2 * pcap-linux.c: Packet capture interface to the Linux kernel
4 * Copyright (c) 2000 Torsten Landschoff <torsten@debian.org>
5 * Sebastian Krahmer <krahmer@cs.uni-potsdam.de>
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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
19 * 3. The names of the authors may not be used to endorse or promote
20 * products derived from this software without specific prior
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.
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>
32 * based on previous works of:
33 * Simon Patarin <patarin@cs.unibo.it>
34 * Phil Wood <cpw@lanl.gov>
36 * Monitor-mode support for mac80211 includes code taken from the iw
37 * command; the copyright notice for that code is
39 * Copyright (c) 2007, 2008 Johannes Berg
40 * Copyright (c) 2007 Andy Lutomirski
41 * Copyright (c) 2007 Mike Kershaw
42 * Copyright (c) 2008 Gábor Stefanik
44 * All rights reserved.
46 * Redistribution and use in source and binary forms, with or without
47 * modification, are permitted provided that the following conditions
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.
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
71 * Known problems with 2.0[.x] kernels:
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
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
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.
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.
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.
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>
141 #include "pcap-int.h"
142 #include "pcap/sll.h"
143 #include "pcap/vlan.h"
146 * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
147 * sockets rather than SOCK_PACKET sockets.
149 * To use them, we include <linux/if_packet.h> rather than
150 * <netpacket/packet.h>; we do so because
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>
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.
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>
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.
171 # include <linux/if_packet.h>
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.
180 * So we check whether PACKET_HOST is defined, and assume that we have
181 * PF_PACKET sockets only if it is defined.
184 # define HAVE_PF_PACKET_SOCKETS
185 # ifdef PACKET_AUXDATA
186 # define HAVE_PACKET_AUXDATA
187 # endif /* PACKET_AUXDATA */
188 # endif /* PACKET_HOST */
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 TPACKET_HDRLEN
195 # define HAVE_PACKET_RING
196 # ifdef TPACKET3_HDRLEN
197 # define HAVE_TPACKET3
198 # endif /* TPACKET3_HDRLEN */
199 # ifdef TPACKET2_HDRLEN
200 # define HAVE_TPACKET2
201 # else /* TPACKET2_HDRLEN */
202 # define TPACKET_V1 0 /* Old kernel with only V1, so no TPACKET_Vn defined */
203 # endif /* TPACKET2_HDRLEN */
204 # endif /* TPACKET_HDRLEN */
205 #endif /* PF_PACKET */
207 #ifdef SO_ATTACH_FILTER
208 #include <linux/types.h>
209 #include <linux/filter.h>
212 #ifdef HAVE_LINUX_NET_TSTAMP_H
213 #include <linux/net_tstamp.h>
217 * Got Wireless Extensions?
219 #ifdef HAVE_LINUX_WIRELESS_H
220 #include <linux/wireless.h>
221 #endif /* HAVE_LINUX_WIRELESS_H */
227 #include <linux/nl80211.h>
229 #include <netlink/genl/genl.h>
230 #include <netlink/genl/family.h>
231 #include <netlink/genl/ctrl.h>
232 #include <netlink/msg.h>
233 #include <netlink/attr.h>
234 #endif /* HAVE_LIBNL */
237 * Got ethtool support?
239 #ifdef HAVE_LINUX_ETHTOOL_H
240 #include <linux/ethtool.h>
243 #ifndef HAVE_SOCKLEN_T
244 typedef int socklen_t;
249 * This is being compiled on a system that lacks MSG_TRUNC; define it
250 * with the value it has in the 2.2 and later kernels, so that, on
251 * those kernels, when we pass it in the flags argument to "recvfrom()"
252 * we're passing the right value and thus get the MSG_TRUNC behavior
253 * we want. (We don't get that behavior on 2.0[.x] kernels, because
254 * they didn't support MSG_TRUNC.)
256 #define MSG_TRUNC 0x20
261 * This is being compiled on a system that lacks SOL_PACKET; define it
262 * with the value it has in the 2.2 and later kernels, so that we can
263 * set promiscuous mode in the good modern way rather than the old
264 * 2.0-kernel crappy way.
266 #define SOL_PACKET 263
269 #define MAX_LINKHEADER_SIZE 256
272 * When capturing on all interfaces we use this as the buffer size.
273 * Should be bigger then all MTUs that occur in real life.
274 * 64kB should be enough for now.
276 #define BIGGER_THAN_ALL_MTUS (64*1024)
279 * Private data for capturing on Linux SOCK_PACKET or PF_PACKET sockets.
282 u_int packets_read; /* count of packets read with recvfrom() */
283 long proc_dropped; /* packets reported dropped by /proc/net/dev */
284 struct pcap_stat stat;
286 char *device; /* device name */
287 int filter_in_userland; /* must filter in userland */
288 int blocks_to_filter_in_userland;
289 int must_do_on_close; /* stuff we must do when we close */
290 int timeout; /* timeout for buffering */
291 int sock_packet; /* using Linux 2.0 compatible interface */
292 int cooked; /* using SOCK_DGRAM rather than SOCK_RAW */
293 int ifindex; /* interface index of device we're bound to */
294 int lo_ifindex; /* interface index of the loopback device */
295 bpf_u_int32 oldmode; /* mode to restore when turning monitor mode off */
296 char *mondevice; /* mac80211 monitor device we created */
297 u_char *mmapbuf; /* memory-mapped region pointer */
298 size_t mmapbuflen; /* size of region */
299 int vlan_offset; /* offset at which to insert vlan tags; if -1, don't insert */
300 u_int tp_version; /* version of tpacket_hdr for mmaped ring */
301 u_int tp_hdrlen; /* hdrlen of tpacket_hdr for mmaped ring */
302 u_char *oneshot_buffer; /* buffer for copy of packet */
304 unsigned char *current_packet; /* Current packet within the TPACKET_V3 block. Move to next block if NULL. */
305 int packets_left; /* Unhandled packets left within the block from previous call to pcap_read_linux_mmap_v3 in case of TPACKET_V3. */
310 * Stuff to do when we close.
312 #define MUST_CLEAR_PROMISC 0x00000001 /* clear promiscuous mode */
313 #define MUST_CLEAR_RFMON 0x00000002 /* clear rfmon (monitor) mode */
314 #define MUST_DELETE_MONIF 0x00000004 /* delete monitor-mode interface */
317 * Prototypes for internal functions and methods.
319 static void map_arphrd_to_dlt(pcap_t *, int, const char *, int);
320 #ifdef HAVE_PF_PACKET_SOCKETS
321 static short int map_packet_type_to_sll_type(short int);
323 static int pcap_activate_linux(pcap_t *);
324 static int activate_old(pcap_t *);
325 static int activate_new(pcap_t *);
326 static int activate_mmap(pcap_t *, int *);
327 static int pcap_can_set_rfmon_linux(pcap_t *);
328 static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *);
329 static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);
330 static int pcap_inject_linux(pcap_t *, const void *, size_t);
331 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
332 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
333 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
334 static int pcap_set_datalink_linux(pcap_t *, int);
335 static void pcap_cleanup_linux(pcap_t *);
338 struct tpacket_hdr *h1;
340 struct tpacket2_hdr *h2;
343 struct tpacket_block_desc *h3;
348 #ifdef HAVE_PACKET_RING
349 #define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset])
351 static void destroy_ring(pcap_t *handle);
352 static int create_ring(pcap_t *handle, int *status);
353 static int prepare_tpacket_socket(pcap_t *handle);
354 static void pcap_cleanup_linux_mmap(pcap_t *);
355 static int pcap_read_linux_mmap_v1(pcap_t *, int, pcap_handler , u_char *);
357 static int pcap_read_linux_mmap_v2(pcap_t *, int, pcap_handler , u_char *);
360 static int pcap_read_linux_mmap_v3(pcap_t *, int, pcap_handler , u_char *);
362 static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *);
363 static int pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf);
364 static int pcap_getnonblock_mmap(pcap_t *p, char *errbuf);
365 static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
366 const u_char *bytes);
370 * Wrap some ioctl calls
372 #ifdef HAVE_PF_PACKET_SOCKETS
373 static int iface_get_id(int fd, const char *device, char *ebuf);
374 #endif /* HAVE_PF_PACKET_SOCKETS */
375 static int iface_get_mtu(int fd, const char *device, char *ebuf);
376 static int iface_get_arptype(int fd, const char *device, char *ebuf);
377 #ifdef HAVE_PF_PACKET_SOCKETS
378 static int iface_bind(int fd, int ifindex, char *ebuf);
379 #ifdef IW_MODE_MONITOR
380 static int has_wext(int sock_fd, const char *device, char *ebuf);
381 #endif /* IW_MODE_MONITOR */
382 static int enter_rfmon_mode(pcap_t *handle, int sock_fd,
384 #endif /* HAVE_PF_PACKET_SOCKETS */
385 static int iface_get_offload(pcap_t *handle);
386 static int iface_bind_old(int fd, const char *device, char *ebuf);
388 #ifdef SO_ATTACH_FILTER
389 static int fix_program(pcap_t *handle, struct sock_fprog *fcode,
391 static int fix_offset(struct bpf_insn *p);
392 static int set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
393 static int reset_kernel_filter(pcap_t *handle);
395 static struct sock_filter total_insn
396 = BPF_STMT(BPF_RET | BPF_K, 0);
397 static struct sock_fprog total_fcode
398 = { 1, &total_insn };
399 #endif /* SO_ATTACH_FILTER */
402 pcap_create_interface(const char *device, char *ebuf)
406 handle = pcap_create_common(device, ebuf, sizeof (struct pcap_linux));
410 handle->activate_op = pcap_activate_linux;
411 handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
412 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
414 * We claim that we support:
416 * software time stamps, with no details about their precision;
417 * hardware time stamps, synced to the host time;
418 * hardware time stamps, not synced to the host time.
420 * XXX - we can't ask a device whether it supports
421 * hardware time stamps, so we just claim all devices do.
423 handle->tstamp_type_count = 3;
424 handle->tstamp_type_list = malloc(3 * sizeof(u_int));
425 if (handle->tstamp_type_list == NULL) {
426 snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
427 pcap_strerror(errno));
431 handle->tstamp_type_list[0] = PCAP_TSTAMP_HOST;
432 handle->tstamp_type_list[1] = PCAP_TSTAMP_ADAPTER;
433 handle->tstamp_type_list[2] = PCAP_TSTAMP_ADAPTER_UNSYNCED;
436 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
438 * We claim that we support microsecond and nanosecond time
441 * XXX - with adapter-supplied time stamps, can we choose
442 * microsecond or nanosecond time stamps on arbitrary
445 handle->tstamp_precision_count = 2;
446 handle->tstamp_precision_list = malloc(2 * sizeof(u_int));
447 if (handle->tstamp_precision_list == NULL) {
448 snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
449 pcap_strerror(errno));
450 if (handle->tstamp_type_list != NULL)
451 free(handle->tstamp_type_list);
455 handle->tstamp_precision_list[0] = PCAP_TSTAMP_PRECISION_MICRO;
456 handle->tstamp_precision_list[1] = PCAP_TSTAMP_PRECISION_NANO;
457 #endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */
464 * If interface {if} is a mac80211 driver, the file
465 * /sys/class/net/{if}/phy80211 is a symlink to
466 * /sys/class/ieee80211/{phydev}, for some {phydev}.
468 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
469 * least, has a "wmaster0" device and a "wlan0" device; the
470 * latter is the one with the IP address. Both show up in
471 * "tcpdump -D" output. Capturing on the wmaster0 device
472 * captures with 802.11 headers.
474 * airmon-ng searches through /sys/class/net for devices named
475 * monN, starting with mon0; as soon as one *doesn't* exist,
476 * it chooses that as the monitor device name. If the "iw"
477 * command exists, it does "iw dev {if} interface add {monif}
478 * type monitor", where {monif} is the monitor device. It
479 * then (sigh) sleeps .1 second, and then configures the
480 * device up. Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
481 * is a file, it writes {mondev}, without a newline, to that file,
482 * and again (sigh) sleeps .1 second, and then iwconfig's that
483 * device into monitor mode and configures it up. Otherwise,
484 * you can't do monitor mode.
486 * All these devices are "glued" together by having the
487 * /sys/class/net/{device}/phy80211 links pointing to the same
488 * place, so, given a wmaster, wlan, or mon device, you can
489 * find the other devices by looking for devices with
490 * the same phy80211 link.
492 * To turn monitor mode off, delete the monitor interface,
493 * either with "iw dev {monif} interface del" or by sending
494 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
496 * Note: if you try to create a monitor device named "monN", and
497 * there's already a "monN" device, it fails, as least with
498 * the netlink interface (which is what iw uses), with a return
499 * value of -ENFILE. (Return values are negative errnos.) We
500 * could probably use that to find an unused device.
502 * Yes, you can have multiple monitor devices for a given
507 * Is this a mac80211 device? If so, fill in the physical device path and
508 * return 1; if not, return 0. On an error, fill in handle->errbuf and
512 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
513 size_t phydev_max_pathlen)
519 * Generate the path string for the symlink to the physical device.
521 if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
522 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
523 "%s: Can't generate path name string for /sys/class/net device",
527 bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
528 if (bytes_read == -1) {
529 if (errno == ENOENT || errno == EINVAL) {
531 * Doesn't exist, or not a symlink; assume that
532 * means it's not a mac80211 device.
537 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
538 "%s: Can't readlink %s: %s", device, pathstr,
544 phydev_path[bytes_read] = '\0';
548 #ifdef HAVE_LIBNL_SOCKETS
549 #define get_nl_errmsg nl_geterror
551 /* libnl 2.x compatibility code */
553 #define nl_sock nl_handle
555 static inline struct nl_handle *
556 nl_socket_alloc(void)
558 return nl_handle_alloc();
562 nl_socket_free(struct nl_handle *h)
564 nl_handle_destroy(h);
567 #define get_nl_errmsg strerror
570 __genl_ctrl_alloc_cache(struct nl_handle *h, struct nl_cache **cache)
572 struct nl_cache *tmp = genl_ctrl_alloc_cache(h);
578 #define genl_ctrl_alloc_cache __genl_ctrl_alloc_cache
579 #endif /* !HAVE_LIBNL_SOCKETS */
581 struct nl80211_state {
582 struct nl_sock *nl_sock;
583 struct nl_cache *nl_cache;
584 struct genl_family *nl80211;
588 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
592 state->nl_sock = nl_socket_alloc();
593 if (!state->nl_sock) {
594 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
595 "%s: failed to allocate netlink handle", device);
599 if (genl_connect(state->nl_sock)) {
600 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
601 "%s: failed to connect to generic netlink", device);
602 goto out_handle_destroy;
605 err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
607 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
608 "%s: failed to allocate generic netlink cache: %s",
609 device, get_nl_errmsg(-err));
610 goto out_handle_destroy;
613 state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
614 if (!state->nl80211) {
615 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
616 "%s: nl80211 not found", device);
623 nl_cache_free(state->nl_cache);
625 nl_socket_free(state->nl_sock);
630 nl80211_cleanup(struct nl80211_state *state)
632 genl_family_put(state->nl80211);
633 nl_cache_free(state->nl_cache);
634 nl_socket_free(state->nl_sock);
638 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
639 const char *device, const char *mondevice)
645 ifindex = iface_get_id(sock_fd, device, handle->errbuf);
651 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
652 "%s: failed to allocate netlink msg", device);
656 genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
657 0, NL80211_CMD_NEW_INTERFACE, 0);
658 NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
659 NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
660 NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
662 err = nl_send_auto_complete(state->nl_sock, msg);
664 #if defined HAVE_LIBNL_NLE
665 if (err == -NLE_FAILURE) {
667 if (err == -ENFILE) {
670 * Device not available; our caller should just
671 * keep trying. (libnl 2.x maps ENFILE to
672 * NLE_FAILURE; it can also map other errors
673 * to that, but there's not much we can do
680 * Real failure, not just "that device is not
683 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
684 "%s: nl_send_auto_complete failed adding %s interface: %s",
685 device, mondevice, get_nl_errmsg(-err));
690 err = nl_wait_for_ack(state->nl_sock);
692 #if defined HAVE_LIBNL_NLE
693 if (err == -NLE_FAILURE) {
695 if (err == -ENFILE) {
698 * Device not available; our caller should just
699 * keep trying. (libnl 2.x maps ENFILE to
700 * NLE_FAILURE; it can also map other errors
701 * to that, but there's not much we can do
708 * Real failure, not just "that device is not
711 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
712 "%s: nl_wait_for_ack failed adding %s interface: %s",
713 device, mondevice, get_nl_errmsg(-err));
726 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
727 "%s: nl_put failed adding %s interface",
734 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
735 const char *device, const char *mondevice)
741 ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
747 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
748 "%s: failed to allocate netlink msg", device);
752 genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
753 0, NL80211_CMD_DEL_INTERFACE, 0);
754 NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
756 err = nl_send_auto_complete(state->nl_sock, msg);
758 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
759 "%s: nl_send_auto_complete failed deleting %s interface: %s",
760 device, mondevice, get_nl_errmsg(-err));
764 err = nl_wait_for_ack(state->nl_sock);
766 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
767 "%s: nl_wait_for_ack failed adding %s interface: %s",
768 device, mondevice, get_nl_errmsg(-err));
780 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
781 "%s: nl_put failed deleting %s interface",
788 enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
790 struct pcap_linux *handlep = handle->priv;
792 char phydev_path[PATH_MAX+1];
793 struct nl80211_state nlstate;
798 * Is this a mac80211 device?
800 ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
802 return ret; /* error */
804 return 0; /* no error, but not mac80211 device */
807 * XXX - is this already a monN device?
809 * Is that determined by old Wireless Extensions ioctls?
813 * OK, it's apparently a mac80211 device.
814 * Try to find an unused monN device for it.
816 ret = nl80211_init(handle, &nlstate, device);
819 for (n = 0; n < UINT_MAX; n++) {
823 char mondevice[3+10+1]; /* mon{UINT_MAX}\0 */
825 snprintf(mondevice, sizeof mondevice, "mon%u", n);
826 ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
828 handlep->mondevice = strdup(mondevice);
833 * Hard failure. Just return ret; handle->errbuf
834 * has already been set.
836 nl80211_cleanup(&nlstate);
841 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
842 "%s: No free monN interfaces", device);
843 nl80211_cleanup(&nlstate);
850 * Sleep for .1 seconds.
853 delay.tv_nsec = 500000000;
854 nanosleep(&delay, NULL);
858 * If we haven't already done so, arrange to have
859 * "pcap_close_all()" called when we exit.
861 if (!pcap_do_addexit(handle)) {
863 * "atexit()" failed; don't put the interface
864 * in rfmon mode, just give up.
866 return PCAP_ERROR_RFMON_NOTSUP;
870 * Now configure the monitor interface up.
