1/*
2 *  pcap-linux.c: Packet capture interface to the Linux kernel
3 *
4 *  Copyright (c) 2000 Torsten Landschoff <torsten@debian.org>
5 *  		       Sebastian Krahmer  <krahmer@cs.uni-potsdam.de>
6 *
7 *  License: BSD
8 *
9 *  Redistribution and use in source and binary forms, with or without
10 *  modification, are permitted provided that the following conditions
11 *  are met:
12 *
13 *  1. Redistributions of source code must retain the above copyright
14 *     notice, this list of conditions and the following disclaimer.
15 *  2. Redistributions in binary form must reproduce the above copyright
16 *     notice, this list of conditions and the following disclaimer in
17 *     the documentation and/or other materials provided with the
18 *     distribution.
19 *  3. The names of the authors may not be used to endorse or promote
20 *     products derived from this software without specific prior
21 *     written permission.
22 *
23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
24 *  IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
25 *  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
26 *
27 *  Modifications:     Added PACKET_MMAP support
28 *                     Paolo Abeni <paolo.abeni@email.it>
29 *                     Added TPACKET_V3 support
30 *                     Gabor Tatarka <gabor.tatarka@ericsson.com>
31 *
32 *                     based on previous works of:
33 *                     Simon Patarin <patarin@cs.unibo.it>
34 *                     Phil Wood <cpw@lanl.gov>
35 *
36 * Monitor-mode support for mac80211 includes code taken from the iw
37 * command; the copyright notice for that code is
38 *
39 * Copyright (c) 2007, 2008	Johannes Berg
40 * Copyright (c) 2007		Andy Lutomirski
41 * Copyright (c) 2007		Mike Kershaw
42 * Copyright (c) 2008		G��bor Stefanik
43 *
44 * All rights reserved.
45 *
46 * Redistribution and use in source and binary forms, with or without
47 * modification, are permitted provided that the following conditions
48 * are met:
49 * 1. Redistributions of source code must retain the above copyright
50 *    notice, this list of conditions and the following disclaimer.
51 * 2. Redistributions in binary form must reproduce the above copyright
52 *    notice, this list of conditions and the following disclaimer in the
53 *    documentation and/or other materials provided with the distribution.
54 * 3. The name of the author may not be used to endorse or promote products
55 *    derived from this software without specific prior written permission.
56 *
57 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
58 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
59 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
60 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
61 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
62 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
63 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
64 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
65 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67 * SUCH DAMAGE.
68 */
69
70/*
71 * Known problems with 2.0[.x] kernels:
72 *
73 *   - The loopback device gives every packet twice; on 2.2[.x] kernels,
74 *     if we use PF_PACKET, we can filter out the transmitted version
75 *     of the packet by using data in the "sockaddr_ll" returned by
76 *     "recvfrom()", but, on 2.0[.x] kernels, we have to use
77 *     PF_INET/SOCK_PACKET, which means "recvfrom()" supplies a
78 *     "sockaddr_pkt" which doesn't give us enough information to let
79 *     us do that.
80 *
81 *   - We have to set the interface's IFF_PROMISC flag ourselves, if
82 *     we're to run in promiscuous mode, which means we have to turn
83 *     it off ourselves when we're done; the kernel doesn't keep track
84 *     of how many sockets are listening promiscuously, which means
85 *     it won't get turned off automatically when no sockets are
86 *     listening promiscuously.  We catch "pcap_close()" and, for
87 *     interfaces we put into promiscuous mode, take them out of
88 *     promiscuous mode - which isn't necessarily the right thing to
89 *     do, if another socket also requested promiscuous mode between
90 *     the time when we opened the socket and the time when we close
91 *     the socket.
92 *
93 *   - MSG_TRUNC isn't supported, so you can't specify that "recvfrom()"
94 *     return the amount of data that you could have read, rather than
95 *     the amount that was returned, so we can't just allocate a buffer
96 *     whose size is the snapshot length and pass the snapshot length
97 *     as the byte count, and also pass MSG_TRUNC, so that the return
98 *     value tells us how long the packet was on the wire.
99 *
100 *     This means that, if we want to get the actual size of the packet,
101 *     so we can return it in the "len" field of the packet header,
102 *     we have to read the entire packet, not just the part that fits
103 *     within the snapshot length, and thus waste CPU time copying data
104 *     from the kernel that our caller won't see.
105 *
106 *     We have to get the actual size, and supply it in "len", because
107 *     otherwise, the IP dissector in tcpdump, for example, will complain
108 *     about "truncated-ip", as the packet will appear to have been
109 *     shorter, on the wire, than the IP header said it should have been.
110 */
111
112
113#define _GNU_SOURCE
114
115#ifdef HAVE_CONFIG_H
116#include <config.h>
117#endif
118
119#include <errno.h>
120#include <stdio.h>
121#include <stdlib.h>
122#include <ctype.h>
123#include <unistd.h>
124#include <fcntl.h>
125#include <string.h>
126#include <limits.h>
127#include <sys/stat.h>
128#include <sys/socket.h>
129#include <sys/ioctl.h>
130#include <sys/utsname.h>
131#include <sys/mman.h>
132#include <linux/if.h>
133#include <linux/if_packet.h>
134#include <linux/sockios.h>
135#include <netinet/in.h>
136#include <linux/if_ether.h>
137#include <net/if_arp.h>
138#include <poll.h>
139#include <dirent.h>
140
141#include "pcap-int.h"
142#include "pcap/sll.h"
143#include "pcap/vlan.h"
144
145/*
146 * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
147 * sockets rather than SOCK_PACKET sockets.
148 *
149 * To use them, we include <linux/if_packet.h> rather than
150 * <netpacket/packet.h>; we do so because
151 *
152 *	some Linux distributions (e.g., Slackware 4.0) have 2.2 or
153 *	later kernels and libc5, and don't provide a <netpacket/packet.h>
154 *	file;
155 *
156 *	not all versions of glibc2 have a <netpacket/packet.h> file
157 *	that defines stuff needed for some of the 2.4-or-later-kernel
158 *	features, so if the system has a 2.4 or later kernel, we
159 *	still can't use those features.
160 *
161 * We're already including a number of other <linux/XXX.h> headers, and
162 * this code is Linux-specific (no other OS has PF_PACKET sockets as
163 * a raw packet capture mechanism), so it's not as if you gain any
164 * useful portability by using <netpacket/packet.h>
165 *
166 * XXX - should we just include <linux/if_packet.h> even if PF_PACKET
167 * isn't defined?  It only defines one data structure in 2.0.x, so
168 * it shouldn't cause any problems.
169 */
170#ifdef PF_PACKET
171# include <linux/if_packet.h>
172
173 /*
174  * On at least some Linux distributions (for example, Red Hat 5.2),
175  * there's no <netpacket/packet.h> file, but PF_PACKET is defined if
176  * you include <sys/socket.h>, but <linux/if_packet.h> doesn't define
177  * any of the PF_PACKET stuff such as "struct sockaddr_ll" or any of
178  * the PACKET_xxx stuff.
179  *
180  * So we check whether PACKET_HOST is defined, and assume that we have
181  * PF_PACKET sockets only if it is defined.
182  */
183# ifdef PACKET_HOST
184#  define HAVE_PF_PACKET_SOCKETS
185#  ifdef PACKET_AUXDATA
186#   define HAVE_PACKET_AUXDATA
187#  endif /* PACKET_AUXDATA */
188# endif /* PACKET_HOST */
189
190
191 /* check for memory mapped access avaibility. We assume every needed
192  * struct is defined if the macro TPACKET_HDRLEN is defined, because it
193  * uses many ring related structs and macros */
194# ifdef PCAP_SUPPORT_PACKET_RING
195# ifdef TPACKET_HDRLEN
196#  define HAVE_PACKET_RING
197#  ifdef TPACKET3_HDRLEN
198#   define HAVE_TPACKET3
199#  endif /* TPACKET3_HDRLEN */
200#  ifdef TPACKET2_HDRLEN
201#   define HAVE_TPACKET2
202#  else  /* TPACKET2_HDRLEN */
203#   define TPACKET_V1	0    /* Old kernel with only V1, so no TPACKET_Vn defined */
204#  endif /* TPACKET2_HDRLEN */
205# endif /* TPACKET_HDRLEN */
206# endif /* PCAP_SUPPORT_PACKET_RING */
207#endif /* PF_PACKET */
208
209#ifdef SO_ATTACH_FILTER
210#include <linux/types.h>
211#include <linux/filter.h>
212#endif
213
214#ifdef HAVE_LINUX_NET_TSTAMP_H
215#include <linux/net_tstamp.h>
216#endif
217
218#ifdef HAVE_LINUX_SOCKIOS_H
219#include <linux/sockios.h>
220#endif
221
222#ifdef HAVE_LINUX_IF_BONDING_H
223#include <linux/if_bonding.h>
224
225/*
226 * The ioctl code to use to check whether a device is a bonding device.
227 */
228#if defined(SIOCBONDINFOQUERY)
229	#define BOND_INFO_QUERY_IOCTL SIOCBONDINFOQUERY
230#elif defined(BOND_INFO_QUERY_OLD)
231	#define BOND_INFO_QUERY_IOCTL BOND_INFO_QUERY_OLD
232#endif
233#endif /* HAVE_LINUX_IF_BONDING_H */
234
235/*
236 * Got Wireless Extensions?
237 */
238#ifdef HAVE_LINUX_WIRELESS_H
239#include <linux/wireless.h>
240#endif /* HAVE_LINUX_WIRELESS_H */
241
242/*
243 * Got libnl?
244 */
245#ifdef HAVE_LIBNL
246#include <linux/nl80211.h>
247
248#include <netlink/genl/genl.h>
249#include <netlink/genl/family.h>
250#include <netlink/genl/ctrl.h>
251#include <netlink/msg.h>
252#include <netlink/attr.h>
253#endif /* HAVE_LIBNL */
254
255/*
256 * Got ethtool support?
257 */
258#ifdef HAVE_LINUX_ETHTOOL_H
259#include <linux/ethtool.h>
260#endif
261
262#ifndef HAVE_SOCKLEN_T
263typedef int		socklen_t;
264#endif
265
266#ifndef MSG_TRUNC
267/*
268 * This is being compiled on a system that lacks MSG_TRUNC; define it
269 * with the value it has in the 2.2 and later kernels, so that, on
270 * those kernels, when we pass it in the flags argument to "recvfrom()"
271 * we're passing the right value and thus get the MSG_TRUNC behavior
272 * we want.  (We don't get that behavior on 2.0[.x] kernels, because
273 * they didn't support MSG_TRUNC.)
274 */
275#define MSG_TRUNC	0x20
276#endif
277
278#ifndef SOL_PACKET
279/*
280 * This is being compiled on a system that lacks SOL_PACKET; define it
281 * with the value it has in the 2.2 and later kernels, so that we can
282 * set promiscuous mode in the good modern way rather than the old
283 * 2.0-kernel crappy way.
284 */
285#define SOL_PACKET	263
286#endif
287
288#define MAX_LINKHEADER_SIZE	256
289
290/*
291 * When capturing on all interfaces we use this as the buffer size.
292 * Should be bigger then all MTUs that occur in real life.
293 * 64kB should be enough for now.
294 */
295#define BIGGER_THAN_ALL_MTUS	(64*1024)
296
297/*
298 * Private data for capturing on Linux SOCK_PACKET or PF_PACKET sockets.
299 */
300struct pcap_linux {
301	u_int	packets_read;	/* count of packets read with recvfrom() */
302	long	proc_dropped;	/* packets reported dropped by /proc/net/dev */
303	struct pcap_stat stat;
304
305	char	*device;	/* device name */
306	int	filter_in_userland; /* must filter in userland */
307	int	blocks_to_filter_in_userland;
308	int	must_do_on_close; /* stuff we must do when we close */
309	int	timeout;	/* timeout for buffering */
310	int	sock_packet;	/* using Linux 2.0 compatible interface */
311	int	cooked;		/* using SOCK_DGRAM rather than SOCK_RAW */
312	int	ifindex;	/* interface index of device we're bound to */
313	int	lo_ifindex;	/* interface index of the loopback device */
314	bpf_u_int32 oldmode;	/* mode to restore when turning monitor mode off */
315	char	*mondevice;	/* mac80211 monitor device we created */
316	u_char	*mmapbuf;	/* memory-mapped region pointer */
317	size_t	mmapbuflen;	/* size of region */
318	int	vlan_offset;	/* offset at which to insert vlan tags; if -1, don't insert */
319	u_int	tp_version;	/* version of tpacket_hdr for mmaped ring */
320	u_int	tp_hdrlen;	/* hdrlen of tpacket_hdr for mmaped ring */
321	u_char	*oneshot_buffer; /* buffer for copy of packet */
322	int	poll_timeout;	/* timeout to use in poll() */
323#ifdef HAVE_TPACKET3
324	unsigned char *current_packet; /* Current packet within the TPACKET_V3 block. Move to next block if NULL. */
325	int packets_left; /* Unhandled packets left within the block from previous call to pcap_read_linux_mmap_v3 in case of TPACKET_V3. */
326#endif
327};
328
329/*
330 * Stuff to do when we close.
331 */
332#define MUST_CLEAR_PROMISC	0x00000001	/* clear promiscuous mode */
333#define MUST_CLEAR_RFMON	0x00000002	/* clear rfmon (monitor) mode */
334#define MUST_DELETE_MONIF	0x00000004	/* delete monitor-mode interface */
335
336/*
337 * Prototypes for internal functions and methods.
338 */
339static int get_if_flags(const char *, bpf_u_int32 *, char *);
340static int is_wifi(int, const char *);
341static void map_arphrd_to_dlt(pcap_t *, int, int, const char *, int);
342#ifdef HAVE_PF_PACKET_SOCKETS
343static short int map_packet_type_to_sll_type(short int);
344#endif
345static int pcap_activate_linux(pcap_t *);
346static int activate_old(pcap_t *);
347static int activate_new(pcap_t *);
348static int activate_mmap(pcap_t *, int *);
349static int pcap_can_set_rfmon_linux(pcap_t *);
350static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *);
351static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);
352static int pcap_inject_linux(pcap_t *, const void *, size_t);
353static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
354static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
355static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
356static int pcap_set_datalink_linux(pcap_t *, int);
357static void pcap_cleanup_linux(pcap_t *);
358
359/*
360 * This is what the header structure looks like in a 64-bit kernel;
361 * we use this, rather than struct tpacket_hdr, if we're using
362 * TPACKET_V1 in 32-bit code running on a 64-bit kernel.
363 */
364struct tpacket_hdr_64 {
365	uint64_t	tp_status;
366	unsigned int	tp_len;
367	unsigned int	tp_snaplen;
368	unsigned short	tp_mac;
369	unsigned short	tp_net;
370	unsigned int	tp_sec;
371	unsigned int	tp_usec;
372};
373
374/*
375 * We use this internally as the tpacket version for TPACKET_V1 in
376 * 32-bit code on a 64-bit kernel.
377 */
378#define TPACKET_V1_64 99
379
380union thdr {
381	struct tpacket_hdr		*h1;
382	struct tpacket_hdr_64		*h1_64;
383#ifdef HAVE_TPACKET2
384	struct tpacket2_hdr		*h2;
385#endif
386#ifdef HAVE_TPACKET3
387	struct tpacket_block_desc	*h3;
388#endif
389	void				*raw;
390};
391
392#ifdef HAVE_PACKET_RING
393#define RING_GET_FRAME_AT(h, offset) (((union thdr **)h->buffer)[(offset)])
394#define RING_GET_CURRENT_FRAME(h) RING_GET_FRAME_AT(h, h->offset)
395
396static void destroy_ring(pcap_t *handle);
397static int create_ring(pcap_t *handle, int *status);
398static int prepare_tpacket_socket(pcap_t *handle);
399static void pcap_cleanup_linux_mmap(pcap_t *);
400static int pcap_read_linux_mmap_v1(pcap_t *, int, pcap_handler , u_char *);
401static int pcap_read_linux_mmap_v1_64(pcap_t *, int, pcap_handler , u_char *);
402#ifdef HAVE_TPACKET2
403static int pcap_read_linux_mmap_v2(pcap_t *, int, pcap_handler , u_char *);
404#endif
405#ifdef HAVE_TPACKET3
406static int pcap_read_linux_mmap_v3(pcap_t *, int, pcap_handler , u_char *);
407#endif
408static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *);
409static int pcap_setnonblock_mmap(pcap_t *p, int nonblock);
410static int pcap_getnonblock_mmap(pcap_t *p);
411static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
412    const u_char *bytes);
413#endif
414
415/*
416 * In pre-3.0 kernels, the tp_vlan_tci field is set to whatever the
417 * vlan_tci field in the skbuff is.  0 can either mean "not on a VLAN"
418 * or "on VLAN 0".  There is no flag set in the tp_status field to
419 * distinguish between them.
420 *
421 * In 3.0 and later kernels, if there's a VLAN tag present, the tp_vlan_tci
422 * field is set to the VLAN tag, and the TP_STATUS_VLAN_VALID flag is set
423 * in the tp_status field, otherwise the tp_vlan_tci field is set to 0 and
424 * the TP_STATUS_VLAN_VALID flag isn't set in the tp_status field.
425 *
426 * With a pre-3.0 kernel, we cannot distinguish between packets with no
427 * VLAN tag and packets on VLAN 0, so we will mishandle some packets, and
428 * there's nothing we can do about that.
429 *
430 * So, on those systems, which never set the TP_STATUS_VLAN_VALID flag, we
431 * continue the behavior of earlier libpcaps, wherein we treated packets
432 * with a VLAN tag of 0 as being packets without a VLAN tag rather than packets
433 * on VLAN 0.  We do this by treating packets with a tp_vlan_tci of 0 and
434 * with the TP_STATUS_VLAN_VALID flag not set in tp_status as not having
435 * VLAN tags.  This does the right thing on 3.0 and later kernels, and
436 * continues the old unfixably-imperfect behavior on pre-3.0 kernels.
437 *
438 * If TP_STATUS_VLAN_VALID isn't defined, we test it as the 0x10 bit; it
439 * has that value in 3.0 and later kernels.
440 */
441#ifdef TP_STATUS_VLAN_VALID
442  #define VLAN_VALID(hdr, hv)	((hv)->tp_vlan_tci != 0 || ((hdr)->tp_status & TP_STATUS_VLAN_VALID))
443#else
444  /*
445   * This is being compiled on a system that lacks TP_STATUS_VLAN_VALID,
446   * so we testwith the value it has in the 3.0 and later kernels, so
447   * we can test it if we're running on a system that has it.  (If we're
448   * running on a system that doesn't have it, it won't be set in the
449   * tp_status field, so the tests of it will always fail; that means
450   * we behave the way we did before we introduced this macro.)
451   */
452  #define VLAN_VALID(hdr, hv)	((hv)->tp_vlan_tci != 0 || ((hdr)->tp_status & 0x10))
453#endif
454
455#ifdef TP_STATUS_VLAN_TPID_VALID
456# define VLAN_TPID(hdr, hv)	(((hv)->tp_vlan_tpid || ((hdr)->tp_status & TP_STATUS_VLAN_TPID_VALID)) ? (hv)->tp_vlan_tpid : ETH_P_8021Q)
457#else
458# define VLAN_TPID(hdr, hv)	ETH_P_8021Q
459#endif
460
461/*
462 * Wrap some ioctl calls
463 */
464#ifdef HAVE_PF_PACKET_SOCKETS
465static int	iface_get_id(int fd, const char *device, char *ebuf);
466#endif /* HAVE_PF_PACKET_SOCKETS */
467static int	iface_get_mtu(int fd, const char *device, char *ebuf);
468static int 	iface_get_arptype(int fd, const char *device, char *ebuf);
469#ifdef HAVE_PF_PACKET_SOCKETS
470static int 	iface_bind(int fd, int ifindex, char *ebuf, int protocol);
471#ifdef IW_MODE_MONITOR
472static int	has_wext(int sock_fd, const char *device, char *ebuf);
473#endif /* IW_MODE_MONITOR */
474static int	enter_rfmon_mode(pcap_t *handle, int sock_fd,
475    const char *device);
476#endif /* HAVE_PF_PACKET_SOCKETS */
477#if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
478static int	iface_ethtool_get_ts_info(const char *device, pcap_t *handle,
479    char *ebuf);
480#endif
481#ifdef HAVE_PACKET_RING
482static int	iface_get_offload(pcap_t *handle);
483#endif
484static int 	iface_bind_old(int fd, const char *device, char *ebuf);
485
486#ifdef SO_ATTACH_FILTER
487static int	fix_program(pcap_t *handle, struct sock_fprog *fcode,
488    int is_mapped);
489static int	fix_offset(struct bpf_insn *p);
490static int	set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
491static int	reset_kernel_filter(pcap_t *handle);
492
493static struct sock_filter	total_insn
494	= BPF_STMT(BPF_RET | BPF_K, 0);
495static struct sock_fprog	total_fcode
496	= { 1, &total_insn };
497#endif /* SO_ATTACH_FILTER */
498
499pcap_t *
500pcap_create_interface(const char *device, char *ebuf)
501{
502	pcap_t *handle;
503
504	handle = pcap_create_common(ebuf, sizeof (struct pcap_linux));
505	if (handle == NULL)
506		return NULL;
507
508	handle->activate_op = pcap_activate_linux;
509	handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
510
511#if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
512	/*
513	 * See what time stamp types we support.
514	 */
515	if (iface_ethtool_get_ts_info(device, handle, ebuf) == -1) {
516		pcap_close(handle);
517		return NULL;
518	}
519#endif
520
521#if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
522	/*
523	 * We claim that we support microsecond and nanosecond time
524	 * stamps.
525	 *
526	 * XXX - with adapter-supplied time stamps, can we choose
527	 * microsecond or nanosecond time stamps on arbitrary
528	 * adapters?
529	 */
530	handle->tstamp_precision_count = 2;
531	handle->tstamp_precision_list = malloc(2 * sizeof(u_int));
532	if (handle->tstamp_precision_list == NULL) {
533		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
534		    errno, "malloc");
535		pcap_close(handle);
536		return NULL;
537	}
538	handle->tstamp_precision_list[0] = PCAP_TSTAMP_PRECISION_MICRO;
539	handle->tstamp_precision_list[1] = PCAP_TSTAMP_PRECISION_NANO;
540#endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */
541
542	return handle;
543}
544
545#ifdef HAVE_LIBNL
546/*
547 * If interface {if} is a mac80211 driver, the file
548 * /sys/class/net/{if}/phy80211 is a symlink to
549 * /sys/class/ieee80211/{phydev}, for some {phydev}.
550 *
551 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
552 * least, has a "wmaster0" device and a "wlan0" device; the
553 * latter is the one with the IP address.  Both show up in
554 * "tcpdump -D" output.  Capturing on the wmaster0 device
555 * captures with 802.11 headers.
556 *
557 * airmon-ng searches through /sys/class/net for devices named
558 * monN, starting with mon0; as soon as one *doesn't* exist,
559 * it chooses that as the monitor device name.  If the "iw"
560 * command exists, it does "iw dev {if} interface add {monif}
561 * type monitor", where {monif} is the monitor device.  It
562 * then (sigh) sleeps .1 second, and then configures the
563 * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
564 * is a file, it writes {mondev}, without a newline, to that file,
565 * and again (sigh) sleeps .1 second, and then iwconfig's that
566 * device into monitor mode and configures it up.  Otherwise,
567 * you can't do monitor mode.
568 *
569 * All these devices are "glued" together by having the
570 * /sys/class/net/{device}/phy80211 links pointing to the same
571 * place, so, given a wmaster, wlan, or mon device, you can
572 * find the other devices by looking for devices with
573 * the same phy80211 link.
574 *
575 * To turn monitor mode off, delete the monitor interface,
576 * either with "iw dev {monif} interface del" or by sending
577 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
578 *
579 * Note: if you try to create a monitor device named "monN", and
580 * there's already a "monN" device, it fails, as least with
581 * the netlink interface (which is what iw uses), with a return
582 * value of -ENFILE.  (Return values are negative errnos.)  We
583 * could probably use that to find an unused device.
584 *
585 * Yes, you can have multiple monitor devices for a given
586 * physical device.
587 */
588
589/*
590 * Is this a mac80211 device?  If so, fill in the physical device path and
591 * return 1; if not, return 0.  On an error, fill in handle->errbuf and
592 * return PCAP_ERROR.
593 */
594static int
595get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
596    size_t phydev_max_pathlen)
597{
598	char *pathstr;
599	ssize_t bytes_read;
600
601	/*
602	 * Generate the path string for the symlink to the physical device.
603	 */
604	if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
605		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
606		    "%s: Can't generate path name string for /sys/class/net device",
607		    device);
608		return PCAP_ERROR;
609	}
610	bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
611	if (bytes_read == -1) {
612		if (errno == ENOENT || errno == EINVAL) {
613			/*
614			 * Doesn't exist, or not a symlink; assume that
615			 * means it's not a mac80211 device.
616			 */
617			free(pathstr);
618			return 0;
619		}
620		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
621		    errno, "%s: Can't readlink %s", device, pathstr);
622		free(pathstr);
623		return PCAP_ERROR;
624	}
625	free(pathstr);
626	phydev_path[bytes_read] = '\0';
627	return 1;
628}
629
630#ifdef HAVE_LIBNL_SOCKETS
631#define get_nl_errmsg	nl_geterror
632#else
633/* libnl 2.x compatibility code */
634
635#define nl_sock nl_handle
636
637static inline struct nl_handle *
638nl_socket_alloc(void)
639{
640	return nl_handle_alloc();
641}
642
643static inline void
644nl_socket_free(struct nl_handle *h)
645{
646	nl_handle_destroy(h);
647}
648
649#define get_nl_errmsg	strerror
650
651static inline int
652__genl_ctrl_alloc_cache(struct nl_handle *h, struct nl_cache **cache)
653{
654	struct nl_cache *tmp = genl_ctrl_alloc_cache(h);
655	if (!tmp)
656		return -ENOMEM;
657	*cache = tmp;
658	return 0;
659}
660#define genl_ctrl_alloc_cache __genl_ctrl_alloc_cache
661#endif /* !HAVE_LIBNL_SOCKETS */
662
663struct nl80211_state {
664	struct nl_sock *nl_sock;
665	struct nl_cache *nl_cache;
666	struct genl_family *nl80211;
667};
668
669static int
670nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
671{
672	int err;
673
674	state->nl_sock = nl_socket_alloc();
675	if (!state->nl_sock) {
676		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
677		    "%s: failed to allocate netlink handle", device);
678		return PCAP_ERROR;
679	}
680
681	if (genl_connect(state->nl_sock)) {
682		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
683		    "%s: failed to connect to generic netlink", device);
684		goto out_handle_destroy;
685	}
686
687	err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
688	if (err < 0) {
689		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
690		    "%s: failed to allocate generic netlink cache: %s",
691		    device, get_nl_errmsg(-err));
692		goto out_handle_destroy;
693	}
694
695	state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
696	if (!state->nl80211) {
697		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
698		    "%s: nl80211 not found", device);
699		goto out_cache_free;
700	}
701
702	return 0;
703
704out_cache_free:
705	nl_cache_free(state->nl_cache);
706out_handle_destroy:
707	nl_socket_free(state->nl_sock);
708	return PCAP_ERROR;
709}
710
711static void
712nl80211_cleanup(struct nl80211_state *state)
713{
714	genl_family_put(state->nl80211);
715	nl_cache_free(state->nl_cache);
716	nl_socket_free(state->nl_sock);
717}
718
719static int
720del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
721    const char *device, const char *mondevice);
722
723static int
724add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
725    const char *device, const char *mondevice)
726{
727	struct pcap_linux *handlep = handle->priv;
728	int ifindex;
729	struct nl_msg *msg;
730	int err;
731
732	ifindex = iface_get_id(sock_fd, device, handle->errbuf);
733	if (ifindex == -1)
734		return PCAP_ERROR;
735
736	msg = nlmsg_alloc();
737	if (!msg) {
738		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
739		    "%s: failed to allocate netlink msg", device);
740		return PCAP_ERROR;
741	}
742
743	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
744		    0, NL80211_CMD_NEW_INTERFACE, 0);
745	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
746	NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
747	NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
748
749	err = nl_send_auto_complete(state->nl_sock, msg);
750	if (err < 0) {
751#if defined HAVE_LIBNL_NLE
752		if (err == -NLE_FAILURE) {
753#else
754		if (err == -ENFILE) {
755#endif
756			/*
757			 * Device not available; our caller should just
758			 * keep trying.  (libnl 2.x maps ENFILE to
759			 * NLE_FAILURE; it can also map other errors
760			 * to that, but there's not much we can do
761			 * about that.)
762			 */
763			nlmsg_free(msg);
764			return 0;
765		} else {
766			/*
767			 * Real failure, not just "that device is not
768			 * available.
769			 */
770			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
771			    "%s: nl_send_auto_complete failed adding %s interface: %s",
772			    device, mondevice, get_nl_errmsg(-err));
773			nlmsg_free(msg);
774			return PCAP_ERROR;
775		}
776	}
777	err = nl_wait_for_ack(state->nl_sock);
778	if (err < 0) {
779#if defined HAVE_LIBNL_NLE
780		if (err == -NLE_FAILURE) {
781#else
782		if (err == -ENFILE) {
783#endif
784			/*
785			 * Device not available; our caller should just
786			 * keep trying.  (libnl 2.x maps ENFILE to
787			 * NLE_FAILURE; it can also map other errors
788			 * to that, but there's not much we can do
789			 * about that.)
790			 */
791			nlmsg_free(msg);
792			return 0;
793		} else {
794			/*
795			 * Real failure, not just "that device is not
796			 * available.
797			 */
798			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
799			    "%s: nl_wait_for_ack failed adding %s interface: %s",
800			    device, mondevice, get_nl_errmsg(-err));
801			nlmsg_free(msg);
802			return PCAP_ERROR;
803		}
804	}
805
806	/*
807	 * Success.
808	 */
809	nlmsg_free(msg);
810
811	/*
812	 * Try to remember the monitor device.
813	 */
814	handlep->mondevice = strdup(mondevice);
815	if (handlep->mondevice == NULL) {
816		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
817		    errno, "strdup");
818		/*
819		 * Get rid of the monitor device.
820		 */
821		del_mon_if(handle, sock_fd, state, device, mondevice);
822		return PCAP_ERROR;
823	}
824	return 1;
825
826nla_put_failure:
827	pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
828	    "%s: nl_put failed adding %s interface",
829	    device, mondevice);
830	nlmsg_free(msg);
831	return PCAP_ERROR;
832}
833
834static int
835del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
836    const char *device, const char *mondevice)
837{
838	int ifindex;
839	struct nl_msg *msg;
840	int err;
841
842	ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
843	if (ifindex == -1)
844		return PCAP_ERROR;
845
846	msg = nlmsg_alloc();
847	if (!msg) {
848		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
849		    "%s: failed to allocate netlink msg", device);
850		return PCAP_ERROR;
851	}
852
853	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
854		    0, NL80211_CMD_DEL_INTERFACE, 0);
855	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
856
857	err = nl_send_auto_complete(state->nl_sock, msg);
858	if (err < 0) {
859		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
860		    "%s: nl_send_auto_complete failed deleting %s interface: %s",
861		    device, mondevice, get_nl_errmsg(-err));
862		nlmsg_free(msg);
863		return PCAP_ERROR;
864	}
865	err = nl_wait_for_ack(state->nl_sock);
866	if (err < 0) {
867		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
868		    "%s: nl_wait_for_ack failed adding %s interface: %s",
869		    device, mondevice, get_nl_errmsg(-err));
870		nlmsg_free(msg);
871		return PCAP_ERROR;
872	}
873
874	/*
875	 * Success.
