_NONE	0xFFFE
2849 #endif
2850 	case ARPHRD_NONE:
2851 		/*
2852 		 * No link-layer header; packets are just IP
2853 		 * packets, so use DLT_RAW.
2854 		 */
2855 		handle->linktype = DLT_RAW;
2856 		break;
2857 
2858 #ifndef ARPHRD_IEEE802154
2859 #define ARPHRD_IEEE802154      804
2860 #endif
2861        case ARPHRD_IEEE802154:
2862                handle->linktype =  DLT_IEEE802_15_4_NOFCS;
2863                break;
2864 
2865 	default:
2866 		handle->linktype = -1;
2867 		break;
2868 	}
2869 }
2870 
2871 /* ===== Functions to interface to the newer kernels ================== */
2872 
2873 /*
2874  * Try to open a packet socket using the new kernel PF_PACKET interface.
2875  * Returns 1 on success, 0 on an error that means the new interface isn't
2876  * present (so the old SOCK_PACKET interface should be tried), and a
2877  * PCAP_ERROR_ value on an error that means that the old mechanism won't
2878  * work either (so it shouldn't be tried).
2879  */
2880 static int
2881 activate_new(pcap_t *handle)
2882 {
2883 #ifdef HAVE_PF_PACKET_SOCKETS
2884 	const char		*device = handle->opt.source;
2885 	int			is_any_device = (strcmp(device, "any") == 0);
2886 	int			sock_fd = -1, arptype;
2887 #ifdef HAVE_PACKET_AUXDATA
2888 	int			val;
2889 #endif
2890 	int			err = 0;
2891 	struct packet_mreq	mr;
2892 
2893 	/*
2894 	 * Open a socket with protocol family packet. If the
2895 	 * "any" device was specified, we open a SOCK_DGRAM
2896 	 * socket for the cooked interface, otherwise we first
2897 	 * try a SOCK_RAW socket for the raw interface.
2898 	 */
2899 	sock_fd = is_any_device ?
2900 		socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) :
2901 		socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
2902 
2903 	if (sock_fd == -1) {
2904 		if (errno == EINVAL || errno == EAFNOSUPPORT) {
2905 			/*
2906 			 * We don't support PF_PACKET/SOCK_whatever
2907 			 * sockets; try the old mechanism.
2908 			 */
2909 			return 0;
2910 		}
2911 
2912 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
2913 			 pcap_strerror(errno) );
2914 		if (errno == EPERM || errno == EACCES) {
2915 			/*
2916 			 * You don't have permission to open the
2917 			 * socket.
2918 			 */
2919 			return PCAP_ERROR_PERM_DENIED;
2920 		} else {
2921 			/*
2922 			 * Other error.
2923 			 */
2924 			return PCAP_ERROR;
2925 		}
2926 	}
2927 
2928 	/* It seems the kernel supports the new interface. */
2929 	handle->md.sock_packet = 0;
2930 
2931 	/*
2932 	 * Get the interface index of the loopback device.
2933 	 * If the attempt fails, don't fail, just set the
2934 	 * "md.lo_ifindex" to -1.
2935 	 *
2936 	 * XXX - can there be more than one device that loops
2937 	 * packets back, i.e. devices other than "lo"?  If so,
2938 	 * we'd need to find them all, and have an array of
2939 	 * indices for them, and check all of them in
2940 	 * "pcap_read_packet()".
2941 	 */
2942 	handle->md.lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
2943 
2944 	/*
2945 	 * Default value for offset to align link-layer payload
2946 	 * on a 4-byte boundary.
2947 	 */
2948 	handle->offset	 = 0;
2949 
2950 	/*
2951 	 * What kind of frames do we have to deal with? Fall back
2952 	 * to cooked mode if we have an unknown interface type
2953 	 * or a type we know doesn't work well in raw mode.
2954 	 */
2955 	if (!is_any_device) {
2956 		/* Assume for now we don't need cooked mode. */
2957 		handle->md.cooked = 0;
2958 
2959 		if (handle->opt.rfmon) {
2960 			/*
2961 			 * We were asked to turn on monitor mode.
2962 			 * Do so before we get the link-layer type,
2963 			 * because entering monitor mode could change
2964 			 * the link-layer type.
2965 			 */
2966 			err = enter_rfmon_mode(handle, sock_fd, device);
2967 			if (err < 0) {
2968 				/* Hard failure */
2969 				close(sock_fd);
2970 				return err;
2971 			}
2972 			if (err == 0) {
2973 				/*
2974 				 * Nothing worked for turning monitor mode
2975 				 * on.
2976 				 */
2977 				close(sock_fd);
2978 				return PCAP_ERROR_RFMON_NOTSUP;
2979 			}
2980 
2981 			/*
2982 			 * Either monitor mode has been turned on for
2983 			 * the device, or we've been given a different
2984 			 * device to open for monitor mode.  If we've
2985 			 * been given a different device, use it.
2986 			 */
2987 			if (handle->md.mondevice != NULL)
2988 				device = handle->md.mondevice;
2989 		}
2990 		arptype	= iface_get_arptype(sock_fd, device, handle->errbuf);
2991 		if (arptype < 0) {
2992 			close(sock_fd);
2993 			return arptype;
2994 		}
2995 		map_arphrd_to_dlt(handle, arptype, 1);
2996 		if (handle->linktype == -1 ||
2997 		    handle->linktype == DLT_LINUX_SLL ||
2998 		    handle->linktype == DLT_LINUX_IRDA ||
2999 		    handle->linktype == DLT_LINUX_LAPD ||
3000 		    (handle->linktype == DLT_EN10MB &&
3001 		     (strncmp("isdn", device, 4) == 0 ||
3002 		      strncmp("isdY", device, 4) == 0))) {
3003 			/*
3004 			 * Unknown interface type (-1), or a
3005 			 * device we explicitly chose to run
3006 			 * in cooked mode (e.g., PPP devices),
3007 			 * or an ISDN device (whose link-layer
3008 			 * type we can only determine by using
3009 			 * APIs that may be different on different
3010 			 * kernels) - reopen in cooked mode.
3011 			 */
3012 			if (close(sock_fd) == -1) {
3013 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3014 					 "close: %s", pcap_strerror(errno));
3015 				return PCAP_ERROR;
3016 			}
3017 			sock_fd = socket(PF_PACKET, SOCK_DGRAM,
3018 			    htons(ETH_P_ALL));
3019 			if (sock_fd == -1) {
3020 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3021 				    "socket: %s", pcap_strerror(errno));
3022 				if (errno == EPERM || errno == EACCES) {
3023 					/*
3024 					 * You don't have permission to
3025 					 * open the socket.
3026 					 */
3027 					return PCAP_ERROR_PERM_DENIED;
3028 				} else {
3029 					/*
3030 					 * Other error.
3031 					 */
3032 					return PCAP_ERROR;
3033 				}
3034 			}
3035 			handle->md.cooked = 1;
3036 
3037 			/*
3038 			 * Get rid of any link-layer type list
3039 			 * we allocated - this only supports cooked
3040 			 * capture.
3041 			 */
3042 			if (handle->dlt_list != NULL) {
3043 				free(handle->dlt_list);
3044 				handle->dlt_list = NULL;
3045 				handle->dlt_count = 0;
3046 			}
3047 
3048 			if (handle->linktype == -1) {
3049 				/*
3050 				 * Warn that we're falling back on
3051 				 * cooked mode; we may want to
3052 				 * update "map_arphrd_to_dlt()"
3053 				 * to handle the new type.
3054 				 */
3055 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3056 					"arptype %d not "
3057 					"supported by libpcap - "
3058 					"falling back to cooked "
3059 					"socket",
3060 					arptype);
3061 			}
3062 
3063 			/*
3064 			 * IrDA capture is not a real "cooked" capture,
3065 			 * it's IrLAP frames, not IP packets.  The
3066 			 * same applies to LAPD capture.
3067 			 */
3068 			if (handle->linktype != DLT_LINUX_IRDA &&
3069 			    handle->linktype != DLT_LINUX_LAPD)
3070 				handle->linktype = DLT_LINUX_SLL;
3071 		}
3072 
3073 		handle->md.ifindex = iface_get_id(sock_fd, device,
3074 		    handle->errbuf);
3075 		if (handle->md.ifindex == -1) {
3076 			close(sock_fd);
3077 			return PCAP_ERROR;
3078 		}
3079 
3080 		if ((err = iface_bind(sock_fd, handle->md.ifindex,
3081 		    handle->errbuf)) != 1) {
3082 		    	close(sock_fd);
3083 			if (err < 0)
3084 				return err;
3085 			else
3086 				return 0;	/* try old mechanism */
3087 		}
3088 	} else {
3089 		/*
3090 		 * The "any" device.
3091 		 */
3092 		if (handle->opt.rfmon) {
3093 			/*
3094 			 * It doesn't support monitor mode.
3095 			 */
3096 			return PCAP_ERROR_RFMON_NOTSUP;
3097 		}
3098 
3099 		/*
3100 		 * It uses cooked mode.
3101 		 */
3102 		handle->md.cooked = 1;
3103 		handle->linktype = DLT_LINUX_SLL;
3104 
3105 		/*
3106 		 * We're not bound to a device.
3107 		 * For now, we're using this as an indication
3108 		 * that we can't transmit; stop doing that only
3109 		 * if we figure out how to transmit in cooked
3110 		 * mode.
3111 		 */
3112 		handle->md.ifindex = -1;
3113 	}
3114 
3115 	/*
3116 	 * Select promiscuous mode on if "promisc" is set.
3117 	 *
3118 	 * Do not turn allmulti mode on if we don't select
3119 	 * promiscuous mode - on some devices (e.g., Orinoco
3120 	 * wireless interfaces), allmulti mode isn't supported
3121 	 * and the driver implements it by turning promiscuous
3122 	 * mode on, and that screws up the operation of the
3123 	 * card as a normal networking interface, and on no
3124 	 * other platform I know of does starting a non-
3125 	 * promiscuous capture affect which multicast packets
3126 	 * are received by the interface.
3127 	 */
3128 
3129 	/*
3130 	 * Hmm, how can we set promiscuous mode on all interfaces?
3131 	 * I am not sure if that is possible at all.  For now, we
3132 	 * silently ignore attempts to turn promiscuous mode on
3133 	 * for the "any" device (so you don't have to explicitly
3134 	 * disable it in programs such as tcpdump).
3135 	 */
3136 
3137 	if (!is_any_device && handle->opt.promisc) {
3138 		memset(&mr, 0, sizeof(mr));
3139 		mr.mr_ifindex = handle->md.ifindex;
3140 		mr.mr_type    = PACKET_MR_PROMISC;
3141 		if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
3142 		    &mr, sizeof(mr)) == -1) {
3143 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3144 				"setsockopt: %s", pcap_strerror(errno));
3145 			close(sock_fd);
3146 			return PCAP_ERROR;
3147 		}
3148 	}
3149 
3150 	/* Enable auxillary data if supported and reserve room for
3151 	 * reconstructing VLAN headers. */
3152 #ifdef HAVE_PACKET_AUXDATA
3153 	val = 1;
3154 	if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
3155 		       sizeof(val)) == -1 && errno != ENOPROTOOPT) {
3156 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3157 			 "setsockopt: %s", pcap_strerror(errno));
3158 		close(sock_fd);
3159 		return PCAP_ERROR;
3160 	}
3161 	handle->offset += VLAN_TAG_LEN;
3162 #endif /* HAVE_PACKET_AUXDATA */
3163 
3164 	/*
3165 	 * This is a 2.2[.x] or later kernel (we know that
3166 	 * because we're not using a SOCK_PACKET socket -
3167 	 * PF_PACKET is supported only in 2.2 and later
3168 	 * kernels).
3169 	 *
3170 	 * We can safely pass "recvfrom()" a byte count
3171 	 * based on the snapshot length.
3172 	 *
3173 	 * If we're in cooked mode, make the snapshot length
3174 	 * large enough to hold a "cooked mode" header plus
3175 	 * 1 byte of packet data (so we don't pass a byte
3176 	 * count of 0 to "recvfrom()").