872 memset(&ifr, 0, sizeof(ifr));
873 strlcpy(ifr.ifr_name, handlep->mondevice, sizeof(ifr.ifr_name));
874 if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
875 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
876 "%s: Can't get flags for %s: %s", device,
877 handlep->mondevice, strerror(errno));
878 del_mon_if(handle, sock_fd, &nlstate, device,
880 nl80211_cleanup(&nlstate);
883 ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
884 if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
885 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
886 "%s: Can't set flags for %s: %s", device,
887 handlep->mondevice, strerror(errno));
888 del_mon_if(handle, sock_fd, &nlstate, device,
890 nl80211_cleanup(&nlstate);
895 * Success. Clean up the libnl state.
897 nl80211_cleanup(&nlstate);
900 * Note that we have to delete the monitor device when we close
903 handlep->must_do_on_close |= MUST_DELETE_MONIF;
906 * Add this to the list of pcaps to close when we exit.
908 pcap_add_to_pcaps_to_close(handle);
912 #endif /* HAVE_LIBNL */
915 pcap_can_set_rfmon_linux(pcap_t *handle)
918 char phydev_path[PATH_MAX+1];
921 #ifdef IW_MODE_MONITOR
926 if (strcmp(handle->opt.source, "any") == 0) {
928 * Monitor mode makes no sense on the "any" device.
935 * Bleah. There doesn't seem to be a way to ask a mac80211
936 * device, through libnl, whether it supports monitor mode;
937 * we'll just check whether the device appears to be a
938 * mac80211 device and, if so, assume the device supports
941 * wmaster devices don't appear to support the Wireless
942 * Extensions, but we can create a mon device for a
943 * wmaster device, so we don't bother checking whether
944 * a mac80211 device supports the Wireless Extensions.
946 ret = get_mac80211_phydev(handle, handle->opt.source, phydev_path,
949 return ret; /* error */
951 return 1; /* mac80211 device */
954 #ifdef IW_MODE_MONITOR
956 * Bleah. There doesn't appear to be an ioctl to use to ask
957 * whether a device supports monitor mode; we'll just do
958 * SIOCGIWMODE and, if it succeeds, assume the device supports
961 * Open a socket on which to attempt to get the mode.
962 * (We assume that if we have Wireless Extensions support
963 * we also have PF_PACKET support.)
965 sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
967 (void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
968 "socket: %s", pcap_strerror(errno));
973 * Attempt to get the current mode.
975 strlcpy(ireq.ifr_ifrn.ifrn_name, handle->opt.source,
976 sizeof ireq.ifr_ifrn.ifrn_name);
977 if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
979 * Well, we got the mode; assume we can set it.
984 if (errno == ENODEV) {
985 /* The device doesn't even exist. */
986 (void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
987 "SIOCGIWMODE failed: %s", pcap_strerror(errno));
989 return PCAP_ERROR_NO_SUCH_DEVICE;
997 * Grabs the number of dropped packets by the interface from /proc/net/dev.
999 * XXX - what about /sys/class/net/{interface name}/rx_*? There are
1000 * individual devices giving, in ASCII, various rx_ and tx_ statistics.
1002 * Or can we get them in binary form from netlink?
1005 linux_if_drops(const char * if_name)
1010 int field_to_convert = 3, if_name_sz = strlen(if_name);
1011 long int dropped_pkts = 0;
1013 file = fopen("/proc/net/dev", "r");
1017 while (!dropped_pkts && fgets( buffer, sizeof(buffer), file ))
1019 /* search for 'bytes' -- if its in there, then
1020 that means we need to grab the fourth field. otherwise
1021 grab the third field. */
1022 if (field_to_convert != 4 && strstr(buffer, "bytes"))
1024 field_to_convert = 4;
1028 /* find iface and make sure it actually matches -- space before the name and : after it */
1029 if ((bufptr = strstr(buffer, if_name)) &&
1030 (bufptr == buffer || *(bufptr-1) == ' ') &&
1031 *(bufptr + if_name_sz) == ':')
1033 bufptr = bufptr + if_name_sz + 1;
1035 /* grab the nth field from it */
1036 while( --field_to_convert && *bufptr != '\0')
1038 while (*bufptr != '\0' && *(bufptr++) == ' ');
1039 while (*bufptr != '\0' && *(bufptr++) != ' ');
1042 /* get rid of any final spaces */
1043 while (*bufptr != '\0' && *bufptr == ' ') bufptr++;
1045 if (*bufptr != '\0')
1046 dropped_pkts = strtol(bufptr, NULL, 10);
1053 return dropped_pkts;
1058 * With older kernels promiscuous mode is kind of interesting because we
1059 * have to reset the interface before exiting. The problem can't really
1060 * be solved without some daemon taking care of managing usage counts.
1061 * If we put the interface into promiscuous mode, we set a flag indicating
1062 * that we must take it out of that mode when the interface is closed,
1063 * and, when closing the interface, if that flag is set we take it out
1064 * of promiscuous mode.
1066 * Even with newer kernels, we have the same issue with rfmon mode.
1069 static void pcap_cleanup_linux( pcap_t *handle )
1071 struct pcap_linux *handlep = handle->priv;
1074 struct nl80211_state nlstate;
1076 #endif /* HAVE_LIBNL */
1077 #ifdef IW_MODE_MONITOR
1080 #endif /* IW_MODE_MONITOR */
1082 if (handlep->must_do_on_close != 0) {
1084 * There's something we have to do when closing this
1087 if (handlep->must_do_on_close & MUST_CLEAR_PROMISC) {
1089 * We put the interface into promiscuous mode;
1090 * take it out of promiscuous mode.
1092 * XXX - if somebody else wants it in promiscuous
1093 * mode, this code cannot know that, so it'll take
1094 * it out of promiscuous mode. That's not fixable
1095 * in 2.0[.x] kernels.
1097 memset(&ifr, 0, sizeof(ifr));
1098 strlcpy(ifr.ifr_name, handlep->device,
1099 sizeof(ifr.ifr_name));
1100 if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
1102 "Can't restore interface %s flags (SIOCGIFFLAGS failed: %s).\n"
1103 "Please adjust manually.\n"
1104 "Hint: This can't happen with Linux >= 2.2.0.\n",
1105 handlep->device, strerror(errno));
1107 if (ifr.ifr_flags & IFF_PROMISC) {
1109 * Promiscuous mode is currently on;
1112 ifr.ifr_flags &= ~IFF_PROMISC;
1113 if (ioctl(handle->fd, SIOCSIFFLAGS,
1116 "Can't restore interface %s flags (SIOCSIFFLAGS failed: %s).\n"
1117 "Please adjust manually.\n"
1118 "Hint: This can't happen with Linux >= 2.2.0.\n",
1127 if (handlep->must_do_on_close & MUST_DELETE_MONIF) {
1128 ret = nl80211_init(handle, &nlstate, handlep->device);
1130 ret = del_mon_if(handle, handle->fd, &nlstate,
1131 handlep->device, handlep->mondevice);
1132 nl80211_cleanup(&nlstate);
1136 "Can't delete monitor interface %s (%s).\n"
1137 "Please delete manually.\n",
1138 handlep->mondevice, handle->errbuf);
1141 #endif /* HAVE_LIBNL */
1143 #ifdef IW_MODE_MONITOR
1144 if (handlep->must_do_on_close & MUST_CLEAR_RFMON) {
1146 * We put the interface into rfmon mode;
1147 * take it out of rfmon mode.
1149 * XXX - if somebody else wants it in rfmon
1150 * mode, this code cannot know that, so it'll take
1151 * it out of rfmon mode.
1155 * First, take the interface down if it's up;
1156 * otherwise, we might get EBUSY.
1157 * If we get errors, just drive on and print
1158 * a warning if we can't restore the mode.
1161 memset(&ifr, 0, sizeof(ifr));
1162 strlcpy(ifr.ifr_name, handlep->device,
1163 sizeof(ifr.ifr_name));
1164 if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) != -1) {
1165 if (ifr.ifr_flags & IFF_UP) {
1166 oldflags = ifr.ifr_flags;
1167 ifr.ifr_flags &= ~IFF_UP;
1168 if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1)
1169 oldflags = 0; /* didn't set, don't restore */
1174 * Now restore the mode.
1176 strlcpy(ireq.ifr_ifrn.ifrn_name, handlep->device,
1177 sizeof ireq.ifr_ifrn.ifrn_name);
1178 ireq.u.mode = handlep->oldmode;
1179 if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
1181 * Scientist, you've failed.
1184 "Can't restore interface %s wireless mode (SIOCSIWMODE failed: %s).\n"
1185 "Please adjust manually.\n",
1186 handlep->device, strerror(errno));
1190 * Now bring the interface back up if we brought
1193 if (oldflags != 0) {
1194 ifr.ifr_flags = oldflags;
1195 if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
1197 "Can't bring interface %s back up (SIOCSIFFLAGS failed: %s).\n"
1198 "Please adjust manually.\n",
1199 handlep->device, strerror(errno));
1203 #endif /* IW_MODE_MONITOR */
1206 * Take this pcap out of the list of pcaps for which we
1207 * have to take the interface out of some mode.
1209 pcap_remove_from_pcaps_to_close(handle);
1212 if (handlep->mondevice != NULL) {
1213 free(handlep->mondevice);
1214 handlep->mondevice = NULL;
1216 if (handlep->device != NULL) {
1217 free(handlep->device);
1218 handlep->device = NULL;
1220 pcap_cleanup_live_common(handle);
1224 * Get a handle for a live capture from the given device. You can
1225 * pass NULL as device to get all packages (without link level
1226 * information of course). If you pass 1 as promisc the interface
1227 * will be set to promiscous mode (XXX: I think this usage should
1228 * be deprecated and functions be added to select that later allow
1229 * modification of that values -- Torsten).
1232 pcap_activate_linux(pcap_t *handle)
1234 struct pcap_linux *handlep = handle->priv;
1240 device = handle->opt.source;
1243 * Make sure the name we were handed will fit into the ioctls we
1244 * might perform on the device; if not, return a "No such device"
1245 * indication, as the Linux kernel shouldn't support creating
1246 * a device whose name won't fit into those ioctls.
1248 * "Will fit" means "will fit, complete with a null terminator",
1249 * so if the length, which does *not* include the null terminator,
1250 * is greater than *or equal to* the size of the field into which
1251 * we'll be copying it, that won't fit.
1253 if (strlen(device) >= sizeof(ifr.ifr_name)) {
1254 status = PCAP_ERROR_NO_SUCH_DEVICE;
1258 handle->inject_op = pcap_inject_linux;
1259 handle->setfilter_op = pcap_setfilter_linux;
1260 handle->setdirection_op = pcap_setdirection_linux;
1261 handle->set_datalink_op = pcap_set_datalink_linux;
1262 handle->getnonblock_op = pcap_getnonblock_fd;
1263 handle->setnonblock_op = pcap_setnonblock_fd;
1264 handle->cleanup_op = pcap_cleanup_linux;
1265 handle->read_op = pcap_read_linux;
1266 handle->stats_op = pcap_stats_linux;
1269 * The "any" device is a special device which causes us not
1270 * to bind to a particular device and thus to look at all
1273 if (strcmp(device, "any") == 0) {
1274 if (handle->opt.promisc) {
1275 handle->opt.promisc = 0;
1276 /* Just a warning. */
1277 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1278 "Promiscuous mode not supported on the \"any\" device");
1279 status = PCAP_WARNING_PROMISC_NOTSUP;
1283 handlep->device = strdup(device);
1284 if (handlep->device == NULL) {
1285 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
1286 pcap_strerror(errno) );
1290 /* copy timeout value */
1291 handlep->timeout = handle->opt.timeout;
1294 * If we're in promiscuous mode, then we probably want
1295 * to see when the interface drops packets too, so get an
1296 * initial count from /proc/net/dev
1298 if (handle->opt.promisc)
1299 handlep->proc_dropped = linux_if_drops(handlep->device);
1302 * Current Linux kernels use the protocol family PF_PACKET to
1303 * allow direct access to all packets on the network while
1304 * older kernels had a special socket type SOCK_PACKET to
1305 * implement this feature.
1306 * While this old implementation is kind of obsolete we need
1307 * to be compatible with older kernels for a while so we are
1308 * trying both methods with the newer method preferred.
1310 ret = activate_new(handle);
1313 * Fatal error with the new way; just fail.
1314 * ret has the error return; if it's PCAP_ERROR,
1315 * handle->errbuf has been set appropriately.
1323 * Try to use memory-mapped access.
1325 switch (activate_mmap(handle, &status)) {
1329 * We succeeded. status has been
1330 * set to the status to return,
1331 * which might be 0, or might be
1332 * a PCAP_WARNING_ value.
1338 * Kernel doesn't support it - just continue
1339 * with non-memory-mapped access.
1345 * We failed to set up to use it, or the kernel
1346 * supports it, but we failed to enable it.
1347 * ret has been set to the error status to
1348 * return and, if it's PCAP_ERROR, handle->errbuf
1349 * contains the error message.
1355 else if (ret == 0) {
1356 /* Non-fatal error; try old way */
1357 if ((ret = activate_old(handle)) != 1) {
1359 * Both methods to open the packet socket failed.
1360 * Tidy up and report our failure (handle->errbuf
1361 * is expected to be set by the functions above).
1369 * We set up the socket, but not with memory-mapped access.
1371 if (handle->opt.buffer_size != 0) {
1373 * Set the socket buffer size to the specified value.
1375 if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF,
1376 &handle->opt.buffer_size,
1377 sizeof(handle->opt.buffer_size)) == -1) {
1378 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1379 "SO_RCVBUF: %s", pcap_strerror(errno));
1380 status = PCAP_ERROR;
1385 /* Allocate the buffer */
1387 handle->buffer = malloc(handle->bufsize + handle->offset);
1388 if (!handle->buffer) {
1389 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1390 "malloc: %s", pcap_strerror(errno));
1391 status = PCAP_ERROR;
1396 * "handle->fd" is a socket, so "select()" and "poll()"
1397 * should work on it.
1399 handle->selectable_fd = handle->fd;
1404 pcap_cleanup_linux(handle);
1409 * Read at most max_packets from the capture stream and call the callback
1410 * for each of them. Returns the number of packets handled or -1 if an
1414 pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
1417 * Currently, on Linux only one packet is delivered per read,
1420 return pcap_read_packet(handle, callback, user);
1424 pcap_set_datalink_linux(pcap_t *handle, int dlt)
1426 handle->linktype = dlt;
1431 * linux_check_direction()
1433 * Do checks based on packet direction.
1436 linux_check_direction(const pcap_t *handle, const struct sockaddr_ll *sll)
1438 struct pcap_linux *handlep = handle->priv;
1440 if (sll->sll_pkttype == PACKET_OUTGOING) {
1443 * If this is from the loopback device, reject it;
1444 * we'll see the packet as an incoming packet as well,
1445 * and we don't want to see it twice.
1447 if (sll->sll_ifindex == handlep->lo_ifindex)
1451 * If the user only wants incoming packets, reject it.
1453 if (handle->direction == PCAP_D_IN)
1458 * If the user only wants outgoing packets, reject it.
1460 if (handle->direction == PCAP_D_OUT)
1467 * Read a packet from the socket calling the handler provided by
1468 * the user. Returns the number of packets received or -1 if an
1472 pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
1474 struct pcap_linux *handlep = handle->priv;
1477 #ifdef HAVE_PF_PACKET_SOCKETS
1478 struct sockaddr_ll from;
1479 struct sll_header *hdrp;
1481 struct sockaddr from;
1483 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1486 struct cmsghdr *cmsg;
1488 struct cmsghdr cmsg;
1489 char buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
1491 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1493 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1494 int packet_len, caplen;
1495 struct pcap_pkthdr pcap_header;
1497 #ifdef HAVE_PF_PACKET_SOCKETS
1499 * If this is a cooked device, leave extra room for a
1500 * fake packet header.
1502 if (handlep->cooked)
1503 offset = SLL_HDR_LEN;
1508 * This system doesn't have PF_PACKET sockets, so it doesn't
1509 * support cooked devices.
1515 * Receive a single packet from the kernel.
1516 * We ignore EINTR, as that might just be due to a signal
1517 * being delivered - if the signal should interrupt the
1518 * loop, the signal handler should call pcap_breakloop()
1519 * to set handle->break_loop (we ignore it on other
1520 * platforms as well).
1521 * We also ignore ENETDOWN, so that we can continue to
1522 * capture traffic if the interface goes down and comes
1523 * back up again; comments in the kernel indicate that
1524 * we'll just block waiting for packets if we try to
1525 * receive from a socket that delivered ENETDOWN, and,
1526 * if we're using a memory-mapped buffer, we won't even
1527 * get notified of "network down" events.
1529 bp = handle->buffer + handle->offset;
1531 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1532 msg.msg_name = &from;
1533 msg.msg_namelen = sizeof(from);
1536 msg.msg_control = &cmsg_buf;
1537 msg.msg_controllen = sizeof(cmsg_buf);
1540 iov.iov_len = handle->bufsize - offset;
1541 iov.iov_base = bp + offset;
1542 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1546 * Has "pcap_breakloop()" been called?
1548 if (handle->break_loop) {
1550 * Yes - clear the flag that indicates that it has,
1551 * and return PCAP_ERROR_BREAK as an indication that
1552 * we were told to break out of the loop.
1554 handle->break_loop = 0;
1555 return PCAP_ERROR_BREAK;
1558 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1559 packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC);
1560 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1561 fromlen = sizeof(from);
1562 packet_len = recvfrom(
1563 handle->fd, bp + offset,
1564 handle->bufsize - offset, MSG_TRUNC,
1565 (struct sockaddr *) &from, &fromlen);
1566 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1567 } while (packet_len == -1 && errno == EINTR);
1569 /* Check if an error occured */
1571 if (packet_len == -1) {
1575 return 0; /* no packet there */
1579 * The device on which we're capturing went away.
1581 * XXX - we should really return
1582 * PCAP_ERROR_IFACE_NOT_UP, but pcap_dispatch()
1583 * etc. aren't defined to return that.
1585 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1586 "The interface went down");
1590 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1591 "recvfrom: %s", pcap_strerror(errno));
1596 #ifdef HAVE_PF_PACKET_SOCKETS
1597 if (!handlep->sock_packet) {
1599 * Unfortunately, there is a window between socket() and
1600 * bind() where the kernel may queue packets from any
1601 * interface. If we're bound to a particular interface,
1602 * discard packets not from that interface.
1604 * (If socket filters are supported, we could do the
1605 * same thing we do when changing the filter; however,
1606 * that won't handle packet sockets without socket
1607 * filter support, and it's a bit more complicated.
1608 * It would save some instructions per packet, however.)
1610 if (handlep->ifindex != -1 &&
1611 from.sll_ifindex != handlep->ifindex)
1615 * Do checks based on packet direction.
1616 * We can only do this if we're using PF_PACKET; the
1617 * address returned for SOCK_PACKET is a "sockaddr_pkt"
1618 * which lacks the relevant packet type information.
1620 if (!linux_check_direction(handle, &from))
1625 #ifdef HAVE_PF_PACKET_SOCKETS
1627 * If this is a cooked device, fill in the fake packet header.
1629 if (handlep->cooked) {
1631 * Add the length of the fake header to the length
1632 * of packet data we read.
