xref: /netbsd-src/external/bsd/libpcap/dist/pcap-linux.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: pcap-linux.c,v 1.4 2017/01/24 22:29:28 christos Exp $	*/
2 
3 /*
4  *  pcap-linux.c: Packet capture interface to the Linux kernel
5  *
6  *  Copyright (c) 2000 Torsten Landschoff <torsten@debian.org>
7  *  		       Sebastian Krahmer  <krahmer@cs.uni-potsdam.de>
8  *
9  *  License: BSD
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in
19  *     the documentation and/or other materials provided with the
20  *     distribution.
21  *  3. The names of the authors may not be used to endorse or promote
22  *     products derived from this software without specific prior
23  *     written permission.
24  *
25  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
26  *  IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
27  *  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
28  *
29  *  Modifications:     Added PACKET_MMAP support
30  *                     Paolo Abeni <paolo.abeni@email.it>
31  *                     Added TPACKET_V3 support
32  *                     Gabor Tatarka <gabor.tatarka@ericsson.com>
33  *
34  *                     based on previous works of:
35  *                     Simon Patarin <patarin@cs.unibo.it>
36  *                     Phil Wood <cpw@lanl.gov>
37  *
38  * Monitor-mode support for mac80211 includes code taken from the iw
39  * command; the copyright notice for that code is
40  *
41  * Copyright (c) 2007, 2008	Johannes Berg
42  * Copyright (c) 2007		Andy Lutomirski
43  * Copyright (c) 2007		Mike Kershaw
44  * Copyright (c) 2008		Gábor Stefanik
45  *
46  * All rights reserved.
47  *
48  * Redistribution and use in source and binary forms, with or without
49  * modification, are permitted provided that the following conditions
50  * are met:
51  * 1. Redistributions of source code must retain the above copyright
52  *    notice, this list of conditions and the following disclaimer.
53  * 2. Redistributions in binary form must reproduce the above copyright
54  *    notice, this list of conditions and the following disclaimer in the
55  *    documentation and/or other materials provided with the distribution.
56  * 3. The name of the author may not be used to endorse or promote products
57  *    derived from this software without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
60  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
61  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
62  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
63  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
64  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
65  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
66  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
67  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  */
71 
72 /*
73  * Known problems with 2.0[.x] kernels:
74  *
75  *   - The loopback device gives every packet twice; on 2.2[.x] kernels,
76  *     if we use PF_PACKET, we can filter out the transmitted version
77  *     of the packet by using data in the "sockaddr_ll" returned by
78  *     "recvfrom()", but, on 2.0[.x] kernels, we have to use
79  *     PF_INET/SOCK_PACKET, which means "recvfrom()" supplies a
80  *     "sockaddr_pkt" which doesn't give us enough information to let
81  *     us do that.
82  *
83  *   - We have to set the interface's IFF_PROMISC flag ourselves, if
84  *     we're to run in promiscuous mode, which means we have to turn
85  *     it off ourselves when we're done; the kernel doesn't keep track
86  *     of how many sockets are listening promiscuously, which means
87  *     it won't get turned off automatically when no sockets are
88  *     listening promiscuously.  We catch "pcap_close()" and, for
89  *     interfaces we put into promiscuous mode, take them out of
90  *     promiscuous mode - which isn't necessarily the right thing to
91  *     do, if another socket also requested promiscuous mode between
92  *     the time when we opened the socket and the time when we close
93  *     the socket.
94  *
95  *   - MSG_TRUNC isn't supported, so you can't specify that "recvfrom()"
96  *     return the amount of data that you could have read, rather than
97  *     the amount that was returned, so we can't just allocate a buffer
98  *     whose size is the snapshot length and pass the snapshot length
99  *     as the byte count, and also pass MSG_TRUNC, so that the return
100  *     value tells us how long the packet was on the wire.
101  *
102  *     This means that, if we want to get the actual size of the packet,
103  *     so we can return it in the "len" field of the packet header,
104  *     we have to read the entire packet, not just the part that fits
105  *     within the snapshot length, and thus waste CPU time copying data
106  *     from the kernel that our caller won't see.
107  *
108  *     We have to get the actual size, and supply it in "len", because
109  *     otherwise, the IP dissector in tcpdump, for example, will complain
110  *     about "truncated-ip", as the packet will appear to have been
111  *     shorter, on the wire, than the IP header said it should have been.
112  */
113 
114 
115 #define _GNU_SOURCE
116 
117 #include <sys/cdefs.h>
118 __RCSID("$NetBSD: pcap-linux.c,v 1.4 2017/01/24 22:29:28 christos Exp $");
119 
120 #ifdef HAVE_CONFIG_H
121 #include "config.h"
122 #endif
123 
124 #include <errno.h>
125 #include <stdio.h>
126 #include <stdlib.h>
127 #include <ctype.h>
128 #include <unistd.h>
129 #include <fcntl.h>
130 #include <string.h>
131 #include <limits.h>
132 #include <sys/stat.h>
133 #include <sys/socket.h>
134 #include <sys/ioctl.h>
135 #include <sys/utsname.h>
136 #include <sys/mman.h>
137 #include <linux/if.h>
138 #include <linux/if_packet.h>
139 #include <linux/sockios.h>
140 #include <netinet/in.h>
141 #include <linux/if_ether.h>
142 #include <net/if_arp.h>
143 #include <poll.h>
144 #include <dirent.h>
145 
146 #include "pcap-int.h"
147 #include "pcap/sll.h"
148 #include "pcap/vlan.h"
149 
150 /*
151  * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
152  * sockets rather than SOCK_PACKET sockets.
153  *
154  * To use them, we include <linux/if_packet.h> rather than
155  * <netpacket/packet.h>; we do so because
156  *
157  *	some Linux distributions (e.g., Slackware 4.0) have 2.2 or
158  *	later kernels and libc5, and don't provide a <netpacket/packet.h>
159  *	file;
160  *
161  *	not all versions of glibc2 have a <netpacket/packet.h> file
162  *	that defines stuff needed for some of the 2.4-or-later-kernel
163  *	features, so if the system has a 2.4 or later kernel, we
164  *	still can't use those features.
165  *
166  * We're already including a number of other <linux/XXX.h> headers, and
167  * this code is Linux-specific (no other OS has PF_PACKET sockets as
168  * a raw packet capture mechanism), so it's not as if you gain any
169  * useful portability by using <netpacket/packet.h>
170  *
171  * XXX - should we just include <linux/if_packet.h> even if PF_PACKET
172  * isn't defined?  It only defines one data structure in 2.0.x, so
173  * it shouldn't cause any problems.
174  */
175 #ifdef PF_PACKET
176 # include <linux/if_packet.h>
177 
178  /*
179   * On at least some Linux distributions (for example, Red Hat 5.2),
180   * there's no <netpacket/packet.h> file, but PF_PACKET is defined if
181   * you include <sys/socket.h>, but <linux/if_packet.h> doesn't define
182   * any of the PF_PACKET stuff such as "struct sockaddr_ll" or any of
183   * the PACKET_xxx stuff.
184   *
185   * So we check whether PACKET_HOST is defined, and assume that we have
186   * PF_PACKET sockets only if it is defined.
187   */
188 # ifdef PACKET_HOST
189 #  define HAVE_PF_PACKET_SOCKETS
190 #  ifdef PACKET_AUXDATA
191 #   define HAVE_PACKET_AUXDATA
192 #  endif /* PACKET_AUXDATA */
193 # endif /* PACKET_HOST */
194 
195 
196  /* check for memory mapped access avaibility. We assume every needed
197   * struct is defined if the macro TPACKET_HDRLEN is defined, because it
198   * uses many ring related structs and macros */
199 # ifdef PCAP_SUPPORT_PACKET_RING
200 # ifdef TPACKET_HDRLEN
201 #  define HAVE_PACKET_RING
202 #  ifdef TPACKET3_HDRLEN
203 #   define HAVE_TPACKET3
204 #  endif /* TPACKET3_HDRLEN */
205 #  ifdef TPACKET2_HDRLEN
206 #   define HAVE_TPACKET2
207 #  else  /* TPACKET2_HDRLEN */
208 #   define TPACKET_V1	0    /* Old kernel with only V1, so no TPACKET_Vn defined */
209 #  endif /* TPACKET2_HDRLEN */
210 # endif /* TPACKET_HDRLEN */
211 # endif /* PCAP_SUPPORT_PACKET_RING */
212 #endif /* PF_PACKET */
213 
214 #ifdef SO_ATTACH_FILTER
215 #include <linux/types.h>
216 #include <linux/filter.h>
217 #endif
218 
219 #ifdef HAVE_LINUX_NET_TSTAMP_H
220 #include <linux/net_tstamp.h>
221 #endif
222 
223 #ifdef HAVE_LINUX_SOCKIOS_H
224 #include <linux/sockios.h>
225 #endif
226 
227 #ifdef HAVE_LINUX_IF_BONDING_H
228 #include <linux/if_bonding.h>
229 
230 /*
231  * The ioctl code to use to check whether a device is a bonding device.
232  */
233 #if defined(SIOCBONDINFOQUERY)
234 	#define BOND_INFO_QUERY_IOCTL SIOCBONDINFOQUERY
235 #elif defined(BOND_INFO_QUERY_OLD)
236 	#define BOND_INFO_QUERY_IOCTL BOND_INFO_QUERY_OLD
237 #endif
238 #endif /* HAVE_LINUX_IF_BONDING_H */
239 
240 /*
241  * Got Wireless Extensions?
242  */
243 #ifdef HAVE_LINUX_WIRELESS_H
244 #include <linux/wireless.h>
245 #endif /* HAVE_LINUX_WIRELESS_H */
246 
247 /*
248  * Got libnl?
249  */
250 #ifdef HAVE_LIBNL
251 #include <linux/nl80211.h>
252 
253 #include <netlink/genl/genl.h>
254 #include <netlink/genl/family.h>
255 #include <netlink/genl/ctrl.h>
256 #include <netlink/msg.h>
257 #include <netlink/attr.h>
258 #endif /* HAVE_LIBNL */
259 
260 /*
261  * Got ethtool support?
262  */
263 #ifdef HAVE_LINUX_ETHTOOL_H
264 #include <linux/ethtool.h>
265 #endif
266 
267 #ifndef HAVE_SOCKLEN_T
268 typedef int		socklen_t;
269 #endif
270 
271 #ifndef MSG_TRUNC
272 /*
273  * This is being compiled on a system that lacks MSG_TRUNC; define it
274  * with the value it has in the 2.2 and later kernels, so that, on
275  * those kernels, when we pass it in the flags argument to "recvfrom()"
276  * we're passing the right value and thus get the MSG_TRUNC behavior
277  * we want.  (We don't get that behavior on 2.0[.x] kernels, because
278  * they didn't support MSG_TRUNC.)
279  */
280 #define MSG_TRUNC	0x20
281 #endif
282 
283 #ifndef SOL_PACKET
284 /*
285  * This is being compiled on a system that lacks SOL_PACKET; define it
286  * with the value it has in the 2.2 and later kernels, so that we can
287  * set promiscuous mode in the good modern way rather than the old
288  * 2.0-kernel crappy way.
289  */
290 #define SOL_PACKET	263
291 #endif
292 
293 #define MAX_LINKHEADER_SIZE	256
294 
295 /*
296  * When capturing on all interfaces we use this as the buffer size.
297  * Should be bigger then all MTUs that occur in real life.
298  * 64kB should be enough for now.
299  */
300 #define BIGGER_THAN_ALL_MTUS	(64*1024)
301 
302 /*
303  * Private data for capturing on Linux SOCK_PACKET or PF_PACKET sockets.
304  */
305 struct pcap_linux {
306 	u_int	packets_read;	/* count of packets read with recvfrom() */
307 	long	proc_dropped;	/* packets reported dropped by /proc/net/dev */
308 	struct pcap_stat stat;
309 
310 	char	*device;	/* device name */
311 	int	filter_in_userland; /* must filter in userland */
312 	int	blocks_to_filter_in_userland;
313 	int	must_do_on_close; /* stuff we must do when we close */
314 	int	timeout;	/* timeout for buffering */
315 	int	sock_packet;	/* using Linux 2.0 compatible interface */
316 	int	cooked;		/* using SOCK_DGRAM rather than SOCK_RAW */
317 	int	ifindex;	/* interface index of device we're bound to */
318 	int	lo_ifindex;	/* interface index of the loopback device */
319 	bpf_u_int32 oldmode;	/* mode to restore when turning monitor mode off */
320 	char	*mondevice;	/* mac80211 monitor device we created */
321 	u_char	*mmapbuf;	/* memory-mapped region pointer */
322 	size_t	mmapbuflen;	/* size of region */
323 	int	vlan_offset;	/* offset at which to insert vlan tags; if -1, don't insert */
324 	u_int	tp_version;	/* version of tpacket_hdr for mmaped ring */
325 	u_int	tp_hdrlen;	/* hdrlen of tpacket_hdr for mmaped ring */
326 	u_char	*oneshot_buffer; /* buffer for copy of packet */
327 	int	poll_timeout;	/* timeout to use in poll() */
328 #ifdef HAVE_TPACKET3
329 	unsigned char *current_packet; /* Current packet within the TPACKET_V3 block. Move to next block if NULL. */
330 	int packets_left; /* Unhandled packets left within the block from previous call to pcap_read_linux_mmap_v3 in case of TPACKET_V3. */
331 #endif
332 };
333 
334 /*
335  * Stuff to do when we close.
336  */
337 #define MUST_CLEAR_PROMISC	0x00000001	/* clear promiscuous mode */
338 #define MUST_CLEAR_RFMON	0x00000002	/* clear rfmon (monitor) mode */
339 #define MUST_DELETE_MONIF	0x00000004	/* delete monitor-mode interface */
340 
341 /*
342  * Prototypes for internal functions and methods.
343  */
344 static void map_arphrd_to_dlt(pcap_t *, int, int, const char *, int);
345 #ifdef HAVE_PF_PACKET_SOCKETS
346 static short int map_packet_type_to_sll_type(short int);
347 #endif
348 static int pcap_activate_linux(pcap_t *);
349 static int activate_old(pcap_t *);
350 static int activate_new(pcap_t *);
351 static int activate_mmap(pcap_t *, int *);
352 static int pcap_can_set_rfmon_linux(pcap_t *);
353 static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *);
354 static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);
355 static int pcap_inject_linux(pcap_t *, const void *, size_t);
356 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
357 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
358 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
359 static int pcap_set_datalink_linux(pcap_t *, int);
360 static void pcap_cleanup_linux(pcap_t *);
361 
362 /*
363  * This is what the header structure looks like in a 64-bit kernel;
364  * we use this, rather than struct tpacket_hdr, if we're using
365  * TPACKET_V1 in 32-bit code running on a 64-bit kernel.
366  */
367 struct tpacket_hdr_64 {
368 	uint64_t	tp_status;
369 	unsigned int	tp_len;
370 	unsigned int	tp_snaplen;
371 	unsigned short	tp_mac;
372 	unsigned short	tp_net;
373 	unsigned int	tp_sec;
374 	unsigned int	tp_usec;
375 };
376 
377 /*
378  * We use this internally as the tpacket version for TPACKET_V1 in
379  * 32-bit code on a 64-bit kernel.
380  */
381 #define TPACKET_V1_64 99
382 
383 union thdr {
384 	struct tpacket_hdr		*h1;
385 	struct tpacket_hdr_64		*h1_64;
386 #ifdef HAVE_TPACKET2
387 	struct tpacket2_hdr		*h2;
388 #endif
389 #ifdef HAVE_TPACKET3
390 	struct tpacket_block_desc	*h3;
391 #endif
392 	void				*raw;
393 };
394 
395 #ifdef HAVE_PACKET_RING
396 #define RING_GET_FRAME_AT(h, offset) (((union thdr **)h->buffer)[(offset)])
397 #define RING_GET_CURRENT_FRAME(h) RING_GET_FRAME_AT(h, h->offset)
398 
399 static void destroy_ring(pcap_t *handle);
400 static int create_ring(pcap_t *handle, int *status);
401 static int prepare_tpacket_socket(pcap_t *handle);
402 static void pcap_cleanup_linux_mmap(pcap_t *);
403 static int pcap_read_linux_mmap_v1(pcap_t *, int, pcap_handler , u_char *);
404 static int pcap_read_linux_mmap_v1_64(pcap_t *, int, pcap_handler , u_char *);
405 #ifdef HAVE_TPACKET2
406 static int pcap_read_linux_mmap_v2(pcap_t *, int, pcap_handler , u_char *);
407 #endif
408 #ifdef HAVE_TPACKET3
409 static int pcap_read_linux_mmap_v3(pcap_t *, int, pcap_handler , u_char *);
410 #endif
411 static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *);
412 static int pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf);
413 static int pcap_getnonblock_mmap(pcap_t *p, char *errbuf);
414 static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
415     const u_char *bytes);
416 #endif
417 
418 #ifdef TP_STATUS_VLAN_TPID_VALID
419 # define VLAN_TPID(hdr, hv)	(((hv)->tp_vlan_tpid || ((hdr)->tp_status & TP_STATUS_VLAN_TPID_VALID)) ? (hv)->tp_vlan_tpid : ETH_P_8021Q)
420 #else
421 # define VLAN_TPID(hdr, hv)	ETH_P_8021Q
422 #endif
423 
424 /*
425  * Wrap some ioctl calls
426  */
427 #ifdef HAVE_PF_PACKET_SOCKETS
428 static int	iface_get_id(int fd, const char *device, char *ebuf);
429 #endif /* HAVE_PF_PACKET_SOCKETS */
430 static int	iface_get_mtu(int fd, const char *device, char *ebuf);
431 static int 	iface_get_arptype(int fd, const char *device, char *ebuf);
432 #ifdef HAVE_PF_PACKET_SOCKETS
433 static int 	iface_bind(int fd, int ifindex, char *ebuf);
434 #ifdef IW_MODE_MONITOR
435 static int	has_wext(int sock_fd, const char *device, char *ebuf);
436 #endif /* IW_MODE_MONITOR */
437 static int	enter_rfmon_mode(pcap_t *handle, int sock_fd,
438     const char *device);
439 #endif /* HAVE_PF_PACKET_SOCKETS */
440 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
441 static int	iface_ethtool_get_ts_info(const char *device, pcap_t *handle,
442     char *ebuf);
443 #endif
444 #ifdef HAVE_PACKET_RING
445 static int	iface_get_offload(pcap_t *handle);
446 #endif
447 static int 	iface_bind_old(int fd, const char *device, char *ebuf);
448 
449 #ifdef SO_ATTACH_FILTER
450 static int	fix_program(pcap_t *handle, struct sock_fprog *fcode,
451     int is_mapped);
452 static int	fix_offset(struct bpf_insn *p);
453 static int	set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
454 static int	reset_kernel_filter(pcap_t *handle);
455 
456 static struct sock_filter	total_insn
457 	= BPF_STMT(BPF_RET | BPF_K, 0);
458 static struct sock_fprog	total_fcode
459 	= { 1, &total_insn };
460 #endif /* SO_ATTACH_FILTER */
461 
462 pcap_t *
463 pcap_create_interface(const char *device, char *ebuf)
464 {
465 	pcap_t *handle;
466 
467 	handle = pcap_create_common(ebuf, sizeof (struct pcap_linux));
468 	if (handle == NULL)
469 		return NULL;
470 
471 	handle->activate_op = pcap_activate_linux;
472 	handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
473 
474 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
475 	/*
476 	 * See what time stamp types we support.
477 	 */
478 	if (iface_ethtool_get_ts_info(device, handle, ebuf) == -1) {
479 		pcap_close(handle);
480 		return NULL;
481 	}
482 #endif
483 
484 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
485 	/*
486 	 * We claim that we support microsecond and nanosecond time
487 	 * stamps.
488 	 *
489 	 * XXX - with adapter-supplied time stamps, can we choose
490 	 * microsecond or nanosecond time stamps on arbitrary
491 	 * adapters?
492 	 */
493 	handle->tstamp_precision_count = 2;
494 	handle->tstamp_precision_list = malloc(2 * sizeof(u_int));
495 	if (handle->tstamp_precision_list == NULL) {
496 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
497 		    pcap_strerror(errno));
498 		pcap_close(handle);
499 		return NULL;
500 	}
501 	handle->tstamp_precision_list[0] = PCAP_TSTAMP_PRECISION_MICRO;
502 	handle->tstamp_precision_list[1] = PCAP_TSTAMP_PRECISION_NANO;
503 #endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */
504 
505 	return handle;
506 }
507 
508 #ifdef HAVE_LIBNL
509 /*
510  * If interface {if} is a mac80211 driver, the file
511  * /sys/class/net/{if}/phy80211 is a symlink to
512  * /sys/class/ieee80211/{phydev}, for some {phydev}.
513  *
514  * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
515  * least, has a "wmaster0" device and a "wlan0" device; the
516  * latter is the one with the IP address.  Both show up in
517  * "tcpdump -D" output.  Capturing on the wmaster0 device
518  * captures with 802.11 headers.
519  *
520  * airmon-ng searches through /sys/class/net for devices named
521  * monN, starting with mon0; as soon as one *doesn't* exist,
522  * it chooses that as the monitor device name.  If the "iw"
523  * command exists, it does "iw dev {if} interface add {monif}
524  * type monitor", where {monif} is the monitor device.  It
525  * then (sigh) sleeps .1 second, and then configures the
526  * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
527  * is a file, it writes {mondev}, without a newline, to that file,
528  * and again (sigh) sleeps .1 second, and then iwconfig's that
529  * device into monitor mode and configures it up.  Otherwise,
530  * you can't do monitor mode.
531  *
532  * All these devices are "glued" together by having the
533  * /sys/class/net/{device}/phy80211 links pointing to the same
534  * place, so, given a wmaster, wlan, or mon device, you can
535  * find the other devices by looking for devices with
536  * the same phy80211 link.
537  *
538  * To turn monitor mode off, delete the monitor interface,
539  * either with "iw dev {monif} interface del" or by sending
540  * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
541  *
542  * Note: if you try to create a monitor device named "monN", and
543  * there's already a "monN" device, it fails, as least with
544  * the netlink interface (which is what iw uses), with a return
545  * value of -ENFILE.  (Return values are negative errnos.)  We
546  * could probably use that to find an unused device.
547  *
548  * Yes, you can have multiple monitor devices for a given
549  * physical device.
550  */
551 
552 /*
553  * Is this a mac80211 device?  If so, fill in the physical device path and
554  * return 1; if not, return 0.  On an error, fill in handle->errbuf and
555  * return PCAP_ERROR.
556  */
557 static int
558 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
559     size_t phydev_max_pathlen)
560 {
561 	char *pathstr;
562 	ssize_t bytes_read;
563 
564 	/*
565 	 * Generate the path string for the symlink to the physical device.
566 	 */
567 	if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
568 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
569 		    "%s: Can't generate path name string for /sys/class/net device",
570 		    device);
571 		return PCAP_ERROR;
572 	}
573 	bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
574 	if (bytes_read == -1) {
575 		if (errno == ENOENT || errno == EINVAL) {
576 			/*
577 			 * Doesn't exist, or not a symlink; assume that
578 			 * means it's not a mac80211 device.
579 			 */
580 			free(pathstr);
581 			return 0;
582 		}
583 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
584 		    "%s: Can't readlink %s: %s", device, pathstr,
585 		    strerror(errno));
586 		free(pathstr);
587 		return PCAP_ERROR;
588 	}
589 	free(pathstr);
590 	phydev_path[bytes_read] = '\0';
591 	return 1;
592 }
593 
594 #ifdef HAVE_LIBNL_SOCKETS
595 #define get_nl_errmsg	nl_geterror
596 #else
597 /* libnl 2.x compatibility code */
598 
599 #define nl_sock nl_handle
600 
601 static inline struct nl_handle *
602 nl_socket_alloc(void)
603 {
604 	return nl_handle_alloc();
605 }
606 
607 static inline void
608 nl_socket_free(struct nl_handle *h)
609 {
610 	nl_handle_destroy(h);
611 }
612 
613 #define get_nl_errmsg	strerror
614 
615 static inline int
616 __genl_ctrl_alloc_cache(struct nl_handle *h, struct nl_cache **cache)
617 {
618 	struct nl_cache *tmp = genl_ctrl_alloc_cache(h);
619 	if (!tmp)
620 		return -ENOMEM;
621 	*cache = tmp;
622 	return 0;
623 }
624 #define genl_ctrl_alloc_cache __genl_ctrl_alloc_cache
625 #endif /* !HAVE_LIBNL_SOCKETS */
626 
627 struct nl80211_state {
628 	struct nl_sock *nl_sock;
629 	struct nl_cache *nl_cache;
630 	struct genl_family *nl80211;
631 };
632 
633 static int
634 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
635 {
636 	int err;
637 
638 	state->nl_sock = nl_socket_alloc();
639 	if (!state->nl_sock) {
640 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
641 		    "%s: failed to allocate netlink handle", device);
642 		return PCAP_ERROR;
643 	}
644 
645 	if (genl_connect(state->nl_sock)) {
646 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
647 		    "%s: failed to connect to generic netlink", device);
648 		goto out_handle_destroy;
649 	}
650 
651 	err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
652 	if (err < 0) {
653 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
654 		    "%s: failed to allocate generic netlink cache: %s",
655 		    device, get_nl_errmsg(-err));
656 		goto out_handle_destroy;
657 	}
658 
659 	state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
660 	if (!state->nl80211) {
661 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
662 		    "%s: nl80211 not found", device);
663 		goto out_cache_free;
664 	}
665 
666 	return 0;
667 
668 out_cache_free:
669 	nl_cache_free(state->nl_cache);
670 out_handle_destroy:
671 	nl_socket_free(state->nl_sock);
672 	return PCAP_ERROR;
673 }
674 
675 static void
676 nl80211_cleanup(struct nl80211_state *state)
677 {
678 	genl_family_put(state->nl80211);
679 	nl_cache_free(state->nl_cache);
680 	nl_socket_free(state->nl_sock);
681 }
682 
683 static int
684 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
685     const char *device, const char *mondevice);
686 
687 static int
688 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
689     const char *device, const char *mondevice)
690 {
691 	struct pcap_linux *handlep = handle->priv;
692 	int ifindex;
693 	struct nl_msg *msg;
694 	int err;
695 
696 	ifindex = iface_get_id(sock_fd, device, handle->errbuf);
697 	if (ifindex == -1)
698 		return PCAP_ERROR;
699 
700 	msg = nlmsg_alloc();
701 	if (!msg) {
702 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
703 		    "%s: failed to allocate netlink msg", device);
704 		return PCAP_ERROR;
705 	}
706 
707 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
708 		    0, NL80211_CMD_NEW_INTERFACE, 0);
709 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
710 	NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
711 	NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
712 
713 	err = nl_send_auto_complete(state->nl_sock, msg);
714 	if (err < 0) {
715 #if defined HAVE_LIBNL_NLE
716 		if (err == -NLE_FAILURE) {
717 #else
718 		if (err == -ENFILE) {
719 #endif
720 			/*
721 			 * Device not available; our caller should just
722 			 * keep trying.  (libnl 2.x maps ENFILE to
723 			 * NLE_FAILURE; it can also map other errors
724 			 * to that, but there's not much we can do
725 			 * about that.)
726 			 */
727 			nlmsg_free(msg);
728 			return 0;
729 		} else {
730 			/*
731 			 * Real failure, not just "that device is not
732 			 * available.
733 			 */
734 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
735 			    "%s: nl_send_auto_complete failed adding %s interface: %s",
736 			    device, mondevice, get_nl_errmsg(-err));
737 			nlmsg_free(msg);
738 			return PCAP_ERROR;
739 		}
740 	}
741 	err = nl_wait_for_ack(state->nl_sock);
742 	if (err < 0) {
743 #if defined HAVE_LIBNL_NLE
744 		if (err == -NLE_FAILURE) {
745 #else
746 		if (err == -ENFILE) {
747 #endif
748 			/*
749 			 * Device not available; our caller should just
750 			 * keep trying.  (libnl 2.x maps ENFILE to
751 			 * NLE_FAILURE; it can also map other errors
752 			 * to that, but there's not much we can do
753 			 * about that.)
754 			 */
755 			nlmsg_free(msg);
756 			return 0;
757 		} else {
758 			/*
759 			 * Real failure, not just "that device is not
760 			 * available.
761 			 */
762 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
763 			    "%s: nl_wait_for_ack failed adding %s interface: %s",
764 			    device, mondevice, get_nl_errmsg(-err));
765 			nlmsg_free(msg);
766 			return PCAP_ERROR;
767 		}
768 	}
769 
770 	/*
771 	 * Success.
772 	 */
773 	nlmsg_free(msg);
774 
775 	/*
776 	 * Try to remember the monitor device.
777 	 */
778 	handlep->mondevice = strdup(mondevice);
779 	if (handlep->mondevice == NULL) {
780 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
781 			 pcap_strerror(errno));
782 		/*
783 		 * Get rid of the monitor device.
784 		 */
785 		del_mon_if(handle, sock_fd, state, device, mondevice);
786 		return PCAP_ERROR;
787 	}
788 	return 1;
789 
790 nla_put_failure:
791 	pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
792 	    "%s: nl_put failed adding %s interface",
793 	    device, mondevice);
794 	nlmsg_free(msg);
795 	return PCAP_ERROR;
796 }
797 
798 static int
799 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
800     const char *device, const char *mondevice)
801 {
802 	int ifindex;
803 	struct nl_msg *msg;
804 	int err;
805 
806 	ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
807 	if (ifindex == -1)
808 		return PCAP_ERROR;
809 
810 	msg = nlmsg_alloc();
811 	if (!msg) {
812 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
813 		    "%s: failed to allocate netlink msg", device);
814 		return PCAP_ERROR;
815 	}
816 
817 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
818 		    0, NL80211_CMD_DEL_INTERFACE, 0);
819 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
820 
821 	err = nl_send_auto_complete(state->nl_sock, msg);
822 	if (err < 0) {
823 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
824 		    "%s: nl_send_auto_complete failed deleting %s interface: %s",
825 		    device, mondevice, get_nl_errmsg(-err));
826 		nlmsg_free(msg);
827 		return PCAP_ERROR;
828 	}
829 	err = nl_wait_for_ack(state->nl_sock);
830 	if (err < 0) {
831 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
832 		    "%s: nl_wait_for_ack failed adding %s interface: %s",
833 		    device, mondevice, get_nl_errmsg(-err));
834 		nlmsg_free(msg);
835 		return PCAP_ERROR;
836 	}
837 
838 	/*
839 	 * Success.
840 	 */
841 	nlmsg_free(msg);
842 	return 1;
843 
844 nla_put_failure:
845 	pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
846 	    "%s: nl_put failed deleting %s interface",
847 	    device, mondevice);
848 	nlmsg_free(msg);
849 	return PCAP_ERROR;
850 }
851 
852 static int
853 enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
854 {
855 	struct pcap_linux *handlep = handle->priv;
856 	int ret;
857 	char phydev_path[PATH_MAX+1];
858 	struct nl80211_state nlstate;
859 	struct ifreq ifr;
860 	u_int n;
861 
862 	/*
863 	 * Is this a mac80211 device?
864 	 */
865 	ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
866 	if (ret < 0)
867 		return ret;	/* error */
868 	if (ret == 0)
869 		return 0;	/* no error, but not mac80211 device */
870 
871 	/*
872 	 * XXX - is this already a monN device?
