xref: /freebsd/contrib/libpcap/pcap-linux.c (revision c243e4902be8df1e643c76b5f18b68bb77cc5268)
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 #ifdef PCAP_SUPPORT_NETFILTER
170 #include "pcap-netfilter-linux.h"
171 #endif
172 
173 /*
174  * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
175  * sockets rather than SOCK_PACKET sockets.
176  *
177  * To use them, we include <linux/if_packet.h> rather than
178  * <netpacket/packet.h>; we do so because
179  *
180  *	some Linux distributions (e.g., Slackware 4.0) have 2.2 or
181  *	later kernels and libc5, and don't provide a <netpacket/packet.h>
182  *	file;
183  *
184  *	not all versions of glibc2 have a <netpacket/packet.h> file
185  *	that defines stuff needed for some of the 2.4-or-later-kernel
186  *	features, so if the system has a 2.4 or later kernel, we
187  *	still can't use those features.
188  *
189  * We're already including a number of other <linux/XXX.h> headers, and
190  * this code is Linux-specific (no other OS has PF_PACKET sockets as
191  * a raw packet capture mechanism), so it's not as if you gain any
192  * useful portability by using <netpacket/packet.h>
193  *
194  * XXX - should we just include <linux/if_packet.h> even if PF_PACKET
195  * isn't defined?  It only defines one data structure in 2.0.x, so
196  * it shouldn't cause any problems.
197  */
198 #ifdef PF_PACKET
199 # include <linux/if_packet.h>
200 
201  /*
202   * On at least some Linux distributions (for example, Red Hat 5.2),
203   * there's no <netpacket/packet.h> file, but PF_PACKET is defined if
204   * you include <sys/socket.h>, but <linux/if_packet.h> doesn't define
205   * any of the PF_PACKET stuff such as "struct sockaddr_ll" or any of
206   * the PACKET_xxx stuff.
207   *
208   * So we check whether PACKET_HOST is defined, and assume that we have
209   * PF_PACKET sockets only if it is defined.
210   */
211 # ifdef PACKET_HOST
212 #  define HAVE_PF_PACKET_SOCKETS
213 #  ifdef PACKET_AUXDATA
214 #   define HAVE_PACKET_AUXDATA
215 #  endif /* PACKET_AUXDATA */
216 # endif /* PACKET_HOST */
217 
218 
219  /* check for memory mapped access avaibility. We assume every needed
220   * struct is defined if the macro TPACKET_HDRLEN is defined, because it
221   * uses many ring related structs and macros */
222 # ifdef TPACKET_HDRLEN
223 #  define HAVE_PACKET_RING
224 #  ifdef TPACKET2_HDRLEN
225 #   define HAVE_TPACKET2
226 #  else
227 #   define TPACKET_V1	0
228 #  endif /* TPACKET2_HDRLEN */
229 # endif /* TPACKET_HDRLEN */
230 #endif /* PF_PACKET */
231 
232 #ifdef SO_ATTACH_FILTER
233 #include <linux/types.h>
234 #include <linux/filter.h>
235 #endif
236 
237 /*
238  * We need linux/sockios.h if we have linux/net_tstamp.h (for time stamp
239  * specification) or linux/ethtool.h (for ethtool ioctls to get offloading
240  * information).
241  */
242 #if defined(HAVE_LINUX_NET_TSTAMP_H) || defined(HAVE_LINUX_ETHTOOL_H)
243 #include <linux/sockios.h>
244 #endif
245 
246 #ifdef HAVE_LINUX_NET_TSTAMP_H
247 #include <linux/net_tstamp.h>
248 #endif
249 
250 /*
251  * Got Wireless Extensions?
252  */
253 #ifdef HAVE_LINUX_WIRELESS_H
254 #include <linux/wireless.h>
255 #endif /* HAVE_LINUX_WIRELESS_H */
256 
257 /*
258  * Got libnl?
259  */
260 #ifdef HAVE_LIBNL
261 #include <linux/nl80211.h>
262 
263 #include <netlink/genl/genl.h>
264 #include <netlink/genl/family.h>
265 #include <netlink/genl/ctrl.h>
266 #include <netlink/msg.h>
267 #include <netlink/attr.h>
268 #endif /* HAVE_LIBNL */
269 
270 /*
271  * Got ethtool support?
272  */
273 #ifdef HAVE_LINUX_ETHTOOL_H
274 #include <linux/ethtool.h>
275 #endif
276 
277 #ifndef HAVE_SOCKLEN_T
278 typedef int		socklen_t;
279 #endif
280 
281 #ifndef MSG_TRUNC
282 /*
283  * This is being compiled on a system that lacks MSG_TRUNC; define it
284  * with the value it has in the 2.2 and later kernels, so that, on
285  * those kernels, when we pass it in the flags argument to "recvfrom()"
286  * we're passing the right value and thus get the MSG_TRUNC behavior
287  * we want.  (We don't get that behavior on 2.0[.x] kernels, because
288  * they didn't support MSG_TRUNC.)
289  */
290 #define MSG_TRUNC	0x20
291 #endif
292 
293 #ifndef SOL_PACKET
294 /*
295  * This is being compiled on a system that lacks SOL_PACKET; define it
296  * with the value it has in the 2.2 and later kernels, so that we can
297  * set promiscuous mode in the good modern way rather than the old
298  * 2.0-kernel crappy way.
299  */
300 #define SOL_PACKET	263
301 #endif
302 
303 #define MAX_LINKHEADER_SIZE	256
304 
305 /*
306  * When capturing on all interfaces we use this as the buffer size.
307  * Should be bigger then all MTUs that occur in real life.
308  * 64kB should be enough for now.
309  */
310 #define BIGGER_THAN_ALL_MTUS	(64*1024)
311 
312 /*
313  * Prototypes for internal functions and methods.
314  */
315 static void map_arphrd_to_dlt(pcap_t *, int, int);
316 #ifdef HAVE_PF_PACKET_SOCKETS
317 static short int map_packet_type_to_sll_type(short int);
318 #endif
319 static int pcap_activate_linux(pcap_t *);
320 static int activate_old(pcap_t *);
321 static int activate_new(pcap_t *);
322 static int activate_mmap(pcap_t *, int *);
323 static int pcap_can_set_rfmon_linux(pcap_t *);
324 static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *);
325 static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);
326 static int pcap_inject_linux(pcap_t *, const void *, size_t);
327 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
328 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
329 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
330 static void pcap_cleanup_linux(pcap_t *);
331 
332 union thdr {
333 	struct tpacket_hdr	*h1;
334 	struct tpacket2_hdr	*h2;
335 	void			*raw;
336 };
337 
338 #ifdef HAVE_PACKET_RING
339 #define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset])
340 
341 static void destroy_ring(pcap_t *handle);
342 static int create_ring(pcap_t *handle, int *status);
343 static int prepare_tpacket_socket(pcap_t *handle);
344 static void pcap_cleanup_linux_mmap(pcap_t *);
345 static int pcap_read_linux_mmap(pcap_t *, int, pcap_handler , u_char *);
346 static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *);
347 static int pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf);
348 static int pcap_getnonblock_mmap(pcap_t *p, char *errbuf);
349 static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
350     const u_char *bytes);
351 #endif
352 
353 /*
354  * Wrap some ioctl calls
355  */
356 #ifdef HAVE_PF_PACKET_SOCKETS
357 static int	iface_get_id(int fd, const char *device, char *ebuf);
358 #endif /* HAVE_PF_PACKET_SOCKETS */
359 static int	iface_get_mtu(int fd, const char *device, char *ebuf);
360 static int 	iface_get_arptype(int fd, const char *device, char *ebuf);
361 #ifdef HAVE_PF_PACKET_SOCKETS
362 static int 	iface_bind(int fd, int ifindex, char *ebuf);
363 #ifdef IW_MODE_MONITOR
364 static int	has_wext(int sock_fd, const char *device, char *ebuf);
365 #endif /* IW_MODE_MONITOR */
366 static int	enter_rfmon_mode(pcap_t *handle, int sock_fd,
367     const char *device);
368 #endif /* HAVE_PF_PACKET_SOCKETS */
369 static int	iface_get_offload(pcap_t *handle);
370 static int 	iface_bind_old(int fd, const char *device, char *ebuf);
371 
372 #ifdef SO_ATTACH_FILTER
373 static int	fix_program(pcap_t *handle, struct sock_fprog *fcode,
374     int is_mapped);
375 static int	fix_offset(struct bpf_insn *p);
376 static int	set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
377 static int	reset_kernel_filter(pcap_t *handle);
378 
379 static struct sock_filter	total_insn
380 	= BPF_STMT(BPF_RET | BPF_K, 0);
381 static struct sock_fprog	total_fcode
382 	= { 1, &total_insn };
383 #endif /* SO_ATTACH_FILTER */
384 
385 pcap_t *
386 pcap_create(const char *device, char *ebuf)
387 {
388 	pcap_t *handle;
389 
390 	/*
391 	 * A null device name is equivalent to the "any" device.
392 	 */
393 	if (device == NULL)
394 		device = "any";
395 
396 #ifdef HAVE_DAG_API
397 	if (strstr(device, "dag")) {
398 		return dag_create(device, ebuf);
399 	}
400 #endif /* HAVE_DAG_API */
401 
402 #ifdef HAVE_SEPTEL_API
403 	if (strstr(device, "septel")) {
404 		return septel_create(device, ebuf);
405 	}
406 #endif /* HAVE_SEPTEL_API */
407 
408 #ifdef HAVE_SNF_API
409         handle = snf_create(device, ebuf);
410         if (strstr(device, "snf") || handle != NULL)
411 		return handle;
412 
413 #endif /* HAVE_SNF_API */
414 
415 #ifdef PCAP_SUPPORT_BT
416 	if (strstr(device, "bluetooth")) {
417 		return bt_create(device, ebuf);
418 	}
419 #endif
420 
421 #ifdef PCAP_SUPPORT_CAN
422 	if (strstr(device, "can") || strstr(device, "vcan")) {
423 		return can_create(device, ebuf);
424 	}
425 #endif
426 
427 #ifdef PCAP_SUPPORT_USB
428 	if (strstr(device, "usbmon")) {
429 		return usb_create(device, ebuf);
430 	}
431 #endif
432 
433 #ifdef PCAP_SUPPORT_NETFILTER
434 	if (strncmp(device, "nflog", strlen("nflog")) == 0) {
435 		return nflog_create(device, ebuf);
436 	}
437 #endif
438 
439 	handle = pcap_create_common(device, ebuf);
440 	if (handle == NULL)
441 		return NULL;
442 
443 	handle->activate_op = pcap_activate_linux;
444 	handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
445 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
446 	/*
447 	 * We claim that we support:
448 	 *
449 	 *	software time stamps, with no details about their precision;
450 	 *	hardware time stamps, synced to the host time;
451 	 *	hardware time stamps, not synced to the host time.
452 	 *
453 	 * XXX - we can't ask a device whether it supports
454 	 * hardware time stamps, so we just claim all devices do.
455 	 */
456 	handle->tstamp_type_count = 3;
457 	handle->tstamp_type_list = malloc(3 * sizeof(u_int));
458 	if (handle->tstamp_type_list == NULL) {
459 		free(handle);
460 		return NULL;
461 	}
462 	handle->tstamp_type_list[0] = PCAP_TSTAMP_HOST;
463 	handle->tstamp_type_list[1] = PCAP_TSTAMP_ADAPTER;
464 	handle->tstamp_type_list[2] = PCAP_TSTAMP_ADAPTER_UNSYNCED;
465 #endif
466 
467 	return handle;
468 }
469 
470 #ifdef HAVE_LIBNL
471 /*
472  * If interface {if} is a mac80211 driver, the file
473  * /sys/class/net/{if}/phy80211 is a symlink to
474  * /sys/class/ieee80211/{phydev}, for some {phydev}.
475  *
476  * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
477  * least, has a "wmaster0" device and a "wlan0" device; the
478  * latter is the one with the IP address.  Both show up in
479  * "tcpdump -D" output.  Capturing on the wmaster0 device
480  * captures with 802.11 headers.
481  *
482  * airmon-ng searches through /sys/class/net for devices named
483  * monN, starting with mon0; as soon as one *doesn't* exist,
484  * it chooses that as the monitor device name.  If the "iw"
485  * command exists, it does "iw dev {if} interface add {monif}
486  * type monitor", where {monif} is the monitor device.  It
487  * then (sigh) sleeps .1 second, and then configures the
488  * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
489  * is a file, it writes {mondev}, without a newline, to that file,
490  * and again (sigh) sleeps .1 second, and then iwconfig's that
491  * device into monitor mode and configures it up.  Otherwise,
492  * you can't do monitor mode.
493  *
494  * All these devices are "glued" together by having the
495  * /sys/class/net/{device}/phy80211 links pointing to the same
496  * place, so, given a wmaster, wlan, or mon device, you can
497  * find the other devices by looking for devices with
498  * the same phy80211 link.
499  *
500  * To turn monitor mode off, delete the monitor interface,
501  * either with "iw dev {monif} interface del" or by sending
502  * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
503  *
504  * Note: if you try to create a monitor device named "monN", and
505  * there's already a "monN" device, it fails, as least with
506  * the netlink interface (which is what iw uses), with a return
507  * value of -ENFILE.  (Return values are negative errnos.)  We
508  * could probably use that to find an unused device.
509  *
510  * Yes, you can have multiple monitor devices for a given
511  * physical device.
512 */
513 
514 /*
515  * Is this a mac80211 device?  If so, fill in the physical device path and
516  * return 1; if not, return 0.  On an error, fill in handle->errbuf and
517  * return PCAP_ERROR.
518  */
519 static int
520 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
521     size_t phydev_max_pathlen)
522 {
523 	char *pathstr;
524 	ssize_t bytes_read;
525 
526 	/*
527 	 * Generate the path string for the symlink to the physical device.
528 	 */
529 	if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
530 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
531 		    "%s: Can't generate path name string for /sys/class/net device",
532 		    device);
533 		return PCAP_ERROR;
534 	}
535 	bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
536 	if (bytes_read == -1) {
537 		if (errno == ENOENT || errno == EINVAL) {
538 			/*
539 			 * Doesn't exist, or not a symlink; assume that
540 			 * means it's not a mac80211 device.
541 			 */
542 			free(pathstr);
543 			return 0;
544 		}
545 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
546 		    "%s: Can't readlink %s: %s", device, pathstr,
547 		    strerror(errno));
548 		free(pathstr);
549 		return PCAP_ERROR;
550 	}
551 	free(pathstr);
552 	phydev_path[bytes_read] = '\0';
553 	return 1;
554 }
555 
556 #ifdef HAVE_LIBNL_2_x
557 #define get_nl_errmsg	nl_geterror
558 #else
559 /* libnl 2.x compatibility code */
560 
561 #define nl_sock nl_handle
562 
563 static inline struct nl_handle *
564 nl_socket_alloc(void)
565 {
566 	return nl_handle_alloc();
567 }
568 
569 static inline void
570 nl_socket_free(struct nl_handle *h)
571 {
572 	nl_handle_destroy(h);
573 }
574 
575 #define get_nl_errmsg	strerror
576 
577 static inline int
578 __genl_ctrl_alloc_cache(struct nl_handle *h, struct nl_cache **cache)
579 {
580 	struct nl_cache *tmp = genl_ctrl_alloc_cache(h);
581 	if (!tmp)
582 		return -ENOMEM;
583 	*cache = tmp;
584 	return 0;
585 }
586 #define genl_ctrl_alloc_cache __genl_ctrl_alloc_cache
587 #endif /* !HAVE_LIBNL_2_x */
588 
589 struct nl80211_state {
590 	struct nl_sock *nl_sock;
591 	struct nl_cache *nl_cache;
592 	struct genl_family *nl80211;
593 };
594 
595 static int
596 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
597 {
598 	int err;
599 
600 	state->nl_sock = nl_socket_alloc();
601 	if (!state->nl_sock) {
602 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
603 		    "%s: failed to allocate netlink handle", device);
604 		return PCAP_ERROR;
605 	}
606 
607 	if (genl_connect(state->nl_sock)) {
608 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
609 		    "%s: failed to connect to generic netlink", device);
610 		goto out_handle_destroy;
611 	}
612 
613 	err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
614 	if (err < 0) {
615 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
616 		    "%s: failed to allocate generic netlink cache: %s",
617 		    device, get_nl_errmsg(-err));
618 		goto out_handle_destroy;
619 	}
620 
621 	state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
622 	if (!state->nl80211) {
623 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
624 		    "%s: nl80211 not found", device);
625 		goto out_cache_free;
626 	}
627 
628 	return 0;
629 
630 out_cache_free:
631 	nl_cache_free(state->nl_cache);
632 out_handle_destroy:
633 	nl_socket_free(state->nl_sock);
634 	return PCAP_ERROR;
635 }
636 
637 static void
638 nl80211_cleanup(struct nl80211_state *state)
639 {
640 	genl_family_put(state->nl80211);
641 	nl_cache_free(state->nl_cache);
642 	nl_socket_free(state->nl_sock);
643 }
644 
645 static int
646 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
647     const char *device, const char *mondevice)
648 {
649 	int ifindex;
650 	struct nl_msg *msg;
651 	int err;
652 
653 	ifindex = iface_get_id(sock_fd, device, handle->errbuf);
654 	if (ifindex == -1)
655 		return PCAP_ERROR;
656 
657 	msg = nlmsg_alloc();
658 	if (!msg) {
659 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
660 		    "%s: failed to allocate netlink msg", device);
661 		return PCAP_ERROR;
662 	}
663 
664 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
665 		    0, NL80211_CMD_NEW_INTERFACE, 0);
666 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
667 	NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
668 	NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
669 
670 	err = nl_send_auto_complete(state->nl_sock, msg);
671 	if (err < 0) {
672 #ifdef HAVE_LIBNL_2_x
673 		if (err == -NLE_FAILURE) {
674 #else
675 		if (err == -ENFILE) {
676 #endif
677 			/*
678 			 * Device not available; our caller should just
679 			 * keep trying.  (libnl 2.x maps ENFILE to
680 			 * NLE_FAILURE; it can also map other errors
681 			 * to that, but there's not much we can do
682 			 * about that.)
683 			 */
684 			nlmsg_free(msg);
685 			return 0;
686 		} else {
687 			/*
688 			 * Real failure, not just "that device is not
689 			 * available.
690 			 */
691 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
692 			    "%s: nl_send_auto_complete failed adding %s interface: %s",
693 			    device, mondevice, get_nl_errmsg(-err));
694 			nlmsg_free(msg);
695 			return PCAP_ERROR;
696 		}
697 	}
698 	err = nl_wait_for_ack(state->nl_sock);
699 	if (err < 0) {
700 #ifdef HAVE_LIBNL_2_x
701 		if (err == -NLE_FAILURE) {
702 #else
703 		if (err == -ENFILE) {
704 #endif
705 			/*
706 			 * Device not available; our caller should just
707 			 * keep trying.  (libnl 2.x maps ENFILE to
708 			 * NLE_FAILURE; it can also map other errors
709 			 * to that, but there's not much we can do
710 			 * about that.)
711 			 */
712 			nlmsg_free(msg);
713 			return 0;
714 		} else {
715 			/*
716 			 * Real failure, not just "that device is not
717 			 * available.
718 			 */
719 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
720 			    "%s: nl_wait_for_ack failed adding %s interface: %s",
721 			    device, mondevice, get_nl_errmsg(-err));
722 			nlmsg_free(msg);
723 			return PCAP_ERROR;
724 		}
725 	}
726 
727 	/*
728 	 * Success.
729 	 */
730 	nlmsg_free(msg);
731 	return 1;
732 
733 nla_put_failure:
734 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
735 	    "%s: nl_put failed adding %s interface",
736 	    device, mondevice);
737 	nlmsg_free(msg);
738 	return PCAP_ERROR;
739 }
740 
741 static int
742 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
743     const char *device, const char *mondevice)
744 {
745 	int ifindex;
746 	struct nl_msg *msg;
747 	int err;
748 
749 	ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
750 	if (ifindex == -1)
751 		return PCAP_ERROR;
752 
753 	msg = nlmsg_alloc();
754 	if (!msg) {
755 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
756 		    "%s: failed to allocate netlink msg", device);
757 		return PCAP_ERROR;
758 	}
759 
760 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
761 		    0, NL80211_CMD_DEL_INTERFACE, 0);
762 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
763 
764 	err = nl_send_auto_complete(state->nl_sock, msg);
765 	if (err < 0) {
766 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
767 		    "%s: nl_send_auto_complete failed deleting %s interface: %s",
768 		    device, mondevice, get_nl_errmsg(-err));
769 		nlmsg_free(msg);
770 		return PCAP_ERROR;
771 	}
772 	err = nl_wait_for_ack(state->nl_sock);
773 	if (err < 0) {
774 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
775 		    "%s: nl_wait_for_ack failed adding %s interface: %s",
776 		    device, mondevice, get_nl_errmsg(-err));
777 		nlmsg_free(msg);
778 		return PCAP_ERROR;
779 	}
780 
781 	/*
782 	 * Success.
783 	 */
784 	nlmsg_free(msg);
785 	return 1;
786 
787 nla_put_failure:
788 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
789 	    "%s: nl_put failed deleting %s interface",
790 	    device, mondevice);
791 	nlmsg_free(msg);
792 	return PCAP_ERROR;
793 }
794 
795 static int
796 enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
797 {
798 	int ret;
799 	char phydev_path[PATH_MAX+1];
800 	struct nl80211_state nlstate;
801 	struct ifreq ifr;
802 	u_int n;
803 
804 	/*
805 	 * Is this a mac80211 device?
806 	 */
807 	ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
808 	if (ret < 0)
809 		return ret;	/* error */
810 	if (ret == 0)
811 		return 0;	/* no error, but not mac80211 device */
812 
813 	/*
814 	 * XXX - is this already a monN device?
815 	 * If so, we're done.
816 	 * Is that determined by old Wireless Extensions ioctls?
817 	 */
818 
819 	/*
820 	 * OK, it's apparently a mac80211 device.
821 	 * Try to find an unused monN device for it.
822 	 */
823 	ret = nl80211_init(handle, &nlstate, device);
824 	if (ret != 0)
825 		return ret;
826 	for (n = 0; n < UINT_MAX; n++) {
827 		/*
828 		 * Try mon{n}.
829 		 */
830 		char mondevice[3+10+1];	/* mon{UINT_MAX}\0 */
831 
832 		snprintf(mondevice, sizeof mondevice, "mon%u", n);
833 		ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
834 		if (ret == 1) {
835 			handle->md.mondevice = strdup(mondevice);
836 			goto added;
837 		}
838 		if (ret < 0) {
839 			/*
840 			 * Hard failure.  Just return ret; handle->errbuf
841 			 * has already been set.
842 			 */
843 			nl80211_cleanup(&nlstate);
844 			return ret;
845 		}
846 	}
847 
848 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
849 	    "%s: No free monN interfaces", device);
850 	nl80211_cleanup(&nlstate);
851 	return PCAP_ERROR;
852 
853 added:
854 
855 #if 0
856 	/*
857 	 * Sleep for .1 seconds.
858 	 */
859 	delay.tv_sec = 0;
860 	delay.tv_nsec = 500000000;
861 	nanosleep(&delay, NULL);
862 #endif
863 
864 	/*
865 	 * If we haven't already done so, arrange to have
866 	 * "pcap_close_all()" called when we exit.
867 	 */
868 	if (!pcap_do_addexit(handle)) {
869 		/*
870 		 * "atexit()" failed; don't put the interface
871 		 * in rfmon mode, just give up.
872 		 */
873 		return PCAP_ERROR_RFMON_NOTSUP;
874 	}
875 
876 	/*
877 	 * Now configure the monitor interface up.