3177 	 */
3178 	if (handle->md.cooked) {
3179 		if (handle->snapshot < SLL_HDR_LEN + 1)
3180 			handle->snapshot = SLL_HDR_LEN + 1;
3181 	}
3182 	handle->bufsize = handle->snapshot;
3183 
3184 	/* Save the socket FD in the pcap structure */
3185 	handle->fd = sock_fd;
3186 
3187 	return 1;
3188 #else
3189 	strncpy(ebuf,
3190 		"New packet capturing interface not supported by build "
3191 		"environment", PCAP_ERRBUF_SIZE);
3192 	return 0;
3193 #endif
3194 }
3195 
3196 #ifdef HAVE_PACKET_RING
3197 /*
3198  * Attempt to activate with memory-mapped access.
3199  *
3200  * On success, returns 1, and sets *status to 0 if there are no warnings
3201  * or to a PCAP_WARNING_ code if there is a warning.
3202  *
3203  * On failure due to lack of support for memory-mapped capture, returns
3204  * 0.
3205  *
3206  * On error, returns -1, and sets *status to the appropriate error code;
3207  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3208  */
3209 static int
3210 activate_mmap(pcap_t *handle, int *status)
3211 {
3212 	int ret;
3213 
3214 	/*
3215 	 * Attempt to allocate a buffer to hold the contents of one
3216 	 * packet, for use by the oneshot callback.
3217 	 */
3218 	handle->md.oneshot_buffer = malloc(handle->snapshot);
3219 	if (handle->md.oneshot_buffer == NULL) {
3220 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3221 			 "can't allocate oneshot buffer: %s",
3222 			 pcap_strerror(errno));
3223 		*status = PCAP_ERROR;
3224 		return -1;
3225 	}
3226 
3227 	if (handle->opt.buffer_size == 0) {
3228 		/* by default request 2M for the ring buffer */
3229 		handle->opt.buffer_size = 2*1024*1024;
3230 	}
3231 	ret = prepare_tpacket_socket(handle);
3232 	if (ret == -1) {
3233 		free(handle->md.oneshot_buffer);
3234 		*status = PCAP_ERROR;
3235 		return ret;
3236 	}
3237 	ret = create_ring(handle, status);
3238 	if (ret == 0) {
3239 		/*
3240 		 * We don't support memory-mapped capture; our caller
3241 		 * will fall back on reading from the socket.
3242 		 */
3243 		free(handle->md.oneshot_buffer);
3244 		return 0;
3245 	}
3246 	if (ret == -1) {
3247 		/*
3248 		 * Error attempting to enable memory-mapped capture;
3249 		 * fail.  create_ring() has set *status.
3250 		 */
3251 		free(handle->md.oneshot_buffer);
3252 		return -1;
3253 	}
3254 
3255 	/*
3256 	 * Success.  *status has been set either to 0 if there are no
3257 	 * warnings or to a PCAP_WARNING_ value if there is a warning.
3258 	 *
3259 	 * Override some defaults and inherit the other fields from
3260 	 * activate_new.
3261 	 * handle->offset is used to get the current position into the rx ring.
3262 	 * handle->cc is used to store the ring size.
3263 	 */
3264 	handle->read_op = pcap_read_linux_mmap;
3265 	handle->cleanup_op = pcap_cleanup_linux_mmap;
3266 	handle->setfilter_op = pcap_setfilter_linux_mmap;
3267 	handle->setnonblock_op = pcap_setnonblock_mmap;
3268 	handle->getnonblock_op = pcap_getnonblock_mmap;
3269 	handle->oneshot_callback = pcap_oneshot_mmap;
3270 	handle->selectable_fd = handle->fd;
3271 	return 1;
3272 }
3273 #else /* HAVE_PACKET_RING */
3274 static int
3275 activate_mmap(pcap_t *handle _U_, int *status _U_)
3276 {
3277 	return 0;
3278 }
3279 #endif /* HAVE_PACKET_RING */
3280 
3281 #ifdef HAVE_PACKET_RING
3282 /*
3283  * Attempt to set the socket to version 2 of the memory-mapped header.
3284  * Return 1 if we succeed or if we fail because version 2 isn't
3285  * supported; return -1 on any other error, and set handle->errbuf.
3286  */
3287 static int
3288 prepare_tpacket_socket(pcap_t *handle)
3289 {
3290 #ifdef HAVE_TPACKET2
3291 	socklen_t len;
3292 	int val;
3293 #endif
3294 
3295 	handle->md.tp_version = TPACKET_V1;
3296 	handle->md.tp_hdrlen = sizeof(struct tpacket_hdr);
3297 
3298 #ifdef HAVE_TPACKET2
3299 	/* Probe whether kernel supports TPACKET_V2 */
3300 	val = TPACKET_V2;
3301 	len = sizeof(val);
3302 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
3303 		if (errno == ENOPROTOOPT)
3304 			return 1;	/* no - just drive on */
3305 
3306 		/* Yes - treat as a failure. */
3307 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3308 		    "can't get TPACKET_V2 header len on packet socket: %s",
3309 		    pcap_strerror(errno));
3310 		return -1;
3311 	}
3312 	handle->md.tp_hdrlen = val;
3313 
3314 	val = TPACKET_V2;
3315 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
3316 		       sizeof(val)) < 0) {
3317 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3318 		    "can't activate TPACKET_V2 on packet socket: %s",
3319 		    pcap_strerror(errno));
3320 		return -1;
3321 	}
3322 	handle->md.tp_version = TPACKET_V2;
3323 
3324 	/* Reserve space for VLAN tag reconstruction */
3325 	val = VLAN_TAG_LEN;
3326 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val,
3327 		       sizeof(val)) < 0) {
3328 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3329 		    "can't set up reserve on packet socket: %s",
3330 		    pcap_strerror(errno));
3331 		return -1;
3332 	}
3333 
3334 #endif /* HAVE_TPACKET2 */
3335 	return 1;
3336 }
3337 
3338 /*
3339  * Attempt to set up memory-mapped access.
3340  *
3341  * On success, returns 1, and sets *status to 0 if there are no warnings
3342  * or to a PCAP_WARNING_ code if there is a warning.
3343  *
3344  * On failure due to lack of support for memory-mapped capture, returns
3345  * 0.
3346  *
3347  * On error, returns -1, and sets *status to the appropriate error code;
3348  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3349  */
3350 static int
3351 create_ring(pcap_t *handle, int *status)
3352 {
3353 	unsigned i, j, frames_per_block;
3354 	struct tpacket_req req;
3355 	socklen_t len;
3356 	unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
3357 	unsigned int frame_size;
3358 
3359 	/*
3360 	 * Start out assuming no warnings or errors.
3361 	 */
3362 	*status = 0;
3363 
3364 	/* Note that with large snapshot length (say 64K, which is the default
3365 	 * for recent versions of tcpdump, the value that "-s 0" has given
3366 	 * for a long time with tcpdump, and the default in Wireshark/TShark),
3367 	 * if we use the snapshot length to calculate the frame length,
3368 	 * only a few frames will be available in the ring even with pretty
3369 	 * large ring size (and a lot of memory will be unused).
3370 	 *
3371 	 * Ideally, we should choose a frame length based on the
3372 	 * minimum of the specified snapshot length and the maximum
3373 	 * packet size.  That's not as easy as it sounds; consider, for
3374 	 * example, an 802.11 interface in monitor mode, where the
3375 	 * frame would include a radiotap header, where the maximum
3376 	 * radiotap header length is device-dependent.
3377 	 *
3378 	 * So, for now, we just do this for Ethernet devices, where
3379 	 * there's no metadata header, and the link-layer header is
3380 	 * fixed length.  We can get the maximum packet size by
3381 	 * adding 18, the Ethernet header length plus the CRC length
3382 	 * (just in case we happen to get the CRC in the packet), to
3383 	 * the MTU of the interface; we fetch the MTU in the hopes
3384 	 * that it reflects support for jumbo frames.  (Even if the
3385 	 * interface is just being used for passive snooping, the driver
3386 	 * might set the size of buffers in the receive ring based on
3387 	 * the MTU, so that the MTU limits the maximum size of packets
3388 	 * that we can receive.)
3389 	 *
3390 	 * We don't do that if segmentation/fragmentation or receive
3391 	 * offload are enabled, so we don't get rudely surprised by
3392 	 * "packets" bigger than the MTU. */
3393 	frame_size = handle->snapshot;
3394 	if (handle->linktype == DLT_EN10MB) {
3395 		int mtu;
3396 		int offload;
3397 
3398 		offload = iface_get_offload(handle);
3399 		if (offload == -1) {
3400 			*status = PCAP_ERROR;
3401 			return -1;
3402 		}
3403 		if (!offload) {
3404 			mtu = iface_get_mtu(handle->fd, handle->opt.source,
3405 			    handle->errbuf);
3406 			if (mtu == -1) {
3407 				*status = PCAP_ERROR;
3408 				return -1;
3409 			}
3410 			if (frame_size > mtu + 18)
3411 				frame_size = mtu + 18;
3412 		}
3413 	}
3414 
3415 	/* NOTE: calculus matching those in tpacket_rcv()
3416 	 * in linux-2.6/net/packet/af_packet.c
3417 	 */
3418 	len = sizeof(sk_type);
3419 	if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type, &len) < 0) {
3420 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "getsockopt: %s", pcap_strerror(errno));
3421 		*status = PCAP_ERROR;
3422 		return -1;
3423 	}
3424 #ifdef PACKET_RESERVE
3425 	len = sizeof(tp_reserve);
3426 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &tp_reserve, &len) < 0) {
3427 		if (errno != ENOPROTOOPT) {
3428 			/*
3429 			 * ENOPROTOOPT means "kernel doesn't support
3430 			 * PACKET_RESERVE", in which case we fall back
3431 			 * as best we can.
3432 			 */
3433 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "getsockopt: %s", pcap_strerror(errno));
3434 			*status = PCAP_ERROR;
3435 			return -1;
3436 		}
3437 		tp_reserve = 0;	/* older kernel, reserve not supported */
3438 	}
3439 #else
3440 	tp_reserve = 0;	/* older kernel, reserve not supported */
3441 #endif
3442 	maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
3443 		/* XXX: in the kernel maclen is calculated from
3444 		 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
3445 		 * in:  packet_snd()           in linux-2.6/net/packet/af_packet.c
3446 		 * then packet_alloc_skb()     in linux-2.6/net/packet/af_packet.c
3447 		 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
3448 		 * but I see no way to get those sizes in userspace,
3449 		 * like for instance with an ifreq ioctl();
3450 		 * the best thing I've found so far is MAX_HEADER in the kernel
3451 		 * part of linux-2.6/include/linux/netdevice.h
3452 		 * which goes up to 128+48=176; since pcap-linux.c defines
3453 		 * a MAX_LINKHEADER_SIZE of 256 which is greater than that,
3454 		 * let's use it.. maybe is it even large enough to directly
3455 		 * replace macoff..
3456 		 */
3457 	tp_hdrlen = TPACKET_ALIGN(handle->md.tp_hdrlen) + sizeof(struct sockaddr_ll) ;
3458 	netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
3459 		/* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN of
3460 		 * netoff, which contradicts
3461 		 * linux-2.6/Documentation/networking/packet_mmap.txt
3462 		 * documenting that:
3463 		 * "- Gap, chosen so that packet data (Start+tp_net)
3464 		 * aligns to TPACKET_ALIGNMENT=16"
3465 		 */
3466 		/* NOTE: in linux-2.6/include/linux/skbuff.h:
3467 		 * "CPUs often take a performance hit
3468 		 *  when accessing unaligned memory locations"
3469 		 */
3470 	macoff = netoff - maclen;
3471 	req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
3472 	req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
3473 
3474 	/* compute the minumum block size that will handle this frame.
3475 	 * The block has to be page size aligned.
3476 	 * The max block size allowed by the kernel is arch-dependent and
3477 	 * it's not explicitly checked here. */
3478 	req.tp_block_size = getpagesize();
3479 	while (req.tp_block_size < req.tp_frame_size)
3480 		req.tp_block_size <<= 1;
3481 
3482 	frames_per_block = req.tp_block_size/req.tp_frame_size;
3483 
3484 	/*
3485 	 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
3486 	 * so we check for PACKET_TIMESTAMP.  We check for
3487 	 * linux/net_tstamp.h just in case a system somehow has
3488 	 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
3489 	 * be unnecessary.