873 	 * If so, we're done.
874 	 * Is that determined by old Wireless Extensions ioctls?
875 	 */
876 
877 	/*
878 	 * OK, it's apparently a mac80211 device.
879 	 * Try to find an unused monN device for it.
880 	 */
881 	ret = nl80211_init(handle, &nlstate, device);
882 	if (ret != 0)
883 		return ret;
884 	for (n = 0; n < UINT_MAX; n++) {
885 		/*
886 		 * Try mon{n}.
887 		 */
888 		char mondevice[3+10+1];	/* mon{UINT_MAX}\0 */
889 
890 		pcap_snprintf(mondevice, sizeof mondevice, "mon%u", n);
891 		ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
892 		if (ret == 1) {
893 			/*
894 			 * Success.  We don't clean up the libnl state
895 			 * yet, as we'll be using it later.
896 			 */
897 			goto added;
898 		}
899 		if (ret < 0) {
900 			/*
901 			 * Hard failure.  Just return ret; handle->errbuf
902 			 * has already been set.
903 			 */
904 			nl80211_cleanup(&nlstate);
905 			return ret;
906 		}
907 	}
908 
909 	pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
910 	    "%s: No free monN interfaces", device);
911 	nl80211_cleanup(&nlstate);
912 	return PCAP_ERROR;
913 
914 added:
915 
916 #if 0
917 	/*
918 	 * Sleep for .1 seconds.
919 	 */
920 	delay.tv_sec = 0;
921 	delay.tv_nsec = 500000000;
922 	nanosleep(&delay, NULL);
923 #endif
924 
925 	/*
926 	 * If we haven't already done so, arrange to have
927 	 * "pcap_close_all()" called when we exit.
928 	 */
929 	if (!pcap_do_addexit(handle)) {
930 		/*
931 		 * "atexit()" failed; don't put the interface
932 		 * in rfmon mode, just give up.
933 		 */
934 		del_mon_if(handle, sock_fd, &nlstate, device,
935 		    handlep->mondevice);
936 		nl80211_cleanup(&nlstate);
937 		return PCAP_ERROR;
938 	}
939 
940 	/*
941 	 * Now configure the monitor interface up.
942 	 */
943 	memset(&ifr, 0, sizeof(ifr));
944 	strlcpy(ifr.ifr_name, handlep->mondevice, sizeof(ifr.ifr_name));
945 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
946 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
947 		    "%s: Can't get flags for %s: %s", device,
948 		    handlep->mondevice, strerror(errno));
949 		del_mon_if(handle, sock_fd, &nlstate, device,
950 		    handlep->mondevice);
951 		nl80211_cleanup(&nlstate);
952 		return PCAP_ERROR;
953 	}
954 	ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
955 	if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
956 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
957 		    "%s: Can't set flags for %s: %s", device,
958 		    handlep->mondevice, strerror(errno));
959 		del_mon_if(handle, sock_fd, &nlstate, device,
960 		    handlep->mondevice);
961 		nl80211_cleanup(&nlstate);
962 		return PCAP_ERROR;
963 	}
964 
965 	/*
966 	 * Success.  Clean up the libnl state.
967 	 */
968 	nl80211_cleanup(&nlstate);
969 
970 	/*
971 	 * Note that we have to delete the monitor device when we close
972 	 * the handle.
973 	 */
974 	handlep->must_do_on_close |= MUST_DELETE_MONIF;
975 
976 	/*
977 	 * Add this to the list of pcaps to close when we exit.
978 	 */
979 	pcap_add_to_pcaps_to_close(handle);
980 
981 	return 1;
982 }
983 #endif /* HAVE_LIBNL */
984 
985 #ifdef IW_MODE_MONITOR
986 /*
987  * Bonding devices mishandle unknown ioctls; they fail with ENODEV
988  * rather than ENOTSUP, EOPNOTSUPP, or ENOTTY, so Wireless Extensions
989  * will fail with ENODEV if we try to do them on a bonding device,
990  * making us return a "no such device" indication rather than just
991  * saying "no Wireless Extensions".
992  *
993  * So we check for bonding devices, if we can, before trying those
994  * ioctls, by trying a bonding device information query ioctl to see
995  * whether it succeeds.
996  */
997 static int
998 is_bonding_device(int fd, const char *device)
999 {
1000 #ifdef BOND_INFO_QUERY_IOCTL
1001 	struct ifreq ifr;
1002 	ifbond ifb;
1003 
1004 	memset(&ifr, 0, sizeof ifr);
1005 	strlcpy(ifr.ifr_name, device, sizeof ifr.ifr_name);
1006 	memset(&ifb, 0, sizeof ifb);
1007 	ifr.ifr_data = (caddr_t)&ifb;
1008 	if (ioctl(fd, BOND_INFO_QUERY_IOCTL, &ifr) == 0)
1009 		return 1;	/* success, so it's a bonding device */
1010 #endif /* BOND_INFO_QUERY_IOCTL */
1011 
1012 	return 0;	/* no, it's not a bonding device */
1013 }
1014 #endif /* IW_MODE_MONITOR */
1015 
1016 static int
1017 pcap_can_set_rfmon_linux(pcap_t *handle)
1018 {
1019 #ifdef HAVE_LIBNL
1020 	char phydev_path[PATH_MAX+1];
1021 	int ret;
1022 #endif
1023 #ifdef IW_MODE_MONITOR
1024 	int sock_fd;
1025 	struct iwreq ireq;
1026 #endif
1027 
1028 	if (strcmp(handle->opt.device, "any") == 0) {
1029 		/*
1030 		 * Monitor mode makes no sense on the "any" device.
1031 		 */
1032 		return 0;
1033 	}
1034 
1035 #ifdef HAVE_LIBNL
1036 	/*
1037 	 * Bleah.  There doesn't seem to be a way to ask a mac80211
1038 	 * device, through libnl, whether it supports monitor mode;
1039 	 * we'll just check whether the device appears to be a
1040 	 * mac80211 device and, if so, assume the device supports
1041 	 * monitor mode.
1042 	 *
1043 	 * wmaster devices don't appear to support the Wireless
1044 	 * Extensions, but we can create a mon device for a
1045 	 * wmaster device, so we don't bother checking whether
1046 	 * a mac80211 device supports the Wireless Extensions.
1047 	 */
1048 	ret = get_mac80211_phydev(handle, handle->opt.device, phydev_path,
1049 	    PATH_MAX);
1050 	if (ret < 0)
1051 		return ret;	/* error */
1052 	if (ret == 1)
1053 		return 1;	/* mac80211 device */
1054 #endif
1055 
1056 #ifdef IW_MODE_MONITOR
1057 	/*
1058 	 * Bleah.  There doesn't appear to be an ioctl to use to ask
1059 	 * whether a device supports monitor mode; we'll just do
1060 	 * SIOCGIWMODE and, if it succeeds, assume the device supports
1061 	 * monitor mode.
1062 	 *
1063 	 * Open a socket on which to attempt to get the mode.
1064 	 * (We assume that if we have Wireless Extensions support
1065 	 * we also have PF_PACKET support.)
1066 	 */
1067 	sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
1068 	if (sock_fd == -1) {
1069 		(void)pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1070 		    "socket: %s", pcap_strerror(errno));
1071 		return PCAP_ERROR;
1072 	}
1073 
1074 	if (is_bonding_device(sock_fd, handle->opt.device)) {
1075 		/* It's a bonding device, so don't even try. */
1076 		close(sock_fd);
1077 		return 0;
1078 	}
1079 
1080 	/*
1081 	 * Attempt to get the current mode.
1082 	 */
1083 	strlcpy(ireq.ifr_ifrn.ifrn_name, handle->opt.device,
1084 	    sizeof ireq.ifr_ifrn.ifrn_name);
1085 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
1086 		/*
1087 		 * Well, we got the mode; assume we can set it.
1088 		 */
1089 		close(sock_fd);
1090 		return 1;
1091 	}
1092 	if (errno == ENODEV) {
1093 		/* The device doesn't even exist. */
1094 		(void)pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1095 		    "SIOCGIWMODE failed: %s", pcap_strerror(errno));
1096 		close(sock_fd);
1097 		return PCAP_ERROR_NO_SUCH_DEVICE;
1098 	}
1099 	close(sock_fd);
1100 #endif
1101 	return 0;
1102 }
1103 
1104 /*
1105  * Grabs the number of dropped packets by the interface from /proc/net/dev.
1106  *
1107  * XXX - what about /sys/class/net/{interface name}/rx_*?  There are
1108  * individual devices giving, in ASCII, various rx_ and tx_ statistics.
1109  *
1110  * Or can we get them in binary form from netlink?
1111  */
1112 static long int
1113 linux_if_drops(const char * if_name)
1114 {
1115 	char buffer[512];
1116 	char * bufptr;
1117 	FILE * file;
1118 	int field_to_convert = 3, if_name_sz = strlen(if_name);
1119 	long int dropped_pkts = 0;
1120 
1121 	file = fopen("/proc/net/dev", "r");
1122 	if (!file)
1123 		return 0;
1124 
1125 	while (!dropped_pkts && fgets( buffer, sizeof(buffer), file ))
1126 	{
1127 		/* 	search for 'bytes' -- if its in there, then
1128 			that means we need to grab the fourth field. otherwise
1129 			grab the third field. */
1130 		if (field_to_convert != 4 && strstr(buffer, "bytes"))
1131 		{
1132 			field_to_convert = 4;
1133 			continue;
1134 		}
1135 
1136 		/* find iface and make sure it actually matches -- space before the name and : after it */
1137 		if ((bufptr = strstr(buffer, if_name)) &&
1138 			(bufptr == buffer || *(bufptr-1) == ' ') &&
1139 			*(bufptr + if_name_sz) == ':')
1140 		{
1141 			bufptr = bufptr + if_name_sz + 1;
1142 
1143 			/* grab the nth field from it */
1144 			while( --field_to_convert && *bufptr != '\0')
1145 			{
1146 				while (*bufptr != '\0' && *(bufptr++) == ' ');
1147 				while (*bufptr != '\0' && *(bufptr++) != ' ');
1148 			}
1149 
1150 			/* get rid of any final spaces */
1151 			while (*bufptr != '\0' && *bufptr == ' ') bufptr++;
1152 
1153 			if (*bufptr != '\0')
1154 				dropped_pkts = strtol(bufptr, NULL, 10);
1155 
1156 			break;
1157 		}
1158 	}
1159 
1160 	fclose(file);
1161 	return dropped_pkts;
1162 }
1163 
1164 
1165 /*
1166  * With older kernels promiscuous mode is kind of interesting because we
1167  * have to reset the interface before exiting. The problem can't really
1168  * be solved without some daemon taking care of managing usage counts.
1169  * If we put the interface into promiscuous mode, we set a flag indicating
1170  * that we must take it out of that mode when the interface is closed,
1171  * and, when closing the interface, if that flag is set we take it out
1172  * of promiscuous mode.
1173  *
1174  * Even with newer kernels, we have the same issue with rfmon mode.
1175  */
1176 
1177 static void	pcap_cleanup_linux( pcap_t *handle )
1178 {
1179 	struct pcap_linux *handlep = handle->priv;
1180 	struct ifreq	ifr;
1181 #ifdef HAVE_LIBNL
1182 	struct nl80211_state nlstate;
1183 	int ret;
1184 #endif /* HAVE_LIBNL */
1185 #ifdef IW_MODE_MONITOR
1186 	int oldflags;
1187 	struct iwreq ireq;
1188 #endif /* IW_MODE_MONITOR */
1189 
1190 	if (handlep->must_do_on_close != 0) {
1191 		/*
1192 		 * There's something we have to do when closing this
1193 		 * pcap_t.
1194 		 */
1195 		if (handlep->must_do_on_close & MUST_CLEAR_PROMISC) {
1196 			/*
1197 			 * We put the interface into promiscuous mode;
1198 			 * take it out of promiscuous mode.
1199 			 *
1200 			 * XXX - if somebody else wants it in promiscuous
1201 			 * mode, this code cannot know that, so it'll take
1202 			 * it out of promiscuous mode.  That's not fixable
1203 			 * in 2.0[.x] kernels.
1204 			 */
1205 			memset(&ifr, 0, sizeof(ifr));
1206 			strlcpy(ifr.ifr_name, handlep->device,
1207 			    sizeof(ifr.ifr_name));
1208 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
1209 				fprintf(stderr,
1210 				    "Can't restore interface %s flags (SIOCGIFFLAGS failed: %s).\n"
1211 				    "Please adjust manually.\n"
1212 				    "Hint: This can't happen with Linux >= 2.2.0.\n",
1213 				    handlep->device, strerror(errno));
1214 			} else {
1215 				if (ifr.ifr_flags & IFF_PROMISC) {
1216 					/*
1217 					 * Promiscuous mode is currently on;
1218 					 * turn it off.
1219 					 */
1220 					ifr.ifr_flags &= ~IFF_PROMISC;
1221 					if (ioctl(handle->fd, SIOCSIFFLAGS,
1222 					    &ifr) == -1) {
1223 						fprintf(stderr,
1224 						    "Can't restore interface %s flags (SIOCSIFFLAGS failed: %s).\n"
1225 						    "Please adjust manually.\n"
1226 						    "Hint: This can't happen with Linux >= 2.2.0.\n",
1227 						    handlep->device,
1228 						    strerror(errno));
1229 					}
1230 				}
1231 			}
1232 		}
1233 
1234 #ifdef HAVE_LIBNL
1235 		if (handlep->must_do_on_close & MUST_DELETE_MONIF) {
1236 			ret = nl80211_init(handle, &nlstate, handlep->device);
1237 			if (ret >= 0) {
1238 				ret = del_mon_if(handle, handle->fd, &nlstate,
1239 				    handlep->device, handlep->mondevice);
1240 				nl80211_cleanup(&nlstate);
1241 			}
1242 			if (ret < 0) {
1243 				fprintf(stderr,
1244 				    "Can't delete monitor interface %s (%s).\n"
1245 				    "Please delete manually.\n",
1246 				    handlep->mondevice, handle->errbuf);
1247 			}
1248 		}
1249 #endif /* HAVE_LIBNL */
1250 
1251 #ifdef IW_MODE_MONITOR
1252 		if (handlep->must_do_on_close & MUST_CLEAR_RFMON) {
1253 			/*
1254 			 * We put the interface into rfmon mode;
1255 			 * take it out of rfmon mode.
1256 			 *
1257 			 * XXX - if somebody else wants it in rfmon
1258 			 * mode, this code cannot know that, so it'll take
1259 			 * it out of rfmon mode.
1260 			 */
1261 
1262 			/*
1263 			 * First, take the interface down if it's up;
1264 			 * otherwise, we might get EBUSY.
1265 			 * If we get errors, just drive on and print
1266 			 * a warning if we can't restore the mode.
1267 			 */
1268 			oldflags = 0;
1269 			memset(&ifr, 0, sizeof(ifr));
1270 			strlcpy(ifr.ifr_name, handlep->device,
1271 			    sizeof(ifr.ifr_name));
1272 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) != -1) {
1273 				if (ifr.ifr_flags & IFF_UP) {
1274 					oldflags = ifr.ifr_flags;
1275 					ifr.ifr_flags &= ~IFF_UP;
1276 					if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1)
1277 						oldflags = 0;	/* didn't set, don't restore */
1278 				}
1279 			}
1280 
1281 			/*
1282 			 * Now restore the mode.
1283 			 */
1284 			strlcpy(ireq.ifr_ifrn.ifrn_name, handlep->device,
1285 			    sizeof ireq.ifr_ifrn.ifrn_name);
1286 			ireq.u.mode = handlep->oldmode;
1287 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
1288 				/*
1289 				 * Scientist, you've failed.
1290 				 */
1291 				fprintf(stderr,
1292 				    "Can't restore interface %s wireless mode (SIOCSIWMODE failed: %s).\n"
1293 				    "Please adjust manually.\n",
1294 				    handlep->device, strerror(errno));
1295 			}
1296 
1297 			/*
1298 			 * Now bring the interface back up if we brought
1299 			 * it down.
1300 			 */
1301 			if (oldflags != 0) {
1302 				ifr.ifr_flags = oldflags;
1303 				if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
1304 					fprintf(stderr,
1305 					    "Can't bring interface %s back up (SIOCSIFFLAGS failed: %s).\n"
1306 					    "Please adjust manually.\n",
1307 					    handlep->device, strerror(errno));
1308 				}
1309 			}
1310 		}
1311 #endif /* IW_MODE_MONITOR */
1312 
1313 		/*
1314 		 * Take this pcap out of the list of pcaps for which we
1315 		 * have to take the interface out of some mode.
1316 		 */
1317 		pcap_remove_from_pcaps_to_close(handle);
1318 	}
1319 
1320 	if (handlep->mondevice != NULL) {
1321 		free(handlep->mondevice);
1322 		handlep->mondevice = NULL;
1323 	}
1324 	if (handlep->device != NULL) {
1325 		free(handlep->device);
1326 		handlep->device = NULL;
1327 	}
1328 	pcap_cleanup_live_common(handle);
1329 }
1330 
1331 /*
1332  * Set the timeout to be used in poll() with memory-mapped packet capture.
1333  */
1334 static void
1335 set_poll_timeout(struct pcap_linux *handlep)
1336 {
1337 #ifdef HAVE_TPACKET3
1338 	struct utsname utsname;
1339 	char *version_component, *endp;
1340 	int major, minor;
1341 	int broken_tpacket_v3 = 1;
1342 
1343 	/*
1344 	 * Some versions of TPACKET_V3 have annoying bugs/misfeatures
1345 	 * around which we have to work.  Determine if we have those
1346 	 * problems or not.
1347 	 */
1348 	if (uname(&utsname) == 0) {
1349 		/*
1350 		 * 3.19 is the first release with a fixed version of
1351 		 * TPACKET_V3.  We treat anything before that as
1352 		 * not haveing a fixed version; that may really mean
1353 		 * it has *no* version.
1354 		 */
1355 		version_component = utsname.release;
1356 		major = strtol(version_component, &endp, 10);
1357 		if (endp != version_component && *endp == '.') {
1358 			/*
1359 			 * OK, that was a valid major version.
1360 			 * Get the minor version.
1361 			 */
1362 			version_component = endp + 1;
1363 			minor = strtol(version_component, &endp, 10);
1364 			if (endp != version_component &&
1365 			    (*endp == '.' || *endp == '\0')) {
1366 				/*
1367 				 * OK, that was a valid minor version.
1368 				 * Is this 3.19 or newer?
1369 				 */
1370 				if (major >= 4 || (major == 3 && minor >= 19)) {
1371 					/* Yes. TPACKET_V3 works correctly. */
1372 					broken_tpacket_v3 = 0;
1373 				}
1374 			}
1375 		}
1376 	}
1377 #endif
1378 	if (handlep->timeout == 0) {
1379 #ifdef HAVE_TPACKET3
1380 		/*
1381 		 * XXX - due to a set of (mis)features in the TPACKET_V3
1382 		 * kernel code prior to the 3.19 kernel, blocking forever
1383 		 * with a TPACKET_V3 socket can, if few packets are
1384 		 * arriving and passing the socket filter, cause most
1385 		 * packets to be dropped.  See libpcap issue #335 for the
1386 		 * full painful story.
1387 		 *
1388 		 * The workaround is to have poll() time out very quickly,
1389 		 * so we grab the frames handed to us, and return them to
1390 		 * the kernel, ASAP.
1391 		 */
1392 		if (handlep->tp_version == TPACKET_V3 && broken_tpacket_v3)
1393 			handlep->poll_timeout = 1;	/* don't block for very long */
1394 		else
1395 #endif
1396 			handlep->poll_timeout = -1;	/* block forever */
1397 	} else if (handlep->timeout > 0) {
1398 #ifdef HAVE_TPACKET3
1399 		/*
1400 		 * For TPACKET_V3, the timeout is handled by the kernel,
1401 		 * so block forever; that way, we don't get extra timeouts.
1402 		 * Don't do that if we have a broken TPACKET_V3, though.
1403 		 */
1404 		if (handlep->tp_version == TPACKET_V3 && !broken_tpacket_v3)
1405 			handlep->poll_timeout = -1;	/* block forever, let TPACKET_V3 wake us up */
1406 		else
1407 #endif
1408 			handlep->poll_timeout = handlep->timeout;	/* block for that amount of time */
1409 	} else {
1410 		/*
1411 		 * Non-blocking mode; we call poll() to pick up error
1412 		 * indications, but we don't want it to wait for
1413 		 * anything.
1414 		 */
1415 		handlep->poll_timeout = 0;
1416 	}
1417 }
1418 
1419 /*
1420  *  Get a handle for a live capture from the given device. You can
1421  *  pass NULL as device to get all packages (without link level
1422  *  information of course). If you pass 1 as promisc the interface
1423  *  will be set to promiscous mode (XXX: I think this usage should
1424  *  be deprecated and functions be added to select that later allow
1425  *  modification of that values -- Torsten).
1426  */
1427 static int
1428 pcap_activate_linux(pcap_t *handle)
1429 {
1430 	struct pcap_linux *handlep = handle->priv;
1431 	const char	*device;
1432 	struct ifreq	ifr;
1433 	int		status = 0;
1434 	int		ret;
1435 
1436 	device = handle->opt.device;
1437 
1438 	/*
1439 	 * Make sure the name we were handed will fit into the ioctls we
1440 	 * might perform on the device; if not, return a "No such device"
1441 	 * indication, as the Linux kernel shouldn't support creating
1442 	 * a device whose name won't fit into those ioctls.
1443 	 *
1444 	 * "Will fit" means "will fit, complete with a null terminator",
1445 	 * so if the length, which does *not* include the null terminator,
1446 	 * is greater than *or equal to* the size of the field into which
1447 	 * we'll be copying it, that won't fit.
1448 	 */
1449 	if (strlen(device) >= sizeof(ifr.ifr_name)) {
1450 		status = PCAP_ERROR_NO_SUCH_DEVICE;
1451 		goto fail;
1452 	}
1453 
1454 	handle->inject_op = pcap_inject_linux;
1455 	handle->setfilter_op = pcap_setfilter_linux;
1456 	handle->setdirection_op = pcap_setdirection_linux;
1457 	handle->set_datalink_op = pcap_set_datalink_linux;
1458 	handle->getnonblock_op = pcap_getnonblock_fd;
1459 	handle->setnonblock_op = pcap_setnonblock_fd;
1460 	handle->cleanup_op = pcap_cleanup_linux;
1461 	handle->read_op = pcap_read_linux;
1462 	handle->stats_op = pcap_stats_linux;
1463 
1464 	/*
1465 	 * The "any" device is a special device which causes us not
1466 	 * to bind to a particular device and thus to look at all
1467 	 * devices.
1468 	 */
1469 	if (strcmp(device, "any") == 0) {
1470 		if (handle->opt.promisc) {
1471 			handle->opt.promisc = 0;
1472 			/* Just a warning. */
1473 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1474 			    "Promiscuous mode not supported on the \"any\" device");
1475 			status = PCAP_WARNING_PROMISC_NOTSUP;
1476 		}
1477 	}
1478 
1479 	handlep->device	= strdup(device);
1480 	if (handlep->device == NULL) {
1481 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
1482 			 pcap_strerror(errno) );
1483 		return PCAP_ERROR;
1484 	}
1485 
1486 	/* copy timeout value */
1487 	handlep->timeout = handle->opt.timeout;
1488 
1489 	/*
1490 	 * If we're in promiscuous mode, then we probably want
1491 	 * to see when the interface drops packets too, so get an
1492 	 * initial count from /proc/net/dev
1493 	 */
1494 	if (handle->opt.promisc)
1495 		handlep->proc_dropped = linux_if_drops(handlep->device);
1496 
1497 	/*
1498 	 * Current Linux kernels use the protocol family PF_PACKET to
1499 	 * allow direct access to all packets on the network while
1500 	 * older kernels had a special socket type SOCK_PACKET to
1501 	 * implement this feature.
1502 	 * While this old implementation is kind of obsolete we need
1503 	 * to be compatible with older kernels for a while so we are
1504 	 * trying both methods with the newer method preferred.
1505 	 */
1506 	ret = activate_new(handle);
1507 	if (ret < 0) {
1508 		/*
1509 		 * Fatal error with the new way; just fail.
1510 		 * ret has the error return; if it's PCAP_ERROR,
1511 		 * handle->errbuf has been set appropriately.
1512 		 */
1513 		status = ret;
1514 		goto fail;
1515 	}
1516 	if (ret == 1) {
1517 		/*
1518 		 * Success.
1519 		 * Try to use memory-mapped access.
1520 		 */
1521 		switch (activate_mmap(handle, &status)) {
1522 
1523 		case 1:
1524 			/*
1525 			 * We succeeded.  status has been
1526 			 * set to the status to return,
1527 			 * which might be 0, or might be
1528 			 * a PCAP_WARNING_ value.
1529 			 *
1530 			 * Set the timeout to use in poll() before
1531 			 * returning.
1532 			 */
1533 			set_poll_timeout(handlep);
1534 			return status;
1535 
1536 		case 0:
1537 			/*
1538 			 * Kernel doesn't support it - just continue
1539 			 * with non-memory-mapped access.
1540 			 */
1541 			break;
1542 
1543 		case -1:
1544 			/*
1545 			 * We failed to set up to use it, or the kernel
1546 			 * supports it, but we failed to enable it.
1547 			 * ret has been set to the error status to
1548 			 * return and, if it's PCAP_ERROR, handle->errbuf
1549 			 * contains the error message.
1550 			 */
1551 			status = ret;
1552 			goto fail;
1553 		}
1554 	}
1555 	else if (ret == 0) {
1556 		/* Non-fatal error; try old way */
1557 		if ((ret = activate_old(handle)) != 1) {
1558 			/*
1559 			 * Both methods to open the packet socket failed.
1560 			 * Tidy up and report our failure (handle->errbuf
1561 			 * is expected to be set by the functions above).
1562 			 */
1563 			status = ret;
1564 			goto fail;
1565 		}
1566 	}
1567 
1568 	/*
1569 	 * We set up the socket, but not with memory-mapped access.
1570 	 */
1571 	if (handle->opt.buffer_size != 0) {
1572 		/*
1573 		 * Set the socket buffer size to the specified value.
1574 		 */
1575 		if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF,
1576 		    &handle->opt.buffer_size,
1577 		    sizeof(handle->opt.buffer_size)) == -1) {
1578 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1579 				 "SO_RCVBUF: %s", pcap_strerror(errno));
1580 			status = PCAP_ERROR;
1581 			goto fail;
1582 		}
1583 	}
1584 
1585 	/* Allocate the buffer */
1586 
1587 	handle->buffer	 = malloc(handle->bufsize + handle->offset);
1588 	if (!handle->buffer) {
1589 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1590 			 "malloc: %s", pcap_strerror(errno));
1591 		status = PCAP_ERROR;
1592 		goto fail;
1593 	}
1594 
1595 	/*
1596 	 * "handle->fd" is a socket, so "select()" and "poll()"
1597 	 * should work on it.
1598 	 */
1599 	handle->selectable_fd = handle->fd;
1600 
1601 	return status;
1602 
1603 fail:
1604 	pcap_cleanup_linux(handle);
1605 	return status;
1606 }
1607 
1608 /*
1609  *  Read at most max_packets from the capture stream and call the callback
1610  *  for each of them. Returns the number of packets handled or -1 if an
1611  *  error occured.
1612  */
1613 static int
1614 pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
1615 {
1616 	/*
1617 	 * Currently, on Linux only one packet is delivered per read,
1618 	 * so we don't loop.
1619 	 */
1620 	return pcap_read_packet(handle, callback, user);
1621 }
1622 
1623 static int
1624 pcap_set_datalink_linux(pcap_t *handle, int dlt)
1625 {
1626 	handle->linktype = dlt;
1627 	return 0;
1628 }
1629 
1630 /*
1631  * linux_check_direction()
1632  *
1633  * Do checks based on packet direction.
1634  */
1635 static inline int
1636 linux_check_direction(const pcap_t *handle, const struct sockaddr_ll *sll)
1637 {
1638 	struct pcap_linux	*handlep = handle->priv;
1639 
1640 	if (sll->sll_pkttype == PACKET_OUTGOING) {
1641 		/*
1642 		 * Outgoing packet.
1643 		 * If this is from the loopback device, reject it;
1644 		 * we'll see the packet as an incoming packet as well,
1645 		 * and we don't want to see it twice.
1646 		 */
1647 		if (sll->sll_ifindex == handlep->lo_ifindex)
1648 			return 0;
1649 
1650 		/*
1651 		 * If this is an outgoing CAN or CAN FD frame, and
1652 		 * the user doesn't only want outgoing packets,
1653 		 * reject it; CAN devices and drivers, and the CAN
1654 		 * stack, always arrange to loop back transmitted
1655 		 * packets, so they also appear as incoming packets.
1656 		 * We don't want duplicate packets, and we can't
1657 		 * easily distinguish packets looped back by the CAN
1658 		 * layer than those received by the CAN layer, so we
1659 		 * eliminate this packet instead.
1660 		 */
1661 		if ((sll->sll_protocol == LINUX_SLL_P_CAN ||
1662 		     sll->sll_protocol == LINUX_SLL_P_CANFD) &&
1663 		     handle->direction != PCAP_D_OUT)
1664 			return 0;
1665 
1666 		/*
1667 		 * If the user only wants incoming packets, reject it.
1668 		 */
1669 		if (handle->direction == PCAP_D_IN)
1670 			return 0;
1671 	} else {
1672 		/*
1673 		 * Incoming packet.
1674 		 * If the user only wants outgoing packets, reject it.
1675 		 */
1676 		if (handle->direction == PCAP_D_OUT)
1677 			return 0;
1678 	}
1679 	return 1;
1680 }
1681 
1682 /*
1683  *  Read a packet from the socket calling the handler provided by
1684  *  the user. Returns the number of packets received or -1 if an
1685  *  error occured.
1686  */
1687 static int
1688 pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
1689 {
1690 	struct pcap_linux	*handlep = handle->priv;
1691 	u_char			*bp;
1692 	int			offset;
1693 #ifdef HAVE_PF_PACKET_SOCKETS
1694 	struct sockaddr_ll	from;
1695 	struct sll_header	*hdrp;
1696 #else
1697 	struct sockaddr		from;
1698 #endif
1699 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1700 	struct iovec		iov;
1701 	struct msghdr		msg;
1702 	struct cmsghdr		*cmsg;
1703 	union {
1704 		struct cmsghdr	cmsg;
1705 		char		buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
1706 	} cmsg_buf;
1707 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1708 	socklen_t		fromlen;
1709 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1710 	int			packet_len, caplen;
1711 	struct pcap_pkthdr	pcap_header;
1712 
1713         struct bpf_aux_data     aux_data;
1714 #ifdef HAVE_PF_PACKET_SOCKETS
1715 	/*
1716 	 * If this is a cooked device, leave extra room for a
1717 	 * fake packet header.
1718 	 */
1719 	if (handlep->cooked)
1720 		offset = SLL_HDR_LEN;
1721 	else
1722 		offset = 0;
1723 #else
1724 	/*
1725 	 * This system doesn't have PF_PACKET sockets, so it doesn't
1726 	 * support cooked devices.
1727 	 */
1728 	offset = 0;
1729 #endif
1730 
1731 	/*
1732 	 * Receive a single packet from the kernel.