876	 */
877	nlmsg_free(msg);
878	return 1;
879
880nla_put_failure:
881	pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
882	    "%s: nl_put failed deleting %s interface",
883	    device, mondevice);
884	nlmsg_free(msg);
885	return PCAP_ERROR;
886}
887
888static int
889enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
890{
891	struct pcap_linux *handlep = handle->priv;
892	int ret;
893	char phydev_path[PATH_MAX+1];
894	struct nl80211_state nlstate;
895	struct ifreq ifr;
896	u_int n;
897
898	/*
899	 * Is this a mac80211 device?
900	 */
901	ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
902	if (ret < 0)
903		return ret;	/* error */
904	if (ret == 0)
905		return 0;	/* no error, but not mac80211 device */
906
907	/*
908	 * XXX - is this already a monN device?
909	 * If so, we're done.
910	 * Is that determined by old Wireless Extensions ioctls?
911	 */
912
913	/*
914	 * OK, it's apparently a mac80211 device.
915	 * Try to find an unused monN device for it.
916	 */
917	ret = nl80211_init(handle, &nlstate, device);
918	if (ret != 0)
919		return ret;
920	for (n = 0; n < UINT_MAX; n++) {
921		/*
922		 * Try mon{n}.
923		 */
924		char mondevice[3+10+1];	/* mon{UINT_MAX}\0 */
925
926		pcap_snprintf(mondevice, sizeof mondevice, "mon%u", n);
927		ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
928		if (ret == 1) {
929			/*
930			 * Success.  We don't clean up the libnl state
931			 * yet, as we'll be using it later.
932			 */
933			goto added;
934		}
935		if (ret < 0) {
936			/*
937			 * Hard failure.  Just return ret; handle->errbuf
938			 * has already been set.
939			 */
940			nl80211_cleanup(&nlstate);
941			return ret;
942		}
943	}
944
945	pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
946	    "%s: No free monN interfaces", device);
947	nl80211_cleanup(&nlstate);
948	return PCAP_ERROR;
949
950added:
951
952#if 0
953	/*
954	 * Sleep for .1 seconds.
955	 */
956	delay.tv_sec = 0;
957	delay.tv_nsec = 500000000;
958	nanosleep(&delay, NULL);
959#endif
960
961	/*
962	 * If we haven't already done so, arrange to have
963	 * "pcap_close_all()" called when we exit.
964	 */
965	if (!pcap_do_addexit(handle)) {
966		/*
967		 * "atexit()" failed; don't put the interface
968		 * in rfmon mode, just give up.
969		 */
970		del_mon_if(handle, sock_fd, &nlstate, device,
971		    handlep->mondevice);
972		nl80211_cleanup(&nlstate);
973		return PCAP_ERROR;
974	}
975
976	/*
977	 * Now configure the monitor interface up.
978	 */
979	memset(&ifr, 0, sizeof(ifr));
980	strlcpy(ifr.ifr_name, handlep->mondevice, sizeof(ifr.ifr_name));
981	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
982		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
983		    errno, "%s: Can't get flags for %s", device,
984		    handlep->mondevice);
985		del_mon_if(handle, sock_fd, &nlstate, device,
986		    handlep->mondevice);
987		nl80211_cleanup(&nlstate);
988		return PCAP_ERROR;
989	}
990	ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
991	if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
992		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
993		    errno, "%s: Can't set flags for %s", device,
994		    handlep->mondevice);
995		del_mon_if(handle, sock_fd, &nlstate, device,
996		    handlep->mondevice);
997		nl80211_cleanup(&nlstate);
998		return PCAP_ERROR;
999	}
1000
1001	/*
1002	 * Success.  Clean up the libnl state.
1003	 */
1004	nl80211_cleanup(&nlstate);
1005
1006	/*
1007	 * Note that we have to delete the monitor device when we close
1008	 * the handle.
1009	 */
1010	handlep->must_do_on_close |= MUST_DELETE_MONIF;
1011
1012	/*
1013	 * Add this to the list of pcaps to close when we exit.
1014	 */
1015	pcap_add_to_pcaps_to_close(handle);
1016
1017	return 1;
1018}
1019#endif /* HAVE_LIBNL */
1020
1021#ifdef IW_MODE_MONITOR
1022/*
1023 * Bonding devices mishandle unknown ioctls; they fail with ENODEV
1024 * rather than ENOTSUP, EOPNOTSUPP, or ENOTTY, so Wireless Extensions
1025 * will fail with ENODEV if we try to do them on a bonding device,
1026 * making us return a "no such device" indication rather than just
1027 * saying "no Wireless Extensions".
1028 *
1029 * So we check for bonding devices, if we can, before trying those
1030 * ioctls, by trying a bonding device information query ioctl to see
1031 * whether it succeeds.
1032 */
1033static int
1034is_bonding_device(int fd, const char *device)
1035{
1036#ifdef BOND_INFO_QUERY_IOCTL
1037	struct ifreq ifr;
1038	ifbond ifb;
1039
1040	memset(&ifr, 0, sizeof ifr);
1041	strlcpy(ifr.ifr_name, device, sizeof ifr.ifr_name);
1042	memset(&ifb, 0, sizeof ifb);
1043	ifr.ifr_data = (caddr_t)&ifb;
1044	if (ioctl(fd, BOND_INFO_QUERY_IOCTL, &ifr) == 0)
1045		return 1;	/* success, so it's a bonding device */
1046#endif /* BOND_INFO_QUERY_IOCTL */
1047
1048	return 0;	/* no, it's not a bonding device */
1049}
1050#endif /* IW_MODE_MONITOR */
1051
1052static int pcap_protocol(pcap_t *handle)
1053{
1054	int protocol;
1055
1056	protocol = handle->opt.protocol;
1057	if (protocol == 0)
1058		protocol = ETH_P_ALL;
1059
1060	return htons(protocol);
1061}
1062
1063static int
1064pcap_can_set_rfmon_linux(pcap_t *handle)
1065{
1066#ifdef HAVE_LIBNL
1067	char phydev_path[PATH_MAX+1];
1068	int ret;
1069#endif
1070#ifdef IW_MODE_MONITOR
1071	int sock_fd;
1072	struct iwreq ireq;
1073#endif
1074
1075	if (strcmp(handle->opt.device, "any") == 0) {
1076		/*
1077		 * Monitor mode makes no sense on the "any" device.
1078		 */
1079		return 0;
1080	}
1081
1082#ifdef HAVE_LIBNL
1083	/*
1084	 * Bleah.  There doesn't seem to be a way to ask a mac80211
1085	 * device, through libnl, whether it supports monitor mode;
1086	 * we'll just check whether the device appears to be a
1087	 * mac80211 device and, if so, assume the device supports
1088	 * monitor mode.
1089	 *
1090	 * wmaster devices don't appear to support the Wireless
1091	 * Extensions, but we can create a mon device for a
1092	 * wmaster device, so we don't bother checking whether
1093	 * a mac80211 device supports the Wireless Extensions.
1094	 */
1095	ret = get_mac80211_phydev(handle, handle->opt.device, phydev_path,
1096	    PATH_MAX);
1097	if (ret < 0)
1098		return ret;	/* error */
1099	if (ret == 1)
1100		return 1;	/* mac80211 device */
1101#endif
1102
1103#ifdef IW_MODE_MONITOR
1104	/*
1105	 * Bleah.  There doesn't appear to be an ioctl to use to ask
1106	 * whether a device supports monitor mode; we'll just do
1107	 * SIOCGIWMODE and, if it succeeds, assume the device supports
1108	 * monitor mode.
1109	 *
1110	 * Open a socket on which to attempt to get the mode.
1111	 * (We assume that if we have Wireless Extensions support
1112	 * we also have PF_PACKET support.)
1113	 */
1114	sock_fd = socket(PF_PACKET, SOCK_RAW, pcap_protocol(handle));
1115	if (sock_fd == -1) {
1116		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1117		    errno, "socket");
1118		return PCAP_ERROR;
1119	}
1120
1121	if (is_bonding_device(sock_fd, handle->opt.device)) {
1122		/* It's a bonding device, so don't even try. */
1123		close(sock_fd);
1124		return 0;
1125	}
1126
1127	/*
1128	 * Attempt to get the current mode.
1129	 */
1130	strlcpy(ireq.ifr_ifrn.ifrn_name, handle->opt.device,
1131	    sizeof ireq.ifr_ifrn.ifrn_name);
1132	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
1133		/*
1134		 * Well, we got the mode; assume we can set it.
1135		 */
1136		close(sock_fd);
1137		return 1;
1138	}
1139	if (errno == ENODEV) {
1140		/* The device doesn't even exist. */
1141		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1142		    errno, "SIOCGIWMODE failed");
1143		close(sock_fd);
1144		return PCAP_ERROR_NO_SUCH_DEVICE;
1145	}
1146	close(sock_fd);
1147#endif
1148	return 0;
1149}
1150
1151/*
1152 * Grabs the number of dropped packets by the interface from /proc/net/dev.
1153 *
1154 * XXX - what about /sys/class/net/{interface name}/rx_*?  There are
1155 * individual devices giving, in ASCII, various rx_ and tx_ statistics.
1156 *
1157 * Or can we get them in binary form from netlink?
1158 */
1159static long int
1160linux_if_drops(const char * if_name)
1161{
1162	char buffer[512];
1163	char * bufptr;
1164	FILE * file;
1165	int field_to_convert = 3, if_name_sz = strlen(if_name);
1166	long int dropped_pkts = 0;
1167
1168	file = fopen("/proc/net/dev", "r");
1169	if (!file)
1170		return 0;
1171
1172	while (!dropped_pkts && fgets( buffer, sizeof(buffer), file ))
1173	{
1174		/* 	search for 'bytes' -- if its in there, then
1175			that means we need to grab the fourth field. otherwise
1176			grab the third field. */
1177		if (field_to_convert != 4 && strstr(buffer, "bytes"))
1178		{
1179			field_to_convert = 4;
1180			continue;
1181		}
1182
1183		/* find iface and make sure it actually matches -- space before the name and : after it */
1184		if ((bufptr = strstr(buffer, if_name)) &&
1185			(bufptr == buffer || *(bufptr-1) == ' ') &&
1186			*(bufptr + if_name_sz) == ':')
1187		{
1188			bufptr = bufptr + if_name_sz + 1;
1189
1190			/* grab the nth field from it */
1191			while( --field_to_convert && *bufptr != '\0')
1192			{
1193				while (*bufptr != '\0' && *(bufptr++) == ' ');
1194				while (*bufptr != '\0' && *(bufptr++) != ' ');
1195			}
1196
1197			/* get rid of any final spaces */
1198			while (*bufptr != '\0' && *bufptr == ' ') bufptr++;
1199
1200			if (*bufptr != '\0')
1201				dropped_pkts = strtol(bufptr, NULL, 10);
1202
1203			break;
1204		}
1205	}
1206
1207	fclose(file);
1208	return dropped_pkts;
1209}
1210
1211
1212/*
1213 * With older kernels promiscuous mode is kind of interesting because we
1214 * have to reset the interface before exiting. The problem can't really
1215 * be solved without some daemon taking care of managing usage counts.
1216 * If we put the interface into promiscuous mode, we set a flag indicating
1217 * that we must take it out of that mode when the interface is closed,
1218 * and, when closing the interface, if that flag is set we take it out
1219 * of promiscuous mode.
1220 *
1221 * Even with newer kernels, we have the same issue with rfmon mode.
1222 */
1223
1224static void	pcap_cleanup_linux( pcap_t *handle )
1225{
1226	struct pcap_linux *handlep = handle->priv;
1227	struct ifreq	ifr;
1228#ifdef HAVE_LIBNL
1229	struct nl80211_state nlstate;
1230	int ret;
1231#endif /* HAVE_LIBNL */
1232#ifdef IW_MODE_MONITOR
1233	int oldflags;
1234	struct iwreq ireq;
1235#endif /* IW_MODE_MONITOR */
1236
1237	if (handlep->must_do_on_close != 0) {
1238		/*
1239		 * There's something we have to do when closing this
1240		 * pcap_t.
1241		 */
1242		if (handlep->must_do_on_close & MUST_CLEAR_PROMISC) {
1243			/*
1244			 * We put the interface into promiscuous mode;
1245			 * take it out of promiscuous mode.
1246			 *
1247			 * XXX - if somebody else wants it in promiscuous
1248			 * mode, this code cannot know that, so it'll take
1249			 * it out of promiscuous mode.  That's not fixable
1250			 * in 2.0[.x] kernels.
1251			 */
1252			memset(&ifr, 0, sizeof(ifr));
1253			strlcpy(ifr.ifr_name, handlep->device,
1254			    sizeof(ifr.ifr_name));
1255			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
1256				fprintf(stderr,
1257				    "Can't restore interface %s flags (SIOCGIFFLAGS failed: %s).\n"
1258				    "Please adjust manually.\n"
1259				    "Hint: This can't happen with Linux >= 2.2.0.\n",
1260				    handlep->device, strerror(errno));
1261			} else {
1262				if (ifr.ifr_flags & IFF_PROMISC) {
1263					/*
1264					 * Promiscuous mode is currently on;
1265					 * turn it off.
1266					 */
1267					ifr.ifr_flags &= ~IFF_PROMISC;
1268					if (ioctl(handle->fd, SIOCSIFFLAGS,
1269					    &ifr) == -1) {
1270						fprintf(stderr,
1271						    "Can't restore interface %s flags (SIOCSIFFLAGS failed: %s).\n"
1272						    "Please adjust manually.\n"
1273						    "Hint: This can't happen with Linux >= 2.2.0.\n",
1274						    handlep->device,
1275						    strerror(errno));
1276					}
1277				}
1278			}
1279		}
1280
1281#ifdef HAVE_LIBNL
1282		if (handlep->must_do_on_close & MUST_DELETE_MONIF) {
1283			ret = nl80211_init(handle, &nlstate, handlep->device);
1284			if (ret >= 0) {
1285				ret = del_mon_if(handle, handle->fd, &nlstate,
1286				    handlep->device, handlep->mondevice);
1287				nl80211_cleanup(&nlstate);
1288			}
1289			if (ret < 0) {
1290				fprintf(stderr,
1291				    "Can't delete monitor interface %s (%s).\n"
1292				    "Please delete manually.\n",
1293				    handlep->mondevice, handle->errbuf);
1294			}
1295		}
1296#endif /* HAVE_LIBNL */
1297
1298#ifdef IW_MODE_MONITOR
1299		if (handlep->must_do_on_close & MUST_CLEAR_RFMON) {
1300			/*
1301			 * We put the interface into rfmon mode;
1302			 * take it out of rfmon mode.
1303			 *
1304			 * XXX - if somebody else wants it in rfmon
1305			 * mode, this code cannot know that, so it'll take
1306			 * it out of rfmon mode.
1307			 */
1308
1309			/*
1310			 * First, take the interface down if it's up;
1311			 * otherwise, we might get EBUSY.
1312			 * If we get errors, just drive on and print
1313			 * a warning if we can't restore the mode.
1314			 */
1315			oldflags = 0;
1316			memset(&ifr, 0, sizeof(ifr));
1317			strlcpy(ifr.ifr_name, handlep->device,
1318			    sizeof(ifr.ifr_name));
1319			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) != -1) {
1320				if (ifr.ifr_flags & IFF_UP) {
1321					oldflags = ifr.ifr_flags;
1322					ifr.ifr_flags &= ~IFF_UP;
1323					if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1)
1324						oldflags = 0;	/* didn't set, don't restore */
1325				}
1326			}
1327
1328			/*
1329			 * Now restore the mode.
1330			 */
1331			strlcpy(ireq.ifr_ifrn.ifrn_name, handlep->device,
1332			    sizeof ireq.ifr_ifrn.ifrn_name);
1333			ireq.u.mode = handlep->oldmode;
1334			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
1335				/*
1336				 * Scientist, you've failed.
1337				 */
1338				fprintf(stderr,
1339				    "Can't restore interface %s wireless mode (SIOCSIWMODE failed: %s).\n"
1340				    "Please adjust manually.\n",
1341				    handlep->device, strerror(errno));
1342			}
1343
1344			/*
1345			 * Now bring the interface back up if we brought
1346			 * it down.
1347			 */
1348			if (oldflags != 0) {
1349				ifr.ifr_flags = oldflags;
1350				if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
1351					fprintf(stderr,
1352					    "Can't bring interface %s back up (SIOCSIFFLAGS failed: %s).\n"
1353					    "Please adjust manually.\n",
1354					    handlep->device, strerror(errno));
1355				}
1356			}
1357		}
1358#endif /* IW_MODE_MONITOR */
1359
1360		/*
1361		 * Take this pcap out of the list of pcaps for which we
1362		 * have to take the interface out of some mode.
1363		 */
1364		pcap_remove_from_pcaps_to_close(handle);
1365	}
1366
1367	if (handlep->mondevice != NULL) {
1368		free(handlep->mondevice);
1369		handlep->mondevice = NULL;
1370	}
1371	if (handlep->device != NULL) {
1372		free(handlep->device);
1373		handlep->device = NULL;
1374	}
1375	pcap_cleanup_live_common(handle);
1376}
1377
1378/*
1379 * Set the timeout to be used in poll() with memory-mapped packet capture.
1380 */
1381static void
1382set_poll_timeout(struct pcap_linux *handlep)
1383{
1384#ifdef HAVE_TPACKET3
1385	struct utsname utsname;
1386	char *version_component, *endp;
1387	int major, minor;
1388	int broken_tpacket_v3 = 1;
1389
1390	/*
1391	 * Some versions of TPACKET_V3 have annoying bugs/misfeatures
1392	 * around which we have to work.  Determine if we have those
1393	 * problems or not.
1394	 */
1395	if (uname(&utsname) == 0) {
1396		/*
1397		 * 3.19 is the first release with a fixed version of
1398		 * TPACKET_V3.  We treat anything before that as
1399		 * not haveing a fixed version; that may really mean
1400		 * it has *no* version.
1401		 */
1402		version_component = utsname.release;
1403		major = strtol(version_component, &endp, 10);
1404		if (endp != version_component && *endp == '.') {
1405			/*
1406			 * OK, that was a valid major version.
1407			 * Get the minor version.
1408			 */
1409			version_component = endp + 1;
1410			minor = strtol(version_component, &endp, 10);
1411			if (endp != version_component &&
1412			    (*endp == '.' || *endp == '\0')) {
1413				/*
1414				 * OK, that was a valid minor version.
1415				 * Is this 3.19 or newer?
1416				 */
1417				if (major >= 4 || (major == 3 && minor >= 19)) {
1418					/* Yes. TPACKET_V3 works correctly. */
1419					broken_tpacket_v3 = 0;
1420				}
1421			}
1422		}
1423	}
1424#endif
1425	if (handlep->timeout == 0) {
1426#ifdef HAVE_TPACKET3
1427		/*
1428		 * XXX - due to a set of (mis)features in the TPACKET_V3
1429		 * kernel code prior to the 3.19 kernel, blocking forever
1430		 * with a TPACKET_V3 socket can, if few packets are
1431		 * arriving and passing the socket filter, cause most
1432		 * packets to be dropped.  See libpcap issue #335 for the
1433		 * full painful story.
1434		 *
1435		 * The workaround is to have poll() time out very quickly,
1436		 * so we grab the frames handed to us, and return them to
1437		 * the kernel, ASAP.
1438		 */
1439		if (handlep->tp_version == TPACKET_V3 && broken_tpacket_v3)
1440			handlep->poll_timeout = 1;	/* don't block for very long */
1441		else
1442#endif
1443			handlep->poll_timeout = -1;	/* block forever */
1444	} else if (handlep->timeout > 0) {
1445#ifdef HAVE_TPACKET3
1446		/*
1447		 * For TPACKET_V3, the timeout is handled by the kernel,
1448		 * so block forever; that way, we don't get extra timeouts.
1449		 * Don't do that if we have a broken TPACKET_V3, though.
1450		 */
1451		if (handlep->tp_version == TPACKET_V3 && !broken_tpacket_v3)
1452			handlep->poll_timeout = -1;	/* block forever, let TPACKET_V3 wake us up */
1453		else
1454#endif
1455			handlep->poll_timeout = handlep->timeout;	/* block for that amount of time */
1456	} else {
1457		/*
1458		 * Non-blocking mode; we call poll() to pick up error
1459		 * indications, but we don't want it to wait for
1460		 * anything.
1461		 */
1462		handlep->poll_timeout = 0;
1463	}
1464}
1465
1466/*
1467 *  Get a handle for a live capture from the given device. You can
1468 *  pass NULL as device to get all packages (without link level
1469 *  information of course). If you pass 1 as promisc the interface
1470 *  will be set to promiscous mode (XXX: I think this usage should
1471 *  be deprecated and functions be added to select that later allow
1472 *  modification of that values -- Torsten).
1473 */
1474static int
1475pcap_activate_linux(pcap_t *handle)
1476{
1477	struct pcap_linux *handlep = handle->priv;
1478	const char	*device;
1479	struct ifreq	ifr;
1480	int		status = 0;
1481	int		ret;
1482
1483	device = handle->opt.device;
1484
1485	/*
1486	 * Make sure the name we were handed will fit into the ioctls we
1487	 * might perform on the device; if not, return a "No such device"
1488	 * indication, as the Linux kernel shouldn't support creating
1489	 * a device whose name won't fit into those ioctls.
1490	 *
1491	 * "Will fit" means "will fit, complete with a null terminator",
1492	 * so if the length, which does *not* include the null terminator,
1493	 * is greater than *or equal to* the size of the field into which
1494	 * we'll be copying it, that won't fit.
1495	 */
1496	if (strlen(device) >= sizeof(ifr.ifr_name)) {
1497		status = PCAP_ERROR_NO_SUCH_DEVICE;
1498		goto fail;
1499	}
1500
1501	/*
1502	 * Turn a negative snapshot value (invalid), a snapshot value of
1503	 * 0 (unspecified), or a value bigger than the normal maximum
1504	 * value, into the maximum allowed value.
1505	 *
1506	 * If some application really *needs* a bigger snapshot
1507	 * length, we should just increase MAXIMUM_SNAPLEN.
1508	 */
1509	if (handle->snapshot <= 0 || handle->snapshot > MAXIMUM_SNAPLEN)
1510		handle->snapshot = MAXIMUM_SNAPLEN;
1511
1512	handle->inject_op = pcap_inject_linux;
1513	handle->setfilter_op = pcap_setfilter_linux;
1514	handle->setdirection_op = pcap_setdirection_linux;
1515	handle->set_datalink_op = pcap_set_datalink_linux;
1516	handle->getnonblock_op = pcap_getnonblock_fd;
1517	handle->setnonblock_op = pcap_setnonblock_fd;
1518	handle->cleanup_op = pcap_cleanup_linux;
1519	handle->read_op = pcap_read_linux;
1520	handle->stats_op = pcap_stats_linux;
1521
1522	/*
1523	 * The "any" device is a special device which causes us not
1524	 * to bind to a particular device and thus to look at all
1525	 * devices.
1526	 */
1527	if (strcmp(device, "any") == 0) {
1528		if (handle->opt.promisc) {
1529			handle->opt.promisc = 0;
1530			/* Just a warning. */
1531			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1532			    "Promiscuous mode not supported on the \"any\" device");
1533			status = PCAP_WARNING_PROMISC_NOTSUP;
1534		}
1535	}
1536
1537	handlep->device	= strdup(device);
1538	if (handlep->device == NULL) {
1539		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1540		    errno, "strdup");
1541		return PCAP_ERROR;
1542	}
1543
1544	/* copy timeout value */
1545	handlep->timeout = handle->opt.timeout;
1546
1547	/*
1548	 * If we're in promiscuous mode, then we probably want
1549	 * to see when the interface drops packets too, so get an
1550	 * initial count from /proc/net/dev
1551	 */
1552	if (handle->opt.promisc)
1553		handlep->proc_dropped = linux_if_drops(handlep->device);
1554
1555	/*
1556	 * Current Linux kernels use the protocol family PF_PACKET to
1557	 * allow direct access to all packets on the network while
1558	 * older kernels had a special socket type SOCK_PACKET to
1559	 * implement this feature.
1560	 * While this old implementation is kind of obsolete we need
1561	 * to be compatible with older kernels for a while so we are
1562	 * trying both methods with the newer method preferred.
1563	 */
1564	ret = activate_new(handle);
1565	if (ret < 0) {
1566		/*
1567		 * Fatal error with the new way; just fail.
1568		 * ret has the error return; if it's PCAP_ERROR,
1569		 * handle->errbuf has been set appropriately.
1570		 */
1571		status = ret;
1572		goto fail;
1573	}
1574	if (ret == 1) {
1575		/*
1576		 * Success.
1577		 * Try to use memory-mapped access.
1578		 */
1579		switch (activate_mmap(handle, &status)) {
1580
1581		case 1:
1582			/*
1583			 * We succeeded.  status has been
1584			 * set to the status to return,
1585			 * which might be 0, or might be
1586			 * a PCAP_WARNING_ value.
1587			 *
1588			 * Set the timeout to use in poll() before
1589			 * returning.
1590			 */
1591			set_poll_timeout(handlep);
1592			return status;
1593
1594		case 0:
1595			/*
1596			 * Kernel doesn't support it - just continue
1597			 * with non-memory-mapped access.
1598			 */
1599			break;
1600
1601		case -1:
1602			/*
1603			 * We failed to set up to use it, or the kernel
1604			 * supports it, but we failed to enable it.
1605			 * ret has been set to the error status to
1606			 * return and, if it's PCAP_ERROR, handle->errbuf
1607			 * contains the error message.
1608			 */
1609			status = ret;
1610			goto fail;
1611		}
1612	}
1613	else if (ret == 0) {
1614		/* Non-fatal error; try old way */
1615		if ((ret = activate_old(handle)) != 1) {
1616			/*
1617			 * Both methods to open the packet socket failed.
1618			 * Tidy up and report our failure (handle->errbuf
1619			 * is expected to be set by the functions above).
1620			 */
1621			status = ret;
1622			goto fail;
1623		}
1624	}
1625
1626	/*
1627	 * We set up the socket, but not with memory-mapped access.
1628	 */
1629	if (handle->opt.buffer_size != 0) {
1630		/*
1631		 * Set the socket buffer size to the specified value.
1632		 */
1633		if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF,
1634		    &handle->opt.buffer_size,
1635		    sizeof(handle->opt.buffer_size)) == -1) {
1636			pcap_fmt_errmsg_for_errno(handle->errbuf,
1637			    PCAP_ERRBUF_SIZE, errno, "SO_RCVBUF");
1638			status = PCAP_ERROR;
1639			goto fail;
1640		}
1641	}
1642
1643	/* Allocate the buffer */
1644
1645	handle->buffer	 = malloc(handle->bufsize + handle->offset);
1646	if (!handle->buffer) {
1647		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1648		    errno, "malloc");
1649		status = PCAP_ERROR;
1650		goto fail;
1651	}
1652
1653	/*
1654	 * "handle->fd" is a socket, so "select()" and "poll()"
1655	 * should work on it.
1656	 */
1657	handle->selectable_fd = handle->fd;
1658
1659	return status;
1660
1661fail:
1662	pcap_cleanup_linux(handle);
1663	return status;
1664}
1665
1666/*
1667 *  Read at most max_packets from the capture stream and call the callback
1668 *  for each of them. Returns the number of packets handled or -1 if an
1669 *  error occured.
1670 */
1671static int
1672pcap_read_linux(pcap_t *handle, int max_packets _U_, pcap_handler callback, u_char *user)
1673{
1674	/*
1675	 * Currently, on Linux only one packet is delivered per read,
1676	 * so we don't loop.
1677	 */
1678	return pcap_read_packet(handle, callback, user);
1679}
1680
1681static int
1682pcap_set_datalink_linux(pcap_t *handle, int dlt)
1683{
1684	handle->linktype = dlt;
1685	return 0;
1686}
1687
1688/*
1689 * linux_check_direction()
1690 *
1691 * Do checks based on packet direction.
1692 */
1693static inline int
1694linux_check_direction(const pcap_t *handle, const struct sockaddr_ll *sll)
1695{
1696	struct pcap_linux	*handlep = handle->priv;
1697
1698	if (sll->sll_pkttype == PACKET_OUTGOING) {
1699		/*
1700		 * Outgoing packet.
1701		 * If this is from the loopback device, reject it;
1702		 * we'll see the packet as an incoming packet as well,
1703		 * and we don't want to see it twice.
1704		 */
1705		if (sll->sll_ifindex == handlep->lo_ifindex)
1706			return 0;
1707
1708		/*
1709		 * If this is an outgoing CAN or CAN FD frame, and
1710		 * the user doesn't only want outgoing packets,
1711		 * reject it; CAN devices and drivers, and the CAN
1712		 * stack, always arrange to loop back transmitted
1713		 * packets, so they also appear as incoming packets.
1714		 * We don't want duplicate packets, and we can't
1715		 * easily distinguish packets looped back by the CAN
1716		 * layer than those received by the CAN layer, so we
1717		 * eliminate this packet instead.
1718		 */
1719		if ((sll->sll_protocol == LINUX_SLL_P_CAN ||
1720		     sll->sll_protocol == LINUX_SLL_P_CANFD) &&
1721		     handle->direction != PCAP_D_OUT)
1722			return 0;
1723
1724		/*
1725		 * If the user only wants incoming packets, reject it.
1726		 */
1727		if (handle->direction == PCAP_D_IN)
1728			return 0;
1729	} else {
1730		/*
1731		 * Incoming packet.
1732		 * If the user only wants outgoing packets, reject it.
1733		 */
1734		if (handle->direction == PCAP_D_OUT)
1735			return 0;
1736	}
1737	return 1;
1738}
1739
1740/*
1741 *  Read a packet from the socket calling the handler provided by
1742 *  the user. Returns the number of packets received or -1 if an
1743 *  error occured.
1744 */
1745static int
1746pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
1747{
1748	struct pcap_linux	*handlep = handle->priv;
1749	u_char			*bp;
1750	int			offset;
1751#ifdef HAVE_PF_PACKET_SOCKETS
1752	struct sockaddr_ll	from;
1753	struct sll_header	*hdrp;
1754#else
1755	struct sockaddr		from;
1756#endif
1757#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI)
1758	struct iovec		iov;
1759	struct msghdr		msg;
1760	struct cmsghdr		*cmsg;
1761	union {
1762		struct cmsghdr	cmsg;
1763		char		buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
1764	} cmsg_buf;
1765#else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI) */
1766	socklen_t		fromlen;
1767#endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI) */
1768	int			packet_len, caplen;
1769	struct pcap_pkthdr	pcap_header;
1770
1771        struct bpf_aux_data     aux_data;
1772#ifdef HAVE_PF_PACKET_SOCKETS
1773	/*
1774	 * If this is a cooked device, leave extra room for a
1775	 * fake packet header.