1634 packet_len += SLL_HDR_LEN;
1636 hdrp = (struct sll_header *)bp;
1637 hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype);
1638 hdrp->sll_hatype = htons(from.sll_hatype);
1639 hdrp->sll_halen = htons(from.sll_halen);
1640 memcpy(hdrp->sll_addr, from.sll_addr,
1641 (from.sll_halen > SLL_ADDRLEN) ?
1644 hdrp->sll_protocol = from.sll_protocol;
1647 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
1648 if (handlep->vlan_offset != -1) {
1649 for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
1650 struct tpacket_auxdata *aux;
1652 struct vlan_tag *tag;
1654 if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
1655 cmsg->cmsg_level != SOL_PACKET ||
1656 cmsg->cmsg_type != PACKET_AUXDATA)
1659 aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
1660 #if defined(TP_STATUS_VLAN_VALID)
1661 if ((aux->tp_vlan_tci == 0) && !(aux->tp_status & TP_STATUS_VLAN_VALID))
1663 if (aux->tp_vlan_tci == 0) /* this is ambigious but without the
1664 TP_STATUS_VLAN_VALID flag, there is
1665 nothing that we can do */
1669 len = packet_len > iov.iov_len ? iov.iov_len : packet_len;
1670 if (len < (unsigned int) handlep->vlan_offset)
1674 memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
1676 tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
1677 tag->vlan_tpid = htons(ETH_P_8021Q);
1678 tag->vlan_tci = htons(aux->tp_vlan_tci);
1680 packet_len += VLAN_TAG_LEN;
1683 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
1684 #endif /* HAVE_PF_PACKET_SOCKETS */
1687 * XXX: According to the kernel source we should get the real
1688 * packet len if calling recvfrom with MSG_TRUNC set. It does
1689 * not seem to work here :(, but it is supported by this code
1691 * To be honest the code RELIES on that feature so this is really
1692 * broken with 2.2.x kernels.
1693 * I spend a day to figure out what's going on and I found out
1694 * that the following is happening:
1696 * The packet comes from a random interface and the packet_rcv
1697 * hook is called with a clone of the packet. That code inserts
1698 * the packet into the receive queue of the packet socket.
1699 * If a filter is attached to that socket that filter is run
1700 * first - and there lies the problem. The default filter always
1701 * cuts the packet at the snaplen:
1706 * So the packet filter cuts down the packet. The recvfrom call
1707 * says "hey, it's only 68 bytes, it fits into the buffer" with
1708 * the result that we don't get the real packet length. This
1709 * is valid at least until kernel 2.2.17pre6.
1711 * We currently handle this by making a copy of the filter
1712 * program, fixing all "ret" instructions with non-zero
1713 * operands to have an operand of MAXIMUM_SNAPLEN so that the
1714 * filter doesn't truncate the packet, and supplying that modified
1715 * filter to the kernel.
1718 caplen = packet_len;
1719 if (caplen > handle->snapshot)
1720 caplen = handle->snapshot;
1722 /* Run the packet filter if not using kernel filter */
1723 if (handlep->filter_in_userland && handle->fcode.bf_insns) {
1724 if (bpf_filter(handle->fcode.bf_insns, bp,
1725 packet_len, caplen) == 0)
1727 /* rejected by filter */
1732 /* Fill in our own header data */
1734 /* get timestamp for this packet */
1735 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
1736 if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
1737 if (ioctl(handle->fd, SIOCGSTAMPNS, &pcap_header.ts) == -1) {
1738 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1739 "SIOCGSTAMPNS: %s", pcap_strerror(errno));
1745 if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
1746 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1747 "SIOCGSTAMP: %s", pcap_strerror(errno));
1752 pcap_header.caplen = caplen;
1753 pcap_header.len = packet_len;
1758 * Arguably, we should count them before we check the filter,
1759 * as on many other platforms "ps_recv" counts packets
1760 * handed to the filter rather than packets that passed
1761 * the filter, but if filtering is done in the kernel, we
1762 * can't get a count of packets that passed the filter,
1763 * and that would mean the meaning of "ps_recv" wouldn't
1764 * be the same on all Linux systems.
1766 * XXX - it's not the same on all systems in any case;
1767 * ideally, we should have a "get the statistics" call
1768 * that supplies more counts and indicates which of them
1769 * it supplies, so that we supply a count of packets
1770 * handed to the filter only on platforms where that
1771 * information is available.
1773 * We count them here even if we can get the packet count
1774 * from the kernel, as we can only determine at run time
1775 * whether we'll be able to get it from the kernel (if
1776 * HAVE_TPACKET_STATS isn't defined, we can't get it from
1777 * the kernel, but if it is defined, the library might
1778 * have been built with a 2.4 or later kernel, but we
1779 * might be running on a 2.2[.x] kernel without Alexey
1780 * Kuznetzov's turbopacket patches, and thus the kernel
1781 * might not be able to supply those statistics). We
1782 * could, I guess, try, when opening the socket, to get
1783 * the statistics, and if we can not increment the count
1784 * here, but it's not clear that always incrementing
1785 * the count is more expensive than always testing a flag
1788 * We keep the count in "handlep->packets_read", and use that
1789 * for "ps_recv" if we can't get the statistics from the kernel.
1790 * We do that because, if we *can* get the statistics from
1791 * the kernel, we use "handlep->stat.ps_recv" and
1792 * "handlep->stat.ps_drop" as running counts, as reading the
1793 * statistics from the kernel resets the kernel statistics,
1794 * and if we directly increment "handlep->stat.ps_recv" here,
1795 * that means it will count packets *twice* on systems where
1796 * we can get kernel statistics - once here, and once in
1797 * pcap_stats_linux().
1799 handlep->packets_read++;
1801 /* Call the user supplied callback function */
1802 callback(userdata, &pcap_header, bp);
1808 pcap_inject_linux(pcap_t *handle, const void *buf, size_t size)
1810 struct pcap_linux *handlep = handle->priv;
1813 #ifdef HAVE_PF_PACKET_SOCKETS
1814 if (!handlep->sock_packet) {
1815 /* PF_PACKET socket */
1816 if (handlep->ifindex == -1) {
1818 * We don't support sending on the "any" device.
1820 strlcpy(handle->errbuf,
1821 "Sending packets isn't supported on the \"any\" device",
1826 if (handlep->cooked) {
1828 * We don't support sending on the "any" device.
1830 * XXX - how do you send on a bound cooked-mode
1832 * Is a "sendto()" required there?
1834 strlcpy(handle->errbuf,
1835 "Sending packets isn't supported in cooked mode",
1842 ret = send(handle->fd, buf, size, 0);
1844 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
1845 pcap_strerror(errno));
1852 * Get the statistics for the given packet capture handle.
1853 * Reports the number of dropped packets iff the kernel supports
1854 * the PACKET_STATISTICS "getsockopt()" argument (2.4 and later
1855 * kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket
1856 * patches); otherwise, that information isn't available, and we lie
1857 * and report 0 as the count of dropped packets.
1860 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
1862 struct pcap_linux *handlep = handle->priv;
1863 #ifdef HAVE_TPACKET_STATS
1864 #ifdef HAVE_TPACKET3
1866 * For sockets using TPACKET_V1 or TPACKET_V2, the extra
1867 * stuff at the end of a struct tpacket_stats_v3 will not
1868 * be filled in, and we don't look at it so this is OK even
1869 * for those sockets. In addition, the PF_PACKET socket
1870 * code in the kernel only uses the length parameter to
1871 * compute how much data to copy out and to indicate how
1872 * much data was copied out, so it's OK to base it on the
1873 * size of a struct tpacket_stats.
1875 * XXX - it's probably OK, in fact, to just use a
1876 * struct tpacket_stats for V3 sockets, as we don't
1877 * care about the tp_freeze_q_cnt stat.
1879 struct tpacket_stats_v3 kstats;
1880 #else /* HAVE_TPACKET3 */
1881 struct tpacket_stats kstats;
1882 #endif /* HAVE_TPACKET3 */
1883 socklen_t len = sizeof (struct tpacket_stats);
1884 #endif /* HAVE_TPACKET_STATS */
1886 long if_dropped = 0;
1889 * To fill in ps_ifdrop, we parse /proc/net/dev for the number
1891 if (handle->opt.promisc)
1893 if_dropped = handlep->proc_dropped;
1894 handlep->proc_dropped = linux_if_drops(handlep->device);
1895 handlep->stat.ps_ifdrop += (handlep->proc_dropped - if_dropped);
1898 #ifdef HAVE_TPACKET_STATS
1900 * Try to get the packet counts from the kernel.
1902 if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
1903 &kstats, &len) > -1) {
1905 * On systems where the PACKET_STATISTICS "getsockopt()"
1906 * argument is supported on PF_PACKET sockets:
1908 * "ps_recv" counts only packets that *passed* the
1909 * filter, not packets that didn't pass the filter.
1910 * This includes packets later dropped because we
1911 * ran out of buffer space.
1913 * "ps_drop" counts packets dropped because we ran
1914 * out of buffer space. It doesn't count packets
1915 * dropped by the interface driver. It counts only
1916 * packets that passed the filter.
1918 * See above for ps_ifdrop.
1920 * Both statistics include packets not yet read from
1921 * the kernel by libpcap, and thus not yet seen by
1924 * In "linux/net/packet/af_packet.c", at least in the
1925 * 2.4.9 kernel, "tp_packets" is incremented for every
1926 * packet that passes the packet filter *and* is
1927 * successfully queued on the socket; "tp_drops" is
1928 * incremented for every packet dropped because there's
1929 * not enough free space in the socket buffer.
1931 * When the statistics are returned for a PACKET_STATISTICS
1932 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
1933 * so that "tp_packets" counts all packets handed to
1934 * the PF_PACKET socket, including packets dropped because
1935 * there wasn't room on the socket buffer - but not
1936 * including packets that didn't pass the filter.
1938 * In the BSD BPF, the count of received packets is
1939 * incremented for every packet handed to BPF, regardless
1940 * of whether it passed the filter.
1942 * We can't make "pcap_stats()" work the same on both
1943 * platforms, but the best approximation is to return
1944 * "tp_packets" as the count of packets and "tp_drops"
1945 * as the count of drops.
1947 * Keep a running total because each call to
1948 * getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
1949 * resets the counters to zero.
1951 handlep->stat.ps_recv += kstats.tp_packets;
1952 handlep->stat.ps_drop += kstats.tp_drops;
1953 *stats = handlep->stat;
1959 * If the error was EOPNOTSUPP, fall through, so that
1960 * if you build the library on a system with
1961 * "struct tpacket_stats" and run it on a system
1962 * that doesn't, it works as it does if the library
1963 * is built on a system without "struct tpacket_stats".
1965 if (errno != EOPNOTSUPP) {
1966 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1967 "pcap_stats: %s", pcap_strerror(errno));
1973 * On systems where the PACKET_STATISTICS "getsockopt()" argument
1974 * is not supported on PF_PACKET sockets:
1976 * "ps_recv" counts only packets that *passed* the filter,
1977 * not packets that didn't pass the filter. It does not
1978 * count packets dropped because we ran out of buffer
1981 * "ps_drop" is not supported.
1983 * "ps_ifdrop" is supported. It will return the number
1984 * of drops the interface reports in /proc/net/dev,
1985 * if that is available.
1987 * "ps_recv" doesn't include packets not yet read from
1988 * the kernel by libpcap.
1990 * We maintain the count of packets processed by libpcap in
1991 * "handlep->packets_read", for reasons described in the comment
1992 * at the end of pcap_read_packet(). We have no idea how many
1993 * packets were dropped by the kernel buffers -- but we know
1994 * how many the interface dropped, so we can return that.
1997 stats->ps_recv = handlep->packets_read;
1999 stats->ps_ifdrop = handlep->stat.ps_ifdrop;
2004 add_linux_if(pcap_if_t **devlistp, const char *ifname, int fd, char *errbuf)
2007 char name[512]; /* XXX - pick a size */
2009 struct ifreq ifrflags;
2012 * Get the interface name.
2016 while (*p != '\0' && isascii(*p) && !isspace(*p)) {
2019 * This could be the separator between a
2020 * name and an alias number, or it could be
2021 * the separator between a name with no
2022 * alias number and the next field.
2024 * If there's a colon after digits, it
2025 * separates the name and the alias number,
2026 * otherwise it separates the name and the
2030 while (isascii(*p) && isdigit(*p))
2034 * That was the next field,
2035 * not the alias number.
2046 * Get the flags for this interface.
2048 strlcpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name));
2049 if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) {
2050 if (errno == ENXIO || errno == ENODEV)
2051 return (0); /* device doesn't actually exist - ignore it */
2052 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2053 "SIOCGIFFLAGS: %.*s: %s",
2054 (int)sizeof(ifrflags.ifr_name),
2056 pcap_strerror(errno));
2061 * Add an entry for this interface, with no addresses.
2063 if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL,
2075 * Get from "/sys/class/net" all interfaces listed there; if they're
2076 * already in the list of interfaces we have, that won't add another
2077 * instance, but if they're not, that'll add them.
2079 * We don't bother getting any addresses for them; it appears you can't
2080 * use SIOCGIFADDR on Linux to get IPv6 addresses for interfaces, and,
2081 * although some other types of addresses can be fetched with SIOCGIFADDR,
2082 * we don't bother with them for now.
2084 * We also don't fail if we couldn't open "/sys/class/net"; we just leave
2085 * the list of interfaces as is, and return 0, so that we can try
2086 * scanning /proc/net/dev.
2088 * Otherwise, we return 1 if we don't get an error and -1 if we do.
2091 scan_sys_class_net(pcap_if_t **devlistp, char *errbuf)
2093 DIR *sys_class_net_d;
2096 char subsystem_path[PATH_MAX+1];
2100 sys_class_net_d = opendir("/sys/class/net");
2101 if (sys_class_net_d == NULL) {
2103 * Don't fail if it doesn't exist at all.
2105 if (errno == ENOENT)
2109 * Fail if we got some other error.
2111 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2112 "Can't open /sys/class/net: %s", pcap_strerror(errno));
2117 * Create a socket from which to fetch interface information.
2119 fd = socket(AF_INET, SOCK_DGRAM, 0);
2121 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2122 "socket: %s", pcap_strerror(errno));
2123 (void)closedir(sys_class_net_d);
2129 ent = readdir(sys_class_net_d);
2132 * Error or EOF; if errno != 0, it's an error.
2138 * Ignore "." and "..".
2140 if (strcmp(ent->d_name, ".") == 0 ||
2141 strcmp(ent->d_name, "..") == 0)
2145 * Ignore plain files; they do not have subdirectories
2146 * and thus have no attributes.
2148 if (ent->d_type == DT_REG)
2152 * Is there an "ifindex" file under that name?
2153 * (We don't care whether it's a directory or
2154 * a symlink; older kernels have directories
2155 * for devices, newer kernels have symlinks to
2158 snprintf(subsystem_path, sizeof subsystem_path,
2159 "/sys/class/net/%s/ifindex", ent->d_name);
2160 if (lstat(subsystem_path, &statb) != 0) {
2162 * Stat failed. Either there was an error
2163 * other than ENOENT, and we don't know if
2164 * this is an interface, or it's ENOENT,
2165 * and either some part of "/sys/class/net/{if}"
2166 * disappeared, in which case it probably means
2167 * the interface disappeared, or there's no
2168 * "ifindex" file, which means it's not a
2169 * network interface.
2175 * Attempt to add the interface.
2177 if (add_linux_if(devlistp, &ent->d_name[0], fd, errbuf) == -1) {
2185 * Well, we didn't fail for any other reason; did we
2186 * fail due to an error reading the directory?
2189 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2190 "Error reading /sys/class/net: %s",
2191 pcap_strerror(errno));
2197 (void)closedir(sys_class_net_d);
2202 * Get from "/proc/net/dev" all interfaces listed there; if they're
2203 * already in the list of interfaces we have, that won't add another
2204 * instance, but if they're not, that'll add them.
2206 * See comments from scan_sys_class_net().
2209 scan_proc_net_dev(pcap_if_t **devlistp, char *errbuf)
2218 proc_net_f = fopen("/proc/net/dev", "r");
2219 if (proc_net_f == NULL) {
2221 * Don't fail if it doesn't exist at all.
2223 if (errno == ENOENT)
2227 * Fail if we got some other error.
2229 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2230 "Can't open /proc/net/dev: %s", pcap_strerror(errno));
2235 * Create a socket from which to fetch interface information.
2237 fd = socket(AF_INET, SOCK_DGRAM, 0);
2239 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2240 "socket: %s", pcap_strerror(errno));
2241 (void)fclose(proc_net_f);
2246 fgets(linebuf, sizeof linebuf, proc_net_f) != NULL; linenum++) {
2248 * Skip the first two lines - they're headers.
2256 * Skip leading white space.
2258 while (*p != '\0' && isascii(*p) && isspace(*p))
2260 if (*p == '\0' || *p == '\n')
2261 continue; /* blank line */
2264 * Attempt to add the interface.
2266 if (add_linux_if(devlistp, p, fd, errbuf) == -1) {
2274 * Well, we didn't fail for any other reason; did we
2275 * fail due to an error reading the file?
2277 if (ferror(proc_net_f)) {
2278 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE,
2279 "Error reading /proc/net/dev: %s",
2280 pcap_strerror(errno));
2286 (void)fclose(proc_net_f);
2291 * Description string for the "any" device.
2293 static const char any_descr[] = "Pseudo-device that captures on all interfaces";
2296 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
2301 * Read "/sys/class/net", and add to the list of interfaces all
2302 * interfaces listed there that we don't already have, because,
2303 * on Linux, SIOCGIFCONF reports only interfaces with IPv4 addresses,
2304 * and even getifaddrs() won't return information about
2305 * interfaces with no addresses, so you need to read "/sys/class/net"
2306 * to get the names of the rest of the interfaces.
2308 ret = scan_sys_class_net(alldevsp, errbuf);
2310 return (-1); /* failed */
2313 * No /sys/class/net; try reading /proc/net/dev instead.
2315 if (scan_proc_net_dev(alldevsp, errbuf) == -1)
2320 * Add the "any" device.
2322 if (pcap_add_if(alldevsp, "any", IFF_UP|IFF_RUNNING,
2323 any_descr, errbuf) < 0)
2330 * Attach the given BPF code to the packet capture device.
2333 pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
2336 struct pcap_linux *handlep;
2337 #ifdef SO_ATTACH_FILTER
2338 struct sock_fprog fcode;
2339 int can_filter_in_kernel;
2346 strlcpy(handle->errbuf, "setfilter: No filter specified",
2351 handlep = handle->priv;
2353 /* Make our private copy of the filter */
2355 if (install_bpf_program(handle, filter) < 0)
2356 /* install_bpf_program() filled in errbuf */
2360 * Run user level packet filter by default. Will be overriden if
2361 * installing a kernel filter succeeds.
2363 handlep->filter_in_userland = 1;
2365 /* Install kernel level filter if possible */
2367 #ifdef SO_ATTACH_FILTER
2369 if (handle->fcode.bf_len > USHRT_MAX) {
2371 * fcode.len is an unsigned short for current kernel.
2372 * I have yet to see BPF-Code with that much
2373 * instructions but still it is possible. So for the
2374 * sake of correctness I added this check.