1733 	 * We ignore EINTR, as that might just be due to a signal
1734 	 * being delivered - if the signal should interrupt the
1735 	 * loop, the signal handler should call pcap_breakloop()
1736 	 * to set handle->break_loop (we ignore it on other
1737 	 * platforms as well).
1738 	 * We also ignore ENETDOWN, so that we can continue to
1739 	 * capture traffic if the interface goes down and comes
1740 	 * back up again; comments in the kernel indicate that
1741 	 * we'll just block waiting for packets if we try to
1742 	 * receive from a socket that delivered ENETDOWN, and,
1743 	 * if we're using a memory-mapped buffer, we won't even
1744 	 * get notified of "network down" events.
1745 	 */
1746 	bp = (u_char *)handle->buffer + handle->offset;
1747 
1748 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1749 	msg.msg_name		= &from;
1750 	msg.msg_namelen		= sizeof(from);
1751 	msg.msg_iov		= &iov;
1752 	msg.msg_iovlen		= 1;
1753 	msg.msg_control		= &cmsg_buf;
1754 	msg.msg_controllen	= sizeof(cmsg_buf);
1755 	msg.msg_flags		= 0;
1756 
1757 	iov.iov_len		= handle->bufsize - offset;
1758 	iov.iov_base		= bp + offset;
1759 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1760 
1761 	do {
1762 		/*
1763 		 * Has "pcap_breakloop()" been called?
1764 		 */
1765 		if (handle->break_loop) {
1766 			/*
1767 			 * Yes - clear the flag that indicates that it has,
1768 			 * and return PCAP_ERROR_BREAK as an indication that
1769 			 * we were told to break out of the loop.
1770 			 */
1771 			handle->break_loop = 0;
1772 			return PCAP_ERROR_BREAK;
1773 		}
1774 
1775 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1776 		packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC);
1777 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1778 		fromlen = sizeof(from);
1779 		packet_len = recvfrom(
1780 			handle->fd, bp + offset,
1781 			handle->bufsize - offset, MSG_TRUNC,
1782 			(struct sockaddr *) &from, &fromlen);
1783 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1784 	} while (packet_len == -1 && errno == EINTR);
1785 
1786 	/* Check if an error occured */
1787 
1788 	if (packet_len == -1) {
1789 		switch (errno) {
1790 
1791 		case EAGAIN:
1792 			return 0;	/* no packet there */
1793 
1794 		case ENETDOWN:
1795 			/*
1796 			 * The device on which we're capturing went away.
1797 			 *
1798 			 * XXX - we should really return
1799 			 * PCAP_ERROR_IFACE_NOT_UP, but pcap_dispatch()
1800 			 * etc. aren't defined to return that.
1801 			 */
1802 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1803 				"The interface went down");
1804 			return PCAP_ERROR;
1805 
1806 		default:
1807 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1808 				 "recvfrom: %s", pcap_strerror(errno));
1809 			return PCAP_ERROR;
1810 		}
1811 	}
1812 
1813 #ifdef HAVE_PF_PACKET_SOCKETS
1814 	if (!handlep->sock_packet) {
1815 		/*
1816 		 * Unfortunately, there is a window between socket() and
1817 		 * bind() where the kernel may queue packets from any
1818 		 * interface.  If we're bound to a particular interface,
1819 		 * discard packets not from that interface.
1820 		 *
1821 		 * (If socket filters are supported, we could do the
1822 		 * same thing we do when changing the filter; however,
1823 		 * that won't handle packet sockets without socket
1824 		 * filter support, and it's a bit more complicated.
1825 		 * It would save some instructions per packet, however.)
1826 		 */
1827 		if (handlep->ifindex != -1 &&
1828 		    from.sll_ifindex != handlep->ifindex)
1829 			return 0;
1830 
1831 		/*
1832 		 * Do checks based on packet direction.
1833 		 * We can only do this if we're using PF_PACKET; the
1834 		 * address returned for SOCK_PACKET is a "sockaddr_pkt"
1835 		 * which lacks the relevant packet type information.
1836 		 */
1837 		if (!linux_check_direction(handle, &from))
1838 			return 0;
1839 	}
1840 #endif
1841 
1842 #ifdef HAVE_PF_PACKET_SOCKETS
1843 	/*
1844 	 * If this is a cooked device, fill in the fake packet header.
1845 	 */
1846 	if (handlep->cooked) {
1847 		/*
1848 		 * Add the length of the fake header to the length
1849 		 * of packet data we read.
1850 		 */
1851 		packet_len += SLL_HDR_LEN;
1852 
1853 		hdrp = (struct sll_header *)bp;
1854 		hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype);
1855 		hdrp->sll_hatype = htons(from.sll_hatype);
1856 		hdrp->sll_halen = htons(from.sll_halen);
1857 		memcpy(hdrp->sll_addr, from.sll_addr,
1858 		    (from.sll_halen > SLL_ADDRLEN) ?
1859 		      SLL_ADDRLEN :
1860 		      from.sll_halen);
1861 		hdrp->sll_protocol = from.sll_protocol;
1862 	}
1863 
1864 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1865 	if (handlep->vlan_offset != -1) {
1866 		for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
1867 			struct tpacket_auxdata *aux;
1868 			unsigned int len;
1869 			struct vlan_tag *tag;
1870 
1871 			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
1872 			    cmsg->cmsg_level != SOL_PACKET ||
1873 			    cmsg->cmsg_type != PACKET_AUXDATA)
1874 				continue;
1875 
1876 			aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
1877 #if defined(TP_STATUS_VLAN_VALID)
1878 			if ((aux->tp_vlan_tci == 0) && !(aux->tp_status & TP_STATUS_VLAN_VALID))
1879 #else
1880 			if (aux->tp_vlan_tci == 0) /* this is ambigious but without the
1881 						TP_STATUS_VLAN_VALID flag, there is
1882 						nothing that we can do */
1883 #endif
1884 				continue;
1885 
1886 			len = (u_int)packet_len > iov.iov_len ? iov.iov_len : (u_int)packet_len;
1887 			if (len < (u_int)handlep->vlan_offset)
1888 				break;
1889 
1890 			/*
1891 			 * Move everything in the header, except the
1892 			 * type field, down VLAN_TAG_LEN bytes, to
1893 			 * allow us to insert the VLAN tag between
1894 			 * that stuff and the type field.
1895 			 */
1896 			bp -= VLAN_TAG_LEN;
1897 			memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
1898 
1899 			/*
1900 			 * Now insert the tag.
1901 			 */
1902 			tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
1903 			tag->vlan_tpid = htons(VLAN_TPID(aux, aux));
1904 			tag->vlan_tci = htons(aux->tp_vlan_tci);
1905 
1906                         /* store vlan tci to bpf_aux_data struct for userland bpf filter */
1907 #if defined(TP_STATUS_VLAN_VALID)
1908                         aux_data.vlan_tag = htons(aux->tp_vlan_tci) & 0x0fff;
1909                         aux_data.vlan_tag_present = (aux->tp_status & TP_STATUS_VLAN_VALID);
1910 #endif
1911 
1912 			/*
1913 			 * Add the tag to the packet lengths.
1914 			 */
1915 			packet_len += VLAN_TAG_LEN;
1916 		}
1917 	}
1918 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1919 #endif /* HAVE_PF_PACKET_SOCKETS */
1920 
1921 	/*
1922 	 * XXX: According to the kernel source we should get the real
1923 	 * packet len if calling recvfrom with MSG_TRUNC set. It does
1924 	 * not seem to work here :(, but it is supported by this code
1925 	 * anyway.
1926 	 * To be honest the code RELIES on that feature so this is really
1927 	 * broken with 2.2.x kernels.
1928 	 * I spend a day to figure out what's going on and I found out
1929 	 * that the following is happening:
1930 	 *
1931 	 * The packet comes from a random interface and the packet_rcv
1932 	 * hook is called with a clone of the packet. That code inserts
1933 	 * the packet into the receive queue of the packet socket.
1934 	 * If a filter is attached to that socket that filter is run
1935 	 * first - and there lies the problem. The default filter always
1936 	 * cuts the packet at the snaplen:
1937 	 *
1938 	 * # tcpdump -d
1939 	 * (000) ret      #68
1940 	 *
1941 	 * So the packet filter cuts down the packet. The recvfrom call
1942 	 * says "hey, it's only 68 bytes, it fits into the buffer" with
1943 	 * the result that we don't get the real packet length. This
1944 	 * is valid at least until kernel 2.2.17pre6.
1945 	 *
1946 	 * We currently handle this by making a copy of the filter
1947 	 * program, fixing all "ret" instructions with non-zero
1948 	 * operands to have an operand of MAXIMUM_SNAPLEN so that the
1949 	 * filter doesn't truncate the packet, and supplying that modified
1950 	 * filter to the kernel.
1951 	 */
1952 
1953 	caplen = packet_len;
1954 	if (caplen > handle->snapshot)
1955 		caplen = handle->snapshot;
1956 
1957 	/* Run the packet filter if not using kernel filter */
1958 	if (handlep->filter_in_userland && handle->fcode.bf_insns) {
1959 		if (bpf_filter_with_aux_data(handle->fcode.bf_insns, bp,
1960 		    packet_len, caplen, &aux_data) == 0) {
1961 			/* rejected by filter */
1962 			return 0;
1963 		}
1964 	}
1965 
1966 	/* Fill in our own header data */
1967 
1968 	/* get timestamp for this packet */
1969 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
1970 	if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
1971 		if (ioctl(handle->fd, SIOCGSTAMPNS, &pcap_header.ts) == -1) {
1972 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1973 					"SIOCGSTAMPNS: %s", pcap_strerror(errno));
1974 			return PCAP_ERROR;
1975 		}
1976         } else
1977 #endif
1978 	{
1979 		if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
1980 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1981 					"SIOCGSTAMP: %s", pcap_strerror(errno));
1982 			return PCAP_ERROR;
1983 		}
1984         }
1985 
1986 	pcap_header.caplen	= caplen;
1987 	pcap_header.len		= packet_len;
1988 
1989 	/*
1990 	 * Count the packet.
1991 	 *
1992 	 * Arguably, we should count them before we check the filter,
1993 	 * as on many other platforms "ps_recv" counts packets
1994 	 * handed to the filter rather than packets that passed
1995 	 * the filter, but if filtering is done in the kernel, we
1996 	 * can't get a count of packets that passed the filter,
1997 	 * and that would mean the meaning of "ps_recv" wouldn't
1998 	 * be the same on all Linux systems.
1999 	 *
2000 	 * XXX - it's not the same on all systems in any case;
2001 	 * ideally, we should have a "get the statistics" call
2002 	 * that supplies more counts and indicates which of them
2003 	 * it supplies, so that we supply a count of packets
2004 	 * handed to the filter only on platforms where that
2005 	 * information is available.
2006 	 *
2007 	 * We count them here even if we can get the packet count
2008 	 * from the kernel, as we can only determine at run time
2009 	 * whether we'll be able to get it from the kernel (if
2010 	 * HAVE_TPACKET_STATS isn't defined, we can't get it from
2011 	 * the kernel, but if it is defined, the library might
2012 	 * have been built with a 2.4 or later kernel, but we
2013 	 * might be running on a 2.2[.x] kernel without Alexey
2014 	 * Kuznetzov's turbopacket patches, and thus the kernel
2015 	 * might not be able to supply those statistics).  We
2016 	 * could, I guess, try, when opening the socket, to get
2017 	 * the statistics, and if we can not increment the count
2018 	 * here, but it's not clear that always incrementing
2019 	 * the count is more expensive than always testing a flag
2020 	 * in memory.
2021 	 *
2022 	 * We keep the count in "handlep->packets_read", and use that
2023 	 * for "ps_recv" if we can't get the statistics from the kernel.
2024 	 * We do that because, if we *can* get the statistics from
2025 	 * the kernel, we use "handlep->stat.ps_recv" and
2026 	 * "handlep->stat.ps_drop" as running counts, as reading the
2027 	 * statistics from the kernel resets the kernel statistics,
2028 	 * and if we directly increment "handlep->stat.ps_recv" here,
2029 	 * that means it will count packets *twice* on systems where
2030 	 * we can get kernel statistics - once here, and once in
2031 	 * pcap_stats_linux().
2032 	 */
2033 	handlep->packets_read++;
2034 
2035 	/* Call the user supplied callback function */
2036 	callback(userdata, &pcap_header, bp);
2037 
2038 	return 1;
2039 }
2040 
2041 static int
2042 pcap_inject_linux(pcap_t *handle, const void *buf, size_t size)
2043 {
2044 	struct pcap_linux *handlep = handle->priv;
2045 	int ret;
2046 
2047 #ifdef HAVE_PF_PACKET_SOCKETS
2048 	if (!handlep->sock_packet) {
2049 		/* PF_PACKET socket */
2050 		if (handlep->ifindex == -1) {
2051 			/*
2052 			 * We don't support sending on the "any" device.
2053 			 */
2054 			strlcpy(handle->errbuf,
2055 			    "Sending packets isn't supported on the \"any\" device",
2056 			    PCAP_ERRBUF_SIZE);
2057 			return (-1);
2058 		}
2059 
2060 		if (handlep->cooked) {
2061 			/*
2062 			 * We don't support sending on the "any" device.
2063 			 *
2064 			 * XXX - how do you send on a bound cooked-mode
2065 			 * socket?
2066 			 * Is a "sendto()" required there?
2067 			 */
2068 			strlcpy(handle->errbuf,
2069 			    "Sending packets isn't supported in cooked mode",
2070 			    PCAP_ERRBUF_SIZE);
2071 			return (-1);
2072 		}
2073 	}
2074 #endif
2075 
2076 	ret = send(handle->fd, buf, size, 0);
2077 	if (ret == -1) {
2078 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
2079 		    pcap_strerror(errno));
2080 		return (-1);
2081 	}
2082 	return (ret);
2083 }
2084 
2085 /*
2086  *  Get the statistics for the given packet capture handle.
2087  *  Reports the number of dropped packets iff the kernel supports
2088  *  the PACKET_STATISTICS "getsockopt()" argument (2.4 and later
2089  *  kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket
2090  *  patches); otherwise, that information isn't available, and we lie
2091  *  and report 0 as the count of dropped packets.
2092  */
2093 static int
2094 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
2095 {
2096 	struct pcap_linux *handlep = handle->priv;
2097 #ifdef HAVE_TPACKET_STATS
2098 #ifdef HAVE_TPACKET3
2099 	/*
2100 	 * For sockets using TPACKET_V1 or TPACKET_V2, the extra
2101 	 * stuff at the end of a struct tpacket_stats_v3 will not
2102 	 * be filled in, and we don't look at it so this is OK even
2103 	 * for those sockets.  In addition, the PF_PACKET socket
2104 	 * code in the kernel only uses the length parameter to
2105 	 * compute how much data to copy out and to indicate how
2106 	 * much data was copied out, so it's OK to base it on the
2107 	 * size of a struct tpacket_stats.
2108 	 *
2109 	 * XXX - it's probably OK, in fact, to just use a
2110 	 * struct tpacket_stats for V3 sockets, as we don't
2111 	 * care about the tp_freeze_q_cnt stat.
2112 	 */
2113 	struct tpacket_stats_v3 kstats;
2114 #else /* HAVE_TPACKET3 */
2115 	struct tpacket_stats kstats;
2116 #endif /* HAVE_TPACKET3 */
2117 	socklen_t len = sizeof (struct tpacket_stats);
2118 #endif /* HAVE_TPACKET_STATS */
2119 
2120 	long if_dropped = 0;
2121 
2122 	/*
2123 	 *	To fill in ps_ifdrop, we parse /proc/net/dev for the number
2124 	 */
2125 	if (handle->opt.promisc)
2126 	{
2127 		if_dropped = handlep->proc_dropped;
2128 		handlep->proc_dropped = linux_if_drops(handlep->device);
2129 		handlep->stat.ps_ifdrop += (handlep->proc_dropped - if_dropped);
2130 	}
2131 
2132 #ifdef HAVE_TPACKET_STATS
2133 	/*
2134 	 * Try to get the packet counts from the kernel.
2135 	 */
2136 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
2137 			&kstats, &len) > -1) {
2138 		/*
2139 		 * On systems where the PACKET_STATISTICS "getsockopt()"
2140 		 * argument is supported on PF_PACKET sockets:
2141 		 *
2142 		 *	"ps_recv" counts only packets that *passed* the
2143 		 *	filter, not packets that didn't pass the filter.
2144 		 *	This includes packets later dropped because we
2145 		 *	ran out of buffer space.
2146 		 *
2147 		 *	"ps_drop" counts packets dropped because we ran
2148 		 *	out of buffer space.  It doesn't count packets
2149 		 *	dropped by the interface driver.  It counts only
2150 		 *	packets that passed the filter.
2151 		 *
2152 		 *	See above for ps_ifdrop.
2153 		 *
2154 		 *	Both statistics include packets not yet read from
2155 		 *	the kernel by libpcap, and thus not yet seen by
2156 		 *	the application.
2157 		 *
2158 		 * In "linux/net/packet/af_packet.c", at least in the
2159 		 * 2.4.9 kernel, "tp_packets" is incremented for every
2160 		 * packet that passes the packet filter *and* is
2161 		 * successfully queued on the socket; "tp_drops" is
2162 		 * incremented for every packet dropped because there's
2163 		 * not enough free space in the socket buffer.
2164 		 *
2165 		 * When the statistics are returned for a PACKET_STATISTICS
2166 		 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
2167 		 * so that "tp_packets" counts all packets handed to
2168 		 * the PF_PACKET socket, including packets dropped because
2169 		 * there wasn't room on the socket buffer - but not
2170 		 * including packets that didn't pass the filter.
2171 		 *
2172 		 * In the BSD BPF, the count of received packets is
2173 		 * incremented for every packet handed to BPF, regardless
2174 		 * of whether it passed the filter.
2175 		 *
2176 		 * We can't make "pcap_stats()" work the same on both
2177 		 * platforms, but the best approximation is to return
2178 		 * "tp_packets" as the count of packets and "tp_drops"
2179 		 * as the count of drops.
2180 		 *
2181 		 * Keep a running total because each call to
2182 		 *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
2183 		 * resets the counters to zero.
2184 		 */
2185 		handlep->stat.ps_recv += kstats.tp_packets;
2186 		handlep->stat.ps_drop += kstats.tp_drops;
2187 		*stats = handlep->stat;
2188 		return 0;
2189 	}
2190 	else
2191 	{
2192 		/*
2193 		 * If the error was EOPNOTSUPP, fall through, so that
2194 		 * if you build the library on a system with
2195 		 * "struct tpacket_stats" and run it on a system
2196 		 * that doesn't, it works as it does if the library
2197 		 * is built on a system without "struct tpacket_stats".
2198 		 */
2199 		if (errno != EOPNOTSUPP) {
2200 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2201 			    "pcap_stats: %s", pcap_strerror(errno));
2202 			return -1;
2203 		}
2204 	}
2205 #endif
2206 	/*
2207 	 * On systems where the PACKET_STATISTICS "getsockopt()" argument
2208 	 * is not supported on PF_PACKET sockets:
2209 	 *
2210 	 *	"ps_recv" counts only packets that *passed* the filter,
2211 	 *	not packets that didn't pass the filter.  It does not
2212 	 *	count packets dropped because we ran out of buffer
2213 	 *	space.
2214 	 *
2215 	 *	"ps_drop" is not supported.
2216 	 *
2217 	 *	"ps_ifdrop" is supported. It will return the number
2218 	 *	of drops the interface reports in /proc/net/dev,
2219 	 *	if that is available.
2220 	 *
2221 	 *	"ps_recv" doesn't include packets not yet read from
2222 	 *	the kernel by libpcap.
2223 	 *
2224 	 * We maintain the count of packets processed by libpcap in
2225 	 * "handlep->packets_read", for reasons described in the comment
2226 	 * at the end of pcap_read_packet().  We have no idea how many
2227 	 * packets were dropped by the kernel buffers -- but we know
2228 	 * how many the interface dropped, so we can return that.
2229 	 */
2230 
2231 	stats->ps_recv = handlep->packets_read;
2232 	stats->ps_drop = 0;
2233 	stats->ps_ifdrop = handlep->stat.ps_ifdrop;
2234 	return 0;
2235 }
2236 
2237 static int
2238 add_linux_if(pcap_if_t **devlistp, const char *ifname, int fd, char *errbuf)
2239 {
2240 	const char *p;
2241 	char name[512];	/* XXX - pick a size */
2242 	char *q, *saveq;
2243 	struct ifreq ifrflags;
2244 
2245 	/*
2246 	 * Get the interface name.
2247 	 */
2248 	p = ifname;
2249 	q = &name[0];
2250 	while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2251 		if (*p == ':') {
2252 			/*
2253 			 * This could be the separator between a
2254 			 * name and an alias number, or it could be
2255 			 * the separator between a name with no
2256 			 * alias number and the next field.
2257 			 *
2258 			 * If there's a colon after digits, it
2259 			 * separates the name and the alias number,
2260 			 * otherwise it separates the name and the
2261 			 * next field.
2262 			 */
2263 			saveq = q;
2264 			while (isascii(*p) && isdigit(*p))
2265 				*q++ = *p++;
2266 			if (*p != ':') {
2267 				/*
2268 				 * That was the next field,
2269 				 * not the alias number.
2270 				 */
2271 				q = saveq;
2272 			}
2273 			break;
2274 		} else
2275 			*q++ = *p++;
2276 	}
2277 	*q = '\0';
2278 
2279 	/*
2280 	 * Get the flags for this interface.
2281 	 */
2282 	strlcpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2283 	if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2284 		if (errno == ENXIO || errno == ENODEV)
2285 			return (0);	/* device doesn't actually exist - ignore it */
2286 		(void)pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
2287 		    "SIOCGIFFLAGS: %.*s: %s",
2288 		    (int)sizeof(ifrflags.ifr_name),
2289 		    ifrflags.ifr_name,
2290 		    pcap_strerror(errno));
2291 		return (-1);
2292 	}
2293 
2294 	/*
2295 	 * Add an entry for this interface, with no addresses.
2296 	 */
2297 	if (pcap_add_if(devlistp, name,
2298 	    if_flags_to_pcap_flags(name, ifrflags.ifr_flags), NULL,
2299 	    errbuf) == -1) {
2300 		/*
2301 		 * Failure.
2302 		 */
2303 		return (-1);
2304 	}
2305 
2306 	return (0);
2307 }
2308 
2309 /*
2310  * Get from "/sys/class/net" all interfaces listed there; if they're
2311  * already in the list of interfaces we have, that won't add another
2312  * instance, but if they're not, that'll add them.
2313  *
2314  * We don't bother getting any addresses for them; it appears you can't
2315  * use SIOCGIFADDR on Linux to get IPv6 addresses for interfaces, and,
2316  * although some other types of addresses can be fetched with SIOCGIFADDR,
2317  * we don't bother with them for now.
2318  *
2319  * We also don't fail if we couldn't open "/sys/class/net"; we just leave
2320  * the list of interfaces as is, and return 0, so that we can try
2321  * scanning /proc/net/dev.
2322  *
2323  * Otherwise, we return 1 if we don't get an error and -1 if we do.
2324  */
2325 static int
2326 scan_sys_class_net(pcap_if_t **devlistp, char *errbuf)
2327 {
2328 	DIR *sys_class_net_d;
2329 	int fd;
2330 	struct dirent *ent;
2331 	char subsystem_path[PATH_MAX+1];
2332 	struct stat statb;
2333 	int ret = 1;
2334 
2335 	sys_class_net_d = opendir("/sys/class/net");
2336 	if (sys_class_net_d == NULL) {
2337 		/*
2338 		 * Don't fail if it doesn't exist at all.
2339 		 */
2340 		if (errno == ENOENT)
2341 			return (0);
2342 
2343 		/*
2344 		 * Fail if we got some other error.
2345 		 */
2346 		(void)pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
2347 		    "Can't open /sys/class/net: %s", pcap_strerror(errno));
2348 		return (-1);
2349 	}
2350 
2351 	/*
2352 	 * Create a socket from which to fetch interface information.
2353 	 */
2354 	fd = socket(PF_UNIX, SOCK_RAW, 0);
2355 	if (fd < 0) {
2356 		(void)pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
2357 		    "socket: %s", pcap_strerror(errno));
2358 		(void)closedir(sys_class_net_d);
2359 		return (-1);
2360 	}
2361 
2362 	for (;;) {
2363 		errno = 0;
2364 		ent = readdir(sys_class_net_d);
2365 		if (ent == NULL) {
2366 			/*
2367 			 * Error or EOF; if errno != 0, it's an error.
2368 			 */
2369 			break;
2370 		}
2371 
2372 		/*
2373 		 * Ignore "." and "..".
2374 		 */
2375 		if (strcmp(ent->d_name, ".") == 0 ||
2376 		    strcmp(ent->d_name, "..") == 0)
2377 			continue;
2378 
2379 		/*
2380 		 * Ignore plain files; they do not have subdirectories
2381 		 * and thus have no attributes.
2382 		 */
2383 		if (ent->d_type == DT_REG)
2384 			continue;
2385 
2386 		/*
2387 		 * Is there an "ifindex" file under that name?
2388 		 * (We don't care whether it's a directory or
2389 		 * a symlink; older kernels have directories
2390 		 * for devices, newer kernels have symlinks to
2391 		 * directories.)
2392 		 */
2393 		pcap_snprintf(subsystem_path, sizeof subsystem_path,
2394 		    "/sys/class/net/%s/ifindex", ent->d_name);
2395 		if (lstat(subsystem_path, &statb) != 0) {
2396 			/*
2397 			 * Stat failed.  Either there was an error
2398 			 * other than ENOENT, and we don't know if
2399 			 * this is an interface, or it's ENOENT,
2400 			 * and either some part of "/sys/class/net/{if}"
2401 			 * disappeared, in which case it probably means
2402 			 * the interface disappeared, or there's no
2403 			 * "ifindex" file, which means it's not a
2404 			 * network interface.
2405 			 */
2406 			continue;
2407 		}
2408 
2409 		/*
2410 		 * Attempt to add the interface.
2411 		 */
2412 		if (add_linux_if(devlistp, &ent->d_name[0], fd, errbuf) == -1) {
2413 			/* Fail. */
2414 			ret = -1;
2415 			break;
2416 		}
2417 	}
2418 	if (ret != -1) {
2419 		/*
2420 		 * Well, we didn't fail for any other reason; did we
2421 		 * fail due to an error reading the directory?
2422 		 */
2423 		if (errno != 0) {
2424 			(void)pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
2425 			    "Error reading /sys/class/net: %s",
2426 			    pcap_strerror(errno));
2427 			ret = -1;
2428 		}
2429 	}
2430 
2431 	(void)close(fd);
2432 	(void)closedir(sys_class_net_d);
2433 	return (ret);
2434 }
2435 
2436 /*
2437  * Get from "/proc/net/dev" all interfaces listed there; if they're
2438  * already in the list of interfaces we have, that won't add another
2439  * instance, but if they're not, that'll add them.
2440  *
2441  * See comments from scan_sys_class_net().
2442  */
2443 static int
2444 scan_proc_net_dev(pcap_if_t **devlistp, char *errbuf)
2445 {
2446 	FILE *proc_net_f;
2447 	int fd;
2448 	char linebuf[512];
2449 	int linenum;
2450 	char *p;
2451 	int ret = 0;
2452 
2453 	proc_net_f = fopen("/proc/net/dev", "r");
2454 	if (proc_net_f == NULL) {
2455 		/*
2456 		 * Don't fail if it doesn't exist at all.
2457 		 */
2458 		if (errno == ENOENT)
2459 			return (0);
2460 
2461 		/*
2462 		 * Fail if we got some other error.
2463 		 */
2464 		(void)pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
2465 		    "Can't open /proc/net/dev: %s", pcap_strerror(errno));
2466 		return (-1);
2467 	}
2468 
2469 	/*
2470 	 * Create a socket from which to fetch interface information.
2471 	 */
2472 	fd = socket(PF_UNIX, SOCK_RAW, 0);
2473 	if (fd < 0) {
2474 		(void)pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
2475 		    "socket: %s", pcap_strerror(errno));
2476 		(void)fclose(proc_net_f);
2477 		return (-1);
2478 	}
2479 
2480 	for (linenum = 1;
2481 	    fgets(linebuf, sizeof linebuf, proc_net_f) != NULL; linenum++) {
2482 		/*
2483 		 * Skip the first two lines - they're headers.
2484 		 */
2485 		if (linenum <= 2)
2486 			continue;
2487 
2488 		p = &linebuf[0];
2489 
2490 		/*
2491 		 * Skip leading white space.
2492 		 */
2493 		while (*p != '\0' && isascii(*p) && isspace(*p))
2494 			p++;
2495 		if (*p == '\0' || *p == '\n')
2496 			continue;	/* blank line */
2497 
2498 		/*
2499 		 * Attempt to add the interface.
2500 		 */
2501 		if (add_linux_if(devlistp, p, fd, errbuf) == -1) {
2502 			/* Fail. */
2503 			ret = -1;
2504 			break;
2505 		}
2506 	}
2507 	if (ret != -1) {
2508 		/*
2509 		 * Well, we didn't fail for any other reason; did we
2510 		 * fail due to an error reading the file?
2511 		 */
2512 		if (ferror(proc_net_f)) {
2513 			(void)pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
2514 			    "Error reading /proc/net/dev: %s",
2515 			    pcap_strerror(errno));
2516 			ret = -1;
2517 		}
2518 	}
2519 
2520 	(void)close(fd);
2521 	(void)fclose(proc_net_f);
2522 	return (ret);
2523 }
2524 
2525 /*
2526  * Description string for the "any" device.
2527  */
2528 static const char any_descr[] = "Pseudo-device that captures on all interfaces";
2529 
2530 /*
2531  * A SOCK_PACKET or PF_PACKET socket can be bound to any network interface.
2532  */
2533 static int
2534 can_be_bound(const char *name _U_)
2535 {
2536 	return (1);
2537 }
2538 
2539 int
2540 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
2541 {
2542 	int ret;
2543 
2544 	/*
2545 	 * Get the list of regular interfaces first.
2546 	 */
2547 	if (pcap_findalldevs_interfaces(alldevsp, errbuf, can_be_bound) == -1)
2548 		return (-1);	/* failure */
2549 
2550 	/*
2551 	 * Read "/sys/class/net", and add to the list of interfaces all
2552 	 * interfaces listed there that we don't already have, because,
2553 	 * on Linux, SIOCGIFCONF reports only interfaces with IPv4 addresses,
2554 	 * and even getifaddrs() won't return information about
2555 	 * interfaces with no addresses, so you need to read "/sys/class/net"
2556 	 * to get the names of the rest of the interfaces.
2557 	 */
2558 	ret = scan_sys_class_net(alldevsp, errbuf);
2559 	if (ret == -1)
2560 		return (-1);	/* failed */
2561 	if (ret == 0) {
2562 		/*
2563 		 * No /sys/class/net; try reading /proc/net/dev instead.
2564 		 */
2565 		if (scan_proc_net_dev(alldevsp, errbuf) == -1)
2566 			return (-1);
2567 	}
2568 
2569 	/*
2570 	 * Add the "any" device.
2571 	 */
2572 	if (pcap_add_if(alldevsp, "any", PCAP_IF_UP|PCAP_IF_RUNNING,
2573 	    any_descr, errbuf) < 0)
2574 		return (-1);
2575 
2576 	return (0);
2577 }
2578 
2579 /*
2580  *  Attach the given BPF code to the packet capture device.