1776	 */
1777	if (handlep->cooked)
1778		offset = SLL_HDR_LEN;
1779	else
1780		offset = 0;
1781#else
1782	/*
1783	 * This system doesn't have PF_PACKET sockets, so it doesn't
1784	 * support cooked devices.
1785	 */
1786	offset = 0;
1787#endif
1788
1789	/*
1790	 * Receive a single packet from the kernel.
1791	 * We ignore EINTR, as that might just be due to a signal
1792	 * being delivered - if the signal should interrupt the
1793	 * loop, the signal handler should call pcap_breakloop()
1794	 * to set handle->break_loop (we ignore it on other
1795	 * platforms as well).
1796	 * We also ignore ENETDOWN, so that we can continue to
1797	 * capture traffic if the interface goes down and comes
1798	 * back up again; comments in the kernel indicate that
1799	 * we'll just block waiting for packets if we try to
1800	 * receive from a socket that delivered ENETDOWN, and,
1801	 * if we're using a memory-mapped buffer, we won't even
1802	 * get notified of "network down" events.
1803	 */
1804	bp = (u_char *)handle->buffer + handle->offset;
1805
1806#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI)
1807	msg.msg_name		= &from;
1808	msg.msg_namelen		= sizeof(from);
1809	msg.msg_iov		= &iov;
1810	msg.msg_iovlen		= 1;
1811	msg.msg_control		= &cmsg_buf;
1812	msg.msg_controllen	= sizeof(cmsg_buf);
1813	msg.msg_flags		= 0;
1814
1815	iov.iov_len		= handle->bufsize - offset;
1816	iov.iov_base		= bp + offset;
1817#endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI) */
1818
1819	do {
1820		/*
1821		 * Has "pcap_breakloop()" been called?
1822		 */
1823		if (handle->break_loop) {
1824			/*
1825			 * Yes - clear the flag that indicates that it has,
1826			 * and return PCAP_ERROR_BREAK as an indication that
1827			 * we were told to break out of the loop.
1828			 */
1829			handle->break_loop = 0;
1830			return PCAP_ERROR_BREAK;
1831		}
1832
1833#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI)
1834		packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC);
1835#else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI) */
1836		fromlen = sizeof(from);
1837		packet_len = recvfrom(
1838			handle->fd, bp + offset,
1839			handle->bufsize - offset, MSG_TRUNC,
1840			(struct sockaddr *) &from, &fromlen);
1841#endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI) */
1842	} while (packet_len == -1 && errno == EINTR);
1843
1844	/* Check if an error occured */
1845
1846	if (packet_len == -1) {
1847		switch (errno) {
1848
1849		case EAGAIN:
1850			return 0;	/* no packet there */
1851
1852		case ENETDOWN:
1853			/*
1854			 * The device on which we're capturing went away.
1855			 *
1856			 * XXX - we should really return
1857			 * PCAP_ERROR_IFACE_NOT_UP, but pcap_dispatch()
1858			 * etc. aren't defined to return that.
1859			 */
1860			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1861				"The interface went down");
1862			return PCAP_ERROR;
1863
1864		default:
1865			pcap_fmt_errmsg_for_errno(handle->errbuf,
1866			    PCAP_ERRBUF_SIZE, errno, "recvfrom");
1867			return PCAP_ERROR;
1868		}
1869	}
1870
1871#ifdef HAVE_PF_PACKET_SOCKETS
1872	if (!handlep->sock_packet) {
1873		/*
1874		 * Unfortunately, there is a window between socket() and
1875		 * bind() where the kernel may queue packets from any
1876		 * interface.  If we're bound to a particular interface,
1877		 * discard packets not from that interface.
1878		 *
1879		 * (If socket filters are supported, we could do the
1880		 * same thing we do when changing the filter; however,
1881		 * that won't handle packet sockets without socket
1882		 * filter support, and it's a bit more complicated.
1883		 * It would save some instructions per packet, however.)
1884		 */
1885		if (handlep->ifindex != -1 &&
1886		    from.sll_ifindex != handlep->ifindex)
1887			return 0;
1888
1889		/*
1890		 * Do checks based on packet direction.
1891		 * We can only do this if we're using PF_PACKET; the
1892		 * address returned for SOCK_PACKET is a "sockaddr_pkt"
1893		 * which lacks the relevant packet type information.
1894		 */
1895		if (!linux_check_direction(handle, &from))
1896			return 0;
1897	}
1898#endif
1899
1900#ifdef HAVE_PF_PACKET_SOCKETS
1901	/*
1902	 * If this is a cooked device, fill in the fake packet header.
1903	 */
1904	if (handlep->cooked) {
1905		/*
1906		 * Add the length of the fake header to the length
1907		 * of packet data we read.
1908		 */
1909		packet_len += SLL_HDR_LEN;
1910
1911		hdrp = (struct sll_header *)bp;
1912		hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype);
1913		hdrp->sll_hatype = htons(from.sll_hatype);
1914		hdrp->sll_halen = htons(from.sll_halen);
1915		memcpy(hdrp->sll_addr, from.sll_addr,
1916		    (from.sll_halen > SLL_ADDRLEN) ?
1917		      SLL_ADDRLEN :
1918		      from.sll_halen);
1919		hdrp->sll_protocol = from.sll_protocol;
1920	}
1921
1922	/*
1923	 * Start out with no VLAN information.
1924	 */
1925	aux_data.vlan_tag_present = 0;
1926	aux_data.vlan_tag = 0;
1927#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI)
1928	if (handlep->vlan_offset != -1) {
1929		for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
1930			struct tpacket_auxdata *aux;
1931			unsigned int len;
1932			struct vlan_tag *tag;
1933
1934			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
1935			    cmsg->cmsg_level != SOL_PACKET ||
1936			    cmsg->cmsg_type != PACKET_AUXDATA) {
1937				/*
1938				 * This isn't a PACKET_AUXDATA auxiliary
1939				 * data item.
1940				 */
1941				continue;
1942			}
1943
1944			aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
1945			if (!VLAN_VALID(aux, aux)) {
1946				/*
1947				 * There is no VLAN information in the
1948				 * auxiliary data.
1949				 */
1950				continue;
1951			}
1952
1953			len = (u_int)packet_len > iov.iov_len ? iov.iov_len : (u_int)packet_len;
1954			if (len < (u_int)handlep->vlan_offset)
1955				break;
1956
1957			/*
1958			 * Move everything in the header, except the
1959			 * type field, down VLAN_TAG_LEN bytes, to
1960			 * allow us to insert the VLAN tag between
1961			 * that stuff and the type field.
1962			 */
1963			bp -= VLAN_TAG_LEN;
1964			memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
1965
1966			/*
1967			 * Now insert the tag.
1968			 */
1969			tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
1970			tag->vlan_tpid = htons(VLAN_TPID(aux, aux));
1971			tag->vlan_tci = htons(aux->tp_vlan_tci);
1972
1973			/*
1974			 * Save a flag indicating that we have a VLAN tag,
1975			 * and the VLAN TCI, to bpf_aux_data struct for
1976			 * use by the BPF filter if we're doing the
1977			 * filtering in userland.
1978			 */
1979			aux_data.vlan_tag_present = 1;
1980			aux_data.vlan_tag = htons(aux->tp_vlan_tci) & 0x0fff;
1981
1982			/*
1983			 * Add the tag to the packet lengths.
1984			 */
1985			packet_len += VLAN_TAG_LEN;
1986		}
1987	}
1988#endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_STRUCT_TPACKET_AUXDATA_TP_VLAN_TCI) */
1989#endif /* HAVE_PF_PACKET_SOCKETS */
1990
1991	/*
1992	 * XXX: According to the kernel source we should get the real
1993	 * packet len if calling recvfrom with MSG_TRUNC set. It does
1994	 * not seem to work here :(, but it is supported by this code
1995	 * anyway.
1996	 * To be honest the code RELIES on that feature so this is really
1997	 * broken with 2.2.x kernels.
1998	 * I spend a day to figure out what's going on and I found out
1999	 * that the following is happening:
2000	 *
2001	 * The packet comes from a random interface and the packet_rcv
2002	 * hook is called with a clone of the packet. That code inserts
2003	 * the packet into the receive queue of the packet socket.
2004	 * If a filter is attached to that socket that filter is run
2005	 * first - and there lies the problem. The default filter always
2006	 * cuts the packet at the snaplen:
2007	 *
2008	 * # tcpdump -d
2009	 * (000) ret      #68
2010	 *
2011	 * So the packet filter cuts down the packet. The recvfrom call
2012	 * says "hey, it's only 68 bytes, it fits into the buffer" with
2013	 * the result that we don't get the real packet length. This
2014	 * is valid at least until kernel 2.2.17pre6.
2015	 *
2016	 * We currently handle this by making a copy of the filter
2017	 * program, fixing all "ret" instructions with non-zero
2018	 * operands to have an operand of MAXIMUM_SNAPLEN so that the
2019	 * filter doesn't truncate the packet, and supplying that modified
2020	 * filter to the kernel.
2021	 */
2022
2023	caplen = packet_len;
2024	if (caplen > handle->snapshot)
2025		caplen = handle->snapshot;
2026
2027	/* Run the packet filter if not using kernel filter */
2028	if (handlep->filter_in_userland && handle->fcode.bf_insns) {
2029		if (bpf_filter_with_aux_data(handle->fcode.bf_insns, bp,
2030		    packet_len, caplen, &aux_data) == 0) {
2031			/* rejected by filter */
2032			return 0;
2033		}
2034	}
2035
2036	/* Fill in our own header data */
2037
2038	/* get timestamp for this packet */
2039#if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
2040	if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
2041		if (ioctl(handle->fd, SIOCGSTAMPNS, &pcap_header.ts) == -1) {
2042			pcap_fmt_errmsg_for_errno(handle->errbuf,
2043			    PCAP_ERRBUF_SIZE, errno, "SIOCGSTAMPNS");
2044			return PCAP_ERROR;
2045		}
2046        } else
2047#endif
2048	{
2049		if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
2050			pcap_fmt_errmsg_for_errno(handle->errbuf,
2051			    PCAP_ERRBUF_SIZE, errno, "SIOCGSTAMP");
2052			return PCAP_ERROR;
2053		}
2054        }
2055
2056	pcap_header.caplen	= caplen;
2057	pcap_header.len		= packet_len;
2058
2059	/*
2060	 * Count the packet.
2061	 *
2062	 * Arguably, we should count them before we check the filter,
2063	 * as on many other platforms "ps_recv" counts packets
2064	 * handed to the filter rather than packets that passed
2065	 * the filter, but if filtering is done in the kernel, we
2066	 * can't get a count of packets that passed the filter,
2067	 * and that would mean the meaning of "ps_recv" wouldn't
2068	 * be the same on all Linux systems.
2069	 *
2070	 * XXX - it's not the same on all systems in any case;
2071	 * ideally, we should have a "get the statistics" call
2072	 * that supplies more counts and indicates which of them
2073	 * it supplies, so that we supply a count of packets
2074	 * handed to the filter only on platforms where that
2075	 * information is available.
2076	 *
2077	 * We count them here even if we can get the packet count
2078	 * from the kernel, as we can only determine at run time
2079	 * whether we'll be able to get it from the kernel (if
2080	 * HAVE_STRUCT_TPACKET_STATS isn't defined, we can't get it from
2081	 * the kernel, but if it is defined, the library might
2082	 * have been built with a 2.4 or later kernel, but we
2083	 * might be running on a 2.2[.x] kernel without Alexey
2084	 * Kuznetzov's turbopacket patches, and thus the kernel
2085	 * might not be able to supply those statistics).  We
2086	 * could, I guess, try, when opening the socket, to get
2087	 * the statistics, and if we can not increment the count
2088	 * here, but it's not clear that always incrementing
2089	 * the count is more expensive than always testing a flag
2090	 * in memory.
2091	 *
2092	 * We keep the count in "handlep->packets_read", and use that
2093	 * for "ps_recv" if we can't get the statistics from the kernel.
2094	 * We do that because, if we *can* get the statistics from
2095	 * the kernel, we use "handlep->stat.ps_recv" and
2096	 * "handlep->stat.ps_drop" as running counts, as reading the
2097	 * statistics from the kernel resets the kernel statistics,
2098	 * and if we directly increment "handlep->stat.ps_recv" here,
2099	 * that means it will count packets *twice* on systems where
2100	 * we can get kernel statistics - once here, and once in
2101	 * pcap_stats_linux().
2102	 */
2103	handlep->packets_read++;
2104
2105	/* Call the user supplied callback function */
2106	callback(userdata, &pcap_header, bp);
2107
2108	return 1;
2109}
2110
2111static int
2112pcap_inject_linux(pcap_t *handle, const void *buf, size_t size)
2113{
2114	struct pcap_linux *handlep = handle->priv;
2115	int ret;
2116
2117#ifdef HAVE_PF_PACKET_SOCKETS
2118	if (!handlep->sock_packet) {
2119		/* PF_PACKET socket */
2120		if (handlep->ifindex == -1) {
2121			/*
2122			 * We don't support sending on the "any" device.
2123			 */
2124			strlcpy(handle->errbuf,
2125			    "Sending packets isn't supported on the \"any\" device",
2126			    PCAP_ERRBUF_SIZE);
2127			return (-1);
2128		}
2129
2130		if (handlep->cooked) {
2131			/*
2132			 * We don't support sending on the "any" device.
2133			 *
2134			 * XXX - how do you send on a bound cooked-mode
2135			 * socket?
2136			 * Is a "sendto()" required there?
2137			 */
2138			strlcpy(handle->errbuf,
2139			    "Sending packets isn't supported in cooked mode",
2140			    PCAP_ERRBUF_SIZE);
2141			return (-1);
2142		}
2143	}
2144#endif
2145
2146	ret = send(handle->fd, buf, size, 0);
2147	if (ret == -1) {
2148		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2149		    errno, "send");
2150		return (-1);
2151	}
2152	return (ret);
2153}
2154
2155/*
2156 *  Get the statistics for the given packet capture handle.
2157 *  Reports the number of dropped packets iff the kernel supports
2158 *  the PACKET_STATISTICS "getsockopt()" argument (2.4 and later
2159 *  kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket
2160 *  patches); otherwise, that information isn't available, and we lie
2161 *  and report 0 as the count of dropped packets.
2162 */
2163static int
2164pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
2165{
2166	struct pcap_linux *handlep = handle->priv;
2167#ifdef HAVE_STRUCT_TPACKET_STATS
2168#ifdef HAVE_TPACKET3
2169	/*
2170	 * For sockets using TPACKET_V1 or TPACKET_V2, the extra
2171	 * stuff at the end of a struct tpacket_stats_v3 will not
2172	 * be filled in, and we don't look at it so this is OK even
2173	 * for those sockets.  In addition, the PF_PACKET socket
2174	 * code in the kernel only uses the length parameter to
2175	 * compute how much data to copy out and to indicate how
2176	 * much data was copied out, so it's OK to base it on the
2177	 * size of a struct tpacket_stats.
2178	 *
2179	 * XXX - it's probably OK, in fact, to just use a
2180	 * struct tpacket_stats for V3 sockets, as we don't
2181	 * care about the tp_freeze_q_cnt stat.
2182	 */
2183	struct tpacket_stats_v3 kstats;
2184#else /* HAVE_TPACKET3 */
2185	struct tpacket_stats kstats;
2186#endif /* HAVE_TPACKET3 */
2187	socklen_t len = sizeof (struct tpacket_stats);
2188#endif /* HAVE_STRUCT_TPACKET_STATS */
2189
2190	long if_dropped = 0;
2191
2192	/*
2193	 *	To fill in ps_ifdrop, we parse /proc/net/dev for the number
2194	 */
2195	if (handle->opt.promisc)
2196	{
2197		if_dropped = handlep->proc_dropped;
2198		handlep->proc_dropped = linux_if_drops(handlep->device);
2199		handlep->stat.ps_ifdrop += (handlep->proc_dropped - if_dropped);
2200	}
2201
2202#ifdef HAVE_STRUCT_TPACKET_STATS
2203	/*
2204	 * Try to get the packet counts from the kernel.
2205	 */
2206	if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
2207			&kstats, &len) > -1) {
2208		/*
2209		 * On systems where the PACKET_STATISTICS "getsockopt()"
2210		 * argument is supported on PF_PACKET sockets:
2211		 *
2212		 *	"ps_recv" counts only packets that *passed* the
2213		 *	filter, not packets that didn't pass the filter.
2214		 *	This includes packets later dropped because we
2215		 *	ran out of buffer space.
2216		 *
2217		 *	"ps_drop" counts packets dropped because we ran
2218		 *	out of buffer space.  It doesn't count packets
2219		 *	dropped by the interface driver.  It counts only
2220		 *	packets that passed the filter.
2221		 *
2222		 *	See above for ps_ifdrop.
2223		 *
2224		 *	Both statistics include packets not yet read from
2225		 *	the kernel by libpcap, and thus not yet seen by
2226		 *	the application.
2227		 *
2228		 * In "linux/net/packet/af_packet.c", at least in the
2229		 * 2.4.9 kernel, "tp_packets" is incremented for every
2230		 * packet that passes the packet filter *and* is
2231		 * successfully queued on the socket; "tp_drops" is
2232		 * incremented for every packet dropped because there's
2233		 * not enough free space in the socket buffer.
2234		 *
2235		 * When the statistics are returned for a PACKET_STATISTICS
2236		 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
2237		 * so that "tp_packets" counts all packets handed to
2238		 * the PF_PACKET socket, including packets dropped because
2239		 * there wasn't room on the socket buffer - but not
2240		 * including packets that didn't pass the filter.
2241		 *
2242		 * In the BSD BPF, the count of received packets is
2243		 * incremented for every packet handed to BPF, regardless
2244		 * of whether it passed the filter.
2245		 *
2246		 * We can't make "pcap_stats()" work the same on both
2247		 * platforms, but the best approximation is to return
2248		 * "tp_packets" as the count of packets and "tp_drops"
2249		 * as the count of drops.
2250		 *
2251		 * Keep a running total because each call to
2252		 *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
2253		 * resets the counters to zero.
2254		 */
2255		handlep->stat.ps_recv += kstats.tp_packets;
2256		handlep->stat.ps_drop += kstats.tp_drops;
2257		*stats = handlep->stat;
2258		return 0;
2259	}
2260	else
2261	{
2262		/*
2263		 * If the error was EOPNOTSUPP, fall through, so that
2264		 * if you build the library on a system with
2265		 * "struct tpacket_stats" and run it on a system
2266		 * that doesn't, it works as it does if the library
2267		 * is built on a system without "struct tpacket_stats".
2268		 */
2269		if (errno != EOPNOTSUPP) {
2270			pcap_fmt_errmsg_for_errno(handle->errbuf,
2271			    PCAP_ERRBUF_SIZE, errno, "pcap_stats");
2272			return -1;
2273		}
2274	}
2275#endif
2276	/*
2277	 * On systems where the PACKET_STATISTICS "getsockopt()" argument
2278	 * is not supported on PF_PACKET sockets:
2279	 *
2280	 *	"ps_recv" counts only packets that *passed* the filter,
2281	 *	not packets that didn't pass the filter.  It does not
2282	 *	count packets dropped because we ran out of buffer
2283	 *	space.
2284	 *
2285	 *	"ps_drop" is not supported.
2286	 *
2287	 *	"ps_ifdrop" is supported. It will return the number
2288	 *	of drops the interface reports in /proc/net/dev,
2289	 *	if that is available.
2290	 *
2291	 *	"ps_recv" doesn't include packets not yet read from
2292	 *	the kernel by libpcap.
2293	 *
2294	 * We maintain the count of packets processed by libpcap in
2295	 * "handlep->packets_read", for reasons described in the comment
2296	 * at the end of pcap_read_packet().  We have no idea how many
2297	 * packets were dropped by the kernel buffers -- but we know
2298	 * how many the interface dropped, so we can return that.
2299	 */
2300
2301	stats->ps_recv = handlep->packets_read;
2302	stats->ps_drop = 0;
2303	stats->ps_ifdrop = handlep->stat.ps_ifdrop;
2304	return 0;
2305}
2306
2307static int
2308add_linux_if(pcap_if_list_t *devlistp, const char *ifname, int fd, char *errbuf)
2309{
2310	const char *p;
2311	char name[512];	/* XXX - pick a size */
2312	char *q, *saveq;
2313	struct ifreq ifrflags;
2314
2315	/*
2316	 * Get the interface name.
2317	 */
2318	p = ifname;
2319	q = &name[0];
2320	while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2321		if (*p == ':') {
2322			/*
2323			 * This could be the separator between a
2324			 * name and an alias number, or it could be
2325			 * the separator between a name with no
2326			 * alias number and the next field.
2327			 *
2328			 * If there's a colon after digits, it
2329			 * separates the name and the alias number,
2330			 * otherwise it separates the name and the
2331			 * next field.
2332			 */
2333			saveq = q;
2334			while (isascii(*p) && isdigit(*p))
2335				*q++ = *p++;
2336			if (*p != ':') {
2337				/*
2338				 * That was the next field,
2339				 * not the alias number.
2340				 */
2341				q = saveq;
2342			}
2343			break;
2344		} else
2345			*q++ = *p++;
2346	}
2347	*q = '\0';
2348
2349	/*
2350	 * Get the flags for this interface.
2351	 */
2352	strlcpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2353	if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2354		if (errno == ENXIO || errno == ENODEV)
2355			return (0);	/* device doesn't actually exist - ignore it */
2356		pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
2357		    errno, "SIOCGIFFLAGS: %.*s",
2358		    (int)sizeof(ifrflags.ifr_name),
2359		    ifrflags.ifr_name);
2360		return (-1);
2361	}
2362
2363	/*
2364	 * Add an entry for this interface, with no addresses, if it's
2365	 * not already in the list.
2366	 */
2367	if (find_or_add_if(devlistp, name, ifrflags.ifr_flags,
2368	    get_if_flags, errbuf) == NULL) {
2369		/*
2370		 * Failure.
2371		 */
2372		return (-1);
2373	}
2374
2375	return (0);
2376}
2377
2378/*
2379 * Get from "/sys/class/net" all interfaces listed there; if they're
2380 * already in the list of interfaces we have, that won't add another
2381 * instance, but if they're not, that'll add them.
2382 *
2383 * We don't bother getting any addresses for them; it appears you can't
2384 * use SIOCGIFADDR on Linux to get IPv6 addresses for interfaces, and,
2385 * although some other types of addresses can be fetched with SIOCGIFADDR,
2386 * we don't bother with them for now.
2387 *
2388 * We also don't fail if we couldn't open "/sys/class/net"; we just leave
2389 * the list of interfaces as is, and return 0, so that we can try
2390 * scanning /proc/net/dev.
2391 *
2392 * Otherwise, we return 1 if we don't get an error and -1 if we do.
2393 */
2394static int
2395scan_sys_class_net(pcap_if_list_t *devlistp, char *errbuf)
2396{
2397	DIR *sys_class_net_d;
2398	int fd;
2399	struct dirent *ent;
2400	char subsystem_path[PATH_MAX+1];
2401	struct stat statb;
2402	int ret = 1;
2403
2404	sys_class_net_d = opendir("/sys/class/net");
2405	if (sys_class_net_d == NULL) {
2406		/*
2407		 * Don't fail if it doesn't exist at all.
2408		 */
2409		if (errno == ENOENT)
2410			return (0);
2411
2412		/*
2413		 * Fail if we got some other error.
2414		 */
2415		pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
2416		    errno, "Can't open /sys/class/net");
2417		return (-1);
2418	}
2419
2420	/*
2421	 * Create a socket from which to fetch interface information.
2422	 */
2423	fd = socket(PF_UNIX, SOCK_RAW, 0);
2424	if (fd < 0) {
2425		pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
2426		    errno, "socket");
2427		(void)closedir(sys_class_net_d);
2428		return (-1);
2429	}
2430
2431	for (;;) {
2432		errno = 0;
2433		ent = readdir(sys_class_net_d);
2434		if (ent == NULL) {
2435			/*
2436			 * Error or EOF; if errno != 0, it's an error.
2437			 */
2438			break;
2439		}
2440
2441		/*
2442		 * Ignore "." and "..".
2443		 */
2444		if (strcmp(ent->d_name, ".") == 0 ||
2445		    strcmp(ent->d_name, "..") == 0)
2446			continue;
2447
2448		/*
2449		 * Ignore plain files; they do not have subdirectories
2450		 * and thus have no attributes.
2451		 */
2452		if (ent->d_type == DT_REG)
2453			continue;
2454
2455		/*
2456		 * Is there an "ifindex" file under that name?
2457		 * (We don't care whether it's a directory or
2458		 * a symlink; older kernels have directories
2459		 * for devices, newer kernels have symlinks to
2460		 * directories.)
2461		 */
2462		pcap_snprintf(subsystem_path, sizeof subsystem_path,
2463		    "/sys/class/net/%s/ifindex", ent->d_name);
2464		if (lstat(subsystem_path, &statb) != 0) {
2465			/*
2466			 * Stat failed.  Either there was an error
2467			 * other than ENOENT, and we don't know if
2468			 * this is an interface, or it's ENOENT,
2469			 * and either some part of "/sys/class/net/{if}"
2470			 * disappeared, in which case it probably means
2471			 * the interface disappeared, or there's no
2472			 * "ifindex" file, which means it's not a
2473			 * network interface.
2474			 */
2475			continue;
2476		}
2477
2478		/*
2479		 * Attempt to add the interface.
2480		 */
2481		if (add_linux_if(devlistp, &ent->d_name[0], fd, errbuf) == -1) {
2482			/* Fail. */
2483			ret = -1;
2484			break;
2485		}
2486	}
2487	if (ret != -1) {
2488		/*
2489		 * Well, we didn't fail for any other reason; did we
2490		 * fail due to an error reading the directory?
2491		 */
2492		if (errno != 0) {
2493			pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
2494			    errno, "Error reading /sys/class/net");
2495			ret = -1;
2496		}
2497	}
2498
2499	(void)close(fd);
2500	(void)closedir(sys_class_net_d);
2501	return (ret);
2502}
2503
2504/*
2505 * Get from "/proc/net/dev" all interfaces listed there; if they're
2506 * already in the list of interfaces we have, that won't add another
2507 * instance, but if they're not, that'll add them.
2508 *
2509 * See comments from scan_sys_class_net().
2510 */
2511static int
2512scan_proc_net_dev(pcap_if_list_t *devlistp, char *errbuf)
2513{
2514	FILE *proc_net_f;
2515	int fd;
2516	char linebuf[512];
2517	int linenum;
2518	char *p;
2519	int ret = 0;
2520
2521	proc_net_f = fopen("/proc/net/dev", "r");
2522	if (proc_net_f == NULL) {
2523		/*
2524		 * Don't fail if it doesn't exist at all.
2525		 */
2526		if (errno == ENOENT)
2527			return (0);
2528
2529		/*
2530		 * Fail if we got some other error.
2531		 */
2532		pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
2533		    errno, "Can't open /proc/net/dev");
2534		return (-1);
2535	}
2536
2537	/*
2538	 * Create a socket from which to fetch interface information.
2539	 */
2540	fd = socket(PF_UNIX, SOCK_RAW, 0);
2541	if (fd < 0) {
2542		pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
2543		    errno, "socket");
2544		(void)fclose(proc_net_f);
2545		return (-1);
2546	}
2547
2548	for (linenum = 1;
2549	    fgets(linebuf, sizeof linebuf, proc_net_f) != NULL; linenum++) {
2550		/*
2551		 * Skip the first two lines - they're headers.
2552		 */
2553		if (linenum <= 2)
2554			continue;
2555
2556		p = &linebuf[0];
2557
2558		/*
2559		 * Skip leading white space.
2560		 */
2561		while (*p != '\0' && isascii(*p) && isspace(*p))
2562			p++;
2563		if (*p == '\0' || *p == '\n')
2564			continue;	/* blank line */
2565
2566		/*
2567		 * Attempt to add the interface.
2568		 */
2569		if (add_linux_if(devlistp, p, fd, errbuf) == -1) {
2570			/* Fail. */
2571			ret = -1;
2572			break;
2573		}
2574	}
2575	if (ret != -1) {
2576		/*
2577		 * Well, we didn't fail for any other reason; did we
2578		 * fail due to an error reading the file?
2579		 */
2580		if (ferror(proc_net_f)) {
2581			pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
2582			    errno, "Error reading /proc/net/dev");
2583			ret = -1;
2584		}
2585	}
2586
2587	(void)close(fd);
2588	(void)fclose(proc_net_f);
2589	return (ret);
2590}
2591
2592/*
2593 * Description string for the "any" device.
2594 */
2595static const char any_descr[] = "Pseudo-device that captures on all interfaces";
2596
2597/*
2598 * A SOCK_PACKET or PF_PACKET socket can be bound to any network interface.
2599 */
2600static int
2601can_be_bound(const char *name _U_)
2602{
2603	return (1);
2604}
2605
2606/*
2607 * Get additional flags for a device, using SIOCGIFMEDIA.
2608 */
2609static int
2610get_if_flags(const char *name, bpf_u_int32 *flags, char *errbuf)
2611{
2612	int sock;
2613	FILE *fh;
2614	unsigned int arptype;
2615	struct ifreq ifr;
2616	struct ethtool_value info;
2617
2618	if (*flags & PCAP_IF_LOOPBACK) {
2619		/*
2620		 * Loopback devices aren't wireless, and "connected"/
2621		 * "disconnected" doesn't apply to them.
2622		 */
2623		*flags |= PCAP_IF_CONNECTION_STATUS_NOT_APPLICABLE;
2624		return 0;
2625	}
2626
2627	sock = socket(AF_INET, SOCK_DGRAM, 0);
2628	if (sock == -1) {
2629		pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE, errno,
2630		    "Can't create socket to get ethtool information for %s",
2631		    name);
2632		return -1;
2633	}
2634
2635	/*
2636	 * OK, what type of network is this?
2637	 * In particular, is it wired or wireless?
2638	 */
2639	if (is_wifi(sock, name)) {
2640		/*
2641		 * Wi-Fi, hence wireless.
2642		 */
2643		*flags |= PCAP_IF_WIRELESS;
2644	} else {
2645		/*
2646		 * OK, what does /sys/class/net/{if}/type contain?
2647		 * (We don't use that for Wi-Fi, as it'll report
2648		 * "Ethernet", i.e. ARPHRD_ETHER, for non-monitor-
2649		 * mode devices.)
2650		 */
2651		char *pathstr;
2652
2653		if (asprintf(&pathstr, "/sys/class/net/%s/type", name) == -1) {
2654			pcap_snprintf(errbuf, PCAP_ERRBUF_SIZE,
2655			    "%s: Can't generate path name string for /sys/class/net device",
2656			    name);
2657			close(sock);
2658			return -1;
2659		}
2660		fh = fopen(pathstr, "r");
2661		if (fh != NULL) {
2662			if (fscanf(fh, "%u", &arptype) == 1) {
2663				/*
2664				 * OK, we got an ARPHRD_ type; what is it?