878 	 */
879 	memset(&ifr, 0, sizeof(ifr));
880 	strncpy(ifr.ifr_name, handle->md.mondevice, sizeof(ifr.ifr_name));
881 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
882 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
883 		    "%s: Can't get flags for %s: %s", device,
884 		    handle->md.mondevice, strerror(errno));
885 		del_mon_if(handle, sock_fd, &nlstate, device,
886 		    handle->md.mondevice);
887 		nl80211_cleanup(&nlstate);
888 		return PCAP_ERROR;
889 	}
890 	ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
891 	if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
892 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
893 		    "%s: Can't set flags for %s: %s", device,
894 		    handle->md.mondevice, strerror(errno));
895 		del_mon_if(handle, sock_fd, &nlstate, device,
896 		    handle->md.mondevice);
897 		nl80211_cleanup(&nlstate);
898 		return PCAP_ERROR;
899 	}
900 
901 	/*
902 	 * Success.  Clean up the libnl state.
903 	 */
904 	nl80211_cleanup(&nlstate);
905 
906 	/*
907 	 * Note that we have to delete the monitor device when we close
908 	 * the handle.
909 	 */
910 	handle->md.must_do_on_close |= MUST_DELETE_MONIF;
911 
912 	/*
913 	 * Add this to the list of pcaps to close when we exit.
914 	 */
915 	pcap_add_to_pcaps_to_close(handle);
916 
917 	return 1;
918 }
919 #endif /* HAVE_LIBNL */
920 
921 static int
922 pcap_can_set_rfmon_linux(pcap_t *handle)
923 {
924 #ifdef HAVE_LIBNL
925 	char phydev_path[PATH_MAX+1];
926 	int ret;
927 #endif
928 #ifdef IW_MODE_MONITOR
929 	int sock_fd;
930 	struct iwreq ireq;
931 #endif
932 
933 	if (strcmp(handle->opt.source, "any") == 0) {
934 		/*
935 		 * Monitor mode makes no sense on the "any" device.
936 		 */
937 		return 0;
938 	}
939 
940 #ifdef HAVE_LIBNL
941 	/*
942 	 * Bleah.  There doesn't seem to be a way to ask a mac80211
943 	 * device, through libnl, whether it supports monitor mode;
944 	 * we'll just check whether the device appears to be a
945 	 * mac80211 device and, if so, assume the device supports
946 	 * monitor mode.
947 	 *
948 	 * wmaster devices don't appear to support the Wireless
949 	 * Extensions, but we can create a mon device for a
950 	 * wmaster device, so we don't bother checking whether
951 	 * a mac80211 device supports the Wireless Extensions.
952 	 */
953 	ret = get_mac80211_phydev(handle, handle->opt.source, phydev_path,
954 	    PATH_MAX);
955 	if (ret < 0)
956 		return ret;	/* error */
957 	if (ret == 1)
958 		return 1;	/* mac80211 device */
959 #endif
960 
961 #ifdef IW_MODE_MONITOR
962 	/*
963 	 * Bleah.  There doesn't appear to be an ioctl to use to ask
964 	 * whether a device supports monitor mode; we'll just do
965 	 * SIOCGIWMODE and, if it succeeds, assume the device supports
966 	 * monitor mode.
967 	 *
968 	 * Open a socket on which to attempt to get the mode.
969 	 * (We assume that if we have Wireless Extensions support
970 	 * we also have PF_PACKET support.)
971 	 */
972 	sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
973 	if (sock_fd == -1) {
974 		(void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
975 		    "socket: %s", pcap_strerror(errno));
976 		return PCAP_ERROR;
977 	}
978 
979 	/*
980 	 * Attempt to get the current mode.
981 	 */
982 	strncpy(ireq.ifr_ifrn.ifrn_name, handle->opt.source,
983 	    sizeof ireq.ifr_ifrn.ifrn_name);
984 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
985 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
986 		/*
987 		 * Well, we got the mode; assume we can set it.
988 		 */
989 		close(sock_fd);
990 		return 1;
991 	}
992 	if (errno == ENODEV) {
993 		/* The device doesn't even exist. */
994 		(void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
995 		    "SIOCGIWMODE failed: %s", pcap_strerror(errno));
996 		close(sock_fd);
997 		return PCAP_ERROR_NO_SUCH_DEVICE;
998 	}
999 	close(sock_fd);
1000 #endif
1001 	return 0;
1002 }
1003 
1004 /*
1005  * Grabs the number of dropped packets by the interface from /proc/net/dev.
1006  *
1007  * XXX - what about /sys/class/net/{interface name}/rx_*?  There are
1008  * individual devices giving, in ASCII, various rx_ and tx_ statistics.
1009  *
1010  * Or can we get them in binary form from netlink?
1011  */
1012 static long int
1013 linux_if_drops(const char * if_name)
1014 {
1015 	char buffer[512];
1016 	char * bufptr;
1017 	FILE * file;
1018 	int field_to_convert = 3, if_name_sz = strlen(if_name);
1019 	long int dropped_pkts = 0;
1020 
1021 	file = fopen("/proc/net/dev", "r");
1022 	if (!file)
1023 		return 0;
1024 
1025 	while (!dropped_pkts && fgets( buffer, sizeof(buffer), file ))
1026 	{
1027 		/* 	search for 'bytes' -- if its in there, then
1028 			that means we need to grab the fourth field. otherwise
1029 			grab the third field. */
1030 		if (field_to_convert != 4 && strstr(buffer, "bytes"))
1031 		{
1032 			field_to_convert = 4;
1033 			continue;
1034 		}
1035 
1036 		/* find iface and make sure it actually matches -- space before the name and : after it */
1037 		if ((bufptr = strstr(buffer, if_name)) &&
1038 			(bufptr == buffer || *(bufptr-1) == ' ') &&
1039 			*(bufptr + if_name_sz) == ':')
1040 		{
1041 			bufptr = bufptr + if_name_sz + 1;
1042 
1043 			/* grab the nth field from it */
1044 			while( --field_to_convert && *bufptr != '\0')
1045 			{
1046 				while (*bufptr != '\0' && *(bufptr++) == ' ');
1047 				while (*bufptr != '\0' && *(bufptr++) != ' ');
1048 			}
1049 
1050 			/* get rid of any final spaces */
1051 			while (*bufptr != '\0' && *bufptr == ' ') bufptr++;
1052 
1053 			if (*bufptr != '\0')
1054 				dropped_pkts = strtol(bufptr, NULL, 10);
1055 
1056 			break;
1057 		}
1058 	}
1059 
1060 	fclose(file);
1061 	return dropped_pkts;
1062 }
1063 
1064 
1065 /*
1066  * With older kernels promiscuous mode is kind of interesting because we
1067  * have to reset the interface before exiting. The problem can't really
1068  * be solved without some daemon taking care of managing usage counts.
1069  * If we put the interface into promiscuous mode, we set a flag indicating
1070  * that we must take it out of that mode when the interface is closed,
1071  * and, when closing the interface, if that flag is set we take it out
1072  * of promiscuous mode.
1073  *
1074  * Even with newer kernels, we have the same issue with rfmon mode.
1075  */
1076 
1077 static void	pcap_cleanup_linux( pcap_t *handle )
1078 {
1079 	struct ifreq	ifr;
1080 #ifdef HAVE_LIBNL
1081 	struct nl80211_state nlstate;
1082 	int ret;
1083 #endif /* HAVE_LIBNL */
1084 #ifdef IW_MODE_MONITOR
1085 	int oldflags;
1086 	struct iwreq ireq;
1087 #endif /* IW_MODE_MONITOR */
1088 
1089 	if (handle->md.must_do_on_close != 0) {
1090 		/*
1091 		 * There's something we have to do when closing this
1092 		 * pcap_t.
1093 		 */
1094 		if (handle->md.must_do_on_close & MUST_CLEAR_PROMISC) {
1095 			/*
1096 			 * We put the interface into promiscuous mode;
1097 			 * take it out of promiscuous mode.
1098 			 *
1099 			 * XXX - if somebody else wants it in promiscuous
1100 			 * mode, this code cannot know that, so it'll take
1101 			 * it out of promiscuous mode.  That's not fixable
1102 			 * in 2.0[.x] kernels.
1103 			 */
1104 			memset(&ifr, 0, sizeof(ifr));
1105 			strncpy(ifr.ifr_name, handle->md.device,
1106 			    sizeof(ifr.ifr_name));
1107 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
1108 				fprintf(stderr,
1109 				    "Can't restore interface %s flags (SIOCGIFFLAGS failed: %s).\n"
1110 				    "Please adjust manually.\n"
1111 				    "Hint: This can't happen with Linux >= 2.2.0.\n",
1112 				    handle->md.device, strerror(errno));
1113 			} else {
1114 				if (ifr.ifr_flags & IFF_PROMISC) {
1115 					/*
1116 					 * Promiscuous mode is currently on;
1117 					 * turn it off.
1118 					 */
1119 					ifr.ifr_flags &= ~IFF_PROMISC;
1120 					if (ioctl(handle->fd, SIOCSIFFLAGS,
1121 					    &ifr) == -1) {
1122 						fprintf(stderr,
1123 						    "Can't restore interface %s flags (SIOCSIFFLAGS failed: %s).\n"
1124 						    "Please adjust manually.\n"
1125 						    "Hint: This can't happen with Linux >= 2.2.0.\n",
1126 						    handle->md.device,
1127 						    strerror(errno));
1128 					}
1129 				}
1130 			}
1131 		}
1132 
1133 #ifdef HAVE_LIBNL
1134 		if (handle->md.must_do_on_close & MUST_DELETE_MONIF) {
1135 			ret = nl80211_init(handle, &nlstate, handle->md.device);
1136 			if (ret >= 0) {
1137 				ret = del_mon_if(handle, handle->fd, &nlstate,
1138 				    handle->md.device, handle->md.mondevice);
1139 				nl80211_cleanup(&nlstate);
1140 			}
1141 			if (ret < 0) {
1142 				fprintf(stderr,
1143 				    "Can't delete monitor interface %s (%s).\n"
1144 				    "Please delete manually.\n",
1145 				    handle->md.mondevice, handle->errbuf);
1146 			}
1147 		}
1148 #endif /* HAVE_LIBNL */
1149 
1150 #ifdef IW_MODE_MONITOR
1151 		if (handle->md.must_do_on_close & MUST_CLEAR_RFMON) {
1152 			/*
1153 			 * We put the interface into rfmon mode;
1154 			 * take it out of rfmon mode.
1155 			 *
1156 			 * XXX - if somebody else wants it in rfmon
1157 			 * mode, this code cannot know that, so it'll take
1158 			 * it out of rfmon mode.
1159 			 */
1160 
1161 			/*
1162 			 * First, take the interface down if it's up;
1163 			 * otherwise, we might get EBUSY.
1164 			 * If we get errors, just drive on and print
1165 			 * a warning if we can't restore the mode.
1166 			 */
1167 			oldflags = 0;
1168 			memset(&ifr, 0, sizeof(ifr));
1169 			strncpy(ifr.ifr_name, handle->md.device,
1170 			    sizeof(ifr.ifr_name));
1171 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) != -1) {
1172 				if (ifr.ifr_flags & IFF_UP) {
1173 					oldflags = ifr.ifr_flags;
1174 					ifr.ifr_flags &= ~IFF_UP;
1175 					if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1)
1176 						oldflags = 0;	/* didn't set, don't restore */
1177 				}
1178 			}
1179 
1180 			/*
1181 			 * Now restore the mode.
1182 			 */
1183 			strncpy(ireq.ifr_ifrn.ifrn_name, handle->md.device,
1184 			    sizeof ireq.ifr_ifrn.ifrn_name);
1185 			ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1]
1186 			    = 0;
1187 			ireq.u.mode = handle->md.oldmode;
1188 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
1189 				/*
1190 				 * Scientist, you've failed.
1191 				 */
1192 				fprintf(stderr,
1193 				    "Can't restore interface %s wireless mode (SIOCSIWMODE failed: %s).\n"
1194 				    "Please adjust manually.\n",
1195 				    handle->md.device, strerror(errno));
1196 			}
1197 
1198 			/*
1199 			 * Now bring the interface back up if we brought
1200 			 * it down.
1201 			 */
1202 			if (oldflags != 0) {
1203 				ifr.ifr_flags = oldflags;
1204 				if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
1205 					fprintf(stderr,
1206 					    "Can't bring interface %s back up (SIOCSIFFLAGS failed: %s).\n"
1207 					    "Please adjust manually.\n",
1208 					    handle->md.device, strerror(errno));
1209 				}
1210 			}
1211 		}
1212 #endif /* IW_MODE_MONITOR */
1213 
1214 		/*
1215 		 * Take this pcap out of the list of pcaps for which we
1216 		 * have to take the interface out of some mode.
1217 		 */
1218 		pcap_remove_from_pcaps_to_close(handle);
1219 	}
1220 
1221 	if (handle->md.mondevice != NULL) {
1222 		free(handle->md.mondevice);
1223 		handle->md.mondevice = NULL;
1224 	}
1225 	if (handle->md.device != NULL) {
1226 		free(handle->md.device);
1227 		handle->md.device = NULL;
1228 	}
1229 	pcap_cleanup_live_common(handle);
1230 }
1231 
1232 /*
1233  *  Get a handle for a live capture from the given device. You can
1234  *  pass NULL as device to get all packages (without link level
1235  *  information of course). If you pass 1 as promisc the interface
1236  *  will be set to promiscous mode (XXX: I think this usage should
1237  *  be deprecated and functions be added to select that later allow
1238  *  modification of that values -- Torsten).
1239  */
1240 static int
1241 pcap_activate_linux(pcap_t *handle)
1242 {
1243 	const char	*device;
1244 	int		status = 0;
1245 
1246 	device = handle->opt.source;
1247 
1248 	handle->inject_op = pcap_inject_linux;
1249 	handle->setfilter_op = pcap_setfilter_linux;
1250 	handle->setdirection_op = pcap_setdirection_linux;
1251 	handle->set_datalink_op = NULL;	/* can't change data link type */
1252 	handle->getnonblock_op = pcap_getnonblock_fd;
1253 	handle->setnonblock_op = pcap_setnonblock_fd;
1254 	handle->cleanup_op = pcap_cleanup_linux;
1255 	handle->read_op = pcap_read_linux;
1256 	handle->stats_op = pcap_stats_linux;
1257 
1258 	/*
1259 	 * The "any" device is a special device which causes us not
1260 	 * to bind to a particular device and thus to look at all
1261 	 * devices.
1262 	 */
1263 	if (strcmp(device, "any") == 0) {
1264 		if (handle->opt.promisc) {
1265 			handle->opt.promisc = 0;
1266 			/* Just a warning. */
1267 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1268 			    "Promiscuous mode not supported on the \"any\" device");
1269 			status = PCAP_WARNING_PROMISC_NOTSUP;
1270 		}
1271 	}
1272 
1273 	handle->md.device	= strdup(device);
1274 	if (handle->md.device == NULL) {
1275 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
1276 			 pcap_strerror(errno) );
1277 		return PCAP_ERROR;
1278 	}
1279 
1280 	/*
1281 	 * If we're in promiscuous mode, then we probably want
1282 	 * to see when the interface drops packets too, so get an
1283 	 * initial count from /proc/net/dev
1284 	 */
1285 	if (handle->opt.promisc)
1286 		handle->md.proc_dropped = linux_if_drops(handle->md.device);
1287 
1288 	/*
1289 	 * Current Linux kernels use the protocol family PF_PACKET to
1290 	 * allow direct access to all packets on the network while
1291 	 * older kernels had a special socket type SOCK_PACKET to
1292 	 * implement this feature.
1293 	 * While this old implementation is kind of obsolete we need
1294 	 * to be compatible with older kernels for a while so we are
1295 	 * trying both methods with the newer method preferred.
1296 	 */
1297 	status = activate_new(handle);
1298 	if (status < 0) {
1299 		/*
1300 		 * Fatal error with the new way; just fail.
1301 		 * status has the error return; if it's PCAP_ERROR,
1302 		 * handle->errbuf has been set appropriately.
1303 		 */
1304 		goto fail;
1305 	}
1306 	if (status == 1) {
1307 		/*
1308 		 * Success.
1309 		 * Try to use memory-mapped access.
1310 		 */
1311 		switch (activate_mmap(handle, &status)) {
1312 
1313 		case 1:
1314 			/*
1315 			 * We succeeded.  status has been
1316 			 * set to the status to return,
1317 			 * which might be 0, or might be
1318 			 * a PCAP_WARNING_ value.
1319 			 */
1320 			return status;
1321 
1322 		case 0:
1323 			/*
1324 			 * Kernel doesn't support it - just continue
1325 			 * with non-memory-mapped access.
1326 			 */
1327 			break;
1328 
1329 		case -1:
1330 			/*
1331 			 * We failed to set up to use it, or the kernel
1332 			 * supports it, but we failed to enable it.
1333 			 * status has been set to the error status to
1334 			 * return and, if it's PCAP_ERROR, handle->errbuf
1335 			 * contains the error message.
1336 			 */
1337 			goto fail;
1338 		}
1339 	}
1340 	else if (status == 0) {
1341 		/* Non-fatal error; try old way */
1342 		if ((status = activate_old(handle)) != 1) {
1343 			/*
1344 			 * Both methods to open the packet socket failed.
1345 			 * Tidy up and report our failure (handle->errbuf
1346 			 * is expected to be set by the functions above).
1347 			 */
1348 			goto fail;
1349 		}
1350 	}
1351 
1352 	/*
1353 	 * We set up the socket, but not with memory-mapped access.
1354 	 */
1355 	status = 0;
1356 	if (handle->opt.buffer_size != 0) {
1357 		/*
1358 		 * Set the socket buffer size to the specified value.
1359 		 */
1360 		if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF,
1361 		    &handle->opt.buffer_size,
1362 		    sizeof(handle->opt.buffer_size)) == -1) {
1363 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1364 				 "SO_RCVBUF: %s", pcap_strerror(errno));
1365 			status = PCAP_ERROR;
1366 			goto fail;
1367 		}
1368 	}
1369 
1370 	/* Allocate the buffer */
1371 
1372 	handle->buffer	 = malloc(handle->bufsize + handle->offset);
1373 	if (!handle->buffer) {
1374 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1375 			 "malloc: %s", pcap_strerror(errno));
1376 		status = PCAP_ERROR;
1377 		goto fail;
1378 	}
1379 
1380 	/*
1381 	 * "handle->fd" is a socket, so "select()" and "poll()"
1382 	 * should work on it.
1383 	 */
1384 	handle->selectable_fd = handle->fd;
1385 
1386 	return status;
1387 
1388 fail:
1389 	pcap_cleanup_linux(handle);
1390 	return status;
1391 }
1392 
1393 /*
1394  *  Read at most max_packets from the capture stream and call the callback
1395  *  for each of them. Returns the number of packets handled or -1 if an
1396  *  error occured.
1397  */
1398 static int
1399 pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
1400 {
1401 	/*
1402 	 * Currently, on Linux only one packet is delivered per read,
1403 	 * so we don't loop.
1404 	 */
1405 	return pcap_read_packet(handle, callback, user);
1406 }
1407 
1408 /*
1409  *  Read a packet from the socket calling the handler provided by
1410  *  the user. Returns the number of packets received or -1 if an
1411  *  error occured.
1412  */
1413 static int
1414 pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
1415 {
1416 	u_char			*bp;
1417 	int			offset;
1418 #ifdef HAVE_PF_PACKET_SOCKETS
1419 	struct sockaddr_ll	from;
1420 	struct sll_header	*hdrp;
1421 #else
1422 	struct sockaddr		from;
1423 #endif
1424 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1425 	struct iovec		iov;
1426 	struct msghdr		msg;
1427 	struct cmsghdr		*cmsg;
1428 	union {
1429 		struct cmsghdr	cmsg;
1430 		char		buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
1431 	} cmsg_buf;
1432 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1433 	socklen_t		fromlen;
1434 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1435 	int			packet_len, caplen;
1436 	struct pcap_pkthdr	pcap_header;
1437 
1438 #ifdef HAVE_PF_PACKET_SOCKETS
1439 	/*
1440 	 * If this is a cooked device, leave extra room for a
1441 	 * fake packet header.
1442 	 */
1443 	if (handle->md.cooked)
1444 		offset = SLL_HDR_LEN;
1445 	else
1446 		offset = 0;
1447 #else
1448 	/*
1449 	 * This system doesn't have PF_PACKET sockets, so it doesn't
1450 	 * support cooked devices.
1451 	 */
1452 	offset = 0;
1453 #endif
1454 
1455 	/*
1456 	 * Receive a single packet from the kernel.
1457 	 * We ignore EINTR, as that might just be due to a signal
1458 	 * being delivered - if the signal should interrupt the
1459 	 * loop, the signal handler should call pcap_breakloop()
1460 	 * to set handle->break_loop (we ignore it on other
1461 	 * platforms as well).
1462 	 * We also ignore ENETDOWN, so that we can continue to
1463 	 * capture traffic if the interface goes down and comes
1464 	 * back up again; comments in the kernel indicate that
1465 	 * we'll just block waiting for packets if we try to
1466 	 * receive from a socket that delivered ENETDOWN, and,
1467 	 * if we're using a memory-mapped buffer, we won't even
1468 	 * get notified of "network down" events.
1469 	 */
1470 	bp = handle->buffer + handle->offset;
1471 
1472 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1473 	msg.msg_name		= &from;
1474 	msg.msg_namelen		= sizeof(from);
1475 	msg.msg_iov		= &iov;
1476 	msg.msg_iovlen		= 1;
1477 	msg.msg_control		= &cmsg_buf;
1478 	msg.msg_controllen	= sizeof(cmsg_buf);
1479 	msg.msg_flags		= 0;
1480 
1481 	iov.iov_len		= handle->bufsize - offset;
1482 	iov.iov_base		= bp + offset;
1483 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1484 
1485 	do {
1486 		/*
1487 		 * Has "pcap_breakloop()" been called?
1488 		 */
1489 		if (handle->break_loop) {
1490 			/*
1491 			 * Yes - clear the flag that indicates that it has,
1492 			 * and return PCAP_ERROR_BREAK as an indication that
1493 			 * we were told to break out of the loop.
1494 			 */
1495 			handle->break_loop = 0;
1496 			return PCAP_ERROR_BREAK;
1497 		}
1498 
1499 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1500 		packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC);
1501 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1502 		fromlen = sizeof(from);
1503 		packet_len = recvfrom(
1504 			handle->fd, bp + offset,
1505 			handle->bufsize - offset, MSG_TRUNC,
1506 			(struct sockaddr *) &from, &fromlen);
1507 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1508 	} while (packet_len == -1 && errno == EINTR);
1509 
1510 	/* Check if an error occured */
1511 
1512 	if (packet_len == -1) {
1513 		switch (errno) {
1514 
1515 		case EAGAIN:
1516 			return 0;	/* no packet there */
1517 
1518 		case ENETDOWN:
1519 			/*
1520 			 * The device on which we're capturing went away.
1521 			 *
1522 			 * XXX - we should really return
1523 			 * PCAP_ERROR_IFACE_NOT_UP, but pcap_dispatch()
1524 			 * etc. aren't defined to return that.
1525 			 */
1526 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1527 				"The interface went down");
1528 			return PCAP_ERROR;
1529 
1530 		default:
1531 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1532 				 "recvfrom: %s", pcap_strerror(errno));
1533 			return PCAP_ERROR;
1534 		}
1535 	}
1536 
1537 #ifdef HAVE_PF_PACKET_SOCKETS
1538 	if (!handle->md.sock_packet) {
1539 		/*
1540 		 * Unfortunately, there is a window between socket() and
1541 		 * bind() where the kernel may queue packets from any
1542 		 * interface.  If we're bound to a particular interface,
1543 		 * discard packets not from that interface.
1544 		 *
1545 		 * (If socket filters are supported, we could do the
1546 		 * same thing we do when changing the filter; however,
1547 		 * that won't handle packet sockets without socket
1548 		 * filter support, and it's a bit more complicated.