2581  */
2582 static int
2583 pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
2584     int is_mmapped)
2585 {
2586 	struct pcap_linux *handlep;
2587 #ifdef SO_ATTACH_FILTER
2588 	struct sock_fprog	fcode;
2589 	int			can_filter_in_kernel;
2590 	int			err = 0;
2591 #endif
2592 
2593 	if (!handle)
2594 		return -1;
2595 	if (!filter) {
2596 	        strlcpy(handle->errbuf, "setfilter: No filter specified",
2597 			PCAP_ERRBUF_SIZE);
2598 		return -1;
2599 	}
2600 
2601 	handlep = handle->priv;
2602 
2603 	/* Make our private copy of the filter */
2604 
2605 	if (install_bpf_program(handle, filter) < 0)
2606 		/* install_bpf_program() filled in errbuf */
2607 		return -1;
2608 
2609 	/*
2610 	 * Run user level packet filter by default. Will be overriden if
2611 	 * installing a kernel filter succeeds.
2612 	 */
2613 	handlep->filter_in_userland = 1;
2614 
2615 	/* Install kernel level filter if possible */
2616 
2617 #ifdef SO_ATTACH_FILTER
2618 #ifdef USHRT_MAX
2619 	if (handle->fcode.bf_len > USHRT_MAX) {
2620 		/*
2621 		 * fcode.len is an unsigned short for current kernel.
2622 		 * I have yet to see BPF-Code with that much
2623 		 * instructions but still it is possible. So for the
2624 		 * sake of correctness I added this check.
2625 		 */
2626 		fprintf(stderr, "Warning: Filter too complex for kernel\n");
2627 		fcode.len = 0;
2628 		fcode.filter = NULL;
2629 		can_filter_in_kernel = 0;
2630 	} else
2631 #endif /* USHRT_MAX */
2632 	{
2633 		/*
2634 		 * Oh joy, the Linux kernel uses struct sock_fprog instead
2635 		 * of struct bpf_program and of course the length field is
2636 		 * of different size. Pointed out by Sebastian
2637 		 *
2638 		 * Oh, and we also need to fix it up so that all "ret"
2639 		 * instructions with non-zero operands have MAXIMUM_SNAPLEN
2640 		 * as the operand if we're not capturing in memory-mapped
2641 		 * mode, and so that, if we're in cooked mode, all memory-
2642 		 * reference instructions use special magic offsets in
2643 		 * references to the link-layer header and assume that the
2644 		 * link-layer payload begins at 0; "fix_program()" will do
2645 		 * that.
2646 		 */
2647 		switch (fix_program(handle, &fcode, is_mmapped)) {
2648 
2649 		case -1:
2650 		default:
2651 			/*
2652 			 * Fatal error; just quit.
2653 			 * (The "default" case shouldn't happen; we
2654 			 * return -1 for that reason.)
2655 			 */
2656 			return -1;
2657 
2658 		case 0:
2659 			/*
2660 			 * The program performed checks that we can't make
2661 			 * work in the kernel.
2662 			 */
2663 			can_filter_in_kernel = 0;
2664 			break;
2665 
2666 		case 1:
2667 			/*
2668 			 * We have a filter that'll work in the kernel.
2669 			 */
2670 			can_filter_in_kernel = 1;
2671 			break;
2672 		}
2673 	}
2674 
2675 	/*
2676 	 * NOTE: at this point, we've set both the "len" and "filter"
2677 	 * fields of "fcode".  As of the 2.6.32.4 kernel, at least,
2678 	 * those are the only members of the "sock_fprog" structure,
2679 	 * so we initialize every member of that structure.
2680 	 *
2681 	 * If there is anything in "fcode" that is not initialized,
2682 	 * it is either a field added in a later kernel, or it's
2683 	 * padding.
2684 	 *
2685 	 * If a new field is added, this code needs to be updated
2686 	 * to set it correctly.
2687 	 *
2688 	 * If there are no other fields, then:
2689 	 *
2690 	 *	if the Linux kernel looks at the padding, it's
2691 	 *	buggy;
2692 	 *
2693 	 *	if the Linux kernel doesn't look at the padding,
2694 	 *	then if some tool complains that we're passing
2695 	 *	uninitialized data to the kernel, then the tool
2696 	 *	is buggy and needs to understand that it's just
2697 	 *	padding.
2698 	 */
2699 	if (can_filter_in_kernel) {
2700 		if ((err = set_kernel_filter(handle, &fcode)) == 0)
2701 		{
2702 			/*
2703 			 * Installation succeded - using kernel filter,
2704 			 * so userland filtering not needed.
2705 			 */
2706 			handlep->filter_in_userland = 0;
2707 		}
2708 		else if (err == -1)	/* Non-fatal error */
2709 		{
2710 			/*
2711 			 * Print a warning if we weren't able to install
2712 			 * the filter for a reason other than "this kernel
2713 			 * isn't configured to support socket filters.
2714 			 */
2715 			if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
2716 				fprintf(stderr,
2717 				    "Warning: Kernel filter failed: %s\n",
2718 					pcap_strerror(errno));
2719 			}
2720 		}
2721 	}
2722 
2723 	/*
2724 	 * If we're not using the kernel filter, get rid of any kernel
2725 	 * filter that might've been there before, e.g. because the
2726 	 * previous filter could work in the kernel, or because some other
2727 	 * code attached a filter to the socket by some means other than
2728 	 * calling "pcap_setfilter()".  Otherwise, the kernel filter may
2729 	 * filter out packets that would pass the new userland filter.
2730 	 */
2731 	if (handlep->filter_in_userland) {
2732 		if (reset_kernel_filter(handle) == -1) {
2733 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2734 			    "can't remove kernel filter: %s",
2735 			    pcap_strerror(errno));
2736 			err = -2;	/* fatal error */
2737 		}
2738 	}
2739 
2740 	/*
2741 	 * Free up the copy of the filter that was made by "fix_program()".
2742 	 */
2743 	if (fcode.filter != NULL)
2744 		free(fcode.filter);
2745 
2746 	if (err == -2)
2747 		/* Fatal error */
2748 		return -1;
2749 #endif /* SO_ATTACH_FILTER */
2750 
2751 	return 0;
2752 }
2753 
2754 static int
2755 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
2756 {
2757 	return pcap_setfilter_linux_common(handle, filter, 0);
2758 }
2759 
2760 
2761 /*
2762  * Set direction flag: Which packets do we accept on a forwarding
2763  * single device? IN, OUT or both?
2764  */
2765 static int
2766 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
2767 {
2768 #ifdef HAVE_PF_PACKET_SOCKETS
2769 	struct pcap_linux *handlep = handle->priv;
2770 
2771 	if (!handlep->sock_packet) {
2772 		handle->direction = d;
2773 		return 0;
2774 	}
2775 #endif
2776 	/*
2777 	 * We're not using PF_PACKET sockets, so we can't determine
2778 	 * the direction of the packet.
2779 	 */
2780 	pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2781 	    "Setting direction is not supported on SOCK_PACKET sockets");
2782 	return -1;
2783 }
2784 
2785 #ifdef HAVE_PF_PACKET_SOCKETS
2786 /*
2787  * Map the PACKET_ value to a LINUX_SLL_ value; we
2788  * want the same numerical value to be used in
2789  * the link-layer header even if the numerical values
2790  * for the PACKET_ #defines change, so that programs
2791  * that look at the packet type field will always be
2792  * able to handle DLT_LINUX_SLL captures.
2793  */
2794 static short int
2795 map_packet_type_to_sll_type(short int sll_pkttype)
2796 {
2797 	switch (sll_pkttype) {
2798 
2799 	case PACKET_HOST:
2800 		return htons(LINUX_SLL_HOST);
2801 
2802 	case PACKET_BROADCAST:
2803 		return htons(LINUX_SLL_BROADCAST);
2804 
2805 	case PACKET_MULTICAST:
2806 		return  htons(LINUX_SLL_MULTICAST);
2807 
2808 	case PACKET_OTHERHOST:
2809 		return htons(LINUX_SLL_OTHERHOST);
2810 
2811 	case PACKET_OUTGOING:
2812 		return htons(LINUX_SLL_OUTGOING);
2813 
2814 	default:
2815 		return -1;
2816 	}
2817 }
2818 #endif
2819 
2820 static int
2821 is_wifi(int sock_fd
2822 #ifndef IW_MODE_MONITOR
2823 _U_
2824 #endif
2825 , const char *device)
2826 {
2827 	char *pathstr;
2828 	struct stat statb;
2829 #ifdef IW_MODE_MONITOR
2830 	char errbuf[PCAP_ERRBUF_SIZE];
2831 #endif
2832 
2833 	/*
2834 	 * See if there's a sysfs wireless directory for it.
2835 	 * If so, it's a wireless interface.
2836 	 */
2837 	if (asprintf(&pathstr, "/sys/class/net/%s/wireless", device) == -1) {
2838 		/*
2839 		 * Just give up here.
2840 		 */
2841 		return 0;
2842 	}
2843 	if (stat(pathstr, &statb) == 0) {
2844 		free(pathstr);
2845 		return 1;
2846 	}
2847 	free(pathstr);
2848 
2849 #ifdef IW_MODE_MONITOR
2850 	/*
2851 	 * OK, maybe it's not wireless, or maybe this kernel doesn't
2852 	 * support sysfs.  Try the wireless extensions.
2853 	 */
2854 	if (has_wext(sock_fd, device, errbuf) == 1) {
2855 		/*
2856 		 * It supports the wireless extensions, so it's a Wi-Fi
2857 		 * device.
2858 		 */
2859 		return 1;
2860 	}
2861 #endif
2862 	return 0;
2863 }
2864 
2865 /*
2866  *  Linux uses the ARP hardware type to identify the type of an
2867  *  interface. pcap uses the DLT_xxx constants for this. This
2868  *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
2869  *  constant, as arguments, and sets "handle->linktype" to the
2870  *  appropriate DLT_XXX constant and sets "handle->offset" to
2871  *  the appropriate value (to make "handle->offset" plus link-layer
2872  *  header length be a multiple of 4, so that the link-layer payload
2873  *  will be aligned on a 4-byte boundary when capturing packets).
2874  *  (If the offset isn't set here, it'll be 0; add code as appropriate
2875  *  for cases where it shouldn't be 0.)
2876  *
2877  *  If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
2878  *  in cooked mode; otherwise, we can't use cooked mode, so we have
2879  *  to pick some type that works in raw mode, or fail.
2880  *
2881  *  Sets the link type to -1 if unable to map the type.
2882  */
2883 static void map_arphrd_to_dlt(pcap_t *handle, int sock_fd, int arptype,
2884 			      const char *device, int cooked_ok)
2885 {
2886 	static const char cdma_rmnet[] = "cdma_rmnet";
2887 
2888 	switch (arptype) {
2889 
2890 	case ARPHRD_ETHER:
2891 		/*
2892 		 * For various annoying reasons having to do with DHCP
2893 		 * software, some versions of Android give the mobile-
2894 		 * phone-network interface an ARPHRD_ value of
2895 		 * ARPHRD_ETHER, even though the packets supplied by
2896 		 * that interface have no link-layer header, and begin
2897 		 * with an IP header, so that the ARPHRD_ value should
2898 		 * be ARPHRD_NONE.
2899 		 *
2900 		 * Detect those devices by checking the device name, and
2901 		 * use DLT_RAW for them.
2902 		 */
2903 		if (strncmp(device, cdma_rmnet, sizeof cdma_rmnet - 1) == 0) {
2904 			handle->linktype = DLT_RAW;
2905 			return;
2906 		}
2907 
2908 		/*
2909 		 * Is this a real Ethernet device?  If so, give it a
2910 		 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
2911 		 * that an application can let you choose it, in case you're
2912 		 * capturing DOCSIS traffic that a Cisco Cable Modem
2913 		 * Termination System is putting out onto an Ethernet (it
2914 		 * doesn't put an Ethernet header onto the wire, it puts raw
2915 		 * DOCSIS frames out on the wire inside the low-level
2916 		 * Ethernet framing).
2917 		 *
2918 		 * XXX - are there any other sorts of "fake Ethernet" that
2919 		 * have ARPHRD_ETHER but that shouldn't offer DLT_DOCSIS as
2920 		 * a Cisco CMTS won't put traffic onto it or get traffic
2921 		 * bridged onto it?  ISDN is handled in "activate_new()",
2922 		 * as we fall back on cooked mode there, and we use
2923 		 * is_wifi() to check for 802.11 devices; are there any
2924 		 * others?
2925 		 */
2926 		if (!is_wifi(sock_fd, device)) {
2927 			/*
2928 			 * It's not a Wi-Fi device; offer DOCSIS.
2929 			 */
2930 			handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2931 			/*
2932 			 * If that fails, just leave the list empty.
2933 			 */
2934 			if (handle->dlt_list != NULL) {
2935 				handle->dlt_list[0] = DLT_EN10MB;
2936 				handle->dlt_list[1] = DLT_DOCSIS;
2937 				handle->dlt_count = 2;
2938 			}
2939 		}
2940 		/* FALLTHROUGH */
2941 
2942 	case ARPHRD_METRICOM:
2943 	case ARPHRD_LOOPBACK:
2944 		handle->linktype = DLT_EN10MB;
2945 		handle->offset = 2;
2946 		break;
2947 
2948 	case ARPHRD_EETHER:
2949 		handle->linktype = DLT_EN3MB;
2950 		break;
2951 
2952 	case ARPHRD_AX25:
2953 		handle->linktype = DLT_AX25_KISS;
2954 		break;
2955 
2956 	case ARPHRD_PRONET:
2957 		handle->linktype = DLT_PRONET;
2958 		break;
2959 
2960 	case ARPHRD_CHAOS:
2961 		handle->linktype = DLT_CHAOS;
2962 		break;
2963 #ifndef ARPHRD_CAN
2964 #define ARPHRD_CAN 280
2965 #endif
2966 	case ARPHRD_CAN:
2967 		/*
2968 		 * Map this to DLT_LINUX_SLL; that way, CAN frames will
2969 		 * have ETH_P_CAN/LINUX_SLL_P_CAN as the protocol and
2970 		 * CAN FD frames will have ETH_P_CANFD/LINUX_SLL_P_CANFD
2971 		 * as the protocol, so they can be distinguished by the
2972 		 * protocol in the SLL header.
2973 		 */
2974 		handle->linktype = DLT_LINUX_SLL;
2975 		break;
2976 
2977 #ifndef ARPHRD_IEEE802_TR
2978 #define ARPHRD_IEEE802_TR 800	/* From Linux 2.4 */
2979 #endif
2980 	case ARPHRD_IEEE802_TR:
2981 	case ARPHRD_IEEE802:
2982 		handle->linktype = DLT_IEEE802;
2983 		handle->offset = 2;
2984 		break;
2985 
2986 	case ARPHRD_ARCNET:
2987 		handle->linktype = DLT_ARCNET_LINUX;
2988 		break;
2989 
2990 #ifndef ARPHRD_FDDI	/* From Linux 2.2.13 */
2991 #define ARPHRD_FDDI	774
2992 #endif
2993 	case ARPHRD_FDDI:
2994 		handle->linktype = DLT_FDDI;
2995 		handle->offset = 3;
2996 		break;
2997 
2998 #ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
2999 #define ARPHRD_ATM 19
3000 #endif
3001 	case ARPHRD_ATM:
3002 		/*
3003 		 * The Classical IP implementation in ATM for Linux
3004 		 * supports both what RFC 1483 calls "LLC Encapsulation",
3005 		 * in which each packet has an LLC header, possibly
3006 		 * with a SNAP header as well, prepended to it, and
3007 		 * what RFC 1483 calls "VC Based Multiplexing", in which
3008 		 * different virtual circuits carry different network
3009 		 * layer protocols, and no header is prepended to packets.
3010 		 *
3011 		 * They both have an ARPHRD_ type of ARPHRD_ATM, so
3012 		 * you can't use the ARPHRD_ type to find out whether
3013 		 * captured packets will have an LLC header, and,
3014 		 * while there's a socket ioctl to *set* the encapsulation
3015 		 * type, there's no ioctl to *get* the encapsulation type.
3016 		 *
3017 		 * This means that
3018 		 *
3019 		 *	programs that dissect Linux Classical IP frames
3020 		 *	would have to check for an LLC header and,
3021 		 *	depending on whether they see one or not, dissect
3022 		 *	the frame as LLC-encapsulated or as raw IP (I
3023 		 *	don't know whether there's any traffic other than
3024 		 *	IP that would show up on the socket, or whether
3025 		 *	there's any support for IPv6 in the Linux
3026 		 *	Classical IP code);
3027 		 *
3028 		 *	filter expressions would have to compile into
3029 		 *	code that checks for an LLC header and does
3030 		 *	the right thing.
3031 		 *
3032 		 * Both of those are a nuisance - and, at least on systems
3033 		 * that support PF_PACKET sockets, we don't have to put
3034 		 * up with those nuisances; instead, we can just capture
3035 		 * in cooked mode.  That's what we'll do, if we can.
3036 		 * Otherwise, we'll just fail.
3037 		 */
3038 		if (cooked_ok)
3039 			handle->linktype = DLT_LINUX_SLL;
3040 		else
3041 			handle->linktype = -1;
3042 		break;
3043 
3044 #ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
3045 #define ARPHRD_IEEE80211 801
3046 #endif
3047 	case ARPHRD_IEEE80211:
3048 		handle->linktype = DLT_IEEE802_11;
3049 		break;
3050 
3051 #ifndef ARPHRD_IEEE80211_PRISM  /* From Linux 2.4.18 */
3052 #define ARPHRD_IEEE80211_PRISM 802
3053 #endif
3054 	case ARPHRD_IEEE80211_PRISM:
3055 		handle->linktype = DLT_PRISM_HEADER;
3056 		break;
3057 
3058 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
3059 #define ARPHRD_IEEE80211_RADIOTAP 803
3060 #endif
3061 	case ARPHRD_IEEE80211_RADIOTAP:
3062 		handle->linktype = DLT_IEEE802_11_RADIO;
3063 		break;
3064 
3065 	case ARPHRD_PPP:
3066 		/*
3067 		 * Some PPP code in the kernel supplies no link-layer
3068 		 * header whatsoever to PF_PACKET sockets; other PPP
3069 		 * code supplies PPP link-layer headers ("syncppp.c");
3070 		 * some PPP code might supply random link-layer
3071 		 * headers (PPP over ISDN - there's code in Ethereal,
3072 		 * for example, to cope with PPP-over-ISDN captures
3073 		 * with which the Ethereal developers have had to cope,
3074 		 * heuristically trying to determine which of the
3075 		 * oddball link-layer headers particular packets have).
3076 		 *
3077 		 * As such, we just punt, and run all PPP interfaces
3078 		 * in cooked mode, if we can; otherwise, we just treat
3079 		 * it as DLT_RAW, for now - if somebody needs to capture,
3080 		 * on a 2.0[.x] kernel, on PPP devices that supply a
3081 		 * link-layer header, they'll have to add code here to
3082 		 * map to the appropriate DLT_ type (possibly adding a
3083 		 * new DLT_ type, if necessary).
3084 		 */
3085 		if (cooked_ok)
3086 			handle->linktype = DLT_LINUX_SLL;
3087 		else {
3088 			/*
3089 			 * XXX - handle ISDN types here?  We can't fall
3090 			 * back on cooked sockets, so we'd have to
3091 			 * figure out from the device name what type of
3092 			 * link-layer encapsulation it's using, and map
3093 			 * that to an appropriate DLT_ value, meaning
3094 			 * we'd map "isdnN" devices to DLT_RAW (they
3095 			 * supply raw IP packets with no link-layer
3096 			 * header) and "isdY" devices to a new DLT_I4L_IP
3097 			 * type that has only an Ethernet packet type as
3098 			 * a link-layer header.
3099 			 *
3100 			 * But sometimes we seem to get random crap
3101 			 * in the link-layer header when capturing on
3102 			 * ISDN devices....
3103 			 */
3104 			handle->linktype = DLT_RAW;
3105 		}
3106 		break;
3107 
3108 #ifndef ARPHRD_CISCO
3109 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
3110 #endif
3111 	case ARPHRD_CISCO:
3112 		handle->linktype = DLT_C_HDLC;
3113 		break;
3114 
3115 	/* Not sure if this is correct for all tunnels, but it
3116 	 * works for CIPE */
3117 	case ARPHRD_TUNNEL:
3118 #ifndef ARPHRD_SIT
3119 #define ARPHRD_SIT 776	/* From Linux 2.2.13 */
3120 #endif
3121 	case ARPHRD_SIT:
3122 	case ARPHRD_CSLIP:
3123 	case ARPHRD_SLIP6:
3124 	case ARPHRD_CSLIP6:
3125 	case ARPHRD_ADAPT:
3126 	case ARPHRD_SLIP:
3127 #ifndef ARPHRD_RAWHDLC
3128 #define ARPHRD_RAWHDLC 518
3129 #endif
3130 	case ARPHRD_RAWHDLC:
3131 #ifndef ARPHRD_DLCI
3132 #define ARPHRD_DLCI 15
3133 #endif
3134 	case ARPHRD_DLCI:
3135 		/*
3136 		 * XXX - should some of those be mapped to DLT_LINUX_SLL
3137 		 * instead?  Should we just map all of them to DLT_LINUX_SLL?
3138 		 */
3139 		handle->linktype = DLT_RAW;
3140 		break;
3141 
3142 #ifndef ARPHRD_FRAD
3143 #define ARPHRD_FRAD 770
3144 #endif
3145 	case ARPHRD_FRAD:
3146 		handle->linktype = DLT_FRELAY;
3147 		break;
3148 
3149 	case ARPHRD_LOCALTLK:
3150 		handle->linktype = DLT_LTALK;
3151 		break;
3152 
3153 	case 18:
3154 		/*
3155 		 * RFC 4338 defines an encapsulation for IP and ARP
3156 		 * packets that's compatible with the RFC 2625
3157 		 * encapsulation, but that uses a different ARP
3158 		 * hardware type and hardware addresses.  That
3159 		 * ARP hardware type is 18; Linux doesn't define
3160 		 * any ARPHRD_ value as 18, but if it ever officially
3161 		 * supports RFC 4338-style IP-over-FC, it should define
3162 		 * one.
3163 		 *
3164 		 * For now, we map it to DLT_IP_OVER_FC, in the hopes
3165 		 * that this will encourage its use in the future,
3166 		 * should Linux ever officially support RFC 4338-style
3167 		 * IP-over-FC.
3168 		 */
3169 		handle->linktype = DLT_IP_OVER_FC;
3170 		break;
3171 
3172 #ifndef ARPHRD_FCPP
3173 #define ARPHRD_FCPP	784
3174 #endif
3175 	case ARPHRD_FCPP:
3176 #ifndef ARPHRD_FCAL
3177 #define ARPHRD_FCAL	785
3178 #endif
3179 	case ARPHRD_FCAL:
3180 #ifndef ARPHRD_FCPL
3181 #define ARPHRD_FCPL	786
3182 #endif
3183 	case ARPHRD_FCPL:
3184 #ifndef ARPHRD_FCFABRIC
3185 #define ARPHRD_FCFABRIC	787
3186 #endif
3187 	case ARPHRD_FCFABRIC:
3188 		/*
3189 		 * Back in 2002, Donald Lee at Cray wanted a DLT_ for
3190 		 * IP-over-FC:
3191 		 *
3192 		 *	http://www.mail-archive.com/tcpdump-workers@sandelman.ottawa.on.ca/msg01043.html
3193 		 *
3194 		 * and one was assigned.
3195 		 *
3196 		 * In a later private discussion (spun off from a message
3197 		 * on the ethereal-users list) on how to get that DLT_
3198 		 * value in libpcap on Linux, I ended up deciding that
3199 		 * the best thing to do would be to have him tweak the
3200 		 * driver to set the ARPHRD_ value to some ARPHRD_FCxx
3201 		 * type, and map all those types to DLT_IP_OVER_FC:
3202 		 *
3203 		 *	I've checked into the libpcap and tcpdump CVS tree
3204 		 *	support for DLT_IP_OVER_FC.  In order to use that,
3205 		 *	you'd have to modify your modified driver to return
3206 		 *	one of the ARPHRD_FCxxx types, in "fcLINUXfcp.c" -
3207 		 *	change it to set "dev->type" to ARPHRD_FCFABRIC, for
3208 		 *	example (the exact value doesn't matter, it can be
3209 		 *	any of ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, or
3210 		 *	ARPHRD_FCFABRIC).
3211 		 *
3212 		 * 11 years later, Christian Svensson wanted to map
3213 		 * various ARPHRD_ values to DLT_FC_2 and
3214 		 * DLT_FC_2_WITH_FRAME_DELIMS for raw Fibre Channel
3215 		 * frames:
3216 		 *
3217 		 *	https://github.com/mcr/libpcap/pull/29
3218 		 *
3219 		 * There doesn't seem to be any network drivers that uses
3220 		 * any of the ARPHRD_FC* values for IP-over-FC, and
3221 		 * it's not exactly clear what the "Dummy types for non
3222 		 * ARP hardware" are supposed to mean (link-layer
3223 		 * header type?  Physical network type?), so it's
3224 		 * not exactly clear why the ARPHRD_FC* types exist
3225 		 * in the first place.
3226 		 *
3227 		 * For now, we map them to DLT_FC_2, and provide an
3228 		 * option of DLT_FC_2_WITH_FRAME_DELIMS, as well as
3229 		 * DLT_IP_OVER_FC just in case there's some old
3230 		 * driver out there that uses one of those types for
3231 		 * IP-over-FC on which somebody wants to capture
3232 		 * packets.
3233 		 */
3234 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
3235 		/*
3236 		 * If that fails, just leave the list empty.
3237 		 */
3238 		if (handle->dlt_list != NULL) {
3239 			handle->dlt_list[0] = DLT_FC_2;
3240 			handle->dlt_list[1] = DLT_FC_2_WITH_FRAME_DELIMS;
3241 			handle->dlt_list[2] = DLT_IP_OVER_FC;
3242 			handle->dlt_count = 3;
3243 		}
3244 		handle->linktype = DLT_FC_2;
3245 		break;
3246 
3247 #ifndef ARPHRD_IRDA
3248 #define ARPHRD_IRDA	783
3249 #endif
3250 	case ARPHRD_IRDA:
3251 		/* Don't expect IP packet out of this interfaces... */
3252 		handle->linktype = DLT_LINUX_IRDA;
3253 		/* We need to save packet direction for IrDA decoding,
3254 		 * so let's use "Linux-cooked" mode. Jean II
3255 		 *
3256 		 * XXX - this is handled in activate_new(). */
3257 		/* handlep->cooked = 1; */
3258 		break;
3259 
3260 	/* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
3261 	 * is needed, please report it to <daniele@orlandi.com> */
3262 #ifndef ARPHRD_LAPD
3263 #define ARPHRD_LAPD	8445
3264 #endif
3265 	case ARPHRD_LAPD:
3266 		/* Don't expect IP packet out of this interfaces... */
3267 		handle->linktype = DLT_LINUX_LAPD;
3268 		break;
3269 
3270 #ifndef ARPHRD_NONE
3271 #define ARPHRD_NONE	0xFFFE
3272 #endif
3273 	case ARPHRD_NONE:
3274 		/*
3275 		 * No link-layer header; packets are just IP
3276 		 * packets, so use DLT_RAW.
3277 		 */
3278 		handle->linktype = DLT_RAW;
3279 		break;
3280 
3281 #ifndef ARPHRD_IEEE802154
3282 #define ARPHRD_IEEE802154      804
3283 #endif
3284        case ARPHRD_IEEE802154:
3285                handle->linktype =  DLT_IEEE802_15_4_NOFCS;
3286                break;
3287 
3288 #ifndef ARPHRD_NETLINK
3289 #define ARPHRD_NETLINK	824
3290 #endif
3291 	case ARPHRD_NETLINK:
3292 		handle->linktype = DLT_NETLINK;
3293 		/*
3294 		 * We need to use cooked mode, so that in sll_protocol we
3295 		 * pick up the netlink protocol type such as NETLINK_ROUTE,
3296 		 * NETLINK_GENERIC, NETLINK_FIB_LOOKUP, etc.
3297 		 *
3298 		 * XXX - this is handled in activate_new().
3299 		 */
3300 		/* handlep->cooked = 1; */
3301 		break;
3302 
3303 	default:
3304 		handle->linktype = -1;
3305 		break;
3306 	}
3307 }
3308 
3309 /* ===== Functions to interface to the newer kernels ================== */
3310 
3311 /*
3312  * Try to open a packet socket using the new kernel PF_PACKET interface.
3313  * Returns 1 on success, 0 on an error that means the new interface isn't
3314  * present (so the old SOCK_PACKET interface should be tried), and a
3315  * PCAP_ERROR_ value on an error that means that the old mechanism won't
3316  * work either (so it shouldn't be tried).
3317  */
3318 static int
3319 activate_new(pcap_t *handle)
3320 {
3321 #ifdef HAVE_PF_PACKET_SOCKETS
3322 	struct pcap_linux *handlep = handle->priv;
3323 	const char		*device = handle->opt.device;
3324 	int			is_any_device = (strcmp(device, "any") == 0);
3325 	int			sock_fd = -1, arptype;
3326 #ifdef HAVE_PACKET_AUXDATA
3327 	int			val;
3328 #endif
3329 	int			err = 0;
3330 	struct packet_mreq	mr;
3331 #if defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT)
3332 	int			bpf_extensions;
3333 	socklen_t		len = sizeof(bpf_extensions);
3334 #endif
3335 
3336 	/*
3337 	 * Open a socket with protocol family packet. If the
3338 	 * "any" device was specified, we open a SOCK_DGRAM
3339 	 * socket for the cooked interface, otherwise we first
3340 	 * try a SOCK_RAW socket for the raw interface.
3341 	 */
3342 	sock_fd = is_any_device ?
3343 		socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) :
3344 		socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
3345 
3346 	if (sock_fd == -1) {
3347 		if (errno == EINVAL || errno == EAFNOSUPPORT) {
3348 			/*
3349 			 * We don't support PF_PACKET/SOCK_whatever
3350 			 * sockets; try the old mechanism.
3351 			 */
3352 			return 0;
3353 		}
3354 
3355 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
3356 			 pcap_strerror(errno) );
3357 		if (errno == EPERM || errno == EACCES) {
3358 			/*
3359 			 * You don't have permission to open the
3360 			 * socket.
3361 			 */
3362 			return PCAP_ERROR_PERM_DENIED;
3363 		} else {
3364 			/*
3365 			 * Other error.
3366 			 */
3367 			return PCAP_ERROR;
3368 		}
3369 	}
3370 
3371 	/* It seems the kernel supports the new interface. */
3372 	handlep->sock_packet = 0;
3373 
3374 	/*
3375 	 * Get the interface index of the loopback device.
3376 	 * If the attempt fails, don't fail, just set the
3377 	 * "handlep->lo_ifindex" to -1.
3378 	 *
3379 	 * XXX - can there be more than one device that loops
3380 	 * packets back, i.e. devices other than "lo"?  If so,
3381 	 * we'd need to find them all, and have an array of
3382 	 * indices for them, and check all of them in
3383 	 * "pcap_read_packet()".
3384 	 */
3385 	handlep->lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
3386 
3387 	/*
3388 	 * Default value for offset to align link-layer payload
3389 	 * on a 4-byte boundary.