2665				 */
2666				switch (arptype) {
2667
2668#ifdef ARPHRD_LOOPBACK
2669				case ARPHRD_LOOPBACK:
2670					/*
2671					 * These are types to which
2672					 * "connected" and "disconnected"
2673					 * don't apply, so don't bother
2674					 * asking about it.
2675					 *
2676					 * XXX - add other types?
2677					 */
2678					close(sock);
2679					fclose(fh);
2680					free(pathstr);
2681					return 0;
2682#endif
2683
2684				case ARPHRD_IRDA:
2685				case ARPHRD_IEEE80211:
2686				case ARPHRD_IEEE80211_PRISM:
2687				case ARPHRD_IEEE80211_RADIOTAP:
2688#ifdef ARPHRD_IEEE802154
2689				case ARPHRD_IEEE802154:
2690#endif
2691#ifdef ARPHRD_IEEE802154_MONITOR
2692				case ARPHRD_IEEE802154_MONITOR:
2693#endif
2694#ifdef ARPHRD_6LOWPAN
2695				case ARPHRD_6LOWPAN:
2696#endif
2697					/*
2698					 * Various wireless types.
2699					 */
2700					*flags |= PCAP_IF_WIRELESS;
2701					break;
2702				}
2703			}
2704			fclose(fh);
2705			free(pathstr);
2706		}
2707	}
2708
2709#ifdef ETHTOOL_GLINK
2710	memset(&ifr, 0, sizeof(ifr));
2711	strlcpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
2712	info.cmd = ETHTOOL_GLINK;
2713	ifr.ifr_data = (caddr_t)&info;
2714	if (ioctl(sock, SIOCETHTOOL, &ifr) == -1) {
2715		int save_errno = errno;
2716
2717		switch (save_errno) {
2718
2719		case EOPNOTSUPP:
2720		case EINVAL:
2721			/*
2722			 * OK, this OS version or driver doesn't support
2723			 * asking for this information.
2724			 * XXX - distinguish between "this doesn't
2725			 * support ethtool at all because it's not
2726			 * that type of device" vs. "this doesn't
2727			 * support ethtool even though it's that
2728			 * type of device", and return "unknown".
2729			 */
2730			*flags |= PCAP_IF_CONNECTION_STATUS_NOT_APPLICABLE;
2731			close(sock);
2732			return 0;
2733
2734		case ENODEV:
2735			/*
2736			 * OK, no such device.
2737			 * The user will find that out when they try to
2738			 * activate the device; just say "OK" and
2739			 * don't set anything.
2740			 */
2741			close(sock);
2742			return 0;
2743
2744		default:
2745			/*
2746			 * Other error.
2747			 */
2748			pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
2749			    save_errno,
2750			    "%s: SIOCETHTOOL(ETHTOOL_GLINK) ioctl failed",
2751			    name);
2752			close(sock);
2753			return -1;
2754		}
2755	}
2756
2757	/*
2758	 * Is it connected?
2759	 */
2760	if (info.data) {
2761		/*
2762		 * It's connected.
2763		 */
2764		*flags |= PCAP_IF_CONNECTION_STATUS_CONNECTED;
2765	} else {
2766		/*
2767		 * It's disconnected.
2768		 */
2769		*flags |= PCAP_IF_CONNECTION_STATUS_DISCONNECTED;
2770	}
2771#endif
2772
2773	close(sock);
2774	return 0;
2775}
2776
2777int
2778pcap_platform_finddevs(pcap_if_list_t *devlistp, char *errbuf)
2779{
2780	int ret;
2781
2782	/*
2783	 * Get the list of regular interfaces first.
2784	 */
2785	if (pcap_findalldevs_interfaces(devlistp, errbuf, can_be_bound,
2786	    get_if_flags) == -1)
2787		return (-1);	/* failure */
2788
2789	/*
2790	 * Read "/sys/class/net", and add to the list of interfaces all
2791	 * interfaces listed there that we don't already have, because,
2792	 * on Linux, SIOCGIFCONF reports only interfaces with IPv4 addresses,
2793	 * and even getifaddrs() won't return information about
2794	 * interfaces with no addresses, so you need to read "/sys/class/net"
2795	 * to get the names of the rest of the interfaces.
2796	 */
2797	ret = scan_sys_class_net(devlistp, errbuf);
2798	if (ret == -1)
2799		return (-1);	/* failed */
2800	if (ret == 0) {
2801		/*
2802		 * No /sys/class/net; try reading /proc/net/dev instead.
2803		 */
2804		if (scan_proc_net_dev(devlistp, errbuf) == -1)
2805			return (-1);
2806	}
2807
2808	/*
2809	 * Add the "any" device.
2810	 * As it refers to all network devices, not to any particular
2811	 * network device, the notion of "connected" vs. "disconnected"
2812	 * doesn't apply.
2813	 */
2814	if (add_dev(devlistp, "any",
2815	    PCAP_IF_UP|PCAP_IF_RUNNING|PCAP_IF_CONNECTION_STATUS_NOT_APPLICABLE,
2816	    any_descr, errbuf) == NULL)
2817		return (-1);
2818
2819	return (0);
2820}
2821
2822/*
2823 *  Attach the given BPF code to the packet capture device.
2824 */
2825static int
2826pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
2827    int is_mmapped)
2828{
2829	struct pcap_linux *handlep;
2830#ifdef SO_ATTACH_FILTER
2831	struct sock_fprog	fcode;
2832	int			can_filter_in_kernel;
2833	int			err = 0;
2834#endif
2835
2836	if (!handle)
2837		return -1;
2838	if (!filter) {
2839	        strlcpy(handle->errbuf, "setfilter: No filter specified",
2840			PCAP_ERRBUF_SIZE);
2841		return -1;
2842	}
2843
2844	handlep = handle->priv;
2845
2846	/* Make our private copy of the filter */
2847
2848	if (install_bpf_program(handle, filter) < 0)
2849		/* install_bpf_program() filled in errbuf */
2850		return -1;
2851
2852	/*
2853	 * Run user level packet filter by default. Will be overriden if
2854	 * installing a kernel filter succeeds.
2855	 */
2856	handlep->filter_in_userland = 1;
2857
2858	/* Install kernel level filter if possible */
2859
2860#ifdef SO_ATTACH_FILTER
2861#ifdef USHRT_MAX
2862	if (handle->fcode.bf_len > USHRT_MAX) {
2863		/*
2864		 * fcode.len is an unsigned short for current kernel.
2865		 * I have yet to see BPF-Code with that much
2866		 * instructions but still it is possible. So for the
2867		 * sake of correctness I added this check.
2868		 */
2869		fprintf(stderr, "Warning: Filter too complex for kernel\n");
2870		fcode.len = 0;
2871		fcode.filter = NULL;
2872		can_filter_in_kernel = 0;
2873	} else
2874#endif /* USHRT_MAX */
2875	{
2876		/*
2877		 * Oh joy, the Linux kernel uses struct sock_fprog instead
2878		 * of struct bpf_program and of course the length field is
2879		 * of different size. Pointed out by Sebastian
2880		 *
2881		 * Oh, and we also need to fix it up so that all "ret"
2882		 * instructions with non-zero operands have MAXIMUM_SNAPLEN
2883		 * as the operand if we're not capturing in memory-mapped
2884		 * mode, and so that, if we're in cooked mode, all memory-
2885		 * reference instructions use special magic offsets in
2886		 * references to the link-layer header and assume that the
2887		 * link-layer payload begins at 0; "fix_program()" will do
2888		 * that.
2889		 */
2890		switch (fix_program(handle, &fcode, is_mmapped)) {
2891
2892		case -1:
2893		default:
2894			/*
2895			 * Fatal error; just quit.
2896			 * (The "default" case shouldn't happen; we
2897			 * return -1 for that reason.)
2898			 */
2899			return -1;
2900
2901		case 0:
2902			/*
2903			 * The program performed checks that we can't make
2904			 * work in the kernel.
2905			 */
2906			can_filter_in_kernel = 0;
2907			break;
2908
2909		case 1:
2910			/*
2911			 * We have a filter that'll work in the kernel.
2912			 */
2913			can_filter_in_kernel = 1;
2914			break;
2915		}
2916	}
2917
2918	/*
2919	 * NOTE: at this point, we've set both the "len" and "filter"
2920	 * fields of "fcode".  As of the 2.6.32.4 kernel, at least,
2921	 * those are the only members of the "sock_fprog" structure,
2922	 * so we initialize every member of that structure.
2923	 *
2924	 * If there is anything in "fcode" that is not initialized,
2925	 * it is either a field added in a later kernel, or it's
2926	 * padding.
2927	 *
2928	 * If a new field is added, this code needs to be updated
2929	 * to set it correctly.
2930	 *
2931	 * If there are no other fields, then:
2932	 *
2933	 *	if the Linux kernel looks at the padding, it's
2934	 *	buggy;
2935	 *
2936	 *	if the Linux kernel doesn't look at the padding,
2937	 *	then if some tool complains that we're passing
2938	 *	uninitialized data to the kernel, then the tool
2939	 *	is buggy and needs to understand that it's just
2940	 *	padding.
2941	 */
2942	if (can_filter_in_kernel) {
2943		if ((err = set_kernel_filter(handle, &fcode)) == 0)
2944		{
2945			/*
2946			 * Installation succeded - using kernel filter,
2947			 * so userland filtering not needed.
2948			 */
2949			handlep->filter_in_userland = 0;
2950		}
2951		else if (err == -1)	/* Non-fatal error */
2952		{
2953			/*
2954			 * Print a warning if we weren't able to install
2955			 * the filter for a reason other than "this kernel
2956			 * isn't configured to support socket filters.
2957			 */
2958			if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
2959				fprintf(stderr,
2960				    "Warning: Kernel filter failed: %s\n",
2961					pcap_strerror(errno));
2962			}
2963		}
2964	}
2965
2966	/*
2967	 * If we're not using the kernel filter, get rid of any kernel
2968	 * filter that might've been there before, e.g. because the
2969	 * previous filter could work in the kernel, or because some other
2970	 * code attached a filter to the socket by some means other than
2971	 * calling "pcap_setfilter()".  Otherwise, the kernel filter may
2972	 * filter out packets that would pass the new userland filter.
2973	 */
2974	if (handlep->filter_in_userland) {
2975		if (reset_kernel_filter(handle) == -1) {
2976			pcap_fmt_errmsg_for_errno(handle->errbuf,
2977			    PCAP_ERRBUF_SIZE, errno,
2978			    "can't remove kernel filter");
2979			err = -2;	/* fatal error */
2980		}
2981	}
2982
2983	/*
2984	 * Free up the copy of the filter that was made by "fix_program()".
2985	 */
2986	if (fcode.filter != NULL)
2987		free(fcode.filter);
2988
2989	if (err == -2)
2990		/* Fatal error */
2991		return -1;
2992#endif /* SO_ATTACH_FILTER */
2993
2994	return 0;
2995}
2996
2997static int
2998pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
2999{
3000	return pcap_setfilter_linux_common(handle, filter, 0);
3001}
3002
3003
3004/*
3005 * Set direction flag: Which packets do we accept on a forwarding
3006 * single device? IN, OUT or both?
3007 */
3008static int
3009pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
3010{
3011#ifdef HAVE_PF_PACKET_SOCKETS
3012	struct pcap_linux *handlep = handle->priv;
3013
3014	if (!handlep->sock_packet) {
3015		handle->direction = d;
3016		return 0;
3017	}
3018#endif
3019	/*
3020	 * We're not using PF_PACKET sockets, so we can't determine
3021	 * the direction of the packet.
3022	 */
3023	pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3024	    "Setting direction is not supported on SOCK_PACKET sockets");
3025	return -1;
3026}
3027
3028#ifdef HAVE_PF_PACKET_SOCKETS
3029/*
3030 * Map the PACKET_ value to a LINUX_SLL_ value; we
3031 * want the same numerical value to be used in
3032 * the link-layer header even if the numerical values
3033 * for the PACKET_ #defines change, so that programs
3034 * that look at the packet type field will always be
3035 * able to handle DLT_LINUX_SLL captures.
3036 */
3037static short int
3038map_packet_type_to_sll_type(short int sll_pkttype)
3039{
3040	switch (sll_pkttype) {
3041
3042	case PACKET_HOST:
3043		return htons(LINUX_SLL_HOST);
3044
3045	case PACKET_BROADCAST:
3046		return htons(LINUX_SLL_BROADCAST);
3047
3048	case PACKET_MULTICAST:
3049		return  htons(LINUX_SLL_MULTICAST);
3050
3051	case PACKET_OTHERHOST:
3052		return htons(LINUX_SLL_OTHERHOST);
3053
3054	case PACKET_OUTGOING:
3055		return htons(LINUX_SLL_OUTGOING);
3056
3057	default:
3058		return -1;
3059	}
3060}
3061#endif
3062
3063static int
3064is_wifi(int sock_fd
3065#ifndef IW_MODE_MONITOR
3066_U_
3067#endif
3068, const char *device)
3069{
3070	char *pathstr;
3071	struct stat statb;
3072#ifdef IW_MODE_MONITOR
3073	char errbuf[PCAP_ERRBUF_SIZE];
3074#endif
3075
3076	/*
3077	 * See if there's a sysfs wireless directory for it.
3078	 * If so, it's a wireless interface.
3079	 */
3080	if (asprintf(&pathstr, "/sys/class/net/%s/wireless", device) == -1) {
3081		/*
3082		 * Just give up here.
3083		 */
3084		return 0;
3085	}
3086	if (stat(pathstr, &statb) == 0) {
3087		free(pathstr);
3088		return 1;
3089	}
3090	free(pathstr);
3091
3092#ifdef IW_MODE_MONITOR
3093	/*
3094	 * OK, maybe it's not wireless, or maybe this kernel doesn't
3095	 * support sysfs.  Try the wireless extensions.
3096	 */
3097	if (has_wext(sock_fd, device, errbuf) == 1) {
3098		/*
3099		 * It supports the wireless extensions, so it's a Wi-Fi
3100		 * device.
3101		 */
3102		return 1;
3103	}
3104#endif
3105	return 0;
3106}
3107
3108/*
3109 *  Linux uses the ARP hardware type to identify the type of an
3110 *  interface. pcap uses the DLT_xxx constants for this. This
3111 *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
3112 *  constant, as arguments, and sets "handle->linktype" to the
3113 *  appropriate DLT_XXX constant and sets "handle->offset" to
3114 *  the appropriate value (to make "handle->offset" plus link-layer
3115 *  header length be a multiple of 4, so that the link-layer payload
3116 *  will be aligned on a 4-byte boundary when capturing packets).
3117 *  (If the offset isn't set here, it'll be 0; add code as appropriate
3118 *  for cases where it shouldn't be 0.)
3119 *
3120 *  If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
3121 *  in cooked mode; otherwise, we can't use cooked mode, so we have
3122 *  to pick some type that works in raw mode, or fail.
3123 *
3124 *  Sets the link type to -1 if unable to map the type.
3125 */
3126static void map_arphrd_to_dlt(pcap_t *handle, int sock_fd, int arptype,
3127			      const char *device, int cooked_ok)
3128{
3129	static const char cdma_rmnet[] = "cdma_rmnet";
3130
3131	switch (arptype) {
3132
3133	case ARPHRD_ETHER:
3134		/*
3135		 * For various annoying reasons having to do with DHCP
3136		 * software, some versions of Android give the mobile-
3137		 * phone-network interface an ARPHRD_ value of
3138		 * ARPHRD_ETHER, even though the packets supplied by
3139		 * that interface have no link-layer header, and begin
3140		 * with an IP header, so that the ARPHRD_ value should
3141		 * be ARPHRD_NONE.
3142		 *
3143		 * Detect those devices by checking the device name, and
3144		 * use DLT_RAW for them.
3145		 */
3146		if (strncmp(device, cdma_rmnet, sizeof cdma_rmnet - 1) == 0) {
3147			handle->linktype = DLT_RAW;
3148			return;
3149		}
3150
3151		/*
3152		 * Is this a real Ethernet device?  If so, give it a
3153		 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
3154		 * that an application can let you choose it, in case you're
3155		 * capturing DOCSIS traffic that a Cisco Cable Modem
3156		 * Termination System is putting out onto an Ethernet (it
3157		 * doesn't put an Ethernet header onto the wire, it puts raw
3158		 * DOCSIS frames out on the wire inside the low-level
3159		 * Ethernet framing).
3160		 *
3161		 * XXX - are there any other sorts of "fake Ethernet" that
3162		 * have ARPHRD_ETHER but that shouldn't offer DLT_DOCSIS as
3163		 * a Cisco CMTS won't put traffic onto it or get traffic
3164		 * bridged onto it?  ISDN is handled in "activate_new()",
3165		 * as we fall back on cooked mode there, and we use
3166		 * is_wifi() to check for 802.11 devices; are there any
3167		 * others?
3168		 */
3169		if (!is_wifi(sock_fd, device)) {
3170			/*
3171			 * It's not a Wi-Fi device; offer DOCSIS.
3172			 */
3173			handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
3174			/*
3175			 * If that fails, just leave the list empty.
3176			 */
3177			if (handle->dlt_list != NULL) {
3178				handle->dlt_list[0] = DLT_EN10MB;
3179				handle->dlt_list[1] = DLT_DOCSIS;
3180				handle->dlt_count = 2;
3181			}
3182		}
3183		/* FALLTHROUGH */
3184
3185	case ARPHRD_METRICOM:
3186	case ARPHRD_LOOPBACK:
3187		handle->linktype = DLT_EN10MB;
3188		handle->offset = 2;
3189		break;
3190
3191	case ARPHRD_EETHER:
3192		handle->linktype = DLT_EN3MB;
3193		break;
3194
3195	case ARPHRD_AX25:
3196		handle->linktype = DLT_AX25_KISS;
3197		break;
3198
3199	case ARPHRD_PRONET:
3200		handle->linktype = DLT_PRONET;
3201		break;
3202
3203	case ARPHRD_CHAOS:
3204		handle->linktype = DLT_CHAOS;
3205		break;
3206#ifndef ARPHRD_CAN
3207#define ARPHRD_CAN 280
3208#endif
3209	case ARPHRD_CAN:
3210		/*
3211		 * Map this to DLT_LINUX_SLL; that way, CAN frames will
3212		 * have ETH_P_CAN/LINUX_SLL_P_CAN as the protocol and
3213		 * CAN FD frames will have ETH_P_CANFD/LINUX_SLL_P_CANFD
3214		 * as the protocol, so they can be distinguished by the
3215		 * protocol in the SLL header.
3216		 */
3217		handle->linktype = DLT_LINUX_SLL;
3218		break;
3219
3220#ifndef ARPHRD_IEEE802_TR
3221#define ARPHRD_IEEE802_TR 800	/* From Linux 2.4 */
3222#endif
3223	case ARPHRD_IEEE802_TR:
3224	case ARPHRD_IEEE802:
3225		handle->linktype = DLT_IEEE802;
3226		handle->offset = 2;
3227		break;
3228
3229	case ARPHRD_ARCNET:
3230		handle->linktype = DLT_ARCNET_LINUX;
3231		break;
3232
3233#ifndef ARPHRD_FDDI	/* From Linux 2.2.13 */
3234#define ARPHRD_FDDI	774
3235#endif
3236	case ARPHRD_FDDI:
3237		handle->linktype = DLT_FDDI;
3238		handle->offset = 3;
3239		break;
3240
3241#ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
3242#define ARPHRD_ATM 19
3243#endif
3244	case ARPHRD_ATM:
3245		/*
3246		 * The Classical IP implementation in ATM for Linux
3247		 * supports both what RFC 1483 calls "LLC Encapsulation",
3248		 * in which each packet has an LLC header, possibly
3249		 * with a SNAP header as well, prepended to it, and
3250		 * what RFC 1483 calls "VC Based Multiplexing", in which
3251		 * different virtual circuits carry different network
3252		 * layer protocols, and no header is prepended to packets.
3253		 *
3254		 * They both have an ARPHRD_ type of ARPHRD_ATM, so
3255		 * you can't use the ARPHRD_ type to find out whether
3256		 * captured packets will have an LLC header, and,
3257		 * while there's a socket ioctl to *set* the encapsulation
3258		 * type, there's no ioctl to *get* the encapsulation type.
3259		 *
3260		 * This means that
3261		 *
3262		 *	programs that dissect Linux Classical IP frames
3263		 *	would have to check for an LLC header and,
3264		 *	depending on whether they see one or not, dissect
3265		 *	the frame as LLC-encapsulated or as raw IP (I
3266		 *	don't know whether there's any traffic other than
3267		 *	IP that would show up on the socket, or whether
3268		 *	there's any support for IPv6 in the Linux
3269		 *	Classical IP code);
3270		 *
3271		 *	filter expressions would have to compile into
3272		 *	code that checks for an LLC header and does
3273		 *	the right thing.
3274		 *
3275		 * Both of those are a nuisance - and, at least on systems
3276		 * that support PF_PACKET sockets, we don't have to put
3277		 * up with those nuisances; instead, we can just capture
3278		 * in cooked mode.  That's what we'll do, if we can.
3279		 * Otherwise, we'll just fail.
3280		 */
3281		if (cooked_ok)
3282			handle->linktype = DLT_LINUX_SLL;
3283		else
3284			handle->linktype = -1;
3285		break;
3286
3287#ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
3288#define ARPHRD_IEEE80211 801
3289#endif
3290	case ARPHRD_IEEE80211:
3291		handle->linktype = DLT_IEEE802_11;
3292		break;
3293
3294#ifndef ARPHRD_IEEE80211_PRISM  /* From Linux 2.4.18 */
3295#define ARPHRD_IEEE80211_PRISM 802
3296#endif
3297	case ARPHRD_IEEE80211_PRISM:
3298		handle->linktype = DLT_PRISM_HEADER;
3299		break;
3300
3301#ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
3302#define ARPHRD_IEEE80211_RADIOTAP 803
3303#endif
3304	case ARPHRD_IEEE80211_RADIOTAP:
3305		handle->linktype = DLT_IEEE802_11_RADIO;
3306		break;
3307
3308	case ARPHRD_PPP:
3309		/*
3310		 * Some PPP code in the kernel supplies no link-layer
3311		 * header whatsoever to PF_PACKET sockets; other PPP
3312		 * code supplies PPP link-layer headers ("syncppp.c");
3313		 * some PPP code might supply random link-layer
3314		 * headers (PPP over ISDN - there's code in Ethereal,
3315		 * for example, to cope with PPP-over-ISDN captures
3316		 * with which the Ethereal developers have had to cope,
3317		 * heuristically trying to determine which of the
3318		 * oddball link-layer headers particular packets have).
3319		 *
3320		 * As such, we just punt, and run all PPP interfaces
3321		 * in cooked mode, if we can; otherwise, we just treat
3322		 * it as DLT_RAW, for now - if somebody needs to capture,
3323		 * on a 2.0[.x] kernel, on PPP devices that supply a
3324		 * link-layer header, they'll have to add code here to
3325		 * map to the appropriate DLT_ type (possibly adding a
3326		 * new DLT_ type, if necessary).
3327		 */
3328		if (cooked_ok)
3329			handle->linktype = DLT_LINUX_SLL;
3330		else {
3331			/*
3332			 * XXX - handle ISDN types here?  We can't fall
3333			 * back on cooked sockets, so we'd have to
3334			 * figure out from the device name what type of
3335			 * link-layer encapsulation it's using, and map
3336			 * that to an appropriate DLT_ value, meaning
3337			 * we'd map "isdnN" devices to DLT_RAW (they
3338			 * supply raw IP packets with no link-layer
3339			 * header) and "isdY" devices to a new DLT_I4L_IP
3340			 * type that has only an Ethernet packet type as
3341			 * a link-layer header.
3342			 *
3343			 * But sometimes we seem to get random crap
3344			 * in the link-layer header when capturing on
3345			 * ISDN devices....
3346			 */
3347			handle->linktype = DLT_RAW;
3348		}
3349		break;
3350
3351#ifndef ARPHRD_CISCO
3352#define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
3353#endif
3354	case ARPHRD_CISCO:
3355		handle->linktype = DLT_C_HDLC;
3356		break;
3357
3358	/* Not sure if this is correct for all tunnels, but it
3359	 * works for CIPE */
3360	case ARPHRD_TUNNEL:
3361#ifndef ARPHRD_SIT
3362#define ARPHRD_SIT 776	/* From Linux 2.2.13 */
3363#endif
3364	case ARPHRD_SIT:
3365	case ARPHRD_CSLIP:
3366	case ARPHRD_SLIP6:
3367	case ARPHRD_CSLIP6:
3368	case ARPHRD_ADAPT:
3369	case ARPHRD_SLIP:
3370#ifndef ARPHRD_RAWHDLC
3371#define ARPHRD_RAWHDLC 518
3372#endif
3373	case ARPHRD_RAWHDLC:
3374#ifndef ARPHRD_DLCI
3375#define ARPHRD_DLCI 15
3376#endif
3377	case ARPHRD_DLCI:
3378		/*
3379		 * XXX - should some of those be mapped to DLT_LINUX_SLL
3380		 * instead?  Should we just map all of them to DLT_LINUX_SLL?
3381		 */
3382		handle->linktype = DLT_RAW;
3383		break;
3384
3385#ifndef ARPHRD_FRAD
3386#define ARPHRD_FRAD 770
3387#endif
3388	case ARPHRD_FRAD:
3389		handle->linktype = DLT_FRELAY;
3390		break;
3391
3392	case ARPHRD_LOCALTLK:
3393		handle->linktype = DLT_LTALK;
3394		break;
3395
3396	case 18:
3397		/*
3398		 * RFC 4338 defines an encapsulation for IP and ARP
3399		 * packets that's compatible with the RFC 2625
3400		 * encapsulation, but that uses a different ARP
3401		 * hardware type and hardware addresses.  That
3402		 * ARP hardware type is 18; Linux doesn't define
3403		 * any ARPHRD_ value as 18, but if it ever officially
3404		 * supports RFC 4338-style IP-over-FC, it should define
3405		 * one.
3406		 *
3407		 * For now, we map it to DLT_IP_OVER_FC, in the hopes
3408		 * that this will encourage its use in the future,
3409		 * should Linux ever officially support RFC 4338-style
3410		 * IP-over-FC.
3411		 */
3412		handle->linktype = DLT_IP_OVER_FC;
3413		break;
3414
3415#ifndef ARPHRD_FCPP
3416#define ARPHRD_FCPP	784
3417#endif
3418	case ARPHRD_FCPP:
3419#ifndef ARPHRD_FCAL
3420#define ARPHRD_FCAL	785
3421#endif
3422	case ARPHRD_FCAL:
3423#ifndef ARPHRD_FCPL
3424#define ARPHRD_FCPL	786
3425#endif
3426	case ARPHRD_FCPL:
3427#ifndef ARPHRD_FCFABRIC
3428#define ARPHRD_FCFABRIC	787
3429#endif
3430	case ARPHRD_FCFABRIC:
3431		/*
3432		 * Back in 2002, Donald Lee at Cray wanted a DLT_ for
3433		 * IP-over-FC:
3434		 *
3435		 *	http://www.mail-archive.com/tcpdump-workers@sandelman.ottawa.on.ca/msg01043.html
3436		 *
3437		 * and one was assigned.
3438		 *
3439		 * In a later private discussion (spun off from a message
3440		 * on the ethereal-users list) on how to get that DLT_
3441		 * value in libpcap on Linux, I ended up deciding that
3442		 * the best thing to do would be to have him tweak the
3443		 * driver to set the ARPHRD_ value to some ARPHRD_FCxx
3444		 * type, and map all those types to DLT_IP_OVER_FC:
3445		 *
3446		 *	I've checked into the libpcap and tcpdump CVS tree
3447		 *	support for DLT_IP_OVER_FC.  In order to use that,
3448		 *	you'd have to modify your modified driver to return
3449		 *	one of the ARPHRD_FCxxx types, in "fcLINUXfcp.c" -
3450		 *	change it to set "dev->type" to ARPHRD_FCFABRIC, for
3451		 *	example (the exact value doesn't matter, it can be
3452		 *	any of ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, or
3453		 *	ARPHRD_FCFABRIC).
3454		 *
3455		 * 11 years later, Christian Svensson wanted to map
3456		 * various ARPHRD_ values to DLT_FC_2 and
3457		 * DLT_FC_2_WITH_FRAME_DELIMS for raw Fibre Channel
3458		 * frames:
3459		 *
3460		 *	https://github.com/mcr/libpcap/pull/29
3461		 *
3462		 * There doesn't seem to be any network drivers that uses
3463		 * any of the ARPHRD_FC* values for IP-over-FC, and
3464		 * it's not exactly clear what the "Dummy types for non
3465		 * ARP hardware" are supposed to mean (link-layer
3466		 * header type?  Physical network type?), so it's
3467		 * not exactly clear why the ARPHRD_FC* types exist
3468		 * in the first place.
3469		 *
3470		 * For now, we map them to DLT_FC_2, and provide an
3471		 * option of DLT_FC_2_WITH_FRAME_DELIMS, as well as
3472		 * DLT_IP_OVER_FC just in case there's some old
3473		 * driver out there that uses one of those types for
3474		 * IP-over-FC on which somebody wants to capture
3475		 * packets.
3476		 */
3477		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 3);
3478		/*
3479		 * If that fails, just leave the list empty.
3480		 */
3481		if (handle->dlt_list != NULL) {
3482			handle->dlt_list[0] = DLT_FC_2;
3483			handle->dlt_list[1] = DLT_FC_2_WITH_FRAME_DELIMS;
3484			handle->dlt_list[2] = DLT_IP_OVER_FC;
3485			handle->dlt_count = 3;
3486		}
3487		handle->linktype = DLT_FC_2;
3488		break;
3489
3490#ifndef ARPHRD_IRDA
3491#define ARPHRD_IRDA	783
3492#endif
3493	case ARPHRD_IRDA:
3494		/* Don't expect IP packet out of this interfaces... */
3495		handle->linktype = DLT_LINUX_IRDA;
3496		/* We need to save packet direction for IrDA decoding,
3497		 * so let's use "Linux-cooked" mode. Jean II
3498		 *
3499		 * XXX - this is handled in activate_new(). */
3500		/* handlep->cooked = 1; */
3501		break;
3502
3503	/* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
3504	 * is needed, please report it to <daniele@orlandi.com> */
3505#ifndef ARPHRD_LAPD
3506#define ARPHRD_LAPD	8445
3507#endif
3508	case ARPHRD_LAPD:
3509		/* Don't expect IP packet out of this interfaces... */
3510		handle->linktype = DLT_LINUX_LAPD;
3511		break;
3512
3513#ifndef ARPHRD_NONE
3514#define ARPHRD_NONE	0xFFFE
3515#endif
3516	case ARPHRD_NONE:
3517		/*
3518		 * No link-layer header; packets are just IP
3519		 * packets, so use DLT_RAW.