3490 	 *
3491 	 * SIOCSHWTSTAMP was introduced in the patch that introduced
3492 	 * linux/net_tstamp.h, so we don't bother checking whether
3493 	 * SIOCSHWTSTAMP is defined (if your Linux system has
3494 	 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
3495 	 * Linux system is badly broken).
3496 	 */
3497 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
3498 	/*
3499 	 * If we were told to do so, ask the kernel and the driver
3500 	 * to use hardware timestamps.
3501 	 *
3502 	 * Hardware timestamps are only supported with mmapped
3503 	 * captures.
3504 	 */
3505 	if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
3506 	    handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
3507 		struct hwtstamp_config hwconfig;
3508 		struct ifreq ifr;
3509 		int timesource;
3510 
3511 		/*
3512 		 * Ask for hardware time stamps on all packets,
3513 		 * including transmitted packets.
3514 		 */
3515 		memset(&hwconfig, 0, sizeof(hwconfig));
3516 		hwconfig.tx_type = HWTSTAMP_TX_ON;
3517 		hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
3518 
3519 		memset(&ifr, 0, sizeof(ifr));
3520 		strcpy(ifr.ifr_name, handle->opt.source);
3521 		ifr.ifr_data = (void *)&hwconfig;
3522 
3523 		if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
3524 			switch (errno) {
3525 
3526 			case EPERM:
3527 				/*
3528 				 * Treat this as an error, as the
3529 				 * user should try to run this
3530 				 * with the appropriate privileges -
3531 				 * and, if they can't, shouldn't
3532 				 * try requesting hardware time stamps.
3533 				 */
3534 				*status = PCAP_ERROR_PERM_DENIED;
3535 				return -1;
3536 
3537 			case EOPNOTSUPP:
3538 				/*
3539 				 * Treat this as a warning, as the
3540 				 * only way to fix the warning is to
3541 				 * get an adapter that supports hardware
3542 				 * time stamps.  We'll just fall back
3543 				 * on the standard host time stamps.
3544 				 */
3545 				*status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
3546 				break;
3547 
3548 			default:
3549 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3550 					"SIOCSHWTSTAMP failed: %s",
3551 					pcap_strerror(errno));
3552 				*status = PCAP_ERROR;
3553 				return -1;
3554 			}
3555 		} else {
3556 			/*
3557 			 * Well, that worked.  Now specify the type of
3558 			 * hardware time stamp we want for this
3559 			 * socket.
3560 			 */
3561 			if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
3562 				/*
3563 				 * Hardware timestamp, synchronized
3564 				 * with the system clock.
3565 				 */
3566 				timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
3567 			} else {
3568 				/*
3569 				 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
3570 				 * timestamp, not synchronized with the
3571 				 * system clock.
3572 				 */
3573 				timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
3574 			}
3575 			if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
3576 				(void *)&timesource, sizeof(timesource))) {
3577 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3578 					"can't set PACKET_TIMESTAMP: %s",
3579 					pcap_strerror(errno));
3580 				*status = PCAP_ERROR;
3581 				return -1;
3582 			}
3583 		}
3584 	}
3585 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
3586 
3587 	/* ask the kernel to create the ring */
3588 retry:
3589 	req.tp_block_nr = req.tp_frame_nr / frames_per_block;
3590 
3591 	/* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
3592 	req.tp_frame_nr = req.tp_block_nr * frames_per_block;
3593 
3594 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3595 					(void *) &req, sizeof(req))) {
3596 		if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
3597 			/*
3598 			 * Memory failure; try to reduce the requested ring
3599 			 * size.
3600 			 *
3601 			 * We used to reduce this by half -- do 5% instead.
3602 			 * That may result in more iterations and a longer
3603 			 * startup, but the user will be much happier with
3604 			 * the resulting buffer size.
3605 			 */
3606 			if (req.tp_frame_nr < 20)
3607 				req.tp_frame_nr -= 1;
3608 			else
3609 				req.tp_frame_nr -= req.tp_frame_nr/20;
3610 			goto retry;
3611 		}
3612 		if (errno == ENOPROTOOPT) {
3613 			/*
3614 			 * We don't have ring buffer support in this kernel.
3615 			 */
3616 			return 0;
3617 		}
3618 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3619 		    "can't create rx ring on packet socket: %s",
3620 		    pcap_strerror(errno));
3621 		*status = PCAP_ERROR;
3622 		return -1;
3623 	}
3624 
3625 	/* memory map the rx ring */
3626 	handle->md.mmapbuflen = req.tp_block_nr * req.tp_block_size;
3627 	handle->md.mmapbuf = mmap(0, handle->md.mmapbuflen,
3628 	    PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
3629 	if (handle->md.mmapbuf == MAP_FAILED) {
3630 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3631 		    "can't mmap rx ring: %s", pcap_strerror(errno));
3632 
3633 		/* clear the allocated ring on error*/
3634 		destroy_ring(handle);
3635 		*status = PCAP_ERROR;
3636 		return -1;
3637 	}
3638 
3639 	/* allocate a ring for each frame header pointer*/
3640 	handle->cc = req.tp_frame_nr;
3641 	handle->buffer = malloc(handle->cc * sizeof(union thdr *));
3642 	if (!handle->buffer) {
3643 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3644 		    "can't allocate ring of frame headers: %s",
3645 		    pcap_strerror(errno));
3646 
3647 		destroy_ring(handle);
3648 		*status = PCAP_ERROR;
3649 		return -1;
3650 	}
3651 
3652 	/* fill the header ring with proper frame ptr*/
3653 	handle->offset = 0;
3654 	for (i=0; i<req.tp_block_nr; ++i) {
3655 		void *base = &handle->md.mmapbuf[i*req.tp_block_size];
3656 		for (j=0; j<frames_per_block; ++j, ++handle->offset) {
3657 			RING_GET_FRAME(handle) = base;
3658 			base += req.tp_frame_size;
3659 		}
3660 	}
3661 
3662 	handle->bufsize = req.tp_frame_size;
3663 	handle->offset = 0;
3664 	return 1;
3665 }
3666 
3667 /* free all ring related resources*/
3668 static void
3669 destroy_ring(pcap_t *handle)
3670 {
3671 	/* tell the kernel to destroy the ring*/
3672 	struct tpacket_req req;
3673 	memset(&req, 0, sizeof(req));
3674 	setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3675 				(void *) &req, sizeof(req));
3676 
3677 	/* if ring is mapped, unmap it*/
3678 	if (handle->md.mmapbuf) {
3679 		/* do not test for mmap failure, as we can't recover from any error */
3680 		munmap(handle->md.mmapbuf, handle->md.mmapbuflen);
3681 		handle->md.mmapbuf = NULL;
3682 	}
3683 }
3684 
3685 /*
3686  * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
3687  * for Linux mmapped capture.
3688  *
3689  * The problem is that pcap_next() and pcap_next_ex() expect the packet
3690  * data handed to the callback to be valid after the callback returns,
3691  * but pcap_read_linux_mmap() has to release that packet as soon as
3692  * the callback returns (otherwise, the kernel thinks there's still
3693  * at least one unprocessed packet available in the ring, so a select()
3694  * will immediately return indicating that there's data to process), so,
3695  * in the callback, we have to make a copy of the packet.
3696  *
3697  * Yes, this means that, if the capture is using the ring buffer, using
3698  * pcap_next() or pcap_next_ex() requires more copies than using
3699  * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
3700  * pcap_next() or pcap_next_ex().
3701  */
3702 static void
3703 pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
3704     const u_char *bytes)
3705 {
3706 	struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
3707 
3708 	*sp->hdr = *h;
3709 	memcpy(sp->pd->md.oneshot_buffer, bytes, h->caplen);
3710 	*sp->pkt = sp->pd->md.oneshot_buffer;
3711 }
3712 
3713 static void
3714 pcap_cleanup_linux_mmap( pcap_t *handle )
3715 {
3716 	destroy_ring(handle);
3717 	if (handle->md.oneshot_buffer != NULL) {
3718 		free(handle->md.oneshot_buffer);
3719 		handle->md.oneshot_buffer = NULL;
3720 	}
3721 	pcap_cleanup_linux(handle);
3722 }
3723 
3724 
3725 static int
3726 pcap_getnonblock_mmap(pcap_t *p, char *errbuf)
3727 {
3728 	/* use negative value of timeout to indicate non blocking ops */
3729 	return (p->md.timeout<0);
3730 }
3731 
3732 static int
3733 pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
3734 {
3735 	/* map each value to the corresponding 2's complement, to
3736 	 * preserve the timeout value provided with pcap_set_timeout */
3737 	if (nonblock) {
3738 		if (p->md.timeout >= 0) {
3739 			/*
3740 			 * Timeout is non-negative, so we're not already
3741 			 * in non-blocking mode; set it to the 2's
3742 			 * complement, to make it negative, as an
3743 			 * indication that we're in non-blocking mode.
3744 			 */
3745 			p->md.timeout = p->md.timeout*-1 - 1;
3746 		}
3747 	} else {
3748 		if (p->md.timeout < 0) {
3749 			/*
3750 			 * Timeout is negative, so we're not already
3751 			 * in blocking mode; reverse the previous
3752 			 * operation, to make the timeout non-negative
3753 			 * again.
3754 			 */
3755 			p->md.timeout = (p->md.timeout+1)*-1;
3756 		}
3757 	}
3758 	return 0;
3759 }
3760 
3761 static inline union thdr *
3762 pcap_get_ring_frame(pcap_t *handle, int status)
3763 {
3764 	union thdr h;
3765 
3766 	h.raw = RING_GET_FRAME(handle);
3767 	switch (handle->md.tp_version) {
3768 	case TPACKET_V1:
3769 		if (status != (h.h1->tp_status ? TP_STATUS_USER :
3770 						TP_STATUS_KERNEL))
3771 			return NULL;
3772 		break;
3773 #ifdef HAVE_TPACKET2
3774 	case TPACKET_V2:
3775 		if (status != (h.h2->tp_status ? TP_STATUS_USER :
3776 						TP_STATUS_KERNEL))
3777 			return NULL;
3778 		break;
3779 #endif
3780 	}
3781 	return h.raw;
3782 }
3783 
3784 #ifndef POLLRDHUP
3785 #define POLLRDHUP 0
3786 #endif
3787 
3788 static int
3789 pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
3790 		u_char *user)
3791 {
3792 	int timeout;
3793 	int pkts = 0;
3794 	char c;
3795 
3796 	/* wait for frames availability.*/
3797 	if (!pcap_get_ring_frame(handle, TP_STATUS_USER)) {
3798 		struct pollfd pollinfo;
3799 		int ret;
3800 
3801 		pollinfo.fd = handle->fd;
3802 		pollinfo.events = POLLIN;
3803 
3804 		if (handle->md.timeout == 0)
3805 			timeout = -1;	/* block forever */
3806 		else if (handle->md.timeout > 0)
3807 			timeout = handle->md.timeout;	/* block for that amount of time */
3808 		else
3809 			timeout = 0;	/* non-blocking mode - poll to pick up errors */
3810 		do {
3811 			ret = poll(&pollinfo, 1, timeout);
3812 			if (ret < 0 && errno != EINTR) {
3813 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3814 					"can't poll on packet socket: %s",
3815 					pcap_strerror(errno));
3816 				return PCAP_ERROR;
3817 			} else if (ret > 0 &&
3818 			    (pollinfo.revents & (POLLHUP|POLLRDHUP|POLLERR|POLLNVAL))) {
3819 				/*
3820 				 * There's some indication other than
3821 				 * "you can read on this descriptor" on
3822 				 * the descriptor.