3390 	 */
3391 	handle->offset	 = 0;
3392 
3393 	/*
3394 	 * What kind of frames do we have to deal with? Fall back
3395 	 * to cooked mode if we have an unknown interface type
3396 	 * or a type we know doesn't work well in raw mode.
3397 	 */
3398 	if (!is_any_device) {
3399 		/* Assume for now we don't need cooked mode. */
3400 		handlep->cooked = 0;
3401 
3402 		if (handle->opt.rfmon) {
3403 			/*
3404 			 * We were asked to turn on monitor mode.
3405 			 * Do so before we get the link-layer type,
3406 			 * because entering monitor mode could change
3407 			 * the link-layer type.
3408 			 */
3409 			err = enter_rfmon_mode(handle, sock_fd, device);
3410 			if (err < 0) {
3411 				/* Hard failure */
3412 				close(sock_fd);
3413 				return err;
3414 			}
3415 			if (err == 0) {
3416 				/*
3417 				 * Nothing worked for turning monitor mode
3418 				 * on.
3419 				 */
3420 				close(sock_fd);
3421 				return PCAP_ERROR_RFMON_NOTSUP;
3422 			}
3423 
3424 			/*
3425 			 * Either monitor mode has been turned on for
3426 			 * the device, or we've been given a different
3427 			 * device to open for monitor mode.  If we've
3428 			 * been given a different device, use it.
3429 			 */
3430 			if (handlep->mondevice != NULL)
3431 				device = handlep->mondevice;
3432 		}
3433 		arptype	= iface_get_arptype(sock_fd, device, handle->errbuf);
3434 		if (arptype < 0) {
3435 			close(sock_fd);
3436 			return arptype;
3437 		}
3438 		map_arphrd_to_dlt(handle, sock_fd, arptype, device, 1);
3439 		if (handle->linktype == -1 ||
3440 		    handle->linktype == DLT_LINUX_SLL ||
3441 		    handle->linktype == DLT_LINUX_IRDA ||
3442 		    handle->linktype == DLT_LINUX_LAPD ||
3443 		    handle->linktype == DLT_NETLINK ||
3444 		    (handle->linktype == DLT_EN10MB &&
3445 		     (strncmp("isdn", device, 4) == 0 ||
3446 		      strncmp("isdY", device, 4) == 0))) {
3447 			/*
3448 			 * Unknown interface type (-1), or a
3449 			 * device we explicitly chose to run
3450 			 * in cooked mode (e.g., PPP devices),
3451 			 * or an ISDN device (whose link-layer
3452 			 * type we can only determine by using
3453 			 * APIs that may be different on different
3454 			 * kernels) - reopen in cooked mode.
3455 			 */
3456 			if (close(sock_fd) == -1) {
3457 				pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3458 					 "close: %s", pcap_strerror(errno));
3459 				return PCAP_ERROR;
3460 			}
3461 			sock_fd = socket(PF_PACKET, SOCK_DGRAM,
3462 			    htons(ETH_P_ALL));
3463 			if (sock_fd == -1) {
3464 				pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3465 				    "socket: %s", pcap_strerror(errno));
3466 				if (errno == EPERM || errno == EACCES) {
3467 					/*
3468 					 * You don't have permission to
3469 					 * open the socket.
3470 					 */
3471 					return PCAP_ERROR_PERM_DENIED;
3472 				} else {
3473 					/*
3474 					 * Other error.
3475 					 */
3476 					return PCAP_ERROR;
3477 				}
3478 			}
3479 			handlep->cooked = 1;
3480 
3481 			/*
3482 			 * Get rid of any link-layer type list
3483 			 * we allocated - this only supports cooked
3484 			 * capture.
3485 			 */
3486 			if (handle->dlt_list != NULL) {
3487 				free(handle->dlt_list);
3488 				handle->dlt_list = NULL;
3489 				handle->dlt_count = 0;
3490 			}
3491 
3492 			if (handle->linktype == -1) {
3493 				/*
3494 				 * Warn that we're falling back on
3495 				 * cooked mode; we may want to
3496 				 * update "map_arphrd_to_dlt()"
3497 				 * to handle the new type.
3498 				 */
3499 				pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3500 					"arptype %d not "
3501 					"supported by libpcap - "
3502 					"falling back to cooked "
3503 					"socket",
3504 					arptype);
3505 			}
3506 
3507 			/*
3508 			 * IrDA capture is not a real "cooked" capture,
3509 			 * it's IrLAP frames, not IP packets.  The
3510 			 * same applies to LAPD capture.
3511 			 */
3512 			if (handle->linktype != DLT_LINUX_IRDA &&
3513 			    handle->linktype != DLT_LINUX_LAPD &&
3514 			    handle->linktype != DLT_NETLINK)
3515 				handle->linktype = DLT_LINUX_SLL;
3516 		}
3517 
3518 		handlep->ifindex = iface_get_id(sock_fd, device,
3519 		    handle->errbuf);
3520 		if (handlep->ifindex == -1) {
3521 			close(sock_fd);
3522 			return PCAP_ERROR;
3523 		}
3524 
3525 		if ((err = iface_bind(sock_fd, handlep->ifindex,
3526 		    handle->errbuf)) != 1) {
3527 		    	close(sock_fd);
3528 			if (err < 0)
3529 				return err;
3530 			else
3531 				return 0;	/* try old mechanism */
3532 		}
3533 	} else {
3534 		/*
3535 		 * The "any" device.
3536 		 */
3537 		if (handle->opt.rfmon) {
3538 			/*
3539 			 * It doesn't support monitor mode.
3540 			 */
3541 			close(sock_fd);
3542 			return PCAP_ERROR_RFMON_NOTSUP;
3543 		}
3544 
3545 		/*
3546 		 * It uses cooked mode.
3547 		 */
3548 		handlep->cooked = 1;
3549 		handle->linktype = DLT_LINUX_SLL;
3550 
3551 		/*
3552 		 * We're not bound to a device.
3553 		 * For now, we're using this as an indication
3554 		 * that we can't transmit; stop doing that only
3555 		 * if we figure out how to transmit in cooked
3556 		 * mode.
3557 		 */
3558 		handlep->ifindex = -1;
3559 	}
3560 
3561 	/*
3562 	 * Select promiscuous mode on if "promisc" is set.
3563 	 *
3564 	 * Do not turn allmulti mode on if we don't select
3565 	 * promiscuous mode - on some devices (e.g., Orinoco
3566 	 * wireless interfaces), allmulti mode isn't supported
3567 	 * and the driver implements it by turning promiscuous
3568 	 * mode on, and that screws up the operation of the
3569 	 * card as a normal networking interface, and on no
3570 	 * other platform I know of does starting a non-
3571 	 * promiscuous capture affect which multicast packets
3572 	 * are received by the interface.
3573 	 */
3574 
3575 	/*
3576 	 * Hmm, how can we set promiscuous mode on all interfaces?
3577 	 * I am not sure if that is possible at all.  For now, we
3578 	 * silently ignore attempts to turn promiscuous mode on
3579 	 * for the "any" device (so you don't have to explicitly
3580 	 * disable it in programs such as tcpdump).
3581 	 */
3582 
3583 	if (!is_any_device && handle->opt.promisc) {
3584 		memset(&mr, 0, sizeof(mr));
3585 		mr.mr_ifindex = handlep->ifindex;
3586 		mr.mr_type    = PACKET_MR_PROMISC;
3587 		if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
3588 		    &mr, sizeof(mr)) == -1) {
3589 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3590 				"setsockopt: %s", pcap_strerror(errno));
3591 			close(sock_fd);
3592 			return PCAP_ERROR;
3593 		}
3594 	}
3595 
3596 	/* Enable auxillary data if supported and reserve room for
3597 	 * reconstructing VLAN headers. */
3598 #ifdef HAVE_PACKET_AUXDATA
3599 	val = 1;
3600 	if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
3601 		       sizeof(val)) == -1 && errno != ENOPROTOOPT) {
3602 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3603 			 "setsockopt: %s", pcap_strerror(errno));
3604 		close(sock_fd);
3605 		return PCAP_ERROR;
3606 	}
3607 	handle->offset += VLAN_TAG_LEN;
3608 #endif /* HAVE_PACKET_AUXDATA */
3609 
3610 	/*
3611 	 * This is a 2.2[.x] or later kernel (we know that
3612 	 * because we're not using a SOCK_PACKET socket -
3613 	 * PF_PACKET is supported only in 2.2 and later
3614 	 * kernels).
3615 	 *
3616 	 * We can safely pass "recvfrom()" a byte count
3617 	 * based on the snapshot length.
3618 	 *
3619 	 * If we're in cooked mode, make the snapshot length
3620 	 * large enough to hold a "cooked mode" header plus
3621 	 * 1 byte of packet data (so we don't pass a byte
3622 	 * count of 0 to "recvfrom()").
3623 	 */
3624 	if (handlep->cooked) {
3625 		if (handle->snapshot < SLL_HDR_LEN + 1)
3626 			handle->snapshot = SLL_HDR_LEN + 1;
3627 	}
3628 	handle->bufsize = handle->snapshot;
3629 
3630 	/*
3631 	 * Set the offset at which to insert VLAN tags.
3632 	 * That should be the offset of the type field.
3633 	 */
3634 	switch (handle->linktype) {
3635 
3636 	case DLT_EN10MB:
3637 		/*
3638 		 * The type field is after the destination and source
3639 		 * MAC address.
3640 		 */
3641 		handlep->vlan_offset = 2 * ETH_ALEN;
3642 		break;
3643 
3644 	case DLT_LINUX_SLL:
3645 		/*
3646 		 * The type field is in the last 2 bytes of the
3647 		 * DLT_LINUX_SLL header.
3648 		 */
3649 		handlep->vlan_offset = SLL_HDR_LEN - 2;
3650 		break;
3651 
3652 	default:
3653 		handlep->vlan_offset = -1; /* unknown */
3654 		break;
3655 	}
3656 
3657 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
3658 	if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
3659 		int nsec_tstamps = 1;
3660 
3661 		if (setsockopt(sock_fd, SOL_SOCKET, SO_TIMESTAMPNS, &nsec_tstamps, sizeof(nsec_tstamps)) < 0) {
3662 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "setsockopt: unable to set SO_TIMESTAMPNS");
3663 			close(sock_fd);
3664 			return PCAP_ERROR;
3665 		}
3666 	}
3667 #endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */
3668 
3669 	/*
3670 	 * We've succeeded. Save the socket FD in the pcap structure.
3671 	 */
3672 	handle->fd = sock_fd;
3673 
3674 #if defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT)
3675 	/*
3676 	 * Can we generate special code for VLAN checks?
3677 	 * (XXX - what if we need the special code but it's not supported
3678 	 * by the OS?  Is that possible?)
3679 	 */
3680 	if (getsockopt(sock_fd, SOL_SOCKET, SO_BPF_EXTENSIONS,
3681 	    &bpf_extensions, &len) == 0) {
3682 		if (bpf_extensions >= SKF_AD_VLAN_TAG_PRESENT) {
3683 			/*
3684 			 * Yes, we can.  Request that we do so.
3685 			 */
3686 			handle->bpf_codegen_flags |= BPF_SPECIAL_VLAN_HANDLING;
3687 		}
3688 	}
3689 #endif /* defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT) */
3690 
3691 	return 1;
3692 #else /* HAVE_PF_PACKET_SOCKETS */
3693 	strlcpy(ebuf,
3694 		"New packet capturing interface not supported by build "
3695 		"environment", PCAP_ERRBUF_SIZE);
3696 	return 0;
3697 #endif /* HAVE_PF_PACKET_SOCKETS */
3698 }
3699 
3700 #ifdef HAVE_PACKET_RING
3701 /*
3702  * Attempt to activate with memory-mapped access.
3703  *
3704  * On success, returns 1, and sets *status to 0 if there are no warnings
3705  * or to a PCAP_WARNING_ code if there is a warning.
3706  *
3707  * On failure due to lack of support for memory-mapped capture, returns
3708  * 0.
3709  *
3710  * On error, returns -1, and sets *status to the appropriate error code;
3711  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3712  */
3713 static int
3714 activate_mmap(pcap_t *handle, int *status)
3715 {
3716 	struct pcap_linux *handlep = handle->priv;
3717 	int ret;
3718 
3719 	/*
3720 	 * Attempt to allocate a buffer to hold the contents of one
3721 	 * packet, for use by the oneshot callback.
3722 	 */
3723 	handlep->oneshot_buffer = malloc(handle->snapshot);
3724 	if (handlep->oneshot_buffer == NULL) {
3725 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3726 			 "can't allocate oneshot buffer: %s",
3727 			 pcap_strerror(errno));
3728 		*status = PCAP_ERROR;
3729 		return -1;
3730 	}
3731 
3732 	if (handle->opt.buffer_size == 0) {
3733 		/* by default request 2M for the ring buffer */
3734 		handle->opt.buffer_size = 2*1024*1024;
3735 	}
3736 	ret = prepare_tpacket_socket(handle);
3737 	if (ret == -1) {
3738 		free(handlep->oneshot_buffer);
3739 		*status = PCAP_ERROR;
3740 		return ret;
3741 	}
3742 	ret = create_ring(handle, status);
3743 	if (ret == 0) {
3744 		/*
3745 		 * We don't support memory-mapped capture; our caller
3746 		 * will fall back on reading from the socket.
3747 		 */
3748 		free(handlep->oneshot_buffer);
3749 		return 0;
3750 	}
3751 	if (ret == -1) {
3752 		/*
3753 		 * Error attempting to enable memory-mapped capture;
3754 		 * fail.  create_ring() has set *status.
3755 		 */
3756 		free(handlep->oneshot_buffer);
3757 		return -1;
3758 	}
3759 
3760 	/*
3761 	 * Success.  *status has been set either to 0 if there are no
3762 	 * warnings or to a PCAP_WARNING_ value if there is a warning.
3763 	 *
3764 	 * Override some defaults and inherit the other fields from
3765 	 * activate_new.
3766 	 * handle->offset is used to get the current position into the rx ring.
3767 	 * handle->cc is used to store the ring size.
3768 	 */
3769 
3770 	switch (handlep->tp_version) {
3771 	case TPACKET_V1:
3772 		handle->read_op = pcap_read_linux_mmap_v1;
3773 		break;
3774 	case TPACKET_V1_64:
3775 		handle->read_op = pcap_read_linux_mmap_v1_64;
3776 		break;
3777 #ifdef HAVE_TPACKET2
3778 	case TPACKET_V2:
3779 		handle->read_op = pcap_read_linux_mmap_v2;
3780 		break;
3781 #endif
3782 #ifdef HAVE_TPACKET3
3783 	case TPACKET_V3:
3784 		handle->read_op = pcap_read_linux_mmap_v3;
3785 		break;
3786 #endif
3787 	}
3788 	handle->cleanup_op = pcap_cleanup_linux_mmap;
3789 	handle->setfilter_op = pcap_setfilter_linux_mmap;
3790 	handle->setnonblock_op = pcap_setnonblock_mmap;
3791 	handle->getnonblock_op = pcap_getnonblock_mmap;
3792 	handle->oneshot_callback = pcap_oneshot_mmap;
3793 	handle->selectable_fd = handle->fd;
3794 	return 1;
3795 }
3796 #else /* HAVE_PACKET_RING */
3797 static int
3798 activate_mmap(pcap_t *handle _U_, int *status _U_)
3799 {
3800 	return 0;
3801 }
3802 #endif /* HAVE_PACKET_RING */
3803 
3804 #ifdef HAVE_PACKET_RING
3805 
3806 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
3807 /*
3808  * Attempt to set the socket to the specified version of the memory-mapped
3809  * header.
3810  *
3811  * Return 0 if we succeed; return 1 if we fail because that version isn't
3812  * supported; return -1 on any other error, and set handle->errbuf.
3813  */
3814 static int
3815 init_tpacket(pcap_t *handle, int version, const char *version_str)
3816 {
3817 	struct pcap_linux *handlep = handle->priv;
3818 	int val = version;
3819 	socklen_t len = sizeof(val);
3820 
3821 	/*
3822 	 * Probe whether kernel supports the specified TPACKET version;
3823 	 * this also gets the length of the header for that version.
3824 	 */
3825 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
3826 		if (errno == ENOPROTOOPT || errno == EINVAL)
3827 			return 1;	/* no */
3828 
3829 		/* Failed to even find out; this is a fatal error. */
3830 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3831 			"can't get %s header len on packet socket: %s",
3832 			version_str,
3833 			pcap_strerror(errno));
3834 		return -1;
3835 	}
3836 	handlep->tp_hdrlen = val;
3837 
3838 	val = version;
3839 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
3840 			   sizeof(val)) < 0) {
3841 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3842 			"can't activate %s on packet socket: %s",
3843 			version_str,
3844 			pcap_strerror(errno));
3845 		return -1;
3846 	}
3847 	handlep->tp_version = version;
3848 
3849 	/* Reserve space for VLAN tag reconstruction */
3850 	val = VLAN_TAG_LEN;
3851 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val,
3852 			   sizeof(val)) < 0) {
3853 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3854 			"can't set up reserve on packet socket: %s",
3855 			pcap_strerror(errno));
3856 		return -1;
3857 	}
3858 
3859 	return 0;
3860 }
3861 #endif /* defined HAVE_TPACKET2 || defined HAVE_TPACKET3 */
3862 
3863 /*
3864  * If the instruction set for which we're compiling has both 32-bit
3865  * and 64-bit versions, and Linux support for the 64-bit version
3866  * predates TPACKET_V2, define ISA_64_BIT as the .machine value
3867  * you get from uname() for the 64-bit version.  Otherwise, leave
3868  * it undefined.  (This includes ARM, which has a 64-bit version,
3869  * but Linux support for it appeared well after TPACKET_V2 support
3870  * did, so there should never be a case where 32-bit ARM code is
3871  * running o a 64-bit kernel that only supports TPACKET_V1.)
3872  *
3873  * If we've omitted your favorite such architecture, please contribute
3874  * a patch.  (No patch is needed for architectures that are 32-bit-only
3875  * or for which Linux has no support for 32-bit userland - or for which,
3876  * as noted, 64-bit support appeared in Linux after TPACKET_V2 support
3877  * did.)
3878  */
3879 #if defined(__i386__)
3880 #define ISA_64_BIT	"x86_64"
3881 #elif defined(__ppc__)
3882 #define ISA_64_BIT	"ppc64"
3883 #elif defined(__sparc__)
3884 #define ISA_64_BIT	"sparc64"
3885 #elif defined(__s390__)
3886 #define ISA_64_BIT	"s390x"
3887 #elif defined(__mips__)
3888 #define ISA_64_BIT	"mips64"
3889 #elif defined(__hppa__)
3890 #define ISA_64_BIT	"parisc64"
3891 #endif
3892 
3893 /*
3894  * Attempt to set the socket to version 3 of the memory-mapped header and,
3895  * if that fails because version 3 isn't supported, attempt to fall
3896  * back to version 2.  If version 2 isn't supported, just leave it at
3897  * version 1.
3898  *
3899  * Return 1 if we succeed or if we fail because neither version 2 nor 3 is
3900  * supported; return -1 on any other error, and set handle->errbuf.
3901  */
3902 static int
3903 prepare_tpacket_socket(pcap_t *handle)
3904 {
3905 	struct pcap_linux *handlep = handle->priv;
3906 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
3907 	int ret;
3908 #endif
3909 
3910 #ifdef HAVE_TPACKET3
3911 	/*
3912 	 * Try setting the version to TPACKET_V3.
3913 	 *
3914 	 * The only mode in which buffering is done on PF_PACKET
3915 	 * sockets, so that packets might not be delivered
3916 	 * immediately, is TPACKET_V3 mode.
3917 	 *
3918 	 * The buffering cannot be disabled in that mode, so
3919 	 * if the user has requested immediate mode, we don't
3920 	 * use TPACKET_V3.
3921 	 */
3922 	if (!handle->opt.immediate) {
3923 		ret = init_tpacket(handle, TPACKET_V3, "TPACKET_V3");
3924 		if (ret == 0) {
3925 			/*
3926 			 * Success.
3927 			 */
3928 			return 1;
3929 		}
3930 		if (ret == -1) {
3931 			/*
3932 			 * We failed for some reason other than "the
3933 			 * kernel doesn't support TPACKET_V3".
3934 			 */
3935 			return -1;
3936 		}
3937 	}
3938 #endif /* HAVE_TPACKET3 */
3939 
3940 #ifdef HAVE_TPACKET2
3941 	/*
3942 	 * Try setting the version to TPACKET_V2.
3943 	 */
3944 	ret = init_tpacket(handle, TPACKET_V2, "TPACKET_V2");
3945 	if (ret == 0) {
3946 		/*
3947 		 * Success.
3948 		 */
3949 		return 1;
3950 	}
3951 	if (ret == -1) {
3952 		/*
3953 		 * We failed for some reason other than "the
3954 		 * kernel doesn't support TPACKET_V2".
3955 		 */
3956 		return -1;
3957 	}
3958 #endif /* HAVE_TPACKET2 */
3959 
3960 	/*
3961 	 * OK, we're using TPACKET_V1, as that's all the kernel supports.
3962 	 */
3963 	handlep->tp_version = TPACKET_V1;
3964 	handlep->tp_hdrlen = sizeof(struct tpacket_hdr);
3965 
3966 #ifdef ISA_64_BIT
3967 	/*
3968 	 * 32-bit userspace + 64-bit kernel + TPACKET_V1 are not compatible with
3969 	 * each other due to platform-dependent data type size differences.
3970 	 *
3971 	 * If we have a 32-bit userland and a 64-bit kernel, use an
3972 	 * internally-defined TPACKET_V1_64, with which we use a 64-bit
3973 	 * version of the data structures.
3974 	 */
3975 	if (sizeof(long) == 4) {
3976 		/*
3977 		 * This is 32-bit code.
3978 		 */
3979 		struct utsname utsname;
3980 
3981 		if (uname(&utsname) == -1) {
3982 			/*
3983 			 * Failed.
3984 			 */
3985 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3986 			    "uname failed: %s", pcap_strerror(errno));
3987 			return -1;
3988 		}
3989 		if (strcmp(utsname.machine, ISA_64_BIT) == 0) {
3990 			/*
3991 			 * uname() tells us the machine is 64-bit,
3992 			 * so we presumably have a 64-bit kernel.
3993 			 *
3994 			 * XXX - this presumes that uname() won't lie
3995 			 * in 32-bit code and claim that the machine
3996 			 * has the 32-bit version of the ISA.
3997 			 */
3998 			handlep->tp_version = TPACKET_V1_64;
3999 			handlep->tp_hdrlen = sizeof(struct tpacket_hdr_64);
4000 		}
4001 	}
4002 #endif
4003 
4004 	return 1;
4005 }
4006 
4007 /*
4008  * Attempt to set up memory-mapped access.
4009  *
4010  * On success, returns 1, and sets *status to 0 if there are no warnings
4011  * or to a PCAP_WARNING_ code if there is a warning.
4012  *
4013  * On failure due to lack of support for memory-mapped capture, returns
4014  * 0.
4015  *
4016  * On error, returns -1, and sets *status to the appropriate error code;
4017  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
4018  */
4019 static int
4020 create_ring(pcap_t *handle, int *status)
4021 {
4022 	struct pcap_linux *handlep = handle->priv;
4023 	unsigned i, j, frames_per_block;
4024 #ifdef HAVE_TPACKET3
4025 	/*
4026 	 * For sockets using TPACKET_V1 or TPACKET_V2, the extra
4027 	 * stuff at the end of a struct tpacket_req3 will be
4028 	 * ignored, so this is OK even for those sockets.
4029 	 */
4030 	struct tpacket_req3 req;
4031 #else
4032 	struct tpacket_req req;
4033 #endif
4034 	socklen_t len;
4035 	unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
4036 	unsigned int frame_size;
4037 
4038 	/*
4039 	 * Start out assuming no warnings or errors.
4040 	 */
4041 	*status = 0;
4042 
4043 	switch (handlep->tp_version) {
4044 
4045 	case TPACKET_V1:
4046 	case TPACKET_V1_64:
4047 #ifdef HAVE_TPACKET2
4048 	case TPACKET_V2:
4049 #endif
4050 		/* Note that with large snapshot length (say 64K, which is
4051 		 * the default for recent versions of tcpdump, the value that
4052 		 * "-s 0" has given for a long time with tcpdump, and the
4053 		 * default in Wireshark/TShark/dumpcap), if we use the snapshot
4054 		 * length to calculate the frame length, only a few frames
4055 		 * will be available in the ring even with pretty
4056 		 * large ring size (and a lot of memory will be unused).
4057 		 *
4058 		 * Ideally, we should choose a frame length based on the
4059 		 * minimum of the specified snapshot length and the maximum
4060 		 * packet size.  That's not as easy as it sounds; consider,
4061 		 * for example, an 802.11 interface in monitor mode, where
4062 		 * the frame would include a radiotap header, where the
4063 		 * maximum radiotap header length is device-dependent.
4064 		 *
4065 		 * So, for now, we just do this for Ethernet devices, where
4066 		 * there's no metadata header, and the link-layer header is
4067 		 * fixed length.  We can get the maximum packet size by
4068 		 * adding 18, the Ethernet header length plus the CRC length
4069 		 * (just in case we happen to get the CRC in the packet), to
4070 		 * the MTU of the interface; we fetch the MTU in the hopes
4071 		 * that it reflects support for jumbo frames.  (Even if the
4072 		 * interface is just being used for passive snooping, the
4073 		 * driver might set the size of buffers in the receive ring
4074 		 * based on the MTU, so that the MTU limits the maximum size
4075 		 * of packets that we can receive.)
4076 		 *
4077 		 * We don't do that if segmentation/fragmentation or receive
4078 		 * offload are enabled, so we don't get rudely surprised by
4079 		 * "packets" bigger than the MTU. */
4080 		frame_size = handle->snapshot;
4081 		if (handle->linktype == DLT_EN10MB) {
4082 			int mtu;
4083 			int offload;
4084 
4085 			offload = iface_get_offload(handle);
4086 			if (offload == -1) {
4087 				*status = PCAP_ERROR;
4088 				return -1;
4089 			}
4090 			if (!offload) {
4091 				mtu = iface_get_mtu(handle->fd, handle->opt.device,
4092 				    handle->errbuf);
4093 				if (mtu == -1) {
4094 					*status = PCAP_ERROR;
4095 					return -1;
4096 				}
4097 				if (frame_size > (unsigned int)mtu + 18)
4098 					frame_size = (unsigned int)mtu + 18;
4099 			}
4100 		}
4101 
4102 		/* NOTE: calculus matching those in tpacket_rcv()
4103 		 * in linux-2.6/net/packet/af_packet.c
4104 		 */
4105 		len = sizeof(sk_type);
4106 		if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type,
4107 		    &len) < 0) {
4108 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4109 			    "getsockopt: %s", pcap_strerror(errno));
4110 			*status = PCAP_ERROR;
4111 			return -1;
4112 		}
4113 #ifdef PACKET_RESERVE
4114 		len = sizeof(tp_reserve);
4115 		if (getsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE,
4116 		    &tp_reserve, &len) < 0) {
4117 			if (errno != ENOPROTOOPT) {
4118 				/*
4119 				 * ENOPROTOOPT means "kernel doesn't support
4120 				 * PACKET_RESERVE", in which case we fall back
4121 				 * as best we can.
4122 				 */
4123 				pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4124 				    "getsockopt: %s", pcap_strerror(errno));
4125 				*status = PCAP_ERROR;
4126 				return -1;
4127 			}
4128 			tp_reserve = 0;	/* older kernel, reserve not supported */
4129 		}
4130 #else
4131 		tp_reserve = 0;	/* older kernel, reserve not supported */
4132 #endif
4133 		maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
4134 			/* XXX: in the kernel maclen is calculated from
4135 			 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
4136 			 * in:  packet_snd()           in linux-2.6/net/packet/af_packet.c
4137 			 * then packet_alloc_skb()     in linux-2.6/net/packet/af_packet.c
4138 			 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
4139 			 * but I see no way to get those sizes in userspace,
4140 			 * like for instance with an ifreq ioctl();
4141 			 * the best thing I've found so far is MAX_HEADER in
4142 			 * the kernel part of linux-2.6/include/linux/netdevice.h
4143 			 * which goes up to 128+48=176; since pcap-linux.c
4144 			 * defines a MAX_LINKHEADER_SIZE of 256 which is
4145 			 * greater than that, let's use it.. maybe is it even
4146 			 * large enough to directly replace macoff..
4147 			 */
4148 		tp_hdrlen = TPACKET_ALIGN(handlep->tp_hdrlen) + sizeof(struct sockaddr_ll) ;
4149 		netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
4150 			/* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN
4151 			 * of netoff, which contradicts
4152 			 * linux-2.6/Documentation/networking/packet_mmap.txt
4153 			 * documenting that:
4154 			 * "- Gap, chosen so that packet data (Start+tp_net)
4155 			 * aligns to TPACKET_ALIGNMENT=16"
4156 			 */
4157 			/* NOTE: in linux-2.6/include/linux/skbuff.h:
4158 			 * "CPUs often take a performance hit
4159 			 *  when accessing unaligned memory locations"
4160 			 */
4161 		macoff = netoff - maclen;
4162 		req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
4163 		req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
4164 		break;
4165 
4166 #ifdef HAVE_TPACKET3
4167 	case TPACKET_V3:
4168 		/* The "frames" for this are actually buffers that
4169 		 * contain multiple variable-sized frames.
4170 		 *
4171 		 * We pick a "frame" size of 128K to leave enough
4172 		 * room for at least one reasonably-sized packet
4173 		 * in the "frame". */
4174 		req.tp_frame_size = MAXIMUM_SNAPLEN;
4175 		req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
4176 		break;
4177 #endif
4178 	default:
4179 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4180 		    "Internal error: unknown TPACKET_ value %u",
4181 		    handlep->tp_version);
4182 		*status = PCAP_ERROR;
4183 		return -1;
4184 	}
4185 
4186 	/* compute the minumum block size that will handle this frame.
4187 	 * The block has to be page size aligned.
4188 	 * The max block size allowed by the kernel is arch-dependent and
4189 	 * it's not explicitly checked here. */
4190 	req.tp_block_size = getpagesize();
4191 	while (req.tp_block_size < req.tp_frame_size)
4192 		req.tp_block_size <<= 1;
4193 
4194 	frames_per_block = req.tp_block_size/req.tp_frame_size;
4195 
4196 	/*
4197 	 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
4198 	 * so we check for PACKET_TIMESTAMP.  We check for
4199 	 * linux/net_tstamp.h just in case a system somehow has
4200 	 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
4201 	 * be unnecessary.
4202 	 *
4203 	 * SIOCSHWTSTAMP was introduced in the patch that introduced
4204 	 * linux/net_tstamp.h, so we don't bother checking whether
4205 	 * SIOCSHWTSTAMP is defined (if your Linux system has
4206 	 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
4207 	 * Linux system is badly broken).
4208 	 */
4209 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
4210 	/*
4211 	 * If we were told to do so, ask the kernel and the driver
4212 	 * to use hardware timestamps.
4213 	 *
4214 	 * Hardware timestamps are only supported with mmapped
4215 	 * captures.
4216 	 */
4217 	if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
4218 	    handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
4219 		struct hwtstamp_config hwconfig;
4220 		struct ifreq ifr;
4221 		int timesource;
4222 
4223 		/*
4224 		 * Ask for hardware time stamps on all packets,
4225 		 * including transmitted packets.