3520		 */
3521		handle->linktype = DLT_RAW;
3522		break;
3523
3524#ifndef ARPHRD_IEEE802154
3525#define ARPHRD_IEEE802154      804
3526#endif
3527       case ARPHRD_IEEE802154:
3528               handle->linktype =  DLT_IEEE802_15_4_NOFCS;
3529               break;
3530
3531#ifndef ARPHRD_NETLINK
3532#define ARPHRD_NETLINK	824
3533#endif
3534	case ARPHRD_NETLINK:
3535		handle->linktype = DLT_NETLINK;
3536		/*
3537		 * We need to use cooked mode, so that in sll_protocol we
3538		 * pick up the netlink protocol type such as NETLINK_ROUTE,
3539		 * NETLINK_GENERIC, NETLINK_FIB_LOOKUP, etc.
3540		 *
3541		 * XXX - this is handled in activate_new().
3542		 */
3543		/* handlep->cooked = 1; */
3544		break;
3545
3546#ifndef ARPHRD_VSOCKMON
3547#define ARPHRD_VSOCKMON	826
3548#endif
3549	case ARPHRD_VSOCKMON:
3550		handle->linktype = DLT_VSOCK;
3551		break;
3552
3553	default:
3554		handle->linktype = -1;
3555		break;
3556	}
3557}
3558
3559/* ===== Functions to interface to the newer kernels ================== */
3560
3561/*
3562 * Try to open a packet socket using the new kernel PF_PACKET interface.
3563 * Returns 1 on success, 0 on an error that means the new interface isn't
3564 * present (so the old SOCK_PACKET interface should be tried), and a
3565 * PCAP_ERROR_ value on an error that means that the old mechanism won't
3566 * work either (so it shouldn't be tried).
3567 */
3568static int
3569activate_new(pcap_t *handle)
3570{
3571#ifdef HAVE_PF_PACKET_SOCKETS
3572	struct pcap_linux *handlep = handle->priv;
3573	const char		*device = handle->opt.device;
3574	int			is_any_device = (strcmp(device, "any") == 0);
3575	int			protocol = pcap_protocol(handle);
3576	int			sock_fd = -1, arptype;
3577#ifdef HAVE_PACKET_AUXDATA
3578	int			val;
3579#endif
3580	int			err = 0;
3581	struct packet_mreq	mr;
3582#if defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT)
3583	int			bpf_extensions;
3584	socklen_t		len = sizeof(bpf_extensions);
3585#endif
3586
3587	/*
3588	 * Open a socket with protocol family packet. If the
3589	 * "any" device was specified, we open a SOCK_DGRAM
3590	 * socket for the cooked interface, otherwise we first
3591	 * try a SOCK_RAW socket for the raw interface.
3592	 */
3593	sock_fd = is_any_device ?
3594		socket(PF_PACKET, SOCK_DGRAM, protocol) :
3595		socket(PF_PACKET, SOCK_RAW, protocol);
3596
3597	if (sock_fd == -1) {
3598		if (errno == EINVAL || errno == EAFNOSUPPORT) {
3599			/*
3600			 * We don't support PF_PACKET/SOCK_whatever
3601			 * sockets; try the old mechanism.
3602			 */
3603			return 0;
3604		}
3605
3606		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3607		    errno, "socket");
3608		if (errno == EPERM || errno == EACCES) {
3609			/*
3610			 * You don't have permission to open the
3611			 * socket.
3612			 */
3613			return PCAP_ERROR_PERM_DENIED;
3614		} else {
3615			/*
3616			 * Other error.
3617			 */
3618			return PCAP_ERROR;
3619		}
3620	}
3621
3622	/* It seems the kernel supports the new interface. */
3623	handlep->sock_packet = 0;
3624
3625	/*
3626	 * Get the interface index of the loopback device.
3627	 * If the attempt fails, don't fail, just set the
3628	 * "handlep->lo_ifindex" to -1.
3629	 *
3630	 * XXX - can there be more than one device that loops
3631	 * packets back, i.e. devices other than "lo"?  If so,
3632	 * we'd need to find them all, and have an array of
3633	 * indices for them, and check all of them in
3634	 * "pcap_read_packet()".
3635	 */
3636	handlep->lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
3637
3638	/*
3639	 * Default value for offset to align link-layer payload
3640	 * on a 4-byte boundary.
3641	 */
3642	handle->offset	 = 0;
3643
3644	/*
3645	 * What kind of frames do we have to deal with? Fall back
3646	 * to cooked mode if we have an unknown interface type
3647	 * or a type we know doesn't work well in raw mode.
3648	 */
3649	if (!is_any_device) {
3650		/* Assume for now we don't need cooked mode. */
3651		handlep->cooked = 0;
3652
3653		if (handle->opt.rfmon) {
3654			/*
3655			 * We were asked to turn on monitor mode.
3656			 * Do so before we get the link-layer type,
3657			 * because entering monitor mode could change
3658			 * the link-layer type.
3659			 */
3660			err = enter_rfmon_mode(handle, sock_fd, device);
3661			if (err < 0) {
3662				/* Hard failure */
3663				close(sock_fd);
3664				return err;
3665			}
3666			if (err == 0) {
3667				/*
3668				 * Nothing worked for turning monitor mode
3669				 * on.
3670				 */
3671				close(sock_fd);
3672				return PCAP_ERROR_RFMON_NOTSUP;
3673			}
3674
3675			/*
3676			 * Either monitor mode has been turned on for
3677			 * the device, or we've been given a different
3678			 * device to open for monitor mode.  If we've
3679			 * been given a different device, use it.
3680			 */
3681			if (handlep->mondevice != NULL)
3682				device = handlep->mondevice;
3683		}
3684		arptype	= iface_get_arptype(sock_fd, device, handle->errbuf);
3685		if (arptype < 0) {
3686			close(sock_fd);
3687			return arptype;
3688		}
3689		map_arphrd_to_dlt(handle, sock_fd, arptype, device, 1);
3690		if (handle->linktype == -1 ||
3691		    handle->linktype == DLT_LINUX_SLL ||
3692		    handle->linktype == DLT_LINUX_IRDA ||
3693		    handle->linktype == DLT_LINUX_LAPD ||
3694		    handle->linktype == DLT_NETLINK ||
3695		    (handle->linktype == DLT_EN10MB &&
3696		     (strncmp("isdn", device, 4) == 0 ||
3697		      strncmp("isdY", device, 4) == 0))) {
3698			/*
3699			 * Unknown interface type (-1), or a
3700			 * device we explicitly chose to run
3701			 * in cooked mode (e.g., PPP devices),
3702			 * or an ISDN device (whose link-layer
3703			 * type we can only determine by using
3704			 * APIs that may be different on different
3705			 * kernels) - reopen in cooked mode.
3706			 */
3707			if (close(sock_fd) == -1) {
3708				pcap_fmt_errmsg_for_errno(handle->errbuf,
3709				    PCAP_ERRBUF_SIZE, errno, "close");
3710				return PCAP_ERROR;
3711			}
3712			sock_fd = socket(PF_PACKET, SOCK_DGRAM, protocol);
3713			if (sock_fd == -1) {
3714				pcap_fmt_errmsg_for_errno(handle->errbuf,
3715				    PCAP_ERRBUF_SIZE, errno, "socket");
3716				if (errno == EPERM || errno == EACCES) {
3717					/*
3718					 * You don't have permission to
3719					 * open the socket.
3720					 */
3721					return PCAP_ERROR_PERM_DENIED;
3722				} else {
3723					/*
3724					 * Other error.
3725					 */
3726					return PCAP_ERROR;
3727				}
3728			}
3729			handlep->cooked = 1;
3730
3731			/*
3732			 * Get rid of any link-layer type list
3733			 * we allocated - this only supports cooked
3734			 * capture.
3735			 */
3736			if (handle->dlt_list != NULL) {
3737				free(handle->dlt_list);
3738				handle->dlt_list = NULL;
3739				handle->dlt_count = 0;
3740			}
3741
3742			if (handle->linktype == -1) {
3743				/*
3744				 * Warn that we're falling back on
3745				 * cooked mode; we may want to
3746				 * update "map_arphrd_to_dlt()"
3747				 * to handle the new type.
3748				 */
3749				pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3750					"arptype %d not "
3751					"supported by libpcap - "
3752					"falling back to cooked "
3753					"socket",
3754					arptype);
3755			}
3756
3757			/*
3758			 * IrDA capture is not a real "cooked" capture,
3759			 * it's IrLAP frames, not IP packets.  The
3760			 * same applies to LAPD capture.
3761			 */
3762			if (handle->linktype != DLT_LINUX_IRDA &&
3763			    handle->linktype != DLT_LINUX_LAPD &&
3764			    handle->linktype != DLT_NETLINK)
3765				handle->linktype = DLT_LINUX_SLL;
3766		}
3767
3768		handlep->ifindex = iface_get_id(sock_fd, device,
3769		    handle->errbuf);
3770		if (handlep->ifindex == -1) {
3771			close(sock_fd);
3772			return PCAP_ERROR;
3773		}
3774
3775		if ((err = iface_bind(sock_fd, handlep->ifindex,
3776		    handle->errbuf, protocol)) != 1) {
3777		    	close(sock_fd);
3778			if (err < 0)
3779				return err;
3780			else
3781				return 0;	/* try old mechanism */
3782		}
3783	} else {
3784		/*
3785		 * The "any" device.
3786		 */
3787		if (handle->opt.rfmon) {
3788			/*
3789			 * It doesn't support monitor mode.
3790			 */
3791			close(sock_fd);
3792			return PCAP_ERROR_RFMON_NOTSUP;
3793		}
3794
3795		/*
3796		 * It uses cooked mode.
3797		 */
3798		handlep->cooked = 1;
3799		handle->linktype = DLT_LINUX_SLL;
3800
3801		/*
3802		 * We're not bound to a device.
3803		 * For now, we're using this as an indication
3804		 * that we can't transmit; stop doing that only
3805		 * if we figure out how to transmit in cooked
3806		 * mode.
3807		 */
3808		handlep->ifindex = -1;
3809	}
3810
3811	/*
3812	 * Select promiscuous mode on if "promisc" is set.
3813	 *
3814	 * Do not turn allmulti mode on if we don't select
3815	 * promiscuous mode - on some devices (e.g., Orinoco
3816	 * wireless interfaces), allmulti mode isn't supported
3817	 * and the driver implements it by turning promiscuous
3818	 * mode on, and that screws up the operation of the
3819	 * card as a normal networking interface, and on no
3820	 * other platform I know of does starting a non-
3821	 * promiscuous capture affect which multicast packets
3822	 * are received by the interface.
3823	 */
3824
3825	/*
3826	 * Hmm, how can we set promiscuous mode on all interfaces?
3827	 * I am not sure if that is possible at all.  For now, we
3828	 * silently ignore attempts to turn promiscuous mode on
3829	 * for the "any" device (so you don't have to explicitly
3830	 * disable it in programs such as tcpdump).
3831	 */
3832
3833	if (!is_any_device && handle->opt.promisc) {
3834		memset(&mr, 0, sizeof(mr));
3835		mr.mr_ifindex = handlep->ifindex;
3836		mr.mr_type    = PACKET_MR_PROMISC;
3837		if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
3838		    &mr, sizeof(mr)) == -1) {
3839			pcap_fmt_errmsg_for_errno(handle->errbuf,
3840			    PCAP_ERRBUF_SIZE, errno, "setsockopt");
3841			close(sock_fd);
3842			return PCAP_ERROR;
3843		}
3844	}
3845
3846	/* Enable auxillary data if supported and reserve room for
3847	 * reconstructing VLAN headers. */
3848#ifdef HAVE_PACKET_AUXDATA
3849	val = 1;
3850	if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
3851		       sizeof(val)) == -1 && errno != ENOPROTOOPT) {
3852		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3853		    errno, "setsockopt");
3854		close(sock_fd);
3855		return PCAP_ERROR;
3856	}
3857	handle->offset += VLAN_TAG_LEN;
3858#endif /* HAVE_PACKET_AUXDATA */
3859
3860	/*
3861	 * This is a 2.2[.x] or later kernel (we know that
3862	 * because we're not using a SOCK_PACKET socket -
3863	 * PF_PACKET is supported only in 2.2 and later
3864	 * kernels).
3865	 *
3866	 * We can safely pass "recvfrom()" a byte count
3867	 * based on the snapshot length.
3868	 *
3869	 * If we're in cooked mode, make the snapshot length
3870	 * large enough to hold a "cooked mode" header plus
3871	 * 1 byte of packet data (so we don't pass a byte
3872	 * count of 0 to "recvfrom()").
3873	 */
3874	if (handlep->cooked) {
3875		if (handle->snapshot < SLL_HDR_LEN + 1)
3876			handle->snapshot = SLL_HDR_LEN + 1;
3877	}
3878	handle->bufsize = handle->snapshot;
3879
3880	/*
3881	 * Set the offset at which to insert VLAN tags.
3882	 * That should be the offset of the type field.
3883	 */
3884	switch (handle->linktype) {
3885
3886	case DLT_EN10MB:
3887		/*
3888		 * The type field is after the destination and source
3889		 * MAC address.
3890		 */
3891		handlep->vlan_offset = 2 * ETH_ALEN;
3892		break;
3893
3894	case DLT_LINUX_SLL:
3895		/*
3896		 * The type field is in the last 2 bytes of the
3897		 * DLT_LINUX_SLL header.
3898		 */
3899		handlep->vlan_offset = SLL_HDR_LEN - 2;
3900		break;
3901
3902	default:
3903		handlep->vlan_offset = -1; /* unknown */
3904		break;
3905	}
3906
3907#if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
3908	if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
3909		int nsec_tstamps = 1;
3910
3911		if (setsockopt(sock_fd, SOL_SOCKET, SO_TIMESTAMPNS, &nsec_tstamps, sizeof(nsec_tstamps)) < 0) {
3912			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "setsockopt: unable to set SO_TIMESTAMPNS");
3913			close(sock_fd);
3914			return PCAP_ERROR;
3915		}
3916	}
3917#endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */
3918
3919	/*
3920	 * We've succeeded. Save the socket FD in the pcap structure.
3921	 */
3922	handle->fd = sock_fd;
3923
3924#if defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT)
3925	/*
3926	 * Can we generate special code for VLAN checks?
3927	 * (XXX - what if we need the special code but it's not supported
3928	 * by the OS?  Is that possible?)
3929	 */
3930	if (getsockopt(sock_fd, SOL_SOCKET, SO_BPF_EXTENSIONS,
3931	    &bpf_extensions, &len) == 0) {
3932		if (bpf_extensions >= SKF_AD_VLAN_TAG_PRESENT) {
3933			/*
3934			 * Yes, we can.  Request that we do so.
3935			 */
3936			handle->bpf_codegen_flags |= BPF_SPECIAL_VLAN_HANDLING;
3937		}
3938	}
3939#endif /* defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT) */
3940
3941	return 1;
3942#else /* HAVE_PF_PACKET_SOCKETS */
3943	strlcpy(ebuf,
3944		"New packet capturing interface not supported by build "
3945		"environment", PCAP_ERRBUF_SIZE);
3946	return 0;
3947#endif /* HAVE_PF_PACKET_SOCKETS */
3948}
3949
3950#ifdef HAVE_PACKET_RING
3951/*
3952 * Attempt to activate with memory-mapped access.
3953 *
3954 * On success, returns 1, and sets *status to 0 if there are no warnings
3955 * or to a PCAP_WARNING_ code if there is a warning.
3956 *
3957 * On failure due to lack of support for memory-mapped capture, returns
3958 * 0.
3959 *
3960 * On error, returns -1, and sets *status to the appropriate error code;
3961 * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3962 */
3963static int
3964activate_mmap(pcap_t *handle, int *status)
3965{
3966	struct pcap_linux *handlep = handle->priv;
3967	int ret;
3968
3969	/*
3970	 * Attempt to allocate a buffer to hold the contents of one
3971	 * packet, for use by the oneshot callback.
3972	 */
3973	handlep->oneshot_buffer = malloc(handle->snapshot);
3974	if (handlep->oneshot_buffer == NULL) {
3975		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3976		    errno, "can't allocate oneshot buffer");
3977		*status = PCAP_ERROR;
3978		return -1;
3979	}
3980
3981	if (handle->opt.buffer_size == 0) {
3982		/* by default request 2M for the ring buffer */
3983		handle->opt.buffer_size = 2*1024*1024;
3984	}
3985	ret = prepare_tpacket_socket(handle);
3986	if (ret == -1) {
3987		free(handlep->oneshot_buffer);
3988		*status = PCAP_ERROR;
3989		return ret;
3990	}
3991	ret = create_ring(handle, status);
3992	if (ret == 0) {
3993		/*
3994		 * We don't support memory-mapped capture; our caller
3995		 * will fall back on reading from the socket.
3996		 */
3997		free(handlep->oneshot_buffer);
3998		return 0;
3999	}
4000	if (ret == -1) {
4001		/*
4002		 * Error attempting to enable memory-mapped capture;
4003		 * fail.  create_ring() has set *status.
4004		 */
4005		free(handlep->oneshot_buffer);
4006		return -1;
4007	}
4008
4009	/*
4010	 * Success.  *status has been set either to 0 if there are no
4011	 * warnings or to a PCAP_WARNING_ value if there is a warning.
4012	 *
4013	 * Override some defaults and inherit the other fields from
4014	 * activate_new.
4015	 * handle->offset is used to get the current position into the rx ring.
4016	 * handle->cc is used to store the ring size.
4017	 */
4018
4019	switch (handlep->tp_version) {
4020	case TPACKET_V1:
4021		handle->read_op = pcap_read_linux_mmap_v1;
4022		break;
4023	case TPACKET_V1_64:
4024		handle->read_op = pcap_read_linux_mmap_v1_64;
4025		break;
4026#ifdef HAVE_TPACKET2
4027	case TPACKET_V2:
4028		handle->read_op = pcap_read_linux_mmap_v2;
4029		break;
4030#endif
4031#ifdef HAVE_TPACKET3
4032	case TPACKET_V3:
4033		handle->read_op = pcap_read_linux_mmap_v3;
4034		break;
4035#endif
4036	}
4037	handle->cleanup_op = pcap_cleanup_linux_mmap;
4038	handle->setfilter_op = pcap_setfilter_linux_mmap;
4039	handle->setnonblock_op = pcap_setnonblock_mmap;
4040	handle->getnonblock_op = pcap_getnonblock_mmap;
4041	handle->oneshot_callback = pcap_oneshot_mmap;
4042	handle->selectable_fd = handle->fd;
4043	return 1;
4044}
4045#else /* HAVE_PACKET_RING */
4046static int
4047activate_mmap(pcap_t *handle _U_, int *status _U_)
4048{
4049	return 0;
4050}
4051#endif /* HAVE_PACKET_RING */
4052
4053#ifdef HAVE_PACKET_RING
4054
4055#if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
4056/*
4057 * Attempt to set the socket to the specified version of the memory-mapped
4058 * header.
4059 *
4060 * Return 0 if we succeed; return 1 if we fail because that version isn't
4061 * supported; return -1 on any other error, and set handle->errbuf.
4062 */
4063static int
4064init_tpacket(pcap_t *handle, int version, const char *version_str)
4065{
4066	struct pcap_linux *handlep = handle->priv;
4067	int val = version;
4068	socklen_t len = sizeof(val);
4069
4070	/*
4071	 * Probe whether kernel supports the specified TPACKET version;
4072	 * this also gets the length of the header for that version.
4073	 */
4074	if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
4075		if (errno == ENOPROTOOPT || errno == EINVAL)
4076			return 1;	/* no */
4077
4078		/* Failed to even find out; this is a fatal error. */
4079		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4080		    errno, "can't get %s header len on packet socket",
4081		    version_str);
4082		return -1;
4083	}
4084	handlep->tp_hdrlen = val;
4085
4086	val = version;
4087	if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
4088			   sizeof(val)) < 0) {
4089		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4090		    errno, "can't activate %s on packet socket", version_str);
4091		return -1;
4092	}
4093	handlep->tp_version = version;
4094
4095	/* Reserve space for VLAN tag reconstruction */
4096	val = VLAN_TAG_LEN;
4097	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val,
4098			   sizeof(val)) < 0) {
4099		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4100		    errno, "can't set up reserve on packet socket");
4101		return -1;
4102	}
4103
4104	return 0;
4105}
4106#endif /* defined HAVE_TPACKET2 || defined HAVE_TPACKET3 */
4107
4108/*
4109 * If the instruction set for which we're compiling has both 32-bit
4110 * and 64-bit versions, and Linux support for the 64-bit version
4111 * predates TPACKET_V2, define ISA_64_BIT as the .machine value
4112 * you get from uname() for the 64-bit version.  Otherwise, leave
4113 * it undefined.  (This includes ARM, which has a 64-bit version,
4114 * but Linux support for it appeared well after TPACKET_V2 support
4115 * did, so there should never be a case where 32-bit ARM code is
4116 * running o a 64-bit kernel that only supports TPACKET_V1.)
4117 *
4118 * If we've omitted your favorite such architecture, please contribute
4119 * a patch.  (No patch is needed for architectures that are 32-bit-only
4120 * or for which Linux has no support for 32-bit userland - or for which,
4121 * as noted, 64-bit support appeared in Linux after TPACKET_V2 support
4122 * did.)
4123 */
4124#if defined(__i386__)
4125#define ISA_64_BIT	"x86_64"
4126#elif defined(__ppc__)
4127#define ISA_64_BIT	"ppc64"
4128#elif defined(__sparc__)
4129#define ISA_64_BIT	"sparc64"
4130#elif defined(__s390__)
4131#define ISA_64_BIT	"s390x"
4132#elif defined(__mips__)
4133#define ISA_64_BIT	"mips64"
4134#elif defined(__hppa__)
4135#define ISA_64_BIT	"parisc64"
4136#endif
4137
4138/*
4139 * Attempt to set the socket to version 3 of the memory-mapped header and,
4140 * if that fails because version 3 isn't supported, attempt to fall
4141 * back to version 2.  If version 2 isn't supported, just leave it at
4142 * version 1.
4143 *
4144 * Return 1 if we succeed or if we fail because neither version 2 nor 3 is
4145 * supported; return -1 on any other error, and set handle->errbuf.
4146 */
4147static int
4148prepare_tpacket_socket(pcap_t *handle)
4149{
4150	struct pcap_linux *handlep = handle->priv;
4151#if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
4152	int ret;
4153#endif
4154
4155#ifdef HAVE_TPACKET3
4156	/*
4157	 * Try setting the version to TPACKET_V3.
4158	 *
4159	 * The only mode in which buffering is done on PF_PACKET
4160	 * sockets, so that packets might not be delivered
4161	 * immediately, is TPACKET_V3 mode.
4162	 *
4163	 * The buffering cannot be disabled in that mode, so
4164	 * if the user has requested immediate mode, we don't
4165	 * use TPACKET_V3.
4166	 */
4167	if (!handle->opt.immediate) {
4168		ret = init_tpacket(handle, TPACKET_V3, "TPACKET_V3");
4169		if (ret == 0) {
4170			/*
4171			 * Success.
4172			 */
4173			return 1;
4174		}
4175		if (ret == -1) {
4176			/*
4177			 * We failed for some reason other than "the
4178			 * kernel doesn't support TPACKET_V3".
4179			 */
4180			return -1;
4181		}
4182	}
4183#endif /* HAVE_TPACKET3 */
4184
4185#ifdef HAVE_TPACKET2
4186	/*
4187	 * Try setting the version to TPACKET_V2.
4188	 */
4189	ret = init_tpacket(handle, TPACKET_V2, "TPACKET_V2");
4190	if (ret == 0) {
4191		/*
4192		 * Success.
4193		 */
4194		return 1;
4195	}
4196	if (ret == -1) {
4197		/*
4198		 * We failed for some reason other than "the
4199		 * kernel doesn't support TPACKET_V2".
4200		 */
4201		return -1;
4202	}
4203#endif /* HAVE_TPACKET2 */
4204
4205	/*
4206	 * OK, we're using TPACKET_V1, as that's all the kernel supports.
4207	 */
4208	handlep->tp_version = TPACKET_V1;
4209	handlep->tp_hdrlen = sizeof(struct tpacket_hdr);
4210
4211#ifdef ISA_64_BIT
4212	/*
4213	 * 32-bit userspace + 64-bit kernel + TPACKET_V1 are not compatible with
4214	 * each other due to platform-dependent data type size differences.
4215	 *
4216	 * If we have a 32-bit userland and a 64-bit kernel, use an
4217	 * internally-defined TPACKET_V1_64, with which we use a 64-bit
4218	 * version of the data structures.
4219	 */
4220	if (sizeof(long) == 4) {
4221		/*
4222		 * This is 32-bit code.
4223		 */
4224		struct utsname utsname;
4225
4226		if (uname(&utsname) == -1) {
4227			/*
4228			 * Failed.
4229			 */
4230			pcap_fmt_errmsg_for_errno(handle->errbuf,
4231			    PCAP_ERRBUF_SIZE, errno, "uname failed");
4232			return -1;
4233		}
4234		if (strcmp(utsname.machine, ISA_64_BIT) == 0) {
4235			/*
4236			 * uname() tells us the machine is 64-bit,
4237			 * so we presumably have a 64-bit kernel.
4238			 *
4239			 * XXX - this presumes that uname() won't lie
4240			 * in 32-bit code and claim that the machine
4241			 * has the 32-bit version of the ISA.
4242			 */
4243			handlep->tp_version = TPACKET_V1_64;
4244			handlep->tp_hdrlen = sizeof(struct tpacket_hdr_64);
4245		}
4246	}
4247#endif
4248
4249	return 1;
4250}
4251
4252#define MAX(a,b) ((a)>(b)?(a):(b))
4253
4254/*
4255 * Attempt to set up memory-mapped access.
4256 *
4257 * On success, returns 1, and sets *status to 0 if there are no warnings
4258 * or to a PCAP_WARNING_ code if there is a warning.
4259 *
4260 * On failure due to lack of support for memory-mapped capture, returns
4261 * 0.
4262 *
4263 * On error, returns -1, and sets *status to the appropriate error code;
4264 * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
4265 */
4266static int
4267create_ring(pcap_t *handle, int *status)
4268{
4269	struct pcap_linux *handlep = handle->priv;
4270	unsigned i, j, frames_per_block;
4271#ifdef HAVE_TPACKET3
4272	/*
4273	 * For sockets using TPACKET_V1 or TPACKET_V2, the extra
4274	 * stuff at the end of a struct tpacket_req3 will be
4275	 * ignored, so this is OK even for those sockets.
4276	 */
4277	struct tpacket_req3 req;
4278#else
4279	struct tpacket_req req;
4280#endif
4281	socklen_t len;
4282	unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
4283	unsigned int frame_size;
4284
4285	/*
4286	 * Start out assuming no warnings or errors.
4287	 */
4288	*status = 0;
4289
4290	switch (handlep->tp_version) {
4291
4292	case TPACKET_V1:
4293	case TPACKET_V1_64:
4294#ifdef HAVE_TPACKET2
4295	case TPACKET_V2:
4296#endif
4297		/* Note that with large snapshot length (say 256K, which is
4298		 * the default for recent versions of tcpdump, Wireshark,
4299		 * TShark, dumpcap or 64K, the value that "-s 0" has given for
4300		 * a long time with tcpdump), if we use the snapshot
4301		 * length to calculate the frame length, only a few frames
4302		 * will be available in the ring even with pretty
4303		 * large ring size (and a lot of memory will be unused).
4304		 *
4305		 * Ideally, we should choose a frame length based on the
4306		 * minimum of the specified snapshot length and the maximum
4307		 * packet size.  That's not as easy as it sounds; consider,
4308		 * for example, an 802.11 interface in monitor mode, where
4309		 * the frame would include a radiotap header, where the
4310		 * maximum radiotap header length is device-dependent.
4311		 *
4312		 * So, for now, we just do this for Ethernet devices, where
4313		 * there's no metadata header, and the link-layer header is
4314		 * fixed length.  We can get the maximum packet size by
4315		 * adding 18, the Ethernet header length plus the CRC length
4316		 * (just in case we happen to get the CRC in the packet), to
4317		 * the MTU of the interface; we fetch the MTU in the hopes
4318		 * that it reflects support for jumbo frames.  (Even if the
4319		 * interface is just being used for passive snooping, the
4320		 * driver might set the size of buffers in the receive ring
4321		 * based on the MTU, so that the MTU limits the maximum size
4322		 * of packets that we can receive.)
4323		 *
4324		 * If segmentation/fragmentation or receive offload are
4325		 * enabled, we can get reassembled/aggregated packets larger
4326		 * than MTU, but bounded to 65535 plus the Ethernet overhead,
4327		 * due to kernel and protocol constraints */
4328		frame_size = handle->snapshot;
4329		if (handle->linktype == DLT_EN10MB) {
4330			unsigned int max_frame_len;
4331			int mtu;
4332			int offload;
4333
4334			mtu = iface_get_mtu(handle->fd, handle->opt.device,
4335			    handle->errbuf);
4336			if (mtu == -1) {
4337				*status = PCAP_ERROR;
4338				return -1;
4339			}
4340			offload = iface_get_offload(handle);
4341			if (offload == -1) {
4342				*status = PCAP_ERROR;
4343				return -1;
4344			}
4345			if (offload)
4346				max_frame_len = MAX(mtu, 65535);
4347			else
4348				max_frame_len = mtu;
4349			max_frame_len += 18;
4350
4351			if (frame_size > max_frame_len)
4352				frame_size = max_frame_len;
4353		}
4354
4355		/* NOTE: calculus matching those in tpacket_rcv()
4356		 * in linux-2.6/net/packet/af_packet.c
4357		 */
4358		len = sizeof(sk_type);
4359		if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type,
4360		    &len) < 0) {
4361			pcap_fmt_errmsg_for_errno(handle->errbuf,
4362			    PCAP_ERRBUF_SIZE, errno, "getsockopt");
4363			*status = PCAP_ERROR;
4364			return -1;
4365		}
4366#ifdef PACKET_RESERVE
4367		len = sizeof(tp_reserve);
4368		if (getsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE,
4369		    &tp_reserve, &len) < 0) {
4370			if (errno != ENOPROTOOPT) {
4371				/*
4372				 * ENOPROTOOPT means "kernel doesn't support
4373				 * PACKET_RESERVE", in which case we fall back
4374				 * as best we can.
4375				 */
4376				pcap_fmt_errmsg_for_errno(handle->errbuf,
4377				    PCAP_ERRBUF_SIZE, errno, "getsockopt");
4378				*status = PCAP_ERROR;
4379				return -1;
4380			}
4381			tp_reserve = 0;	/* older kernel, reserve not supported */
4382		}
4383#else
4384		tp_reserve = 0;	/* older kernel, reserve not supported */
4385#endif
4386		maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
4387			/* XXX: in the kernel maclen is calculated from
4388			 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
4389			 * in:  packet_snd()           in linux-2.6/net/packet/af_packet.c
4390			 * then packet_alloc_skb()     in linux-2.6/net/packet/af_packet.c
4391			 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
4392			 * but I see no way to get those sizes in userspace,
4393			 * like for instance with an ifreq ioctl();
4394			 * the best thing I've found so far is MAX_HEADER in
4395			 * the kernel part of linux-2.6/include/linux/netdevice.h
4396			 * which goes up to 128+48=176; since pcap-linux.c
4397			 * defines a MAX_LINKHEADER_SIZE of 256 which is
4398			 * greater than that, let's use it.. maybe is it even
4399			 * large enough to directly replace macoff..