3823 				 */
3824 				if (pollinfo.revents & (POLLHUP | POLLRDHUP)) {
3825 					snprintf(handle->errbuf,
3826 						PCAP_ERRBUF_SIZE,
3827 						"Hangup on packet socket");
3828 					return PCAP_ERROR;
3829 				}
3830 				if (pollinfo.revents & POLLERR) {
3831 					/*
3832 					 * A recv() will give us the
3833 					 * actual error code.
3834 					 *
3835 					 * XXX - make the socket non-blocking?
3836 					 */
3837 					if (recv(handle->fd, &c, sizeof c,
3838 					    MSG_PEEK) != -1)
3839 						continue;	/* what, no error? */
3840 					if (errno == ENETDOWN) {
3841 						/*
3842 						 * The device on which we're
3843 						 * capturing went away.
3844 						 *
3845 						 * XXX - we should really return
3846 						 * PCAP_ERROR_IFACE_NOT_UP,
3847 						 * but pcap_dispatch() etc.
3848 						 * aren't defined to return
3849 						 * that.
3850 						 */
3851 						snprintf(handle->errbuf,
3852 							PCAP_ERRBUF_SIZE,
3853 							"The interface went down");
3854 					} else {
3855 						snprintf(handle->errbuf,
3856 							PCAP_ERRBUF_SIZE,
3857 							"Error condition on packet socket: %s",
3858 							strerror(errno));
3859 					}
3860 					return PCAP_ERROR;
3861 				}
3862 				if (pollinfo.revents & POLLNVAL) {
3863 					snprintf(handle->errbuf,
3864 						PCAP_ERRBUF_SIZE,
3865 						"Invalid polling request on packet socket");
3866 					return PCAP_ERROR;
3867 				}
3868   			}
3869 			/* check for break loop condition on interrupted syscall*/
3870 			if (handle->break_loop) {
3871 				handle->break_loop = 0;
3872 				return PCAP_ERROR_BREAK;
3873 			}
3874 		} while (ret < 0);
3875 	}
3876 
3877 	/* non-positive values of max_packets are used to require all
3878 	 * packets currently available in the ring */
3879 	while ((pkts < max_packets) || (max_packets <= 0)) {
3880 		int run_bpf;
3881 		struct sockaddr_ll *sll;
3882 		struct pcap_pkthdr pcaphdr;
3883 		unsigned char *bp;
3884 		union thdr h;
3885 		unsigned int tp_len;
3886 		unsigned int tp_mac;
3887 		unsigned int tp_snaplen;
3888 		unsigned int tp_sec;
3889 		unsigned int tp_usec;
3890 
3891 		h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
3892 		if (!h.raw)
3893 			break;
3894 
3895 		switch (handle->md.tp_version) {
3896 		case TPACKET_V1:
3897 			tp_len	   = h.h1->tp_len;
3898 			tp_mac	   = h.h1->tp_mac;
3899 			tp_snaplen = h.h1->tp_snaplen;
3900 			tp_sec	   = h.h1->tp_sec;
3901 			tp_usec	   = h.h1->tp_usec;
3902 			break;
3903 #ifdef HAVE_TPACKET2
3904 		case TPACKET_V2:
3905 			tp_len	   = h.h2->tp_len;
3906 			tp_mac	   = h.h2->tp_mac;
3907 			tp_snaplen = h.h2->tp_snaplen;
3908 			tp_sec	   = h.h2->tp_sec;
3909 			tp_usec	   = h.h2->tp_nsec / 1000;
3910 			break;
3911 #endif
3912 		default:
3913 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3914 				"unsupported tpacket version %d",
3915 				handle->md.tp_version);
3916 			return -1;
3917 		}
3918 		/* perform sanity check on internal offset. */
3919 		if (tp_mac + tp_snaplen > handle->bufsize) {
3920 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3921 				"corrupted frame on kernel ring mac "
3922 				"offset %d + caplen %d > frame len %d",
3923 				tp_mac, tp_snaplen, handle->bufsize);
3924 			return -1;
3925 		}
3926 
3927 		/* run filter on received packet
3928 		 * If the kernel filtering is enabled we need to run the
3929 		 * filter until all the frames present into the ring
3930 		 * at filter creation time are processed.
3931 		 * In such case md.use_bpf is used as a counter for the
3932 		 * packet we need to filter.
3933 		 * Note: alternatively it could be possible to stop applying
3934 		 * the filter when the ring became empty, but it can possibly
3935 		 * happen a lot later... */
3936 		bp = (unsigned char*)h.raw + tp_mac;
3937 		run_bpf = (!handle->md.use_bpf) ||
3938 			((handle->md.use_bpf>1) && handle->md.use_bpf--);
3939 		if (run_bpf && handle->fcode.bf_insns &&
3940 				(bpf_filter(handle->fcode.bf_insns, bp,
3941 					tp_len, tp_snaplen) == 0))
3942 			goto skip;
3943 
3944 		/*
3945 		 * Do checks based on packet direction.
3946 		 */
3947 		sll = (void *)h.raw + TPACKET_ALIGN(handle->md.tp_hdrlen);
3948 		if (sll->sll_pkttype == PACKET_OUTGOING) {
3949 			/*
3950 			 * Outgoing packet.
3951 			 * If this is from the loopback device, reject it;
3952 			 * we'll see the packet as an incoming packet as well,
3953 			 * and we don't want to see it twice.
3954 			 */
3955 			if (sll->sll_ifindex == handle->md.lo_ifindex)
3956 				goto skip;
3957 
3958 			/*
3959 			 * If the user only wants incoming packets, reject it.
3960 			 */
3961 			if (handle->direction == PCAP_D_IN)
3962 				goto skip;
3963 		} else {
3964 			/*
3965 			 * Incoming packet.
3966 			 * If the user only wants outgoing packets, reject it.
3967 			 */
3968 			if (handle->direction == PCAP_D_OUT)
3969 				goto skip;
3970 		}
3971 
3972 		/* get required packet info from ring header */
3973 		pcaphdr.ts.tv_sec = tp_sec;
3974 		pcaphdr.ts.tv_usec = tp_usec;
3975 		pcaphdr.caplen = tp_snaplen;
3976 		pcaphdr.len = tp_len;
3977 
3978 		/* if required build in place the sll header*/
3979 		if (handle->md.cooked) {
3980 			struct sll_header *hdrp;
3981 
3982 			/*
3983 			 * The kernel should have left us with enough
3984 			 * space for an sll header; back up the packet
3985 			 * data pointer into that space, as that'll be
3986 			 * the beginning of the packet we pass to the
3987 			 * callback.
3988 			 */
3989 			bp -= SLL_HDR_LEN;
3990 
3991 			/*
3992 			 * Let's make sure that's past the end of
3993 			 * the tpacket header, i.e. >=
3994 			 * ((u_char *)thdr + TPACKET_HDRLEN), so we
3995 			 * don't step on the header when we construct
3996 			 * the sll header.
3997 			 */
3998 			if (bp < (u_char *)h.raw +
3999 					   TPACKET_ALIGN(handle->md.tp_hdrlen) +
4000 					   sizeof(struct sockaddr_ll)) {
4001 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4002 					"cooked-mode frame doesn't have room for sll header");
4003 				return -1;
4004 			}
4005 
4006 			/*
4007 			 * OK, that worked; construct the sll header.
4008 			 */
4009 			hdrp = (struct sll_header *)bp;
4010 			hdrp->sll_pkttype = map_packet_type_to_sll_type(
4011 							sll->sll_pkttype);
4012 			hdrp->sll_hatype = htons(sll->sll_hatype);
4013 			hdrp->sll_halen = htons(sll->sll_halen);
4014 			memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
4015 			hdrp->sll_protocol = sll->sll_protocol;
4016 
4017 			/* update packet len */
4018 			pcaphdr.caplen += SLL_HDR_LEN;
4019 			pcaphdr.len += SLL_HDR_LEN;
4020 		}
4021 
4022 #ifdef HAVE_TPACKET2
4023 		if (handle->md.tp_version == TPACKET_V2 && h.h2->tp_vlan_tci &&
4024 		    tp_snaplen >= 2 * ETH_ALEN) {
4025 			struct vlan_tag *tag;
4026 
4027 			bp -= VLAN_TAG_LEN;
4028 			memmove(bp, bp + VLAN_TAG_LEN, 2 * ETH_ALEN);
4029 
4030 			tag = (struct vlan_tag *)(bp + 2 * ETH_ALEN);
4031 			tag->vlan_tpid = htons(ETH_P_8021Q);
4032 			tag->vlan_tci = htons(h.h2->tp_vlan_tci);
4033 
4034 			pcaphdr.caplen += VLAN_TAG_LEN;
4035 			pcaphdr.len += VLAN_TAG_LEN;
4036 		}
4037 #endif
4038 
4039 		/*
4040 		 * The only way to tell the kernel to cut off the
4041 		 * packet at a snapshot length is with a filter program;
4042 		 * if there's no filter program, the kernel won't cut
4043 		 * the packet off.
4044 		 *
4045 		 * Trim the snapshot length to be no longer than the
4046 		 * specified snapshot length.
4047 		 */
4048 		if (pcaphdr.caplen > handle->snapshot)
4049 			pcaphdr.caplen = handle->snapshot;
4050 
4051 		/* pass the packet to the user */
4052 		pkts++;
4053 		callback(user, &pcaphdr, bp);
4054 		handle->md.packets_read++;
4055 
4056 skip:
4057 		/* next packet */
4058 		switch (handle->md.tp_version) {
4059 		case TPACKET_V1:
4060 			h.h1->tp_status = TP_STATUS_KERNEL;
4061 			break;
4062 #ifdef HAVE_TPACKET2
4063 		case TPACKET_V2:
4064 			h.h2->tp_status = TP_STATUS_KERNEL;
4065 			break;
4066 #endif
4067 		}
4068 		if (++handle->offset >= handle->cc)
4069 			handle->offset = 0;
4070 
4071 		/* check for break loop condition*/
4072 		if (handle->break_loop) {
4073 			handle->break_loop = 0;
4074 			return PCAP_ERROR_BREAK;
4075 		}
4076 	}
4077 	return pkts;
4078 }
4079 
4080 static int
4081 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
4082 {
4083 	int n, offset;
4084 	int ret;
4085 
4086 	/*
4087 	 * Don't rewrite "ret" instructions; we don't need to, as
4088 	 * we're not reading packets with recvmsg(), and we don't
4089 	 * want to, as, by not rewriting them, the kernel can avoid
4090 	 * copying extra data.
4091 	 */
4092 	ret = pcap_setfilter_linux_common(handle, filter, 1);
4093 	if (ret < 0)
4094 		return ret;
4095 
4096 	/* if the kernel filter is enabled, we need to apply the filter on
4097 	 * all packets present into the ring. Get an upper bound of their number
4098 	 */
4099 	if (!handle->md.use_bpf)
4100 		return ret;
4101 
4102 	/* walk the ring backward and count the free slot */
4103 	offset = handle->offset;
4104 	if (--handle->offset < 0)
4105 		handle->offset = handle->cc - 1;
4106 	for (n=0; n < handle->cc; ++n) {
4107 		if (--handle->offset < 0)
4108 			handle->offset = handle->cc - 1;
4109 		if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL))
4110 			break;
4111 	}
4112 
4113 	/* be careful to not change current ring position */
4114 	handle->offset = offset;
4115 
4116 	/* store the number of packets currently present in the ring */
4117 	handle->md.use_bpf = 1 + (handle->cc - n);
4118 	return ret;
4119 }
4120 
4121 #endif /* HAVE_PACKET_RING */
4122 
4123 
4124 #ifdef HAVE_PF_PACKET_SOCKETS
4125 /*
4126  *  Return the index of the given device name. Fill ebuf and return
4127  *  -1 on failure.
4128  */
4129 static int
4130 iface_get_id(int fd, const char *device, char *ebuf)
4131 {
4132 	struct ifreq	ifr;
4133 
4134 	memset(&ifr, 0, sizeof(ifr));
4135 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4136 
4137 	if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
4138 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4139 			 "SIOCGIFINDEX: %s", pcap_strerror(errno));
4140 		return -1;
4141 	}
4142 
4143 	return ifr.ifr_ifindex;
4144 }
4145 
4146 /*
4147  *  Bind the socket associated with FD to the given device.