1549 		 * It would save some instructions per packet, however.)
1550 		 */
1551 		if (handle->md.ifindex != -1 &&
1552 		    from.sll_ifindex != handle->md.ifindex)
1553 			return 0;
1554 
1555 		/*
1556 		 * Do checks based on packet direction.
1557 		 * We can only do this if we're using PF_PACKET; the
1558 		 * address returned for SOCK_PACKET is a "sockaddr_pkt"
1559 		 * which lacks the relevant packet type information.
1560 		 */
1561 		if (from.sll_pkttype == PACKET_OUTGOING) {
1562 			/*
1563 			 * Outgoing packet.
1564 			 * If this is from the loopback device, reject it;
1565 			 * we'll see the packet as an incoming packet as well,
1566 			 * and we don't want to see it twice.
1567 			 */
1568 			if (from.sll_ifindex == handle->md.lo_ifindex)
1569 				return 0;
1570 
1571 			/*
1572 			 * If the user only wants incoming packets, reject it.
1573 			 */
1574 			if (handle->direction == PCAP_D_IN)
1575 				return 0;
1576 		} else {
1577 			/*
1578 			 * Incoming packet.
1579 			 * If the user only wants outgoing packets, reject it.
1580 			 */
1581 			if (handle->direction == PCAP_D_OUT)
1582 				return 0;
1583 		}
1584 	}
1585 #endif
1586 
1587 #ifdef HAVE_PF_PACKET_SOCKETS
1588 	/*
1589 	 * If this is a cooked device, fill in the fake packet header.
1590 	 */
1591 	if (handle->md.cooked) {
1592 		/*
1593 		 * Add the length of the fake header to the length
1594 		 * of packet data we read.
1595 		 */
1596 		packet_len += SLL_HDR_LEN;
1597 
1598 		hdrp = (struct sll_header *)bp;
1599 		hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype);
1600 		hdrp->sll_hatype = htons(from.sll_hatype);
1601 		hdrp->sll_halen = htons(from.sll_halen);
1602 		memcpy(hdrp->sll_addr, from.sll_addr,
1603 		    (from.sll_halen > SLL_ADDRLEN) ?
1604 		      SLL_ADDRLEN :
1605 		      from.sll_halen);
1606 		hdrp->sll_protocol = from.sll_protocol;
1607 	}
1608 
1609 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1610 	for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
1611 		struct tpacket_auxdata *aux;
1612 		unsigned int len;
1613 		struct vlan_tag *tag;
1614 
1615 		if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
1616 		    cmsg->cmsg_level != SOL_PACKET ||
1617 		    cmsg->cmsg_type != PACKET_AUXDATA)
1618 			continue;
1619 
1620 		aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
1621 		if (aux->tp_vlan_tci == 0)
1622 			continue;
1623 
1624 		len = packet_len > iov.iov_len ? iov.iov_len : packet_len;
1625 		if (len < 2 * ETH_ALEN)
1626 			break;
1627 
1628 		bp -= VLAN_TAG_LEN;
1629 		memmove(bp, bp + VLAN_TAG_LEN, 2 * ETH_ALEN);
1630 
1631 		tag = (struct vlan_tag *)(bp + 2 * ETH_ALEN);
1632 		tag->vlan_tpid = htons(ETH_P_8021Q);
1633 		tag->vlan_tci = htons(aux->tp_vlan_tci);
1634 
1635 		packet_len += VLAN_TAG_LEN;
1636 	}
1637 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1638 #endif /* HAVE_PF_PACKET_SOCKETS */
1639 
1640 	/*
1641 	 * XXX: According to the kernel source we should get the real
1642 	 * packet len if calling recvfrom with MSG_TRUNC set. It does
1643 	 * not seem to work here :(, but it is supported by this code
1644 	 * anyway.
1645 	 * To be honest the code RELIES on that feature so this is really
1646 	 * broken with 2.2.x kernels.
1647 	 * I spend a day to figure out what's going on and I found out
1648 	 * that the following is happening:
1649 	 *
1650 	 * The packet comes from a random interface and the packet_rcv
1651 	 * hook is called with a clone of the packet. That code inserts
1652 	 * the packet into the receive queue of the packet socket.
1653 	 * If a filter is attached to that socket that filter is run
1654 	 * first - and there lies the problem. The default filter always
1655 	 * cuts the packet at the snaplen:
1656 	 *
1657 	 * # tcpdump -d
1658 	 * (000) ret      #68
1659 	 *
1660 	 * So the packet filter cuts down the packet. The recvfrom call
1661 	 * says "hey, it's only 68 bytes, it fits into the buffer" with
1662 	 * the result that we don't get the real packet length. This
1663 	 * is valid at least until kernel 2.2.17pre6.
1664 	 *
1665 	 * We currently handle this by making a copy of the filter
1666 	 * program, fixing all "ret" instructions with non-zero
1667 	 * operands to have an operand of 65535 so that the filter
1668 	 * doesn't truncate the packet, and supplying that modified
1669 	 * filter to the kernel.
1670 	 */
1671 
1672 	caplen = packet_len;
1673 	if (caplen > handle->snapshot)
1674 		caplen = handle->snapshot;
1675 
1676 	/* Run the packet filter if not using kernel filter */
1677 	if (!handle->md.use_bpf && handle->fcode.bf_insns) {
1678 		if (bpf_filter(handle->fcode.bf_insns, bp,
1679 		                packet_len, caplen) == 0)
1680 		{
1681 			/* rejected by filter */
1682 			return 0;
1683 		}
1684 	}
1685 
1686 	/* Fill in our own header data */
1687 
1688 	if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
1689 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1690 			 "SIOCGSTAMP: %s", pcap_strerror(errno));
1691 		return PCAP_ERROR;
1692 	}
1693 	pcap_header.caplen	= caplen;
1694 	pcap_header.len		= packet_len;
1695 
1696 	/*
1697 	 * Count the packet.
1698 	 *
1699 	 * Arguably, we should count them before we check the filter,
1700 	 * as on many other platforms "ps_recv" counts packets
1701 	 * handed to the filter rather than packets that passed
1702 	 * the filter, but if filtering is done in the kernel, we
1703 	 * can't get a count of packets that passed the filter,
1704 	 * and that would mean the meaning of "ps_recv" wouldn't
1705 	 * be the same on all Linux systems.
1706 	 *
1707 	 * XXX - it's not the same on all systems in any case;
1708 	 * ideally, we should have a "get the statistics" call
1709 	 * that supplies more counts and indicates which of them
1710 	 * it supplies, so that we supply a count of packets
1711 	 * handed to the filter only on platforms where that
1712 	 * information is available.
1713 	 *
1714 	 * We count them here even if we can get the packet count
1715 	 * from the kernel, as we can only determine at run time
1716 	 * whether we'll be able to get it from the kernel (if
1717 	 * HAVE_TPACKET_STATS isn't defined, we can't get it from
1718 	 * the kernel, but if it is defined, the library might
1719 	 * have been built with a 2.4 or later kernel, but we
1720 	 * might be running on a 2.2[.x] kernel without Alexey
1721 	 * Kuznetzov's turbopacket patches, and thus the kernel
1722 	 * might not be able to supply those statistics).  We
1723 	 * could, I guess, try, when opening the socket, to get
1724 	 * the statistics, and if we can not increment the count
1725 	 * here, but it's not clear that always incrementing
1726 	 * the count is more expensive than always testing a flag
1727 	 * in memory.
1728 	 *
1729 	 * We keep the count in "md.packets_read", and use that for
1730 	 * "ps_recv" if we can't get the statistics from the kernel.
1731 	 * We do that because, if we *can* get the statistics from
1732 	 * the kernel, we use "md.stat.ps_recv" and "md.stat.ps_drop"
1733 	 * as running counts, as reading the statistics from the
1734 	 * kernel resets the kernel statistics, and if we directly
1735 	 * increment "md.stat.ps_recv" here, that means it will
1736 	 * count packets *twice* on systems where we can get kernel
1737 	 * statistics - once here, and once in pcap_stats_linux().
1738 	 */
1739 	handle->md.packets_read++;
1740 
1741 	/* Call the user supplied callback function */
1742 	callback(userdata, &pcap_header, bp);
1743 
1744 	return 1;
1745 }
1746 
1747 static int
1748 pcap_inject_linux(pcap_t *handle, const void *buf, size_t size)
1749 {
1750 	int ret;
1751 
1752 #ifdef HAVE_PF_PACKET_SOCKETS
1753 	if (!handle->md.sock_packet) {
1754 		/* PF_PACKET socket */
1755 		if (handle->md.ifindex == -1) {
1756 			/*
1757 			 * We don't support sending on the "any" device.
1758 			 */
1759 			strlcpy(handle->errbuf,
1760 			    "Sending packets isn't supported on the \"any\" device",
1761 			    PCAP_ERRBUF_SIZE);
1762 			return (-1);
1763 		}
1764 
1765 		if (handle->md.cooked) {
1766 			/*
1767 			 * We don't support sending on the "any" device.
1768 			 *
1769 			 * XXX - how do you send on a bound cooked-mode
1770 			 * socket?
1771 			 * Is a "sendto()" required there?
1772 			 */
1773 			strlcpy(handle->errbuf,
1774 			    "Sending packets isn't supported in cooked mode",
1775 			    PCAP_ERRBUF_SIZE);
1776 			return (-1);
1777 		}
1778 	}
1779 #endif
1780 
1781 	ret = send(handle->fd, buf, size, 0);
1782 	if (ret == -1) {
1783 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
1784 		    pcap_strerror(errno));
1785 		return (-1);
1786 	}
1787 	return (ret);
1788 }
1789 
1790 /*
1791  *  Get the statistics for the given packet capture handle.
1792  *  Reports the number of dropped packets iff the kernel supports
1793  *  the PACKET_STATISTICS "getsockopt()" argument (2.4 and later
1794  *  kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket
1795  *  patches); otherwise, that information isn't available, and we lie
1796  *  and report 0 as the count of dropped packets.
1797  */
1798 static int
1799 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
1800 {
1801 #ifdef HAVE_TPACKET_STATS
1802 	struct tpacket_stats kstats;
1803 	socklen_t len = sizeof (struct tpacket_stats);
1804 #endif
1805 
1806 	long if_dropped = 0;
1807 
1808 	/*
1809 	 *	To fill in ps_ifdrop, we parse /proc/net/dev for the number
1810 	 */
1811 	if (handle->opt.promisc)
1812 	{
1813 		if_dropped = handle->md.proc_dropped;
1814 		handle->md.proc_dropped = linux_if_drops(handle->md.device);
1815 		handle->md.stat.ps_ifdrop += (handle->md.proc_dropped - if_dropped);
1816 	}
1817 
1818 #ifdef HAVE_TPACKET_STATS
1819 	/*
1820 	 * Try to get the packet counts from the kernel.
1821 	 */
1822 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
1823 			&kstats, &len) > -1) {
1824 		/*
1825 		 * On systems where the PACKET_STATISTICS "getsockopt()"
1826 		 * argument is supported on PF_PACKET sockets:
1827 		 *
1828 		 *	"ps_recv" counts only packets that *passed* the
1829 		 *	filter, not packets that didn't pass the filter.
1830 		 *	This includes packets later dropped because we
1831 		 *	ran out of buffer space.
1832 		 *
1833 		 *	"ps_drop" counts packets dropped because we ran
1834 		 *	out of buffer space.  It doesn't count packets
1835 		 *	dropped by the interface driver.  It counts only
1836 		 *	packets that passed the filter.
1837 		 *
1838 		 *	See above for ps_ifdrop.
1839 		 *
1840 		 *	Both statistics include packets not yet read from
1841 		 *	the kernel by libpcap, and thus not yet seen by
1842 		 *	the application.
1843 		 *
1844 		 * In "linux/net/packet/af_packet.c", at least in the
1845 		 * 2.4.9 kernel, "tp_packets" is incremented for every
1846 		 * packet that passes the packet filter *and* is
1847 		 * successfully queued on the socket; "tp_drops" is
1848 		 * incremented for every packet dropped because there's
1849 		 * not enough free space in the socket buffer.
1850 		 *
1851 		 * When the statistics are returned for a PACKET_STATISTICS
1852 		 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
1853 		 * so that "tp_packets" counts all packets handed to
1854 		 * the PF_PACKET socket, including packets dropped because
1855 		 * there wasn't room on the socket buffer - but not
1856 		 * including packets that didn't pass the filter.
1857 		 *
1858 		 * In the BSD BPF, the count of received packets is
1859 		 * incremented for every packet handed to BPF, regardless
1860 		 * of whether it passed the filter.
1861 		 *
1862 		 * We can't make "pcap_stats()" work the same on both
1863 		 * platforms, but the best approximation is to return
1864 		 * "tp_packets" as the count of packets and "tp_drops"
1865 		 * as the count of drops.
1866 		 *
1867 		 * Keep a running total because each call to
1868 		 *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
1869 		 * resets the counters to zero.
1870 		 */
1871 		handle->md.stat.ps_recv += kstats.tp_packets;
1872 		handle->md.stat.ps_drop += kstats.tp_drops;
1873 		*stats = handle->md.stat;
1874 		return 0;
1875 	}
1876 	else
1877 	{
1878 		/*
1879 		 * If the error was EOPNOTSUPP, fall through, so that
1880 		 * if you build the library on a system with
1881 		 * "struct tpacket_stats" and run it on a system
1882 		 * that doesn't, it works as it does if the library
1883 		 * is built on a system without "struct tpacket_stats".
1884 		 */
1885 		if (errno != EOPNOTSUPP) {
1886 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1887 			    "pcap_stats: %s", pcap_strerror(errno));
1888 			return -1;
1889 		}
1890 	}
1891 #endif
1892 	/*
1893 	 * On systems where the PACKET_STATISTICS "getsockopt()" argument
1894 	 * is not supported on PF_PACKET sockets:
1895 	 *
1896 	 *	"ps_recv" counts only packets that *passed* the filter,
1897 	 *	not packets that didn't pass the filter.  It does not
1898 	 *	count packets dropped because we ran out of buffer
1899 	 *	space.
1900 	 *
1901 	 *	"ps_drop" is not supported.
1902 	 *
1903 	 *	"ps_ifdrop" is supported. It will return the number
1904 	 *	of drops the interface reports in /proc/net/dev,
1905 	 *	if that is available.
1906 	 *
1907 	 *	"ps_recv" doesn't include packets not yet read from
1908 	 *	the kernel by libpcap.
1909 	 *
1910 	 * We maintain the count of packets processed by libpcap in
1911 	 * "md.packets_read", for reasons described in the comment
1912 	 * at the end of pcap_read_packet().  We have no idea how many
1913 	 * packets were dropped by the kernel buffers -- but we know
1914 	 * how many the interface dropped, so we can return that.
1915 	 */
1916 
1917 	stats->ps_recv = handle->md.packets_read;
1918 	stats->ps_drop = 0;
1919 	stats->ps_ifdrop = handle->md.stat.ps_ifdrop;
1920 	return 0;
1921 }
1922 
1923 /*
1924  * Get from "/sys/class/net" all interfaces listed there; if they're
1925  * already in the list of interfaces we have, that won't add another
1926  * instance, but if they're not, that'll add them.
1927  *
1928  * We don't bother getting any addresses for them; it appears you can't
1929  * use SIOCGIFADDR on Linux to get IPv6 addresses for interfaces, and,
1930  * although some other types of addresses can be fetched with SIOCGIFADDR,
1931  * we don't bother with them for now.
1932  *
1933  * We also don't fail if we couldn't open "/sys/class/net"; we just leave
1934  * the list of interfaces as is, and return 0, so that we can try
1935  * scanning /proc/net/dev.
1936  */
1937 static int
1938 scan_sys_class_net(pcap_if_t **devlistp, char *errbuf)
1939 {
1940 	DIR *sys_class_net_d;
1941 	int fd;
1942 	struct dirent *ent;
1943 	char *p;
1944 	char name[512];	/* XXX - pick a size */
1945 	char *q, *saveq;
1946 	struct ifreq ifrflags;
1947 	int ret = 1;
1948 
1949 	sys_class_net_d = opendir("/sys/class/net");
1950 	if (sys_class_net_d == NULL) {
1951 		/*
1952 		 * Don't fail if it doesn't exist at all.
1953 		 */
1954 		if (errno == ENOENT)
1955 			return (0);
1956 
1957 		/*
1958 		 * Fail if we got some other error.
1959 		 */
1960 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
1961 		    "Can't open /sys/class/net: %s", pcap_strerror(errno));
1962 		return (-1);
1963 	}
1964 
1965 	/*
1966 	 * Create a socket from which to fetch interface information.
1967 	 */
1968 	fd = socket(AF_INET, SOCK_DGRAM, 0);
1969 	if (fd < 0) {
1970 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
1971 		    "socket: %s", pcap_strerror(errno));
1972 		(void)closedir(sys_class_net_d);
1973 		return (-1);
1974 	}
1975 
1976 	for (;;) {
1977 		errno = 0;
1978 		ent = readdir(sys_class_net_d);
1979 		if (ent == NULL) {
1980 			/*
1981 			 * Error or EOF; if errno != 0, it's an error.
1982 			 */
1983 			break;
1984 		}
1985 
1986 		/*
1987 		 * Ignore directories (".", "..", and any subdirectories).
1988 		 */
1989 		if (ent->d_type == DT_DIR)
1990 			continue;
1991 
1992 		/*
1993 		 * Get the interface name.
1994 		 */
1995 		p = &ent->d_name[0];
1996 		q = &name[0];
1997 		while (*p != '\0' && isascii(*p) && !isspace(*p)) {
1998 			if (*p == ':') {
1999 				/*
2000 				 * This could be the separator between a
2001 				 * name and an alias number, or it could be
2002 				 * the separator between a name with no
2003 				 * alias number and the next field.
2004 				 *
2005 				 * If there's a colon after digits, it
2006 				 * separates the name and the alias number,
2007 				 * otherwise it separates the name and the
2008 				 * next field.
2009 				 */
2010 				saveq = q;
2011 				while (isascii(*p) && isdigit(*p))
2012 					*q++ = *p++;
2013 				if (*p != ':') {
2014 					/*
2015 					 * That was the next field,
2016 					 * not the alias number.
2017 					 */
2018 					q = saveq;
2019 				}
2020 				break;
2021 			} else
2022 				*q++ = *p++;
2023 		}
2024 		*q = '\0';
2025 
2026 		/*
2027 		 * Get the flags for this interface, and skip it if
2028 		 * it's not up.
2029 		 */
2030 		strncpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2031 		if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2032 			if (errno == ENXIO || errno == ENODEV)
2033 				continue;
2034 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2035 			    "SIOCGIFFLAGS: %.*s: %s",
2036 			    (int)sizeof(ifrflags.ifr_name),
2037 			    ifrflags.ifr_name,
2038 			    pcap_strerror(errno));
2039 			ret = -1;
2040 			break;
2041 		}
2042 		if (!(ifrflags.ifr_flags & IFF_UP))
2043 			continue;
2044 
2045 		/*
2046 		 * Add an entry for this interface, with no addresses.
2047 		 */
2048 		if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2049 		    errbuf) == -1) {
2050 			/*
2051 			 * Failure.
2052 			 */
2053 			ret = -1;
2054 			break;
2055 		}
2056 	}
2057 	if (ret != -1) {
2058 		/*
2059 		 * Well, we didn't fail for any other reason; did we
2060 		 * fail due to an error reading the directory?
2061 		 */
2062 		if (errno != 0) {
2063 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2064 			    "Error reading /sys/class/net: %s",
2065 			    pcap_strerror(errno));
2066 			ret = -1;
2067 		}
2068 	}
2069 
2070 	(void)close(fd);
2071 	(void)closedir(sys_class_net_d);
2072 	return (ret);
2073 }
2074 
2075 /*
2076  * Get from "/proc/net/dev" all interfaces listed there; if they're
2077  * already in the list of interfaces we have, that won't add another
2078  * instance, but if they're not, that'll add them.
2079  *
2080  * See comments from scan_sys_class_net().
2081  */
2082 static int
2083 scan_proc_net_dev(pcap_if_t **devlistp, char *errbuf)
2084 {
2085 	FILE *proc_net_f;
2086 	int fd;
2087 	char linebuf[512];
2088 	int linenum;
2089 	char *p;
2090 	char name[512];	/* XXX - pick a size */
2091 	char *q, *saveq;
2092 	struct ifreq ifrflags;
2093 	int ret = 0;
2094 
2095 	proc_net_f = fopen("/proc/net/dev", "r");
2096 	if (proc_net_f == NULL) {
2097 		/*
2098 		 * Don't fail if it doesn't exist at all.
2099 		 */
2100 		if (errno == ENOENT)
2101 			return (0);
2102 
2103 		/*
2104 		 * Fail if we got some other error.
2105 		 */
2106 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2107 		    "Can't open /proc/net/dev: %s", pcap_strerror(errno));
2108 		return (-1);
2109 	}
2110 
2111 	/*
2112 	 * Create a socket from which to fetch interface information.
2113 	 */
2114 	fd = socket(AF_INET, SOCK_DGRAM, 0);
2115 	if (fd < 0) {
2116 		(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2117 		    "socket: %s", pcap_strerror(errno));
2118 		(void)fclose(proc_net_f);
2119 		return (-1);
2120 	}
2121 
2122 	for (linenum = 1;
2123 	    fgets(linebuf, sizeof linebuf, proc_net_f) != NULL; linenum++) {
2124 		/*
2125 		 * Skip the first two lines - they're headers.
2126 		 */
2127 		if (linenum <= 2)
2128 			continue;
2129 
2130 		p = &linebuf[0];
2131 
2132 		/*
2133 		 * Skip leading white space.
2134 		 */
2135 		while (*p != '\0' && isascii(*p) && isspace(*p))
2136 			p++;
2137 		if (*p == '\0' || *p == '\n')
2138 			continue;	/* blank line */
2139 
2140 		/*
2141 		 * Get the interface name.
2142 		 */
2143 		q = &name[0];
2144 		while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2145 			if (*p == ':') {
2146 				/*
2147 				 * This could be the separator between a
2148 				 * name and an alias number, or it could be
2149 				 * the separator between a name with no
2150 				 * alias number and the next field.
2151 				 *
2152 				 * If there's a colon after digits, it
2153 				 * separates the name and the alias number,
2154 				 * otherwise it separates the name and the
2155 				 * next field.
2156 				 */
2157 				saveq = q;
2158 				while (isascii(*p) && isdigit(*p))
2159 					*q++ = *p++;
2160 				if (*p != ':') {
2161 					/*
2162 					 * That was the next field,
2163 					 * not the alias number.
2164 					 */
2165 					q = saveq;
2166 				}
2167 				break;
2168 			} else
2169 				*q++ = *p++;
2170 		}
2171 		*q = '\0';
2172 
2173 		/*
2174 		 * Get the flags for this interface, and skip it if
2175 		 * it's not up.
2176 		 */
2177 		strncpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2178 		if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2179 			if (errno == ENXIO)
2180 				continue;
2181 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2182 			    "SIOCGIFFLAGS: %.*s: %s",
2183 			    (int)sizeof(ifrflags.ifr_name),
2184 			    ifrflags.ifr_name,
2185 			    pcap_strerror(errno));
2186 			ret = -1;
2187 			break;
2188 		}
2189 		if (!(ifrflags.ifr_flags & IFF_UP))
2190 			continue;
2191 
2192 		/*
2193 		 * Add an entry for this interface, with no addresses.
2194 		 */
2195 		if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2196 		    errbuf) == -1) {
2197 			/*
2198 			 * Failure.
2199 			 */
2200 			ret = -1;
2201 			break;
2202 		}
2203 	}
2204 	if (ret != -1) {
2205 		/*
2206 		 * Well, we didn't fail for any other reason; did we
2207 		 * fail due to an error reading the file?