2376 fprintf(stderr, "Warning: Filter too complex for kernel\n");
2378 fcode.filter = NULL;
2379 can_filter_in_kernel = 0;
2381 #endif /* USHRT_MAX */
2384 * Oh joy, the Linux kernel uses struct sock_fprog instead
2385 * of struct bpf_program and of course the length field is
2386 * of different size. Pointed out by Sebastian
2388 * Oh, and we also need to fix it up so that all "ret"
2389 * instructions with non-zero operands have MAXIMUM_SNAPLEN
2390 * as the operand if we're not capturing in memory-mapped
2391 * mode, and so that, if we're in cooked mode, all memory-
2392 * reference instructions use special magic offsets in
2393 * references to the link-layer header and assume that the
2394 * link-layer payload begins at 0; "fix_program()" will do
2397 switch (fix_program(handle, &fcode, is_mmapped)) {
2402 * Fatal error; just quit.
2403 * (The "default" case shouldn't happen; we
2404 * return -1 for that reason.)
2410 * The program performed checks that we can't make
2411 * work in the kernel.
2413 can_filter_in_kernel = 0;
2418 * We have a filter that'll work in the kernel.
2420 can_filter_in_kernel = 1;
2426 * NOTE: at this point, we've set both the "len" and "filter"
2427 * fields of "fcode". As of the 2.6.32.4 kernel, at least,
2428 * those are the only members of the "sock_fprog" structure,
2429 * so we initialize every member of that structure.
2431 * If there is anything in "fcode" that is not initialized,
2432 * it is either a field added in a later kernel, or it's
2435 * If a new field is added, this code needs to be updated
2436 * to set it correctly.
2438 * If there are no other fields, then:
2440 * if the Linux kernel looks at the padding, it's
2443 * if the Linux kernel doesn't look at the padding,
2444 * then if some tool complains that we're passing
2445 * uninitialized data to the kernel, then the tool
2446 * is buggy and needs to understand that it's just
2449 if (can_filter_in_kernel) {
2450 if ((err = set_kernel_filter(handle, &fcode)) == 0)
2453 * Installation succeded - using kernel filter,
2454 * so userland filtering not needed.
2456 handlep->filter_in_userland = 0;
2458 else if (err == -1) /* Non-fatal error */
2461 * Print a warning if we weren't able to install
2462 * the filter for a reason other than "this kernel
2463 * isn't configured to support socket filters.
2465 if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
2467 "Warning: Kernel filter failed: %s\n",
2468 pcap_strerror(errno));
2474 * If we're not using the kernel filter, get rid of any kernel
2475 * filter that might've been there before, e.g. because the
2476 * previous filter could work in the kernel, or because some other
2477 * code attached a filter to the socket by some means other than
2478 * calling "pcap_setfilter()". Otherwise, the kernel filter may
2479 * filter out packets that would pass the new userland filter.
2481 if (handlep->filter_in_userland)
2482 reset_kernel_filter(handle);
2485 * Free up the copy of the filter that was made by "fix_program()".
2487 if (fcode.filter != NULL)
2493 #endif /* SO_ATTACH_FILTER */
2499 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
2501 return pcap_setfilter_linux_common(handle, filter, 0);
2506 * Set direction flag: Which packets do we accept on a forwarding
2507 * single device? IN, OUT or both?
2510 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
2512 #ifdef HAVE_PF_PACKET_SOCKETS
2513 struct pcap_linux *handlep = handle->priv;
2515 if (!handlep->sock_packet) {
2516 handle->direction = d;
2521 * We're not using PF_PACKET sockets, so we can't determine
2522 * the direction of the packet.
2524 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2525 "Setting direction is not supported on SOCK_PACKET sockets");
2529 #ifdef HAVE_PF_PACKET_SOCKETS
2531 * Map the PACKET_ value to a LINUX_SLL_ value; we
2532 * want the same numerical value to be used in
2533 * the link-layer header even if the numerical values
2534 * for the PACKET_ #defines change, so that programs
2535 * that look at the packet type field will always be
2536 * able to handle DLT_LINUX_SLL captures.
2539 map_packet_type_to_sll_type(short int sll_pkttype)
2541 switch (sll_pkttype) {
2544 return htons(LINUX_SLL_HOST);
2546 case PACKET_BROADCAST:
2547 return htons(LINUX_SLL_BROADCAST);
2549 case PACKET_MULTICAST:
2550 return htons(LINUX_SLL_MULTICAST);
2552 case PACKET_OTHERHOST:
2553 return htons(LINUX_SLL_OTHERHOST);
2555 case PACKET_OUTGOING:
2556 return htons(LINUX_SLL_OUTGOING);
2565 * Linux uses the ARP hardware type to identify the type of an
2566 * interface. pcap uses the DLT_xxx constants for this. This
2567 * function takes a pointer to a "pcap_t", and an ARPHRD_xxx
2568 * constant, as arguments, and sets "handle->linktype" to the
2569 * appropriate DLT_XXX constant and sets "handle->offset" to
2570 * the appropriate value (to make "handle->offset" plus link-layer
2571 * header length be a multiple of 4, so that the link-layer payload
2572 * will be aligned on a 4-byte boundary when capturing packets).
2573 * (If the offset isn't set here, it'll be 0; add code as appropriate
2574 * for cases where it shouldn't be 0.)
2576 * If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
2577 * in cooked mode; otherwise, we can't use cooked mode, so we have
2578 * to pick some type that works in raw mode, or fail.
2580 * Sets the link type to -1 if unable to map the type.
2582 static void map_arphrd_to_dlt(pcap_t *handle, int arptype, const char *device,
2585 static const char cdma_rmnet[] = "cdma_rmnet";
2591 * For various annoying reasons having to do with DHCP
2592 * software, some versions of Android give the mobile-
2593 * phone-network interface an ARPHRD_ value of
2594 * ARPHRD_ETHER, even though the packet supplied by
2595 * that interface have no link-layer header, and begin
2596 * with an IP header, so that the ARPHRD_ value should
2599 * Detect those devices by checking the device name, and
2600 * use DLT_RAW for them.
2602 if (strncmp(device, cdma_rmnet, sizeof cdma_rmnet - 1) == 0) {
2603 handle->linktype = DLT_RAW;
2608 * This is (presumably) a real Ethernet capture; give it a
2609 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
2610 * that an application can let you choose it, in case you're
2611 * capturing DOCSIS traffic that a Cisco Cable Modem
2612 * Termination System is putting out onto an Ethernet (it
2613 * doesn't put an Ethernet header onto the wire, it puts raw
2614 * DOCSIS frames out on the wire inside the low-level
2615 * Ethernet framing).
2617 * XXX - are there any sorts of "fake Ethernet" that have
2618 * ARPHRD_ETHER but that *shouldn't offer DLT_DOCSIS as
2619 * a Cisco CMTS won't put traffic onto it or get traffic
2620 * bridged onto it? ISDN is handled in "activate_new()",
2621 * as we fall back on cooked mode there; are there any
2624 handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2626 * If that fails, just leave the list empty.
2628 if (handle->dlt_list != NULL) {
2629 handle->dlt_list[0] = DLT_EN10MB;
2630 handle->dlt_list[1] = DLT_DOCSIS;
2631 handle->dlt_count = 2;
2635 case ARPHRD_METRICOM:
2636 case ARPHRD_LOOPBACK:
2637 handle->linktype = DLT_EN10MB;
2642 handle->linktype = DLT_EN3MB;
2646 handle->linktype = DLT_AX25_KISS;
2650 handle->linktype = DLT_PRONET;
2654 handle->linktype = DLT_CHAOS;
2657 #define ARPHRD_CAN 280
2660 handle->linktype = DLT_CAN_SOCKETCAN;
2663 #ifndef ARPHRD_IEEE802_TR
2664 #define ARPHRD_IEEE802_TR 800 /* From Linux 2.4 */
2666 case ARPHRD_IEEE802_TR:
2667 case ARPHRD_IEEE802:
2668 handle->linktype = DLT_IEEE802;
2673 handle->linktype = DLT_ARCNET_LINUX;
2676 #ifndef ARPHRD_FDDI /* From Linux 2.2.13 */
2677 #define ARPHRD_FDDI 774
2680 handle->linktype = DLT_FDDI;
2684 #ifndef ARPHRD_ATM /* FIXME: How to #include this? */
2685 #define ARPHRD_ATM 19
2689 * The Classical IP implementation in ATM for Linux
2690 * supports both what RFC 1483 calls "LLC Encapsulation",
2691 * in which each packet has an LLC header, possibly
2692 * with a SNAP header as well, prepended to it, and
2693 * what RFC 1483 calls "VC Based Multiplexing", in which
2694 * different virtual circuits carry different network
2695 * layer protocols, and no header is prepended to packets.
2697 * They both have an ARPHRD_ type of ARPHRD_ATM, so
2698 * you can't use the ARPHRD_ type to find out whether
2699 * captured packets will have an LLC header, and,
2700 * while there's a socket ioctl to *set* the encapsulation
2701 * type, there's no ioctl to *get* the encapsulation type.
2705 * programs that dissect Linux Classical IP frames
2706 * would have to check for an LLC header and,
2707 * depending on whether they see one or not, dissect
2708 * the frame as LLC-encapsulated or as raw IP (I
2709 * don't know whether there's any traffic other than
2710 * IP that would show up on the socket, or whether
2711 * there's any support for IPv6 in the Linux
2712 * Classical IP code);
2714 * filter expressions would have to compile into
2715 * code that checks for an LLC header and does
2718 * Both of those are a nuisance - and, at least on systems
2719 * that support PF_PACKET sockets, we don't have to put
2720 * up with those nuisances; instead, we can just capture
2721 * in cooked mode. That's what we'll do, if we can.
2722 * Otherwise, we'll just fail.
2725 handle->linktype = DLT_LINUX_SLL;
2727 handle->linktype = -1;
2730 #ifndef ARPHRD_IEEE80211 /* From Linux 2.4.6 */
2731 #define ARPHRD_IEEE80211 801
2733 case ARPHRD_IEEE80211:
2734 handle->linktype = DLT_IEEE802_11;
2737 #ifndef ARPHRD_IEEE80211_PRISM /* From Linux 2.4.18 */
2738 #define ARPHRD_IEEE80211_PRISM 802
2740 case ARPHRD_IEEE80211_PRISM:
2741 handle->linktype = DLT_PRISM_HEADER;
2744 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
2745 #define ARPHRD_IEEE80211_RADIOTAP 803
2747 case ARPHRD_IEEE80211_RADIOTAP:
2748 handle->linktype = DLT_IEEE802_11_RADIO;
2753 * Some PPP code in the kernel supplies no link-layer
2754 * header whatsoever to PF_PACKET sockets; other PPP
2755 * code supplies PPP link-layer headers ("syncppp.c");
2756 * some PPP code might supply random link-layer
2757 * headers (PPP over ISDN - there's code in Ethereal,
2758 * for example, to cope with PPP-over-ISDN captures
2759 * with which the Ethereal developers have had to cope,
2760 * heuristically trying to determine which of the
2761 * oddball link-layer headers particular packets have).
2763 * As such, we just punt, and run all PPP interfaces
2764 * in cooked mode, if we can; otherwise, we just treat
2765 * it as DLT_RAW, for now - if somebody needs to capture,
2766 * on a 2.0[.x] kernel, on PPP devices that supply a
2767 * link-layer header, they'll have to add code here to
2768 * map to the appropriate DLT_ type (possibly adding a
2769 * new DLT_ type, if necessary).
2772 handle->linktype = DLT_LINUX_SLL;
2775 * XXX - handle ISDN types here? We can't fall
2776 * back on cooked sockets, so we'd have to
2777 * figure out from the device name what type of
2778 * link-layer encapsulation it's using, and map
2779 * that to an appropriate DLT_ value, meaning
2780 * we'd map "isdnN" devices to DLT_RAW (they
2781 * supply raw IP packets with no link-layer
2782 * header) and "isdY" devices to a new DLT_I4L_IP
2783 * type that has only an Ethernet packet type as
2784 * a link-layer header.
2786 * But sometimes we seem to get random crap
2787 * in the link-layer header when capturing on
2790 handle->linktype = DLT_RAW;
2794 #ifndef ARPHRD_CISCO
2795 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
2798 handle->linktype = DLT_C_HDLC;
2801 /* Not sure if this is correct for all tunnels, but it
2805 #define ARPHRD_SIT 776 /* From Linux 2.2.13 */
2813 #ifndef ARPHRD_RAWHDLC
2814 #define ARPHRD_RAWHDLC 518
2816 case ARPHRD_RAWHDLC:
2818 #define ARPHRD_DLCI 15
2822 * XXX - should some of those be mapped to DLT_LINUX_SLL
2823 * instead? Should we just map all of them to DLT_LINUX_SLL?
2825 handle->linktype = DLT_RAW;
2829 #define ARPHRD_FRAD 770
2832 handle->linktype = DLT_FRELAY;
2835 case ARPHRD_LOCALTLK:
2836 handle->linktype = DLT_LTALK;
2841 * RFC 4338 defines an encapsulation for IP and ARP
2842 * packets that's compatible with the RFC 2625
2843 * encapsulation, but that uses a different ARP
2844 * hardware type and hardware addresses. That
2845 * ARP hardware type is 18; Linux doesn't define
2846 * any ARPHRD_ value as 18, but if it ever officially
2847 * supports RFC 4338-style IP-over-FC, it should define
2850 * For now, we map it to DLT_IP_OVER_FC, in the hopes
2851 * that this will encourage its use in the future,
2852 * should Linux ever officially support RFC 4338-style
2855 handle->linktype = DLT_IP_OVER_FC;
2859 #define ARPHRD_FCPP 784
2863 #define ARPHRD_FCAL 785
2867 #define ARPHRD_FCPL 786
2870 #ifndef ARPHRD_FCFABRIC
2871 #define ARPHRD_FCFABRIC 787
2873 case ARPHRD_FCFABRIC:
2875 * Back in 2002, Donald Lee at Cray wanted a DLT_ for
2878 * http://www.mail-archive.com/tcpdump-workers@sandelman.ottawa.on.ca/msg01043.html
2880 * and one was assigned.
2882 * In a later private discussion (spun off from a message
2883 * on the ethereal-users list) on how to get that DLT_
2884 * value in libpcap on Linux, I ended up deciding that
2885 * the best thing to do would be to have him tweak the
2886 * driver to set the ARPHRD_ value to some ARPHRD_FCxx
2887 * type, and map all those types to DLT_IP_OVER_FC:
2889 * I've checked into the libpcap and tcpdump CVS tree
2890 * support for DLT_IP_OVER_FC. In order to use that,
2891 * you'd have to modify your modified driver to return
2892 * one of the ARPHRD_FCxxx types, in "fcLINUXfcp.c" -
2893 * change it to set "dev->type" to ARPHRD_FCFABRIC, for
2894 * example (the exact value doesn't matter, it can be
2895 * any of ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, or
2898 * 11 years later, Christian Svensson wanted to map
2899 * various ARPHRD_ values to DLT_FC_2 and
2900 * DLT_FC_2_WITH_FRAME_DELIMS for raw Fibre Channel
2903 * https://github.com/mcr/libpcap/pull/29
2905 * There doesn't seem to be any network drivers that uses
2906 * any of the ARPHRD_FC* values for IP-over-FC, and
2907 * it's not exactly clear what the "Dummy types for non
2908 * ARP hardware" are supposed to mean (link-layer
2909 * header type? Physical network type?), so it's
2910 * not exactly clear why the ARPHRD_FC* types exist
2911 * in the first place.
2913 * For now, we map them to DLT_FC_2, and provide an
2914 * option of DLT_FC_2_WITH_FRAME_DELIMS, as well as
2915 * DLT_IP_OVER_FC just in case there's some old
2916 * driver out there that uses one of those types for
2917 * IP-over-FC on which somebody wants to capture
2920 handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2922 * If that fails, just leave the list empty.
2924 if (handle->dlt_list != NULL) {
2925 handle->dlt_list[0] = DLT_FC_2;
2926 handle->dlt_list[1] = DLT_FC_2_WITH_FRAME_DELIMS;
2927 handle->dlt_list[2] = DLT_IP_OVER_FC;
2928 handle->dlt_count = 3;
2930 handle->linktype = DLT_FC_2;
2934 #define ARPHRD_IRDA 783
2937 /* Don't expect IP packet out of this interfaces... */
2938 handle->linktype = DLT_LINUX_IRDA;
2939 /* We need to save packet direction for IrDA decoding,
2940 * so let's use "Linux-cooked" mode. Jean II
2942 * XXX - this is handled in activate_new(). */
2943 //handlep->cooked = 1;
2946 /* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
2947 * is needed, please report it to <daniele@orlandi.com> */
2949 #define ARPHRD_LAPD 8445
2952 /* Don't expect IP packet out of this interfaces... */
2953 handle->linktype = DLT_LINUX_LAPD;
2957 #define ARPHRD_NONE 0xFFFE
2961 * No link-layer header; packets are just IP
2962 * packets, so use DLT_RAW.
2964 handle->linktype = DLT_RAW;
2967 #ifndef ARPHRD_IEEE802154
2968 #define ARPHRD_IEEE802154 804
2970 case ARPHRD_IEEE802154:
2971 handle->linktype = DLT_IEEE802_15_4_NOFCS;
2974 #ifndef ARPHRD_NETLINK
2975 #define ARPHRD_NETLINK 824
2977 case ARPHRD_NETLINK:
2978 handle->linktype = DLT_NETLINK;
2980 * We need to use cooked mode, so that in sll_protocol we
2981 * pick up the netlink protocol type such as NETLINK_ROUTE,
2982 * NETLINK_GENERIC, NETLINK_FIB_LOOKUP, etc.
2984 * XXX - this is handled in activate_new().
2986 //handlep->cooked = 1;
2990 handle->linktype = -1;
2995 /* ===== Functions to interface to the newer kernels ================== */
2998 * Try to open a packet socket using the new kernel PF_PACKET interface.
2999 * Returns 1 on success, 0 on an error that means the new interface isn't
3000 * present (so the old SOCK_PACKET interface should be tried), and a
3001 * PCAP_ERROR_ value on an error that means that the old mechanism won't
3002 * work either (so it shouldn't be tried).
3005 activate_new(pcap_t *handle)
3007 #ifdef HAVE_PF_PACKET_SOCKETS
3008 struct pcap_linux *handlep = handle->priv;
3009 const char *device = handle->opt.source;
3010 int is_any_device = (strcmp(device, "any") == 0);
3011 int sock_fd = -1, arptype;
3012 #ifdef HAVE_PACKET_AUXDATA
3016 struct packet_mreq mr;
3019 * Open a socket with protocol family packet. If the
3020 * "any" device was specified, we open a SOCK_DGRAM
3021 * socket for the cooked interface, otherwise we first
3022 * try a SOCK_RAW socket for the raw interface.
3024 sock_fd = is_any_device ?
3025 socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) :
3026 socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
3028 if (sock_fd == -1) {
3029 if (errno == EINVAL || errno == EAFNOSUPPORT) {
3031 * We don't support PF_PACKET/SOCK_whatever
3032 * sockets; try the old mechanism.