4148  *  Return 1 on success, 0 if we should try a SOCK_PACKET socket,
4149  *  or a PCAP_ERROR_ value on a hard error.
4150  */
4151 static int
4152 iface_bind(int fd, int ifindex, char *ebuf)
4153 {
4154 	struct sockaddr_ll	sll;
4155 	int			err;
4156 	socklen_t		errlen = sizeof(err);
4157 
4158 	memset(&sll, 0, sizeof(sll));
4159 	sll.sll_family		= AF_PACKET;
4160 	sll.sll_ifindex		= ifindex;
4161 	sll.sll_protocol	= htons(ETH_P_ALL);
4162 
4163 	if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
4164 		if (errno == ENETDOWN) {
4165 			/*
4166 			 * Return a "network down" indication, so that
4167 			 * the application can report that rather than
4168 			 * saying we had a mysterious failure and
4169 			 * suggest that they report a problem to the
4170 			 * libpcap developers.
4171 			 */
4172 			return PCAP_ERROR_IFACE_NOT_UP;
4173 		} else {
4174 			snprintf(ebuf, PCAP_ERRBUF_SIZE,
4175 				 "bind: %s", pcap_strerror(errno));
4176 			return PCAP_ERROR;
4177 		}
4178 	}
4179 
4180 	/* Any pending errors, e.g., network is down? */
4181 
4182 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
4183 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4184 			"getsockopt: %s", pcap_strerror(errno));
4185 		return 0;
4186 	}
4187 
4188 	if (err == ENETDOWN) {
4189 		/*
4190 		 * Return a "network down" indication, so that
4191 		 * the application can report that rather than
4192 		 * saying we had a mysterious failure and
4193 		 * suggest that they report a problem to the
4194 		 * libpcap developers.
4195 		 */
4196 		return PCAP_ERROR_IFACE_NOT_UP;
4197 	} else if (err > 0) {
4198 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4199 			"bind: %s", pcap_strerror(err));
4200 		return 0;
4201 	}
4202 
4203 	return 1;
4204 }
4205 
4206 #ifdef IW_MODE_MONITOR
4207 /*
4208  * Check whether the device supports the Wireless Extensions.
4209  * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE
4210  * if the device doesn't even exist.
4211  */
4212 static int
4213 has_wext(int sock_fd, const char *device, char *ebuf)
4214 {
4215 	struct iwreq ireq;
4216 
4217 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4218 	    sizeof ireq.ifr_ifrn.ifrn_name);
4219 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4220 	if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
4221 		return 1;	/* yes */
4222 	snprintf(ebuf, PCAP_ERRBUF_SIZE,
4223 	    "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
4224 	if (errno == ENODEV)
4225 		return PCAP_ERROR_NO_SUCH_DEVICE;
4226 	return 0;
4227 }
4228 
4229 /*
4230  * Per me si va ne la citta dolente,
4231  * Per me si va ne l'etterno dolore,
4232  *	...
4233  * Lasciate ogne speranza, voi ch'intrate.
4234  *
4235  * XXX - airmon-ng does special stuff with the Orinoco driver and the
4236  * wlan-ng driver.
4237  */
4238 typedef enum {
4239 	MONITOR_WEXT,
4240 	MONITOR_HOSTAP,
4241 	MONITOR_PRISM,
4242 	MONITOR_PRISM54,
4243 	MONITOR_ACX100,
4244 	MONITOR_RT2500,
4245 	MONITOR_RT2570,
4246 	MONITOR_RT73,
4247 	MONITOR_RTL8XXX
4248 } monitor_type;
4249 
4250 /*
4251  * Use the Wireless Extensions, if we have them, to try to turn monitor mode
4252  * on if it's not already on.
4253  *
4254  * Returns 1 on success, 0 if we don't support the Wireless Extensions
4255  * on this device, or a PCAP_ERROR_ value if we do support them but
4256  * we weren't able to turn monitor mode on.
4257  */
4258 static int
4259 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
4260 {
4261 	/*
4262 	 * XXX - at least some adapters require non-Wireless Extensions
4263 	 * mechanisms to turn monitor mode on.
4264 	 *
4265 	 * Atheros cards might require that a separate "monitor virtual access
4266 	 * point" be created, with later versions of the madwifi driver.
4267 	 * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode
4268 	 * monitor -bssid", which apparently spits out a line "athN"
4269 	 * where "athN" is the monitor mode device.  To leave monitor
4270 	 * mode, it destroys the monitor mode device.
4271 	 *
4272 	 * Some Intel Centrino adapters might require private ioctls to get
4273 	 * radio headers; the ipw2200 and ipw3945 drivers allow you to
4274 	 * configure a separate "rtapN" interface to capture in monitor
4275 	 * mode without preventing the adapter from operating normally.
4276 	 * (airmon-ng doesn't appear to use that, though.)
4277 	 *
4278 	 * It would be Truly Wonderful if mac80211 and nl80211 cleaned this
4279 	 * up, and if all drivers were converted to mac80211 drivers.
4280 	 *
4281 	 * If interface {if} is a mac80211 driver, the file
4282 	 * /sys/class/net/{if}/phy80211 is a symlink to
4283 	 * /sys/class/ieee80211/{phydev}, for some {phydev}.
4284 	 *
4285 	 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
4286 	 * least, has a "wmaster0" device and a "wlan0" device; the
4287 	 * latter is the one with the IP address.  Both show up in
4288 	 * "tcpdump -D" output.  Capturing on the wmaster0 device
4289 	 * captures with 802.11 headers.
4290 	 *
4291 	 * airmon-ng searches through /sys/class/net for devices named
4292 	 * monN, starting with mon0; as soon as one *doesn't* exist,
4293 	 * it chooses that as the monitor device name.  If the "iw"
4294 	 * command exists, it does "iw dev {if} interface add {monif}
4295 	 * type monitor", where {monif} is the monitor device.  It
4296 	 * then (sigh) sleeps .1 second, and then configures the
4297 	 * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
4298 	 * is a file, it writes {mondev}, without a newline, to that file,
4299 	 * and again (sigh) sleeps .1 second, and then iwconfig's that
4300 	 * device into monitor mode and configures it up.  Otherwise,
4301 	 * you can't do monitor mode.
4302 	 *
4303 	 * All these devices are "glued" together by having the
4304 	 * /sys/class/net/{device}/phy80211 links pointing to the same
4305 	 * place, so, given a wmaster, wlan, or mon device, you can
4306 	 * find the other devices by looking for devices with
4307 	 * the same phy80211 link.
4308 	 *
4309 	 * To turn monitor mode off, delete the monitor interface,
4310 	 * either with "iw dev {monif} interface del" or by sending
4311 	 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
4312 	 *
4313 	 * Note: if you try to create a monitor device named "monN", and
4314 	 * there's already a "monN" device, it fails, as least with
4315 	 * the netlink interface (which is what iw uses), with a return
4316 	 * value of -ENFILE.  (Return values are negative errnos.)  We
4317 	 * could probably use that to find an unused device.
4318 	 */
4319 	int err;
4320 	struct iwreq ireq;
4321 	struct iw_priv_args *priv;
4322 	monitor_type montype;
4323 	int i;
4324 	__u32 cmd;
4325 	struct ifreq ifr;
4326 	int oldflags;
4327 	int args[2];
4328 	int channel;
4329 
4330 	/*
4331 	 * Does this device *support* the Wireless Extensions?
4332 	 */
4333 	err = has_wext(sock_fd, device, handle->errbuf);
4334 	if (err <= 0)
4335 		return err;	/* either it doesn't or the device doesn't even exist */
4336 	/*
4337 	 * Start out assuming we have no private extensions to control
4338 	 * radio metadata.
4339 	 */
4340 	montype = MONITOR_WEXT;
4341 	cmd = 0;
4342 
4343 	/*
4344 	 * Try to get all the Wireless Extensions private ioctls
4345 	 * supported by this device.
4346 	 *
4347 	 * First, get the size of the buffer we need, by supplying no
4348 	 * buffer and a length of 0.  If the device supports private
4349 	 * ioctls, it should return E2BIG, with ireq.u.data.length set
4350 	 * to the length we need.  If it doesn't support them, it should
4351 	 * return EOPNOTSUPP.
4352 	 */
4353 	memset(&ireq, 0, sizeof ireq);
4354 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4355 	    sizeof ireq.ifr_ifrn.ifrn_name);
4356 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4357 	ireq.u.data.pointer = (void *)args;
4358 	ireq.u.data.length = 0;
4359 	ireq.u.data.flags = 0;
4360 	if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
4361 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4362 		    "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!",
4363 		    device);
4364 		return PCAP_ERROR;
4365 	}
4366 	if (errno != EOPNOTSUPP) {
4367 		/*
4368 		 * OK, it's not as if there are no private ioctls.
4369 		 */
4370 		if (errno != E2BIG) {
4371 			/*
4372 			 * Failed.
4373 			 */
4374 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4375 			    "%s: SIOCGIWPRIV: %s", device,
4376 			    pcap_strerror(errno));
4377 			return PCAP_ERROR;
4378 		}
4379 
4380 		/*
4381 		 * OK, try to get the list of private ioctls.
4382 		 */
4383 		priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args));
4384 		if (priv == NULL) {
4385 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4386 			    "malloc: %s", pcap_strerror(errno));
4387 			return PCAP_ERROR;
4388 		}
4389 		ireq.u.data.pointer = (void *)priv;
4390 		if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) {
4391 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4392 			    "%s: SIOCGIWPRIV: %s", device,
4393 			    pcap_strerror(errno));
4394 			free(priv);
4395 			return PCAP_ERROR;
4396 		}
4397 
4398 		/*
4399 		 * Look for private ioctls to turn monitor mode on or, if
4400 		 * monitor mode is on, to set the header type.
4401 		 */
4402 		for (i = 0; i < ireq.u.data.length; i++) {
4403 			if (strcmp(priv[i].name, "monitor_type") == 0) {
4404 				/*
4405 				 * Hostap driver, use this one.
4406 				 * Set monitor mode first.
4407 				 * You can set it to 0 to get DLT_IEEE80211,
4408 				 * 1 to get DLT_PRISM, 2 to get
4409 				 * DLT_IEEE80211_RADIO_AVS, and, with more
4410 				 * recent versions of the driver, 3 to get
4411 				 * DLT_IEEE80211_RADIO.
4412 				 */
4413 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4414 					break;
4415 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4416 					break;
4417 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4418 					break;
4419 				montype = MONITOR_HOSTAP;
4420 				cmd = priv[i].cmd;
4421 				break;
4422 			}
4423 			if (strcmp(priv[i].name, "set_prismhdr") == 0) {
4424 				/*
4425 				 * Prism54 driver, use this one.
4426 				 * Set monitor mode first.
4427 				 * You can set it to 2 to get DLT_IEEE80211
4428 				 * or 3 or get DLT_PRISM.
4429 				 */
4430 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4431 					break;
4432 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4433 					break;
4434 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4435 					break;
4436 				montype = MONITOR_PRISM54;
4437 				cmd = priv[i].cmd;
4438 				break;
4439 			}
4440 			if (strcmp(priv[i].name, "forceprismheader") == 0) {
4441 				/*
4442 				 * RT2570 driver, use this one.
4443 				 * Do this after turning monitor mode on.
4444 				 * You can set it to 1 to get DLT_PRISM or 2
4445 				 * to get DLT_IEEE80211.
4446 				 */
4447 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4448 					break;
4449 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4450 					break;
4451 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4452 					break;
4453 				montype = MONITOR_RT2570;
4454 				cmd = priv[i].cmd;
4455 				break;
4456 			}
4457 			if (strcmp(priv[i].name, "forceprism") == 0) {
4458 				/*
4459 				 * RT73 driver, use this one.
4460 				 * Do this after turning monitor mode on.
4461 				 * Its argument is a *string*; you can
4462 				 * set it to "1" to get DLT_PRISM or "2"
4463 				 * to get DLT_IEEE80211.