2208 		 */
2209 		if (ferror(proc_net_f)) {
2210 			(void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2211 			    "Error reading /proc/net/dev: %s",
2212 			    pcap_strerror(errno));
2213 			ret = -1;
2214 		}
2215 	}
2216 
2217 	(void)close(fd);
2218 	(void)fclose(proc_net_f);
2219 	return (ret);
2220 }
2221 
2222 /*
2223  * Description string for the "any" device.
2224  */
2225 static const char any_descr[] = "Pseudo-device that captures on all interfaces";
2226 
2227 int
2228 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
2229 {
2230 	int ret;
2231 
2232 	/*
2233 	 * Read "/sys/class/net", and add to the list of interfaces all
2234 	 * interfaces listed there that we don't already have, because,
2235 	 * on Linux, SIOCGIFCONF reports only interfaces with IPv4 addresses,
2236 	 * and even getifaddrs() won't return information about
2237 	 * interfaces with no addresses, so you need to read "/sys/class/net"
2238 	 * to get the names of the rest of the interfaces.
2239 	 */
2240 	ret = scan_sys_class_net(alldevsp, errbuf);
2241 	if (ret == -1)
2242 		return (-1);	/* failed */
2243 	if (ret == 0) {
2244 		/*
2245 		 * No /sys/class/net; try reading /proc/net/dev instead.
2246 		 */
2247 		if (scan_proc_net_dev(alldevsp, errbuf) == -1)
2248 			return (-1);
2249 	}
2250 
2251 	/*
2252 	 * Add the "any" device.
2253 	 */
2254 	if (pcap_add_if(alldevsp, "any", 0, any_descr, errbuf) < 0)
2255 		return (-1);
2256 
2257 #ifdef HAVE_DAG_API
2258 	/*
2259 	 * Add DAG devices.
2260 	 */
2261 	if (dag_platform_finddevs(alldevsp, errbuf) < 0)
2262 		return (-1);
2263 #endif /* HAVE_DAG_API */
2264 
2265 #ifdef HAVE_SEPTEL_API
2266 	/*
2267 	 * Add Septel devices.
2268 	 */
2269 	if (septel_platform_finddevs(alldevsp, errbuf) < 0)
2270 		return (-1);
2271 #endif /* HAVE_SEPTEL_API */
2272 
2273 #ifdef HAVE_SNF_API
2274 	if (snf_platform_finddevs(alldevsp, errbuf) < 0)
2275 		return (-1);
2276 #endif /* HAVE_SNF_API */
2277 
2278 #ifdef PCAP_SUPPORT_BT
2279 	/*
2280 	 * Add Bluetooth devices.
2281 	 */
2282 	if (bt_platform_finddevs(alldevsp, errbuf) < 0)
2283 		return (-1);
2284 #endif
2285 
2286 #ifdef PCAP_SUPPORT_USB
2287 	/*
2288 	 * Add USB devices.
2289 	 */
2290 	if (usb_platform_finddevs(alldevsp, errbuf) < 0)
2291 		return (-1);
2292 #endif
2293 
2294 #ifdef PCAP_SUPPORT_NETFILTER
2295 	/*
2296 	 * Add netfilter devices.
2297 	 */
2298 	if (netfilter_platform_finddevs(alldevsp, errbuf) < 0)
2299 		return (-1);
2300 #endif
2301 
2302 	return (0);
2303 }
2304 
2305 /*
2306  *  Attach the given BPF code to the packet capture device.
2307  */
2308 static int
2309 pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
2310     int is_mmapped)
2311 {
2312 #ifdef SO_ATTACH_FILTER
2313 	struct sock_fprog	fcode;
2314 	int			can_filter_in_kernel;
2315 	int			err = 0;
2316 #endif
2317 
2318 	if (!handle)
2319 		return -1;
2320 	if (!filter) {
2321 	        strncpy(handle->errbuf, "setfilter: No filter specified",
2322 			PCAP_ERRBUF_SIZE);
2323 		return -1;
2324 	}
2325 
2326 	/* Make our private copy of the filter */
2327 
2328 	if (install_bpf_program(handle, filter) < 0)
2329 		/* install_bpf_program() filled in errbuf */
2330 		return -1;
2331 
2332 	/*
2333 	 * Run user level packet filter by default. Will be overriden if
2334 	 * installing a kernel filter succeeds.
2335 	 */
2336 	handle->md.use_bpf = 0;
2337 
2338 	/* Install kernel level filter if possible */
2339 
2340 #ifdef SO_ATTACH_FILTER
2341 #ifdef USHRT_MAX
2342 	if (handle->fcode.bf_len > USHRT_MAX) {
2343 		/*
2344 		 * fcode.len is an unsigned short for current kernel.
2345 		 * I have yet to see BPF-Code with that much
2346 		 * instructions but still it is possible. So for the
2347 		 * sake of correctness I added this check.
2348 		 */
2349 		fprintf(stderr, "Warning: Filter too complex for kernel\n");
2350 		fcode.len = 0;
2351 		fcode.filter = NULL;
2352 		can_filter_in_kernel = 0;
2353 	} else
2354 #endif /* USHRT_MAX */
2355 	{
2356 		/*
2357 		 * Oh joy, the Linux kernel uses struct sock_fprog instead
2358 		 * of struct bpf_program and of course the length field is
2359 		 * of different size. Pointed out by Sebastian
2360 		 *
2361 		 * Oh, and we also need to fix it up so that all "ret"
2362 		 * instructions with non-zero operands have 65535 as the
2363 		 * operand if we're not capturing in memory-mapped modee,
2364 		 * and so that, if we're in cooked mode, all memory-reference
2365 		 * instructions use special magic offsets in references to
2366 		 * the link-layer header and assume that the link-layer
2367 		 * payload begins at 0; "fix_program()" will do that.
2368 		 */
2369 		switch (fix_program(handle, &fcode, is_mmapped)) {
2370 
2371 		case -1:
2372 		default:
2373 			/*
2374 			 * Fatal error; just quit.
2375 			 * (The "default" case shouldn't happen; we
2376 			 * return -1 for that reason.)
2377 			 */
2378 			return -1;
2379 
2380 		case 0:
2381 			/*
2382 			 * The program performed checks that we can't make
2383 			 * work in the kernel.
2384 			 */
2385 			can_filter_in_kernel = 0;
2386 			break;
2387 
2388 		case 1:
2389 			/*
2390 			 * We have a filter that'll work in the kernel.
2391 			 */
2392 			can_filter_in_kernel = 1;
2393 			break;
2394 		}
2395 	}
2396 
2397 	/*
2398 	 * NOTE: at this point, we've set both the "len" and "filter"
2399 	 * fields of "fcode".  As of the 2.6.32.4 kernel, at least,
2400 	 * those are the only members of the "sock_fprog" structure,
2401 	 * so we initialize every member of that structure.
2402 	 *
2403 	 * If there is anything in "fcode" that is not initialized,
2404 	 * it is either a field added in a later kernel, or it's
2405 	 * padding.
2406 	 *
2407 	 * If a new field is added, this code needs to be updated
2408 	 * to set it correctly.
2409 	 *
2410 	 * If there are no other fields, then:
2411 	 *
2412 	 *	if the Linux kernel looks at the padding, it's
2413 	 *	buggy;
2414 	 *
2415 	 *	if the Linux kernel doesn't look at the padding,
2416 	 *	then if some tool complains that we're passing
2417 	 *	uninitialized data to the kernel, then the tool
2418 	 *	is buggy and needs to understand that it's just
2419 	 *	padding.
2420 	 */
2421 	if (can_filter_in_kernel) {
2422 		if ((err = set_kernel_filter(handle, &fcode)) == 0)
2423 		{
2424 			/* Installation succeded - using kernel filter. */
2425 			handle->md.use_bpf = 1;
2426 		}
2427 		else if (err == -1)	/* Non-fatal error */
2428 		{
2429 			/*
2430 			 * Print a warning if we weren't able to install
2431 			 * the filter for a reason other than "this kernel
2432 			 * isn't configured to support socket filters.
2433 			 */
2434 			if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
2435 				fprintf(stderr,
2436 				    "Warning: Kernel filter failed: %s\n",
2437 					pcap_strerror(errno));
2438 			}
2439 		}
2440 	}
2441 
2442 	/*
2443 	 * If we're not using the kernel filter, get rid of any kernel
2444 	 * filter that might've been there before, e.g. because the
2445 	 * previous filter could work in the kernel, or because some other
2446 	 * code attached a filter to the socket by some means other than
2447 	 * calling "pcap_setfilter()".  Otherwise, the kernel filter may
2448 	 * filter out packets that would pass the new userland filter.
2449 	 */
2450 	if (!handle->md.use_bpf)
2451 		reset_kernel_filter(handle);
2452 
2453 	/*
2454 	 * Free up the copy of the filter that was made by "fix_program()".
2455 	 */
2456 	if (fcode.filter != NULL)
2457 		free(fcode.filter);
2458 
2459 	if (err == -2)
2460 		/* Fatal error */
2461 		return -1;
2462 #endif /* SO_ATTACH_FILTER */
2463 
2464 	return 0;
2465 }
2466 
2467 static int
2468 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
2469 {
2470 	return pcap_setfilter_linux_common(handle, filter, 0);
2471 }
2472 
2473 
2474 /*
2475  * Set direction flag: Which packets do we accept on a forwarding
2476  * single device? IN, OUT or both?
2477  */
2478 static int
2479 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
2480 {
2481 #ifdef HAVE_PF_PACKET_SOCKETS
2482 	if (!handle->md.sock_packet) {
2483 		handle->direction = d;
2484 		return 0;
2485 	}
2486 #endif
2487 	/*
2488 	 * We're not using PF_PACKET sockets, so we can't determine
2489 	 * the direction of the packet.
2490 	 */
2491 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2492 	    "Setting direction is not supported on SOCK_PACKET sockets");
2493 	return -1;
2494 }
2495 
2496 #ifdef HAVE_PF_PACKET_SOCKETS
2497 /*
2498  * Map the PACKET_ value to a LINUX_SLL_ value; we
2499  * want the same numerical value to be used in
2500  * the link-layer header even if the numerical values
2501  * for the PACKET_ #defines change, so that programs
2502  * that look at the packet type field will always be
2503  * able to handle DLT_LINUX_SLL captures.
2504  */
2505 static short int
2506 map_packet_type_to_sll_type(short int sll_pkttype)
2507 {
2508 	switch (sll_pkttype) {
2509 
2510 	case PACKET_HOST:
2511 		return htons(LINUX_SLL_HOST);
2512 
2513 	case PACKET_BROADCAST:
2514 		return htons(LINUX_SLL_BROADCAST);
2515 
2516 	case PACKET_MULTICAST:
2517 		return  htons(LINUX_SLL_MULTICAST);
2518 
2519 	case PACKET_OTHERHOST:
2520 		return htons(LINUX_SLL_OTHERHOST);
2521 
2522 	case PACKET_OUTGOING:
2523 		return htons(LINUX_SLL_OUTGOING);
2524 
2525 	default:
2526 		return -1;
2527 	}
2528 }
2529 #endif
2530 
2531 /*
2532  *  Linux uses the ARP hardware type to identify the type of an
2533  *  interface. pcap uses the DLT_xxx constants for this. This
2534  *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
2535  *  constant, as arguments, and sets "handle->linktype" to the
2536  *  appropriate DLT_XXX constant and sets "handle->offset" to
2537  *  the appropriate value (to make "handle->offset" plus link-layer
2538  *  header length be a multiple of 4, so that the link-layer payload
2539  *  will be aligned on a 4-byte boundary when capturing packets).
2540  *  (If the offset isn't set here, it'll be 0; add code as appropriate
2541  *  for cases where it shouldn't be 0.)
2542  *
2543  *  If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
2544  *  in cooked mode; otherwise, we can't use cooked mode, so we have
2545  *  to pick some type that works in raw mode, or fail.
2546  *
2547  *  Sets the link type to -1 if unable to map the type.
2548  */
2549 static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
2550 {
2551 	switch (arptype) {
2552 
2553 	case ARPHRD_ETHER:
2554 		/*
2555 		 * This is (presumably) a real Ethernet capture; give it a
2556 		 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
2557 		 * that an application can let you choose it, in case you're
2558 		 * capturing DOCSIS traffic that a Cisco Cable Modem
2559 		 * Termination System is putting out onto an Ethernet (it
2560 		 * doesn't put an Ethernet header onto the wire, it puts raw
2561 		 * DOCSIS frames out on the wire inside the low-level
2562 		 * Ethernet framing).
2563 		 *
2564 		 * XXX - are there any sorts of "fake Ethernet" that have
2565 		 * ARPHRD_ETHER but that *shouldn't offer DLT_DOCSIS as
2566 		 * a Cisco CMTS won't put traffic onto it or get traffic
2567 		 * bridged onto it?  ISDN is handled in "activate_new()",
2568 		 * as we fall back on cooked mode there; are there any
2569 		 * others?
2570 		 */
2571 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2572 		/*
2573 		 * If that fails, just leave the list empty.
2574 		 */
2575 		if (handle->dlt_list != NULL) {
2576 			handle->dlt_list[0] = DLT_EN10MB;
2577 			handle->dlt_list[1] = DLT_DOCSIS;
2578 			handle->dlt_count = 2;
2579 		}
2580 		/* FALLTHROUGH */
2581 
2582 	case ARPHRD_METRICOM:
2583 	case ARPHRD_LOOPBACK:
2584 		handle->linktype = DLT_EN10MB;
2585 		handle->offset = 2;
2586 		break;
2587 
2588 	case ARPHRD_EETHER:
2589 		handle->linktype = DLT_EN3MB;
2590 		break;
2591 
2592 	case ARPHRD_AX25:
2593 		handle->linktype = DLT_AX25_KISS;
2594 		break;
2595 
2596 	case ARPHRD_PRONET:
2597 		handle->linktype = DLT_PRONET;
2598 		break;
2599 
2600 	case ARPHRD_CHAOS:
2601 		handle->linktype = DLT_CHAOS;
2602 		break;
2603 #ifndef ARPHRD_CAN
2604 #define ARPHRD_CAN 280
2605 #endif
2606 	case ARPHRD_CAN:
2607 		handle->linktype = DLT_CAN_SOCKETCAN;
2608 		break;
2609 
2610 #ifndef ARPHRD_IEEE802_TR
2611 #define ARPHRD_IEEE802_TR 800	/* From Linux 2.4 */
2612 #endif
2613 	case ARPHRD_IEEE802_TR:
2614 	case ARPHRD_IEEE802:
2615 		handle->linktype = DLT_IEEE802;
2616 		handle->offset = 2;
2617 		break;
2618 
2619 	case ARPHRD_ARCNET:
2620 		handle->linktype = DLT_ARCNET_LINUX;
2621 		break;
2622 
2623 #ifndef ARPHRD_FDDI	/* From Linux 2.2.13 */
2624 #define ARPHRD_FDDI	774
2625 #endif
2626 	case ARPHRD_FDDI:
2627 		handle->linktype = DLT_FDDI;
2628 		handle->offset = 3;
2629 		break;
2630 
2631 #ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
2632 #define ARPHRD_ATM 19
2633 #endif
2634 	case ARPHRD_ATM:
2635 		/*
2636 		 * The Classical IP implementation in ATM for Linux
2637 		 * supports both what RFC 1483 calls "LLC Encapsulation",
2638 		 * in which each packet has an LLC header, possibly
2639 		 * with a SNAP header as well, prepended to it, and
2640 		 * what RFC 1483 calls "VC Based Multiplexing", in which
2641 		 * different virtual circuits carry different network
2642 		 * layer protocols, and no header is prepended to packets.
2643 		 *
2644 		 * They both have an ARPHRD_ type of ARPHRD_ATM, so
2645 		 * you can't use the ARPHRD_ type to find out whether
2646 		 * captured packets will have an LLC header, and,
2647 		 * while there's a socket ioctl to *set* the encapsulation
2648 		 * type, there's no ioctl to *get* the encapsulation type.
2649 		 *
2650 		 * This means that
2651 		 *
2652 		 *	programs that dissect Linux Classical IP frames
2653 		 *	would have to check for an LLC header and,
2654 		 *	depending on whether they see one or not, dissect
2655 		 *	the frame as LLC-encapsulated or as raw IP (I
2656 		 *	don't know whether there's any traffic other than
2657 		 *	IP that would show up on the socket, or whether
2658 		 *	there's any support for IPv6 in the Linux
2659 		 *	Classical IP code);
2660 		 *
2661 		 *	filter expressions would have to compile into
2662 		 *	code that checks for an LLC header and does
2663 		 *	the right thing.
2664 		 *
2665 		 * Both of those are a nuisance - and, at least on systems
2666 		 * that support PF_PACKET sockets, we don't have to put
2667 		 * up with those nuisances; instead, we can just capture
2668 		 * in cooked mode.  That's what we'll do, if we can.
2669 		 * Otherwise, we'll just fail.
2670 		 */
2671 		if (cooked_ok)
2672 			handle->linktype = DLT_LINUX_SLL;
2673 		else
2674 			handle->linktype = -1;
2675 		break;
2676 
2677 #ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
2678 #define ARPHRD_IEEE80211 801
2679 #endif
2680 	case ARPHRD_IEEE80211:
2681 		handle->linktype = DLT_IEEE802_11;
2682 		break;
2683 
2684 #ifndef ARPHRD_IEEE80211_PRISM  /* From Linux 2.4.18 */
2685 #define ARPHRD_IEEE80211_PRISM 802
2686 #endif
2687 	case ARPHRD_IEEE80211_PRISM:
2688 		handle->linktype = DLT_PRISM_HEADER;
2689 		break;
2690 
2691 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
2692 #define ARPHRD_IEEE80211_RADIOTAP 803
2693 #endif
2694 	case ARPHRD_IEEE80211_RADIOTAP:
2695 		handle->linktype = DLT_IEEE802_11_RADIO;
2696 		break;
2697 
2698 	case ARPHRD_PPP:
2699 		/*
2700 		 * Some PPP code in the kernel supplies no link-layer
2701 		 * header whatsoever to PF_PACKET sockets; other PPP
2702 		 * code supplies PPP link-layer headers ("syncppp.c");
2703 		 * some PPP code might supply random link-layer
2704 		 * headers (PPP over ISDN - there's code in Ethereal,
2705 		 * for example, to cope with PPP-over-ISDN captures
2706 		 * with which the Ethereal developers have had to cope,
2707 		 * heuristically trying to determine which of the
2708 		 * oddball link-layer headers particular packets have).
2709 		 *
2710 		 * As such, we just punt, and run all PPP interfaces
2711 		 * in cooked mode, if we can; otherwise, we just treat
2712 		 * it as DLT_RAW, for now - if somebody needs to capture,
2713 		 * on a 2.0[.x] kernel, on PPP devices that supply a
2714 		 * link-layer header, they'll have to add code here to
2715 		 * map to the appropriate DLT_ type (possibly adding a
2716 		 * new DLT_ type, if necessary).
2717 		 */
2718 		if (cooked_ok)
2719 			handle->linktype = DLT_LINUX_SLL;
2720 		else {
2721 			/*
2722 			 * XXX - handle ISDN types here?  We can't fall
2723 			 * back on cooked sockets, so we'd have to
2724 			 * figure out from the device name what type of
2725 			 * link-layer encapsulation it's using, and map
2726 			 * that to an appropriate DLT_ value, meaning
2727 			 * we'd map "isdnN" devices to DLT_RAW (they
2728 			 * supply raw IP packets with no link-layer
2729 			 * header) and "isdY" devices to a new DLT_I4L_IP
2730 			 * type that has only an Ethernet packet type as
2731 			 * a link-layer header.
2732 			 *
2733 			 * But sometimes we seem to get random crap
2734 			 * in the link-layer header when capturing on
2735 			 * ISDN devices....
2736 			 */
2737 			handle->linktype = DLT_RAW;
2738 		}
2739 		break;
2740 
2741 #ifndef ARPHRD_CISCO
2742 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
2743 #endif
2744 	case ARPHRD_CISCO:
2745 		handle->linktype = DLT_C_HDLC;
2746 		break;
2747 
2748 	/* Not sure if this is correct for all tunnels, but it
2749 	 * works for CIPE */
2750 	case ARPHRD_TUNNEL:
2751 #ifndef ARPHRD_SIT
2752 #define ARPHRD_SIT 776	/* From Linux 2.2.13 */
2753 #endif
2754 	case ARPHRD_SIT:
2755 	case ARPHRD_CSLIP:
2756 	case ARPHRD_SLIP6:
2757 	case ARPHRD_CSLIP6:
2758 	case ARPHRD_ADAPT:
2759 	case ARPHRD_SLIP:
2760 #ifndef ARPHRD_RAWHDLC
2761 #define ARPHRD_RAWHDLC 518
2762 #endif
2763 	case ARPHRD_RAWHDLC:
2764 #ifndef ARPHRD_DLCI
2765 #define ARPHRD_DLCI 15
2766 #endif
2767 	case ARPHRD_DLCI:
2768 		/*
2769 		 * XXX - should some of those be mapped to DLT_LINUX_SLL
2770 		 * instead?  Should we just map all of them to DLT_LINUX_SLL?
2771 		 */
2772 		handle->linktype = DLT_RAW;
2773 		break;
2774 
2775 #ifndef ARPHRD_FRAD
2776 #define ARPHRD_FRAD 770
2777 #endif
2778 	case ARPHRD_FRAD:
2779 		handle->linktype = DLT_FRELAY;
2780 		break;
2781 
2782 	case ARPHRD_LOCALTLK:
2783 		handle->linktype = DLT_LTALK;
2784 		break;
2785 
2786 #ifndef ARPHRD_FCPP
2787 #define ARPHRD_FCPP	784
2788 #endif
2789 	case ARPHRD_FCPP:
2790 #ifndef ARPHRD_FCAL
2791 #define ARPHRD_FCAL	785
2792 #endif
2793 	case ARPHRD_FCAL:
2794 #ifndef ARPHRD_FCPL
2795 #define ARPHRD_FCPL	786
2796 #endif
2797 	case ARPHRD_FCPL:
2798 #ifndef ARPHRD_FCFABRIC
2799 #define ARPHRD_FCFABRIC	787
2800 #endif
2801 	case ARPHRD_FCFABRIC:
2802 		/*
2803 		 * We assume that those all mean RFC 2625 IP-over-
2804 		 * Fibre Channel, with the RFC 2625 header at
2805 		 * the beginning of the packet.
2806 		 */
2807 		handle->linktype = DLT_IP_OVER_FC;
2808 		break;
2809 
2810 #ifndef ARPHRD_IRDA
2811 #define ARPHRD_IRDA	783
2812 #endif
2813 	case ARPHRD_IRDA:
2814 		/* Don't expect IP packet out of this interfaces... */
2815 		handle->linktype = DLT_LINUX_IRDA;
2816 		/* We need to save packet direction for IrDA decoding,
2817 		 * so let's use "Linux-cooked" mode. Jean II */
2818 		//handle->md.cooked = 1;
2819 		break;
2820 
2821 	/* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
2822 	 * is needed, please report it to <daniele@orlandi.com> */
2823 #ifndef ARPHRD_LAPD
2824 #define ARPHRD_LAPD	8445
2825 #endif
2826 	case ARPHRD_LAPD:
2827 		/* Don't expect IP packet out of this interfaces... */
2828 		handle->linktype = DLT_LINUX_LAPD;
2829 		break;
2830 
2831 #ifndef ARPHRD_NONE
2832 #define ARPHRD_NONE	0xFFFE
2833 #endif
2834 	case ARPHRD_NONE:
2835 		/*
2836 		 * No link-layer header; packets are just IP
2837 		 * packets, so use DLT_RAW.