3037 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
3038 pcap_strerror(errno) );
3039 if (errno == EPERM || errno == EACCES) {
3041 * You don't have permission to open the
3044 return PCAP_ERROR_PERM_DENIED;
3053 /* It seems the kernel supports the new interface. */
3054 handlep->sock_packet = 0;
3057 * Get the interface index of the loopback device.
3058 * If the attempt fails, don't fail, just set the
3059 * "handlep->lo_ifindex" to -1.
3061 * XXX - can there be more than one device that loops
3062 * packets back, i.e. devices other than "lo"? If so,
3063 * we'd need to find them all, and have an array of
3064 * indices for them, and check all of them in
3065 * "pcap_read_packet()".
3067 handlep->lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
3070 * Default value for offset to align link-layer payload
3071 * on a 4-byte boundary.
3076 * What kind of frames do we have to deal with? Fall back
3077 * to cooked mode if we have an unknown interface type
3078 * or a type we know doesn't work well in raw mode.
3080 if (!is_any_device) {
3081 /* Assume for now we don't need cooked mode. */
3082 handlep->cooked = 0;
3084 if (handle->opt.rfmon) {
3086 * We were asked to turn on monitor mode.
3087 * Do so before we get the link-layer type,
3088 * because entering monitor mode could change
3089 * the link-layer type.
3091 err = enter_rfmon_mode(handle, sock_fd, device);
3099 * Nothing worked for turning monitor mode
3103 return PCAP_ERROR_RFMON_NOTSUP;
3107 * Either monitor mode has been turned on for
3108 * the device, or we've been given a different
3109 * device to open for monitor mode. If we've
3110 * been given a different device, use it.
3112 if (handlep->mondevice != NULL)
3113 device = handlep->mondevice;
3115 arptype = iface_get_arptype(sock_fd, device, handle->errbuf);
3120 map_arphrd_to_dlt(handle, arptype, device, 1);
3121 if (handle->linktype == -1 ||
3122 handle->linktype == DLT_LINUX_SLL ||
3123 handle->linktype == DLT_LINUX_IRDA ||
3124 handle->linktype == DLT_LINUX_LAPD ||
3125 handle->linktype == DLT_NETLINK ||
3126 (handle->linktype == DLT_EN10MB &&
3127 (strncmp("isdn", device, 4) == 0 ||
3128 strncmp("isdY", device, 4) == 0))) {
3130 * Unknown interface type (-1), or a
3131 * device we explicitly chose to run
3132 * in cooked mode (e.g., PPP devices),
3133 * or an ISDN device (whose link-layer
3134 * type we can only determine by using
3135 * APIs that may be different on different
3136 * kernels) - reopen in cooked mode.
3138 if (close(sock_fd) == -1) {
3139 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3140 "close: %s", pcap_strerror(errno));
3143 sock_fd = socket(PF_PACKET, SOCK_DGRAM,
3145 if (sock_fd == -1) {
3146 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3147 "socket: %s", pcap_strerror(errno));
3148 if (errno == EPERM || errno == EACCES) {
3150 * You don't have permission to
3153 return PCAP_ERROR_PERM_DENIED;
3161 handlep->cooked = 1;
3164 * Get rid of any link-layer type list
3165 * we allocated - this only supports cooked
3168 if (handle->dlt_list != NULL) {
3169 free(handle->dlt_list);
3170 handle->dlt_list = NULL;
3171 handle->dlt_count = 0;
3174 if (handle->linktype == -1) {
3176 * Warn that we're falling back on
3177 * cooked mode; we may want to
3178 * update "map_arphrd_to_dlt()"
3179 * to handle the new type.
3181 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3183 "supported by libpcap - "
3184 "falling back to cooked "
3190 * IrDA capture is not a real "cooked" capture,
3191 * it's IrLAP frames, not IP packets. The
3192 * same applies to LAPD capture.
3194 if (handle->linktype != DLT_LINUX_IRDA &&
3195 handle->linktype != DLT_LINUX_LAPD &&
3196 handle->linktype != DLT_NETLINK)
3197 handle->linktype = DLT_LINUX_SLL;
3200 handlep->ifindex = iface_get_id(sock_fd, device,
3202 if (handlep->ifindex == -1) {
3207 if ((err = iface_bind(sock_fd, handlep->ifindex,
3208 handle->errbuf)) != 1) {
3213 return 0; /* try old mechanism */
3219 if (handle->opt.rfmon) {
3221 * It doesn't support monitor mode.
3224 return PCAP_ERROR_RFMON_NOTSUP;
3228 * It uses cooked mode.
3230 handlep->cooked = 1;
3231 handle->linktype = DLT_LINUX_SLL;
3234 * We're not bound to a device.
3235 * For now, we're using this as an indication
3236 * that we can't transmit; stop doing that only
3237 * if we figure out how to transmit in cooked
3240 handlep->ifindex = -1;
3244 * Select promiscuous mode on if "promisc" is set.
3246 * Do not turn allmulti mode on if we don't select
3247 * promiscuous mode - on some devices (e.g., Orinoco
3248 * wireless interfaces), allmulti mode isn't supported
3249 * and the driver implements it by turning promiscuous
3250 * mode on, and that screws up the operation of the
3251 * card as a normal networking interface, and on no
3252 * other platform I know of does starting a non-
3253 * promiscuous capture affect which multicast packets
3254 * are received by the interface.
3258 * Hmm, how can we set promiscuous mode on all interfaces?
3259 * I am not sure if that is possible at all. For now, we
3260 * silently ignore attempts to turn promiscuous mode on
3261 * for the "any" device (so you don't have to explicitly
3262 * disable it in programs such as tcpdump).
3265 if (!is_any_device && handle->opt.promisc) {
3266 memset(&mr, 0, sizeof(mr));
3267 mr.mr_ifindex = handlep->ifindex;
3268 mr.mr_type = PACKET_MR_PROMISC;
3269 if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
3270 &mr, sizeof(mr)) == -1) {
3271 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3272 "setsockopt: %s", pcap_strerror(errno));
3278 /* Enable auxillary data if supported and reserve room for
3279 * reconstructing VLAN headers. */
3280 #ifdef HAVE_PACKET_AUXDATA
3282 if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
3283 sizeof(val)) == -1 && errno != ENOPROTOOPT) {
3284 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3285 "setsockopt: %s", pcap_strerror(errno));
3289 handle->offset += VLAN_TAG_LEN;
3290 #endif /* HAVE_PACKET_AUXDATA */
3293 * This is a 2.2[.x] or later kernel (we know that
3294 * because we're not using a SOCK_PACKET socket -
3295 * PF_PACKET is supported only in 2.2 and later
3298 * We can safely pass "recvfrom()" a byte count
3299 * based on the snapshot length.
3301 * If we're in cooked mode, make the snapshot length
3302 * large enough to hold a "cooked mode" header plus
3303 * 1 byte of packet data (so we don't pass a byte
3304 * count of 0 to "recvfrom()").
3306 if (handlep->cooked) {
3307 if (handle->snapshot < SLL_HDR_LEN + 1)
3308 handle->snapshot = SLL_HDR_LEN + 1;
3310 handle->bufsize = handle->snapshot;
3313 * Set the offset at which to insert VLAN tags.
3315 switch (handle->linktype) {
3318 handlep->vlan_offset = 2 * ETH_ALEN;
3322 handlep->vlan_offset = 14;
3326 handlep->vlan_offset = -1; /* unknown */
3330 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS)
3331 if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
3332 int nsec_tstamps = 1;
3334 if (setsockopt(sock_fd, SOL_SOCKET, SO_TIMESTAMPNS, &nsec_tstamps, sizeof(nsec_tstamps)) < 0) {
3335 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "setsockopt: unable to set SO_TIMESTAMPNS");
3340 #endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */
3343 * We've succeeded. Save the socket FD in the pcap structure.
3345 handle->fd = sock_fd;
3348 #else /* HAVE_PF_PACKET_SOCKETS */
3350 "New packet capturing interface not supported by build "
3351 "environment", PCAP_ERRBUF_SIZE);
3353 #endif /* HAVE_PF_PACKET_SOCKETS */
3356 #ifdef HAVE_PACKET_RING
3358 * Attempt to activate with memory-mapped access.
3360 * On success, returns 1, and sets *status to 0 if there are no warnings
3361 * or to a PCAP_WARNING_ code if there is a warning.
3363 * On failure due to lack of support for memory-mapped capture, returns
3366 * On error, returns -1, and sets *status to the appropriate error code;
3367 * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3370 activate_mmap(pcap_t *handle, int *status)
3372 struct pcap_linux *handlep = handle->priv;
3376 * Attempt to allocate a buffer to hold the contents of one
3377 * packet, for use by the oneshot callback.
3379 handlep->oneshot_buffer = malloc(handle->snapshot);
3380 if (handlep->oneshot_buffer == NULL) {
3381 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3382 "can't allocate oneshot buffer: %s",
3383 pcap_strerror(errno));
3384 *status = PCAP_ERROR;
3388 if (handle->opt.buffer_size == 0) {
3389 /* by default request 2M for the ring buffer */
3390 handle->opt.buffer_size = 2*1024*1024;
3392 ret = prepare_tpacket_socket(handle);
3394 free(handlep->oneshot_buffer);
3395 *status = PCAP_ERROR;
3398 ret = create_ring(handle, status);
3401 * We don't support memory-mapped capture; our caller
3402 * will fall back on reading from the socket.
3404 free(handlep->oneshot_buffer);
3409 * Error attempting to enable memory-mapped capture;
3410 * fail. create_ring() has set *status.
3412 free(handlep->oneshot_buffer);
3417 * Success. *status has been set either to 0 if there are no
3418 * warnings or to a PCAP_WARNING_ value if there is a warning.
3420 * Override some defaults and inherit the other fields from
3422 * handle->offset is used to get the current position into the rx ring.
3423 * handle->cc is used to store the ring size.
3426 switch (handlep->tp_version) {
3428 handle->read_op = pcap_read_linux_mmap_v1;
3430 #ifdef HAVE_TPACKET2
3432 handle->read_op = pcap_read_linux_mmap_v2;
3435 #ifdef HAVE_TPACKET3
3437 handle->read_op = pcap_read_linux_mmap_v3;
3441 handle->cleanup_op = pcap_cleanup_linux_mmap;
3442 handle->setfilter_op = pcap_setfilter_linux_mmap;
3443 handle->setnonblock_op = pcap_setnonblock_mmap;
3444 handle->getnonblock_op = pcap_getnonblock_mmap;
3445 handle->oneshot_callback = pcap_oneshot_mmap;
3446 handle->selectable_fd = handle->fd;
3449 #else /* HAVE_PACKET_RING */
3451 activate_mmap(pcap_t *handle _U_, int *status _U_)
3455 #endif /* HAVE_PACKET_RING */
3457 #ifdef HAVE_PACKET_RING
3459 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
3461 * Attempt to set the socket to the specified version of the memory-mapped
3464 * Return 0 if we succeed; return 1 if we fail because that version isn't
3465 * supported; return -1 on any other error, and set handle->errbuf.
3468 init_tpacket(pcap_t *handle, int version, const char *version_str)
3470 struct pcap_linux *handlep = handle->priv;
3472 socklen_t len = sizeof(val);
3474 /* Probe whether kernel supports the specified TPACKET version */
3475 if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
3476 if (errno == ENOPROTOOPT || errno == EINVAL)
3479 /* Failed to even find out; this is a fatal error. */
3480 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3481 "can't get %s header len on packet socket: %s",
3483 pcap_strerror(errno));
3486 handlep->tp_hdrlen = val;
3489 if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
3491 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3492 "can't activate %s on packet socket: %s",
3494 pcap_strerror(errno));
3497 handlep->tp_version = version;
3499 /* Reserve space for VLAN tag reconstruction */
3501 if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val,
3503 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3504 "can't set up reserve on packet socket: %s",
3505 pcap_strerror(errno));
3511 #endif /* defined HAVE_TPACKET2 || defined HAVE_TPACKET3 */
3514 * Attempt to set the socket to version 3 of the memory-mapped header and,
3515 * if that fails because version 3 isn't supported, attempt to fall
3516 * back to version 2. If version 2 isn't supported, just leave it at
3519 * Return 1 if we succeed or if we fail because neither version 2 nor 3 is
3520 * supported; return -1 on any other error, and set handle->errbuf.
3523 prepare_tpacket_socket(pcap_t *handle)
3525 struct pcap_linux *handlep = handle->priv;
3526 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
3530 handlep->tp_version = TPACKET_V1;
3531 handlep->tp_hdrlen = sizeof(struct tpacket_hdr);
3533 #ifdef HAVE_TPACKET3
3535 * The only mode in which buffering is done on PF_PACKET
3536 * sockets, so that packets might not be delivered
3537 * immediately, is TPACKET_V3 mode.
3539 * The buffering cannot be disabled in that mode, so
3540 * if the user has requested immediate mode, we don't
3543 if (handle->opt.immediate)
3544 ret = 1; /* pretend TPACKET_V3 couldn't be set */
3546 ret = init_tpacket(handle, TPACKET_V3, "TPACKET_V3");
3548 /* Error during setting up TPACKET_V3. */
3550 } else if (1 == ret) {
3551 /* TPACKET_V3 not supported - fall back to TPACKET_V2. */
3552 #endif /* HAVE_TPACKET3 */
3554 #ifdef HAVE_TPACKET2
3555 ret = init_tpacket(handle, TPACKET_V2, "TPACKET_V2");
3557 /* Error during setting up TPACKET_V2. */
3560 #endif /* HAVE_TPACKET2 */
3562 #ifdef HAVE_TPACKET3
3564 #endif /* HAVE_TPACKET3 */
3570 * Attempt to set up memory-mapped access.
3572 * On success, returns 1, and sets *status to 0 if there are no warnings
3573 * or to a PCAP_WARNING_ code if there is a warning.
3575 * On failure due to lack of support for memory-mapped capture, returns
3578 * On error, returns -1, and sets *status to the appropriate error code;
3579 * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
3582 create_ring(pcap_t *handle, int *status)
3584 struct pcap_linux *handlep = handle->priv;
3585 unsigned i, j, frames_per_block;
3586 #ifdef HAVE_TPACKET3
3588 * For sockets using TPACKET_V1 or TPACKET_V2, the extra
3589 * stuff at the end of a struct tpacket_req3 will be
3590 * ignored, so this is OK even for those sockets.
3592 struct tpacket_req3 req;
3594 struct tpacket_req req;
3597 unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
3598 unsigned int frame_size;
3601 * Start out assuming no warnings or errors.
3605 switch (handlep->tp_version) {
3608 #ifdef HAVE_TPACKET2
3611 /* Note that with large snapshot length (say 64K, which is
3612 * the default for recent versions of tcpdump, the value that
3613 * "-s 0" has given for a long time with tcpdump, and the
3614 * default in Wireshark/TShark/dumpcap), if we use the snapshot
3615 * length to calculate the frame length, only a few frames
3616 * will be available in the ring even with pretty
3617 * large ring size (and a lot of memory will be unused).
3619 * Ideally, we should choose a frame length based on the
3620 * minimum of the specified snapshot length and the maximum
3621 * packet size. That's not as easy as it sounds; consider,
3622 * for example, an 802.11 interface in monitor mode, where
3623 * the frame would include a radiotap header, where the
3624 * maximum radiotap header length is device-dependent.
3626 * So, for now, we just do this for Ethernet devices, where
3627 * there's no metadata header, and the link-layer header is
3628 * fixed length. We can get the maximum packet size by
3629 * adding 18, the Ethernet header length plus the CRC length
3630 * (just in case we happen to get the CRC in the packet), to
3631 * the MTU of the interface; we fetch the MTU in the hopes
3632 * that it reflects support for jumbo frames. (Even if the
3633 * interface is just being used for passive snooping, the
3634 * driver might set the size of buffers in the receive ring
3635 * based on the MTU, so that the MTU limits the maximum size
3636 * of packets that we can receive.)
3638 * We don't do that if segmentation/fragmentation or receive
3639 * offload are enabled, so we don't get rudely surprised by
3640 * "packets" bigger than the MTU. */
3641 frame_size = handle->snapshot;
3642 if (handle->linktype == DLT_EN10MB) {
3646 offload = iface_get_offload(handle);
3647 if (offload == -1) {
3648 *status = PCAP_ERROR;
3652 mtu = iface_get_mtu(handle->fd, handle->opt.source,
3655 *status = PCAP_ERROR;
3658 if (frame_size > mtu + 18)
3659 frame_size = mtu + 18;
3663 /* NOTE: calculus matching those in tpacket_rcv()
3664 * in linux-2.6/net/packet/af_packet.c
3666 len = sizeof(sk_type);
3667 if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type,
3669 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3670 "getsockopt: %s", pcap_strerror(errno));
3671 *status = PCAP_ERROR;
3674 #ifdef PACKET_RESERVE
3675 len = sizeof(tp_reserve);
3676 if (getsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE,
3677 &tp_reserve, &len) < 0) {
3678 if (errno != ENOPROTOOPT) {
3680 * ENOPROTOOPT means "kernel doesn't support
3681 * PACKET_RESERVE", in which case we fall back
3684 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3685 "getsockopt: %s", pcap_strerror(errno));
3686 *status = PCAP_ERROR;
3689 tp_reserve = 0; /* older kernel, reserve not supported */
3692 tp_reserve = 0; /* older kernel, reserve not supported */
3694 maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
3695 /* XXX: in the kernel maclen is calculated from
3696 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
3697 * in: packet_snd() in linux-2.6/net/packet/af_packet.c
3698 * then packet_alloc_skb() in linux-2.6/net/packet/af_packet.c
3699 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
3700 * but I see no way to get those sizes in userspace,
3701 * like for instance with an ifreq ioctl();
3702 * the best thing I've found so far is MAX_HEADER in
3703 * the kernel part of linux-2.6/include/linux/netdevice.h
3704 * which goes up to 128+48=176; since pcap-linux.c
3705 * defines a MAX_LINKHEADER_SIZE of 256 which is
3706 * greater than that, let's use it.. maybe is it even
3707 * large enough to directly replace macoff..
3709 tp_hdrlen = TPACKET_ALIGN(handlep->tp_hdrlen) + sizeof(struct sockaddr_ll) ;
3710 netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
3711 /* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN
3712 * of netoff, which contradicts
3713 * linux-2.6/Documentation/networking/packet_mmap.txt
3715 * "- Gap, chosen so that packet data (Start+tp_net)
3716 * aligns to TPACKET_ALIGNMENT=16"
3718 /* NOTE: in linux-2.6/include/linux/skbuff.h:
3719 * "CPUs often take a performance hit
3720 * when accessing unaligned memory locations"
3722 macoff = netoff - maclen;
3723 req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
3724 req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
3727 #ifdef HAVE_TPACKET3
3729 /* The "frames" for this are actually buffers that
3730 * contain multiple variable-sized frames.
3732 * We pick a "frame" size of 128K to leave enough
3733 * room for at least one reasonably-sized packet
3734 * in the "frame". */
3735 req.tp_frame_size = MAXIMUM_SNAPLEN;
3736 req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
3741 /* compute the minumum block size that will handle this frame.