4400			 */
4401		tp_hdrlen = TPACKET_ALIGN(handlep->tp_hdrlen) + sizeof(struct sockaddr_ll) ;
4402		netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
4403			/* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN
4404			 * of netoff, which contradicts
4405			 * linux-2.6/Documentation/networking/packet_mmap.txt
4406			 * documenting that:
4407			 * "- Gap, chosen so that packet data (Start+tp_net)
4408			 * aligns to TPACKET_ALIGNMENT=16"
4409			 */
4410			/* NOTE: in linux-2.6/include/linux/skbuff.h:
4411			 * "CPUs often take a performance hit
4412			 *  when accessing unaligned memory locations"
4413			 */
4414		macoff = netoff - maclen;
4415		req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
4416		/*
4417		 * Round the buffer size up to a multiple of the
4418		 * frame size (rather than rounding down, which
4419		 * would give a buffer smaller than our caller asked
4420		 * for, and possibly give zero frames if the requested
4421		 * buffer size is too small for one frame).
4422		 */
4423		req.tp_frame_nr = (handle->opt.buffer_size + req.tp_frame_size - 1)/req.tp_frame_size;
4424		break;
4425
4426#ifdef HAVE_TPACKET3
4427	case TPACKET_V3:
4428		/* The "frames" for this are actually buffers that
4429		 * contain multiple variable-sized frames.
4430		 *
4431		 * We pick a "frame" size of MAXIMUM_SNAPLEN to leave
4432		 * enough room for at least one reasonably-sized packet
4433		 * in the "frame". */
4434		req.tp_frame_size = MAXIMUM_SNAPLEN;
4435		/*
4436		 * Round the buffer size up to a multiple of the
4437		 * "frame" size (rather than rounding down, which
4438		 * would give a buffer smaller than our caller asked
4439		 * for, and possibly give zero "frames" if the requested
4440		 * buffer size is too small for one "frame").
4441		 */
4442		req.tp_frame_nr = (handle->opt.buffer_size + req.tp_frame_size - 1)/req.tp_frame_size;
4443		break;
4444#endif
4445	default:
4446		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4447		    "Internal error: unknown TPACKET_ value %u",
4448		    handlep->tp_version);
4449		*status = PCAP_ERROR;
4450		return -1;
4451	}
4452
4453	/* compute the minumum block size that will handle this frame.
4454	 * The block has to be page size aligned.
4455	 * The max block size allowed by the kernel is arch-dependent and
4456	 * it's not explicitly checked here. */
4457	req.tp_block_size = getpagesize();
4458	while (req.tp_block_size < req.tp_frame_size)
4459		req.tp_block_size <<= 1;
4460
4461	frames_per_block = req.tp_block_size/req.tp_frame_size;
4462
4463	/*
4464	 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
4465	 * so we check for PACKET_TIMESTAMP.  We check for
4466	 * linux/net_tstamp.h just in case a system somehow has
4467	 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
4468	 * be unnecessary.
4469	 *
4470	 * SIOCSHWTSTAMP was introduced in the patch that introduced
4471	 * linux/net_tstamp.h, so we don't bother checking whether
4472	 * SIOCSHWTSTAMP is defined (if your Linux system has
4473	 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
4474	 * Linux system is badly broken).
4475	 */
4476#if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
4477	/*
4478	 * If we were told to do so, ask the kernel and the driver
4479	 * to use hardware timestamps.
4480	 *
4481	 * Hardware timestamps are only supported with mmapped
4482	 * captures.
4483	 */
4484	if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
4485	    handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
4486		struct hwtstamp_config hwconfig;
4487		struct ifreq ifr;
4488		int timesource;
4489
4490		/*
4491		 * Ask for hardware time stamps on all packets,
4492		 * including transmitted packets.
4493		 */
4494		memset(&hwconfig, 0, sizeof(hwconfig));
4495		hwconfig.tx_type = HWTSTAMP_TX_ON;
4496		hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
4497
4498		memset(&ifr, 0, sizeof(ifr));
4499		strlcpy(ifr.ifr_name, handle->opt.device, sizeof(ifr.ifr_name));
4500		ifr.ifr_data = (void *)&hwconfig;
4501
4502		if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
4503			switch (errno) {
4504
4505			case EPERM:
4506				/*
4507				 * Treat this as an error, as the
4508				 * user should try to run this
4509				 * with the appropriate privileges -
4510				 * and, if they can't, shouldn't
4511				 * try requesting hardware time stamps.
4512				 */
4513				*status = PCAP_ERROR_PERM_DENIED;
4514				return -1;
4515
4516			case EOPNOTSUPP:
4517			case ERANGE:
4518				/*
4519				 * Treat this as a warning, as the
4520				 * only way to fix the warning is to
4521				 * get an adapter that supports hardware
4522				 * time stamps for *all* packets.
4523				 * (ERANGE means "we support hardware
4524				 * time stamps, but for packets matching
4525				 * that particular filter", so it means
4526				 * "we don't support hardware time stamps
4527				 * for all incoming packets" here.)
4528				 *
4529				 * We'll just fall back on the standard
4530				 * host time stamps.
4531				 */
4532				*status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
4533				break;
4534
4535			default:
4536				pcap_fmt_errmsg_for_errno(handle->errbuf,
4537				    PCAP_ERRBUF_SIZE, errno,
4538				    "SIOCSHWTSTAMP failed");
4539				*status = PCAP_ERROR;
4540				return -1;
4541			}
4542		} else {
4543			/*
4544			 * Well, that worked.  Now specify the type of
4545			 * hardware time stamp we want for this
4546			 * socket.
4547			 */
4548			if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
4549				/*
4550				 * Hardware timestamp, synchronized
4551				 * with the system clock.
4552				 */
4553				timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
4554			} else {
4555				/*
4556				 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
4557				 * timestamp, not synchronized with the
4558				 * system clock.
4559				 */
4560				timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
4561			}
4562			if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
4563				(void *)&timesource, sizeof(timesource))) {
4564				pcap_fmt_errmsg_for_errno(handle->errbuf,
4565				    PCAP_ERRBUF_SIZE, errno,
4566				    "can't set PACKET_TIMESTAMP");
4567				*status = PCAP_ERROR;
4568				return -1;
4569			}
4570		}
4571	}
4572#endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
4573
4574	/* ask the kernel to create the ring */
4575retry:
4576	req.tp_block_nr = req.tp_frame_nr / frames_per_block;
4577
4578	/* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
4579	req.tp_frame_nr = req.tp_block_nr * frames_per_block;
4580
4581#ifdef HAVE_TPACKET3
4582	/* timeout value to retire block - use the configured buffering timeout, or default if <0. */
4583	req.tp_retire_blk_tov = (handlep->timeout>=0)?handlep->timeout:0;
4584	/* private data not used */
4585	req.tp_sizeof_priv = 0;
4586	/* Rx ring - feature request bits - none (rxhash will not be filled) */
4587	req.tp_feature_req_word = 0;
4588#endif
4589
4590	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
4591					(void *) &req, sizeof(req))) {
4592		if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
4593			/*
4594			 * Memory failure; try to reduce the requested ring
4595			 * size.
4596			 *
4597			 * We used to reduce this by half -- do 5% instead.
4598			 * That may result in more iterations and a longer
4599			 * startup, but the user will be much happier with
4600			 * the resulting buffer size.
4601			 */
4602			if (req.tp_frame_nr < 20)
4603				req.tp_frame_nr -= 1;
4604			else
4605				req.tp_frame_nr -= req.tp_frame_nr/20;
4606			goto retry;
4607		}
4608		if (errno == ENOPROTOOPT) {
4609			/*
4610			 * We don't have ring buffer support in this kernel.
4611			 */
4612			return 0;
4613		}
4614		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4615		    errno, "can't create rx ring on packet socket");
4616		*status = PCAP_ERROR;
4617		return -1;
4618	}
4619
4620	/* memory map the rx ring */
4621	handlep->mmapbuflen = req.tp_block_nr * req.tp_block_size;
4622	handlep->mmapbuf = mmap(0, handlep->mmapbuflen,
4623	    PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
4624	if (handlep->mmapbuf == MAP_FAILED) {
4625		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4626		    errno, "can't mmap rx ring");
4627
4628		/* clear the allocated ring on error*/
4629		destroy_ring(handle);
4630		*status = PCAP_ERROR;
4631		return -1;
4632	}
4633
4634	/* allocate a ring for each frame header pointer*/
4635	handle->cc = req.tp_frame_nr;
4636	handle->buffer = malloc(handle->cc * sizeof(union thdr *));
4637	if (!handle->buffer) {
4638		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4639		    errno, "can't allocate ring of frame headers");
4640
4641		destroy_ring(handle);
4642		*status = PCAP_ERROR;
4643		return -1;
4644	}
4645
4646	/* fill the header ring with proper frame ptr*/
4647	handle->offset = 0;
4648	for (i=0; i<req.tp_block_nr; ++i) {
4649		void *base = &handlep->mmapbuf[i*req.tp_block_size];
4650		for (j=0; j<frames_per_block; ++j, ++handle->offset) {
4651			RING_GET_CURRENT_FRAME(handle) = base;
4652			base += req.tp_frame_size;
4653		}
4654	}
4655
4656	handle->bufsize = req.tp_frame_size;
4657	handle->offset = 0;
4658	return 1;
4659}
4660
4661/* free all ring related resources*/
4662static void
4663destroy_ring(pcap_t *handle)
4664{
4665	struct pcap_linux *handlep = handle->priv;
4666
4667	/* tell the kernel to destroy the ring*/
4668	struct tpacket_req req;
4669	memset(&req, 0, sizeof(req));
4670	/* do not test for setsockopt failure, as we can't recover from any error */
4671	(void)setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
4672				(void *) &req, sizeof(req));
4673
4674	/* if ring is mapped, unmap it*/
4675	if (handlep->mmapbuf) {
4676		/* do not test for mmap failure, as we can't recover from any error */
4677		(void)munmap(handlep->mmapbuf, handlep->mmapbuflen);
4678		handlep->mmapbuf = NULL;
4679	}
4680}
4681
4682/*
4683 * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
4684 * for Linux mmapped capture.
4685 *
4686 * The problem is that pcap_next() and pcap_next_ex() expect the packet
4687 * data handed to the callback to be valid after the callback returns,
4688 * but pcap_read_linux_mmap() has to release that packet as soon as
4689 * the callback returns (otherwise, the kernel thinks there's still
4690 * at least one unprocessed packet available in the ring, so a select()
4691 * will immediately return indicating that there's data to process), so,
4692 * in the callback, we have to make a copy of the packet.
4693 *
4694 * Yes, this means that, if the capture is using the ring buffer, using
4695 * pcap_next() or pcap_next_ex() requires more copies than using
4696 * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
4697 * pcap_next() or pcap_next_ex().
4698 */
4699static void
4700pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
4701    const u_char *bytes)
4702{
4703	struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
4704	pcap_t *handle = sp->pd;
4705	struct pcap_linux *handlep = handle->priv;
4706
4707	*sp->hdr = *h;
4708	memcpy(handlep->oneshot_buffer, bytes, h->caplen);
4709	*sp->pkt = handlep->oneshot_buffer;
4710}
4711
4712static void
4713pcap_cleanup_linux_mmap( pcap_t *handle )
4714{
4715	struct pcap_linux *handlep = handle->priv;
4716
4717	destroy_ring(handle);
4718	if (handlep->oneshot_buffer != NULL) {
4719		free(handlep->oneshot_buffer);
4720		handlep->oneshot_buffer = NULL;
4721	}
4722	pcap_cleanup_linux(handle);
4723}
4724
4725
4726static int
4727pcap_getnonblock_mmap(pcap_t *handle)
4728{
4729	struct pcap_linux *handlep = handle->priv;
4730
4731	/* use negative value of timeout to indicate non blocking ops */
4732	return (handlep->timeout<0);
4733}
4734
4735static int
4736pcap_setnonblock_mmap(pcap_t *handle, int nonblock)
4737{
4738	struct pcap_linux *handlep = handle->priv;
4739
4740	/*
4741	 * Set the file descriptor to non-blocking mode, as we use
4742	 * it for sending packets.
4743	 */
4744	if (pcap_setnonblock_fd(handle, nonblock) == -1)
4745		return -1;
4746
4747	/*
4748	 * Map each value to their corresponding negation to
4749	 * preserve the timeout value provided with pcap_set_timeout.
4750	 */
4751	if (nonblock) {
4752		if (handlep->timeout >= 0) {
4753			/*
4754			 * Indicate that we're switching to
4755			 * non-blocking mode.
4756			 */
4757			handlep->timeout = ~handlep->timeout;
4758		}
4759	} else {
4760		if (handlep->timeout < 0) {
4761			handlep->timeout = ~handlep->timeout;
4762		}
4763	}
4764	/* Update the timeout to use in poll(). */
4765	set_poll_timeout(handlep);
4766	return 0;
4767}
4768
4769/*
4770 * Get the status field of the ring buffer frame at a specified offset.
4771 */
4772static inline int
4773pcap_get_ring_frame_status(pcap_t *handle, int offset)
4774{
4775	struct pcap_linux *handlep = handle->priv;
4776	union thdr h;
4777
4778	h.raw = RING_GET_FRAME_AT(handle, offset);
4779	switch (handlep->tp_version) {
4780	case TPACKET_V1:
4781		return (h.h1->tp_status);
4782		break;
4783	case TPACKET_V1_64:
4784		return (h.h1_64->tp_status);
4785		break;
4786#ifdef HAVE_TPACKET2
4787	case TPACKET_V2:
4788		return (h.h2->tp_status);
4789		break;
4790#endif
4791#ifdef HAVE_TPACKET3
4792	case TPACKET_V3:
4793		return (h.h3->hdr.bh1.block_status);
4794		break;
4795#endif
4796	}
4797	/* This should not happen. */
4798	return 0;
4799}
4800
4801#ifndef POLLRDHUP
4802#define POLLRDHUP 0
4803#endif
4804
4805/*
4806 * Block waiting for frames to be available.
4807 */
4808static int pcap_wait_for_frames_mmap(pcap_t *handle)
4809{
4810	struct pcap_linux *handlep = handle->priv;
4811	char c;
4812	struct pollfd pollinfo;
4813	int ret;
4814
4815	pollinfo.fd = handle->fd;
4816	pollinfo.events = POLLIN;
4817
4818	do {
4819		/*
4820		 * Yes, we do this even in non-blocking mode, as it's
4821		 * the only way to get error indications from a
4822		 * tpacket socket.
4823		 *
4824		 * The timeout is 0 in non-blocking mode, so poll()
4825		 * returns immediately.
4826		 */
4827		ret = poll(&pollinfo, 1, handlep->poll_timeout);
4828		if (ret < 0 && errno != EINTR) {
4829			pcap_fmt_errmsg_for_errno(handle->errbuf,
4830			    PCAP_ERRBUF_SIZE, errno,
4831			    "can't poll on packet socket");
4832			return PCAP_ERROR;
4833		} else if (ret > 0 &&
4834			(pollinfo.revents & (POLLHUP|POLLRDHUP|POLLERR|POLLNVAL))) {
4835			/*
4836			 * There's some indication other than
4837			 * "you can read on this descriptor" on
4838			 * the descriptor.
4839			 */
4840			if (pollinfo.revents & (POLLHUP | POLLRDHUP)) {
4841				pcap_snprintf(handle->errbuf,
4842					PCAP_ERRBUF_SIZE,
4843					"Hangup on packet socket");
4844				return PCAP_ERROR;
4845			}
4846			if (pollinfo.revents & POLLERR) {
4847				/*
4848				 * A recv() will give us the actual error code.
4849				 *
4850				 * XXX - make the socket non-blocking?
4851				 */
4852				if (recv(handle->fd, &c, sizeof c,
4853					MSG_PEEK) != -1)
4854					continue;	/* what, no error? */
4855				if (errno == ENETDOWN) {
4856					/*
4857					 * The device on which we're
4858					 * capturing went away.
4859					 *
4860					 * XXX - we should really return
4861					 * PCAP_ERROR_IFACE_NOT_UP, but
4862					 * pcap_dispatch() etc. aren't
4863					 * defined to return that.
4864					 */
4865					pcap_snprintf(handle->errbuf,
4866						PCAP_ERRBUF_SIZE,
4867						"The interface went down");
4868				} else {
4869					pcap_fmt_errmsg_for_errno(handle->errbuf,
4870					    PCAP_ERRBUF_SIZE, errno,
4871					    "Error condition on packet socket");
4872				}
4873				return PCAP_ERROR;
4874			}
4875			if (pollinfo.revents & POLLNVAL) {
4876				pcap_snprintf(handle->errbuf,
4877					PCAP_ERRBUF_SIZE,
4878					"Invalid polling request on packet socket");
4879				return PCAP_ERROR;
4880			}
4881		}
4882		/* check for break loop condition on interrupted syscall*/
4883		if (handle->break_loop) {
4884			handle->break_loop = 0;
4885			return PCAP_ERROR_BREAK;
4886		}
4887	} while (ret < 0);
4888	return 0;
4889}
4890
4891/* handle a single memory mapped packet */
4892static int pcap_handle_packet_mmap(
4893		pcap_t *handle,
4894		pcap_handler callback,
4895		u_char *user,
4896		unsigned char *frame,
4897		unsigned int tp_len,
4898		unsigned int tp_mac,
4899		unsigned int tp_snaplen,
4900		unsigned int tp_sec,
4901		unsigned int tp_usec,
4902		int tp_vlan_tci_valid,
4903		__u16 tp_vlan_tci,
4904		__u16 tp_vlan_tpid)
4905{
4906	struct pcap_linux *handlep = handle->priv;
4907	unsigned char *bp;
4908	struct sockaddr_ll *sll;
4909	struct pcap_pkthdr pcaphdr;
4910	unsigned int snaplen = tp_snaplen;
4911
4912	/* perform sanity check on internal offset. */
4913	if (tp_mac + tp_snaplen > handle->bufsize) {
4914		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4915			"corrupted frame on kernel ring mac "
4916			"offset %u + caplen %u > frame len %d",
4917			tp_mac, tp_snaplen, handle->bufsize);
4918		return -1;
4919	}
4920
4921	/* run filter on received packet
4922	 * If the kernel filtering is enabled we need to run the
4923	 * filter until all the frames present into the ring
4924	 * at filter creation time are processed.
4925	 * In this case, blocks_to_filter_in_userland is used
4926	 * as a counter for the packet we need to filter.
4927	 * Note: alternatively it could be possible to stop applying
4928	 * the filter when the ring became empty, but it can possibly
4929	 * happen a lot later... */
4930	bp = frame + tp_mac;
4931
4932	/* if required build in place the sll header*/
4933	sll = (void *)frame + TPACKET_ALIGN(handlep->tp_hdrlen);
4934	if (handlep->cooked) {
4935		struct sll_header *hdrp;
4936
4937		/*
4938		 * The kernel should have left us with enough
4939		 * space for an sll header; back up the packet
4940		 * data pointer into that space, as that'll be
4941		 * the beginning of the packet we pass to the
4942		 * callback.
4943		 */
4944		bp -= SLL_HDR_LEN;
4945
4946		/*
4947		 * Let's make sure that's past the end of
4948		 * the tpacket header, i.e. >=
4949		 * ((u_char *)thdr + TPACKET_HDRLEN), so we
4950		 * don't step on the header when we construct
4951		 * the sll header.
4952		 */
4953		if (bp < (u_char *)frame +
4954				   TPACKET_ALIGN(handlep->tp_hdrlen) +
4955				   sizeof(struct sockaddr_ll)) {
4956			pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4957				"cooked-mode frame doesn't have room for sll header");
4958			return -1;
4959		}
4960
4961		/*
4962		 * OK, that worked; construct the sll header.
4963		 */
4964		hdrp = (struct sll_header *)bp;
4965		hdrp->sll_pkttype = map_packet_type_to_sll_type(
4966						sll->sll_pkttype);
4967		hdrp->sll_hatype = htons(sll->sll_hatype);
4968		hdrp->sll_halen = htons(sll->sll_halen);
4969		memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
4970		hdrp->sll_protocol = sll->sll_protocol;
4971
4972		snaplen += sizeof(struct sll_header);
4973	}
4974
4975	if (handlep->filter_in_userland && handle->fcode.bf_insns) {
4976		struct bpf_aux_data aux_data;
4977
4978		aux_data.vlan_tag_present = tp_vlan_tci_valid;
4979		aux_data.vlan_tag = tp_vlan_tci & 0x0fff;
4980
4981		if (bpf_filter_with_aux_data(handle->fcode.bf_insns,
4982					     bp,
4983					     tp_len,
4984					     snaplen,
4985					     &aux_data) == 0)
4986			return 0;
4987	}
4988
4989	if (!linux_check_direction(handle, sll))
4990		return 0;
4991
4992	/* get required packet info from ring header */
4993	pcaphdr.ts.tv_sec = tp_sec;
4994	pcaphdr.ts.tv_usec = tp_usec;
4995	pcaphdr.caplen = tp_snaplen;
4996	pcaphdr.len = tp_len;
4997
4998	/* if required build in place the sll header*/
4999	if (handlep->cooked) {
5000		/* update packet len */
5001		pcaphdr.caplen += SLL_HDR_LEN;
5002		pcaphdr.len += SLL_HDR_LEN;
5003	}
5004
5005#if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
5006	if (tp_vlan_tci_valid &&
5007		handlep->vlan_offset != -1 &&
5008		tp_snaplen >= (unsigned int) handlep->vlan_offset)
5009	{
5010		struct vlan_tag *tag;
5011
5012		/*
5013		 * Move everything in the header, except the type field,
5014		 * down VLAN_TAG_LEN bytes, to allow us to insert the
5015		 * VLAN tag between that stuff and the type field.
5016		 */
5017		bp -= VLAN_TAG_LEN;
5018		memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
5019
5020		/*
5021		 * Now insert the tag.
5022		 */
5023		tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
5024		tag->vlan_tpid = htons(tp_vlan_tpid);
5025		tag->vlan_tci = htons(tp_vlan_tci);
5026
5027		/*
5028		 * Add the tag to the packet lengths.
5029		 */
5030		pcaphdr.caplen += VLAN_TAG_LEN;
5031		pcaphdr.len += VLAN_TAG_LEN;
5032	}
5033#endif
5034
5035	/*
5036	 * The only way to tell the kernel to cut off the
5037	 * packet at a snapshot length is with a filter program;
5038	 * if there's no filter program, the kernel won't cut
5039	 * the packet off.
5040	 *
5041	 * Trim the snapshot length to be no longer than the
5042	 * specified snapshot length.
5043	 */
5044	if (pcaphdr.caplen > (bpf_u_int32)handle->snapshot)
5045		pcaphdr.caplen = handle->snapshot;
5046
5047	/* pass the packet to the user */
5048	callback(user, &pcaphdr, bp);
5049
5050	return 1;
5051}
5052
5053static int
5054pcap_read_linux_mmap_v1(pcap_t *handle, int max_packets, pcap_handler callback,
5055		u_char *user)
5056{
5057	struct pcap_linux *handlep = handle->priv;
5058	union thdr h;
5059	int pkts = 0;
5060	int ret;
5061
5062	/* wait for frames availability.*/
5063	h.raw = RING_GET_CURRENT_FRAME(handle);
5064	if (h.h1->tp_status == TP_STATUS_KERNEL) {
5065		/*
5066		 * The current frame is owned by the kernel; wait for
5067		 * a frame to be handed to us.
5068		 */
5069		ret = pcap_wait_for_frames_mmap(handle);
5070		if (ret) {
5071			return ret;
5072		}
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		/*
5079		 * Get the current ring buffer frame, and break if
5080		 * it's still owned by the kernel.
5081		 */
5082		h.raw = RING_GET_CURRENT_FRAME(handle);
5083		if (h.h1->tp_status == TP_STATUS_KERNEL)
5084			break;
5085
5086		ret = pcap_handle_packet_mmap(
5087				handle,
5088				callback,
5089				user,
5090				h.raw,
5091				h.h1->tp_len,
5092				h.h1->tp_mac,
5093				h.h1->tp_snaplen,
5094				h.h1->tp_sec,
5095				h.h1->tp_usec,
5096				0,
5097				0,
5098				0);
5099		if (ret == 1) {
5100			pkts++;
5101			handlep->packets_read++;
5102		} else if (ret < 0) {
5103			return ret;
5104		}
5105
5106		/*
5107		 * Hand this block back to the kernel, and, if we're
5108		 * counting blocks that need to be filtered in userland
5109		 * after having been filtered by the kernel, count
5110		 * the one we've just processed.
5111		 */
5112		h.h1->tp_status = TP_STATUS_KERNEL;
5113		if (handlep->blocks_to_filter_in_userland > 0) {
5114			handlep->blocks_to_filter_in_userland--;
5115			if (handlep->blocks_to_filter_in_userland == 0) {
5116				/*
5117				 * No more blocks need to be filtered
5118				 * in userland.
5119				 */
5120				handlep->filter_in_userland = 0;
5121			}
5122		}
5123
5124		/* next block */
5125		if (++handle->offset >= handle->cc)
5126			handle->offset = 0;
5127
5128		/* check for break loop condition*/
5129		if (handle->break_loop) {
5130			handle->break_loop = 0;
5131			return PCAP_ERROR_BREAK;
5132		}
5133	}
5134	return pkts;
5135}
5136
5137static int
5138pcap_read_linux_mmap_v1_64(pcap_t *handle, int max_packets, pcap_handler callback,
5139		u_char *user)
5140{
5141	struct pcap_linux *handlep = handle->priv;
5142	union thdr h;
5143	int pkts = 0;
5144	int ret;
5145
5146	/* wait for frames availability.*/
5147	h.raw = RING_GET_CURRENT_FRAME(handle);
5148	if (h.h1_64->tp_status == TP_STATUS_KERNEL) {
5149		/*
5150		 * The current frame is owned by the kernel; wait for
5151		 * a frame to be handed to us.
5152		 */
5153		ret = pcap_wait_for_frames_mmap(handle);
5154		if (ret) {
5155			return ret;
5156		}
5157	}
5158
5159	/* non-positive values of max_packets are used to require all
5160	 * packets currently available in the ring */
5161	while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
5162		/*
5163		 * Get the current ring buffer frame, and break if
5164		 * it's still owned by the kernel.
5165		 */
5166		h.raw = RING_GET_CURRENT_FRAME(handle);
5167		if (h.h1_64->tp_status == TP_STATUS_KERNEL)
5168			break;
5169
5170		ret = pcap_handle_packet_mmap(
5171				handle,
5172				callback,
5173				user,
5174				h.raw,
5175				h.h1_64->tp_len,
5176				h.h1_64->tp_mac,
5177				h.h1_64->tp_snaplen,
5178				h.h1_64->tp_sec,
5179				h.h1_64->tp_usec,
5180				0,
5181				0,
5182				0);
5183		if (ret == 1) {
5184			pkts++;
5185			handlep->packets_read++;
5186		} else if (ret < 0) {
5187			return ret;
5188		}
5189
5190		/*
5191		 * Hand this block back to the kernel, and, if we're
5192		 * counting blocks that need to be filtered in userland
5193		 * after having been filtered by the kernel, count
5194		 * the one we've just processed.
5195		 */
5196		h.h1_64->tp_status = TP_STATUS_KERNEL;
5197		if (handlep->blocks_to_filter_in_userland > 0) {
5198			handlep->blocks_to_filter_in_userland--;
5199			if (handlep->blocks_to_filter_in_userland == 0) {
5200				/*
5201				 * No more blocks need to be filtered
5202				 * in userland.
5203				 */
5204				handlep->filter_in_userland = 0;
5205			}
5206		}
5207
5208		/* next block */
5209		if (++handle->offset >= handle->cc)
5210			handle->offset = 0;
5211
5212		/* check for break loop condition*/
5213		if (handle->break_loop) {
5214			handle->break_loop = 0;
5215			return PCAP_ERROR_BREAK;
5216		}
5217	}
5218	return pkts;
5219}
5220
5221#ifdef HAVE_TPACKET2
5222static int
5223pcap_read_linux_mmap_v2(pcap_t *handle, int max_packets, pcap_handler callback,
5224		u_char *user)
5225{
5226	struct pcap_linux *handlep = handle->priv;
5227	union thdr h;
5228	int pkts = 0;
5229	int ret;
5230
5231	/* wait for frames availability.*/
5232	h.raw = RING_GET_CURRENT_FRAME(handle);
5233	if (h.h2->tp_status == TP_STATUS_KERNEL) {
5234		/*
5235		 * The current frame is owned by the kernel; wait for
5236		 * a frame to be handed to us.
5237		 */
5238		ret = pcap_wait_for_frames_mmap(handle);
5239		if (ret) {
5240			return ret;
5241		}
5242	}
5243
5244	/* non-positive values of max_packets are used to require all
5245	 * packets currently available in the ring */
5246	while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
5247		/*
5248		 * Get the current ring buffer frame, and break if
5249		 * it's still owned by the kernel.
5250		 */
5251		h.raw = RING_GET_CURRENT_FRAME(handle);
5252		if (h.h2->tp_status == TP_STATUS_KERNEL)
5253			break;
5254
5255		ret = pcap_handle_packet_mmap(
5256				handle,
5257				callback,
5258				user,
5259				h.raw,
5260				h.h2->tp_len,
5261				h.h2->tp_mac,
5262				h.h2->tp_snaplen,
5263				h.h2->tp_sec,
5264				handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? h.h2->tp_nsec : h.h2->tp_nsec / 1000,
5265				VLAN_VALID(h.h2, h.h2),
5266				h.h2->tp_vlan_tci,
5267				VLAN_TPID(h.h2, h.h2));
5268		if (ret == 1) {
5269			pkts++;
5270			handlep->packets_read++;
5271		} else if (ret < 0) {
5272			return ret;
5273		}
5274
5275		/*
5276		 * Hand this block back to the kernel, and, if we're
5277		 * counting blocks that need to be filtered in userland
5278		 * after having been filtered by the kernel, count
5279		 * the one we've just processed.
5280		 */
5281		h.h2->tp_status = TP_STATUS_KERNEL;
5282		if (handlep->blocks_to_filter_in_userland > 0) {
5283			handlep->blocks_to_filter_in_userland--;
5284			if (handlep->blocks_to_filter_in_userland == 0) {
5285				/*
5286				 * No more blocks need to be filtered
5287				 * in userland.
5288				 */
5289				handlep->filter_in_userland = 0;
5290			}
5291		}
5292
5293		/* next block */
5294		if (++handle->offset >= handle->cc)
5295			handle->offset = 0;
5296
5297		/* check for break loop condition*/
5298		if (handle->break_loop) {
5299			handle->break_loop = 0;
5300			return PCAP_ERROR_BREAK;
5301		}
5302	}
5303	return pkts;
5304}
5305#endif /* HAVE_TPACKET2 */
5306
5307#ifdef HAVE_TPACKET3
5308static int
5309pcap_read_linux_mmap_v3(pcap_t *handle, int max_packets, pcap_handler callback,
5310		u_char *user)
5311{
5312	struct pcap_linux *handlep = handle->priv;
5313	union thdr h;
5314	int pkts = 0;
5315	int ret;
5316
5317again:
5318	if (handlep->current_packet == NULL) {
5319		/* wait for frames availability.*/
5320		h.raw = RING_GET_CURRENT_FRAME(handle);
5321		if (h.h3->hdr.bh1.block_status == TP_STATUS_KERNEL) {
5322			/*
5323			 * The current frame is owned by the kernel; wait
5324			 * for a frame to be handed to us.