2838 		 */
2839 		handle->linktype = DLT_RAW;
2840 		break;
2841 
2842 #ifndef ARPHRD_IEEE802154
2843 #define ARPHRD_IEEE802154      804
2844 #endif
2845        case ARPHRD_IEEE802154:
2846                handle->linktype =  DLT_IEEE802_15_4_NOFCS;
2847                break;
2848 
2849 	default:
2850 		handle->linktype = -1;
2851 		break;
2852 	}
2853 }
2854 
2855 /* ===== Functions to interface to the newer kernels ================== */
2856 
2857 /*
2858  * Try to open a packet socket using the new kernel PF_PACKET interface.
2859  * Returns 1 on success, 0 on an error that means the new interface isn't
2860  * present (so the old SOCK_PACKET interface should be tried), and a
2861  * PCAP_ERROR_ value on an error that means that the old mechanism won't
2862  * work either (so it shouldn't be tried).
2863  */
2864 static int
2865 activate_new(pcap_t *handle)
2866 {
2867 #ifdef HAVE_PF_PACKET_SOCKETS
2868 	const char		*device = handle->opt.source;
2869 	int			is_any_device = (strcmp(device, "any") == 0);
2870 	int			sock_fd = -1, arptype;
2871 #ifdef HAVE_PACKET_AUXDATA
2872 	int			val;
2873 #endif
2874 	int			err = 0;
2875 	struct packet_mreq	mr;
2876 
2877 	/*
2878 	 * Open a socket with protocol family packet. If the
2879 	 * "any" device was specified, we open a SOCK_DGRAM
2880 	 * socket for the cooked interface, otherwise we first
2881 	 * try a SOCK_RAW socket for the raw interface.
2882 	 */
2883 	sock_fd = is_any_device ?
2884 		socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) :
2885 		socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
2886 
2887 	if (sock_fd == -1) {
2888 		if (errno == EINVAL || errno == EAFNOSUPPORT) {
2889 			/*
2890 			 * We don't support PF_PACKET/SOCK_whatever
2891 			 * sockets; try the old mechanism.
2892 			 */
2893 			return 0;
2894 		}
2895 
2896 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
2897 			 pcap_strerror(errno) );
2898 		if (errno == EPERM || errno == EACCES) {
2899 			/*
2900 			 * You don't have permission to open the
2901 			 * socket.
2902 			 */
2903 			return PCAP_ERROR_PERM_DENIED;
2904 		} else {
2905 			/*
2906 			 * Other error.
2907 			 */
2908 			return PCAP_ERROR;
2909 		}
2910 	}
2911 
2912 	/* It seems the kernel supports the new interface. */
2913 	handle->md.sock_packet = 0;
2914 
2915 	/*
2916 	 * Get the interface index of the loopback device.
2917 	 * If the attempt fails, don't fail, just set the
2918 	 * "md.lo_ifindex" to -1.
2919 	 *
2920 	 * XXX - can there be more than one device that loops
2921 	 * packets back, i.e. devices other than "lo"?  If so,
2922 	 * we'd need to find them all, and have an array of
2923 	 * indices for them, and check all of them in
2924 	 * "pcap_read_packet()".
2925 	 */
2926 	handle->md.lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
2927 
2928 	/*
2929 	 * Default value for offset to align link-layer payload
2930 	 * on a 4-byte boundary.
2931 	 */
2932 	handle->offset	 = 0;
2933 
2934 	/*
2935 	 * What kind of frames do we have to deal with? Fall back
2936 	 * to cooked mode if we have an unknown interface type
2937 	 * or a type we know doesn't work well in raw mode.
2938 	 */
2939 	if (!is_any_device) {
2940 		/* Assume for now we don't need cooked mode. */
2941 		handle->md.cooked = 0;
2942 
2943 		if (handle->opt.rfmon) {
2944 			/*
2945 			 * We were asked to turn on monitor mode.
2946 			 * Do so before we get the link-layer type,
2947 			 * because entering monitor mode could change
2948 			 * the link-layer type.
2949 			 */
2950 			err = enter_rfmon_mode(handle, sock_fd, device);
2951 			if (err < 0) {
2952 				/* Hard failure */
2953 				close(sock_fd);
2954 				return err;
2955 			}
2956 			if (err == 0) {
2957 				/*
2958 				 * Nothing worked for turning monitor mode
2959 				 * on.
2960 				 */
2961 				close(sock_fd);
2962 				return PCAP_ERROR_RFMON_NOTSUP;
2963 			}
2964 
2965 			/*
2966 			 * Either monitor mode has been turned on for
2967 			 * the device, or we've been given a different
2968 			 * device to open for monitor mode.  If we've
2969 			 * been given a different device, use it.
2970 			 */
2971 			if (handle->md.mondevice != NULL)
2972 				device = handle->md.mondevice;
2973 		}
2974 		arptype	= iface_get_arptype(sock_fd, device, handle->errbuf);
2975 		if (arptype < 0) {
2976 			close(sock_fd);
2977 			return arptype;
2978 		}
2979 		map_arphrd_to_dlt(handle, arptype, 1);
2980 		if (handle->linktype == -1 ||
2981 		    handle->linktype == DLT_LINUX_SLL ||
2982 		    handle->linktype == DLT_LINUX_IRDA ||
2983 		    handle->linktype == DLT_LINUX_LAPD ||
2984 		    (handle->linktype == DLT_EN10MB &&
2985 		     (strncmp("isdn", device, 4) == 0 ||
2986 		      strncmp("isdY", device, 4) == 0))) {
2987 			/*
2988 			 * Unknown interface type (-1), or a
2989 			 * device we explicitly chose to run
2990 			 * in cooked mode (e.g., PPP devices),
2991 			 * or an ISDN device (whose link-layer
2992 			 * type we can only determine by using
2993 			 * APIs that may be different on different
2994 			 * kernels) - reopen in cooked mode.
2995 			 */
2996 			if (close(sock_fd) == -1) {
2997 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2998 					 "close: %s", pcap_strerror(errno));
2999 				return PCAP_ERROR;
3000 			}
3001 			sock_fd = socket(PF_PACKET, SOCK_DGRAM,
3002 			    htons(ETH_P_ALL));
3003 			if (sock_fd == -1) {
3004 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3005 				    "socket: %s", pcap_strerror(errno));
3006 				if (errno == EPERM || errno == EACCES) {
3007 					/*
3008 					 * You don't have permission to
3009 					 * open the socket.
3010 					 */
3011 					return PCAP_ERROR_PERM_DENIED;
3012 				} else {
3013 					/*
3014 					 * Other error.
3015 					 */
3016 					return PCAP_ERROR;
3017 				}
3018 			}
3019 			handle->md.cooked = 1;
3020 
3021 			/*
3022 			 * Get rid of any link-layer type list
3023 			 * we allocated - this only supports cooked
3024 			 * capture.
3025 			 */
3026 			if (handle->dlt_list != NULL) {
3027 				free(handle->dlt_list);
3028 				handle->dlt_list = NULL;
3029 				handle->dlt_count = 0;
3030 			}
3031 
3032 			if (handle->linktype == -1) {
3033 				/*
3034 				 * Warn that we're falling back on
3035 				 * cooked mode; we may want to
3036 				 * update "map_arphrd_to_dlt()"
3037 				 * to handle the new type.
3038 				 */
3039 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3040 					"arptype %d not "
3041 					"supported by libpcap - "
3042 					"falling back to cooked "
3043 					"socket",
3044 					arptype);
3045 			}
3046 
3047 			/*
3048 			 * IrDA capture is not a real "cooked" capture,
3049 			 * it's IrLAP frames, not IP packets.  The
3050 			 * same applies to LAPD capture.
3051 			 */
3052 			if (handle->linktype != DLT_LINUX_IRDA &&
3053 			    handle->linktype != DLT_LINUX_LAPD)
3054 				handle->linktype = DLT_LINUX_SLL;
3055 		}
3056 
3057 		handle->md.ifindex = iface_get_id(sock_fd, device,
3058 		    handle->errbuf);
3059 		if (handle->md.ifindex == -1) {
3060 			close(sock_fd);
3061 			return PCAP_ERROR;
3062 		}
3063 
3064 		if ((err = iface_bind(sock_fd, handle->md.ifindex,
3065 		    handle->errbuf)) != 1) {
3066 		    	close(sock_fd);
3067 			if (err < 0)
3068 				return err;
3069 			else
3070 				return 0;	/* try old mechanism */
3071 		}
3072 	} else {
3073 		/*
3074 		 * The "any" device.
3075 		 */
3076 		if (handle->opt.rfmon) {
3077 			/*
3078 			 * It doesn't support monitor mode.
3079 			 */
3080 			return PCAP_ERROR_RFMON_NOTSUP;
3081 		}
3082 
3083 		/*
3084 		 * It uses cooked mode.
3085 		 */
3086 		handle->md.cooked = 1;
3087 		handle->linktype = DLT_LINUX_SLL;
3088 
3089 		/*
3090 		 * We're not bound to a device.
3091 		 * For now, we're using this as an indication
3092 		 * that we can't transmit; stop doing that only
3093 		 * if we figure out how to transmit in cooked
3094 		 * mode.
3095 		 */
3096 		handle->md.ifindex = -1;
3097 	}
3098 
3099 	/*
3100 	 * Select promiscuous mode on if "promisc" is set.
3101 	 *
3102 	 * Do not turn allmulti mode on if we don't select
3103 	 * promiscuous mode - on some devices (e.g., Orinoco
3104 	 * wireless interfaces), allmulti mode isn't supported
3105 	 * and the driver implements it by turning promiscuous
3106 	 * mode on, and that screws up the operation of the
3107 	 * card as a normal networking interface, and on no
3108 	 * other platform I know of does starting a non-
3109 	 * promiscuous capture affect which multicast packets
3110 	 * are received by the interface.
3111 	 */
3112 
3113 	/*
3114 	 * Hmm, how can we set promiscuous mode on all interfaces?
3115 	 * I am not sure if that is possible at all.  For now, we
3116 	 * silently ignore attempts to turn promiscuous mode on
3117 	 * for the "any" device (so you don't have to explicitly
3118 	 * disable it in programs such as tcpdump).
3119 	 */
3120 
3121 	if (!is_any_device && handle->opt.promisc) {
3122 		memset(&mr, 0, sizeof(mr));
3123 		mr.mr_ifindex = handle->md.ifindex;
3124 		mr.mr_type    = PACKET_MR_PROMISC;
3125 		if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
3126 		    &mr, sizeof(mr)) == -1) {
3127 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3128 				"setsockopt: %s", pcap_strerror(errno));
3129 			close(sock_fd);
3130 			return PCAP_ERROR;
3131 		}
3132 	}
3133 
3134 	/* Enable auxillary data if supported and reserve room for
3135 	 * reconstructing VLAN headers. */
3136 #ifdef HAVE_PACKET_AUXDATA
3137 	val = 1;
3138 	if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
3139 		       sizeof(val)) == -1 && errno != ENOPROTOOPT) {
3140 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3141 			 "setsockopt: %s", pcap_strerror(errno));
3142 		close(sock_fd);
3143 		return PCAP_ERROR;
3144 	}
3145 	handle->offset += VLAN_TAG_LEN;
3146 #endif /* HAVE_PACKET_AUXDATA */
3147 
3148 	/*
3149 	 * This is a 2.2[.x] or later kernel (we know that
3150 	 * because we're not using a SOCK_PACKET socket -
3151 	 * PF_PACKET is supported only in 2.2 and later
3152 	 * kernels).
3153 	 *
3154 	 * We can safely pass "recvfrom()" a byte count
3155 	 * based on the snapshot length.
3156 	 *
3157 	 * If we're in cooked mode, make the snapshot length
3158 	 * large enough to hold a "cooked mode" header plus
3159 	 * 1 byte of packet data (so we don't pass a byte
3160 	 * count of 0 to "recvfrom()").
3161 	 */
3162 	if (handle->md.cooked) {
3163 		if (handle->snapshot < SLL_HDR_LEN + 1)
3164 			handle->snapshot = SLL_HDR_LEN + 1;
3165 	}
3166 	handle->bufsize = handle->snapshot;
3167 
3168 	/* Save the socket FD in the pcap structure */
3169 	handle->fd = sock_fd;
3170 
3171 	return 1;
3172 #else
3173 	strncpy(ebuf,
3174 		"New packet capturing interface not supported by build "
3175 		"environment", PCAP_ERRBUF_SIZE);
3176 	return 0;
3177 #endif
3178 }
3179 
3180 #ifdef HAVE_PACKET_RING
3181 /*
3182  * Attempt to activate with memory-mapped access.
3183  *
3184  * On success, returns 1, and sets *status to 0 if there are no warnings
3185  * or to a PCAP_WARNING_ code if there is a warning.
3186  *
3187  * On failure due to lack of support for memory-mapped capture, returns
3188  * 0.
3189  *
3190  * On error, returns -1, and sets *status to the appropriate error code;
3191  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3192  */
3193 static int
3194 activate_mmap(pcap_t *handle, int *status)
3195 {
3196 	int ret;
3197 
3198 	/*
3199 	 * Attempt to allocate a buffer to hold the contents of one
3200 	 * packet, for use by the oneshot callback.
3201 	 */
3202 	handle->md.oneshot_buffer = malloc(handle->snapshot);
3203 	if (handle->md.oneshot_buffer == NULL) {
3204 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3205 			 "can't allocate oneshot buffer: %s",
3206 			 pcap_strerror(errno));
3207 		*status = PCAP_ERROR;
3208 		return -1;
3209 	}
3210 
3211 	if (handle->opt.buffer_size == 0) {
3212 		/* by default request 2M for the ring buffer */
3213 		handle->opt.buffer_size = 2*1024*1024;
3214 	}
3215 	ret = prepare_tpacket_socket(handle);
3216 	if (ret == -1) {
3217 		free(handle->md.oneshot_buffer);
3218 		*status = PCAP_ERROR;
3219 		return ret;
3220 	}
3221 	ret = create_ring(handle, status);
3222 	if (ret == 0) {
3223 		/*
3224 		 * We don't support memory-mapped capture; our caller
3225 		 * will fall back on reading from the socket.
3226 		 */
3227 		free(handle->md.oneshot_buffer);
3228 		return 0;
3229 	}
3230 	if (ret == -1) {
3231 		/*
3232 		 * Error attempting to enable memory-mapped capture;
3233 		 * fail.  create_ring() has set *status.
3234 		 */
3235 		free(handle->md.oneshot_buffer);
3236 		return -1;
3237 	}
3238 
3239 	/*
3240 	 * Success.  *status has been set either to 0 if there are no
3241 	 * warnings or to a PCAP_WARNING_ value if there is a warning.
3242 	 *
3243 	 * Override some defaults and inherit the other fields from
3244 	 * activate_new.
3245 	 * handle->offset is used to get the current position into the rx ring.
3246 	 * handle->cc is used to store the ring size.
3247 	 */
3248 	handle->read_op = pcap_read_linux_mmap;
3249 	handle->cleanup_op = pcap_cleanup_linux_mmap;
3250 	handle->setfilter_op = pcap_setfilter_linux_mmap;
3251 	handle->setnonblock_op = pcap_setnonblock_mmap;
3252 	handle->getnonblock_op = pcap_getnonblock_mmap;
3253 	handle->oneshot_callback = pcap_oneshot_mmap;
3254 	handle->selectable_fd = handle->fd;
3255 	return 1;
3256 }
3257 #else /* HAVE_PACKET_RING */
3258 static int
3259 activate_mmap(pcap_t *handle _U_, int *status _U_)
3260 {
3261 	return 0;
3262 }
3263 #endif /* HAVE_PACKET_RING */
3264 
3265 #ifdef HAVE_PACKET_RING
3266 /*
3267  * Attempt to set the socket to version 2 of the memory-mapped header.
3268  * Return 1 if we succeed or if we fail because version 2 isn't
3269  * supported; return -1 on any other error, and set handle->errbuf.
3270  */
3271 static int
3272 prepare_tpacket_socket(pcap_t *handle)
3273 {
3274 #ifdef HAVE_TPACKET2
3275 	socklen_t len;
3276 	int val;
3277 #endif
3278 
3279 	handle->md.tp_version = TPACKET_V1;
3280 	handle->md.tp_hdrlen = sizeof(struct tpacket_hdr);
3281 
3282 #ifdef HAVE_TPACKET2
3283 	/* Probe whether kernel supports TPACKET_V2 */
3284 	val = TPACKET_V2;
3285 	len = sizeof(val);
3286 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
3287 		if (errno == ENOPROTOOPT)
3288 			return 1;	/* no - just drive on */
3289 
3290 		/* Yes - treat as a failure. */
3291 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3292 		    "can't get TPACKET_V2 header len on packet socket: %s",
3293 		    pcap_strerror(errno));
3294 		return -1;
3295 	}
3296 	handle->md.tp_hdrlen = val;
3297 
3298 	val = TPACKET_V2;
3299 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
3300 		       sizeof(val)) < 0) {
3301 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3302 		    "can't activate TPACKET_V2 on packet socket: %s",
3303 		    pcap_strerror(errno));
3304 		return -1;
3305 	}
3306 	handle->md.tp_version = TPACKET_V2;
3307 
3308 	/* Reserve space for VLAN tag reconstruction */
3309 	val = VLAN_TAG_LEN;
3310 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val,
3311 		       sizeof(val)) < 0) {
3312 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3313 		    "can't set up reserve on packet socket: %s",
3314 		    pcap_strerror(errno));
3315 		return -1;
3316 	}
3317 
3318 #endif /* HAVE_TPACKET2 */
3319 	return 1;
3320 }
3321 
3322 /*
3323  * Attempt to set up memory-mapped access.
3324  *
3325  * On success, returns 1, and sets *status to 0 if there are no warnings
3326  * or to a PCAP_WARNING_ code if there is a warning.
3327  *
3328  * On failure due to lack of support for memory-mapped capture, returns
3329  * 0.
3330  *
3331  * On error, returns -1, and sets *status to the appropriate error code;
3332  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3333  */
3334 static int
3335 create_ring(pcap_t *handle, int *status)
3336 {
3337 	unsigned i, j, frames_per_block;
3338 	struct tpacket_req req;
3339 	socklen_t len;
3340 	unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
3341 	unsigned int frame_size;
3342 
3343 	/*
3344 	 * Start out assuming no warnings or errors.
3345 	 */
3346 	*status = 0;
3347 
3348 	/* Note that with large snapshot length (say 64K, which is the default
3349 	 * for recent versions of tcpdump, the value that "-s 0" has given
3350 	 * for a long time with tcpdump, and the default in Wireshark/TShark),
3351 	 * if we use the snapshot length to calculate the frame length,
3352 	 * only a few frames will be available in the ring even with pretty
3353 	 * large ring size (and a lot of memory will be unused).
3354 	 *
3355 	 * Ideally, we should choose a frame length based on the
3356 	 * minimum of the specified snapshot length and the maximum
3357 	 * packet size.  That's not as easy as it sounds; consider, for
3358 	 * example, an 802.11 interface in monitor mode, where the
3359 	 * frame would include a radiotap header, where the maximum
3360 	 * radiotap header length is device-dependent.
3361 	 *
3362 	 * So, for now, we just do this for Ethernet devices, where
3363 	 * there's no metadata header, and the link-layer header is
3364 	 * fixed length.  We can get the maximum packet size by
3365 	 * adding 18, the Ethernet header length plus the CRC length
3366 	 * (just in case we happen to get the CRC in the packet), to
3367 	 * the MTU of the interface; we fetch the MTU in the hopes
3368 	 * that it reflects support for jumbo frames.  (Even if the
3369 	 * interface is just being used for passive snooping, the driver
3370 	 * might set the size of buffers in the receive ring based on
3371 	 * the MTU, so that the MTU limits the maximum size of packets
3372 	 * that we can receive.)
3373 	 *
3374 	 * We don't do that if segmentation/fragmentation or receive
3375 	 * offload are enabled, so we don't get rudely surprised by
3376 	 * "packets" bigger than the MTU. */
3377 	frame_size = handle->snapshot;
3378 	if (handle->linktype == DLT_EN10MB) {
3379 		int mtu;
3380 		int offload;
3381 
3382 		offload = iface_get_offload(handle);
3383 		if (offload == -1) {
3384 			*status = PCAP_ERROR;
3385 			return -1;
3386 		}
3387 		if (!offload) {
3388 			mtu = iface_get_mtu(handle->fd, handle->opt.source,
3389 			    handle->errbuf);
3390 			if (mtu == -1) {
3391 				*status = PCAP_ERROR;
3392 				return -1;
3393 			}
3394 			if (frame_size > mtu + 18)
3395 				frame_size = mtu + 18;
3396 		}
3397 	}
3398 
3399 	/* NOTE: calculus matching those in tpacket_rcv()
3400 	 * in linux-2.6/net/packet/af_packet.c
3401 	 */
3402 	len = sizeof(sk_type);
3403 	if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type, &len) < 0) {
3404 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "getsockopt: %s", pcap_strerror(errno));
3405 		*status = PCAP_ERROR;
3406 		return -1;
3407 	}
3408 #ifdef PACKET_RESERVE
3409 	len = sizeof(tp_reserve);
3410 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &tp_reserve, &len) < 0) {
3411 		if (errno != ENOPROTOOPT) {
3412 			/*
3413 			 * ENOPROTOOPT means "kernel doesn't support
3414 			 * PACKET_RESERVE", in which case we fall back
3415 			 * as best we can.
3416 			 */
3417 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "getsockopt: %s", pcap_strerror(errno));
3418 			*status = PCAP_ERROR;
3419 			return -1;
3420 		}
3421 		tp_reserve = 0;	/* older kernel, reserve not supported */
3422 	}
3423 #else
3424 	tp_reserve = 0;	/* older kernel, reserve not supported */
3425 #endif
3426 	maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
3427 		/* XXX: in the kernel maclen is calculated from
3428 		 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
3429 		 * in:  packet_snd()           in linux-2.6/net/packet/af_packet.c
3430 		 * then packet_alloc_skb()     in linux-2.6/net/packet/af_packet.c
3431 		 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
3432 		 * but I see no way to get those sizes in userspace,
3433 		 * like for instance with an ifreq ioctl();
3434 		 * the best thing I've found so far is MAX_HEADER in the kernel
3435 		 * part of linux-2.6/include/linux/netdevice.h
3436 		 * which goes up to 128+48=176; since pcap-linux.c defines
3437 		 * a MAX_LINKHEADER_SIZE of 256 which is greater than that,
3438 		 * let's use it.. maybe is it even large enough to directly
3439 		 * replace macoff..
3440 		 */
3441 	tp_hdrlen = TPACKET_ALIGN(handle->md.tp_hdrlen) + sizeof(struct sockaddr_ll) ;
3442 	netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
3443 		/* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN of
3444 		 * netoff, which contradicts
3445 		 * linux-2.6/Documentation/networking/packet_mmap.txt
3446 		 * documenting that:
3447 		 * "- Gap, chosen so that packet data (Start+tp_net)
3448 		 * aligns to TPACKET_ALIGNMENT=16"
3449 		 */
3450 		/* NOTE: in linux-2.6/include/linux/skbuff.h:
3451 		 * "CPUs often take a performance hit
3452 		 *  when accessing unaligned memory locations"
3453 		 */
3454 	macoff = netoff - maclen;
3455 	req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
3456 	req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
3457 
3458 	/* compute the minumum block size that will handle this frame.
3459 	 * The block has to be page size aligned.
3460 	 * The max block size allowed by the kernel is arch-dependent and
3461 	 * it's not explicitly checked here. */
3462 	req.tp_block_size = getpagesize();
3463 	while (req.tp_block_size < req.tp_frame_size)
3464 		req.tp_block_size <<= 1;
3465 
3466 	frames_per_block = req.tp_block_size/req.tp_frame_size;
3467 
3468 	/*
3469 	 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
3470 	 * so we check for PACKET_TIMESTAMP.  We check for
3471 	 * linux/net_tstamp.h just in case a system somehow has
3472 	 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
3473 	 * be unnecessary.