4464 				 */
4465 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR)
4466 					break;
4467 				if (priv[i].set_args & IW_PRIV_SIZE_FIXED)
4468 					break;
4469 				montype = MONITOR_RT73;
4470 				cmd = priv[i].cmd;
4471 				break;
4472 			}
4473 			if (strcmp(priv[i].name, "prismhdr") == 0) {
4474 				/*
4475 				 * One of the RTL8xxx drivers, use this one.
4476 				 * It can only be done after monitor mode
4477 				 * has been turned on.  You can set it to 1
4478 				 * to get DLT_PRISM or 0 to get DLT_IEEE80211.
4479 				 */
4480 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4481 					break;
4482 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4483 					break;
4484 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4485 					break;
4486 				montype = MONITOR_RTL8XXX;
4487 				cmd = priv[i].cmd;
4488 				break;
4489 			}
4490 			if (strcmp(priv[i].name, "rfmontx") == 0) {
4491 				/*
4492 				 * RT2500 or RT61 driver, use this one.
4493 				 * It has one one-byte parameter; set
4494 				 * u.data.length to 1 and u.data.pointer to
4495 				 * point to the parameter.
4496 				 * It doesn't itself turn monitor mode on.
4497 				 * You can set it to 1 to allow transmitting
4498 				 * in monitor mode(?) and get DLT_IEEE80211,
4499 				 * or set it to 0 to disallow transmitting in
4500 				 * monitor mode(?) and get DLT_PRISM.
4501 				 */
4502 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4503 					break;
4504 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2)
4505 					break;
4506 				montype = MONITOR_RT2500;
4507 				cmd = priv[i].cmd;
4508 				break;
4509 			}
4510 			if (strcmp(priv[i].name, "monitor") == 0) {
4511 				/*
4512 				 * Either ACX100 or hostap, use this one.
4513 				 * It turns monitor mode on.
4514 				 * If it takes two arguments, it's ACX100;
4515 				 * the first argument is 1 for DLT_PRISM
4516 				 * or 2 for DLT_IEEE80211, and the second
4517 				 * argument is the channel on which to
4518 				 * run.  If it takes one argument, it's
4519 				 * HostAP, and the argument is 2 for
4520 				 * DLT_IEEE80211 and 3 for DLT_PRISM.
4521 				 *
4522 				 * If we see this, we don't quit, as this
4523 				 * might be a version of the hostap driver
4524 				 * that also supports "monitor_type".
4525 				 */
4526 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4527 					break;
4528 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4529 					break;
4530 				switch (priv[i].set_args & IW_PRIV_SIZE_MASK) {
4531 
4532 				case 1:
4533 					montype = MONITOR_PRISM;
4534 					cmd = priv[i].cmd;
4535 					break;
4536 
4537 				case 2:
4538 					montype = MONITOR_ACX100;
4539 					cmd = priv[i].cmd;
4540 					break;
4541 
4542 				default:
4543 					break;
4544 				}
4545 			}
4546 		}
4547 		free(priv);
4548 	}
4549 
4550 	/*
4551 	 * XXX - ipw3945?  islism?
4552 	 */
4553 
4554 	/*
4555 	 * Get the old mode.
4556 	 */
4557 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4558 	    sizeof ireq.ifr_ifrn.ifrn_name);
4559 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4560 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) {
4561 		/*
4562 		 * We probably won't be able to set the mode, either.
4563 		 */
4564 		return PCAP_ERROR_RFMON_NOTSUP;
4565 	}
4566 
4567 	/*
4568 	 * Is it currently in monitor mode?
4569 	 */
4570 	if (ireq.u.mode == IW_MODE_MONITOR) {
4571 		/*
4572 		 * Yes.  Just leave things as they are.
4573 		 * We don't offer multiple link-layer types, as
4574 		 * changing the link-layer type out from under
4575 		 * somebody else capturing in monitor mode would
4576 		 * be considered rude.
4577 		 */
4578 		return 1;
4579 	}
4580 	/*
4581 	 * No.  We have to put the adapter into rfmon mode.
4582 	 */
4583 
4584 	/*
4585 	 * If we haven't already done so, arrange to have
4586 	 * "pcap_close_all()" called when we exit.
4587 	 */
4588 	if (!pcap_do_addexit(handle)) {
4589 		/*
4590 		 * "atexit()" failed; don't put the interface
4591 		 * in rfmon mode, just give up.
4592 		 */
4593 		return PCAP_ERROR_RFMON_NOTSUP;
4594 	}
4595 
4596 	/*
4597 	 * Save the old mode.
4598 	 */
4599 	handle->md.oldmode = ireq.u.mode;
4600 
4601 	/*
4602 	 * Put the adapter in rfmon mode.  How we do this depends
4603 	 * on whether we have a special private ioctl or not.
4604 	 */
4605 	if (montype == MONITOR_PRISM) {
4606 		/*
4607 		 * We have the "monitor" private ioctl, but none of
4608 		 * the other private ioctls.  Use this, and select
4609 		 * the Prism header.
4610 		 *
4611 		 * If it fails, just fall back on SIOCSIWMODE.
4612 		 */
4613 		memset(&ireq, 0, sizeof ireq);
4614 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4615 		    sizeof ireq.ifr_ifrn.ifrn_name);
4616 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4617 		ireq.u.data.length = 1;	/* 1 argument */
4618 		args[0] = 3;	/* request Prism header */
4619 		memcpy(ireq.u.name, args, IFNAMSIZ);
4620 		if (ioctl(sock_fd, cmd, &ireq) != -1) {
4621 			/*
4622 			 * Success.
4623 			 * Note that we have to put the old mode back
4624 			 * when we close the device.
4625 			 */
4626 			handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
4627 
4628 			/*
4629 			 * Add this to the list of pcaps to close
4630 			 * when we exit.
4631 			 */
4632 			pcap_add_to_pcaps_to_close(handle);
4633 
4634 			return 1;
4635 		}
4636 
4637 		/*
4638 		 * Failure.  Fall back on SIOCSIWMODE.
4639 		 */
4640 	}
4641 
4642 	/*
4643 	 * First, take the interface down if it's up; otherwise, we
4644 	 * might get EBUSY.
4645 	 */
4646 	memset(&ifr, 0, sizeof(ifr));
4647 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4648 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
4649 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4650 		    "%s: Can't get flags: %s", device, strerror(errno));
4651 		return PCAP_ERROR;
4652 	}
4653 	oldflags = 0;
4654 	if (ifr.ifr_flags & IFF_UP) {
4655 		oldflags = ifr.ifr_flags;
4656 		ifr.ifr_flags &= ~IFF_UP;
4657 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4658 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4659 			    "%s: Can't set flags: %s", device, strerror(errno));
4660 			return PCAP_ERROR;
4661 		}
4662 	}
4663 
4664 	/*
4665 	 * Then turn monitor mode on.
4666 	 */
4667 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4668 	    sizeof ireq.ifr_ifrn.ifrn_name);
4669 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4670 	ireq.u.mode = IW_MODE_MONITOR;
4671 	if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) {
4672 		/*
4673 		 * Scientist, you've failed.
4674 		 * Bring the interface back up if we shut it down.
4675 		 */
4676 		ifr.ifr_flags = oldflags;
4677 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4678 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4679 			    "%s: Can't set flags: %s", device, strerror(errno));
4680 			return PCAP_ERROR;
4681 		}
4682 		return PCAP_ERROR_RFMON_NOTSUP;
4683 	}
4684 
4685 	/*
4686 	 * XXX - airmon-ng does "iwconfig {if} key off" after setting
4687 	 * monitor mode and setting the channel, and then does
4688 	 * "iwconfig up".
4689 	 */
4690 
4691 	/*
4692 	 * Now select the appropriate radio header.
4693 	 */
4694 	switch (montype) {
4695 
4696 	case MONITOR_WEXT:
4697 		/*
4698 		 * We don't have any private ioctl to set the header.
4699 		 */
4700 		break;
4701 
4702 	case MONITOR_HOSTAP:
4703 		/*
4704 		 * Try to select the radiotap header.
4705 		 */
4706 		memset(&ireq, 0, sizeof ireq);
4707 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4708 		    sizeof ireq.ifr_ifrn.ifrn_name);
4709 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4710 		args[0] = 3;	/* request radiotap header */
4711 		memcpy(ireq.u.name, args, sizeof (int));
4712 		if (ioctl(sock_fd, cmd, &ireq) != -1)
4713 			break;	/* success */
4714 
4715 		/*
4716 		 * That failed.  Try to select the AVS header.
4717 		 */
4718 		memset(&ireq, 0, sizeof ireq);
4719 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4720 		    sizeof ireq.ifr_ifrn.ifrn_name);
4721 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4722 		args[0] = 2;	/* request AVS header */
4723 		memcpy(ireq.u.name, args, sizeof (int));
4724 		if (ioctl(sock_fd, cmd, &ireq) != -1)
4725 			break;	/* success */
4726 
4727 		/*
4728 		 * That failed.  Try to select the Prism header.
4729 		 */
4730 		memset(&ireq, 0, sizeof ireq);
4731 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4732 		    sizeof ireq.ifr_ifrn.ifrn_name);
4733 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4734 		args[0] = 1;	/* request Prism header */
4735 		memcpy(ireq.u.name, args, sizeof (int));
4736 		ioctl(sock_fd, cmd, &ireq);
4737 		break;
4738 
4739 	case MONITOR_PRISM:
4740 		/*
4741 		 * The private ioctl failed.
4742 		 */
4743 		break;
4744 
4745 	case MONITOR_PRISM54:
4746 		/*
4747 		 * Select the Prism header.
4748 		 */
4749 		memset(&ireq, 0, sizeof ireq);
4750 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4751 		    sizeof ireq.ifr_ifrn.ifrn_name);
4752 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4753 		args[0] = 3;	/* request Prism header */
4754 		memcpy(ireq.u.name, args, sizeof (int));
4755 		ioctl(sock_fd, cmd, &ireq);
4756 		break;
4757 
4758 	case MONITOR_ACX100:
4759 		/*
4760 		 * Get the current channel.
4761 		 */
4762 		memset(&ireq, 0, sizeof ireq);
4763 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4764 		    sizeof ireq.ifr_ifrn.ifrn_name);
4765 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4766 		if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) {
4767 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4768 			    "%s: SIOCGIWFREQ: %s", device,
4769 			    pcap_strerror(errno));
4770 			return PCAP_ERROR;
4771 		}
4772 		channel = ireq.u.freq.m;
4773 
4774 		/*
4775 		 * Select the Prism header, and set the channel to the
4776 		 * current value.
4777 		 */
4778 		memset(&ireq, 0, sizeof ireq);
4779 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4780 		    sizeof ireq.ifr_ifrn.ifrn_name);
4781 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4782 		args[0] = 1;		/* request Prism header */
4783 		args[1] = channel;	/* set channel */
4784 		memcpy(ireq.u.name, args, 2*sizeof (int));
4785 		ioctl(sock_fd, cmd, &ireq);
4786 		break;
4787 
4788 	case MONITOR_RT2500:
4789 		/*
4790 		 * Disallow transmission - that turns on the
4791 		 * Prism header.
4792 		 */
4793 		memset(&ireq, 0, sizeof ireq);
4794 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4795 		    sizeof ireq.ifr_ifrn.ifrn_name);
4796 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4797 		args[0] = 0;	/* disallow transmitting */
4798 		memcpy(ireq.u.name, args, sizeof (int));
4799 		ioctl(sock_fd, cmd, &ireq);
4800 		break;
4801 
4802 	case MONITOR_RT2570:
4803 		/*
4804 		 * Force the Prism header.
4805 		 */
4806 		memset(&ireq, 0, sizeof ireq);
4807 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4808 		    sizeof ireq.ifr_ifrn.ifrn_name);
4809 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4810 		args[0] = 1;	/* request Prism header */
4811 		memcpy(ireq.u.name, args, sizeof (int));
4812 		ioctl(sock_fd, cmd, &ireq);
4813 		break;
4814 
4815 	case MONITOR_RT73:
4816 		/*
4817 		 * Force the Prism header.