5325			 */
5326			ret = pcap_wait_for_frames_mmap(handle);
5327			if (ret) {
5328				return ret;
5329			}
5330		}
5331	}
5332	h.raw = RING_GET_CURRENT_FRAME(handle);
5333	if (h.h3->hdr.bh1.block_status == TP_STATUS_KERNEL) {
5334		if (pkts == 0 && handlep->timeout == 0) {
5335			/* Block until we see a packet. */
5336			goto again;
5337		}
5338		return pkts;
5339	}
5340
5341	/* non-positive values of max_packets are used to require all
5342	 * packets currently available in the ring */
5343	while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
5344		int packets_to_read;
5345
5346		if (handlep->current_packet == NULL) {
5347			h.raw = RING_GET_CURRENT_FRAME(handle);
5348			if (h.h3->hdr.bh1.block_status == TP_STATUS_KERNEL)
5349				break;
5350
5351			handlep->current_packet = h.raw + h.h3->hdr.bh1.offset_to_first_pkt;
5352			handlep->packets_left = h.h3->hdr.bh1.num_pkts;
5353		}
5354		packets_to_read = handlep->packets_left;
5355
5356		if (!PACKET_COUNT_IS_UNLIMITED(max_packets) &&
5357		    packets_to_read > (max_packets - pkts)) {
5358			/*
5359			 * We've been given a maximum number of packets
5360			 * to process, and there are more packets in
5361			 * this buffer than that.  Only process enough
5362			 * of them to get us up to that maximum.
5363			 */
5364			packets_to_read = max_packets - pkts;
5365		}
5366
5367		while (packets_to_read-- && !handle->break_loop) {
5368			struct tpacket3_hdr* tp3_hdr = (struct tpacket3_hdr*) handlep->current_packet;
5369			ret = pcap_handle_packet_mmap(
5370					handle,
5371					callback,
5372					user,
5373					handlep->current_packet,
5374					tp3_hdr->tp_len,
5375					tp3_hdr->tp_mac,
5376					tp3_hdr->tp_snaplen,
5377					tp3_hdr->tp_sec,
5378					handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? tp3_hdr->tp_nsec : tp3_hdr->tp_nsec / 1000,
5379					VLAN_VALID(tp3_hdr, &tp3_hdr->hv1),
5380					tp3_hdr->hv1.tp_vlan_tci,
5381					VLAN_TPID(tp3_hdr, &tp3_hdr->hv1));
5382			if (ret == 1) {
5383				pkts++;
5384				handlep->packets_read++;
5385			} else if (ret < 0) {
5386				handlep->current_packet = NULL;
5387				return ret;
5388			}
5389			handlep->current_packet += tp3_hdr->tp_next_offset;
5390			handlep->packets_left--;
5391		}
5392
5393		if (handlep->packets_left <= 0) {
5394			/*
5395			 * Hand this block back to the kernel, and, if
5396			 * we're counting blocks that need to be
5397			 * filtered in userland after having been
5398			 * filtered by the kernel, count the one we've
5399			 * just processed.
5400			 */
5401			h.h3->hdr.bh1.block_status = TP_STATUS_KERNEL;
5402			if (handlep->blocks_to_filter_in_userland > 0) {
5403				handlep->blocks_to_filter_in_userland--;
5404				if (handlep->blocks_to_filter_in_userland == 0) {
5405					/*
5406					 * No more blocks need to be filtered
5407					 * in userland.
5408					 */
5409					handlep->filter_in_userland = 0;
5410				}
5411			}
5412
5413			/* next block */
5414			if (++handle->offset >= handle->cc)
5415				handle->offset = 0;
5416
5417			handlep->current_packet = NULL;
5418		}
5419
5420		/* check for break loop condition*/
5421		if (handle->break_loop) {
5422			handle->break_loop = 0;
5423			return PCAP_ERROR_BREAK;
5424		}
5425	}
5426	if (pkts == 0 && handlep->timeout == 0) {
5427		/* Block until we see a packet. */
5428		goto again;
5429	}
5430	return pkts;
5431}
5432#endif /* HAVE_TPACKET3 */
5433
5434static int
5435pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
5436{
5437	struct pcap_linux *handlep = handle->priv;
5438	int n, offset;
5439	int ret;
5440
5441	/*
5442	 * Don't rewrite "ret" instructions; we don't need to, as
5443	 * we're not reading packets with recvmsg(), and we don't
5444	 * want to, as, by not rewriting them, the kernel can avoid
5445	 * copying extra data.
5446	 */
5447	ret = pcap_setfilter_linux_common(handle, filter, 1);
5448	if (ret < 0)
5449		return ret;
5450
5451	/*
5452	 * If we're filtering in userland, there's nothing to do;
5453	 * the new filter will be used for the next packet.
5454	 */
5455	if (handlep->filter_in_userland)
5456		return ret;
5457
5458	/*
5459	 * We're filtering in the kernel; the packets present in
5460	 * all blocks currently in the ring were already filtered
5461	 * by the old filter, and so will need to be filtered in
5462	 * userland by the new filter.
5463	 *
5464	 * Get an upper bound for the number of such blocks; first,
5465	 * walk the ring backward and count the free blocks.
5466	 */
5467	offset = handle->offset;
5468	if (--offset < 0)
5469		offset = handle->cc - 1;
5470	for (n=0; n < handle->cc; ++n) {
5471		if (--offset < 0)
5472			offset = handle->cc - 1;
5473		if (pcap_get_ring_frame_status(handle, offset) != TP_STATUS_KERNEL)
5474			break;
5475	}
5476
5477	/*
5478	 * If we found free blocks, decrement the count of free
5479	 * blocks by 1, just in case we lost a race with another
5480	 * thread of control that was adding a packet while
5481	 * we were counting and that had run the filter before
5482	 * we changed it.
5483	 *
5484	 * XXX - could there be more than one block added in
5485	 * this fashion?
5486	 *
5487	 * XXX - is there a way to avoid that race, e.g. somehow
5488	 * wait for all packets that passed the old filter to
5489	 * be added to the ring?
5490	 */
5491	if (n != 0)
5492		n--;
5493
5494	/*
5495	 * Set the count of blocks worth of packets to filter
5496	 * in userland to the total number of blocks in the
5497	 * ring minus the number of free blocks we found, and
5498	 * turn on userland filtering.  (The count of blocks
5499	 * worth of packets to filter in userland is guaranteed
5500	 * not to be zero - n, above, couldn't be set to a
5501	 * value > handle->cc, and if it were equal to
5502	 * handle->cc, it wouldn't be zero, and thus would
5503	 * be decremented to handle->cc - 1.)
5504	 */
5505	handlep->blocks_to_filter_in_userland = handle->cc - n;
5506	handlep->filter_in_userland = 1;
5507	return ret;
5508}
5509
5510#endif /* HAVE_PACKET_RING */
5511
5512
5513#ifdef HAVE_PF_PACKET_SOCKETS
5514/*
5515 *  Return the index of the given device name. Fill ebuf and return
5516 *  -1 on failure.
5517 */
5518static int
5519iface_get_id(int fd, const char *device, char *ebuf)
5520{
5521	struct ifreq	ifr;
5522
5523	memset(&ifr, 0, sizeof(ifr));
5524	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5525
5526	if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
5527		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5528		    errno, "SIOCGIFINDEX");
5529		return -1;
5530	}
5531
5532	return ifr.ifr_ifindex;
5533}
5534
5535/*
5536 *  Bind the socket associated with FD to the given device.
5537 *  Return 1 on success, 0 if we should try a SOCK_PACKET socket,
5538 *  or a PCAP_ERROR_ value on a hard error.
5539 */
5540static int
5541iface_bind(int fd, int ifindex, char *ebuf, int protocol)
5542{
5543	struct sockaddr_ll	sll;
5544	int			err;
5545	socklen_t		errlen = sizeof(err);
5546
5547	memset(&sll, 0, sizeof(sll));
5548	sll.sll_family		= AF_PACKET;
5549	sll.sll_ifindex		= ifindex;
5550	sll.sll_protocol	= protocol;
5551
5552	if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
5553		if (errno == ENETDOWN) {
5554			/*
5555			 * Return a "network down" indication, so that
5556			 * the application can report that rather than
5557			 * saying we had a mysterious failure and
5558			 * suggest that they report a problem to the
5559			 * libpcap developers.
5560			 */
5561			return PCAP_ERROR_IFACE_NOT_UP;
5562		} else {
5563			pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5564			    errno, "bind");
5565			return PCAP_ERROR;
5566		}
5567	}
5568
5569	/* Any pending errors, e.g., network is down? */
5570
5571	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
5572		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5573		    errno, "getsockopt");
5574		return 0;
5575	}
5576
5577	if (err == ENETDOWN) {
5578		/*
5579		 * Return a "network down" indication, so that
5580		 * the application can report that rather than
5581		 * saying we had a mysterious failure and
5582		 * suggest that they report a problem to the
5583		 * libpcap developers.
5584		 */
5585		return PCAP_ERROR_IFACE_NOT_UP;
5586	} else if (err > 0) {
5587		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5588		    err, "bind");
5589		return 0;
5590	}
5591
5592	return 1;
5593}
5594
5595#ifdef IW_MODE_MONITOR
5596/*
5597 * Check whether the device supports the Wireless Extensions.
5598 * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE
5599 * if the device doesn't even exist.
5600 */
5601static int
5602has_wext(int sock_fd, const char *device, char *ebuf)
5603{
5604	struct iwreq ireq;
5605
5606	if (is_bonding_device(sock_fd, device))
5607		return 0;	/* bonding device, so don't even try */
5608
5609	strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5610	    sizeof ireq.ifr_ifrn.ifrn_name);
5611	if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
5612		return 1;	/* yes */
5613	pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE, errno,
5614	    "%s: SIOCGIWNAME", device);
5615	if (errno == ENODEV)
5616		return PCAP_ERROR_NO_SUCH_DEVICE;
5617	return 0;
5618}
5619
5620/*
5621 * Per me si va ne la citta dolente,
5622 * Per me si va ne l'etterno dolore,
5623 *	...
5624 * Lasciate ogne speranza, voi ch'intrate.
5625 *
5626 * XXX - airmon-ng does special stuff with the Orinoco driver and the
5627 * wlan-ng driver.
5628 */
5629typedef enum {
5630	MONITOR_WEXT,
5631	MONITOR_HOSTAP,
5632	MONITOR_PRISM,
5633	MONITOR_PRISM54,
5634	MONITOR_ACX100,
5635	MONITOR_RT2500,
5636	MONITOR_RT2570,
5637	MONITOR_RT73,
5638	MONITOR_RTL8XXX
5639} monitor_type;
5640
5641/*
5642 * Use the Wireless Extensions, if we have them, to try to turn monitor mode
5643 * on if it's not already on.
5644 *
5645 * Returns 1 on success, 0 if we don't support the Wireless Extensions
5646 * on this device, or a PCAP_ERROR_ value if we do support them but
5647 * we weren't able to turn monitor mode on.
5648 */
5649static int
5650enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
5651{
5652	/*
5653	 * XXX - at least some adapters require non-Wireless Extensions
5654	 * mechanisms to turn monitor mode on.
5655	 *
5656	 * Atheros cards might require that a separate "monitor virtual access
5657	 * point" be created, with later versions of the madwifi driver.
5658	 * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode
5659	 * monitor -bssid", which apparently spits out a line "athN"
5660	 * where "athN" is the monitor mode device.  To leave monitor
5661	 * mode, it destroys the monitor mode device.
5662	 *
5663	 * Some Intel Centrino adapters might require private ioctls to get
5664	 * radio headers; the ipw2200 and ipw3945 drivers allow you to
5665	 * configure a separate "rtapN" interface to capture in monitor
5666	 * mode without preventing the adapter from operating normally.
5667	 * (airmon-ng doesn't appear to use that, though.)
5668	 *
5669	 * It would be Truly Wonderful if mac80211 and nl80211 cleaned this
5670	 * up, and if all drivers were converted to mac80211 drivers.
5671	 *
5672	 * If interface {if} is a mac80211 driver, the file
5673	 * /sys/class/net/{if}/phy80211 is a symlink to
5674	 * /sys/class/ieee80211/{phydev}, for some {phydev}.
5675	 *
5676	 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
5677	 * least, has a "wmaster0" device and a "wlan0" device; the
5678	 * latter is the one with the IP address.  Both show up in
5679	 * "tcpdump -D" output.  Capturing on the wmaster0 device
5680	 * captures with 802.11 headers.
5681	 *
5682	 * airmon-ng searches through /sys/class/net for devices named
5683	 * monN, starting with mon0; as soon as one *doesn't* exist,
5684	 * it chooses that as the monitor device name.  If the "iw"
5685	 * command exists, it does "iw dev {if} interface add {monif}
5686	 * type monitor", where {monif} is the monitor device.  It
5687	 * then (sigh) sleeps .1 second, and then configures the
5688	 * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
5689	 * is a file, it writes {mondev}, without a newline, to that file,
5690	 * and again (sigh) sleeps .1 second, and then iwconfig's that
5691	 * device into monitor mode and configures it up.  Otherwise,
5692	 * you can't do monitor mode.
5693	 *
5694	 * All these devices are "glued" together by having the
5695	 * /sys/class/net/{device}/phy80211 links pointing to the same
5696	 * place, so, given a wmaster, wlan, or mon device, you can
5697	 * find the other devices by looking for devices with
5698	 * the same phy80211 link.
5699	 *
5700	 * To turn monitor mode off, delete the monitor interface,
5701	 * either with "iw dev {monif} interface del" or by sending
5702	 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
5703	 *
5704	 * Note: if you try to create a monitor device named "monN", and
5705	 * there's already a "monN" device, it fails, as least with
5706	 * the netlink interface (which is what iw uses), with a return
5707	 * value of -ENFILE.  (Return values are negative errnos.)  We
5708	 * could probably use that to find an unused device.
5709	 */
5710	struct pcap_linux *handlep = handle->priv;
5711	int err;
5712	struct iwreq ireq;
5713	struct iw_priv_args *priv;
5714	monitor_type montype;
5715	int i;
5716	__u32 cmd;
5717	struct ifreq ifr;
5718	int oldflags;
5719	int args[2];
5720	int channel;
5721
5722	/*
5723	 * Does this device *support* the Wireless Extensions?
5724	 */
5725	err = has_wext(sock_fd, device, handle->errbuf);
5726	if (err <= 0)
5727		return err;	/* either it doesn't or the device doesn't even exist */
5728	/*
5729	 * Start out assuming we have no private extensions to control
5730	 * radio metadata.
5731	 */
5732	montype = MONITOR_WEXT;
5733	cmd = 0;
5734
5735	/*
5736	 * Try to get all the Wireless Extensions private ioctls
5737	 * supported by this device.
5738	 *
5739	 * First, get the size of the buffer we need, by supplying no
5740	 * buffer and a length of 0.  If the device supports private
5741	 * ioctls, it should return E2BIG, with ireq.u.data.length set
5742	 * to the length we need.  If it doesn't support them, it should
5743	 * return EOPNOTSUPP.
5744	 */
5745	memset(&ireq, 0, sizeof ireq);
5746	strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5747	    sizeof ireq.ifr_ifrn.ifrn_name);
5748	ireq.u.data.pointer = (void *)args;
5749	ireq.u.data.length = 0;
5750	ireq.u.data.flags = 0;
5751	if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
5752		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5753		    "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!",
5754		    device);
5755		return PCAP_ERROR;
5756	}
5757	if (errno != EOPNOTSUPP) {
5758		/*
5759		 * OK, it's not as if there are no private ioctls.
5760		 */
5761		if (errno != E2BIG) {
5762			/*
5763			 * Failed.
5764			 */
5765			pcap_fmt_errmsg_for_errno(handle->errbuf,
5766			    PCAP_ERRBUF_SIZE, errno, "%s: SIOCGIWPRIV", device);
5767			return PCAP_ERROR;
5768		}
5769
5770		/*
5771		 * OK, try to get the list of private ioctls.
5772		 */
5773		priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args));
5774		if (priv == NULL) {
5775			pcap_fmt_errmsg_for_errno(handle->errbuf,
5776			    PCAP_ERRBUF_SIZE, errno, "malloc");
5777			return PCAP_ERROR;
5778		}
5779		ireq.u.data.pointer = (void *)priv;
5780		if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) {
5781			pcap_fmt_errmsg_for_errno(handle->errbuf,
5782			    PCAP_ERRBUF_SIZE, errno, "%s: SIOCGIWPRIV", device);
5783			free(priv);
5784			return PCAP_ERROR;
5785		}
5786
5787		/*
5788		 * Look for private ioctls to turn monitor mode on or, if
5789		 * monitor mode is on, to set the header type.
5790		 */
5791		for (i = 0; i < ireq.u.data.length; i++) {
5792			if (strcmp(priv[i].name, "monitor_type") == 0) {
5793				/*
5794				 * Hostap driver, use this one.
5795				 * Set monitor mode first.
5796				 * You can set it to 0 to get DLT_IEEE80211,
5797				 * 1 to get DLT_PRISM, 2 to get
5798				 * DLT_IEEE80211_RADIO_AVS, and, with more
5799				 * recent versions of the driver, 3 to get
5800				 * DLT_IEEE80211_RADIO.
5801				 */
5802				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5803					break;
5804				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5805					break;
5806				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5807					break;
5808				montype = MONITOR_HOSTAP;
5809				cmd = priv[i].cmd;
5810				break;
5811			}
5812			if (strcmp(priv[i].name, "set_prismhdr") == 0) {
5813				/*
5814				 * Prism54 driver, use this one.
5815				 * Set monitor mode first.
5816				 * You can set it to 2 to get DLT_IEEE80211
5817				 * or 3 or get DLT_PRISM.
5818				 */
5819				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5820					break;
5821				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5822					break;
5823				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5824					break;
5825				montype = MONITOR_PRISM54;
5826				cmd = priv[i].cmd;
5827				break;
5828			}
5829			if (strcmp(priv[i].name, "forceprismheader") == 0) {
5830				/*
5831				 * RT2570 driver, use this one.
5832				 * Do this after turning monitor mode on.
5833				 * You can set it to 1 to get DLT_PRISM or 2
5834				 * to get DLT_IEEE80211.
5835				 */
5836				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5837					break;
5838				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5839					break;
5840				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5841					break;
5842				montype = MONITOR_RT2570;
5843				cmd = priv[i].cmd;
5844				break;
5845			}
5846			if (strcmp(priv[i].name, "forceprism") == 0) {
5847				/*
5848				 * RT73 driver, use this one.
5849				 * Do this after turning monitor mode on.
5850				 * Its argument is a *string*; you can
5851				 * set it to "1" to get DLT_PRISM or "2"
5852				 * to get DLT_IEEE80211.
5853				 */
5854				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR)
5855					break;
5856				if (priv[i].set_args & IW_PRIV_SIZE_FIXED)
5857					break;
5858				montype = MONITOR_RT73;
5859				cmd = priv[i].cmd;
5860				break;
5861			}
5862			if (strcmp(priv[i].name, "prismhdr") == 0) {
5863				/*
5864				 * One of the RTL8xxx drivers, use this one.
5865				 * It can only be done after monitor mode
5866				 * has been turned on.  You can set it to 1
5867				 * to get DLT_PRISM or 0 to get DLT_IEEE80211.
5868				 */
5869				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5870					break;
5871				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5872					break;
5873				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5874					break;
5875				montype = MONITOR_RTL8XXX;
5876				cmd = priv[i].cmd;
5877				break;
5878			}
5879			if (strcmp(priv[i].name, "rfmontx") == 0) {
5880				/*
5881				 * RT2500 or RT61 driver, use this one.
5882				 * It has one one-byte parameter; set
5883				 * u.data.length to 1 and u.data.pointer to
5884				 * point to the parameter.
5885				 * It doesn't itself turn monitor mode on.
5886				 * You can set it to 1 to allow transmitting
5887				 * in monitor mode(?) and get DLT_IEEE80211,
5888				 * or set it to 0 to disallow transmitting in
5889				 * monitor mode(?) and get DLT_PRISM.
5890				 */
5891				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5892					break;
5893				if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2)
5894					break;
5895				montype = MONITOR_RT2500;
5896				cmd = priv[i].cmd;
5897				break;
5898			}
5899			if (strcmp(priv[i].name, "monitor") == 0) {
5900				/*
5901				 * Either ACX100 or hostap, use this one.
5902				 * It turns monitor mode on.
5903				 * If it takes two arguments, it's ACX100;
5904				 * the first argument is 1 for DLT_PRISM
5905				 * or 2 for DLT_IEEE80211, and the second
5906				 * argument is the channel on which to
5907				 * run.  If it takes one argument, it's
5908				 * HostAP, and the argument is 2 for
5909				 * DLT_IEEE80211 and 3 for DLT_PRISM.
5910				 *
5911				 * If we see this, we don't quit, as this
5912				 * might be a version of the hostap driver
5913				 * that also supports "monitor_type".
5914				 */
5915				if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5916					break;
5917				if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5918					break;
5919				switch (priv[i].set_args & IW_PRIV_SIZE_MASK) {
5920
5921				case 1:
5922					montype = MONITOR_PRISM;
5923					cmd = priv[i].cmd;
5924					break;
5925
5926				case 2:
5927					montype = MONITOR_ACX100;
5928					cmd = priv[i].cmd;
5929					break;
5930
5931				default:
5932					break;
5933				}
5934			}
5935		}
5936		free(priv);
5937	}
5938
5939	/*
5940	 * XXX - ipw3945?  islism?
5941	 */
5942
5943	/*
5944	 * Get the old mode.
5945	 */
5946	strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5947	    sizeof ireq.ifr_ifrn.ifrn_name);
5948	if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) {
5949		/*
5950		 * We probably won't be able to set the mode, either.
5951		 */
5952		return PCAP_ERROR_RFMON_NOTSUP;
5953	}
5954
5955	/*
5956	 * Is it currently in monitor mode?
5957	 */
5958	if (ireq.u.mode == IW_MODE_MONITOR) {
5959		/*
5960		 * Yes.  Just leave things as they are.
5961		 * We don't offer multiple link-layer types, as
5962		 * changing the link-layer type out from under
5963		 * somebody else capturing in monitor mode would
5964		 * be considered rude.
5965		 */
5966		return 1;
5967	}
5968	/*
5969	 * No.  We have to put the adapter into rfmon mode.
5970	 */
5971
5972	/*
5973	 * If we haven't already done so, arrange to have
5974	 * "pcap_close_all()" called when we exit.
5975	 */
5976	if (!pcap_do_addexit(handle)) {
5977		/*
5978		 * "atexit()" failed; don't put the interface
5979		 * in rfmon mode, just give up.
5980		 */
5981		return PCAP_ERROR_RFMON_NOTSUP;
5982	}
5983
5984	/*
5985	 * Save the old mode.
5986	 */
5987	handlep->oldmode = ireq.u.mode;
5988
5989	/*
5990	 * Put the adapter in rfmon mode.  How we do this depends
5991	 * on whether we have a special private ioctl or not.
5992	 */
5993	if (montype == MONITOR_PRISM) {
5994		/*
5995		 * We have the "monitor" private ioctl, but none of
5996		 * the other private ioctls.  Use this, and select
5997		 * the Prism header.
5998		 *
5999		 * If it fails, just fall back on SIOCSIWMODE.
6000		 */
6001		memset(&ireq, 0, sizeof ireq);
6002		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6003		    sizeof ireq.ifr_ifrn.ifrn_name);
6004		ireq.u.data.length = 1;	/* 1 argument */
6005		args[0] = 3;	/* request Prism header */
6006		memcpy(ireq.u.name, args, sizeof (int));
6007		if (ioctl(sock_fd, cmd, &ireq) != -1) {
6008			/*
6009			 * Success.
6010			 * Note that we have to put the old mode back
6011			 * when we close the device.
6012			 */
6013			handlep->must_do_on_close |= MUST_CLEAR_RFMON;
6014
6015			/*
6016			 * Add this to the list of pcaps to close
6017			 * when we exit.
6018			 */
6019			pcap_add_to_pcaps_to_close(handle);
6020
6021			return 1;
6022		}
6023
6024		/*
6025		 * Failure.  Fall back on SIOCSIWMODE.
6026		 */
6027	}
6028
6029	/*
6030	 * First, take the interface down if it's up; otherwise, we
6031	 * might get EBUSY.
6032	 */
6033	memset(&ifr, 0, sizeof(ifr));
6034	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6035	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
6036		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
6037		    errno, "%s: Can't get flags", device);
6038		return PCAP_ERROR;
6039	}
6040	oldflags = 0;
6041	if (ifr.ifr_flags & IFF_UP) {
6042		oldflags = ifr.ifr_flags;
6043		ifr.ifr_flags &= ~IFF_UP;
6044		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
6045			pcap_fmt_errmsg_for_errno(handle->errbuf,
6046			    PCAP_ERRBUF_SIZE, errno, "%s: Can't set flags",
6047			    device);
6048			return PCAP_ERROR;
6049		}
6050	}
6051
6052	/*
6053	 * Then turn monitor mode on.
6054	 */
6055	strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6056	    sizeof ireq.ifr_ifrn.ifrn_name);
6057	ireq.u.mode = IW_MODE_MONITOR;
6058	if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) {
6059		/*
6060		 * Scientist, you've failed.
6061		 * Bring the interface back up if we shut it down.
6062		 */
6063		ifr.ifr_flags = oldflags;
6064		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
6065			pcap_fmt_errmsg_for_errno(handle->errbuf,
6066			    PCAP_ERRBUF_SIZE, errno, "%s: Can't set flags",
6067			    device);
6068			return PCAP_ERROR;
6069		}
6070		return PCAP_ERROR_RFMON_NOTSUP;
6071	}
6072
6073	/*
6074	 * XXX - airmon-ng does "iwconfig {if} key off" after setting
6075	 * monitor mode and setting the channel, and then does
6076	 * "iwconfig up".
6077	 */
6078
6079	/*
6080	 * Now select the appropriate radio header.
6081	 */
6082	switch (montype) {
6083
6084	case MONITOR_WEXT:
6085		/*
6086		 * We don't have any private ioctl to set the header.
6087		 */
6088		break;
6089
6090	case MONITOR_HOSTAP:
6091		/*
6092		 * Try to select the radiotap header.
6093		 */
6094		memset(&ireq, 0, sizeof ireq);
6095		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6096		    sizeof ireq.ifr_ifrn.ifrn_name);
6097		args[0] = 3;	/* request radiotap header */
6098		memcpy(ireq.u.name, args, sizeof (int));
6099		if (ioctl(sock_fd, cmd, &ireq) != -1)
6100			break;	/* success */
6101
6102		/*
6103		 * That failed.  Try to select the AVS header.
6104		 */
6105		memset(&ireq, 0, sizeof ireq);
6106		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6107		    sizeof ireq.ifr_ifrn.ifrn_name);
6108		args[0] = 2;	/* request AVS header */
6109		memcpy(ireq.u.name, args, sizeof (int));
6110		if (ioctl(sock_fd, cmd, &ireq) != -1)
6111			break;	/* success */
6112
6113		/*
6114		 * That failed.  Try to select the Prism header.
6115		 */
6116		memset(&ireq, 0, sizeof ireq);
6117		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6118		    sizeof ireq.ifr_ifrn.ifrn_name);
6119		args[0] = 1;	/* request Prism header */
6120		memcpy(ireq.u.name, args, sizeof (int));
6121		ioctl(sock_fd, cmd, &ireq);
6122		break;
6123
6124	case MONITOR_PRISM:
6125		/*
6126		 * The private ioctl failed.
6127		 */
6128		break;
6129
6130	case MONITOR_PRISM54:
6131		/*
6132		 * Select the Prism header.
6133		 */
6134		memset(&ireq, 0, sizeof ireq);
6135		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6136		    sizeof ireq.ifr_ifrn.ifrn_name);
6137		args[0] = 3;	/* request Prism header */
6138		memcpy(ireq.u.name, args, sizeof (int));
6139		ioctl(sock_fd, cmd, &ireq);
6140		break;
6141
6142	case MONITOR_ACX100:
6143		/*
6144		 * Get the current channel.
6145		 */
6146		memset(&ireq, 0, sizeof ireq);
6147		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6148		    sizeof ireq.ifr_ifrn.ifrn_name);
6149		if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) {
6150			pcap_fmt_errmsg_for_errno(handle->errbuf,
6151			    PCAP_ERRBUF_SIZE, errno, "%s: SIOCGIWFREQ", device);
6152			return PCAP_ERROR;
6153		}
6154		channel = ireq.u.freq.m;
6155
6156		/*
6157		 * Select the Prism header, and set the channel to the
6158		 * current value.
6159		 */
6160		memset(&ireq, 0, sizeof ireq);
6161		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6162		    sizeof ireq.ifr_ifrn.ifrn_name);
6163		args[0] = 1;		/* request Prism header */
6164		args[1] = channel;	/* set channel */
6165		memcpy(ireq.u.name, args, 2*sizeof (int));
6166		ioctl(sock_fd, cmd, &ireq);
6167		break;
6168
6169	case MONITOR_RT2500:
6170		/*
6171		 * Disallow transmission - that turns on the
6172		 * Prism header.
6173		 */
6174		memset(&ireq, 0, sizeof ireq);
6175		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6176		    sizeof ireq.ifr_ifrn.ifrn_name);
6177		args[0] = 0;	/* disallow transmitting */
6178		memcpy(ireq.u.name, args, sizeof (int));
6179		ioctl(sock_fd, cmd, &ireq);
6180		break;
6181
6182	case MONITOR_RT2570:
6183		/*
6184		 * Force the Prism header.
6185		 */
6186		memset(&ireq, 0, sizeof ireq);
6187		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6188		    sizeof ireq.ifr_ifrn.ifrn_name);
6189		args[0] = 1;	/* request Prism header */
6190		memcpy(ireq.u.name, args, sizeof (int));
6191		ioctl(sock_fd, cmd, &ireq);
6192		break;
6193
6194	case MONITOR_RT73:
6195		/*
6196		 * Force the Prism header.
6197		 */
6198		memset(&ireq, 0, sizeof ireq);
6199		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6200		    sizeof ireq.ifr_ifrn.ifrn_name);
6201		ireq.u.data.length = 1;	/* 1 argument */
6202		ireq.u.data.pointer = "1";
6203		ireq.u.data.flags = 0;
6204		ioctl(sock_fd, cmd, &ireq);
6205		break;
6206
6207	case MONITOR_RTL8XXX:
6208		/*
6209		 * Force the Prism header.
6210		 */
6211		memset(&ireq, 0, sizeof ireq);
6212		strlcpy(ireq.ifr_ifrn.ifrn_name, device,
6213		    sizeof ireq.ifr_ifrn.ifrn_name);
6214		args[0] = 1;	/* request Prism header */
6215		memcpy(ireq.u.name, args, sizeof (int));
6216		ioctl(sock_fd, cmd, &ireq);
6217		break;
6218	}
6219
6220	/*
6221	 * Now bring the interface back up if we brought it down.