4226 		 */
4227 		memset(&hwconfig, 0, sizeof(hwconfig));
4228 		hwconfig.tx_type = HWTSTAMP_TX_ON;
4229 		hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
4230 
4231 		memset(&ifr, 0, sizeof(ifr));
4232 		strlcpy(ifr.ifr_name, handle->opt.device, sizeof(ifr.ifr_name));
4233 		ifr.ifr_data = (void *)&hwconfig;
4234 
4235 		if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
4236 			switch (errno) {
4237 
4238 			case EPERM:
4239 				/*
4240 				 * Treat this as an error, as the
4241 				 * user should try to run this
4242 				 * with the appropriate privileges -
4243 				 * and, if they can't, shouldn't
4244 				 * try requesting hardware time stamps.
4245 				 */
4246 				*status = PCAP_ERROR_PERM_DENIED;
4247 				return -1;
4248 
4249 			case EOPNOTSUPP:
4250 			case ERANGE:
4251 				/*
4252 				 * Treat this as a warning, as the
4253 				 * only way to fix the warning is to
4254 				 * get an adapter that supports hardware
4255 				 * time stamps for *all* packets.
4256 				 * (ERANGE means "we support hardware
4257 				 * time stamps, but for packets matching
4258 				 * that particular filter", so it means
4259 				 * "we don't support hardware time stamps
4260 				 * for all incoming packets" here.)
4261 				 *
4262 				 * We'll just fall back on the standard
4263 				 * host time stamps.
4264 				 */
4265 				*status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
4266 				break;
4267 
4268 			default:
4269 				pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4270 					"SIOCSHWTSTAMP failed: %s",
4271 					pcap_strerror(errno));
4272 				*status = PCAP_ERROR;
4273 				return -1;
4274 			}
4275 		} else {
4276 			/*
4277 			 * Well, that worked.  Now specify the type of
4278 			 * hardware time stamp we want for this
4279 			 * socket.
4280 			 */
4281 			if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
4282 				/*
4283 				 * Hardware timestamp, synchronized
4284 				 * with the system clock.
4285 				 */
4286 				timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
4287 			} else {
4288 				/*
4289 				 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
4290 				 * timestamp, not synchronized with the
4291 				 * system clock.
4292 				 */
4293 				timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
4294 			}
4295 			if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
4296 				(void *)&timesource, sizeof(timesource))) {
4297 				pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4298 					"can't set PACKET_TIMESTAMP: %s",
4299 					pcap_strerror(errno));
4300 				*status = PCAP_ERROR;
4301 				return -1;
4302 			}
4303 		}
4304 	}
4305 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
4306 
4307 	/* ask the kernel to create the ring */
4308 retry:
4309 	req.tp_block_nr = req.tp_frame_nr / frames_per_block;
4310 
4311 	/* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
4312 	req.tp_frame_nr = req.tp_block_nr * frames_per_block;
4313 
4314 #ifdef HAVE_TPACKET3
4315 	/* timeout value to retire block - use the configured buffering timeout, or default if <0. */
4316 	req.tp_retire_blk_tov = (handlep->timeout>=0)?handlep->timeout:0;
4317 	/* private data not used */
4318 	req.tp_sizeof_priv = 0;
4319 	/* Rx ring - feature request bits - none (rxhash will not be filled) */
4320 	req.tp_feature_req_word = 0;
4321 #endif
4322 
4323 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
4324 					(void *) &req, sizeof(req))) {
4325 		if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
4326 			/*
4327 			 * Memory failure; try to reduce the requested ring
4328 			 * size.
4329 			 *
4330 			 * We used to reduce this by half -- do 5% instead.
4331 			 * That may result in more iterations and a longer
4332 			 * startup, but the user will be much happier with
4333 			 * the resulting buffer size.
4334 			 */
4335 			if (req.tp_frame_nr < 20)
4336 				req.tp_frame_nr -= 1;
4337 			else
4338 				req.tp_frame_nr -= req.tp_frame_nr/20;
4339 			goto retry;
4340 		}
4341 		if (errno == ENOPROTOOPT) {
4342 			/*
4343 			 * We don't have ring buffer support in this kernel.
4344 			 */
4345 			return 0;
4346 		}
4347 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4348 		    "can't create rx ring on packet socket: %s",
4349 		    pcap_strerror(errno));
4350 		*status = PCAP_ERROR;
4351 		return -1;
4352 	}
4353 
4354 	/* memory map the rx ring */
4355 	handlep->mmapbuflen = req.tp_block_nr * req.tp_block_size;
4356 	handlep->mmapbuf = mmap(0, handlep->mmapbuflen,
4357 	    PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
4358 	if (handlep->mmapbuf == MAP_FAILED) {
4359 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4360 		    "can't mmap rx ring: %s", pcap_strerror(errno));
4361 
4362 		/* clear the allocated ring on error*/
4363 		destroy_ring(handle);
4364 		*status = PCAP_ERROR;
4365 		return -1;
4366 	}
4367 
4368 	/* allocate a ring for each frame header pointer*/
4369 	handle->cc = req.tp_frame_nr;
4370 	handle->buffer = malloc(handle->cc * sizeof(union thdr *));
4371 	if (!handle->buffer) {
4372 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4373 		    "can't allocate ring of frame headers: %s",
4374 		    pcap_strerror(errno));
4375 
4376 		destroy_ring(handle);
4377 		*status = PCAP_ERROR;
4378 		return -1;
4379 	}
4380 
4381 	/* fill the header ring with proper frame ptr*/
4382 	handle->offset = 0;
4383 	for (i=0; i<req.tp_block_nr; ++i) {
4384 		void *base = &handlep->mmapbuf[i*req.tp_block_size];
4385 		for (j=0; j<frames_per_block; ++j, ++handle->offset) {
4386 			RING_GET_CURRENT_FRAME(handle) = base;
4387 			base += req.tp_frame_size;
4388 		}
4389 	}
4390 
4391 	handle->bufsize = req.tp_frame_size;
4392 	handle->offset = 0;
4393 	return 1;
4394 }
4395 
4396 /* free all ring related resources*/
4397 static void
4398 destroy_ring(pcap_t *handle)
4399 {
4400 	struct pcap_linux *handlep = handle->priv;
4401 
4402 	/* tell the kernel to destroy the ring*/
4403 	struct tpacket_req req;
4404 	memset(&req, 0, sizeof(req));
4405 	/* do not test for setsockopt failure, as we can't recover from any error */
4406 	(void)setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
4407 				(void *) &req, sizeof(req));
4408 
4409 	/* if ring is mapped, unmap it*/
4410 	if (handlep->mmapbuf) {
4411 		/* do not test for mmap failure, as we can't recover from any error */
4412 		(void)munmap(handlep->mmapbuf, handlep->mmapbuflen);
4413 		handlep->mmapbuf = NULL;
4414 	}
4415 }
4416 
4417 /*
4418  * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
4419  * for Linux mmapped capture.
4420  *
4421  * The problem is that pcap_next() and pcap_next_ex() expect the packet
4422  * data handed to the callback to be valid after the callback returns,
4423  * but pcap_read_linux_mmap() has to release that packet as soon as
4424  * the callback returns (otherwise, the kernel thinks there's still
4425  * at least one unprocessed packet available in the ring, so a select()
4426  * will immediately return indicating that there's data to process), so,
4427  * in the callback, we have to make a copy of the packet.
4428  *
4429  * Yes, this means that, if the capture is using the ring buffer, using
4430  * pcap_next() or pcap_next_ex() requires more copies than using
4431  * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
4432  * pcap_next() or pcap_next_ex().
4433  */
4434 static void
4435 pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
4436     const u_char *bytes)
4437 {
4438 	struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
4439 	pcap_t *handle = sp->pd;
4440 	struct pcap_linux *handlep = handle->priv;
4441 
4442 	*sp->hdr = *h;
4443 	memcpy(handlep->oneshot_buffer, bytes, h->caplen);
4444 	*sp->pkt = handlep->oneshot_buffer;
4445 }
4446 
4447 static void
4448 pcap_cleanup_linux_mmap( pcap_t *handle )
4449 {
4450 	struct pcap_linux *handlep = handle->priv;
4451 
4452 	destroy_ring(handle);
4453 	if (handlep->oneshot_buffer != NULL) {
4454 		free(handlep->oneshot_buffer);
4455 		handlep->oneshot_buffer = NULL;
4456 	}
4457 	pcap_cleanup_linux(handle);
4458 }
4459 
4460 
4461 static int
4462 pcap_getnonblock_mmap(pcap_t *p, char *errbuf)
4463 {
4464 	struct pcap_linux *handlep = p->priv;
4465 
4466 	/* use negative value of timeout to indicate non blocking ops */
4467 	return (handlep->timeout<0);
4468 }
4469 
4470 static int
4471 pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
4472 {
4473 	struct pcap_linux *handlep = p->priv;
4474 
4475 	/*
4476 	 * Set the file descriptor to non-blocking mode, as we use
4477 	 * it for sending packets.
4478 	 */
4479 	if (pcap_setnonblock_fd(p, nonblock, errbuf) == -1)
4480 		return -1;
4481 
4482 	/*
4483 	 * Map each value to their corresponding negation to
4484 	 * preserve the timeout value provided with pcap_set_timeout.
4485 	 */
4486 	if (nonblock) {
4487 		if (handlep->timeout >= 0) {
4488 			/*
4489 			 * Indicate that we're switching to
4490 			 * non-blocking mode.
4491 			 */
4492 			handlep->timeout = ~handlep->timeout;
4493 		}
4494 	} else {
4495 		if (handlep->timeout < 0) {
4496 			handlep->timeout = ~handlep->timeout;
4497 		}
4498 	}
4499 	/* Update the timeout to use in poll(). */
4500 	set_poll_timeout(handlep);
4501 	return 0;
4502 }
4503 
4504 /*
4505  * Get the status field of the ring buffer frame at a specified offset.
4506  */
4507 static inline int
4508 pcap_get_ring_frame_status(pcap_t *handle, int offset)
4509 {
4510 	struct pcap_linux *handlep = handle->priv;
4511 	union thdr h;
4512 
4513 	h.raw = RING_GET_FRAME_AT(handle, offset);
4514 	switch (handlep->tp_version) {
4515 	case TPACKET_V1:
4516 		return (h.h1->tp_status);
4517 		break;
4518 	case TPACKET_V1_64:
4519 		return (h.h1_64->tp_status);
4520 		break;
4521 #ifdef HAVE_TPACKET2
4522 	case TPACKET_V2:
4523 		return (h.h2->tp_status);
4524 		break;
4525 #endif
4526 #ifdef HAVE_TPACKET3
4527 	case TPACKET_V3:
4528 		return (h.h3->hdr.bh1.block_status);
4529 		break;
4530 #endif
4531 	}
4532 	/* This should not happen. */
4533 	return 0;
4534 }
4535 
4536 #ifndef POLLRDHUP
4537 #define POLLRDHUP 0
4538 #endif
4539 
4540 /*
4541  * Block waiting for frames to be available.
4542  */
4543 static int pcap_wait_for_frames_mmap(pcap_t *handle)
4544 {
4545 	struct pcap_linux *handlep = handle->priv;
4546 	char c;
4547 	struct pollfd pollinfo;
4548 	int ret;
4549 
4550 	pollinfo.fd = handle->fd;
4551 	pollinfo.events = POLLIN;
4552 
4553 	do {
4554 		/*
4555 		 * Yes, we do this even in non-blocking mode, as it's
4556 		 * the only way to get error indications from a
4557 		 * tpacket socket.
4558 		 *
4559 		 * The timeout is 0 in non-blocking mode, so poll()
4560 		 * returns immediately.
4561 		 */
4562 		ret = poll(&pollinfo, 1, handlep->poll_timeout);
4563 		if (ret < 0 && errno != EINTR) {
4564 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4565 				"can't poll on packet socket: %s",
4566 				pcap_strerror(errno));
4567 			return PCAP_ERROR;
4568 		} else if (ret > 0 &&
4569 			(pollinfo.revents & (POLLHUP|POLLRDHUP|POLLERR|POLLNVAL))) {
4570 			/*
4571 			 * There's some indication other than
4572 			 * "you can read on this descriptor" on
4573 			 * the descriptor.
4574 			 */
4575 			if (pollinfo.revents & (POLLHUP | POLLRDHUP)) {
4576 				pcap_snprintf(handle->errbuf,
4577 					PCAP_ERRBUF_SIZE,
4578 					"Hangup on packet socket");
4579 				return PCAP_ERROR;
4580 			}
4581 			if (pollinfo.revents & POLLERR) {
4582 				/*
4583 				 * A recv() will give us the actual error code.
4584 				 *
4585 				 * XXX - make the socket non-blocking?
4586 				 */
4587 				if (recv(handle->fd, &c, sizeof c,
4588 					MSG_PEEK) != -1)
4589 					continue;	/* what, no error? */
4590 				if (errno == ENETDOWN) {
4591 					/*
4592 					 * The device on which we're
4593 					 * capturing went away.
4594 					 *
4595 					 * XXX - we should really return
4596 					 * PCAP_ERROR_IFACE_NOT_UP, but
4597 					 * pcap_dispatch() etc. aren't
4598 					 * defined to return that.
4599 					 */
4600 					pcap_snprintf(handle->errbuf,
4601 						PCAP_ERRBUF_SIZE,
4602 						"The interface went down");
4603 				} else {
4604 					pcap_snprintf(handle->errbuf,
4605 						PCAP_ERRBUF_SIZE,
4606 						"Error condition on packet socket: %s",
4607 						strerror(errno));
4608 				}
4609 				return PCAP_ERROR;
4610 			}
4611 			if (pollinfo.revents & POLLNVAL) {
4612 				pcap_snprintf(handle->errbuf,
4613 					PCAP_ERRBUF_SIZE,
4614 					"Invalid polling request on packet socket");
4615 				return PCAP_ERROR;
4616 			}
4617 		}
4618 		/* check for break loop condition on interrupted syscall*/
4619 		if (handle->break_loop) {
4620 			handle->break_loop = 0;
4621 			return PCAP_ERROR_BREAK;
4622 		}
4623 	} while (ret < 0);
4624 	return 0;
4625 }
4626 
4627 /* handle a single memory mapped packet */
4628 static int pcap_handle_packet_mmap(
4629 		pcap_t *handle,
4630 		pcap_handler callback,
4631 		u_char *user,
4632 		unsigned char *frame,
4633 		unsigned int tp_len,
4634 		unsigned int tp_mac,
4635 		unsigned int tp_snaplen,
4636 		unsigned int tp_sec,
4637 		unsigned int tp_usec,
4638 		int tp_vlan_tci_valid,
4639 		__u16 tp_vlan_tci,
4640 		__u16 tp_vlan_tpid)
4641 {
4642 	struct pcap_linux *handlep = handle->priv;
4643 	unsigned char *bp;
4644 	struct sockaddr_ll *sll;
4645 	struct pcap_pkthdr pcaphdr;
4646 
4647 	/* perform sanity check on internal offset. */
4648 	if (tp_mac + tp_snaplen > handle->bufsize) {
4649 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4650 			"corrupted frame on kernel ring mac "
4651 			"offset %u + caplen %u > frame len %d",
4652 			tp_mac, tp_snaplen, handle->bufsize);
4653 		return -1;
4654 	}
4655 
4656 	/* run filter on received packet
4657 	 * If the kernel filtering is enabled we need to run the
4658 	 * filter until all the frames present into the ring
4659 	 * at filter creation time are processed.
4660 	 * In this case, blocks_to_filter_in_userland is used
4661 	 * as a counter for the packet we need to filter.
4662 	 * Note: alternatively it could be possible to stop applying
4663 	 * the filter when the ring became empty, but it can possibly
4664 	 * happen a lot later... */
4665 	bp = frame + tp_mac;
4666 
4667 	/* if required build in place the sll header*/
4668 	sll = (void *)frame + TPACKET_ALIGN(handlep->tp_hdrlen);
4669 	if (handlep->cooked) {
4670 		struct sll_header *hdrp;
4671 
4672 		/*
4673 		 * The kernel should have left us with enough
4674 		 * space for an sll header; back up the packet
4675 		 * data pointer into that space, as that'll be
4676 		 * the beginning of the packet we pass to the
4677 		 * callback.
4678 		 */
4679 		bp -= SLL_HDR_LEN;
4680 
4681 		/*
4682 		 * Let's make sure that's past the end of
4683 		 * the tpacket header, i.e. >=
4684 		 * ((u_char *)thdr + TPACKET_HDRLEN), so we
4685 		 * don't step on the header when we construct
4686 		 * the sll header.
4687 		 */
4688 		if (bp < (u_char *)frame +
4689 				   TPACKET_ALIGN(handlep->tp_hdrlen) +
4690 				   sizeof(struct sockaddr_ll)) {
4691 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4692 				"cooked-mode frame doesn't have room for sll header");
4693 			return -1;
4694 		}
4695 
4696 		/*
4697 		 * OK, that worked; construct the sll header.
4698 		 */
4699 		hdrp = (struct sll_header *)bp;
4700 		hdrp->sll_pkttype = map_packet_type_to_sll_type(
4701 						sll->sll_pkttype);
4702 		hdrp->sll_hatype = htons(sll->sll_hatype);
4703 		hdrp->sll_halen = htons(sll->sll_halen);
4704 		memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
4705 		hdrp->sll_protocol = sll->sll_protocol;
4706 	}
4707 
4708 	if (handlep->filter_in_userland && handle->fcode.bf_insns) {
4709 		struct bpf_aux_data aux_data;
4710 
4711 		aux_data.vlan_tag = tp_vlan_tci & 0x0fff;
4712 		aux_data.vlan_tag_present = tp_vlan_tci_valid;
4713 
4714 		if (bpf_filter_with_aux_data(handle->fcode.bf_insns, bp,
4715 		    tp_len, tp_snaplen, &aux_data) == 0)
4716 			return 0;
4717 	}
4718 
4719 	if (!linux_check_direction(handle, sll))
4720 		return 0;
4721 
4722 	/* get required packet info from ring header */
4723 	pcaphdr.ts.tv_sec = tp_sec;
4724 	pcaphdr.ts.tv_usec = tp_usec;
4725 	pcaphdr.caplen = tp_snaplen;
4726 	pcaphdr.len = tp_len;
4727 
4728 	/* if required build in place the sll header*/
4729 	if (handlep->cooked) {
4730 		/* update packet len */
4731 		pcaphdr.caplen += SLL_HDR_LEN;
4732 		pcaphdr.len += SLL_HDR_LEN;
4733 	}
4734 
4735 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
4736 	if (tp_vlan_tci_valid &&
4737 		handlep->vlan_offset != -1 &&
4738 		tp_snaplen >= (unsigned int) handlep->vlan_offset)
4739 	{
4740 		struct vlan_tag *tag;
4741 
4742 		/*
4743 		 * Move everything in the header, except the type field,
4744 		 * down VLAN_TAG_LEN bytes, to allow us to insert the
4745 		 * VLAN tag between that stuff and the type field.
4746 		 */
4747 		bp -= VLAN_TAG_LEN;
4748 		memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
4749 
4750 		/*
4751 		 * Now insert the tag.
4752 		 */
4753 		tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
4754 		tag->vlan_tpid = htons(tp_vlan_tpid);
4755 		tag->vlan_tci = htons(tp_vlan_tci);
4756 
4757 		/*
4758 		 * Add the tag to the packet lengths.
4759 		 */
4760 		pcaphdr.caplen += VLAN_TAG_LEN;
4761 		pcaphdr.len += VLAN_TAG_LEN;
4762 	}
4763 #endif
4764 
4765 	/*
4766 	 * The only way to tell the kernel to cut off the
4767 	 * packet at a snapshot length is with a filter program;
4768 	 * if there's no filter program, the kernel won't cut
4769 	 * the packet off.
4770 	 *
4771 	 * Trim the snapshot length to be no longer than the
4772 	 * specified snapshot length.
4773 	 */
4774 	if (pcaphdr.caplen > (bpf_u_int32)handle->snapshot)
4775 		pcaphdr.caplen = handle->snapshot;
4776 
4777 	/* pass the packet to the user */
4778 	callback(user, &pcaphdr, bp);
4779 
4780 	return 1;
4781 }
4782 
4783 static int
4784 pcap_read_linux_mmap_v1(pcap_t *handle, int max_packets, pcap_handler callback,
4785 		u_char *user)
4786 {
4787 	struct pcap_linux *handlep = handle->priv;
4788 	union thdr h;
4789 	int pkts = 0;
4790 	int ret;
4791 
4792 	/* wait for frames availability.*/
4793 	h.raw = RING_GET_CURRENT_FRAME(handle);
4794 	if (h.h1->tp_status == TP_STATUS_KERNEL) {
4795 		/*
4796 		 * The current frame is owned by the kernel; wait for
4797 		 * a frame to be handed to us.
4798 		 */
4799 		ret = pcap_wait_for_frames_mmap(handle);
4800 		if (ret) {
4801 			return ret;
4802 		}
4803 	}
4804 
4805 	/* non-positive values of max_packets are used to require all
4806 	 * packets currently available in the ring */
4807 	while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
4808 		/*
4809 		 * Get the current ring buffer frame, and break if
4810 		 * it's still owned by the kernel.
4811 		 */
4812 		h.raw = RING_GET_CURRENT_FRAME(handle);
4813 		if (h.h1->tp_status == TP_STATUS_KERNEL)
4814 			break;
4815 
4816 		ret = pcap_handle_packet_mmap(
4817 				handle,
4818 				callback,
4819 				user,
4820 				h.raw,
4821 				h.h1->tp_len,
4822 				h.h1->tp_mac,
4823 				h.h1->tp_snaplen,
4824 				h.h1->tp_sec,
4825 				h.h1->tp_usec,
4826 				0,
4827 				0,
4828 				0);
4829 		if (ret == 1) {
4830 			pkts++;
4831 			handlep->packets_read++;
4832 		} else if (ret < 0) {
4833 			return ret;
4834 		}
4835 
4836 		/*
4837 		 * Hand this block back to the kernel, and, if we're
4838 		 * counting blocks that need to be filtered in userland
4839 		 * after having been filtered by the kernel, count
4840 		 * the one we've just processed.
4841 		 */
4842 		h.h1->tp_status = TP_STATUS_KERNEL;
4843 		if (handlep->blocks_to_filter_in_userland > 0) {
4844 			handlep->blocks_to_filter_in_userland--;
4845 			if (handlep->blocks_to_filter_in_userland == 0) {
4846 				/*
4847 				 * No more blocks need to be filtered
4848 				 * in userland.
4849 				 */
4850 				handlep->filter_in_userland = 0;
4851 			}
4852 		}
4853 
4854 		/* next block */
4855 		if (++handle->offset >= handle->cc)
4856 			handle->offset = 0;
4857 
4858 		/* check for break loop condition*/
4859 		if (handle->break_loop) {
4860 			handle->break_loop = 0;
4861 			return PCAP_ERROR_BREAK;
4862 		}
4863 	}
4864 	return pkts;
4865 }
4866 
4867 static int
4868 pcap_read_linux_mmap_v1_64(pcap_t *handle, int max_packets, pcap_handler callback,
4869 		u_char *user)
4870 {
4871 	struct pcap_linux *handlep = handle->priv;
4872 	union thdr h;
4873 	int pkts = 0;
4874 	int ret;
4875 
4876 	/* wait for frames availability.*/
4877 	h.raw = RING_GET_CURRENT_FRAME(handle);
4878 	if (h.h1_64->tp_status == TP_STATUS_KERNEL) {
4879 		/*
4880 		 * The current frame is owned by the kernel; wait for
4881 		 * a frame to be handed to us.
4882 		 */
4883 		ret = pcap_wait_for_frames_mmap(handle);
4884 		if (ret) {
4885 			return ret;
4886 		}
4887 	}
4888 
4889 	/* non-positive values of max_packets are used to require all
4890 	 * packets currently available in the ring */
4891 	while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
4892 		/*
4893 		 * Get the current ring buffer frame, and break if
4894 		 * it's still owned by the kernel.
4895 		 */
4896 		h.raw = RING_GET_CURRENT_FRAME(handle);
4897 		if (h.h1_64->tp_status == TP_STATUS_KERNEL)
4898 			break;
4899 
4900 		ret = pcap_handle_packet_mmap(
4901 				handle,
4902 				callback,
4903 				user,
4904 				h.raw,
4905 				h.h1_64->tp_len,
4906 				h.h1_64->tp_mac,
4907 				h.h1_64->tp_snaplen,
4908 				h.h1_64->tp_sec,
4909 				h.h1_64->tp_usec,
4910 				0,
4911 				0,
4912 				0);
4913 		if (ret == 1) {
4914 			pkts++;
4915 			handlep->packets_read++;
4916 		} else if (ret < 0) {
4917 			return ret;
4918 		}
4919 
4920 		/*
4921 		 * Hand this block back to the kernel, and, if we're
4922 		 * counting blocks that need to be filtered in userland
4923 		 * after having been filtered by the kernel, count
4924 		 * the one we've just processed.
4925 		 */
4926 		h.h1_64->tp_status = TP_STATUS_KERNEL;
4927 		if (handlep->blocks_to_filter_in_userland > 0) {
4928 			handlep->blocks_to_filter_in_userland--;
4929 			if (handlep->blocks_to_filter_in_userland == 0) {
4930 				/*
4931 				 * No more blocks need to be filtered
4932 				 * in userland.
4933 				 */
4934 				handlep->filter_in_userland = 0;
4935 			}
4936 		}
4937 
4938 		/* next block */
4939 		if (++handle->offset >= handle->cc)
4940 			handle->offset = 0;
4941 
4942 		/* check for break loop condition*/
4943 		if (handle->break_loop) {
4944 			handle->break_loop = 0;
4945 			return PCAP_ERROR_BREAK;
4946 		}
4947 	}
4948 	return pkts;
4949 }
4950 
4951 #ifdef HAVE_TPACKET2
4952 static int
4953 pcap_read_linux_mmap_v2(pcap_t *handle, int max_packets, pcap_handler callback,
4954 		u_char *user)
4955 {
4956 	struct pcap_linux *handlep = handle->priv;
4957 	union thdr h;
4958 	int pkts = 0;
4959 	int ret;
4960 
4961 	/* wait for frames availability.*/
4962 	h.raw = RING_GET_CURRENT_FRAME(handle);
4963 	if (h.h2->tp_status == TP_STATUS_KERNEL) {
4964 		/*
4965 		 * The current frame is owned by the kernel; wait for
4966 		 * a frame to be handed to us.
4967 		 */
4968 		ret = pcap_wait_for_frames_mmap(handle);
4969 		if (ret) {
4970 			return ret;
4971 		}
4972 	}
4973 
4974 	/* non-positive values of max_packets are used to require all
4975 	 * packets currently available in the ring */
4976 	while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
4977 		/*
4978 		 * Get the current ring buffer frame, and break if
4979 		 * it's still owned by the kernel.
4980 		 */
4981 		h.raw = RING_GET_CURRENT_FRAME(handle);
4982 		if (h.h2->tp_status == TP_STATUS_KERNEL)
4983 			break;
4984 
4985 		ret = pcap_handle_packet_mmap(
4986 				handle,
4987 				callback,
4988 				user,
4989 				h.raw,
4990 				h.h2->tp_len,
4991 				h.h2->tp_mac,
4992 				h.h2->tp_snaplen,
4993 				h.h2->tp_sec,
4994 				handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? h.h2->tp_nsec : h.h2->tp_nsec / 1000,
4995 #if defined(TP_STATUS_VLAN_VALID)
4996 				(h.h2->tp_vlan_tci || (h.h2->tp_status & TP_STATUS_VLAN_VALID)),
4997 #else
4998 				h.h2->tp_vlan_tci != 0,
4999 #endif
5000 				h.h2->tp_vlan_tci,
5001 				VLAN_TPID(h.h2, h.h2));
5002 		if (ret == 1) {
5003 			pkts++;
5004 			handlep->packets_read++;
5005 		} else if (ret < 0) {
5006 			return ret;
5007 		}
5008 
5009 		/*
5010 		 * Hand this block back to the kernel, and, if we're
5011 		 * counting blocks that need to be filtered in userland
5012 		 * after having been filtered by the kernel, count
5013 		 * the one we've just processed.
5014 		 */
5015 		h.h2->tp_status = TP_STATUS_KERNEL;
5016 		if (handlep->blocks_to_filter_in_userland > 0) {
5017 			handlep->blocks_to_filter_in_userland--;
5018 			if (handlep->blocks_to_filter_in_userland == 0) {
5019 				/*
5020 				 * No more blocks need to be filtered
5021 				 * in userland.
5022 				 */
5023 				handlep->filter_in_userland = 0;
5024 			}
5025 		}
5026 
5027 		/* next block */
5028 		if (++handle->offset >= handle->cc)
5029 			handle->offset = 0;
5030 
5031 		/* check for break loop condition*/
5032 		if (handle->break_loop) {
5033 			handle->break_loop = 0;
5034 			return PCAP_ERROR_BREAK;
5035 		}
5036 	}
5037 	return pkts;
5038 }
5039 #endif /* HAVE_TPACKET2 */
5040 
5041 #ifdef HAVE_TPACKET3
5042 static int
5043 pcap_read_linux_mmap_v3(pcap_t *handle, int max_packets, pcap_handler callback,
5044 		u_char *user)
5045 {
5046 	struct pcap_linux *handlep = handle->priv;
5047 	union thdr h;
5048 	int pkts = 0;
5049 	int ret;
5050 
5051 again:
5052 	if (handlep->current_packet == NULL) {
5053 		/* wait for frames availability.*/
5054 		h.raw = RING_GET_CURRENT_FRAME(handle);
5055 		if (h.h3->hdr.bh1.block_status == TP_STATUS_KERNEL) {
5056 			/*
5057 			 * The current frame is owned by the kernel; wait
5058 			 * for a frame to be handed to us.
5059 			 */
5060 			ret = pcap_wait_for_frames_mmap(handle);
5061 			if (ret) {
5062 				return ret;
5063 			}
5064 		}
5065 	}
5066 	h.raw = RING_GET_CURRENT_FRAME(handle);
5067 	if (h.h3->hdr.bh1.block_status == TP_STATUS_KERNEL) {
5068 		if (pkts == 0 && handlep->timeout == 0) {
5069 			/* Block until we see a packet. */
5070 			goto again;
5071 		}
5072 		return pkts;
5073 	}
5074 
5075 	/* non-positive values of max_packets are used to require all
5076 	 * packets currently available in the ring */
5077 	while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
5078 		int packets_to_read;
5079 
5080 		if (handlep->current_packet == NULL) {
5081 			h.raw = RING_GET_CURRENT_FRAME(handle);
5082 			if (h.h3->hdr.bh1.block_status == TP_STATUS_KERNEL)
5083 				break;
5084 
5085 			handlep->current_packet = h.raw + h.h3->hdr.bh1.offset_to_first_pkt;
5086 			handlep->packets_left = h.h3->hdr.bh1.num_pkts;
5087 		}
5088 		packets_to_read = handlep->packets_left;
5089 
5090 		if (!PACKET_COUNT_IS_UNLIMITED(max_packets) &&
5091 		    packets_to_read > (max_packets - pkts)) {
5092 			/*
5093 			 * We've been given a maximum number of packets
5094 			 * to process, and there are more packets in
5095 			 * this buffer than that.  Only process enough
5096 			 * of them to get us up to that maximum.