4818 		 */
4819 		memset(&ireq, 0, sizeof ireq);
4820 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4821 		    sizeof ireq.ifr_ifrn.ifrn_name);
4822 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4823 		ireq.u.data.length = 1;	/* 1 argument */
4824 		ireq.u.data.pointer = "1";
4825 		ireq.u.data.flags = 0;
4826 		ioctl(sock_fd, cmd, &ireq);
4827 		break;
4828 
4829 	case MONITOR_RTL8XXX:
4830 		/*
4831 		 * Force the Prism header.
4832 		 */
4833 		memset(&ireq, 0, sizeof ireq);
4834 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4835 		    sizeof ireq.ifr_ifrn.ifrn_name);
4836 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4837 		args[0] = 1;	/* request Prism header */
4838 		memcpy(ireq.u.name, args, sizeof (int));
4839 		ioctl(sock_fd, cmd, &ireq);
4840 		break;
4841 	}
4842 
4843 	/*
4844 	 * Now bring the interface back up if we brought it down.
4845 	 */
4846 	if (oldflags != 0) {
4847 		ifr.ifr_flags = oldflags;
4848 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4849 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4850 			    "%s: Can't set flags: %s", device, strerror(errno));
4851 
4852 			/*
4853 			 * At least try to restore the old mode on the
4854 			 * interface.
4855 			 */
4856 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
4857 				/*
4858 				 * Scientist, you've failed.
4859 				 */
4860 				fprintf(stderr,
4861 				    "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n"
4862 				    "Please adjust manually.\n",
4863 				    strerror(errno));
4864 			}
4865 			return PCAP_ERROR;
4866 		}
4867 	}
4868 
4869 	/*
4870 	 * Note that we have to put the old mode back when we
4871 	 * close the device.
4872 	 */
4873 	handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
4874 
4875 	/*
4876 	 * Add this to the list of pcaps to close when we exit.
4877 	 */
4878 	pcap_add_to_pcaps_to_close(handle);
4879 
4880 	return 1;
4881 }
4882 #endif /* IW_MODE_MONITOR */
4883 
4884 /*
4885  * Try various mechanisms to enter monitor mode.
4886  */
4887 static int
4888 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
4889 {
4890 #if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
4891 	int ret;
4892 #endif
4893 
4894 #ifdef HAVE_LIBNL
4895 	ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
4896 	if (ret < 0)
4897 		return ret;	/* error attempting to do so */
4898 	if (ret == 1)
4899 		return 1;	/* success */
4900 #endif /* HAVE_LIBNL */
4901 
4902 #ifdef IW_MODE_MONITOR
4903 	ret = enter_rfmon_mode_wext(handle, sock_fd, device);
4904 	if (ret < 0)
4905 		return ret;	/* error attempting to do so */
4906 	if (ret == 1)
4907 		return 1;	/* success */
4908 #endif /* IW_MODE_MONITOR */
4909 
4910 	/*
4911 	 * Either none of the mechanisms we know about work or none
4912 	 * of those mechanisms are available, so we can't do monitor
4913 	 * mode.
4914 	 */
4915 	return 0;
4916 }
4917 
4918 /*
4919  * Find out if we have any form of fragmentation/reassembly offloading.
4920  *
4921  * We do so using SIOCETHTOOL checking for various types of offloading;
4922  * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
4923  * of the types of offloading, there's nothing we can do to check, so
4924  * we just say "no, we don't".
4925  */
4926 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
4927 static int
4928 iface_ethtool_ioctl(pcap_t *handle, int cmd, const char *cmdname)
4929 {
4930 	struct ifreq	ifr;
4931 	struct ethtool_value eval;
4932 
4933 	memset(&ifr, 0, sizeof(ifr));
4934 	strncpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name));
4935 	eval.cmd = cmd;
4936 	ifr.ifr_data = (caddr_t)&eval;
4937 	if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
4938 		if (errno == EOPNOTSUPP) {
4939 			/*
4940 			 * OK, let's just return 0, which, in our
4941 			 * case, either means "no, what we're asking
4942 			 * about is not enabled" or "all the flags
4943 			 * are clear (i.e., nothing is enabled)".
4944 			 */
4945 			return 0;
4946 		}
4947 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4948 		    "%s: SIOETHTOOL(%s) ioctl failed: %s", handle->opt.source,
4949 		    cmdname, strerror(errno));
4950 		return -1;
4951 	}
4952 	return eval.data;
4953 }
4954 
4955 static int
4956 iface_get_offload(pcap_t *handle)
4957 {
4958 	int ret;
4959 
4960 #ifdef ETHTOOL_GTSO
4961 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO");
4962 	if (ret == -1)
4963 		return -1;
4964 	if (ret)
4965 		return 1;	/* TCP segmentation offloading on */
4966 #endif
4967 
4968 #ifdef ETHTOOL_GUFO
4969 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO");
4970 	if (ret == -1)
4971 		return -1;
4972 	if (ret)
4973 		return 1;	/* UDP fragmentation offloading on */
4974 #endif
4975 
4976 #ifdef ETHTOOL_GGSO
4977 	/*
4978 	 * XXX - will this cause large unsegmented packets to be
4979 	 * handed to PF_PACKET sockets on transmission?  If not,
4980 	 * this need not be checked.
4981 	 */
4982 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO");
4983 	if (ret == -1)
4984 		return -1;
4985 	if (ret)
4986 		return 1;	/* generic segmentation offloading on */
4987 #endif
4988 
4989 #ifdef ETHTOOL_GFLAGS
4990 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4991 	if (ret == -1)
4992 		return -1;
4993 	if (ret & ETH_FLAG_LRO)
4994 		return 1;	/* large receive offloading on */
4995 #endif
4996 
4997 #ifdef ETHTOOL_GGRO
4998 	/*
4999 	 * XXX - will this cause large reassembled packets to be
5000 	 * handed to PF_PACKET sockets on receipt?  If not,
5001 	 * this need not be checked.
5002 	 */
5003 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO");
5004 	if (ret == -1)
5005 		return -1;
5006 	if (ret)
5007 		return 1;	/* generic (large) receive offloading on */
5008 #endif
5009 
5010 	return 0;
5011 }
5012 #else /* SIOCETHTOOL */
5013 static int
5014 iface_get_offload(pcap_t *handle _U_)
5015 {
5016 	/*
5017 	 * XXX - do we need to get this information if we don't
5018 	 * have the ethtool ioctls?  If so, how do we do that?
5019 	 */
5020 	return 0;
5021 }
5022 #endif /* SIOCETHTOOL */
5023 
5024 #endif /* HAVE_PF_PACKET_SOCKETS */
5025 
5026 /* ===== Functions to interface to the older kernels ================== */
5027 
5028 /*
5029  * Try to open a packet socket using the old kernel interface.
5030  * Returns 1 on success and a PCAP_ERROR_ value on an error.
5031  */
5032 static int
5033 activate_old(pcap_t *handle)
5034 {
5035 	int		arptype;
5036 	struct ifreq	ifr;
5037 	const char	*device = handle->opt.source;
5038 	struct utsname	utsname;
5039 	int		mtu;
5040 
5041 	/* Open the socket */
5042 
5043 	handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
5044 	if (handle->fd == -1) {
5045 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5046 			 "socket: %s", pcap_strerror(errno));
5047 		if (errno == EPERM || errno == EACCES) {
5048 			/*
5049 			 * You don't have permission to open the
5050 			 * socket.
5051 			 */
5052 			return PCAP_ERROR_PERM_DENIED;
5053 		} else {
5054 			/*
5055 			 * Other error.
5056 			 */
5057 			return PCAP_ERROR;
5058 		}
5059 	}
5060 
5061 	/* It worked - we are using the old interface */
5062 	handle->md.sock_packet = 1;
5063 
5064 	/* ...which means we get the link-layer header. */
5065 	handle->md.cooked = 0;
5066 
5067 	/* Bind to the given device */
5068 
5069 	if (strcmp(device, "any") == 0) {
5070 		strncpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
5071 			PCAP_ERRBUF_SIZE);
5072 		return PCAP_ERROR;
5073 	}
5074 	if (iface_bind_old(handle->fd, device, handle->errbuf) == -1)
5075 		return PCAP_ERROR;
5076 
5077 	/*
5078 	 * Try to get the link-layer type.
5079 	 */
5080 	arptype = iface_get_arptype(handle->fd, device, handle->errbuf);
5081 	if (arptype < 0)
5082 		return PCAP_ERROR;
5083 
5084 	/*
5085 	 * Try to find the DLT_ type corresponding to that
5086 	 * link-layer type.
5087 	 */
5088 	map_arphrd_to_dlt(handle, arptype, 0);
5089 	if (handle->linktype == -1) {
5090 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5091 			 "unknown arptype %d", arptype);
5092 		return PCAP_ERROR;
5093 	}
5094 
5095 	/* Go to promisc mode if requested */
5096 
5097 	if (handle->opt.promisc) {
5098 		memset(&ifr, 0, sizeof(ifr));
5099 		strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5100 		if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
5101 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5102 				 "SIOCGIFFLAGS: %s", pcap_strerror(errno));
5103 			return PCAP_ERROR;
5104 		}
5105 		if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
5106 			/*
5107 			 * Promiscuous mode isn't currently on,
5108 			 * so turn it on, and remember that
5109 			 * we should turn it off when the
5110 			 * pcap_t is closed.
5111 			 */
5112 
5113 			/*
5114 			 * If we haven't already done so, arrange
5115 			 * to have "pcap_close_all()" called when
5116 			 * we exit.
5117 			 */
5118 			if (!pcap_do_addexit(handle)) {
5119 				/*
5120 				 * "atexit()" failed; don't put
5121 				 * the interface in promiscuous
5122 				 * mode, just give up.
5123 				 */
5124 				return PCAP_ERROR;
5125 			}
5126 
5127 			ifr.ifr_flags |= IFF_PROMISC;
5128 			if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
5129 			        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5130 					 "SIOCSIFFLAGS: %s",
5131 					 pcap_strerror(errno));
5132 				return PCAP_ERROR;
5133 			}
5134 			handle->md.must_do_on_close |= MUST_CLEAR_PROMISC;
5135 
5136 			/*
5137 			 * Add this to the list of pcaps
5138 			 * to close when we exit.
5139 			 */
5140 			pcap_add_to_pcaps_to_close(handle);
5141 		}
5142 	}
5143 
5144 	/*
5145 	 * Compute the buffer size.
5146 	 *
5147 	 * We're using SOCK_PACKET, so this might be a 2.0[.x]
5148 	 * kernel, and might require special handling - check.
5149 	 */
5150 	if (uname(&utsname) < 0 ||
5151 	    strncmp(utsname.release, "2.0", 3) == 0) {
5152 		/*
5153 		 * Either we couldn't find out what kernel release
5154 		 * this is, or it's a 2.0[.x] kernel.
5155 		 *
5156 		 * In the 2.0[.x] kernel, a "recvfrom()" on
5157 		 * a SOCK_PACKET socket, with MSG_TRUNC set, will
5158 		 * return the number of bytes read, so if we pass
5159 		 * a length based on the snapshot length, it'll
5160 		 * return the number of bytes from the packet
5161 		 * copied to userland, not the actual length
5162 		 * of the packet.
5163 		 *
5164 		 * This means that, for example, the IP dissector
5165 		 * in tcpdump will get handed a packet length less
5166 		 * than the length in the IP header, and will
5167 		 * complain about "truncated-ip".
5168 		 *
5169 		 * So we don't bother trying to copy from the
5170 		 * kernel only the bytes in which we're interested,
5171 		 * but instead copy them all, just as the older
5172 		 * versions of libpcap for Linux did.
5173 		 *
5174 		 * The buffer therefore needs to be big enough to
5175 		 * hold the largest packet we can get from this
5176 		 * device.  Unfortunately, we can't get the MRU
5177 		 * of the network; we can only get the MTU.  The
5178 		 * MTU may be too small, in which case a packet larger
5179 		 * than the buffer size will be truncated *and* we
5180 		 * won't get the actual packet size.
5181 		 *
5182 		 * However, if the snapshot length is larger than
5183 		 * the buffer size based on the MTU, we use the
5184 		 * snapshot length as the buffer size, instead;
5185 		 * this means that with a sufficiently large snapshot
5186 		 * length we won't artificially truncate packets
5187 		 * to the MTU-based size.