3474 	 *
3475 	 * SIOCSHWTSTAMP was introduced in the patch that introduced
3476 	 * linux/net_tstamp.h, so we don't bother checking whether
3477 	 * SIOCSHWTSTAMP is defined (if your Linux system has
3478 	 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
3479 	 * Linux system is badly broken).
3480 	 */
3481 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
3482 	/*
3483 	 * If we were told to do so, ask the kernel and the driver
3484 	 * to use hardware timestamps.
3485 	 *
3486 	 * Hardware timestamps are only supported with mmapped
3487 	 * captures.
3488 	 */
3489 	if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
3490 	    handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
3491 		struct hwtstamp_config hwconfig;
3492 		struct ifreq ifr;
3493 		int timesource;
3494 
3495 		/*
3496 		 * Ask for hardware time stamps on all packets,
3497 		 * including transmitted packets.
3498 		 */
3499 		memset(&hwconfig, 0, sizeof(hwconfig));
3500 		hwconfig.tx_type = HWTSTAMP_TX_ON;
3501 		hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
3502 
3503 		memset(&ifr, 0, sizeof(ifr));
3504 		strcpy(ifr.ifr_name, handle->opt.source);
3505 		ifr.ifr_data = (void *)&hwconfig;
3506 
3507 		if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
3508 			switch (errno) {
3509 
3510 			case EPERM:
3511 				/*
3512 				 * Treat this as an error, as the
3513 				 * user should try to run this
3514 				 * with the appropriate privileges -
3515 				 * and, if they can't, shouldn't
3516 				 * try requesting hardware time stamps.
3517 				 */
3518 				*status = PCAP_ERROR_PERM_DENIED;
3519 				return -1;
3520 
3521 			case EOPNOTSUPP:
3522 				/*
3523 				 * Treat this as a warning, as the
3524 				 * only way to fix the warning is to
3525 				 * get an adapter that supports hardware
3526 				 * time stamps.  We'll just fall back
3527 				 * on the standard host time stamps.
3528 				 */
3529 				*status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
3530 				break;
3531 
3532 			default:
3533 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3534 					"SIOCSHWTSTAMP failed: %s",
3535 					pcap_strerror(errno));
3536 				*status = PCAP_ERROR;
3537 				return -1;
3538 			}
3539 		} else {
3540 			/*
3541 			 * Well, that worked.  Now specify the type of
3542 			 * hardware time stamp we want for this
3543 			 * socket.
3544 			 */
3545 			if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
3546 				/*
3547 				 * Hardware timestamp, synchronized
3548 				 * with the system clock.
3549 				 */
3550 				timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
3551 			} else {
3552 				/*
3553 				 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
3554 				 * timestamp, not synchronized with the
3555 				 * system clock.
3556 				 */
3557 				timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
3558 			}
3559 			if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
3560 				(void *)&timesource, sizeof(timesource))) {
3561 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3562 					"can't set PACKET_TIMESTAMP: %s",
3563 					pcap_strerror(errno));
3564 				*status = PCAP_ERROR;
3565 				return -1;
3566 			}
3567 		}
3568 	}
3569 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
3570 
3571 	/* ask the kernel to create the ring */
3572 retry:
3573 	req.tp_block_nr = req.tp_frame_nr / frames_per_block;
3574 
3575 	/* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
3576 	req.tp_frame_nr = req.tp_block_nr * frames_per_block;
3577 
3578 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3579 					(void *) &req, sizeof(req))) {
3580 		if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
3581 			/*
3582 			 * Memory failure; try to reduce the requested ring
3583 			 * size.
3584 			 *
3585 			 * We used to reduce this by half -- do 5% instead.
3586 			 * That may result in more iterations and a longer
3587 			 * startup, but the user will be much happier with
3588 			 * the resulting buffer size.
3589 			 */
3590 			if (req.tp_frame_nr < 20)
3591 				req.tp_frame_nr -= 1;
3592 			else
3593 				req.tp_frame_nr -= req.tp_frame_nr/20;
3594 			goto retry;
3595 		}
3596 		if (errno == ENOPROTOOPT) {
3597 			/*
3598 			 * We don't have ring buffer support in this kernel.
3599 			 */
3600 			return 0;
3601 		}
3602 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3603 		    "can't create rx ring on packet socket: %s",
3604 		    pcap_strerror(errno));
3605 		*status = PCAP_ERROR;
3606 		return -1;
3607 	}
3608 
3609 	/* memory map the rx ring */
3610 	handle->md.mmapbuflen = req.tp_block_nr * req.tp_block_size;
3611 	handle->md.mmapbuf = mmap(0, handle->md.mmapbuflen,
3612 	    PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
3613 	if (handle->md.mmapbuf == MAP_FAILED) {
3614 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3615 		    "can't mmap rx ring: %s", pcap_strerror(errno));
3616 
3617 		/* clear the allocated ring on error*/
3618 		destroy_ring(handle);
3619 		*status = PCAP_ERROR;
3620 		return -1;
3621 	}
3622 
3623 	/* allocate a ring for each frame header pointer*/
3624 	handle->cc = req.tp_frame_nr;
3625 	handle->buffer = malloc(handle->cc * sizeof(union thdr *));
3626 	if (!handle->buffer) {
3627 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3628 		    "can't allocate ring of frame headers: %s",
3629 		    pcap_strerror(errno));
3630 
3631 		destroy_ring(handle);
3632 		*status = PCAP_ERROR;
3633 		return -1;
3634 	}
3635 
3636 	/* fill the header ring with proper frame ptr*/
3637 	handle->offset = 0;
3638 	for (i=0; i<req.tp_block_nr; ++i) {
3639 		void *base = &handle->md.mmapbuf[i*req.tp_block_size];
3640 		for (j=0; j<frames_per_block; ++j, ++handle->offset) {
3641 			RING_GET_FRAME(handle) = base;
3642 			base += req.tp_frame_size;
3643 		}
3644 	}
3645 
3646 	handle->bufsize = req.tp_frame_size;
3647 	handle->offset = 0;
3648 	return 1;
3649 }
3650 
3651 /* free all ring related resources*/
3652 static void
3653 destroy_ring(pcap_t *handle)
3654 {
3655 	/* tell the kernel to destroy the ring*/
3656 	struct tpacket_req req;
3657 	memset(&req, 0, sizeof(req));
3658 	setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3659 				(void *) &req, sizeof(req));
3660 
3661 	/* if ring is mapped, unmap it*/
3662 	if (handle->md.mmapbuf) {
3663 		/* do not test for mmap failure, as we can't recover from any error */
3664 		munmap(handle->md.mmapbuf, handle->md.mmapbuflen);
3665 		handle->md.mmapbuf = NULL;
3666 	}
3667 }
3668 
3669 /*
3670  * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
3671  * for Linux mmapped capture.
3672  *
3673  * The problem is that pcap_next() and pcap_next_ex() expect the packet
3674  * data handed to the callback to be valid after the callback returns,
3675  * but pcap_read_linux_mmap() has to release that packet as soon as
3676  * the callback returns (otherwise, the kernel thinks there's still
3677  * at least one unprocessed packet available in the ring, so a select()
3678  * will immediately return indicating that there's data to process), so,
3679  * in the callback, we have to make a copy of the packet.
3680  *
3681  * Yes, this means that, if the capture is using the ring buffer, using
3682  * pcap_next() or pcap_next_ex() requires more copies than using
3683  * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
3684  * pcap_next() or pcap_next_ex().
3685  */
3686 static void
3687 pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
3688     const u_char *bytes)
3689 {
3690 	struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
3691 
3692 	*sp->hdr = *h;
3693 	memcpy(sp->pd->md.oneshot_buffer, bytes, h->caplen);
3694 	*sp->pkt = sp->pd->md.oneshot_buffer;
3695 }
3696 
3697 static void
3698 pcap_cleanup_linux_mmap( pcap_t *handle )
3699 {
3700 	destroy_ring(handle);
3701 	if (handle->md.oneshot_buffer != NULL) {
3702 		free(handle->md.oneshot_buffer);
3703 		handle->md.oneshot_buffer = NULL;
3704 	}
3705 	pcap_cleanup_linux(handle);
3706 }
3707 
3708 
3709 static int
3710 pcap_getnonblock_mmap(pcap_t *p, char *errbuf)
3711 {
3712 	/* use negative value of timeout to indicate non blocking ops */
3713 	return (p->md.timeout<0);
3714 }
3715 
3716 static int
3717 pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
3718 {
3719 	/* map each value to the corresponding 2's complement, to
3720 	 * preserve the timeout value provided with pcap_set_timeout */
3721 	if (nonblock) {
3722 		if (p->md.timeout >= 0) {
3723 			/*
3724 			 * Timeout is non-negative, so we're not already
3725 			 * in non-blocking mode; set it to the 2's
3726 			 * complement, to make it negative, as an
3727 			 * indication that we're in non-blocking mode.
3728 			 */
3729 			p->md.timeout = p->md.timeout*-1 - 1;
3730 		}
3731 	} else {
3732 		if (p->md.timeout < 0) {
3733 			/*
3734 			 * Timeout is negative, so we're not already
3735 			 * in blocking mode; reverse the previous
3736 			 * operation, to make the timeout non-negative
3737 			 * again.
3738 			 */
3739 			p->md.timeout = (p->md.timeout+1)*-1;
3740 		}
3741 	}
3742 	return 0;
3743 }
3744 
3745 static inline union thdr *
3746 pcap_get_ring_frame(pcap_t *handle, int status)
3747 {
3748 	union thdr h;
3749 
3750 	h.raw = RING_GET_FRAME(handle);
3751 	switch (handle->md.tp_version) {
3752 	case TPACKET_V1:
3753 		if (status != (h.h1->tp_status ? TP_STATUS_USER :
3754 						TP_STATUS_KERNEL))
3755 			return NULL;
3756 		break;
3757 #ifdef HAVE_TPACKET2
3758 	case TPACKET_V2:
3759 		if (status != (h.h2->tp_status ? TP_STATUS_USER :
3760 						TP_STATUS_KERNEL))
3761 			return NULL;
3762 		break;
3763 #endif
3764 	}
3765 	return h.raw;
3766 }
3767 
3768 #ifndef POLLRDHUP
3769 #define POLLRDHUP 0
3770 #endif
3771 
3772 static int
3773 pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
3774 		u_char *user)
3775 {
3776 	int timeout;
3777 	int pkts = 0;
3778 	char c;
3779 
3780 	/* wait for frames availability.*/
3781 	if (!pcap_get_ring_frame(handle, TP_STATUS_USER)) {
3782 		struct pollfd pollinfo;
3783 		int ret;
3784 
3785 		pollinfo.fd = handle->fd;
3786 		pollinfo.events = POLLIN;
3787 
3788 		if (handle->md.timeout == 0)
3789 			timeout = -1;	/* block forever */
3790 		else if (handle->md.timeout > 0)
3791 			timeout = handle->md.timeout;	/* block for that amount of time */
3792 		else
3793 			timeout = 0;	/* non-blocking mode - poll to pick up errors */
3794 		do {
3795 			ret = poll(&pollinfo, 1, timeout);
3796 			if (ret < 0 && errno != EINTR) {
3797 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3798 					"can't poll on packet socket: %s",
3799 					pcap_strerror(errno));
3800 				return PCAP_ERROR;
3801 			} else if (ret > 0 &&
3802 			    (pollinfo.revents & (POLLHUP|POLLRDHUP|POLLERR|POLLNVAL))) {
3803 				/*
3804 				 * There's some indication other than
3805 				 * "you can read on this descriptor" on
3806 				 * the descriptor.
3807 				 */
3808 				if (pollinfo.revents & (POLLHUP | POLLRDHUP)) {
3809 					snprintf(handle->errbuf,
3810 						PCAP_ERRBUF_SIZE,
3811 						"Hangup on packet socket");
3812 					return PCAP_ERROR;
3813 				}
3814 				if (pollinfo.revents & POLLERR) {
3815 					/*
3816 					 * A recv() will give us the
3817 					 * actual error code.
3818 					 *
3819 					 * XXX - make the socket non-blocking?
3820 					 */
3821 					if (recv(handle->fd, &c, sizeof c,
3822 					    MSG_PEEK) != -1)
3823 						continue;	/* what, no error? */
3824 					if (errno == ENETDOWN) {
3825 						/*
3826 						 * The device on which we're
3827 						 * capturing went away.
3828 						 *
3829 						 * XXX - we should really return
3830 						 * PCAP_ERROR_IFACE_NOT_UP,
3831 						 * but pcap_dispatch() etc.
3832 						 * aren't defined to return
3833 						 * that.
3834 						 */
3835 						snprintf(handle->errbuf,
3836 							PCAP_ERRBUF_SIZE,
3837 							"The interface went down");
3838 					} else {
3839 						snprintf(handle->errbuf,
3840 							PCAP_ERRBUF_SIZE,
3841 							"Error condition on packet socket: %s",
3842 							strerror(errno));
3843 					}
3844 					return PCAP_ERROR;
3845 				}
3846 				if (pollinfo.revents & POLLNVAL) {
3847 					snprintf(handle->errbuf,
3848 						PCAP_ERRBUF_SIZE,
3849 						"Invalid polling request on packet socket");
3850 					return PCAP_ERROR;
3851 				}
3852   			}
3853 			/* check for break loop condition on interrupted syscall*/
3854 			if (handle->break_loop) {
3855 				handle->break_loop = 0;
3856 				return PCAP_ERROR_BREAK;
3857 			}
3858 		} while (ret < 0);
3859 	}
3860 
3861 	/* non-positive values of max_packets are used to require all
3862 	 * packets currently available in the ring */
3863 	while ((pkts < max_packets) || (max_packets <= 0)) {
3864 		int run_bpf;
3865 		struct sockaddr_ll *sll;
3866 		struct pcap_pkthdr pcaphdr;
3867 		unsigned char *bp;
3868 		union thdr h;
3869 		unsigned int tp_len;
3870 		unsigned int tp_mac;
3871 		unsigned int tp_snaplen;
3872 		unsigned int tp_sec;
3873 		unsigned int tp_usec;
3874 
3875 		h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
3876 		if (!h.raw)
3877 			break;
3878 
3879 		switch (handle->md.tp_version) {
3880 		case TPACKET_V1:
3881 			tp_len	   = h.h1->tp_len;
3882 			tp_mac	   = h.h1->tp_mac;
3883 			tp_snaplen = h.h1->tp_snaplen;
3884 			tp_sec	   = h.h1->tp_sec;
3885 			tp_usec	   = h.h1->tp_usec;
3886 			break;
3887 #ifdef HAVE_TPACKET2
3888 		case TPACKET_V2:
3889 			tp_len	   = h.h2->tp_len;
3890 			tp_mac	   = h.h2->tp_mac;
3891 			tp_snaplen = h.h2->tp_snaplen;
3892 			tp_sec	   = h.h2->tp_sec;
3893 			tp_usec	   = h.h2->tp_nsec / 1000;
3894 			break;
3895 #endif
3896 		default:
3897 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3898 				"unsupported tpacket version %d",
3899 				handle->md.tp_version);
3900 			return -1;
3901 		}
3902 		/* perform sanity check on internal offset. */
3903 		if (tp_mac + tp_snaplen > handle->bufsize) {
3904 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3905 				"corrupted frame on kernel ring mac "
3906 				"offset %d + caplen %d > frame len %d",
3907 				tp_mac, tp_snaplen, handle->bufsize);
3908 			return -1;
3909 		}
3910 
3911 		/* run filter on received packet
3912 		 * If the kernel filtering is enabled we need to run the
3913 		 * filter until all the frames present into the ring
3914 		 * at filter creation time are processed.
3915 		 * In such case md.use_bpf is used as a counter for the
3916 		 * packet we need to filter.
3917 		 * Note: alternatively it could be possible to stop applying
3918 		 * the filter when the ring became empty, but it can possibly
3919 		 * happen a lot later... */
3920 		bp = (unsigned char*)h.raw + tp_mac;
3921 		run_bpf = (!handle->md.use_bpf) ||
3922 			((handle->md.use_bpf>1) && handle->md.use_bpf--);
3923 		if (run_bpf && handle->fcode.bf_insns &&
3924 				(bpf_filter(handle->fcode.bf_insns, bp,
3925 					tp_len, tp_snaplen) == 0))
3926 			goto skip;
3927 
3928 		/*
3929 		 * Do checks based on packet direction.
3930 		 */
3931 		sll = (void *)h.raw + TPACKET_ALIGN(handle->md.tp_hdrlen);
3932 		if (sll->sll_pkttype == PACKET_OUTGOING) {
3933 			/*
3934 			 * Outgoing packet.
3935 			 * If this is from the loopback device, reject it;
3936 			 * we'll see the packet as an incoming packet as well,
3937 			 * and we don't want to see it twice.
3938 			 */
3939 			if (sll->sll_ifindex == handle->md.lo_ifindex)
3940 				goto skip;
3941 
3942 			/*
3943 			 * If the user only wants incoming packets, reject it.
3944 			 */
3945 			if (handle->direction == PCAP_D_IN)
3946 				goto skip;
3947 		} else {
3948 			/*
3949 			 * Incoming packet.
3950 			 * If the user only wants outgoing packets, reject it.
3951 			 */
3952 			if (handle->direction == PCAP_D_OUT)
3953 				goto skip;
3954 		}
3955 
3956 		/* get required packet info from ring header */
3957 		pcaphdr.ts.tv_sec = tp_sec;
3958 		pcaphdr.ts.tv_usec = tp_usec;
3959 		pcaphdr.caplen = tp_snaplen;
3960 		pcaphdr.len = tp_len;
3961 
3962 		/* if required build in place the sll header*/
3963 		if (handle->md.cooked) {
3964 			struct sll_header *hdrp;
3965 
3966 			/*
3967 			 * The kernel should have left us with enough
3968 			 * space for an sll header; back up the packet
3969 			 * data pointer into that space, as that'll be
3970 			 * the beginning of the packet we pass to the
3971 			 * callback.
3972 			 */
3973 			bp -= SLL_HDR_LEN;
3974 
3975 			/*
3976 			 * Let's make sure that's past the end of
3977 			 * the tpacket header, i.e. >=
3978 			 * ((u_char *)thdr + TPACKET_HDRLEN), so we
3979 			 * don't step on the header when we construct
3980 			 * the sll header.
3981 			 */
3982 			if (bp < (u_char *)h.raw +
3983 					   TPACKET_ALIGN(handle->md.tp_hdrlen) +
3984 					   sizeof(struct sockaddr_ll)) {
3985 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3986 					"cooked-mode frame doesn't have room for sll header");
3987 				return -1;
3988 			}
3989 
3990 			/*
3991 			 * OK, that worked; construct the sll header.
3992 			 */
3993 			hdrp = (struct sll_header *)bp;
3994 			hdrp->sll_pkttype = map_packet_type_to_sll_type(
3995 							sll->sll_pkttype);
3996 			hdrp->sll_hatype = htons(sll->sll_hatype);
3997 			hdrp->sll_halen = htons(sll->sll_halen);
3998 			memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
3999 			hdrp->sll_protocol = sll->sll_protocol;
4000 
4001 			/* update packet len */
4002 			pcaphdr.caplen += SLL_HDR_LEN;
4003 			pcaphdr.len += SLL_HDR_LEN;
4004 		}
4005 
4006 #ifdef HAVE_TPACKET2
4007 		if (handle->md.tp_version == TPACKET_V2 && h.h2->tp_vlan_tci &&
4008 		    tp_snaplen >= 2 * ETH_ALEN) {
4009 			struct vlan_tag *tag;
4010 
4011 			bp -= VLAN_TAG_LEN;
4012 			memmove(bp, bp + VLAN_TAG_LEN, 2 * ETH_ALEN);
4013 
4014 			tag = (struct vlan_tag *)(bp + 2 * ETH_ALEN);
4015 			tag->vlan_tpid = htons(ETH_P_8021Q);
4016 			tag->vlan_tci = htons(h.h2->tp_vlan_tci);
4017 
4018 			pcaphdr.caplen += VLAN_TAG_LEN;
4019 			pcaphdr.len += VLAN_TAG_LEN;
4020 		}
4021 #endif
4022 
4023 		/*
4024 		 * The only way to tell the kernel to cut off the
4025 		 * packet at a snapshot length is with a filter program;
4026 		 * if there's no filter program, the kernel won't cut
4027 		 * the packet off.
4028 		 *
4029 		 * Trim the snapshot length to be no longer than the
4030 		 * specified snapshot length.
4031 		 */
4032 		if (pcaphdr.caplen > handle->snapshot)
4033 			pcaphdr.caplen = handle->snapshot;
4034 
4035 		/* pass the packet to the user */
4036 		pkts++;
4037 		callback(user, &pcaphdr, bp);
4038 		handle->md.packets_read++;
4039 
4040 skip:
4041 		/* next packet */
4042 		switch (handle->md.tp_version) {
4043 		case TPACKET_V1:
4044 			h.h1->tp_status = TP_STATUS_KERNEL;
4045 			break;
4046 #ifdef HAVE_TPACKET2
4047 		case TPACKET_V2:
4048 			h.h2->tp_status = TP_STATUS_KERNEL;
4049 			break;
4050 #endif
4051 		}
4052 		if (++handle->offset >= handle->cc)
4053 			handle->offset = 0;
4054 
4055 		/* check for break loop condition*/
4056 		if (handle->break_loop) {
4057 			handle->break_loop = 0;
4058 			return PCAP_ERROR_BREAK;
4059 		}
4060 	}
4061 	return pkts;
4062 }
4063 
4064 static int
4065 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
4066 {
4067 	int n, offset;
4068 	int ret;
4069 
4070 	/*
4071 	 * Don't rewrite "ret" instructions; we don't need to, as
4072 	 * we're not reading packets with recvmsg(), and we don't
4073 	 * want to, as, by not rewriting them, the kernel can avoid
4074 	 * copying extra data.
4075 	 */
4076 	ret = pcap_setfilter_linux_common(handle, filter, 1);
4077 	if (ret < 0)
4078 		return ret;
4079 
4080 	/* if the kernel filter is enabled, we need to apply the filter on
4081 	 * all packets present into the ring. Get an upper bound of their number
4082 	 */
4083 	if (!handle->md.use_bpf)
4084 		return ret;
4085 
4086 	/* walk the ring backward and count the free slot */
4087 	offset = handle->offset;
4088 	if (--handle->offset < 0)
4089 		handle->offset = handle->cc - 1;
4090 	for (n=0; n < handle->cc; ++n) {
4091 		if (--handle->offset < 0)
4092 			handle->offset = handle->cc - 1;
4093 		if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL))
4094 			break;
4095 	}
4096 
4097 	/* be careful to not change current ring position */
4098 	handle->offset = offset;
4099 
4100 	/* store the number of packets currently present in the ring */
4101 	handle->md.use_bpf = 1 + (handle->cc - n);
4102 	return ret;
4103 }
4104 
4105 #endif /* HAVE_PACKET_RING */
4106 
4107 
4108 #ifdef HAVE_PF_PACKET_SOCKETS
4109 /*
4110  *  Return the index of the given device name. Fill ebuf and return
4111  *  -1 on failure.
4112  */
4113 static int
4114 iface_get_id(int fd, const char *device, char *ebuf)
4115 {
4116 	struct ifreq	ifr;
4117 
4118 	memset(&ifr, 0, sizeof(ifr));
4119 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4120 
4121 	if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
4122 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4123 			 "SIOCGIFINDEX: %s", pcap_strerror(errno));
4124 		return -1;
4125 	}
4126 
4127 	return ifr.ifr_ifindex;
4128 }
4129 
4130 /*
4131  *  Bind the socket associated with FD to the given device.