5097 			 */
5098 			packets_to_read = max_packets - pkts;
5099 		}
5100 
5101 		while (packets_to_read-- && !handle->break_loop) {
5102 			struct tpacket3_hdr* tp3_hdr = (struct tpacket3_hdr*) handlep->current_packet;
5103 			ret = pcap_handle_packet_mmap(
5104 					handle,
5105 					callback,
5106 					user,
5107 					handlep->current_packet,
5108 					tp3_hdr->tp_len,
5109 					tp3_hdr->tp_mac,
5110 					tp3_hdr->tp_snaplen,
5111 					tp3_hdr->tp_sec,
5112 					handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? tp3_hdr->tp_nsec : tp3_hdr->tp_nsec / 1000,
5113 #if defined(TP_STATUS_VLAN_VALID)
5114 					(tp3_hdr->hv1.tp_vlan_tci || (tp3_hdr->tp_status & TP_STATUS_VLAN_VALID)),
5115 #else
5116 					tp3_hdr->hv1.tp_vlan_tci != 0,
5117 #endif
5118 					tp3_hdr->hv1.tp_vlan_tci,
5119 					VLAN_TPID(tp3_hdr, &tp3_hdr->hv1));
5120 			if (ret == 1) {
5121 				pkts++;
5122 				handlep->packets_read++;
5123 			} else if (ret < 0) {
5124 				handlep->current_packet = NULL;
5125 				return ret;
5126 			}
5127 			handlep->current_packet += tp3_hdr->tp_next_offset;
5128 			handlep->packets_left--;
5129 		}
5130 
5131 		if (handlep->packets_left <= 0) {
5132 			/*
5133 			 * Hand this block back to the kernel, and, if
5134 			 * we're counting blocks that need to be
5135 			 * filtered in userland after having been
5136 			 * filtered by the kernel, count the one we've
5137 			 * just processed.
5138 			 */
5139 			h.h3->hdr.bh1.block_status = TP_STATUS_KERNEL;
5140 			if (handlep->blocks_to_filter_in_userland > 0) {
5141 				handlep->blocks_to_filter_in_userland--;
5142 				if (handlep->blocks_to_filter_in_userland == 0) {
5143 					/*
5144 					 * No more blocks need to be filtered
5145 					 * in userland.
5146 					 */
5147 					handlep->filter_in_userland = 0;
5148 				}
5149 			}
5150 
5151 			/* next block */
5152 			if (++handle->offset >= handle->cc)
5153 				handle->offset = 0;
5154 
5155 			handlep->current_packet = NULL;
5156 		}
5157 
5158 		/* check for break loop condition*/
5159 		if (handle->break_loop) {
5160 			handle->break_loop = 0;
5161 			return PCAP_ERROR_BREAK;
5162 		}
5163 	}
5164 	if (pkts == 0 && handlep->timeout == 0) {
5165 		/* Block until we see a packet. */
5166 		goto again;
5167 	}
5168 	return pkts;
5169 }
5170 #endif /* HAVE_TPACKET3 */
5171 
5172 static int
5173 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
5174 {
5175 	struct pcap_linux *handlep = handle->priv;
5176 	int n, offset;
5177 	int ret;
5178 
5179 	/*
5180 	 * Don't rewrite "ret" instructions; we don't need to, as
5181 	 * we're not reading packets with recvmsg(), and we don't
5182 	 * want to, as, by not rewriting them, the kernel can avoid
5183 	 * copying extra data.
5184 	 */
5185 	ret = pcap_setfilter_linux_common(handle, filter, 1);
5186 	if (ret < 0)
5187 		return ret;
5188 
5189 	/*
5190 	 * If we're filtering in userland, there's nothing to do;
5191 	 * the new filter will be used for the next packet.
5192 	 */
5193 	if (handlep->filter_in_userland)
5194 		return ret;
5195 
5196 	/*
5197 	 * We're filtering in the kernel; the packets present in
5198 	 * all blocks currently in the ring were already filtered
5199 	 * by the old filter, and so will need to be filtered in
5200 	 * userland by the new filter.
5201 	 *
5202 	 * Get an upper bound for the number of such blocks; first,
5203 	 * walk the ring backward and count the free blocks.
5204 	 */
5205 	offset = handle->offset;
5206 	if (--offset < 0)
5207 		offset = handle->cc - 1;
5208 	for (n=0; n < handle->cc; ++n) {
5209 		if (--offset < 0)
5210 			offset = handle->cc - 1;
5211 		if (pcap_get_ring_frame_status(handle, offset) != TP_STATUS_KERNEL)
5212 			break;
5213 	}
5214 
5215 	/*
5216 	 * If we found free blocks, decrement the count of free
5217 	 * blocks by 1, just in case we lost a race with another
5218 	 * thread of control that was adding a packet while
5219 	 * we were counting and that had run the filter before
5220 	 * we changed it.
5221 	 *
5222 	 * XXX - could there be more than one block added in
5223 	 * this fashion?
5224 	 *
5225 	 * XXX - is there a way to avoid that race, e.g. somehow
5226 	 * wait for all packets that passed the old filter to
5227 	 * be added to the ring?
5228 	 */
5229 	if (n != 0)
5230 		n--;
5231 
5232 	/*
5233 	 * Set the count of blocks worth of packets to filter
5234 	 * in userland to the total number of blocks in the
5235 	 * ring minus the number of free blocks we found, and
5236 	 * turn on userland filtering.  (The count of blocks
5237 	 * worth of packets to filter in userland is guaranteed
5238 	 * not to be zero - n, above, couldn't be set to a
5239 	 * value > handle->cc, and if it were equal to
5240 	 * handle->cc, it wouldn't be zero, and thus would
5241 	 * be decremented to handle->cc - 1.)
5242 	 */
5243 	handlep->blocks_to_filter_in_userland = handle->cc - n;
5244 	handlep->filter_in_userland = 1;
5245 	return ret;
5246 }
5247 
5248 #endif /* HAVE_PACKET_RING */
5249 
5250 
5251 #ifdef HAVE_PF_PACKET_SOCKETS
5252 /*
5253  *  Return the index of the given device name. Fill ebuf and return
5254  *  -1 on failure.
5255  */
5256 static int
5257 iface_get_id(int fd, const char *device, char *ebuf)
5258 {
5259 	struct ifreq	ifr;
5260 
5261 	memset(&ifr, 0, sizeof(ifr));
5262 	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5263 
5264 	if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
5265 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
5266 			 "SIOCGIFINDEX: %s", pcap_strerror(errno));
5267 		return -1;
5268 	}
5269 
5270 	return ifr.ifr_ifindex;
5271 }
5272 
5273 /*
5274  *  Bind the socket associated with FD to the given device.
5275  *  Return 1 on success, 0 if we should try a SOCK_PACKET socket,
5276  *  or a PCAP_ERROR_ value on a hard error.
5277  */
5278 static int
5279 iface_bind(int fd, int ifindex, char *ebuf)
5280 {
5281 	struct sockaddr_ll	sll;
5282 	int			err;
5283 	socklen_t		errlen = sizeof(err);
5284 
5285 	memset(&sll, 0, sizeof(sll));
5286 	sll.sll_family		= AF_PACKET;
5287 	sll.sll_ifindex		= ifindex;
5288 	sll.sll_protocol	= htons(ETH_P_ALL);
5289 
5290 	if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
5291 		if (errno == ENETDOWN) {
5292 			/*
5293 			 * Return a "network down" indication, so that
5294 			 * the application can report that rather than
5295 			 * saying we had a mysterious failure and
5296 			 * suggest that they report a problem to the
5297 			 * libpcap developers.
5298 			 */
5299 			return PCAP_ERROR_IFACE_NOT_UP;
5300 		} else {
5301 			pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
5302 				 "bind: %s", pcap_strerror(errno));
5303 			return PCAP_ERROR;
5304 		}
5305 	}
5306 
5307 	/* Any pending errors, e.g., network is down? */
5308 
5309 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
5310 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
5311 			"getsockopt: %s", pcap_strerror(errno));
5312 		return 0;
5313 	}
5314 
5315 	if (err == ENETDOWN) {
5316 		/*
5317 		 * Return a "network down" indication, so that
5318 		 * the application can report that rather than
5319 		 * saying we had a mysterious failure and
5320 		 * suggest that they report a problem to the
5321 		 * libpcap developers.
5322 		 */
5323 		return PCAP_ERROR_IFACE_NOT_UP;
5324 	} else if (err > 0) {
5325 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
5326 			"bind: %s", pcap_strerror(err));
5327 		return 0;
5328 	}
5329 
5330 	return 1;
5331 }
5332 
5333 #ifdef IW_MODE_MONITOR
5334 /*
5335  * Check whether the device supports the Wireless Extensions.
5336  * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE
5337  * if the device doesn't even exist.
5338  */
5339 static int
5340 has_wext(int sock_fd, const char *device, char *ebuf)
5341 {
5342 	struct iwreq ireq;
5343 
5344 	if (is_bonding_device(sock_fd, device))
5345 		return 0;	/* bonding device, so don't even try */
5346 
5347 	strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5348 	    sizeof ireq.ifr_ifrn.ifrn_name);
5349 	if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
5350 		return 1;	/* yes */
5351 	pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
5352 	    "%s: SIOCGIWNAME: %s", device, pcap_strerror(errno));
5353 	if (errno == ENODEV)
5354 		return PCAP_ERROR_NO_SUCH_DEVICE;
5355 	return 0;
5356 }
5357 
5358 /*
5359  * Per me si va ne la citta dolente,
5360  * Per me si va ne l'etterno dolore,
5361  *	...
5362  * Lasciate ogne speranza, voi ch'intrate.
5363  *
5364  * XXX - airmon-ng does special stuff with the Orinoco driver and the
5365  * wlan-ng driver.
5366  */
5367 typedef enum {
5368 	MONITOR_WEXT,
5369 	MONITOR_HOSTAP,
5370 	MONITOR_PRISM,
5371 	MONITOR_PRISM54,
5372 	MONITOR_ACX100,
5373 	MONITOR_RT2500,
5374 	MONITOR_RT2570,
5375 	MONITOR_RT73,
5376 	MONITOR_RTL8XXX
5377 } monitor_type;
5378 
5379 /*
5380  * Use the Wireless Extensions, if we have them, to try to turn monitor mode
5381  * on if it's not already on.
5382  *
5383  * Returns 1 on success, 0 if we don't support the Wireless Extensions
5384  * on this device, or a PCAP_ERROR_ value if we do support them but
5385  * we weren't able to turn monitor mode on.
5386  */
5387 static int
5388 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
5389 {
5390 	/*
5391 	 * XXX - at least some adapters require non-Wireless Extensions
5392 	 * mechanisms to turn monitor mode on.
5393 	 *
5394 	 * Atheros cards might require that a separate "monitor virtual access
5395 	 * point" be created, with later versions of the madwifi driver.
5396 	 * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode
5397 	 * monitor -bssid", which apparently spits out a line "athN"
5398 	 * where "athN" is the monitor mode device.  To leave monitor
5399 	 * mode, it destroys the monitor mode device.
5400 	 *
5401 	 * Some Intel Centrino adapters might require private ioctls to get
5402 	 * radio headers; the ipw2200 and ipw3945 drivers allow you to
5403 	 * configure a separate "rtapN" interface to capture in monitor
5404 	 * mode without preventing the adapter from operating normally.
5405 	 * (airmon-ng doesn't appear to use that, though.)
5406 	 *
5407 	 * It would be Truly Wonderful if mac80211 and nl80211 cleaned this
5408 	 * up, and if all drivers were converted to mac80211 drivers.
5409 	 *
5410 	 * If interface {if} is a mac80211 driver, the file
5411 	 * /sys/class/net/{if}/phy80211 is a symlink to
5412 	 * /sys/class/ieee80211/{phydev}, for some {phydev}.
5413 	 *
5414 	 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
5415 	 * least, has a "wmaster0" device and a "wlan0" device; the
5416 	 * latter is the one with the IP address.  Both show up in
5417 	 * "tcpdump -D" output.  Capturing on the wmaster0 device
5418 	 * captures with 802.11 headers.
5419 	 *
5420 	 * airmon-ng searches through /sys/class/net for devices named
5421 	 * monN, starting with mon0; as soon as one *doesn't* exist,
5422 	 * it chooses that as the monitor device name.  If the "iw"
5423 	 * command exists, it does "iw dev {if} interface add {monif}
5424 	 * type monitor", where {monif} is the monitor device.  It
5425 	 * then (sigh) sleeps .1 second, and then configures the
5426 	 * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
5427 	 * is a file, it writes {mondev}, without a newline, to that file,
5428 	 * and again (sigh) sleeps .1 second, and then iwconfig's that
5429 	 * device into monitor mode and configures it up.  Otherwise,
5430 	 * you can't do monitor mode.
5431 	 *
5432 	 * All these devices are "glued" together by having the
5433 	 * /sys/class/net/{device}/phy80211 links pointing to the same
5434 	 * place, so, given a wmaster, wlan, or mon device, you can
5435 	 * find the other devices by looking for devices with
5436 	 * the same phy80211 link.
5437 	 *
5438 	 * To turn monitor mode off, delete the monitor interface,
5439 	 * either with "iw dev {monif} interface del" or by sending
5440 	 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
5441 	 *
5442 	 * Note: if you try to create a monitor device named "monN", and
5443 	 * there's already a "monN" device, it fails, as least with
5444 	 * the netlink interface (which is what iw uses), with a return
5445 	 * value of -ENFILE.  (Return values are negative errnos.)  We
5446 	 * could probably use that to find an unused device.
5447 	 */
5448 	struct pcap_linux *handlep = handle->priv;
5449 	int err;
5450 	struct iwreq ireq;
5451 	struct iw_priv_args *priv;
5452 	monitor_type montype;
5453 	int i;
5454 	__u32 cmd;
5455 	struct ifreq ifr;
5456 	int oldflags;
5457 	int args[2];
5458 	int channel;
5459 
5460 	/*
5461 	 * Does this device *support* the Wireless Extensions?
5462 	 */
5463 	err = has_wext(sock_fd, device, handle->errbuf);
5464 	if (err <= 0)
5465 		return err;	/* either it doesn't or the device doesn't even exist */
5466 	/*
5467 	 * Start out assuming we have no private extensions to control
5468 	 * radio metadata.
5469 	 */
5470 	montype = MONITOR_WEXT;
5471 	cmd = 0;
5472 
5473 	/*
5474 	 * Try to get all the Wireless Extensions private ioctls
5475 	 * supported by this device.
5476 	 *
5477 	 * First, get the size of the buffer we need, by supplying no
5478 	 * buffer and a length of 0.  If the device supports private
5479 	 * ioctls, it should return E2BIG, with ireq.u.data.length set
5480 	 * to the length we need.  If it doesn't support them, it should
5481 	 * return EOPNOTSUPP.
5482 	 */
5483 	memset(&ireq, 0, sizeof ireq);
5484 	strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5485 	    sizeof ireq.ifr_ifrn.ifrn_name);
5486 	ireq.u.data.pointer = (void *)args;
5487 	ireq.u.data.length = 0;
5488 	ireq.u.data.flags = 0;
5489 	if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
5490 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5491 		    "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!",
5492 		    device);
5493 		return PCAP_ERROR;
5494 	}
5495 	if (errno != EOPNOTSUPP) {
5496 		/*
5497 		 * OK, it's not as if there are no private ioctls.
5498 		 */
5499 		if (errno != E2BIG) {
5500 			/*
5501 			 * Failed.
5502 			 */
5503 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5504 			    "%s: SIOCGIWPRIV: %s", device,
5505 			    pcap_strerror(errno));
5506 			return PCAP_ERROR;
5507 		}
5508 
5509 		/*
5510 		 * OK, try to get the list of private ioctls.
5511 		 */
5512 		priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args));
5513 		if (priv == NULL) {
5514 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5515 			    "malloc: %s", pcap_strerror(errno));
5516 			return PCAP_ERROR;
5517 		}
5518 		ireq.u.data.pointer = (void *)priv;
5519 		if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) {
5520 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5521 			    "%s: SIOCGIWPRIV: %s", device,
5522 			    pcap_strerror(errno));
5523 			free(priv);
5524 			return PCAP_ERROR;
5525 		}
5526 
5527 		/*
5528 		 * Look for private ioctls to turn monitor mode on or, if
5529 		 * monitor mode is on, to set the header type.
5530 		 */
5531 		for (i = 0; i < ireq.u.data.length; i++) {
5532 			if (strcmp(priv[i].name, "monitor_type") == 0) {
5533 				/*
5534 				 * Hostap driver, use this one.
5535 				 * Set monitor mode first.
5536 				 * You can set it to 0 to get DLT_IEEE80211,
5537 				 * 1 to get DLT_PRISM, 2 to get
5538 				 * DLT_IEEE80211_RADIO_AVS, and, with more
5539 				 * recent versions of the driver, 3 to get
5540 				 * DLT_IEEE80211_RADIO.
5541 				 */
5542 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5543 					break;
5544 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5545 					break;
5546 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5547 					break;
5548 				montype = MONITOR_HOSTAP;
5549 				cmd = priv[i].cmd;
5550 				break;
5551 			}
5552 			if (strcmp(priv[i].name, "set_prismhdr") == 0) {
5553 				/*
5554 				 * Prism54 driver, use this one.
5555 				 * Set monitor mode first.
5556 				 * You can set it to 2 to get DLT_IEEE80211
5557 				 * or 3 or get DLT_PRISM.
5558 				 */
5559 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5560 					break;
5561 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5562 					break;
5563 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5564 					break;
5565 				montype = MONITOR_PRISM54;
5566 				cmd = priv[i].cmd;
5567 				break;
5568 			}
5569 			if (strcmp(priv[i].name, "forceprismheader") == 0) {
5570 				/*
5571 				 * RT2570 driver, use this one.
5572 				 * Do this after turning monitor mode on.
5573 				 * You can set it to 1 to get DLT_PRISM or 2
5574 				 * to get DLT_IEEE80211.
5575 				 */
5576 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5577 					break;
5578 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5579 					break;
5580 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5581 					break;
5582 				montype = MONITOR_RT2570;
5583 				cmd = priv[i].cmd;
5584 				break;
5585 			}
5586 			if (strcmp(priv[i].name, "forceprism") == 0) {
5587 				/*
5588 				 * RT73 driver, use this one.
5589 				 * Do this after turning monitor mode on.
5590 				 * Its argument is a *string*; you can
5591 				 * set it to "1" to get DLT_PRISM or "2"
5592 				 * to get DLT_IEEE80211.
5593 				 */
5594 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR)
5595 					break;
5596 				if (priv[i].set_args & IW_PRIV_SIZE_FIXED)
5597 					break;
5598 				montype = MONITOR_RT73;
5599 				cmd = priv[i].cmd;
5600 				break;
5601 			}
5602 			if (strcmp(priv[i].name, "prismhdr") == 0) {
5603 				/*
5604 				 * One of the RTL8xxx drivers, use this one.
5605 				 * It can only be done after monitor mode
5606 				 * has been turned on.  You can set it to 1
5607 				 * to get DLT_PRISM or 0 to get DLT_IEEE80211.
5608 				 */
5609 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5610 					break;
5611 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5612 					break;
5613 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5614 					break;
5615 				montype = MONITOR_RTL8XXX;
5616 				cmd = priv[i].cmd;
5617 				break;
5618 			}
5619 			if (strcmp(priv[i].name, "rfmontx") == 0) {
5620 				/*
5621 				 * RT2500 or RT61 driver, use this one.
5622 				 * It has one one-byte parameter; set
5623 				 * u.data.length to 1 and u.data.pointer to
5624 				 * point to the parameter.
5625 				 * It doesn't itself turn monitor mode on.
5626 				 * You can set it to 1 to allow transmitting
5627 				 * in monitor mode(?) and get DLT_IEEE80211,
5628 				 * or set it to 0 to disallow transmitting in
5629 				 * monitor mode(?) and get DLT_PRISM.
5630 				 */
5631 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5632 					break;
5633 				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2)
5634 					break;
5635 				montype = MONITOR_RT2500;
5636 				cmd = priv[i].cmd;
5637 				break;
5638 			}
5639 			if (strcmp(priv[i].name, "monitor") == 0) {
5640 				/*
5641 				 * Either ACX100 or hostap, use this one.
5642 				 * It turns monitor mode on.
5643 				 * If it takes two arguments, it's ACX100;
5644 				 * the first argument is 1 for DLT_PRISM
5645 				 * or 2 for DLT_IEEE80211, and the second
5646 				 * argument is the channel on which to
5647 				 * run.  If it takes one argument, it's
5648 				 * HostAP, and the argument is 2 for
5649 				 * DLT_IEEE80211 and 3 for DLT_PRISM.
5650 				 *
5651 				 * If we see this, we don't quit, as this
5652 				 * might be a version of the hostap driver
5653 				 * that also supports "monitor_type".
5654 				 */
5655 				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5656 					break;
5657 				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5658 					break;
5659 				switch (priv[i].set_args & IW_PRIV_SIZE_MASK) {
5660 
5661 				case 1:
5662 					montype = MONITOR_PRISM;
5663 					cmd = priv[i].cmd;
5664 					break;
5665 
5666 				case 2:
5667 					montype = MONITOR_ACX100;
5668 					cmd = priv[i].cmd;
5669 					break;
5670 
5671 				default:
5672 					break;
5673 				}
5674 			}
5675 		}
5676 		free(priv);
5677 	}
5678 
5679 	/*
5680 	 * XXX - ipw3945?  islism?
5681 	 */
5682 
5683 	/*
5684 	 * Get the old mode.
5685 	 */
5686 	strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5687 	    sizeof ireq.ifr_ifrn.ifrn_name);
5688 	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) {
5689 		/*
5690 		 * We probably won't be able to set the mode, either.
5691 		 */
5692 		return PCAP_ERROR_RFMON_NOTSUP;
5693 	}
5694 
5695 	/*
5696 	 * Is it currently in monitor mode?
5697 	 */
5698 	if (ireq.u.mode == IW_MODE_MONITOR) {
5699 		/*
5700 		 * Yes.  Just leave things as they are.
5701 		 * We don't offer multiple link-layer types, as
5702 		 * changing the link-layer type out from under
5703 		 * somebody else capturing in monitor mode would
5704 		 * be considered rude.
5705 		 */
5706 		return 1;
5707 	}
5708 	/*
5709 	 * No.  We have to put the adapter into rfmon mode.
5710 	 */
5711 
5712 	/*
5713 	 * If we haven't already done so, arrange to have
5714 	 * "pcap_close_all()" called when we exit.
5715 	 */
5716 	if (!pcap_do_addexit(handle)) {
5717 		/*
5718 		 * "atexit()" failed; don't put the interface
5719 		 * in rfmon mode, just give up.
5720 		 */
5721 		return PCAP_ERROR_RFMON_NOTSUP;
5722 	}
5723 
5724 	/*
5725 	 * Save the old mode.
5726 	 */
5727 	handlep->oldmode = ireq.u.mode;
5728 
5729 	/*
5730 	 * Put the adapter in rfmon mode.  How we do this depends
5731 	 * on whether we have a special private ioctl or not.
5732 	 */
5733 	if (montype == MONITOR_PRISM) {
5734 		/*
5735 		 * We have the "monitor" private ioctl, but none of
5736 		 * the other private ioctls.  Use this, and select
5737 		 * the Prism header.
5738 		 *
5739 		 * If it fails, just fall back on SIOCSIWMODE.
5740 		 */
5741 		memset(&ireq, 0, sizeof ireq);
5742 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5743 		    sizeof ireq.ifr_ifrn.ifrn_name);
5744 		ireq.u.data.length = 1;	/* 1 argument */
5745 		args[0] = 3;	/* request Prism header */
5746 		memcpy(ireq.u.name, args, sizeof (int));
5747 		if (ioctl(sock_fd, cmd, &ireq) != -1) {
5748 			/*
5749 			 * Success.
5750 			 * Note that we have to put the old mode back
5751 			 * when we close the device.
5752 			 */
5753 			handlep->must_do_on_close |= MUST_CLEAR_RFMON;
5754 
5755 			/*
5756 			 * Add this to the list of pcaps to close
5757 			 * when we exit.
5758 			 */
5759 			pcap_add_to_pcaps_to_close(handle);
5760 
5761 			return 1;
5762 		}
5763 
5764 		/*
5765 		 * Failure.  Fall back on SIOCSIWMODE.
5766 		 */
5767 	}
5768 
5769 	/*
5770 	 * First, take the interface down if it's up; otherwise, we
5771 	 * might get EBUSY.
5772 	 */
5773 	memset(&ifr, 0, sizeof(ifr));
5774 	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5775 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
5776 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5777 		    "%s: Can't get flags: %s", device, strerror(errno));
5778 		return PCAP_ERROR;
5779 	}
5780 	oldflags = 0;
5781 	if (ifr.ifr_flags & IFF_UP) {
5782 		oldflags = ifr.ifr_flags;
5783 		ifr.ifr_flags &= ~IFF_UP;
5784 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5785 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5786 			    "%s: Can't set flags: %s", device, strerror(errno));
5787 			return PCAP_ERROR;
5788 		}
5789 	}
5790 
5791 	/*
5792 	 * Then turn monitor mode on.
5793 	 */
5794 	strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5795 	    sizeof ireq.ifr_ifrn.ifrn_name);
5796 	ireq.u.mode = IW_MODE_MONITOR;
5797 	if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) {
5798 		/*
5799 		 * Scientist, you've failed.
5800 		 * Bring the interface back up if we shut it down.
5801 		 */
5802 		ifr.ifr_flags = oldflags;
5803 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5804 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5805 			    "%s: Can't set flags: %s", device, strerror(errno));
5806 			return PCAP_ERROR;
5807 		}
5808 		return PCAP_ERROR_RFMON_NOTSUP;
5809 	}
5810 
5811 	/*
5812 	 * XXX - airmon-ng does "iwconfig {if} key off" after setting
5813 	 * monitor mode and setting the channel, and then does
5814 	 * "iwconfig up".
5815 	 */
5816 
5817 	/*
5818 	 * Now select the appropriate radio header.
5819 	 */
5820 	switch (montype) {
5821 
5822 	case MONITOR_WEXT:
5823 		/*
5824 		 * We don't have any private ioctl to set the header.
5825 		 */
5826 		break;
5827 
5828 	case MONITOR_HOSTAP:
5829 		/*
5830 		 * Try to select the radiotap header.
5831 		 */
5832 		memset(&ireq, 0, sizeof ireq);
5833 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5834 		    sizeof ireq.ifr_ifrn.ifrn_name);
5835 		args[0] = 3;	/* request radiotap header */
5836 		memcpy(ireq.u.name, args, sizeof (int));
5837 		if (ioctl(sock_fd, cmd, &ireq) != -1)
5838 			break;	/* success */
5839 
5840 		/*
5841 		 * That failed.  Try to select the AVS header.
5842 		 */
5843 		memset(&ireq, 0, sizeof ireq);
5844 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5845 		    sizeof ireq.ifr_ifrn.ifrn_name);
5846 		args[0] = 2;	/* request AVS header */
5847 		memcpy(ireq.u.name, args, sizeof (int));
5848 		if (ioctl(sock_fd, cmd, &ireq) != -1)
5849 			break;	/* success */
5850 
5851 		/*
5852 		 * That failed.  Try to select the Prism header.
5853 		 */
5854 		memset(&ireq, 0, sizeof ireq);
5855 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5856 		    sizeof ireq.ifr_ifrn.ifrn_name);
5857 		args[0] = 1;	/* request Prism header */
5858 		memcpy(ireq.u.name, args, sizeof (int));
5859 		ioctl(sock_fd, cmd, &ireq);
5860 		break;
5861 
5862 	case MONITOR_PRISM:
5863 		/*
5864 		 * The private ioctl failed.
5865 		 */
5866 		break;
5867 
5868 	case MONITOR_PRISM54:
5869 		/*
5870 		 * Select the Prism header.
5871 		 */
5872 		memset(&ireq, 0, sizeof ireq);
5873 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5874 		    sizeof ireq.ifr_ifrn.ifrn_name);
5875 		args[0] = 3;	/* request Prism header */
5876 		memcpy(ireq.u.name, args, sizeof (int));
5877 		ioctl(sock_fd, cmd, &ireq);
5878 		break;
5879 
5880 	case MONITOR_ACX100:
5881 		/*
5882 		 * Get the current channel.
5883 		 */
5884 		memset(&ireq, 0, sizeof ireq);
5885 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5886 		    sizeof ireq.ifr_ifrn.ifrn_name);
5887 		if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) {
5888 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5889 			    "%s: SIOCGIWFREQ: %s", device,
5890 			    pcap_strerror(errno));
5891 			return PCAP_ERROR;
5892 		}
5893 		channel = ireq.u.freq.m;
5894 
5895 		/*
5896 		 * Select the Prism header, and set the channel to the
5897 		 * current value.
5898 		 */
5899 		memset(&ireq, 0, sizeof ireq);
5900 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5901 		    sizeof ireq.ifr_ifrn.ifrn_name);
5902 		args[0] = 1;		/* request Prism header */
5903 		args[1] = channel;	/* set channel */
5904 		memcpy(ireq.u.name, args, 2*sizeof (int));
5905 		ioctl(sock_fd, cmd, &ireq);
5906 		break;
5907 
5908 	case MONITOR_RT2500:
5909 		/*
5910 		 * Disallow transmission - that turns on the
5911 		 * Prism header.
5912 		 */
5913 		memset(&ireq, 0, sizeof ireq);
5914 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5915 		    sizeof ireq.ifr_ifrn.ifrn_name);
5916 		args[0] = 0;	/* disallow transmitting */
5917 		memcpy(ireq.u.name, args, sizeof (int));
5918 		ioctl(sock_fd, cmd, &ireq);
5919 		break;
5920 
5921 	case MONITOR_RT2570:
5922 		/*
5923 		 * Force the Prism header.
5924 		 */
5925 		memset(&ireq, 0, sizeof ireq);
5926 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5927 		    sizeof ireq.ifr_ifrn.ifrn_name);
5928 		args[0] = 1;	/* request Prism header */
5929 		memcpy(ireq.u.name, args, sizeof (int));
5930 		ioctl(sock_fd, cmd, &ireq);
5931 		break;
5932 
5933 	case MONITOR_RT73:
5934 		/*
5935 		 * Force the Prism header.
5936 		 */
5937 		memset(&ireq, 0, sizeof ireq);
5938 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5939 		    sizeof ireq.ifr_ifrn.ifrn_name);
5940 		ireq.u.data.length = 1;	/* 1 argument */
5941 		ireq.u.data.pointer = "1";
5942 		ireq.u.data.flags = 0;
5943 		ioctl(sock_fd, cmd, &ireq);
5944 		break;
5945 
5946 	case MONITOR_RTL8XXX:
5947 		/*
5948 		 * Force the Prism header.
5949 		 */
5950 		memset(&ireq, 0, sizeof ireq);
5951 		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5952 		    sizeof ireq.ifr_ifrn.ifrn_name);
5953 		args[0] = 1;	/* request Prism header */
5954 		memcpy(ireq.u.name, args, sizeof (int));
5955 		ioctl(sock_fd, cmd, &ireq);
5956 		break;
5957 	}
5958 
5959 	/*
5960 	 * Now bring the interface back up if we brought it down.