3742 * The block has to be page size aligned.
3743 * The max block size allowed by the kernel is arch-dependent and
3744 * it's not explicitly checked here. */
3745 req.tp_block_size = getpagesize();
3746 while (req.tp_block_size < req.tp_frame_size)
3747 req.tp_block_size <<= 1;
3749 frames_per_block = req.tp_block_size/req.tp_frame_size;
3752 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
3753 * so we check for PACKET_TIMESTAMP. We check for
3754 * linux/net_tstamp.h just in case a system somehow has
3755 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
3758 * SIOCSHWTSTAMP was introduced in the patch that introduced
3759 * linux/net_tstamp.h, so we don't bother checking whether
3760 * SIOCSHWTSTAMP is defined (if your Linux system has
3761 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
3762 * Linux system is badly broken).
3764 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
3766 * If we were told to do so, ask the kernel and the driver
3767 * to use hardware timestamps.
3769 * Hardware timestamps are only supported with mmapped
3772 if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
3773 handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
3774 struct hwtstamp_config hwconfig;
3779 * Ask for hardware time stamps on all packets,
3780 * including transmitted packets.
3782 memset(&hwconfig, 0, sizeof(hwconfig));
3783 hwconfig.tx_type = HWTSTAMP_TX_ON;
3784 hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
3786 memset(&ifr, 0, sizeof(ifr));
3787 strlcpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name));
3788 ifr.ifr_data = (void *)&hwconfig;
3790 if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
3795 * Treat this as an error, as the
3796 * user should try to run this
3797 * with the appropriate privileges -
3798 * and, if they can't, shouldn't
3799 * try requesting hardware time stamps.
3801 *status = PCAP_ERROR_PERM_DENIED;
3806 * Treat this as a warning, as the
3807 * only way to fix the warning is to
3808 * get an adapter that supports hardware
3809 * time stamps. We'll just fall back
3810 * on the standard host time stamps.
3812 *status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
3816 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3817 "SIOCSHWTSTAMP failed: %s",
3818 pcap_strerror(errno));
3819 *status = PCAP_ERROR;
3824 * Well, that worked. Now specify the type of
3825 * hardware time stamp we want for this
3828 if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
3830 * Hardware timestamp, synchronized
3831 * with the system clock.
3833 timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
3836 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
3837 * timestamp, not synchronized with the
3840 timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
3842 if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
3843 (void *)×ource, sizeof(timesource))) {
3844 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3845 "can't set PACKET_TIMESTAMP: %s",
3846 pcap_strerror(errno));
3847 *status = PCAP_ERROR;
3852 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
3854 /* ask the kernel to create the ring */
3856 req.tp_block_nr = req.tp_frame_nr / frames_per_block;
3858 /* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
3859 req.tp_frame_nr = req.tp_block_nr * frames_per_block;
3861 #ifdef HAVE_TPACKET3
3862 /* timeout value to retire block - use the configured buffering timeout, or default if <0. */
3863 req.tp_retire_blk_tov = (handlep->timeout>=0)?handlep->timeout:0;
3864 /* private data not used */
3865 req.tp_sizeof_priv = 0;
3866 /* Rx ring - feature request bits - none (rxhash will not be filled) */
3867 req.tp_feature_req_word = 0;
3870 if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3871 (void *) &req, sizeof(req))) {
3872 if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
3874 * Memory failure; try to reduce the requested ring
3877 * We used to reduce this by half -- do 5% instead.
3878 * That may result in more iterations and a longer
3879 * startup, but the user will be much happier with
3880 * the resulting buffer size.
3882 if (req.tp_frame_nr < 20)
3883 req.tp_frame_nr -= 1;
3885 req.tp_frame_nr -= req.tp_frame_nr/20;
3888 if (errno == ENOPROTOOPT) {
3890 * We don't have ring buffer support in this kernel.
3894 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3895 "can't create rx ring on packet socket: %s",
3896 pcap_strerror(errno));
3897 *status = PCAP_ERROR;
3901 /* memory map the rx ring */
3902 handlep->mmapbuflen = req.tp_block_nr * req.tp_block_size;
3903 handlep->mmapbuf = mmap(0, handlep->mmapbuflen,
3904 PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
3905 if (handlep->mmapbuf == MAP_FAILED) {
3906 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3907 "can't mmap rx ring: %s", pcap_strerror(errno));
3909 /* clear the allocated ring on error*/
3910 destroy_ring(handle);
3911 *status = PCAP_ERROR;
3915 /* allocate a ring for each frame header pointer*/
3916 handle->cc = req.tp_frame_nr;
3917 handle->buffer = malloc(handle->cc * sizeof(union thdr *));
3918 if (!handle->buffer) {
3919 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3920 "can't allocate ring of frame headers: %s",
3921 pcap_strerror(errno));
3923 destroy_ring(handle);
3924 *status = PCAP_ERROR;
3928 /* fill the header ring with proper frame ptr*/
3930 for (i=0; i<req.tp_block_nr; ++i) {
3931 void *base = &handlep->mmapbuf[i*req.tp_block_size];
3932 for (j=0; j<frames_per_block; ++j, ++handle->offset) {
3933 RING_GET_FRAME(handle) = base;
3934 base += req.tp_frame_size;
3938 handle->bufsize = req.tp_frame_size;
3943 /* free all ring related resources*/
3945 destroy_ring(pcap_t *handle)
3947 struct pcap_linux *handlep = handle->priv;
3949 /* tell the kernel to destroy the ring*/
3950 struct tpacket_req req;
3951 memset(&req, 0, sizeof(req));
3952 setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3953 (void *) &req, sizeof(req));
3955 /* if ring is mapped, unmap it*/
3956 if (handlep->mmapbuf) {
3957 /* do not test for mmap failure, as we can't recover from any error */
3958 munmap(handlep->mmapbuf, handlep->mmapbuflen);
3959 handlep->mmapbuf = NULL;
3964 * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
3965 * for Linux mmapped capture.
3967 * The problem is that pcap_next() and pcap_next_ex() expect the packet
3968 * data handed to the callback to be valid after the callback returns,
3969 * but pcap_read_linux_mmap() has to release that packet as soon as
3970 * the callback returns (otherwise, the kernel thinks there's still
3971 * at least one unprocessed packet available in the ring, so a select()
3972 * will immediately return indicating that there's data to process), so,
3973 * in the callback, we have to make a copy of the packet.
3975 * Yes, this means that, if the capture is using the ring buffer, using
3976 * pcap_next() or pcap_next_ex() requires more copies than using
3977 * pcap_loop() or pcap_dispatch(). If that bothers you, don't use
3978 * pcap_next() or pcap_next_ex().
3981 pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
3982 const u_char *bytes)
3984 struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
3985 pcap_t *handle = sp->pd;
3986 struct pcap_linux *handlep = handle->priv;
3989 memcpy(handlep->oneshot_buffer, bytes, h->caplen);
3990 *sp->pkt = handlep->oneshot_buffer;
3994 pcap_cleanup_linux_mmap( pcap_t *handle )
3996 struct pcap_linux *handlep = handle->priv;
3998 destroy_ring(handle);
3999 if (handlep->oneshot_buffer != NULL) {
4000 free(handlep->oneshot_buffer);
4001 handlep->oneshot_buffer = NULL;
4003 pcap_cleanup_linux(handle);
4008 pcap_getnonblock_mmap(pcap_t *p, char *errbuf)
4010 struct pcap_linux *handlep = p->priv;
4012 /* use negative value of timeout to indicate non blocking ops */
4013 return (handlep->timeout<0);
4017 pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
4019 struct pcap_linux *handlep = p->priv;
4022 * Set the file descriptor to non-blocking mode, as we use
4023 * it for sending packets.
4025 if (pcap_setnonblock_fd(p, nonblock, errbuf) == -1)
4029 * Map each value to their corresponding negation to
4030 * preserve the timeout value provided with pcap_set_timeout.
4033 if (handlep->timeout >= 0) {
4035 * Indicate that we're switching to
4036 * non-blocking mode.
4038 handlep->timeout = ~handlep->timeout;
4041 if (handlep->timeout < 0) {
4042 handlep->timeout = ~handlep->timeout;
4048 static inline union thdr *
4049 pcap_get_ring_frame(pcap_t *handle, int status)
4051 struct pcap_linux *handlep = handle->priv;
4054 h.raw = RING_GET_FRAME(handle);
4055 switch (handlep->tp_version) {
4057 if (status != (h.h1->tp_status ? TP_STATUS_USER :
4061 #ifdef HAVE_TPACKET2
4063 if (status != (h.h2->tp_status ? TP_STATUS_USER :
4068 #ifdef HAVE_TPACKET3
4070 if (status != (h.h3->hdr.bh1.block_status ? TP_STATUS_USER :
4083 /* wait for frames availability.*/
4084 static int pcap_wait_for_frames_mmap(pcap_t *handle)
4086 if (!pcap_get_ring_frame(handle, TP_STATUS_USER)) {
4087 struct pcap_linux *handlep = handle->priv;
4090 struct pollfd pollinfo;
4093 pollinfo.fd = handle->fd;
4094 pollinfo.events = POLLIN;
4096 if (handlep->timeout == 0) {
4097 #ifdef HAVE_TPACKET3
4099 * XXX - due to a set of (mis)features in the
4100 * TPACKET_V3 kernel code, blocking forever with
4101 * a TPACKET_V3 socket can, if few packets
4102 * are arriving and passing the socket filter,
4103 * cause most packets to be dropped. See
4104 * libpcap issue #335 for the full painful
4105 * story. The workaround is to have poll()
4106 * time out very quickly, so we grab the
4107 * frames handed to us, and return them to
4110 * If those issues are ever fixed, we might
4111 * want to check the kernel version and block
4112 * forever with TPACKET_V3 if we're running
4113 * with a kernel that has the fix.
4115 if (handlep->tp_version == TPACKET_V3)
4116 timeout = 1; /* don't block for very long */
4119 timeout = -1; /* block forever */
4120 } else if (handlep->timeout > 0)
4121 timeout = handlep->timeout; /* block for that amount of time */
4123 timeout = 0; /* non-blocking mode - poll to pick up errors */
4125 ret = poll(&pollinfo, 1, timeout);
4126 if (ret < 0 && errno != EINTR) {
4127 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4128 "can't poll on packet socket: %s",
4129 pcap_strerror(errno));
4131 } else if (ret > 0 &&
4132 (pollinfo.revents & (POLLHUP|POLLRDHUP|POLLERR|POLLNVAL))) {
4134 * There's some indication other than
4135 * "you can read on this descriptor" on
4138 if (pollinfo.revents & (POLLHUP | POLLRDHUP)) {
4139 snprintf(handle->errbuf,
4141 "Hangup on packet socket");
4144 if (pollinfo.revents & POLLERR) {
4146 * A recv() will give us the
4147 * actual error code.
4149 * XXX - make the socket non-blocking?
4151 if (recv(handle->fd, &c, sizeof c,
4153 continue; /* what, no error? */
4154 if (errno == ENETDOWN) {
4156 * The device on which we're
4157 * capturing went away.
4159 * XXX - we should really return
4160 * PCAP_ERROR_IFACE_NOT_UP,
4161 * but pcap_dispatch() etc.
4162 * aren't defined to return
4165 snprintf(handle->errbuf,
4167 "The interface went down");
4169 snprintf(handle->errbuf,
4171 "Error condition on packet socket: %s",
4176 if (pollinfo.revents & POLLNVAL) {
4177 snprintf(handle->errbuf,
4179 "Invalid polling request on packet socket");
4183 /* check for break loop condition on interrupted syscall*/
4184 if (handle->break_loop) {
4185 handle->break_loop = 0;
4186 return PCAP_ERROR_BREAK;
4193 /* handle a single memory mapped packet */
4194 static int pcap_handle_packet_mmap(
4196 pcap_handler callback,
4198 unsigned char *frame,
4199 unsigned int tp_len,
4200 unsigned int tp_mac,
4201 unsigned int tp_snaplen,
4202 unsigned int tp_sec,
4203 unsigned int tp_usec,
4204 int tp_vlan_tci_valid,
4207 struct pcap_linux *handlep = handle->priv;
4209 struct sockaddr_ll *sll;
4210 struct pcap_pkthdr pcaphdr;
4212 /* perform sanity check on internal offset. */
4213 if (tp_mac + tp_snaplen > handle->bufsize) {
4214 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4215 "corrupted frame on kernel ring mac "
4216 "offset %d + caplen %d > frame len %d",
4217 tp_mac, tp_snaplen, handle->bufsize);
4221 /* run filter on received packet
4222 * If the kernel filtering is enabled we need to run the
4223 * filter until all the frames present into the ring
4224 * at filter creation time are processed.
4225 * In this case, blocks_to_filter_in_userland is used
4226 * as a counter for the packet we need to filter.
4227 * Note: alternatively it could be possible to stop applying
4228 * the filter when the ring became empty, but it can possibly
4229 * happen a lot later... */
4230 bp = frame + tp_mac;
4231 if (handlep->filter_in_userland && handle->fcode.bf_insns &&
4232 (bpf_filter(handle->fcode.bf_insns, bp,
4233 tp_len, tp_snaplen) == 0))
4236 sll = (void *)frame + TPACKET_ALIGN(handlep->tp_hdrlen);
4237 if (!linux_check_direction(handle, sll))
4240 /* get required packet info from ring header */
4241 pcaphdr.ts.tv_sec = tp_sec;
4242 pcaphdr.ts.tv_usec = tp_usec;
4243 pcaphdr.caplen = tp_snaplen;
4244 pcaphdr.len = tp_len;
4246 /* if required build in place the sll header*/
4247 if (handlep->cooked) {
4248 struct sll_header *hdrp;
4251 * The kernel should have left us with enough
4252 * space for an sll header; back up the packet
4253 * data pointer into that space, as that'll be
4254 * the beginning of the packet we pass to the
4260 * Let's make sure that's past the end of
4261 * the tpacket header, i.e. >=
4262 * ((u_char *)thdr + TPACKET_HDRLEN), so we
4263 * don't step on the header when we construct
4266 if (bp < (u_char *)frame +
4267 TPACKET_ALIGN(handlep->tp_hdrlen) +
4268 sizeof(struct sockaddr_ll)) {
4269 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4270 "cooked-mode frame doesn't have room for sll header");
4275 * OK, that worked; construct the sll header.
4277 hdrp = (struct sll_header *)bp;
4278 hdrp->sll_pkttype = map_packet_type_to_sll_type(
4280 hdrp->sll_hatype = htons(sll->sll_hatype);
4281 hdrp->sll_halen = htons(sll->sll_halen);
4282 memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
4283 hdrp->sll_protocol = sll->sll_protocol;
4285 /* update packet len */
4286 pcaphdr.caplen += SLL_HDR_LEN;
4287 pcaphdr.len += SLL_HDR_LEN;
4290 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3)
4291 if (tp_vlan_tci_valid &&
4292 handlep->vlan_offset != -1 &&
4293 tp_snaplen >= (unsigned int) handlep->vlan_offset)
4295 struct vlan_tag *tag;
4298 memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
4300 tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
4301 tag->vlan_tpid = htons(ETH_P_8021Q);
4302 tag->vlan_tci = htons(tp_vlan_tci);
4304 pcaphdr.caplen += VLAN_TAG_LEN;
4305 pcaphdr.len += VLAN_TAG_LEN;
4310 * The only way to tell the kernel to cut off the
4311 * packet at a snapshot length is with a filter program;
4312 * if there's no filter program, the kernel won't cut
4315 * Trim the snapshot length to be no longer than the
4316 * specified snapshot length.
4318 if (pcaphdr.caplen > handle->snapshot)
4319 pcaphdr.caplen = handle->snapshot;
4321 /* pass the packet to the user */
4322 callback(user, &pcaphdr, bp);
4328 pcap_read_linux_mmap_v1(pcap_t *handle, int max_packets, pcap_handler callback,
4331 struct pcap_linux *handlep = handle->priv;
4335 /* wait for frames availability.*/
4336 ret = pcap_wait_for_frames_mmap(handle);
4341 /* non-positive values of max_packets are used to require all
4342 * packets currently available in the ring */
4343 while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
4346 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
4350 ret = pcap_handle_packet_mmap(
4364 handlep->packets_read++;
4365 } else if (ret < 0) {
4370 * Hand this block back to the kernel, and, if we're
4371 * counting blocks that need to be filtered in userland
4372 * after having been filtered by the kernel, count
4373 * the one we've just processed.
4375 h.h1->tp_status = TP_STATUS_KERNEL;
4376 if (handlep->blocks_to_filter_in_userland > 0) {
4377 handlep->blocks_to_filter_in_userland--;
4378 if (handlep->blocks_to_filter_in_userland == 0) {
4380 * No more blocks need to be filtered
4383 handlep->filter_in_userland = 0;
4388 if (++handle->offset >= handle->cc)
4391 /* check for break loop condition*/
4392 if (handle->break_loop) {
4393 handle->break_loop = 0;
4394 return PCAP_ERROR_BREAK;
4400 #ifdef HAVE_TPACKET2
4402 pcap_read_linux_mmap_v2(pcap_t *handle, int max_packets, pcap_handler callback,
4405 struct pcap_linux *handlep = handle->priv;
4409 /* wait for frames availability.*/
4410 ret = pcap_wait_for_frames_mmap(handle);
4415 /* non-positive values of max_packets are used to require all
4416 * packets currently available in the ring */
4417 while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
4420 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
4424 ret = pcap_handle_packet_mmap(
4433 handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? h.h2->tp_nsec : h.h2->tp_nsec / 1000,
4434 #if defined(TP_STATUS_VLAN_VALID)
4435 (h.h2->tp_vlan_tci || (h.h2->tp_status & TP_STATUS_VLAN_VALID)),
4437 h.h2->tp_vlan_tci != 0,
4442 handlep->packets_read++;
4443 } else if (ret < 0) {
4448 * Hand this block back to the kernel, and, if we're
4449 * counting blocks that need to be filtered in userland
4450 * after having been filtered by the kernel, count
4451 * the one we've just processed.