5188 		 *
5189 		 * This mess just one of many problems with packet
5190 		 * capture on 2.0[.x] kernels; you really want a
5191 		 * 2.2[.x] or later kernel if you want packet capture
5192 		 * to work well.
5193 		 */
5194 		mtu = iface_get_mtu(handle->fd, device, handle->errbuf);
5195 		if (mtu == -1)
5196 			return PCAP_ERROR;
5197 		handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
5198 		if (handle->bufsize < handle->snapshot)
5199 			handle->bufsize = handle->snapshot;
5200 	} else {
5201 		/*
5202 		 * This is a 2.2[.x] or later kernel.
5203 		 *
5204 		 * We can safely pass "recvfrom()" a byte count
5205 		 * based on the snapshot length.
5206 		 */
5207 		handle->bufsize = handle->snapshot;
5208 	}
5209 
5210 	/*
5211 	 * Default value for offset to align link-layer payload
5212 	 * on a 4-byte boundary.
5213 	 */
5214 	handle->offset	 = 0;
5215 
5216 	return 1;
5217 }
5218 
5219 /*
5220  *  Bind the socket associated with FD to the given device using the
5221  *  interface of the old kernels.
5222  */
5223 static int
5224 iface_bind_old(int fd, const char *device, char *ebuf)
5225 {
5226 	struct sockaddr	saddr;
5227 	int		err;
5228 	socklen_t	errlen = sizeof(err);
5229 
5230 	memset(&saddr, 0, sizeof(saddr));
5231 	strncpy(saddr.sa_data, device, sizeof(saddr.sa_data));
5232 	if (bind(fd, &saddr, sizeof(saddr)) == -1) {
5233 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5234 			 "bind: %s", pcap_strerror(errno));
5235 		return -1;
5236 	}
5237 
5238 	/* Any pending errors, e.g., network is down? */
5239 
5240 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
5241 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5242 			"getsockopt: %s", pcap_strerror(errno));
5243 		return -1;
5244 	}
5245 
5246 	if (err > 0) {
5247 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5248 			"bind: %s", pcap_strerror(err));
5249 		return -1;
5250 	}
5251 
5252 	return 0;
5253 }
5254 
5255 
5256 /* ===== System calls available on all supported kernels ============== */
5257 
5258 /*
5259  *  Query the kernel for the MTU of the given interface.
5260  */
5261 static int
5262 iface_get_mtu(int fd, const char *device, char *ebuf)
5263 {
5264 	struct ifreq	ifr;
5265 
5266 	if (!device)
5267 		return BIGGER_THAN_ALL_MTUS;
5268 
5269 	memset(&ifr, 0, sizeof(ifr));
5270 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5271 
5272 	if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
5273 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5274 			 "SIOCGIFMTU: %s", pcap_strerror(errno));
5275 		return -1;
5276 	}
5277 
5278 	return ifr.ifr_mtu;
5279 }
5280 
5281 /*
5282  *  Get the hardware type of the given interface as ARPHRD_xxx constant.
5283  */
5284 static int
5285 iface_get_arptype(int fd, const char *device, char *ebuf)
5286 {
5287 	struct ifreq	ifr;
5288 
5289 	memset(&ifr, 0, sizeof(ifr));
5290 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5291 
5292 	if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
5293 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5294 			 "SIOCGIFHWADDR: %s", pcap_strerror(errno));
5295 		if (errno == ENODEV) {
5296 			/*
5297 			 * No such device.
5298 			 */
5299 			return PCAP_ERROR_NO_SUCH_DEVICE;
5300 		}
5301 		return PCAP_ERROR;
5302 	}
5303 
5304 	return ifr.ifr_hwaddr.sa_family;
5305 }
5306 
5307 #ifdef SO_ATTACH_FILTER
5308 static int
5309 fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
5310 {
5311 	size_t prog_size;
5312 	register int i;
5313 	register struct bpf_insn *p;
5314 	struct bpf_insn *f;
5315 	int len;
5316 
5317 	/*
5318 	 * Make a copy of the filter, and modify that copy if
5319 	 * necessary.
5320 	 */
5321 	prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
5322 	len = handle->fcode.bf_len;
5323 	f = (struct bpf_insn *)malloc(prog_size);
5324 	if (f == NULL) {
5325 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5326 			 "malloc: %s", pcap_strerror(errno));
5327 		return -1;
5328 	}
5329 	memcpy(f, handle->fcode.bf_insns, prog_size);
5330 	fcode->len = len;
5331 	fcode->filter = (struct sock_filter *) f;
5332 
5333 	for (i = 0; i < len; ++i) {
5334 		p = &f[i];
5335 		/*
5336 		 * What type of instruction is this?
5337 		 */
5338 		switch (BPF_CLASS(p->code)) {
5339 
5340 		case BPF_RET:
5341 			/*
5342 			 * It's a return instruction; are we capturing
5343 			 * in memory-mapped mode?
5344 			 */
5345 			if (!is_mmapped) {
5346 				/*
5347 				 * No; is the snapshot length a constant,
5348 				 * rather than the contents of the
5349 				 * accumulator?
5350 				 */
5351 				if (BPF_MODE(p->code) == BPF_K) {
5352 					/*
5353 					 * Yes - if the value to be returned,
5354 					 * i.e. the snapshot length, is
5355 					 * anything other than 0, make it
5356 					 * 65535, so that the packet is
5357 					 * truncated by "recvfrom()",
5358 					 * not by the filter.
5359 					 *
5360 					 * XXX - there's nothing we can
5361 					 * easily do if it's getting the
5362 					 * value from the accumulator; we'd
5363 					 * have to insert code to force
5364 					 * non-zero values to be 65535.
5365 					 */
5366 					if (p->k != 0)
5367 						p->k = 65535;
5368 				}
5369 			}
5370 			break;
5371 
5372 		case BPF_LD:
5373 		case BPF_LDX:
5374 			/*
5375 			 * It's a load instruction; is it loading
5376 			 * from the packet?
5377 			 */
5378 			switch (BPF_MODE(p->code)) {
5379 
5380 			case BPF_ABS:
5381 			case BPF_IND:
5382 			case BPF_MSH:
5383 				/*
5384 				 * Yes; are we in cooked mode?
5385 				 */
5386 				if (handle->md.cooked) {
5387 					/*
5388 					 * Yes, so we need to fix this
5389 					 * instruction.
5390 					 */
5391 					if (fix_offset(p) < 0) {
5392 						/*
5393 						 * We failed to do so.
5394 						 * Return 0, so our caller
5395 						 * knows to punt to userland.
5396 						 */
5397 						return 0;
5398 					}
5399 				}
5400 				break;
5401 			}
5402 			break;
5403 		}
5404 	}
5405 	return 1;	/* we succeeded */
5406 }
5407 
5408 static int
5409 fix_offset(struct bpf_insn *p)
5410 {
5411 	/*
5412 	 * What's the offset?
5413 	 */
5414 	if (p->k >= SLL_HDR_LEN) {
5415 		/*
5416 		 * It's within the link-layer payload; that starts at an
5417 		 * offset of 0, as far as the kernel packet filter is
5418 		 * concerned, so subtract the length of the link-layer
5419 		 * header.
5420 		 */
5421 		p->k -= SLL_HDR_LEN;
5422 	} else if (p->k == 0) {
5423 		/*
5424 		 * It's the packet type field; map it to the special magic
5425 		 * kernel offset for that field.
5426 		 */
5427 		p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
5428 	} else if (p->k == 14) {
5429 		/*
5430 		 * It's the protocol field; map it to the special magic
5431 		 * kernel offset for that field.
5432 		 */
5433 		p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5434 	} else if ((bpf_int32)(p->k) > 0) {
5435 		/*
5436 		 * It's within the header, but it's not one of those
5437 		 * fields; we can't do that in the kernel, so punt
5438 		 * to userland.
5439 		 */
5440 		return -1;
5441 	}
5442 	return 0;
5443 }
5444 
5445 static int
5446 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
5447 {
5448 	int total_filter_on = 0;
5449 	int save_mode;
5450 	int ret;
5451 	int save_errno;
5452 
5453 	/*
5454 	 * The socket filter code doesn't discard all packets queued
5455 	 * up on the socket when the filter is changed; this means
5456 	 * that packets that don't match the new filter may show up
5457 	 * after the new filter is put onto the socket, if those
5458 	 * packets haven't yet been read.
5459 	 *
5460 	 * This means, for example, that if you do a tcpdump capture
5461 	 * with a filter, the first few packets in the capture might
5462 	 * be packets that wouldn't have passed the filter.
5463 	 *
5464 	 * We therefore discard all packets queued up on the socket
5465 	 * when setting a kernel filter.  (This isn't an issue for
5466 	 * userland filters, as the userland filtering is done after
5467 	 * packets are queued up.)
5468 	 *
5469 	 * To flush those packets, we put the socket in read-only mode,
5470 	 * and read packets from the socket until there are no more to
5471 	 * read.
5472 	 *
5473 	 * In order to keep that from being an infinite loop - i.e.,
5474 	 * to keep more packets from arriving while we're draining
5475 	 * the queue - we put the "total filter", which is a filter
5476 	 * that rejects all packets, onto the socket before draining
5477 	 * the queue.
5478 	 *
5479 	 * This code deliberately ignores any errors, so that you may
5480 	 * get bogus packets if an error occurs, rather than having
5481 	 * the filtering done in userland even if it could have been
5482 	 * done in the kernel.
5483 	 */
5484 	if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5485 		       &total_fcode, sizeof(total_fcode)) == 0) {
5486 		char drain[1];
5487 
5488 		/*
5489 		 * Note that we've put the total filter onto the socket.
5490 		 */
5491 		total_filter_on = 1;
5492 
5493 		/*
5494 		 * Save the socket's current mode, and put it in
5495 		 * non-blocking mode; we drain it by reading packets
5496 		 * until we get an error (which is normally a
5497 		 * "nothing more to be read" error).
5498 		 */
5499 		save_mode = fcntl(handle->fd, F_GETFL, 0);
5500 		if (save_mode != -1 &&
5501 		    fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) >= 0) {
5502 			while (recv(handle->fd, &drain, sizeof drain,
5503 			       MSG_TRUNC) >= 0)
5504 				;
5505 			save_errno = errno;
5506 			fcntl(handle->fd, F_SETFL, save_mode);
5507 			if (save_errno != EAGAIN) {
5508 				/* Fatal error */
5509 				reset_kernel_filter(handle);
5510 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5511 				 "recv: %s", pcap_strerror(save_errno));
5512 				return -2;
5513 			}
5514 		}
5515 	}
5516 
5517 	/*
5518 	 * Now attach the new filter.
5519 	 */
5520 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5521 			 fcode, sizeof(*fcode));
5522 	if (ret == -1 && total_filter_on) {
5523 		/*
5524 		 * Well, we couldn't set that filter on the socket,
5525 		 * but we could set the total filter on the socket.
5526 		 *
5527 		 * This could, for example, mean that the filter was
5528 		 * too big to put into the kernel, so we'll have to
5529 		 * filter in userland; in any case, we'll be doing
5530 		 * filtering in userland, so we need to remove the
5531 		 * total filter so we see packets.
5532 		 */
5533 		save_errno = errno;
5534 
5535 		/*
5536 		 * XXX - if this fails, we're really screwed;
5537 		 * we have the total filter on the socket,
5538 		 * and it won't come off.  What do we do then?
5539 		 */
5540 		reset_kernel_filter(handle);
5541 
5542 		errno = save_errno;
5543 	}
5544 	return ret;
5545 }
5546 
5547 static int
5548 reset_kernel_filter(pcap_t *handle)
5549 {
5550 	/*
5551 	 * setsockopt() barfs unless it get a dummy parameter.
5552 	 * valgrind whines unless the value is initialized,
5553 	 * as it has no idea that setsockopt() ignores its
5554 	 * parameter.
5555 	 */
5556 	int dummy = 0;
5557 
5558 	return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
5559 				   &dummy, sizeof(dummy));
5560 }
5561 #endif
5562