4132  *  Return 1 on success, 0 if we should try a SOCK_PACKET socket,
4133  *  or a PCAP_ERROR_ value on a hard error.
4134  */
4135 static int
4136 iface_bind(int fd, int ifindex, char *ebuf)
4137 {
4138 	struct sockaddr_ll	sll;
4139 	int			err;
4140 	socklen_t		errlen = sizeof(err);
4141 
4142 	memset(&sll, 0, sizeof(sll));
4143 	sll.sll_family		= AF_PACKET;
4144 	sll.sll_ifindex		= ifindex;
4145 	sll.sll_protocol	= htons(ETH_P_ALL);
4146 
4147 	if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
4148 		if (errno == ENETDOWN) {
4149 			/*
4150 			 * Return a "network down" indication, so that
4151 			 * the application can report that rather than
4152 			 * saying we had a mysterious failure and
4153 			 * suggest that they report a problem to the
4154 			 * libpcap developers.
4155 			 */
4156 			return PCAP_ERROR_IFACE_NOT_UP;
4157 		} else {
4158 			snprintf(ebuf, PCAP_ERRBUF_SIZE,
4159 				 "bind: %s", pcap_strerror(errno));
4160 			return PCAP_ERROR;
4161 		}
4162 	}
4163 
4164 	/* Any pending errors, e.g., network is down? */
4165 
4166 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
4167 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4168 			"getsockopt: %s", pcap_strerror(errno));
4169 		return 0;
4170 	}
4171 
4172 	if (err == ENETDOWN) {
4173 		/*
4174 		 * Return a "network down" indication, so that
4175 		 * the application can report that rather than
4176 		 * saying we had a mysterious failure and
4177 		 * suggest that they report a problem to the
4178 		 * libpcap developers.
4179 		 */
4180 		return PCAP_ERROR_IFACE_NOT_UP;
4181 	} else if (err > 0) {
4182 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
4183 			"bind: %s", pcap_strerror(err));
4184 		return 0;
4185 	}
4186 
4187 	return 1;
4188 }
4189 
4190 #ifdef IW_MODE_MONITOR
4191 /*
4192  * Check whether the device supports the Wireless Extensions.
4193  * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE
4194  * if the device doesn't even exist.
4195  */
4196 static int
4197 has_wext(int sock_fd, const char *device, char *ebuf)
4198 {
4199 	struct iwreq ireq;
4200 
4201 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4202 	    sizeof ireq.ifr_ifrn.ifrn_name);
4203 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4204 	if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
4205 		return 1;	/* yes */
4206 	snprintf(ebuf, PCAP_ERRBUF_SIZE,
4207 	    "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
4208 	if (errno == ENODEV)
4209 		return PCAP_ERROR_NO_SUCH_DEVICE;
4210 	return 0;
4211 }
4212 
4213 /*
4214  * Per me si va ne la citta dolente,
4215  * Per me si va ne l'etterno dolore,
4216  *	...
4217  * Lasciate ogne speranza, voi ch'intrate.
4218  *
4219  * XXX - airmon-ng does special stuff with the Orinoco driver and the
4220  * wlan-ng driver.
4221  */
4222 typedef enum {
4223 	MONITOR_WEXT,
4224 	MONITOR_HOSTAP,
4225 	MONITOR_PRISM,
4226 	MONITOR_PRISM54,
4227 	MONITOR_ACX100,
4228 	MONITOR_RT2500,
4229 	MONITOR_RT2570,
4230 	MONITOR_RT73,
4231 	MONITOR_RTL8XXX
4232 } monitor_type;
4233 
4234 /*
4235  * Use the Wireless Extensions, if we have them, to try to turn monitor mode
4236  * on if it's not already on.
4237  *
4238  * Returns 1 on success, 0 if we don't support the Wireless Extensions
4239  * on this device, or a PCAP_ERROR_ value if we do support them but
4240  * we weren't able to turn monitor mode on.
4241  */
4242 static int
4243 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
4244 {
4245 	/*
4246 	 * XXX - at least some adapters require non-Wireless Extensions
4247 	 * mechanisms to turn monitor mode on.
4248 	 *
4249 	 * Atheros cards might require that a separate "monitor virtual access
4250 	 * point" be created, with later versions of the madwifi driver.
4251 	 * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode
4252 	 * monitor -bssid", which apparently spits out a line "athN"
4253 	 * where "athN" is the monitor mode device.  To leave monitor
4254 	 * mode, it destroys the monitor mode device.
4255 	 *
4256 	 * Some Intel Centrino adapters might require private ioctls to get
4257 	 * radio headers; the ipw2200 and ipw3945 drivers allow you to
4258 	 * configure a separate "rtapN" interface to capture in monitor
4259 	 * mode without preventing the adapter from operating normally.
4260 	 * (airmon-ng doesn't appear to use that, though.)
4261 	 *
4262 	 * It would be Truly Wonderful if mac80211 and nl80211 cleaned this
4263 	 * up, and if all drivers were converted to mac80211 drivers.
4264 	 *
4265 	 * If interface {if} is a mac80211 driver, the file
4266 	 * /sys/class/net/{if}/phy80211 is a symlink to
4267 	 * /sys/class/ieee80211/{phydev}, for some {phydev}.
4268 	 *
4269 	 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
4270 	 * least, has a "wmaster0" device and a "wlan0" device; the
4271 	 * latter is the one with the IP address.  Both show up in
4272 	 * "tcpdump -D" output.  Capturing on the wmaster0 device
4273 	 * captures with 802.11 headers.
4274 	 *
4275 	 * airmon-ng searches through /sys/class/net for devices named
4276 	 * monN, starting with mon0; as soon as one *doesn't* exist,
4277 	 * it chooses that as the monitor device name.  If the "iw"
4278 	 * command exists, it does "iw dev {if} interface add {monif}
4279 	 * type monitor", where {monif} is the monitor device.  It
4280 	 * then (sigh) sleeps .1 second, and then configures the
4281 	 * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
4282 	 * is a file, it writes {mondev}, without a newline, to that file,
4283 	 * and again (sigh) sleeps .1 second, and then iwconfig's that
4284 	 * device into monitor mode and configures it up.  Otherwise,
4285 	 * you can't do monitor mode.
4286 	 *
4287 	 * All these devices are "glued" together by having the
4288 	 * /sys/class/net/{device}/phy80211 links pointing to the same
4289 	 * place, so, given a wmaster, wlan, or mon device, you can
4290 	 * find the other devices by looking for devices with
4291 	 * the same phy80211 link.
4292 	 *
4293 	 * To turn monitor mode off, delete the monitor interface,
4294 	 * either with "iw dev {monif} interface del" or by sending
4295 	 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
4296 	 *
4297 	 * Note: if you try to create a monitor device named "monN", and
4298 	 * there's already a "monN" device, it fails, as least with
4299 	 * the netlink interface (which is what iw uses), with a return
4300 	 * value of -ENFILE.  (Return values are negative errnos.)  We
4301 	 * could probably use that to find an unused device.
4302 	 */
4303 	int err;
4304 	struct iwreq ireq;
4305 	struct iw_priv_args *priv;
4306 	monitor_type montype;
4307 	int i;
4308 	__u32 cmd;
4309 	struct ifreq ifr;
4310 	int oldflags;
4311 	int args[2];
4312 	int channel;
4313 
4314 	/*
4315 	 * Does this device *support* the Wireless Extensions?
4316 	 */
4317 	err = has_wext(sock_fd, device, handle->errbuf);
4318 	if (err <= 0)
4319 		return err;	/* either it doesn't or the device doesn't even exist */
4320 	/*
4321 	 * Start out assuming we have no private extensions to control
4322 	 * radio metadata.
4323 	 */
4324 	montype = MONITOR_WEXT;
4325 	cmd = 0;
4326 
4327 	/*
4328 	 * Try to get all the Wireless Extensions private ioctls
4329 	 * supported by this device.
4330 	 *
4331 	 * First, get the size of the buffer we need, by supplying no
4332 	 * buffer and a length of 0.  If the device supports private
4333 	 * ioctls, it should return E2BIG, with ireq.u.data.length set
4334 	 * to the length we need.  If it doesn't support them, it should
4335 	 * return EOPNOTSUPP.
4336 	 */
4337 	memset(&ireq, 0, sizeof ireq);
4338 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4339 	    sizeof ireq.ifr_ifrn.ifrn_name);
4340 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4341 	ireq.u.data.pointer = (void *)args;
4342 	ireq.u.data.length = 0;
4343 	ireq.u.data.flags = 0;
4344 	if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
4345 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4346 		    "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!",
4347 		    device);
4348 		return PCAP_ERROR;
4349 	}
4350 	if (errno != EOPNOTSUPP) {
4351 		/*
4352 		 * OK, it's not as if there are no private ioctls.
4353 		 */
4354 		if (errno != E2BIG) {
4355 			/*
4356 			 * Failed.
4357 			 */
4358 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4359 			    "%s: SIOCGIWPRIV: %s", device,
4360 			    pcap_strerror(errno));
4361 			return PCAP_ERROR;
4362 		}
4363 
4364 		/*
4365 		 * OK, try to get the list of private ioctls.
4366 		 */
4367 		priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args));
4368 		if (priv == NULL) {
4369 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4370 			    "malloc: %s", pcap_strerror(errno));
4371 			return PCAP_ERROR;
4372 		}
4373 		ireq.u.data.pointer = (void *)priv;
4374 		if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) {
4375 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4376 			    "%s: SIOCGIWPRIV: %s", device,
4377 			    pcap_strerror(errno));
4378 			free(priv);
4379 			return PCAP_ERROR;
4380 		}
4381 
4382 		/*
4383 		 * Look for private ioctls to turn monitor mode on or, if
4384 		 * monitor mode is on, to set the header type.
4385 		 */
4386 		for (i = 0; i < ireq.u.data.length; i++) {
4387 			if (strcmp(priv[i].name, "monitor_type") == 0) {
4388 				/*
4389 				 * Hostap driver, use this one.
4390 				 * Set monitor mode first.
4391 				 * You can set it to 0 to get DLT_IEEE80211,
4392 				 * 1 to get DLT_PRISM, 2 to get
4393 				 * DLT_IEEE80211_RADIO_AVS, and, with more
4394 				 * recent versions of the driver, 3 to get
4395 				 * DLT_IEEE80211_RADIO.
4396 				 */
4397 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4398 					break;
4399 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4400 					break;
4401 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4402 					break;
4403 				montype = MONITOR_HOSTAP;
4404 				cmd = priv[i].cmd;
4405 				break;
4406 			}
4407 			if (strcmp(priv[i].name, "set_prismhdr") == 0) {
4408 				/*
4409 				 * Prism54 driver, use this one.
4410 				 * Set monitor mode first.
4411 				 * You can set it to 2 to get DLT_IEEE80211
4412 				 * or 3 or get DLT_PRISM.
4413 				 */
4414 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4415 					break;
4416 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4417 					break;
4418 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4419 					break;
4420 				montype = MONITOR_PRISM54;
4421 				cmd = priv[i].cmd;
4422 				break;
4423 			}
4424 			if (strcmp(priv[i].name, "forceprismheader") == 0) {
4425 				/*
4426 				 * RT2570 driver, use this one.
4427 				 * Do this after turning monitor mode on.
4428 				 * You can set it to 1 to get DLT_PRISM or 2
4429 				 * to get DLT_IEEE80211.
4430 				 */
4431 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4432 					break;
4433 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4434 					break;
4435 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4436 					break;
4437 				montype = MONITOR_RT2570;
4438 				cmd = priv[i].cmd;
4439 				break;
4440 			}
4441 			if (strcmp(priv[i].name, "forceprism") == 0) {
4442 				/*
4443 				 * RT73 driver, use this one.
4444 				 * Do this after turning monitor mode on.
4445 				 * Its argument is a *string*; you can
4446 				 * set it to "1" to get DLT_PRISM or "2"
4447 				 * to get DLT_IEEE80211.
4448 				 */
4449 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR)
4450 					break;
4451 				if (priv[i].set_args & IW_PRIV_SIZE_FIXED)
4452 					break;
4453 				montype = MONITOR_RT73;
4454 				cmd = priv[i].cmd;
4455 				break;
4456 			}
4457 			if (strcmp(priv[i].name, "prismhdr") == 0) {
4458 				/*
4459 				 * One of the RTL8xxx drivers, use this one.
4460 				 * It can only be done after monitor mode
4461 				 * has been turned on.  You can set it to 1
4462 				 * to get DLT_PRISM or 0 to get DLT_IEEE80211.
4463 				 */
4464 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4465 					break;
4466 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4467 					break;
4468 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4469 					break;
4470 				montype = MONITOR_RTL8XXX;
4471 				cmd = priv[i].cmd;
4472 				break;
4473 			}
4474 			if (strcmp(priv[i].name, "rfmontx") == 0) {
4475 				/*
4476 				 * RT2500 or RT61 driver, use this one.
4477 				 * It has one one-byte parameter; set
4478 				 * u.data.length to 1 and u.data.pointer to
4479 				 * point to the parameter.
4480 				 * It doesn't itself turn monitor mode on.
4481 				 * You can set it to 1 to allow transmitting
4482 				 * in monitor mode(?) and get DLT_IEEE80211,
4483 				 * or set it to 0 to disallow transmitting in
4484 				 * monitor mode(?) and get DLT_PRISM.
4485 				 */
4486 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4487 					break;
4488 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2)
4489 					break;
4490 				montype = MONITOR_RT2500;
4491 				cmd = priv[i].cmd;
4492 				break;
4493 			}
4494 			if (strcmp(priv[i].name, "monitor") == 0) {
4495 				/*
4496 				 * Either ACX100 or hostap, use this one.
4497 				 * It turns monitor mode on.
4498 				 * If it takes two arguments, it's ACX100;
4499 				 * the first argument is 1 for DLT_PRISM
4500 				 * or 2 for DLT_IEEE80211, and the second
4501 				 * argument is the channel on which to
4502 				 * run.  If it takes one argument, it's
4503 				 * HostAP, and the argument is 2 for
4504 				 * DLT_IEEE80211 and 3 for DLT_PRISM.
4505 				 *
4506 				 * If we see this, we don't quit, as this
4507 				 * might be a version of the hostap driver
4508 				 * that also supports "monitor_type".
4509 				 */
4510 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4511 					break;
4512 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4513 					break;
4514 				switch (priv[i].set_args & IW_PRIV_SIZE_MASK) {
4515 
4516 				case 1:
4517 					montype = MONITOR_PRISM;
4518 					cmd = priv[i].cmd;
4519 					break;
4520 
4521 				case 2:
4522 					montype = MONITOR_ACX100;
4523 					cmd = priv[i].cmd;
4524 					break;
4525 
4526 				default:
4527 					break;
4528 				}
4529 			}
4530 		}
4531 		free(priv);
4532 	}
4533 
4534 	/*
4535 	 * XXX - ipw3945?  islism?
4536 	 */
4537 
4538 	/*
4539 	 * Get the old mode.
4540 	 */
4541 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4542 	    sizeof ireq.ifr_ifrn.ifrn_name);
4543 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4544 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) {
4545 		/*
4546 		 * We probably won't be able to set the mode, either.
4547 		 */
4548 		return PCAP_ERROR_RFMON_NOTSUP;
4549 	}
4550 
4551 	/*
4552 	 * Is it currently in monitor mode?
4553 	 */
4554 	if (ireq.u.mode == IW_MODE_MONITOR) {
4555 		/*
4556 		 * Yes.  Just leave things as they are.
4557 		 * We don't offer multiple link-layer types, as
4558 		 * changing the link-layer type out from under
4559 		 * somebody else capturing in monitor mode would
4560 		 * be considered rude.
4561 		 */
4562 		return 1;
4563 	}
4564 	/*
4565 	 * No.  We have to put the adapter into rfmon mode.
4566 	 */
4567 
4568 	/*
4569 	 * If we haven't already done so, arrange to have
4570 	 * "pcap_close_all()" called when we exit.
4571 	 */
4572 	if (!pcap_do_addexit(handle)) {
4573 		/*
4574 		 * "atexit()" failed; don't put the interface
4575 		 * in rfmon mode, just give up.
4576 		 */
4577 		return PCAP_ERROR_RFMON_NOTSUP;
4578 	}
4579 
4580 	/*
4581 	 * Save the old mode.
4582 	 */
4583 	handle->md.oldmode = ireq.u.mode;
4584 
4585 	/*
4586 	 * Put the adapter in rfmon mode.  How we do this depends
4587 	 * on whether we have a special private ioctl or not.
4588 	 */
4589 	if (montype == MONITOR_PRISM) {
4590 		/*
4591 		 * We have the "monitor" private ioctl, but none of
4592 		 * the other private ioctls.  Use this, and select
4593 		 * the Prism header.
4594 		 *
4595 		 * If it fails, just fall back on SIOCSIWMODE.
4596 		 */
4597 		memset(&ireq, 0, sizeof ireq);
4598 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4599 		    sizeof ireq.ifr_ifrn.ifrn_name);
4600 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4601 		ireq.u.data.length = 1;	/* 1 argument */
4602 		args[0] = 3;	/* request Prism header */
4603 		memcpy(ireq.u.name, args, IFNAMSIZ);
4604 		if (ioctl(sock_fd, cmd, &ireq) != -1) {
4605 			/*
4606 			 * Success.
4607 			 * Note that we have to put the old mode back
4608 			 * when we close the device.
4609 			 */
4610 			handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
4611 
4612 			/*
4613 			 * Add this to the list of pcaps to close
4614 			 * when we exit.
4615 			 */
4616 			pcap_add_to_pcaps_to_close(handle);
4617 
4618 			return 1;
4619 		}
4620 
4621 		/*
4622 		 * Failure.  Fall back on SIOCSIWMODE.
4623 		 */
4624 	}
4625 
4626 	/*
4627 	 * First, take the interface down if it's up; otherwise, we
4628 	 * might get EBUSY.
4629 	 */
4630 	memset(&ifr, 0, sizeof(ifr));
4631 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4632 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
4633 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4634 		    "%s: Can't get flags: %s", device, strerror(errno));
4635 		return PCAP_ERROR;
4636 	}
4637 	oldflags = 0;
4638 	if (ifr.ifr_flags & IFF_UP) {
4639 		oldflags = ifr.ifr_flags;
4640 		ifr.ifr_flags &= ~IFF_UP;
4641 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4642 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4643 			    "%s: Can't set flags: %s", device, strerror(errno));
4644 			return PCAP_ERROR;
4645 		}
4646 	}
4647 
4648 	/*
4649 	 * Then turn monitor mode on.
4650 	 */
4651 	strncpy(ireq.ifr_ifrn.ifrn_name, device,
4652 	    sizeof ireq.ifr_ifrn.ifrn_name);
4653 	ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4654 	ireq.u.mode = IW_MODE_MONITOR;
4655 	if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) {
4656 		/*
4657 		 * Scientist, you've failed.
4658 		 * Bring the interface back up if we shut it down.
4659 		 */
4660 		ifr.ifr_flags = oldflags;
4661 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4662 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4663 			    "%s: Can't set flags: %s", device, strerror(errno));
4664 			return PCAP_ERROR;
4665 		}
4666 		return PCAP_ERROR_RFMON_NOTSUP;
4667 	}
4668 
4669 	/*
4670 	 * XXX - airmon-ng does "iwconfig {if} key off" after setting
4671 	 * monitor mode and setting the channel, and then does
4672 	 * "iwconfig up".
4673 	 */
4674 
4675 	/*
4676 	 * Now select the appropriate radio header.
4677 	 */
4678 	switch (montype) {
4679 
4680 	case MONITOR_WEXT:
4681 		/*
4682 		 * We don't have any private ioctl to set the header.
4683 		 */
4684 		break;
4685 
4686 	case MONITOR_HOSTAP:
4687 		/*
4688 		 * Try to select the radiotap header.
4689 		 */
4690 		memset(&ireq, 0, sizeof ireq);
4691 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4692 		    sizeof ireq.ifr_ifrn.ifrn_name);
4693 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4694 		args[0] = 3;	/* request radiotap header */
4695 		memcpy(ireq.u.name, args, sizeof (int));
4696 		if (ioctl(sock_fd, cmd, &ireq) != -1)
4697 			break;	/* success */
4698 
4699 		/*
4700 		 * That failed.  Try to select the AVS header.
4701 		 */
4702 		memset(&ireq, 0, sizeof ireq);
4703 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4704 		    sizeof ireq.ifr_ifrn.ifrn_name);
4705 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4706 		args[0] = 2;	/* request AVS header */
4707 		memcpy(ireq.u.name, args, sizeof (int));
4708 		if (ioctl(sock_fd, cmd, &ireq) != -1)
4709 			break;	/* success */
4710 
4711 		/*
4712 		 * That failed.  Try to select the Prism header.
4713 		 */
4714 		memset(&ireq, 0, sizeof ireq);
4715 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4716 		    sizeof ireq.ifr_ifrn.ifrn_name);
4717 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4718 		args[0] = 1;	/* request Prism header */
4719 		memcpy(ireq.u.name, args, sizeof (int));
4720 		ioctl(sock_fd, cmd, &ireq);
4721 		break;
4722 
4723 	case MONITOR_PRISM:
4724 		/*
4725 		 * The private ioctl failed.
4726 		 */
4727 		break;
4728 
4729 	case MONITOR_PRISM54:
4730 		/*
4731 		 * Select the Prism header.
4732 		 */
4733 		memset(&ireq, 0, sizeof ireq);
4734 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4735 		    sizeof ireq.ifr_ifrn.ifrn_name);
4736 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4737 		args[0] = 3;	/* request Prism header */
4738 		memcpy(ireq.u.name, args, sizeof (int));
4739 		ioctl(sock_fd, cmd, &ireq);
4740 		break;
4741 
4742 	case MONITOR_ACX100:
4743 		/*
4744 		 * Get the current channel.
4745 		 */
4746 		memset(&ireq, 0, sizeof ireq);
4747 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4748 		    sizeof ireq.ifr_ifrn.ifrn_name);
4749 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4750 		if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) {
4751 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4752 			    "%s: SIOCGIWFREQ: %s", device,
4753 			    pcap_strerror(errno));
4754 			return PCAP_ERROR;
4755 		}
4756 		channel = ireq.u.freq.m;
4757 
4758 		/*
4759 		 * Select the Prism header, and set the channel to the
4760 		 * current value.
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 		args[0] = 1;		/* request Prism header */
4767 		args[1] = channel;	/* set channel */
4768 		memcpy(ireq.u.name, args, 2*sizeof (int));
4769 		ioctl(sock_fd, cmd, &ireq);
4770 		break;
4771 
4772 	case MONITOR_RT2500:
4773 		/*
4774 		 * Disallow transmission - that turns on the
4775 		 * Prism header.
4776 		 */
4777 		memset(&ireq, 0, sizeof ireq);
4778 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4779 		    sizeof ireq.ifr_ifrn.ifrn_name);
4780 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4781 		args[0] = 0;	/* disallow transmitting */
4782 		memcpy(ireq.u.name, args, sizeof (int));
4783 		ioctl(sock_fd, cmd, &ireq);
4784 		break;
4785 
4786 	case MONITOR_RT2570:
4787 		/*
4788 		 * Force the Prism header.
4789 		 */
4790 		memset(&ireq, 0, sizeof ireq);
4791 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4792 		    sizeof ireq.ifr_ifrn.ifrn_name);
4793 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4794 		args[0] = 1;	/* request Prism header */
4795 		memcpy(ireq.u.name, args, sizeof (int));
4796 		ioctl(sock_fd, cmd, &ireq);
4797 		break;
4798 
4799 	case MONITOR_RT73:
4800 		/*
4801 		 * Force the Prism header.