4453 h.h2->tp_status = TP_STATUS_KERNEL;
4454 if (handlep->blocks_to_filter_in_userland > 0) {
4455 handlep->blocks_to_filter_in_userland--;
4456 if (handlep->blocks_to_filter_in_userland == 0) {
4458 * No more blocks need to be filtered
4461 handlep->filter_in_userland = 0;
4466 if (++handle->offset >= handle->cc)
4469 /* check for break loop condition*/
4470 if (handle->break_loop) {
4471 handle->break_loop = 0;
4472 return PCAP_ERROR_BREAK;
4477 #endif /* HAVE_TPACKET2 */
4479 #ifdef HAVE_TPACKET3
4481 pcap_read_linux_mmap_v3(pcap_t *handle, int max_packets, pcap_handler callback,
4484 struct pcap_linux *handlep = handle->priv;
4490 if (handlep->current_packet == NULL) {
4491 /* wait for frames availability.*/
4492 ret = pcap_wait_for_frames_mmap(handle);
4497 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
4499 if (pkts == 0 && handlep->timeout == 0) {
4500 /* Block until we see a packet. */
4506 /* non-positive values of max_packets are used to require all
4507 * packets currently available in the ring */
4508 while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) {
4509 if (handlep->current_packet == NULL) {
4510 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
4514 handlep->current_packet = h.raw + h.h3->hdr.bh1.offset_to_first_pkt;
4515 handlep->packets_left = h.h3->hdr.bh1.num_pkts;
4517 int packets_to_read = handlep->packets_left;
4519 if (!PACKET_COUNT_IS_UNLIMITED(max_packets) && packets_to_read > max_packets) {
4520 packets_to_read = max_packets;
4523 while(packets_to_read--) {
4524 struct tpacket3_hdr* tp3_hdr = (struct tpacket3_hdr*) handlep->current_packet;
4525 ret = pcap_handle_packet_mmap(
4529 handlep->current_packet,
4532 tp3_hdr->tp_snaplen,
4534 handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? tp3_hdr->tp_nsec : tp3_hdr->tp_nsec / 1000,
4535 #if defined(TP_STATUS_VLAN_VALID)
4536 (tp3_hdr->hv1.tp_vlan_tci || (tp3_hdr->tp_status & TP_STATUS_VLAN_VALID)),
4538 tp3_hdr->hv1.tp_vlan_tci != 0,
4540 tp3_hdr->hv1.tp_vlan_tci);
4543 handlep->packets_read++;
4544 } else if (ret < 0) {
4545 handlep->current_packet = NULL;
4548 handlep->current_packet += tp3_hdr->tp_next_offset;
4549 handlep->packets_left--;
4552 if (handlep->packets_left <= 0) {
4554 * Hand this block back to the kernel, and, if
4555 * we're counting blocks that need to be
4556 * filtered in userland after having been
4557 * filtered by the kernel, count the one we've
4560 h.h3->hdr.bh1.block_status = TP_STATUS_KERNEL;
4561 if (handlep->blocks_to_filter_in_userland > 0) {
4562 handlep->blocks_to_filter_in_userland--;
4563 if (handlep->blocks_to_filter_in_userland == 0) {
4565 * No more blocks need to be filtered
4568 handlep->filter_in_userland = 0;
4573 if (++handle->offset >= handle->cc)
4576 handlep->current_packet = NULL;
4579 /* check for break loop condition*/
4580 if (handle->break_loop) {
4581 handle->break_loop = 0;
4582 return PCAP_ERROR_BREAK;
4585 if (pkts == 0 && handlep->timeout == 0) {
4586 /* Block until we see a packet. */
4591 #endif /* HAVE_TPACKET3 */
4594 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
4596 struct pcap_linux *handlep = handle->priv;
4601 * Don't rewrite "ret" instructions; we don't need to, as
4602 * we're not reading packets with recvmsg(), and we don't
4603 * want to, as, by not rewriting them, the kernel can avoid
4604 * copying extra data.
4606 ret = pcap_setfilter_linux_common(handle, filter, 1);
4611 * If we're filtering in userland, there's nothing to do;
4612 * the new filter will be used for the next packet.
4614 if (handlep->filter_in_userland)
4618 * We're filtering in the kernel; the packets present in
4619 * all blocks currently in the ring were already filtered
4620 * by the old filter, and so will need to be filtered in
4621 * userland by the new filter.
4623 * Get an upper bound for the number of such blocks; first,
4624 * walk the ring backward and count the free blocks.
4626 offset = handle->offset;
4627 if (--handle->offset < 0)
4628 handle->offset = handle->cc - 1;
4629 for (n=0; n < handle->cc; ++n) {
4630 if (--handle->offset < 0)
4631 handle->offset = handle->cc - 1;
4632 if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL))
4637 * If we found free blocks, decrement the count of free
4638 * blocks by 1, just in case we lost a race with another
4639 * thread of control that was adding a packet while
4640 * we were counting and that had run the filter before
4643 * XXX - could there be more than one block added in
4646 * XXX - is there a way to avoid that race, e.g. somehow
4647 * wait for all packets that passed the old filter to
4648 * be added to the ring?
4653 /* be careful to not change current ring position */
4654 handle->offset = offset;
4657 * Set the count of blocks worth of packets to filter
4658 * in userland to the total number of blocks in the
4659 * ring minus the number of free blocks we found, and
4660 * turn on userland filtering. (The count of blocks
4661 * worth of packets to filter in userland is guaranteed
4662 * not to be zero - n, above, couldn't be set to a
4663 * value > handle->cc, and if it were equal to
4664 * handle->cc, it wouldn't be zero, and thus would
4665 * be decremented to handle->cc - 1.)
4667 handlep->blocks_to_filter_in_userland = handle->cc - n;
4668 handlep->filter_in_userland = 1;
4672 #endif /* HAVE_PACKET_RING */
4675 #ifdef HAVE_PF_PACKET_SOCKETS
4677 * Return the index of the given device name. Fill ebuf and return
4681 iface_get_id(int fd, const char *device, char *ebuf)
4685 memset(&ifr, 0, sizeof(ifr));
4686 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4688 if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
4689 snprintf(ebuf, PCAP_ERRBUF_SIZE,
4690 "SIOCGIFINDEX: %s", pcap_strerror(errno));
4694 return ifr.ifr_ifindex;
4698 * Bind the socket associated with FD to the given device.
4699 * Return 1 on success, 0 if we should try a SOCK_PACKET socket,
4700 * or a PCAP_ERROR_ value on a hard error.
4703 iface_bind(int fd, int ifindex, char *ebuf)
4705 struct sockaddr_ll sll;
4707 socklen_t errlen = sizeof(err);
4709 memset(&sll, 0, sizeof(sll));
4710 sll.sll_family = AF_PACKET;
4711 sll.sll_ifindex = ifindex;
4712 sll.sll_protocol = htons(ETH_P_ALL);
4714 if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
4715 if (errno == ENETDOWN) {
4717 * Return a "network down" indication, so that
4718 * the application can report that rather than
4719 * saying we had a mysterious failure and
4720 * suggest that they report a problem to the
4721 * libpcap developers.
4723 return PCAP_ERROR_IFACE_NOT_UP;
4725 snprintf(ebuf, PCAP_ERRBUF_SIZE,
4726 "bind: %s", pcap_strerror(errno));
4731 /* Any pending errors, e.g., network is down? */
4733 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
4734 snprintf(ebuf, PCAP_ERRBUF_SIZE,
4735 "getsockopt: %s", pcap_strerror(errno));
4739 if (err == ENETDOWN) {
4741 * Return a "network down" indication, so that
4742 * the application can report that rather than
4743 * saying we had a mysterious failure and
4744 * suggest that they report a problem to the
4745 * libpcap developers.
4747 return PCAP_ERROR_IFACE_NOT_UP;
4748 } else if (err > 0) {
4749 snprintf(ebuf, PCAP_ERRBUF_SIZE,
4750 "bind: %s", pcap_strerror(err));
4757 #ifdef IW_MODE_MONITOR
4759 * Check whether the device supports the Wireless Extensions.
4760 * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE
4761 * if the device doesn't even exist.
4764 has_wext(int sock_fd, const char *device, char *ebuf)
4768 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
4769 sizeof ireq.ifr_ifrn.ifrn_name);
4770 if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
4772 snprintf(ebuf, PCAP_ERRBUF_SIZE,
4773 "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
4774 if (errno == ENODEV)
4775 return PCAP_ERROR_NO_SUCH_DEVICE;
4780 * Per me si va ne la citta dolente,
4781 * Per me si va ne l'etterno dolore,
4783 * Lasciate ogne speranza, voi ch'intrate.
4785 * XXX - airmon-ng does special stuff with the Orinoco driver and the
4801 * Use the Wireless Extensions, if we have them, to try to turn monitor mode
4802 * on if it's not already on.
4804 * Returns 1 on success, 0 if we don't support the Wireless Extensions
4805 * on this device, or a PCAP_ERROR_ value if we do support them but
4806 * we weren't able to turn monitor mode on.
4809 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
4812 * XXX - at least some adapters require non-Wireless Extensions
4813 * mechanisms to turn monitor mode on.
4815 * Atheros cards might require that a separate "monitor virtual access
4816 * point" be created, with later versions of the madwifi driver.
4817 * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode
4818 * monitor -bssid", which apparently spits out a line "athN"
4819 * where "athN" is the monitor mode device. To leave monitor
4820 * mode, it destroys the monitor mode device.
4822 * Some Intel Centrino adapters might require private ioctls to get
4823 * radio headers; the ipw2200 and ipw3945 drivers allow you to
4824 * configure a separate "rtapN" interface to capture in monitor
4825 * mode without preventing the adapter from operating normally.
4826 * (airmon-ng doesn't appear to use that, though.)
4828 * It would be Truly Wonderful if mac80211 and nl80211 cleaned this
4829 * up, and if all drivers were converted to mac80211 drivers.
4831 * If interface {if} is a mac80211 driver, the file
4832 * /sys/class/net/{if}/phy80211 is a symlink to
4833 * /sys/class/ieee80211/{phydev}, for some {phydev}.
4835 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
4836 * least, has a "wmaster0" device and a "wlan0" device; the
4837 * latter is the one with the IP address. Both show up in
4838 * "tcpdump -D" output. Capturing on the wmaster0 device
4839 * captures with 802.11 headers.
4841 * airmon-ng searches through /sys/class/net for devices named
4842 * monN, starting with mon0; as soon as one *doesn't* exist,
4843 * it chooses that as the monitor device name. If the "iw"
4844 * command exists, it does "iw dev {if} interface add {monif}
4845 * type monitor", where {monif} is the monitor device. It
4846 * then (sigh) sleeps .1 second, and then configures the
4847 * device up. Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
4848 * is a file, it writes {mondev}, without a newline, to that file,
4849 * and again (sigh) sleeps .1 second, and then iwconfig's that
4850 * device into monitor mode and configures it up. Otherwise,
4851 * you can't do monitor mode.
4853 * All these devices are "glued" together by having the
4854 * /sys/class/net/{device}/phy80211 links pointing to the same
4855 * place, so, given a wmaster, wlan, or mon device, you can
4856 * find the other devices by looking for devices with
4857 * the same phy80211 link.
4859 * To turn monitor mode off, delete the monitor interface,
4860 * either with "iw dev {monif} interface del" or by sending
4861 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
4863 * Note: if you try to create a monitor device named "monN", and
4864 * there's already a "monN" device, it fails, as least with
4865 * the netlink interface (which is what iw uses), with a return
4866 * value of -ENFILE. (Return values are negative errnos.) We
4867 * could probably use that to find an unused device.
4869 struct pcap_linux *handlep = handle->priv;
4872 struct iw_priv_args *priv;
4873 monitor_type montype;
4882 * Does this device *support* the Wireless Extensions?
4884 err = has_wext(sock_fd, device, handle->errbuf);
4886 return err; /* either it doesn't or the device doesn't even exist */
4888 * Start out assuming we have no private extensions to control
4891 montype = MONITOR_WEXT;
4895 * Try to get all the Wireless Extensions private ioctls
4896 * supported by this device.
4898 * First, get the size of the buffer we need, by supplying no
4899 * buffer and a length of 0. If the device supports private
4900 * ioctls, it should return E2BIG, with ireq.u.data.length set
4901 * to the length we need. If it doesn't support them, it should
4902 * return EOPNOTSUPP.
4904 memset(&ireq, 0, sizeof ireq);
4905 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
4906 sizeof ireq.ifr_ifrn.ifrn_name);
4907 ireq.u.data.pointer = (void *)args;
4908 ireq.u.data.length = 0;
4909 ireq.u.data.flags = 0;
4910 if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
4911 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4912 "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!",
4916 if (errno != EOPNOTSUPP) {
4918 * OK, it's not as if there are no private ioctls.
4920 if (errno != E2BIG) {
4924 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4925 "%s: SIOCGIWPRIV: %s", device,
4926 pcap_strerror(errno));
4931 * OK, try to get the list of private ioctls.
4933 priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args));
4935 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4936 "malloc: %s", pcap_strerror(errno));
4939 ireq.u.data.pointer = (void *)priv;
4940 if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) {
4941 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4942 "%s: SIOCGIWPRIV: %s", device,
4943 pcap_strerror(errno));
4949 * Look for private ioctls to turn monitor mode on or, if
4950 * monitor mode is on, to set the header type.
4952 for (i = 0; i < ireq.u.data.length; i++) {
4953 if (strcmp(priv[i].name, "monitor_type") == 0) {
4955 * Hostap driver, use this one.
4956 * Set monitor mode first.
4957 * You can set it to 0 to get DLT_IEEE80211,
4958 * 1 to get DLT_PRISM, 2 to get
4959 * DLT_IEEE80211_RADIO_AVS, and, with more
4960 * recent versions of the driver, 3 to get
4961 * DLT_IEEE80211_RADIO.
4963 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4965 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4967 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4969 montype = MONITOR_HOSTAP;
4973 if (strcmp(priv[i].name, "set_prismhdr") == 0) {
4975 * Prism54 driver, use this one.
4976 * Set monitor mode first.
4977 * You can set it to 2 to get DLT_IEEE80211
4978 * or 3 or get DLT_PRISM.
4980 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4982 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
4984 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
4986 montype = MONITOR_PRISM54;
4990 if (strcmp(priv[i].name, "forceprismheader") == 0) {
4992 * RT2570 driver, use this one.
4993 * Do this after turning monitor mode on.
4994 * You can set it to 1 to get DLT_PRISM or 2
4995 * to get DLT_IEEE80211.
4997 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
4999 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5001 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5003 montype = MONITOR_RT2570;
5007 if (strcmp(priv[i].name, "forceprism") == 0) {
5009 * RT73 driver, use this one.
5010 * Do this after turning monitor mode on.
5011 * Its argument is a *string*; you can
5012 * set it to "1" to get DLT_PRISM or "2"
5013 * to get DLT_IEEE80211.
5015 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR)
5017 if (priv[i].set_args & IW_PRIV_SIZE_FIXED)
5019 montype = MONITOR_RT73;
5023 if (strcmp(priv[i].name, "prismhdr") == 0) {
5025 * One of the RTL8xxx drivers, use this one.
5026 * It can only be done after monitor mode
5027 * has been turned on. You can set it to 1
5028 * to get DLT_PRISM or 0 to get DLT_IEEE80211.
5030 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5032 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5034 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
5036 montype = MONITOR_RTL8XXX;
5040 if (strcmp(priv[i].name, "rfmontx") == 0) {
5042 * RT2500 or RT61 driver, use this one.
5043 * It has one one-byte parameter; set
5044 * u.data.length to 1 and u.data.pointer to
5045 * point to the parameter.
5046 * It doesn't itself turn monitor mode on.
5047 * You can set it to 1 to allow transmitting
5048 * in monitor mode(?) and get DLT_IEEE80211,
5049 * or set it to 0 to disallow transmitting in
5050 * monitor mode(?) and get DLT_PRISM.
5052 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5054 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2)
5056 montype = MONITOR_RT2500;
5060 if (strcmp(priv[i].name, "monitor") == 0) {
5062 * Either ACX100 or hostap, use this one.
5063 * It turns monitor mode on.
5064 * If it takes two arguments, it's ACX100;
5065 * the first argument is 1 for DLT_PRISM
5066 * or 2 for DLT_IEEE80211, and the second
5067 * argument is the channel on which to
5068 * run. If it takes one argument, it's
5069 * HostAP, and the argument is 2 for
5070 * DLT_IEEE80211 and 3 for DLT_PRISM.
5072 * If we see this, we don't quit, as this
5073 * might be a version of the hostap driver
5074 * that also supports "monitor_type".
5076 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
5078 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
5080 switch (priv[i].set_args & IW_PRIV_SIZE_MASK) {
5083 montype = MONITOR_PRISM;
5088 montype = MONITOR_ACX100;
5101 * XXX - ipw3945? islism?
5107 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5108 sizeof ireq.ifr_ifrn.ifrn_name);
5109 if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) {
5111 * We probably won't be able to set the mode, either.
5113 return PCAP_ERROR_RFMON_NOTSUP;
5117 * Is it currently in monitor mode?
5119 if (ireq.u.mode == IW_MODE_MONITOR) {
5121 * Yes. Just leave things as they are.
5122 * We don't offer multiple link-layer types, as
5123 * changing the link-layer type out from under
5124 * somebody else capturing in monitor mode would
5125 * be considered rude.
5130 * No. We have to put the adapter into rfmon mode.
5134 * If we haven't already done so, arrange to have
5135 * "pcap_close_all()" called when we exit.
5137 if (!pcap_do_addexit(handle)) {
5139 * "atexit()" failed; don't put the interface
5140 * in rfmon mode, just give up.
5142 return PCAP_ERROR_RFMON_NOTSUP;
5146 * Save the old mode.
5148 handlep->oldmode = ireq.u.mode;
5151 * Put the adapter in rfmon mode. How we do this depends
5152 * on whether we have a special private ioctl or not.
5154 if (montype == MONITOR_PRISM) {
5156 * We have the "monitor" private ioctl, but none of
5157 * the other private ioctls. Use this, and select
5160 * If it fails, just fall back on SIOCSIWMODE.
5162 memset(&ireq, 0, sizeof ireq);
5163 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5164 sizeof ireq.ifr_ifrn.ifrn_name);
5165 ireq.u.data.length = 1; /* 1 argument */
5166 args[0] = 3; /* request Prism header */
5167 memcpy(ireq.u.name, args, sizeof (int));
5168 if (ioctl(sock_fd, cmd, &ireq) != -1) {
5171 * Note that we have to put the old mode back
5172 * when we close the device.
5174 handlep->must_do_on_close |= MUST_CLEAR_RFMON;
5177 * Add this to the list of pcaps to close
5180 pcap_add_to_pcaps_to_close(handle);
5186 * Failure. Fall back on SIOCSIWMODE.
5191 * First, take the interface down if it's up; otherwise, we
5194 memset(&ifr, 0, sizeof(ifr));
5195 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5196 if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
5197 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5198 "%s: Can't get flags: %s", device, strerror(errno));
5202 if (ifr.ifr_flags & IFF_UP) {
5203 oldflags = ifr.ifr_flags;
5204 ifr.ifr_flags &= ~IFF_UP;
5205 if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5206 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5207 "%s: Can't set flags: %s", device, strerror(errno));
5213 * Then turn monitor mode on.
5215 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5216 sizeof ireq.ifr_ifrn.ifrn_name);
5217 ireq.u.mode = IW_MODE_MONITOR;
5218 if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) {
5220 * Scientist, you've failed.
5221 * Bring the interface back up if we shut it down.
5223 ifr.ifr_flags = oldflags;
5224 if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5225 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5226 "%s: Can't set flags: %s", device, strerror(errno));
5229 return PCAP_ERROR_RFMON_NOTSUP;
5233 * XXX - airmon-ng does "iwconfig {if} key off" after setting
5234 * monitor mode and setting the channel, and then does
5239 * Now select the appropriate radio header.
5245 * We don't have any private ioctl to set the header.
5249 case MONITOR_HOSTAP:
5251 * Try to select the radiotap header.