6222	 */
6223	if (oldflags != 0) {
6224		ifr.ifr_flags = oldflags;
6225		if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
6226			pcap_fmt_errmsg_for_errno(handle->errbuf,
6227			    PCAP_ERRBUF_SIZE, errno, "%s: Can't set flags",
6228			    device);
6229
6230			/*
6231			 * At least try to restore the old mode on the
6232			 * interface.
6233			 */
6234			if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
6235				/*
6236				 * Scientist, you've failed.
6237				 */
6238				fprintf(stderr,
6239				    "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n"
6240				    "Please adjust manually.\n",
6241				    strerror(errno));
6242			}
6243			return PCAP_ERROR;
6244		}
6245	}
6246
6247	/*
6248	 * Note that we have to put the old mode back when we
6249	 * close the device.
6250	 */
6251	handlep->must_do_on_close |= MUST_CLEAR_RFMON;
6252
6253	/*
6254	 * Add this to the list of pcaps to close when we exit.
6255	 */
6256	pcap_add_to_pcaps_to_close(handle);
6257
6258	return 1;
6259}
6260#endif /* IW_MODE_MONITOR */
6261
6262/*
6263 * Try various mechanisms to enter monitor mode.
6264 */
6265static int
6266enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
6267{
6268#if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
6269	int ret;
6270#endif
6271
6272#ifdef HAVE_LIBNL
6273	ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
6274	if (ret < 0)
6275		return ret;	/* error attempting to do so */
6276	if (ret == 1)
6277		return 1;	/* success */
6278#endif /* HAVE_LIBNL */
6279
6280#ifdef IW_MODE_MONITOR
6281	ret = enter_rfmon_mode_wext(handle, sock_fd, device);
6282	if (ret < 0)
6283		return ret;	/* error attempting to do so */
6284	if (ret == 1)
6285		return 1;	/* success */
6286#endif /* IW_MODE_MONITOR */
6287
6288	/*
6289	 * Either none of the mechanisms we know about work or none
6290	 * of those mechanisms are available, so we can't do monitor
6291	 * mode.
6292	 */
6293	return 0;
6294}
6295
6296#if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
6297/*
6298 * Map SOF_TIMESTAMPING_ values to PCAP_TSTAMP_ values.
6299 */
6300static const struct {
6301	int soft_timestamping_val;
6302	int pcap_tstamp_val;
6303} sof_ts_type_map[3] = {
6304	{ SOF_TIMESTAMPING_SOFTWARE, PCAP_TSTAMP_HOST },
6305	{ SOF_TIMESTAMPING_SYS_HARDWARE, PCAP_TSTAMP_ADAPTER },
6306	{ SOF_TIMESTAMPING_RAW_HARDWARE, PCAP_TSTAMP_ADAPTER_UNSYNCED }
6307};
6308#define NUM_SOF_TIMESTAMPING_TYPES	(sizeof sof_ts_type_map / sizeof sof_ts_type_map[0])
6309
6310/*
6311 * Set the list of time stamping types to include all types.
6312 */
6313static void
6314iface_set_all_ts_types(pcap_t *handle)
6315{
6316	u_int i;
6317
6318	handle->tstamp_type_count = NUM_SOF_TIMESTAMPING_TYPES;
6319	handle->tstamp_type_list = malloc(NUM_SOF_TIMESTAMPING_TYPES * sizeof(u_int));
6320	for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++)
6321		handle->tstamp_type_list[i] = sof_ts_type_map[i].pcap_tstamp_val;
6322}
6323
6324#ifdef ETHTOOL_GET_TS_INFO
6325/*
6326 * Get a list of time stamping capabilities.
6327 */
6328static int
6329iface_ethtool_get_ts_info(const char *device, pcap_t *handle, char *ebuf)
6330{
6331	int fd;
6332	struct ifreq ifr;
6333	struct ethtool_ts_info info;
6334	int num_ts_types;
6335	u_int i, j;
6336
6337	/*
6338	 * This doesn't apply to the "any" device; you can't say "turn on
6339	 * hardware time stamping for all devices that exist now and arrange
6340	 * that it be turned on for any device that appears in the future",
6341	 * and not all devices even necessarily *support* hardware time
6342	 * stamping, so don't report any time stamp types.
6343	 */
6344	if (strcmp(device, "any") == 0) {
6345		handle->tstamp_type_list = NULL;
6346		return 0;
6347	}
6348
6349	/*
6350	 * Create a socket from which to fetch time stamping capabilities.
6351	 */
6352	fd = socket(PF_UNIX, SOCK_RAW, 0);
6353	if (fd < 0) {
6354		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
6355		    errno, "socket for SIOCETHTOOL(ETHTOOL_GET_TS_INFO)");
6356		return -1;
6357	}
6358
6359	memset(&ifr, 0, sizeof(ifr));
6360	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6361	memset(&info, 0, sizeof(info));
6362	info.cmd = ETHTOOL_GET_TS_INFO;
6363	ifr.ifr_data = (caddr_t)&info;
6364	if (ioctl(fd, SIOCETHTOOL, &ifr) == -1) {
6365		int save_errno = errno;
6366
6367		close(fd);
6368		switch (save_errno) {
6369
6370		case EOPNOTSUPP:
6371		case EINVAL:
6372			/*
6373			 * OK, this OS version or driver doesn't support
6374			 * asking for the time stamping types, so let's
6375			 * just return all the possible types.
6376			 */
6377			iface_set_all_ts_types(handle);
6378			return 0;
6379
6380		case ENODEV:
6381			/*
6382			 * OK, no such device.
6383			 * The user will find that out when they try to
6384			 * activate the device; just return an empty
6385			 * list of time stamp types.
6386			 */
6387			handle->tstamp_type_list = NULL;
6388			return 0;
6389
6390		default:
6391			/*
6392			 * Other error.
6393			 */
6394			pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
6395			    save_errno,
6396			    "%s: SIOCETHTOOL(ETHTOOL_GET_TS_INFO) ioctl failed",
6397			    device);
6398			return -1;
6399		}
6400	}
6401	close(fd);
6402
6403	/*
6404	 * Do we support hardware time stamping of *all* packets?
6405	 */
6406	if (!(info.rx_filters & (1 << HWTSTAMP_FILTER_ALL))) {
6407		/*
6408		 * No, so don't report any time stamp types.
6409		 *
6410		 * XXX - some devices either don't report
6411		 * HWTSTAMP_FILTER_ALL when they do support it, or
6412		 * report HWTSTAMP_FILTER_ALL but map it to only
6413		 * time stamping a few PTP packets.  See
6414		 * http://marc.info/?l=linux-netdev&m=146318183529571&w=2
6415		 */
6416		handle->tstamp_type_list = NULL;
6417		return 0;
6418	}
6419
6420	num_ts_types = 0;
6421	for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) {
6422		if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val)
6423			num_ts_types++;
6424	}
6425	handle->tstamp_type_count = num_ts_types;
6426	if (num_ts_types != 0) {
6427		handle->tstamp_type_list = malloc(num_ts_types * sizeof(u_int));
6428		for (i = 0, j = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) {
6429			if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val) {
6430				handle->tstamp_type_list[j] = sof_ts_type_map[i].pcap_tstamp_val;
6431				j++;
6432			}
6433		}
6434	} else
6435		handle->tstamp_type_list = NULL;
6436
6437	return 0;
6438}
6439#else /* ETHTOOL_GET_TS_INFO */
6440static int
6441iface_ethtool_get_ts_info(const char *device, pcap_t *handle, char *ebuf _U_)
6442{
6443	/*
6444	 * This doesn't apply to the "any" device; you can't say "turn on
6445	 * hardware time stamping for all devices that exist now and arrange
6446	 * that it be turned on for any device that appears in the future",
6447	 * and not all devices even necessarily *support* hardware time
6448	 * stamping, so don't report any time stamp types.
6449	 */
6450	if (strcmp(device, "any") == 0) {
6451		handle->tstamp_type_list = NULL;
6452		return 0;
6453	}
6454
6455	/*
6456	 * We don't have an ioctl to use to ask what's supported,
6457	 * so say we support everything.
6458	 */
6459	iface_set_all_ts_types(handle);
6460	return 0;
6461}
6462#endif /* ETHTOOL_GET_TS_INFO */
6463
6464#endif /* defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) */
6465
6466#ifdef HAVE_PACKET_RING
6467/*
6468 * Find out if we have any form of fragmentation/reassembly offloading.
6469 *
6470 * We do so using SIOCETHTOOL checking for various types of offloading;
6471 * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
6472 * of the types of offloading, there's nothing we can do to check, so
6473 * we just say "no, we don't".
6474 */
6475#if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
6476static int
6477iface_ethtool_flag_ioctl(pcap_t *handle, int cmd, const char *cmdname)
6478{
6479	struct ifreq	ifr;
6480	struct ethtool_value eval;
6481
6482	memset(&ifr, 0, sizeof(ifr));
6483	strlcpy(ifr.ifr_name, handle->opt.device, sizeof(ifr.ifr_name));
6484	eval.cmd = cmd;
6485	eval.data = 0;
6486	ifr.ifr_data = (caddr_t)&eval;
6487	if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
6488		if (errno == EOPNOTSUPP || errno == EINVAL) {
6489			/*
6490			 * OK, let's just return 0, which, in our
6491			 * case, either means "no, what we're asking
6492			 * about is not enabled" or "all the flags
6493			 * are clear (i.e., nothing is enabled)".
6494			 */
6495			return 0;
6496		}
6497		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
6498		    errno, "%s: SIOCETHTOOL(%s) ioctl failed",
6499		    handle->opt.device, cmdname);
6500		return -1;
6501	}
6502	return eval.data;
6503}
6504
6505static int
6506iface_get_offload(pcap_t *handle)
6507{
6508	int ret;
6509
6510#ifdef ETHTOOL_GTSO
6511	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO");
6512	if (ret == -1)
6513		return -1;
6514	if (ret)
6515		return 1;	/* TCP segmentation offloading on */
6516#endif
6517
6518#ifdef ETHTOOL_GUFO
6519	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO");
6520	if (ret == -1)
6521		return -1;
6522	if (ret)
6523		return 1;	/* UDP fragmentation offloading on */
6524#endif
6525
6526#ifdef ETHTOOL_GGSO
6527	/*
6528	 * XXX - will this cause large unsegmented packets to be
6529	 * handed to PF_PACKET sockets on transmission?  If not,
6530	 * this need not be checked.
6531	 */
6532	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO");
6533	if (ret == -1)
6534		return -1;
6535	if (ret)
6536		return 1;	/* generic segmentation offloading on */
6537#endif
6538
6539#ifdef ETHTOOL_GFLAGS
6540	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
6541	if (ret == -1)
6542		return -1;
6543	if (ret & ETH_FLAG_LRO)
6544		return 1;	/* large receive offloading on */
6545#endif
6546
6547#ifdef ETHTOOL_GGRO
6548	/*
6549	 * XXX - will this cause large reassembled packets to be
6550	 * handed to PF_PACKET sockets on receipt?  If not,
6551	 * this need not be checked.
6552	 */
6553	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO");
6554	if (ret == -1)
6555		return -1;
6556	if (ret)
6557		return 1;	/* generic (large) receive offloading on */
6558#endif
6559
6560	return 0;
6561}
6562#else /* SIOCETHTOOL */
6563static int
6564iface_get_offload(pcap_t *handle _U_)
6565{
6566	/*
6567	 * XXX - do we need to get this information if we don't
6568	 * have the ethtool ioctls?  If so, how do we do that?
6569	 */
6570	return 0;
6571}
6572#endif /* SIOCETHTOOL */
6573
6574#endif /* HAVE_PACKET_RING */
6575
6576#endif /* HAVE_PF_PACKET_SOCKETS */
6577
6578/* ===== Functions to interface to the older kernels ================== */
6579
6580/*
6581 * Try to open a packet socket using the old kernel interface.
6582 * Returns 1 on success and a PCAP_ERROR_ value on an error.
6583 */
6584static int
6585activate_old(pcap_t *handle)
6586{
6587	struct pcap_linux *handlep = handle->priv;
6588	int		err;
6589	int		arptype;
6590	struct ifreq	ifr;
6591	const char	*device = handle->opt.device;
6592	struct utsname	utsname;
6593	int		mtu;
6594
6595	/* Open the socket */
6596
6597	handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
6598	if (handle->fd == -1) {
6599		err = errno;
6600		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
6601		    err, "socket");
6602		if (err == EPERM || err == EACCES) {
6603			/*
6604			 * You don't have permission to open the
6605			 * socket.
6606			 */
6607			return PCAP_ERROR_PERM_DENIED;
6608		} else {
6609			/*
6610			 * Other error.
6611			 */
6612			return PCAP_ERROR;
6613		}
6614	}
6615
6616	/* It worked - we are using the old interface */
6617	handlep->sock_packet = 1;
6618
6619	/* ...which means we get the link-layer header. */
6620	handlep->cooked = 0;
6621
6622	/* Bind to the given device */
6623
6624	if (strcmp(device, "any") == 0) {
6625		strlcpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
6626			PCAP_ERRBUF_SIZE);
6627		return PCAP_ERROR;
6628	}
6629	if (iface_bind_old(handle->fd, device, handle->errbuf) == -1)
6630		return PCAP_ERROR;
6631
6632	/*
6633	 * Try to get the link-layer type.
6634	 */
6635	arptype = iface_get_arptype(handle->fd, device, handle->errbuf);
6636	if (arptype < 0)
6637		return PCAP_ERROR;
6638
6639	/*
6640	 * Try to find the DLT_ type corresponding to that
6641	 * link-layer type.
6642	 */
6643	map_arphrd_to_dlt(handle, handle->fd, arptype, device, 0);
6644	if (handle->linktype == -1) {
6645		pcap_snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6646			 "unknown arptype %d", arptype);
6647		return PCAP_ERROR;
6648	}
6649
6650	/* Go to promisc mode if requested */
6651
6652	if (handle->opt.promisc) {
6653		memset(&ifr, 0, sizeof(ifr));
6654		strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6655		if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
6656			pcap_fmt_errmsg_for_errno(handle->errbuf,
6657			    PCAP_ERRBUF_SIZE, errno, "SIOCGIFFLAGS");
6658			return PCAP_ERROR;
6659		}
6660		if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
6661			/*
6662			 * Promiscuous mode isn't currently on,
6663			 * so turn it on, and remember that
6664			 * we should turn it off when the
6665			 * pcap_t is closed.
6666			 */
6667
6668			/*
6669			 * If we haven't already done so, arrange
6670			 * to have "pcap_close_all()" called when
6671			 * we exit.
6672			 */
6673			if (!pcap_do_addexit(handle)) {
6674				/*
6675				 * "atexit()" failed; don't put
6676				 * the interface in promiscuous
6677				 * mode, just give up.
6678				 */
6679				return PCAP_ERROR;
6680			}
6681
6682			ifr.ifr_flags |= IFF_PROMISC;
6683			if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
6684				pcap_fmt_errmsg_for_errno(handle->errbuf,
6685				    PCAP_ERRBUF_SIZE, errno, "SIOCSIFFLAGS");
6686				return PCAP_ERROR;
6687			}
6688			handlep->must_do_on_close |= MUST_CLEAR_PROMISC;
6689
6690			/*
6691			 * Add this to the list of pcaps
6692			 * to close when we exit.
6693			 */
6694			pcap_add_to_pcaps_to_close(handle);
6695		}
6696	}
6697
6698	/*
6699	 * Compute the buffer size.
6700	 *
6701	 * We're using SOCK_PACKET, so this might be a 2.0[.x]
6702	 * kernel, and might require special handling - check.
6703	 */
6704	if (uname(&utsname) < 0 ||
6705	    strncmp(utsname.release, "2.0", 3) == 0) {
6706		/*
6707		 * Either we couldn't find out what kernel release
6708		 * this is, or it's a 2.0[.x] kernel.
6709		 *
6710		 * In the 2.0[.x] kernel, a "recvfrom()" on
6711		 * a SOCK_PACKET socket, with MSG_TRUNC set, will
6712		 * return the number of bytes read, so if we pass
6713		 * a length based on the snapshot length, it'll
6714		 * return the number of bytes from the packet
6715		 * copied to userland, not the actual length
6716		 * of the packet.
6717		 *
6718		 * This means that, for example, the IP dissector
6719		 * in tcpdump will get handed a packet length less
6720		 * than the length in the IP header, and will
6721		 * complain about "truncated-ip".
6722		 *
6723		 * So we don't bother trying to copy from the
6724		 * kernel only the bytes in which we're interested,
6725		 * but instead copy them all, just as the older
6726		 * versions of libpcap for Linux did.
6727		 *
6728		 * The buffer therefore needs to be big enough to
6729		 * hold the largest packet we can get from this
6730		 * device.  Unfortunately, we can't get the MRU
6731		 * of the network; we can only get the MTU.  The
6732		 * MTU may be too small, in which case a packet larger
6733		 * than the buffer size will be truncated *and* we
6734		 * won't get the actual packet size.
6735		 *
6736		 * However, if the snapshot length is larger than
6737		 * the buffer size based on the MTU, we use the
6738		 * snapshot length as the buffer size, instead;
6739		 * this means that with a sufficiently large snapshot
6740		 * length we won't artificially truncate packets
6741		 * to the MTU-based size.
6742		 *
6743		 * This mess just one of many problems with packet
6744		 * capture on 2.0[.x] kernels; you really want a
6745		 * 2.2[.x] or later kernel if you want packet capture
6746		 * to work well.
6747		 */
6748		mtu = iface_get_mtu(handle->fd, device, handle->errbuf);
6749		if (mtu == -1)
6750			return PCAP_ERROR;
6751		handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
6752		if (handle->bufsize < (u_int)handle->snapshot)
6753			handle->bufsize = (u_int)handle->snapshot;
6754	} else {
6755		/*
6756		 * This is a 2.2[.x] or later kernel.
6757		 *
6758		 * We can safely pass "recvfrom()" a byte count
6759		 * based on the snapshot length.
6760		 */
6761		handle->bufsize = (u_int)handle->snapshot;
6762	}
6763
6764	/*
6765	 * Default value for offset to align link-layer payload
6766	 * on a 4-byte boundary.
6767	 */
6768	handle->offset	 = 0;
6769
6770	/*
6771	 * SOCK_PACKET sockets don't supply information from
6772	 * stripped VLAN tags.
6773	 */
6774	handlep->vlan_offset = -1; /* unknown */
6775
6776	return 1;
6777}
6778
6779/*
6780 *  Bind the socket associated with FD to the given device using the
6781 *  interface of the old kernels.
6782 */
6783static int
6784iface_bind_old(int fd, const char *device, char *ebuf)
6785{
6786	struct sockaddr	saddr;
6787	int		err;
6788	socklen_t	errlen = sizeof(err);
6789
6790	memset(&saddr, 0, sizeof(saddr));
6791	strlcpy(saddr.sa_data, device, sizeof(saddr.sa_data));
6792	if (bind(fd, &saddr, sizeof(saddr)) == -1) {
6793		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
6794		    errno, "bind");
6795		return -1;
6796	}
6797
6798	/* Any pending errors, e.g., network is down? */
6799
6800	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
6801		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
6802		    errno, "getsockopt");
6803		return -1;
6804	}
6805
6806	if (err > 0) {
6807		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
6808		    err, "bind");
6809		return -1;
6810	}
6811
6812	return 0;
6813}
6814
6815
6816/* ===== System calls available on all supported kernels ============== */
6817
6818/*
6819 *  Query the kernel for the MTU of the given interface.
6820 */
6821static int
6822iface_get_mtu(int fd, const char *device, char *ebuf)
6823{
6824	struct ifreq	ifr;
6825
6826	if (!device)
6827		return BIGGER_THAN_ALL_MTUS;
6828
6829	memset(&ifr, 0, sizeof(ifr));
6830	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6831
6832	if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
6833		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
6834		    errno, "SIOCGIFMTU");
6835		return -1;
6836	}
6837
6838	return ifr.ifr_mtu;
6839}
6840
6841/*
6842 *  Get the hardware type of the given interface as ARPHRD_xxx constant.
6843 */
6844static int
6845iface_get_arptype(int fd, const char *device, char *ebuf)
6846{
6847	struct ifreq	ifr;
6848
6849	memset(&ifr, 0, sizeof(ifr));
6850	strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
6851
6852	if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
6853		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
6854		    errno, "SIOCGIFHWADDR");
6855		if (errno == ENODEV) {
6856			/*
6857			 * No such device.
6858			 */
6859			return PCAP_ERROR_NO_SUCH_DEVICE;
6860		}
6861		return PCAP_ERROR;
6862	}
6863
6864	return ifr.ifr_hwaddr.sa_family;
6865}
6866
6867#ifdef SO_ATTACH_FILTER
6868static int
6869fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
6870{
6871	struct pcap_linux *handlep = handle->priv;
6872	size_t prog_size;
6873	register int i;
6874	register struct bpf_insn *p;
6875	struct bpf_insn *f;
6876	int len;
6877
6878	/*
6879	 * Make a copy of the filter, and modify that copy if
6880	 * necessary.
6881	 */
6882	prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
6883	len = handle->fcode.bf_len;
6884	f = (struct bpf_insn *)malloc(prog_size);
6885	if (f == NULL) {
6886		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
6887		    errno, "malloc");
6888		return -1;
6889	}
6890	memcpy(f, handle->fcode.bf_insns, prog_size);
6891	fcode->len = len;
6892	fcode->filter = (struct sock_filter *) f;
6893
6894	for (i = 0; i < len; ++i) {
6895		p = &f[i];
6896		/*
6897		 * What type of instruction is this?
6898		 */
6899		switch (BPF_CLASS(p->code)) {
6900
6901		case BPF_RET:
6902			/*
6903			 * It's a return instruction; are we capturing
6904			 * in memory-mapped mode?
6905			 */
6906			if (!is_mmapped) {
6907				/*
6908				 * No; is the snapshot length a constant,
6909				 * rather than the contents of the
6910				 * accumulator?
6911				 */
6912				if (BPF_MODE(p->code) == BPF_K) {
6913					/*
6914					 * Yes - if the value to be returned,
6915					 * i.e. the snapshot length, is
6916					 * anything other than 0, make it
6917					 * MAXIMUM_SNAPLEN, so that the packet
6918					 * is truncated by "recvfrom()",
6919					 * not by the filter.
6920					 *
6921					 * XXX - there's nothing we can
6922					 * easily do if it's getting the
6923					 * value from the accumulator; we'd
6924					 * have to insert code to force
6925					 * non-zero values to be
6926					 * MAXIMUM_SNAPLEN.
6927					 */
6928					if (p->k != 0)
6929						p->k = MAXIMUM_SNAPLEN;
6930				}
6931			}
6932			break;
6933
6934		case BPF_LD:
6935		case BPF_LDX:
6936			/*
6937			 * It's a load instruction; is it loading
6938			 * from the packet?
6939			 */
6940			switch (BPF_MODE(p->code)) {
6941
6942			case BPF_ABS:
6943			case BPF_IND:
6944			case BPF_MSH:
6945				/*
6946				 * Yes; are we in cooked mode?
6947				 */
6948				if (handlep->cooked) {
6949					/*
6950					 * Yes, so we need to fix this
6951					 * instruction.
6952					 */
6953					if (fix_offset(p) < 0) {
6954						/*
6955						 * We failed to do so.
6956						 * Return 0, so our caller
6957						 * knows to punt to userland.
6958						 */
6959						return 0;
6960					}
6961				}
6962				break;
6963			}
6964			break;
6965		}
6966	}
6967	return 1;	/* we succeeded */
6968}
6969
6970static int
6971fix_offset(struct bpf_insn *p)
6972{
6973	/*
6974	 * What's the offset?
6975	 */
6976	if (p->k >= SLL_HDR_LEN) {
6977		/*
6978		 * It's within the link-layer payload; that starts at an
6979		 * offset of 0, as far as the kernel packet filter is
6980		 * concerned, so subtract the length of the link-layer
6981		 * header.
6982		 */
6983		p->k -= SLL_HDR_LEN;
6984	} else if (p->k == 0) {
6985		/*
6986		 * It's the packet type field; map it to the special magic
6987		 * kernel offset for that field.
6988		 */
6989		p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
6990	} else if (p->k == 14) {
6991		/*
6992		 * It's the protocol field; map it to the special magic
6993		 * kernel offset for that field.
6994		 */
6995		p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
6996	} else if ((bpf_int32)(p->k) > 0) {
6997		/*
6998		 * It's within the header, but it's not one of those
6999		 * fields; we can't do that in the kernel, so punt
7000		 * to userland.
7001		 */
7002		return -1;
7003	}
7004	return 0;
7005}
7006
7007static int
7008set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
7009{
7010	int total_filter_on = 0;
7011	int save_mode;
7012	int ret;
7013	int save_errno;
7014
7015	/*
7016	 * The socket filter code doesn't discard all packets queued
7017	 * up on the socket when the filter is changed; this means
7018	 * that packets that don't match the new filter may show up
7019	 * after the new filter is put onto the socket, if those
7020	 * packets haven't yet been read.
7021	 *
7022	 * This means, for example, that if you do a tcpdump capture
7023	 * with a filter, the first few packets in the capture might
7024	 * be packets that wouldn't have passed the filter.
7025	 *
7026	 * We therefore discard all packets queued up on the socket
7027	 * when setting a kernel filter.  (This isn't an issue for
7028	 * userland filters, as the userland filtering is done after
7029	 * packets are queued up.)
7030	 *
7031	 * To flush those packets, we put the socket in read-only mode,
7032	 * and read packets from the socket until there are no more to
7033	 * read.
7034	 *
7035	 * In order to keep that from being an infinite loop - i.e.,
7036	 * to keep more packets from arriving while we're draining
7037	 * the queue - we put the "total filter", which is a filter
7038	 * that rejects all packets, onto the socket before draining
7039	 * the queue.
7040	 *
7041	 * This code deliberately ignores any errors, so that you may
7042	 * get bogus packets if an error occurs, rather than having
7043	 * the filtering done in userland even if it could have been
7044	 * done in the kernel.
7045	 */
7046	if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
7047		       &total_fcode, sizeof(total_fcode)) == 0) {
7048		char drain[1];
7049
7050		/*
7051		 * Note that we've put the total filter onto the socket.
7052		 */
7053		total_filter_on = 1;
7054
7055		/*
7056		 * Save the socket's current mode, and put it in
7057		 * non-blocking mode; we drain it by reading packets
7058		 * until we get an error (which is normally a
7059		 * "nothing more to be read" error).
7060		 */
7061		save_mode = fcntl(handle->fd, F_GETFL, 0);
7062		if (save_mode == -1) {
7063			pcap_fmt_errmsg_for_errno(handle->errbuf,
7064			    PCAP_ERRBUF_SIZE, errno,
7065			    "can't get FD flags when changing filter");
7066			return -2;
7067		}
7068		if (fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) < 0) {
7069			pcap_fmt_errmsg_for_errno(handle->errbuf,
7070			    PCAP_ERRBUF_SIZE, errno,
7071			    "can't set nonblocking mode when changing filter");
7072			return -2;
7073		}
7074		while (recv(handle->fd, &drain, sizeof drain, MSG_TRUNC) >= 0)
7075			;
7076		save_errno = errno;
7077		if (save_errno != EAGAIN) {
7078			/*
7079			 * Fatal error.
7080			 *
7081			 * If we can't restore the mode or reset the
7082			 * kernel filter, there's nothing we can do.
7083			 */
7084			(void)fcntl(handle->fd, F_SETFL, save_mode);
7085			(void)reset_kernel_filter(handle);
7086			pcap_fmt_errmsg_for_errno(handle->errbuf,
7087			    PCAP_ERRBUF_SIZE, save_errno,
7088			    "recv failed when changing filter");
7089			return -2;
7090		}
7091		if (fcntl(handle->fd, F_SETFL, save_mode) == -1) {
7092			pcap_fmt_errmsg_for_errno(handle->errbuf,
7093			    PCAP_ERRBUF_SIZE, errno,
7094			    "can't restore FD flags when changing filter");
7095			return -2;
7096		}
7097	}
7098
7099	/*
7100	 * Now attach the new filter.
7101	 */
7102	ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
7103			 fcode, sizeof(*fcode));
7104	if (ret == -1 && total_filter_on) {
7105		/*
7106		 * Well, we couldn't set that filter on the socket,
7107		 * but we could set the total filter on the socket.
7108		 *
7109		 * This could, for example, mean that the filter was
7110		 * too big to put into the kernel, so we'll have to
7111		 * filter in userland; in any case, we'll be doing
7112		 * filtering in userland, so we need to remove the
7113		 * total filter so we see packets.
7114		 */
7115		save_errno = errno;
7116
7117		/*
7118		 * If this fails, we're really screwed; we have the
7119		 * total filter on the socket, and it won't come off.
7120		 * Report it as a fatal error.
7121		 */
7122		if (reset_kernel_filter(handle) == -1) {
7123			pcap_fmt_errmsg_for_errno(handle->errbuf,
7124			    PCAP_ERRBUF_SIZE, errno,
7125			    "can't remove kernel total filter");
7126			return -2;	/* fatal error */
7127		}
7128
7129		errno = save_errno;
7130	}
7131	return ret;
7132}
7133
7134static int
7135reset_kernel_filter(pcap_t *handle)
7136{
7137	int ret;
7138	/*
7139	 * setsockopt() barfs unless it get a dummy parameter.
7140	 * valgrind whines unless the value is initialized,
7141	 * as it has no idea that setsockopt() ignores its
7142	 * parameter.
7143	 */
7144	int dummy = 0;
7145
7146	ret = setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
7147				   &dummy, sizeof(dummy));
7148	/*
7149	 * Ignore ENOENT - it means "we don't have a filter", so there
7150	 * was no filter to remove, and there's still no filter.
7151	 *
7152	 * Also ignore ENONET, as a lot of kernel versions had a
7153	 * typo where ENONET, rather than ENOENT, was returned.
7154	 */
7155	if (ret == -1 && errno != ENOENT && errno != ENONET)
7156		return -1;
7157	return 0;
7158}
7159#endif
7160
7161int
7162pcap_set_protocol(pcap_t *p, int protocol)
7163{
7164	if (pcap_check_activated(p))
7165		return (PCAP_ERROR_ACTIVATED);
7166	p->opt.protocol = protocol;
7167	return (0);
7168}
7169
7170/*
7171 * Libpcap version string.
7172 */
7173const char *
7174pcap_lib_version(void)
7175{
7176#ifdef HAVE_PACKET_RING
7177 #if defined(HAVE_TPACKET3)
7178	return (PCAP_VERSION_STRING " (with TPACKET_V3)");
7179 #elif defined(HAVE_TPACKET2)
7180	return (PCAP_VERSION_STRING " (with TPACKET_V2)");
7181 #else
7182	return (PCAP_VERSION_STRING " (with TPACKET_V1)");
7183 #endif
7184#else
7185	return (PCAP_VERSION_STRING " (without TPACKET)");
7186#endif
7187}
7188