5961 	 */
5962 	if (oldflags != 0) {
5963 		ifr.ifr_flags = oldflags;
5964 		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5965 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5966 			    "%s: Can't set flags: %s", device, strerror(errno));
5967 
5968 			/*
5969 			 * At least try to restore the old mode on the
5970 			 * interface.
5971 			 */
5972 			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
5973 				/*
5974 				 * Scientist, you've failed.
5975 				 */
5976 				fprintf(stderr,
5977 				    "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n"
5978 				    "Please adjust manually.\n",
5979 				    strerror(errno));
5980 			}
5981 			return PCAP_ERROR;
5982 		}
5983 	}
5984 
5985 	/*
5986 	 * Note that we have to put the old mode back when we
5987 	 * close the device.
5988 	 */
5989 	handlep->must_do_on_close |= MUST_CLEAR_RFMON;
5990 
5991 	/*
5992 	 * Add this to the list of pcaps to close when we exit.
5993 	 */
5994 	pcap_add_to_pcaps_to_close(handle);
5995 
5996 	return 1;
5997 }
5998 #endif /* IW_MODE_MONITOR */
5999 
6000 /*
6001  * Try various mechanisms to enter monitor mode.
6002  */
6003 static int
6004 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
6005 {
6006 #if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
6007 	int ret;
6008 #endif
6009 
6010 #ifdef HAVE_LIBNL
6011 	ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
6012 	if (ret < 0)
6013 		return ret;	/* error attempting to do so */
6014 	if (ret == 1)
6015 		return 1;	/* success */
6016 #endif /* HAVE_LIBNL */
6017 
6018 #ifdef IW_MODE_MONITOR
6019 	ret = enter_rfmon_mode_wext(handle, sock_fd, device);
6020 	if (ret < 0)
6021 		return ret;	/* error attempting to do so */
6022 	if (ret == 1)
6023 		return 1;	/* success */
6024 #endif /* IW_MODE_MONITOR */
6025 
6026 	/*
6027 	 * Either none of the mechanisms we know about work or none
6028 	 * of those mechanisms are available, so we can't do monitor
6029 	 * mode.
6030 	 */
6031 	return 0;
6032 }
6033 
6034 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
6035 /*
6036  * Map SOF_TIMESTAMPING_ values to PCAP_TSTAMP_ values.
6037  */
6038 static const struct {
6039 	int soft_timestamping_val;
6040 	int pcap_tstamp_val;
6041 } sof_ts_type_map[3] = {
6042 	{ SOF_TIMESTAMPING_SOFTWARE, PCAP_TSTAMP_HOST },
6043 	{ SOF_TIMESTAMPING_SYS_HARDWARE, PCAP_TSTAMP_ADAPTER },
6044 	{ SOF_TIMESTAMPING_RAW_HARDWARE, PCAP_TSTAMP_ADAPTER_UNSYNCED }
6045 };
6046 #define NUM_SOF_TIMESTAMPING_TYPES	(sizeof sof_ts_type_map / sizeof sof_ts_type_map[0])
6047 
6048 /*
6049  * Set the list of time stamping types to include all types.
6050  */
6051 static void
6052 iface_set_all_ts_types(pcap_t *handle)
6053 {
6054 	u_int i;
6055 
6056 	handle->tstamp_type_count = NUM_SOF_TIMESTAMPING_TYPES;
6057 	handle->tstamp_type_list = malloc(NUM_SOF_TIMESTAMPING_TYPES * sizeof(u_int));
6058 	for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++)
6059 		handle->tstamp_type_list[i] = sof_ts_type_map[i].pcap_tstamp_val;
6060 }
6061 
6062 #ifdef ETHTOOL_GET_TS_INFO
6063 /*
6064  * Get a list of time stamping capabilities.
6065  */
6066 static int
6067 iface_ethtool_get_ts_info(const char *device, pcap_t *handle, char *ebuf)
6068 {
6069 	int fd;
6070 	struct ifreq ifr;
6071 	struct ethtool_ts_info info;
6072 	int num_ts_types;
6073 	u_int i, j;
6074 
6075 	/*
6076 	 * This doesn't apply to the "any" device; you can't say "turn on
6077 	 * hardware time stamping for all devices that exist now and arrange
6078 	 * that it be turned on for any device that appears in the future",
6079 	 * and not all devices even necessarily *support* hardware time
6080 	 * stamping, so don't report any time stamp types.
6081 	 */
6082 	if (strcmp(device, "any") == 0) {
6083 		handle->tstamp_type_list = NULL;
6084 		return 0;
6085 	}
6086 
6087 	/*
6088 	 * Create a socket from which to fetch time stamping capabilities.
6089 	 */
6090 	fd = socket(PF_UNIX, SOCK_RAW, 0);
6091 	if (fd < 0) {
6092 		(void)pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
6093 		    "socket for SIOCETHTOOL(ETHTOOL_GET_TS_INFO): %s", pcap_strerror(errno));
6094 		return -1;
6095 	}
6096 
6097 	memset(&ifr, 0, sizeof(ifr));
6098 	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6099 	memset(&info, 0, sizeof(info));
6100 	info.cmd = ETHTOOL_GET_TS_INFO;
6101 	ifr.ifr_data = (caddr_t)&info;
6102 	if (ioctl(fd, SIOCETHTOOL, &ifr) == -1) {
6103 		int save_errno = errno;
6104 
6105 		close(fd);
6106 		switch (save_errno) {
6107 
6108 		case EOPNOTSUPP:
6109 		case EINVAL:
6110 			/*
6111 			 * OK, this OS version or driver doesn't support
6112 			 * asking for the time stamping types, so let's
6113 			 * just return all the possible types.
6114 			 */
6115 			iface_set_all_ts_types(handle);
6116 			return 0;
6117 
6118 		case ENODEV:
6119 			/*
6120 			 * OK, no such device.
6121 			 * The user will find that out when they try to
6122 			 * activate the device; just return an empty
6123 			 * list of time stamp types.
6124 			 */
6125 			handle->tstamp_type_list = NULL;
6126 			return 0;
6127 
6128 		default:
6129 			/*
6130 			 * Other error.
6131 			 */
6132 			pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
6133 			    "%s: SIOCETHTOOL(ETHTOOL_GET_TS_INFO) ioctl failed: %s", device,
6134 			    strerror(save_errno));
6135 			return -1;
6136 		}
6137 	}
6138 	close(fd);
6139 
6140 	/*
6141 	 * Do we support hardware time stamping of *all* packets?
6142 	 */
6143 	if (!(info.rx_filters & (1 << HWTSTAMP_FILTER_ALL))) {
6144 		/*
6145 		 * No, so don't report any time stamp types.
6146 		 *
6147 		 * XXX - some devices either don't report
6148 		 * HWTSTAMP_FILTER_ALL when they do support it, or
6149 		 * report HWTSTAMP_FILTER_ALL but map it to only
6150 		 * time stamping a few PTP packets.  See
6151 		 * http://marc.info/?l=linux-netdev&m=146318183529571&w=2
6152 		 */
6153 		handle->tstamp_type_list = NULL;
6154 		return 0;
6155 	}
6156 
6157 	num_ts_types = 0;
6158 	for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) {
6159 		if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val)
6160 			num_ts_types++;
6161 	}
6162 	handle->tstamp_type_count = num_ts_types;
6163 	if (num_ts_types != 0) {
6164 		handle->tstamp_type_list = malloc(num_ts_types * sizeof(u_int));
6165 		for (i = 0, j = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) {
6166 			if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val) {
6167 				handle->tstamp_type_list[j] = sof_ts_type_map[i].pcap_tstamp_val;
6168 				j++;
6169 			}
6170 		}
6171 	} else
6172 		handle->tstamp_type_list = NULL;
6173 
6174 	return 0;
6175 }
6176 #else /* ETHTOOL_GET_TS_INFO */
6177 static int
6178 iface_ethtool_get_ts_info(const char *device, pcap_t *handle, char *ebuf _U_)
6179 {
6180 	/*
6181 	 * This doesn't apply to the "any" device; you can't say "turn on
6182 	 * hardware time stamping for all devices that exist now and arrange
6183 	 * that it be turned on for any device that appears in the future",
6184 	 * and not all devices even necessarily *support* hardware time
6185 	 * stamping, so don't report any time stamp types.
6186 	 */
6187 	if (strcmp(device, "any") == 0) {
6188 		handle->tstamp_type_list = NULL;
6189 		return 0;
6190 	}
6191 
6192 	/*
6193 	 * We don't have an ioctl to use to ask what's supported,
6194 	 * so say we support everything.
6195 	 */
6196 	iface_set_all_ts_types(handle);
6197 	return 0;
6198 }
6199 #endif /* ETHTOOL_GET_TS_INFO */
6200 
6201 #endif /* defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) */
6202 
6203 #ifdef HAVE_PACKET_RING
6204 /*
6205  * Find out if we have any form of fragmentation/reassembly offloading.
6206  *
6207  * We do so using SIOCETHTOOL checking for various types of offloading;
6208  * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
6209  * of the types of offloading, there's nothing we can do to check, so
6210  * we just say "no, we don't".
6211  */
6212 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
6213 static int
6214 iface_ethtool_flag_ioctl(pcap_t *handle, int cmd, const char *cmdname)
6215 {
6216 	struct ifreq	ifr;
6217 	struct ethtool_value eval;
6218 
6219 	memset(&ifr, 0, sizeof(ifr));
6220 	strlcpy(ifr.ifr_name, handle->opt.device, sizeof(ifr.ifr_name));
6221 	eval.cmd = cmd;
6222 	eval.data = 0;
6223 	ifr.ifr_data = (caddr_t)&eval;
6224 	if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
6225 		if (errno == EOPNOTSUPP || errno == EINVAL) {
6226 			/*
6227 			 * OK, let's just return 0, which, in our
6228 			 * case, either means "no, what we're asking
6229 			 * about is not enabled" or "all the flags
6230 			 * are clear (i.e., nothing is enabled)".
6231 			 */
6232 			return 0;
6233 		}
6234 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6235 		    "%s: SIOCETHTOOL(%s) ioctl failed: %s", handle->opt.device,
6236 		    cmdname, strerror(errno));
6237 		return -1;
6238 	}
6239 	return eval.data;
6240 }
6241 
6242 static int
6243 iface_get_offload(pcap_t *handle)
6244 {
6245 	int ret;
6246 
6247 #ifdef ETHTOOL_GTSO
6248 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO");
6249 	if (ret == -1)
6250 		return -1;
6251 	if (ret)
6252 		return 1;	/* TCP segmentation offloading on */
6253 #endif
6254 
6255 #ifdef ETHTOOL_GUFO
6256 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO");
6257 	if (ret == -1)
6258 		return -1;
6259 	if (ret)
6260 		return 1;	/* UDP fragmentation offloading on */
6261 #endif
6262 
6263 #ifdef ETHTOOL_GGSO
6264 	/*
6265 	 * XXX - will this cause large unsegmented packets to be
6266 	 * handed to PF_PACKET sockets on transmission?  If not,
6267 	 * this need not be checked.
6268 	 */
6269 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO");
6270 	if (ret == -1)
6271 		return -1;
6272 	if (ret)
6273 		return 1;	/* generic segmentation offloading on */
6274 #endif
6275 
6276 #ifdef ETHTOOL_GFLAGS
6277 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
6278 	if (ret == -1)
6279 		return -1;
6280 	if (ret & ETH_FLAG_LRO)
6281 		return 1;	/* large receive offloading on */
6282 #endif
6283 
6284 #ifdef ETHTOOL_GGRO
6285 	/*
6286 	 * XXX - will this cause large reassembled packets to be
6287 	 * handed to PF_PACKET sockets on receipt?  If not,
6288 	 * this need not be checked.
6289 	 */
6290 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO");
6291 	if (ret == -1)
6292 		return -1;
6293 	if (ret)
6294 		return 1;	/* generic (large) receive offloading on */
6295 #endif
6296 
6297 	return 0;
6298 }
6299 #else /* SIOCETHTOOL */
6300 static int
6301 iface_get_offload(pcap_t *handle _U_)
6302 {
6303 	/*
6304 	 * XXX - do we need to get this information if we don't
6305 	 * have the ethtool ioctls?  If so, how do we do that?
6306 	 */
6307 	return 0;
6308 }
6309 #endif /* SIOCETHTOOL */
6310 
6311 #endif /* HAVE_PACKET_RING */
6312 
6313 #endif /* HAVE_PF_PACKET_SOCKETS */
6314 
6315 /* ===== Functions to interface to the older kernels ================== */
6316 
6317 /*
6318  * Try to open a packet socket using the old kernel interface.
6319  * Returns 1 on success and a PCAP_ERROR_ value on an error.
6320  */
6321 static int
6322 activate_old(pcap_t *handle)
6323 {
6324 	struct pcap_linux *handlep = handle->priv;
6325 	int		arptype;
6326 	struct ifreq	ifr;
6327 	const char	*device = handle->opt.device;
6328 	struct utsname	utsname;
6329 	int		mtu;
6330 
6331 	/* Open the socket */
6332 
6333 	handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
6334 	if (handle->fd == -1) {
6335 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6336 			 "socket: %s", pcap_strerror(errno));
6337 		if (errno == EPERM || errno == EACCES) {
6338 			/*
6339 			 * You don't have permission to open the
6340 			 * socket.
6341 			 */
6342 			return PCAP_ERROR_PERM_DENIED;
6343 		} else {
6344 			/*
6345 			 * Other error.
6346 			 */
6347 			return PCAP_ERROR;
6348 		}
6349 	}
6350 
6351 	/* It worked - we are using the old interface */
6352 	handlep->sock_packet = 1;
6353 
6354 	/* ...which means we get the link-layer header. */
6355 	handlep->cooked = 0;
6356 
6357 	/* Bind to the given device */
6358 
6359 	if (strcmp(device, "any") == 0) {
6360 		strlcpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
6361 			PCAP_ERRBUF_SIZE);
6362 		return PCAP_ERROR;
6363 	}
6364 	if (iface_bind_old(handle->fd, device, handle->errbuf) == -1)
6365 		return PCAP_ERROR;
6366 
6367 	/*
6368 	 * Try to get the link-layer type.
6369 	 */
6370 	arptype = iface_get_arptype(handle->fd, device, handle->errbuf);
6371 	if (arptype < 0)
6372 		return PCAP_ERROR;
6373 
6374 	/*
6375 	 * Try to find the DLT_ type corresponding to that
6376 	 * link-layer type.
6377 	 */
6378 	map_arphrd_to_dlt(handle, handle->fd, arptype, device, 0);
6379 	if (handle->linktype == -1) {
6380 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6381 			 "unknown arptype %d", arptype);
6382 		return PCAP_ERROR;
6383 	}
6384 
6385 	/* Go to promisc mode if requested */
6386 
6387 	if (handle->opt.promisc) {
6388 		memset(&ifr, 0, sizeof(ifr));
6389 		strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6390 		if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
6391 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6392 				 "SIOCGIFFLAGS: %s", pcap_strerror(errno));
6393 			return PCAP_ERROR;
6394 		}
6395 		if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
6396 			/*
6397 			 * Promiscuous mode isn't currently on,
6398 			 * so turn it on, and remember that
6399 			 * we should turn it off when the
6400 			 * pcap_t is closed.
6401 			 */
6402 
6403 			/*
6404 			 * If we haven't already done so, arrange
6405 			 * to have "pcap_close_all()" called when
6406 			 * we exit.
6407 			 */
6408 			if (!pcap_do_addexit(handle)) {
6409 				/*
6410 				 * "atexit()" failed; don't put
6411 				 * the interface in promiscuous
6412 				 * mode, just give up.
6413 				 */
6414 				return PCAP_ERROR;
6415 			}
6416 
6417 			ifr.ifr_flags |= IFF_PROMISC;
6418 			if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
6419 			        pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6420 					 "SIOCSIFFLAGS: %s",
6421 					 pcap_strerror(errno));
6422 				return PCAP_ERROR;
6423 			}
6424 			handlep->must_do_on_close |= MUST_CLEAR_PROMISC;
6425 
6426 			/*
6427 			 * Add this to the list of pcaps
6428 			 * to close when we exit.
6429 			 */
6430 			pcap_add_to_pcaps_to_close(handle);
6431 		}
6432 	}
6433 
6434 	/*
6435 	 * Compute the buffer size.
6436 	 *
6437 	 * We're using SOCK_PACKET, so this might be a 2.0[.x]
6438 	 * kernel, and might require special handling - check.
6439 	 */
6440 	if (uname(&utsname) < 0 ||
6441 	    strncmp(utsname.release, "2.0", 3) == 0) {
6442 		/*
6443 		 * Either we couldn't find out what kernel release
6444 		 * this is, or it's a 2.0[.x] kernel.
6445 		 *
6446 		 * In the 2.0[.x] kernel, a "recvfrom()" on
6447 		 * a SOCK_PACKET socket, with MSG_TRUNC set, will
6448 		 * return the number of bytes read, so if we pass
6449 		 * a length based on the snapshot length, it'll
6450 		 * return the number of bytes from the packet
6451 		 * copied to userland, not the actual length
6452 		 * of the packet.
6453 		 *
6454 		 * This means that, for example, the IP dissector
6455 		 * in tcpdump will get handed a packet length less
6456 		 * than the length in the IP header, and will
6457 		 * complain about "truncated-ip".
6458 		 *
6459 		 * So we don't bother trying to copy from the
6460 		 * kernel only the bytes in which we're interested,
6461 		 * but instead copy them all, just as the older
6462 		 * versions of libpcap for Linux did.
6463 		 *
6464 		 * The buffer therefore needs to be big enough to
6465 		 * hold the largest packet we can get from this
6466 		 * device.  Unfortunately, we can't get the MRU
6467 		 * of the network; we can only get the MTU.  The
6468 		 * MTU may be too small, in which case a packet larger
6469 		 * than the buffer size will be truncated *and* we
6470 		 * won't get the actual packet size.
6471 		 *
6472 		 * However, if the snapshot length is larger than
6473 		 * the buffer size based on the MTU, we use the
6474 		 * snapshot length as the buffer size, instead;
6475 		 * this means that with a sufficiently large snapshot
6476 		 * length we won't artificially truncate packets
6477 		 * to the MTU-based size.
6478 		 *
6479 		 * This mess just one of many problems with packet
6480 		 * capture on 2.0[.x] kernels; you really want a
6481 		 * 2.2[.x] or later kernel if you want packet capture
6482 		 * to work well.
6483 		 */
6484 		mtu = iface_get_mtu(handle->fd, device, handle->errbuf);
6485 		if (mtu == -1)
6486 			return PCAP_ERROR;
6487 		handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
6488 		if (handle->bufsize < (u_int)handle->snapshot)
6489 			handle->bufsize = (u_int)handle->snapshot;
6490 	} else {
6491 		/*
6492 		 * This is a 2.2[.x] or later kernel.
6493 		 *
6494 		 * We can safely pass "recvfrom()" a byte count
6495 		 * based on the snapshot length.
6496 		 */
6497 		handle->bufsize = (u_int)handle->snapshot;
6498 	}
6499 
6500 	/*
6501 	 * Default value for offset to align link-layer payload
6502 	 * on a 4-byte boundary.
6503 	 */
6504 	handle->offset	 = 0;
6505 
6506 	/*
6507 	 * SOCK_PACKET sockets don't supply information from
6508 	 * stripped VLAN tags.
6509 	 */
6510 	handlep->vlan_offset = -1; /* unknown */
6511 
6512 	return 1;
6513 }
6514 
6515 /*
6516  *  Bind the socket associated with FD to the given device using the
6517  *  interface of the old kernels.
6518  */
6519 static int
6520 iface_bind_old(int fd, const char *device, char *ebuf)
6521 {
6522 	struct sockaddr	saddr;
6523 	int		err;
6524 	socklen_t	errlen = sizeof(err);
6525 
6526 	memset(&saddr, 0, sizeof(saddr));
6527 	strlcpy(saddr.sa_data, device, sizeof(saddr.sa_data));
6528 	if (bind(fd, &saddr, sizeof(saddr)) == -1) {
6529 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
6530 			 "bind: %s", pcap_strerror(errno));
6531 		return -1;
6532 	}
6533 
6534 	/* Any pending errors, e.g., network is down? */
6535 
6536 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
6537 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
6538 			"getsockopt: %s", pcap_strerror(errno));
6539 		return -1;
6540 	}
6541 
6542 	if (err > 0) {
6543 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
6544 			"bind: %s", pcap_strerror(err));
6545 		return -1;
6546 	}
6547 
6548 	return 0;
6549 }
6550 
6551 
6552 /* ===== System calls available on all supported kernels ============== */
6553 
6554 /*
6555  *  Query the kernel for the MTU of the given interface.
6556  */
6557 static int
6558 iface_get_mtu(int fd, const char *device, char *ebuf)
6559 {
6560 	struct ifreq	ifr;
6561 
6562 	if (!device)
6563 		return BIGGER_THAN_ALL_MTUS;
6564 
6565 	memset(&ifr, 0, sizeof(ifr));
6566 	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6567 
6568 	if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
6569 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
6570 			 "SIOCGIFMTU: %s", pcap_strerror(errno));
6571 		return -1;
6572 	}
6573 
6574 	return ifr.ifr_mtu;
6575 }
6576 
6577 /*
6578  *  Get the hardware type of the given interface as ARPHRD_xxx constant.
6579  */
6580 static int
6581 iface_get_arptype(int fd, const char *device, char *ebuf)
6582 {
6583 	struct ifreq	ifr;
6584 
6585 	memset(&ifr, 0, sizeof(ifr));
6586 	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6587 
6588 	if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
6589 		pcap_snprintf(ebuf, PCAP_ERRBUF_SIZE,
6590 			 "SIOCGIFHWADDR: %s", pcap_strerror(errno));
6591 		if (errno == ENODEV) {
6592 			/*
6593 			 * No such device.
6594 			 */
6595 			return PCAP_ERROR_NO_SUCH_DEVICE;
6596 		}
6597 		return PCAP_ERROR;
6598 	}
6599 
6600 	return ifr.ifr_hwaddr.sa_family;
6601 }
6602 
6603 #ifdef SO_ATTACH_FILTER
6604 static int
6605 fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
6606 {
6607 	struct pcap_linux *handlep = handle->priv;
6608 	size_t prog_size;
6609 	register int i;
6610 	register struct bpf_insn *p;
6611 	struct bpf_insn *f;
6612 	int len;
6613 
6614 	/*
6615 	 * Make a copy of the filter, and modify that copy if
6616 	 * necessary.
6617 	 */
6618 	prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
6619 	len = handle->fcode.bf_len;
6620 	f = (struct bpf_insn *)malloc(prog_size);
6621 	if (f == NULL) {
6622 		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6623 			 "malloc: %s", pcap_strerror(errno));
6624 		return -1;
6625 	}
6626 	memcpy(f, handle->fcode.bf_insns, prog_size);
6627 	fcode->len = len;
6628 	fcode->filter = (struct sock_filter *) f;
6629 
6630 	for (i = 0; i < len; ++i) {
6631 		p = &f[i];
6632 		/*
6633 		 * What type of instruction is this?
6634 		 */
6635 		switch (BPF_CLASS(p->code)) {
6636 
6637 		case BPF_RET:
6638 			/*
6639 			 * It's a return instruction; are we capturing
6640 			 * in memory-mapped mode?
6641 			 */
6642 			if (!is_mmapped) {
6643 				/*
6644 				 * No; is the snapshot length a constant,
6645 				 * rather than the contents of the
6646 				 * accumulator?
6647 				 */
6648 				if (BPF_MODE(p->code) == BPF_K) {
6649 					/*
6650 					 * Yes - if the value to be returned,
6651 					 * i.e. the snapshot length, is
6652 					 * anything other than 0, make it
6653 					 * MAXIMUM_SNAPLEN, so that the packet
6654 					 * is truncated by "recvfrom()",
6655 					 * not by the filter.
6656 					 *
6657 					 * XXX - there's nothing we can
6658 					 * easily do if it's getting the
6659 					 * value from the accumulator; we'd
6660 					 * have to insert code to force
6661 					 * non-zero values to be
6662 					 * MAXIMUM_SNAPLEN.
6663 					 */
6664 					if (p->k != 0)
6665 						p->k = MAXIMUM_SNAPLEN;
6666 				}
6667 			}
6668 			break;
6669 
6670 		case BPF_LD:
6671 		case BPF_LDX:
6672 			/*
6673 			 * It's a load instruction; is it loading
6674 			 * from the packet?
6675 			 */
6676 			switch (BPF_MODE(p->code)) {
6677 
6678 			case BPF_ABS:
6679 			case BPF_IND:
6680 			case BPF_MSH:
6681 				/*
6682 				 * Yes; are we in cooked mode?
6683 				 */
6684 				if (handlep->cooked) {
6685 					/*
6686 					 * Yes, so we need to fix this
6687 					 * instruction.
6688 					 */
6689 					if (fix_offset(p) < 0) {
6690 						/*
6691 						 * We failed to do so.
6692 						 * Return 0, so our caller
6693 						 * knows to punt to userland.
6694 						 */
6695 						return 0;
6696 					}
6697 				}
6698 				break;
6699 			}
6700 			break;
6701 		}
6702 	}
6703 	return 1;	/* we succeeded */
6704 }
6705 
6706 static int
6707 fix_offset(struct bpf_insn *p)
6708 {
6709 	/*
6710 	 * What's the offset?
6711 	 */
6712 	if (p->k >= SLL_HDR_LEN) {
6713 		/*
6714 		 * It's within the link-layer payload; that starts at an
6715 		 * offset of 0, as far as the kernel packet filter is
6716 		 * concerned, so subtract the length of the link-layer
6717 		 * header.
6718 		 */
6719 		p->k -= SLL_HDR_LEN;
6720 	} else if (p->k == 0) {
6721 		/*
6722 		 * It's the packet type field; map it to the special magic
6723 		 * kernel offset for that field.
6724 		 */
6725 		p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
6726 	} else if (p->k == 14) {
6727 		/*
6728 		 * It's the protocol field; map it to the special magic
6729 		 * kernel offset for that field.
6730 		 */
6731 		p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
6732 	} else if ((bpf_int32)(p->k) > 0) {
6733 		/*
6734 		 * It's within the header, but it's not one of those
6735 		 * fields; we can't do that in the kernel, so punt
6736 		 * to userland.
6737 		 */
6738 		return -1;
6739 	}
6740 	return 0;
6741 }
6742 
6743 static int
6744 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
6745 {
6746 	int total_filter_on = 0;
6747 	int save_mode;
6748 	int ret;
6749 	int save_errno;
6750 
6751 	/*
6752 	 * The socket filter code doesn't discard all packets queued
6753 	 * up on the socket when the filter is changed; this means
6754 	 * that packets that don't match the new filter may show up
6755 	 * after the new filter is put onto the socket, if those
6756 	 * packets haven't yet been read.
6757 	 *
6758 	 * This means, for example, that if you do a tcpdump capture
6759 	 * with a filter, the first few packets in the capture might
6760 	 * be packets that wouldn't have passed the filter.
6761 	 *
6762 	 * We therefore discard all packets queued up on the socket
6763 	 * when setting a kernel filter.  (This isn't an issue for
6764 	 * userland filters, as the userland filtering is done after
6765 	 * packets are queued up.)
6766 	 *
6767 	 * To flush those packets, we put the socket in read-only mode,
6768 	 * and read packets from the socket until there are no more to
6769 	 * read.
6770 	 *
6771 	 * In order to keep that from being an infinite loop - i.e.,
6772 	 * to keep more packets from arriving while we're draining
6773 	 * the queue - we put the "total filter", which is a filter
6774 	 * that rejects all packets, onto the socket before draining
6775 	 * the queue.
6776 	 *
6777 	 * This code deliberately ignores any errors, so that you may
6778 	 * get bogus packets if an error occurs, rather than having
6779 	 * the filtering done in userland even if it could have been
6780 	 * done in the kernel.
6781 	 */
6782 	if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
6783 		       &total_fcode, sizeof(total_fcode)) == 0) {
6784 		char drain[1];
6785 
6786 		/*
6787 		 * Note that we've put the total filter onto the socket.
6788 		 */
6789 		total_filter_on = 1;
6790 
6791 		/*
6792 		 * Save the socket's current mode, and put it in
6793 		 * non-blocking mode; we drain it by reading packets
6794 		 * until we get an error (which is normally a
6795 		 * "nothing more to be read" error).
6796 		 */
6797 		save_mode = fcntl(handle->fd, F_GETFL, 0);
6798 		if (save_mode == -1) {
6799 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6800 			    "can't get FD flags when changing filter: %s",
6801 			    pcap_strerror(errno));
6802 			return -2;
6803 		}
6804 		if (fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) < 0) {
6805 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6806 			    "can't set nonblocking mode when changing filter: %s",
6807 			    pcap_strerror(errno));
6808 			return -2;
6809 		}
6810 		while (recv(handle->fd, &drain, sizeof drain, MSG_TRUNC) >= 0)
6811 			;
6812 		save_errno = errno;
6813 		if (save_errno != EAGAIN) {
6814 			/*
6815 			 * Fatal error.
6816 			 *
6817 			 * If we can't restore the mode or reset the
6818 			 * kernel filter, there's nothing we can do.
6819 			 */
6820 			(void)fcntl(handle->fd, F_SETFL, save_mode);
6821 			(void)reset_kernel_filter(handle);
6822 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6823 			    "recv failed when changing filter: %s",
6824 			    pcap_strerror(save_errno));
6825 			return -2;
6826 		}
6827 		if (fcntl(handle->fd, F_SETFL, save_mode) == -1) {
6828 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6829 			    "can't restore FD flags when changing filter: %s",
6830 			    pcap_strerror(save_errno));
6831 			return -2;
6832 		}
6833 	}
6834 
6835 	/*
6836 	 * Now attach the new filter.
6837 	 */
6838 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
6839 			 fcode, sizeof(*fcode));
6840 	if (ret == -1 && total_filter_on) {
6841 		/*
6842 		 * Well, we couldn't set that filter on the socket,
6843 		 * but we could set the total filter on the socket.
6844 		 *
6845 		 * This could, for example, mean that the filter was
6846 		 * too big to put into the kernel, so we'll have to
6847 		 * filter in userland; in any case, we'll be doing
6848 		 * filtering in userland, so we need to remove the
6849 		 * total filter so we see packets.
6850 		 */
6851 		save_errno = errno;
6852 
6853 		/*
6854 		 * If this fails, we're really screwed; we have the
6855 		 * total filter on the socket, and it won't come off.
6856 		 * Report it as a fatal error.
6857 		 */
6858 		if (reset_kernel_filter(handle) == -1) {
6859 			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6860 			    "can't remove kernel total filter: %s",
6861 			    pcap_strerror(errno));
6862 			return -2;	/* fatal error */
6863 		}
6864 
6865 		errno = save_errno;
6866 	}
6867 	return ret;
6868 }
6869 
6870 static int
6871 reset_kernel_filter(pcap_t *handle)
6872 {
6873 	/*
6874 	 * setsockopt() barfs unless it get a dummy parameter.
6875 	 * valgrind whines unless the value is initialized,
6876 	 * as it has no idea that setsockopt() ignores its
6877 	 * parameter.
6878 	 */
6879 	int dummy = 0;
6880 
6881 	return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
6882 				   &dummy, sizeof(dummy));
6883 }
6884 #endif
6885