4802 		 */
4803 		memset(&ireq, 0, sizeof ireq);
4804 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4805 		    sizeof ireq.ifr_ifrn.ifrn_name);
4806 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4807 		ireq.u.data.length = 1;	/* 1 argument */
4808 		ireq.u.data.pointer = "1";
4809 		ireq.u.data.flags = 0;
4810 		ioctl(sock_fd, cmd, &ireq);
4811 		break;
4812 
4813 	case MONITOR_RTL8XXX:
4814 		/*
4815 		 * Force the Prism header.
4816 		 */
4817 		memset(&ireq, 0, sizeof ireq);
4818 		strncpy(ireq.ifr_ifrn.ifrn_name, device,
4819 		    sizeof ireq.ifr_ifrn.ifrn_name);
4820 		ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
4821 		args[0] = 1;	/* request Prism header */
4822 		memcpy(ireq.u.name, args, sizeof (int));
4823 		ioctl(sock_fd, cmd, &ireq);
4824 		break;
4825 	}
4826 
4827 	/*
4828 	 * Now bring the interface back up if we brought it down.
4829 	 */
4830 	if (oldflags != 0) {
4831 		ifr.ifr_flags = oldflags;
4832 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4833 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4834 			    "%s: Can't set flags: %s", device, strerror(errno));
4835 
4836 			/*
4837 			 * At least try to restore the old mode on the
4838 			 * interface.
4839 			 */
4840 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
4841 				/*
4842 				 * Scientist, you've failed.
4843 				 */
4844 				fprintf(stderr,
4845 				    "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n"
4846 				    "Please adjust manually.\n",
4847 				    strerror(errno));
4848 			}
4849 			return PCAP_ERROR;
4850 		}
4851 	}
4852 
4853 	/*
4854 	 * Note that we have to put the old mode back when we
4855 	 * close the device.
4856 	 */
4857 	handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
4858 
4859 	/*
4860 	 * Add this to the list of pcaps to close when we exit.
4861 	 */
4862 	pcap_add_to_pcaps_to_close(handle);
4863 
4864 	return 1;
4865 }
4866 #endif /* IW_MODE_MONITOR */
4867 
4868 /*
4869  * Try various mechanisms to enter monitor mode.
4870  */
4871 static int
4872 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
4873 {
4874 #if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
4875 	int ret;
4876 #endif
4877 
4878 #ifdef HAVE_LIBNL
4879 	ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
4880 	if (ret < 0)
4881 		return ret;	/* error attempting to do so */
4882 	if (ret == 1)
4883 		return 1;	/* success */
4884 #endif /* HAVE_LIBNL */
4885 
4886 #ifdef IW_MODE_MONITOR
4887 	ret = enter_rfmon_mode_wext(handle, sock_fd, device);
4888 	if (ret < 0)
4889 		return ret;	/* error attempting to do so */
4890 	if (ret == 1)
4891 		return 1;	/* success */
4892 #endif /* IW_MODE_MONITOR */
4893 
4894 	/*
4895 	 * Either none of the mechanisms we know about work or none
4896 	 * of those mechanisms are available, so we can't do monitor
4897 	 * mode.
4898 	 */
4899 	return 0;
4900 }
4901 
4902 /*
4903  * Find out if we have any form of fragmentation/reassembly offloading.
4904  *
4905  * We do so using SIOCETHTOOL checking for various types of offloading;
4906  * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
4907  * of the types of offloading, there's nothing we can do to check, so
4908  * we just say "no, we don't".
4909  */
4910 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
4911 static int
4912 iface_ethtool_ioctl(pcap_t *handle, int cmd, const char *cmdname)
4913 {
4914 	struct ifreq	ifr;
4915 	struct ethtool_value eval;
4916 
4917 	memset(&ifr, 0, sizeof(ifr));
4918 	strncpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name));
4919 	eval.cmd = cmd;
4920 	ifr.ifr_data = (caddr_t)&eval;
4921 	if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
4922 		if (errno == EOPNOTSUPP) {
4923 			/*
4924 			 * OK, let's just return 0, which, in our
4925 			 * case, either means "no, what we're asking
4926 			 * about is not enabled" or "all the flags
4927 			 * are clear (i.e., nothing is enabled)".
4928 			 */
4929 			return 0;
4930 		}
4931 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4932 		    "%s: SIOETHTOOL(%s) ioctl failed: %s", handle->opt.source,
4933 		    cmdname, strerror(errno));
4934 		return -1;
4935 	}
4936 	return eval.data;
4937 }
4938 
4939 static int
4940 iface_get_offload(pcap_t *handle)
4941 {
4942 	int ret;
4943 
4944 #ifdef ETHTOOL_GTSO
4945 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO");
4946 	if (ret == -1)
4947 		return -1;
4948 	if (ret)
4949 		return 1;	/* TCP segmentation offloading on */
4950 #endif
4951 
4952 #ifdef ETHTOOL_GUFO
4953 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO");
4954 	if (ret == -1)
4955 		return -1;
4956 	if (ret)
4957 		return 1;	/* UDP fragmentation offloading on */
4958 #endif
4959 
4960 #ifdef ETHTOOL_GGSO
4961 	/*
4962 	 * XXX - will this cause large unsegmented packets to be
4963 	 * handed to PF_PACKET sockets on transmission?  If not,
4964 	 * this need not be checked.
4965 	 */
4966 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO");
4967 	if (ret == -1)
4968 		return -1;
4969 	if (ret)
4970 		return 1;	/* generic segmentation offloading on */
4971 #endif
4972 
4973 #ifdef ETHTOOL_GFLAGS
4974 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
4975 	if (ret == -1)
4976 		return -1;
4977 	if (ret & ETH_FLAG_LRO)
4978 		return 1;	/* large receive offloading on */
4979 #endif
4980 
4981 #ifdef ETHTOOL_GGRO
4982 	/*
4983 	 * XXX - will this cause large reassembled packets to be
4984 	 * handed to PF_PACKET sockets on receipt?  If not,
4985 	 * this need not be checked.
4986 	 */
4987 	ret = iface_ethtool_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO");
4988 	if (ret == -1)
4989 		return -1;
4990 	if (ret)
4991 		return 1;	/* generic (large) receive offloading on */
4992 #endif
4993 
4994 	return 0;
4995 }
4996 #else /* SIOCETHTOOL */
4997 static int
4998 iface_get_offload(pcap_t *handle _U_)
4999 {
5000 	/*
5001 	 * XXX - do we need to get this information if we don't
5002 	 * have the ethtool ioctls?  If so, how do we do that?
5003 	 */
5004 	return 0;
5005 }
5006 #endif /* SIOCETHTOOL */
5007 
5008 #endif /* HAVE_PF_PACKET_SOCKETS */
5009 
5010 /* ===== Functions to interface to the older kernels ================== */
5011 
5012 /*
5013  * Try to open a packet socket using the old kernel interface.
5014  * Returns 1 on success and a PCAP_ERROR_ value on an error.
5015  */
5016 static int
5017 activate_old(pcap_t *handle)
5018 {
5019 	int		arptype;
5020 	struct ifreq	ifr;
5021 	const char	*device = handle->opt.source;
5022 	struct utsname	utsname;
5023 	int		mtu;
5024 
5025 	/* Open the socket */
5026 
5027 	handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
5028 	if (handle->fd == -1) {
5029 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5030 			 "socket: %s", pcap_strerror(errno));
5031 		if (errno == EPERM || errno == EACCES) {
5032 			/*
5033 			 * You don't have permission to open the
5034 			 * socket.
5035 			 */
5036 			return PCAP_ERROR_PERM_DENIED;
5037 		} else {
5038 			/*
5039 			 * Other error.
5040 			 */
5041 			return PCAP_ERROR;
5042 		}
5043 	}
5044 
5045 	/* It worked - we are using the old interface */
5046 	handle->md.sock_packet = 1;
5047 
5048 	/* ...which means we get the link-layer header. */
5049 	handle->md.cooked = 0;
5050 
5051 	/* Bind to the given device */
5052 
5053 	if (strcmp(device, "any") == 0) {
5054 		strncpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
5055 			PCAP_ERRBUF_SIZE);
5056 		return PCAP_ERROR;
5057 	}
5058 	if (iface_bind_old(handle->fd, device, handle->errbuf) == -1)
5059 		return PCAP_ERROR;
5060 
5061 	/*
5062 	 * Try to get the link-layer type.
5063 	 */
5064 	arptype = iface_get_arptype(handle->fd, device, handle->errbuf);
5065 	if (arptype < 0)
5066 		return PCAP_ERROR;
5067 
5068 	/*
5069 	 * Try to find the DLT_ type corresponding to that
5070 	 * link-layer type.
5071 	 */
5072 	map_arphrd_to_dlt(handle, arptype, 0);
5073 	if (handle->linktype == -1) {
5074 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5075 			 "unknown arptype %d", arptype);
5076 		return PCAP_ERROR;
5077 	}
5078 
5079 	/* Go to promisc mode if requested */
5080 
5081 	if (handle->opt.promisc) {
5082 		memset(&ifr, 0, sizeof(ifr));
5083 		strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5084 		if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
5085 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5086 				 "SIOCGIFFLAGS: %s", pcap_strerror(errno));
5087 			return PCAP_ERROR;
5088 		}
5089 		if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
5090 			/*
5091 			 * Promiscuous mode isn't currently on,
5092 			 * so turn it on, and remember that
5093 			 * we should turn it off when the
5094 			 * pcap_t is closed.
5095 			 */
5096 
5097 			/*
5098 			 * If we haven't already done so, arrange
5099 			 * to have "pcap_close_all()" called when
5100 			 * we exit.
5101 			 */
5102 			if (!pcap_do_addexit(handle)) {
5103 				/*
5104 				 * "atexit()" failed; don't put
5105 				 * the interface in promiscuous
5106 				 * mode, just give up.
5107 				 */
5108 				return PCAP_ERROR;
5109 			}
5110 
5111 			ifr.ifr_flags |= IFF_PROMISC;
5112 			if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
5113 			        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5114 					 "SIOCSIFFLAGS: %s",
5115 					 pcap_strerror(errno));
5116 				return PCAP_ERROR;
5117 			}
5118 			handle->md.must_do_on_close |= MUST_CLEAR_PROMISC;
5119 
5120 			/*
5121 			 * Add this to the list of pcaps
5122 			 * to close when we exit.
5123 			 */
5124 			pcap_add_to_pcaps_to_close(handle);
5125 		}
5126 	}
5127 
5128 	/*
5129 	 * Compute the buffer size.
5130 	 *
5131 	 * We're using SOCK_PACKET, so this might be a 2.0[.x]
5132 	 * kernel, and might require special handling - check.
5133 	 */
5134 	if (uname(&utsname) < 0 ||
5135 	    strncmp(utsname.release, "2.0", 3) == 0) {
5136 		/*
5137 		 * Either we couldn't find out what kernel release
5138 		 * this is, or it's a 2.0[.x] kernel.
5139 		 *
5140 		 * In the 2.0[.x] kernel, a "recvfrom()" on
5141 		 * a SOCK_PACKET socket, with MSG_TRUNC set, will
5142 		 * return the number of bytes read, so if we pass
5143 		 * a length based on the snapshot length, it'll
5144 		 * return the number of bytes from the packet
5145 		 * copied to userland, not the actual length
5146 		 * of the packet.
5147 		 *
5148 		 * This means that, for example, the IP dissector
5149 		 * in tcpdump will get handed a packet length less
5150 		 * than the length in the IP header, and will
5151 		 * complain about "truncated-ip".
5152 		 *
5153 		 * So we don't bother trying to copy from the
5154 		 * kernel only the bytes in which we're interested,
5155 		 * but instead copy them all, just as the older
5156 		 * versions of libpcap for Linux did.
5157 		 *
5158 		 * The buffer therefore needs to be big enough to
5159 		 * hold the largest packet we can get from this
5160 		 * device.  Unfortunately, we can't get the MRU
5161 		 * of the network; we can only get the MTU.  The
5162 		 * MTU may be too small, in which case a packet larger
5163 		 * than the buffer size will be truncated *and* we
5164 		 * won't get the actual packet size.
5165 		 *
5166 		 * However, if the snapshot length is larger than
5167 		 * the buffer size based on the MTU, we use the
5168 		 * snapshot length as the buffer size, instead;
5169 		 * this means that with a sufficiently large snapshot
5170 		 * length we won't artificially truncate packets
5171 		 * to the MTU-based size.
5172 		 *
5173 		 * This mess just one of many problems with packet
5174 		 * capture on 2.0[.x] kernels; you really want a
5175 		 * 2.2[.x] or later kernel if you want packet capture
5176 		 * to work well.
5177 		 */
5178 		mtu = iface_get_mtu(handle->fd, device, handle->errbuf);
5179 		if (mtu == -1)
5180 			return PCAP_ERROR;
5181 		handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
5182 		if (handle->bufsize < handle->snapshot)
5183 			handle->bufsize = handle->snapshot;
5184 	} else {
5185 		/*
5186 		 * This is a 2.2[.x] or later kernel.
5187 		 *
5188 		 * We can safely pass "recvfrom()" a byte count
5189 		 * based on the snapshot length.
5190 		 */
5191 		handle->bufsize = handle->snapshot;
5192 	}
5193 
5194 	/*
5195 	 * Default value for offset to align link-layer payload
5196 	 * on a 4-byte boundary.
5197 	 */
5198 	handle->offset	 = 0;
5199 
5200 	return 1;
5201 }
5202 
5203 /*
5204  *  Bind the socket associated with FD to the given device using the
5205  *  interface of the old kernels.
5206  */
5207 static int
5208 iface_bind_old(int fd, const char *device, char *ebuf)
5209 {
5210 	struct sockaddr	saddr;
5211 	int		err;
5212 	socklen_t	errlen = sizeof(err);
5213 
5214 	memset(&saddr, 0, sizeof(saddr));
5215 	strncpy(saddr.sa_data, device, sizeof(saddr.sa_data));
5216 	if (bind(fd, &saddr, sizeof(saddr)) == -1) {
5217 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5218 			 "bind: %s", pcap_strerror(errno));
5219 		return -1;
5220 	}
5221 
5222 	/* Any pending errors, e.g., network is down? */
5223 
5224 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
5225 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5226 			"getsockopt: %s", pcap_strerror(errno));
5227 		return -1;
5228 	}
5229 
5230 	if (err > 0) {
5231 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5232 			"bind: %s", pcap_strerror(err));
5233 		return -1;
5234 	}
5235 
5236 	return 0;
5237 }
5238 
5239 
5240 /* ===== System calls available on all supported kernels ============== */
5241 
5242 /*
5243  *  Query the kernel for the MTU of the given interface.
5244  */
5245 static int
5246 iface_get_mtu(int fd, const char *device, char *ebuf)
5247 {
5248 	struct ifreq	ifr;
5249 
5250 	if (!device)
5251 		return BIGGER_THAN_ALL_MTUS;
5252 
5253 	memset(&ifr, 0, sizeof(ifr));
5254 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5255 
5256 	if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
5257 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5258 			 "SIOCGIFMTU: %s", pcap_strerror(errno));
5259 		return -1;
5260 	}
5261 
5262 	return ifr.ifr_mtu;
5263 }
5264 
5265 /*
5266  *  Get the hardware type of the given interface as ARPHRD_xxx constant.
5267  */
5268 static int
5269 iface_get_arptype(int fd, const char *device, char *ebuf)
5270 {
5271 	struct ifreq	ifr;
5272 
5273 	memset(&ifr, 0, sizeof(ifr));
5274 	strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5275 
5276 	if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
5277 		snprintf(ebuf, PCAP_ERRBUF_SIZE,
5278 			 "SIOCGIFHWADDR: %s", pcap_strerror(errno));
5279 		if (errno == ENODEV) {
5280 			/*
5281 			 * No such device.
5282 			 */
5283 			return PCAP_ERROR_NO_SUCH_DEVICE;
5284 		}
5285 		return PCAP_ERROR;
5286 	}
5287 
5288 	return ifr.ifr_hwaddr.sa_family;
5289 }
5290 
5291 #ifdef SO_ATTACH_FILTER
5292 static int
5293 fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
5294 {
5295 	size_t prog_size;
5296 	register int i;
5297 	register struct bpf_insn *p;
5298 	struct bpf_insn *f;
5299 	int len;
5300 
5301 	/*
5302 	 * Make a copy of the filter, and modify that copy if
5303 	 * necessary.
5304 	 */
5305 	prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
5306 	len = handle->fcode.bf_len;
5307 	f = (struct bpf_insn *)malloc(prog_size);
5308 	if (f == NULL) {
5309 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5310 			 "malloc: %s", pcap_strerror(errno));
5311 		return -1;
5312 	}
5313 	memcpy(f, handle->fcode.bf_insns, prog_size);
5314 	fcode->len = len;
5315 	fcode->filter = (struct sock_filter *) f;
5316 
5317 	for (i = 0; i < len; ++i) {
5318 		p = &f[i];
5319 		/*
5320 		 * What type of instruction is this?
5321 		 */
5322 		switch (BPF_CLASS(p->code)) {
5323 
5324 		case BPF_RET:
5325 			/*
5326 			 * It's a return instruction; are we capturing
5327 			 * in memory-mapped mode?
5328 			 */
5329 			if (!is_mmapped) {
5330 				/*
5331 				 * No; is the snapshot length a constant,
5332 				 * rather than the contents of the
5333 				 * accumulator?
5334 				 */
5335 				if (BPF_MODE(p->code) == BPF_K) {
5336 					/*
5337 					 * Yes - if the value to be returned,
5338 					 * i.e. the snapshot length, is
5339 					 * anything other than 0, make it
5340 					 * 65535, so that the packet is
5341 					 * truncated by "recvfrom()",
5342 					 * not by the filter.
5343 					 *
5344 					 * XXX - there's nothing we can
5345 					 * easily do if it's getting the
5346 					 * value from the accumulator; we'd
5347 					 * have to insert code to force
5348 					 * non-zero values to be 65535.
5349 					 */
5350 					if (p->k != 0)
5351 						p->k = 65535;
5352 				}
5353 			}
5354 			break;
5355 
5356 		case BPF_LD:
5357 		case BPF_LDX:
5358 			/*
5359 			 * It's a load instruction; is it loading
5360 			 * from the packet?
5361 			 */
5362 			switch (BPF_MODE(p->code)) {
5363 
5364 			case BPF_ABS:
5365 			case BPF_IND:
5366 			case BPF_MSH:
5367 				/*
5368 				 * Yes; are we in cooked mode?
5369 				 */
5370 				if (handle->md.cooked) {
5371 					/*
5372 					 * Yes, so we need to fix this
5373 					 * instruction.
5374 					 */
5375 					if (fix_offset(p) < 0) {
5376 						/*
5377 						 * We failed to do so.
5378 						 * Return 0, so our caller
5379 						 * knows to punt to userland.
5380 						 */
5381 						return 0;
5382 					}
5383 				}
5384 				break;
5385 			}
5386 			break;
5387 		}
5388 	}
5389 	return 1;	/* we succeeded */
5390 }
5391 
5392 static int
5393 fix_offset(struct bpf_insn *p)
5394 {
5395 	/*
5396 	 * What's the offset?
5397 	 */
5398 	if (p->k >= SLL_HDR_LEN) {
5399 		/*
5400 		 * It's within the link-layer payload; that starts at an
5401 		 * offset of 0, as far as the kernel packet filter is
5402 		 * concerned, so subtract the length of the link-layer
5403 		 * header.
5404 		 */
5405 		p->k -= SLL_HDR_LEN;
5406 	} else if (p->k == 14) {
5407 		/*
5408 		 * It's the protocol field; map it to the special magic
5409 		 * kernel offset for that field.
5410 		 */
5411 		p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5412 	} else {
5413 		/*
5414 		 * It's within the header, but it's not one of those
5415 		 * fields; we can't do that in the kernel, so punt
5416 		 * to userland.
5417 		 */
5418 		return -1;
5419 	}
5420 	return 0;
5421 }
5422 
5423 static int
5424 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
5425 {
5426 	int total_filter_on = 0;
5427 	int save_mode;
5428 	int ret;
5429 	int save_errno;
5430 
5431 	/*
5432 	 * The socket filter code doesn't discard all packets queued
5433 	 * up on the socket when the filter is changed; this means
5434 	 * that packets that don't match the new filter may show up
5435 	 * after the new filter is put onto the socket, if those
5436 	 * packets haven't yet been read.
5437 	 *
5438 	 * This means, for example, that if you do a tcpdump capture
5439 	 * with a filter, the first few packets in the capture might
5440 	 * be packets that wouldn't have passed the filter.
5441 	 *
5442 	 * We therefore discard all packets queued up on the socket
5443 	 * when setting a kernel filter.  (This isn't an issue for
5444 	 * userland filters, as the userland filtering is done after
5445 	 * packets are queued up.)
5446 	 *
5447 	 * To flush those packets, we put the socket in read-only mode,
5448 	 * and read packets from the socket until there are no more to
5449 	 * read.
5450 	 *
5451 	 * In order to keep that from being an infinite loop - i.e.,
5452 	 * to keep more packets from arriving while we're draining
5453 	 * the queue - we put the "total filter", which is a filter
5454 	 * that rejects all packets, onto the socket before draining
5455 	 * the queue.
5456 	 *
5457 	 * This code deliberately ignores any errors, so that you may
5458 	 * get bogus packets if an error occurs, rather than having
5459 	 * the filtering done in userland even if it could have been
5460 	 * done in the kernel.
5461 	 */
5462 	if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5463 		       &total_fcode, sizeof(total_fcode)) == 0) {
5464 		char drain[1];
5465 
5466 		/*
5467 		 * Note that we've put the total filter onto the socket.
5468 		 */
5469 		total_filter_on = 1;
5470 
5471 		/*
5472 		 * Save the socket's current mode, and put it in
5473 		 * non-blocking mode; we drain it by reading packets
5474 		 * until we get an error (which is normally a
5475 		 * "nothing more to be read" error).
5476 		 */
5477 		save_mode = fcntl(handle->fd, F_GETFL, 0);
5478 		if (save_mode != -1 &&
5479 		    fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) >= 0) {
5480 			while (recv(handle->fd, &drain, sizeof drain,
5481 			       MSG_TRUNC) >= 0)
5482 				;
5483 			save_errno = errno;
5484 			fcntl(handle->fd, F_SETFL, save_mode);
5485 			if (save_errno != EAGAIN) {
5486 				/* Fatal error */
5487 				reset_kernel_filter(handle);
5488 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5489 				 "recv: %s", pcap_strerror(save_errno));
5490 				return -2;
5491 			}
5492 		}
5493 	}
5494 
5495 	/*
5496 	 * Now attach the new filter.
5497 	 */
5498 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5499 			 fcode, sizeof(*fcode));
5500 	if (ret == -1 && total_filter_on) {
5501 		/*
5502 		 * Well, we couldn't set that filter on the socket,
5503 		 * but we could set the total filter on the socket.
5504 		 *
5505 		 * This could, for example, mean that the filter was
5506 		 * too big to put into the kernel, so we'll have to
5507 		 * filter in userland; in any case, we'll be doing
5508 		 * filtering in userland, so we need to remove the
5509 		 * total filter so we see packets.
5510 		 */
5511 		save_errno = errno;
5512 
5513 		/*
5514 		 * XXX - if this fails, we're really screwed;
5515 		 * we have the total filter on the socket,
5516 		 * and it won't come off.  What do we do then?
5517 		 */
5518 		reset_kernel_filter(handle);
5519 
5520 		errno = save_errno;
5521 	}
5522 	return ret;
5523 }
5524 
5525 static int
5526 reset_kernel_filter(pcap_t *handle)
5527 {
5528 	/*
5529 	 * setsockopt() barfs unless it get a dummy parameter.
5530 	 * valgrind whines unless the value is initialized,
5531 	 * as it has no idea that setsockopt() ignores its
5532 	 * parameter.
5533 	 */
5534 	int dummy = 0;
5535 
5536 	return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
5537 				   &dummy, sizeof(dummy));
5538 }
5539 #endif
5540