5253 memset(&ireq, 0, sizeof ireq);
5254 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5255 sizeof ireq.ifr_ifrn.ifrn_name);
5256 args[0] = 3; /* request radiotap header */
5257 memcpy(ireq.u.name, args, sizeof (int));
5258 if (ioctl(sock_fd, cmd, &ireq) != -1)
5259 break; /* success */
5262 * That failed. Try to select the AVS header.
5264 memset(&ireq, 0, sizeof ireq);
5265 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5266 sizeof ireq.ifr_ifrn.ifrn_name);
5267 args[0] = 2; /* request AVS header */
5268 memcpy(ireq.u.name, args, sizeof (int));
5269 if (ioctl(sock_fd, cmd, &ireq) != -1)
5270 break; /* success */
5273 * That failed. Try to select the Prism header.
5275 memset(&ireq, 0, sizeof ireq);
5276 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5277 sizeof ireq.ifr_ifrn.ifrn_name);
5278 args[0] = 1; /* request Prism header */
5279 memcpy(ireq.u.name, args, sizeof (int));
5280 ioctl(sock_fd, cmd, &ireq);
5285 * The private ioctl failed.
5289 case MONITOR_PRISM54:
5291 * Select the Prism header.
5293 memset(&ireq, 0, sizeof ireq);
5294 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5295 sizeof ireq.ifr_ifrn.ifrn_name);
5296 args[0] = 3; /* request Prism header */
5297 memcpy(ireq.u.name, args, sizeof (int));
5298 ioctl(sock_fd, cmd, &ireq);
5301 case MONITOR_ACX100:
5303 * Get the current channel.
5305 memset(&ireq, 0, sizeof ireq);
5306 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5307 sizeof ireq.ifr_ifrn.ifrn_name);
5308 if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) {
5309 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5310 "%s: SIOCGIWFREQ: %s", device,
5311 pcap_strerror(errno));
5314 channel = ireq.u.freq.m;
5317 * Select the Prism header, and set the channel to the
5320 memset(&ireq, 0, sizeof ireq);
5321 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5322 sizeof ireq.ifr_ifrn.ifrn_name);
5323 args[0] = 1; /* request Prism header */
5324 args[1] = channel; /* set channel */
5325 memcpy(ireq.u.name, args, 2*sizeof (int));
5326 ioctl(sock_fd, cmd, &ireq);
5329 case MONITOR_RT2500:
5331 * Disallow transmission - that turns on the
5334 memset(&ireq, 0, sizeof ireq);
5335 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5336 sizeof ireq.ifr_ifrn.ifrn_name);
5337 args[0] = 0; /* disallow transmitting */
5338 memcpy(ireq.u.name, args, sizeof (int));
5339 ioctl(sock_fd, cmd, &ireq);
5342 case MONITOR_RT2570:
5344 * Force the Prism header.
5346 memset(&ireq, 0, sizeof ireq);
5347 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5348 sizeof ireq.ifr_ifrn.ifrn_name);
5349 args[0] = 1; /* request Prism header */
5350 memcpy(ireq.u.name, args, sizeof (int));
5351 ioctl(sock_fd, cmd, &ireq);
5356 * Force the Prism header.
5358 memset(&ireq, 0, sizeof ireq);
5359 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5360 sizeof ireq.ifr_ifrn.ifrn_name);
5361 ireq.u.data.length = 1; /* 1 argument */
5362 ireq.u.data.pointer = "1";
5363 ireq.u.data.flags = 0;
5364 ioctl(sock_fd, cmd, &ireq);
5367 case MONITOR_RTL8XXX:
5369 * Force the Prism header.
5371 memset(&ireq, 0, sizeof ireq);
5372 strlcpy(ireq.ifr_ifrn.ifrn_name, device,
5373 sizeof ireq.ifr_ifrn.ifrn_name);
5374 args[0] = 1; /* request Prism header */
5375 memcpy(ireq.u.name, args, sizeof (int));
5376 ioctl(sock_fd, cmd, &ireq);
5381 * Now bring the interface back up if we brought it down.
5383 if (oldflags != 0) {
5384 ifr.ifr_flags = oldflags;
5385 if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
5386 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5387 "%s: Can't set flags: %s", device, strerror(errno));
5390 * At least try to restore the old mode on the
5393 if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
5395 * Scientist, you've failed.
5398 "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n"
5399 "Please adjust manually.\n",
5407 * Note that we have to put the old mode back when we
5410 handlep->must_do_on_close |= MUST_CLEAR_RFMON;
5413 * Add this to the list of pcaps to close when we exit.
5415 pcap_add_to_pcaps_to_close(handle);
5419 #endif /* IW_MODE_MONITOR */
5422 * Try various mechanisms to enter monitor mode.
5425 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
5427 #if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
5432 ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
5434 return ret; /* error attempting to do so */
5436 return 1; /* success */
5437 #endif /* HAVE_LIBNL */
5439 #ifdef IW_MODE_MONITOR
5440 ret = enter_rfmon_mode_wext(handle, sock_fd, device);
5442 return ret; /* error attempting to do so */
5444 return 1; /* success */
5445 #endif /* IW_MODE_MONITOR */
5448 * Either none of the mechanisms we know about work or none
5449 * of those mechanisms are available, so we can't do monitor
5456 * Find out if we have any form of fragmentation/reassembly offloading.
5458 * We do so using SIOCETHTOOL checking for various types of offloading;
5459 * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
5460 * of the types of offloading, there's nothing we can do to check, so
5461 * we just say "no, we don't".
5463 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
5465 iface_ethtool_ioctl(pcap_t *handle, int cmd, const char *cmdname)
5468 struct ethtool_value eval;
5470 memset(&ifr, 0, sizeof(ifr));
5471 strlcpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name));
5474 ifr.ifr_data = (caddr_t)&eval;
5475 if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
5476 if (errno == EOPNOTSUPP || errno == EINVAL) {
5478 * OK, let's just return 0, which, in our
5479 * case, either means "no, what we're asking
5480 * about is not enabled" or "all the flags
5481 * are clear (i.e., nothing is enabled)".
5485 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5486 "%s: SIOETHTOOL(%s) ioctl failed: %s", handle->opt.source,
5487 cmdname, strerror(errno));
5494 iface_get_offload(pcap_t *handle)
5499 ret = iface_ethtool_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO");
5503 return 1; /* TCP segmentation offloading on */
5507 ret = iface_ethtool_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO");
5511 return 1; /* UDP fragmentation offloading on */
5516 * XXX - will this cause large unsegmented packets to be
5517 * handed to PF_PACKET sockets on transmission? If not,
5518 * this need not be checked.
5520 ret = iface_ethtool_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO");
5524 return 1; /* generic segmentation offloading on */
5527 #ifdef ETHTOOL_GFLAGS
5528 ret = iface_ethtool_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS");
5531 if (ret & ETH_FLAG_LRO)
5532 return 1; /* large receive offloading on */
5537 * XXX - will this cause large reassembled packets to be
5538 * handed to PF_PACKET sockets on receipt? If not,
5539 * this need not be checked.
5541 ret = iface_ethtool_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO");
5545 return 1; /* generic (large) receive offloading on */
5550 #else /* SIOCETHTOOL */
5552 iface_get_offload(pcap_t *handle _U_)
5555 * XXX - do we need to get this information if we don't
5556 * have the ethtool ioctls? If so, how do we do that?
5560 #endif /* SIOCETHTOOL */
5562 #endif /* HAVE_PF_PACKET_SOCKETS */
5564 /* ===== Functions to interface to the older kernels ================== */
5567 * Try to open a packet socket using the old kernel interface.
5568 * Returns 1 on success and a PCAP_ERROR_ value on an error.
5571 activate_old(pcap_t *handle)
5573 struct pcap_linux *handlep = handle->priv;
5576 const char *device = handle->opt.source;
5577 struct utsname utsname;
5580 /* Open the socket */
5582 handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
5583 if (handle->fd == -1) {
5584 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5585 "socket: %s", pcap_strerror(errno));
5586 if (errno == EPERM || errno == EACCES) {
5588 * You don't have permission to open the
5591 return PCAP_ERROR_PERM_DENIED;
5600 /* It worked - we are using the old interface */
5601 handlep->sock_packet = 1;
5603 /* ...which means we get the link-layer header. */
5604 handlep->cooked = 0;
5606 /* Bind to the given device */
5608 if (strcmp(device, "any") == 0) {
5609 strlcpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
5613 if (iface_bind_old(handle->fd, device, handle->errbuf) == -1)
5617 * Try to get the link-layer type.
5619 arptype = iface_get_arptype(handle->fd, device, handle->errbuf);
5624 * Try to find the DLT_ type corresponding to that
5627 map_arphrd_to_dlt(handle, arptype, device, 0);
5628 if (handle->linktype == -1) {
5629 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5630 "unknown arptype %d", arptype);
5634 /* Go to promisc mode if requested */
5636 if (handle->opt.promisc) {
5637 memset(&ifr, 0, sizeof(ifr));
5638 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5639 if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
5640 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5641 "SIOCGIFFLAGS: %s", pcap_strerror(errno));
5644 if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
5646 * Promiscuous mode isn't currently on,
5647 * so turn it on, and remember that
5648 * we should turn it off when the
5653 * If we haven't already done so, arrange
5654 * to have "pcap_close_all()" called when
5657 if (!pcap_do_addexit(handle)) {
5659 * "atexit()" failed; don't put
5660 * the interface in promiscuous
5661 * mode, just give up.
5666 ifr.ifr_flags |= IFF_PROMISC;
5667 if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
5668 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5670 pcap_strerror(errno));
5673 handlep->must_do_on_close |= MUST_CLEAR_PROMISC;
5676 * Add this to the list of pcaps
5677 * to close when we exit.
5679 pcap_add_to_pcaps_to_close(handle);
5684 * Compute the buffer size.
5686 * We're using SOCK_PACKET, so this might be a 2.0[.x]
5687 * kernel, and might require special handling - check.
5689 if (uname(&utsname) < 0 ||
5690 strncmp(utsname.release, "2.0", 3) == 0) {
5692 * Either we couldn't find out what kernel release
5693 * this is, or it's a 2.0[.x] kernel.
5695 * In the 2.0[.x] kernel, a "recvfrom()" on
5696 * a SOCK_PACKET socket, with MSG_TRUNC set, will
5697 * return the number of bytes read, so if we pass
5698 * a length based on the snapshot length, it'll
5699 * return the number of bytes from the packet
5700 * copied to userland, not the actual length
5703 * This means that, for example, the IP dissector
5704 * in tcpdump will get handed a packet length less
5705 * than the length in the IP header, and will
5706 * complain about "truncated-ip".
5708 * So we don't bother trying to copy from the
5709 * kernel only the bytes in which we're interested,
5710 * but instead copy them all, just as the older
5711 * versions of libpcap for Linux did.
5713 * The buffer therefore needs to be big enough to
5714 * hold the largest packet we can get from this
5715 * device. Unfortunately, we can't get the MRU
5716 * of the network; we can only get the MTU. The
5717 * MTU may be too small, in which case a packet larger
5718 * than the buffer size will be truncated *and* we
5719 * won't get the actual packet size.
5721 * However, if the snapshot length is larger than
5722 * the buffer size based on the MTU, we use the
5723 * snapshot length as the buffer size, instead;
5724 * this means that with a sufficiently large snapshot
5725 * length we won't artificially truncate packets
5726 * to the MTU-based size.
5728 * This mess just one of many problems with packet
5729 * capture on 2.0[.x] kernels; you really want a
5730 * 2.2[.x] or later kernel if you want packet capture
5733 mtu = iface_get_mtu(handle->fd, device, handle->errbuf);
5736 handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
5737 if (handle->bufsize < handle->snapshot)
5738 handle->bufsize = handle->snapshot;
5741 * This is a 2.2[.x] or later kernel.
5743 * We can safely pass "recvfrom()" a byte count
5744 * based on the snapshot length.
5746 handle->bufsize = handle->snapshot;
5750 * Default value for offset to align link-layer payload
5751 * on a 4-byte boundary.
5756 * SOCK_PACKET sockets don't supply information from
5757 * stripped VLAN tags.
5759 handlep->vlan_offset = -1; /* unknown */
5765 * Bind the socket associated with FD to the given device using the
5766 * interface of the old kernels.
5769 iface_bind_old(int fd, const char *device, char *ebuf)
5771 struct sockaddr saddr;
5773 socklen_t errlen = sizeof(err);
5775 memset(&saddr, 0, sizeof(saddr));
5776 strlcpy(saddr.sa_data, device, sizeof(saddr.sa_data));
5777 if (bind(fd, &saddr, sizeof(saddr)) == -1) {
5778 snprintf(ebuf, PCAP_ERRBUF_SIZE,
5779 "bind: %s", pcap_strerror(errno));
5783 /* Any pending errors, e.g., network is down? */
5785 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
5786 snprintf(ebuf, PCAP_ERRBUF_SIZE,
5787 "getsockopt: %s", pcap_strerror(errno));
5792 snprintf(ebuf, PCAP_ERRBUF_SIZE,
5793 "bind: %s", pcap_strerror(err));
5801 /* ===== System calls available on all supported kernels ============== */
5804 * Query the kernel for the MTU of the given interface.
5807 iface_get_mtu(int fd, const char *device, char *ebuf)
5812 return BIGGER_THAN_ALL_MTUS;
5814 memset(&ifr, 0, sizeof(ifr));
5815 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5817 if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
5818 snprintf(ebuf, PCAP_ERRBUF_SIZE,
5819 "SIOCGIFMTU: %s", pcap_strerror(errno));
5827 * Get the hardware type of the given interface as ARPHRD_xxx constant.
5830 iface_get_arptype(int fd, const char *device, char *ebuf)
5834 memset(&ifr, 0, sizeof(ifr));
5835 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5837 if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
5838 snprintf(ebuf, PCAP_ERRBUF_SIZE,
5839 "SIOCGIFHWADDR: %s", pcap_strerror(errno));
5840 if (errno == ENODEV) {
5844 return PCAP_ERROR_NO_SUCH_DEVICE;
5849 return ifr.ifr_hwaddr.sa_family;
5852 #ifdef SO_ATTACH_FILTER
5854 fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
5856 struct pcap_linux *handlep = handle->priv;
5859 register struct bpf_insn *p;
5864 * Make a copy of the filter, and modify that copy if
5867 prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
5868 len = handle->fcode.bf_len;
5869 f = (struct bpf_insn *)malloc(prog_size);
5871 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5872 "malloc: %s", pcap_strerror(errno));
5875 memcpy(f, handle->fcode.bf_insns, prog_size);
5877 fcode->filter = (struct sock_filter *) f;
5879 for (i = 0; i < len; ++i) {
5882 * What type of instruction is this?
5884 switch (BPF_CLASS(p->code)) {
5888 * It's a return instruction; are we capturing
5889 * in memory-mapped mode?
5893 * No; is the snapshot length a constant,
5894 * rather than the contents of the
5897 if (BPF_MODE(p->code) == BPF_K) {
5899 * Yes - if the value to be returned,
5900 * i.e. the snapshot length, is
5901 * anything other than 0, make it
5902 * MAXIMUM_SNAPLEN, so that the packet
5903 * is truncated by "recvfrom()",
5904 * not by the filter.
5906 * XXX - there's nothing we can
5907 * easily do if it's getting the
5908 * value from the accumulator; we'd
5909 * have to insert code to force
5910 * non-zero values to be
5914 p->k = MAXIMUM_SNAPLEN;
5922 * It's a load instruction; is it loading
5925 switch (BPF_MODE(p->code)) {
5931 * Yes; are we in cooked mode?
5933 if (handlep->cooked) {
5935 * Yes, so we need to fix this
5938 if (fix_offset(p) < 0) {
5940 * We failed to do so.
5941 * Return 0, so our caller
5942 * knows to punt to userland.
5952 return 1; /* we succeeded */
5956 fix_offset(struct bpf_insn *p)
5959 * What's the offset?
5961 if (p->k >= SLL_HDR_LEN) {
5963 * It's within the link-layer payload; that starts at an
5964 * offset of 0, as far as the kernel packet filter is
5965 * concerned, so subtract the length of the link-layer
5968 p->k -= SLL_HDR_LEN;
5969 } else if (p->k == 0) {
5971 * It's the packet type field; map it to the special magic
5972 * kernel offset for that field.
5974 p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
5975 } else if (p->k == 14) {
5977 * It's the protocol field; map it to the special magic
5978 * kernel offset for that field.
5980 p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5981 } else if ((bpf_int32)(p->k) > 0) {
5983 * It's within the header, but it's not one of those
5984 * fields; we can't do that in the kernel, so punt
5993 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
5995 int total_filter_on = 0;
6001 * The socket filter code doesn't discard all packets queued
6002 * up on the socket when the filter is changed; this means
6003 * that packets that don't match the new filter may show up
6004 * after the new filter is put onto the socket, if those
6005 * packets haven't yet been read.
6007 * This means, for example, that if you do a tcpdump capture
6008 * with a filter, the first few packets in the capture might
6009 * be packets that wouldn't have passed the filter.
6011 * We therefore discard all packets queued up on the socket
6012 * when setting a kernel filter. (This isn't an issue for
6013 * userland filters, as the userland filtering is done after
6014 * packets are queued up.)
6016 * To flush those packets, we put the socket in read-only mode,
6017 * and read packets from the socket until there are no more to
6020 * In order to keep that from being an infinite loop - i.e.,
6021 * to keep more packets from arriving while we're draining
6022 * the queue - we put the "total filter", which is a filter
6023 * that rejects all packets, onto the socket before draining
6026 * This code deliberately ignores any errors, so that you may
6027 * get bogus packets if an error occurs, rather than having
6028 * the filtering done in userland even if it could have been
6029 * done in the kernel.
6031 if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
6032 &total_fcode, sizeof(total_fcode)) == 0) {
6036 * Note that we've put the total filter onto the socket.
6038 total_filter_on = 1;
6041 * Save the socket's current mode, and put it in
6042 * non-blocking mode; we drain it by reading packets
6043 * until we get an error (which is normally a
6044 * "nothing more to be read" error).
6046 save_mode = fcntl(handle->fd, F_GETFL, 0);
6047 if (save_mode != -1 &&
6048 fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) >= 0) {
6049 while (recv(handle->fd, &drain, sizeof drain,
6053 fcntl(handle->fd, F_SETFL, save_mode);
6054 if (save_errno != EAGAIN) {
6056 reset_kernel_filter(handle);
6057 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
6058 "recv: %s", pcap_strerror(save_errno));
6065 * Now attach the new filter.
6067 ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
6068 fcode, sizeof(*fcode));
6069 if (ret == -1 && total_filter_on) {
6071 * Well, we couldn't set that filter on the socket,
6072 * but we could set the total filter on the socket.
6074 * This could, for example, mean that the filter was
6075 * too big to put into the kernel, so we'll have to
6076 * filter in userland; in any case, we'll be doing
6077 * filtering in userland, so we need to remove the
6078 * total filter so we see packets.
6083 * XXX - if this fails, we're really screwed;
6084 * we have the total filter on the socket,
6085 * and it won't come off. What do we do then?
6087 reset_kernel_filter(handle);
6095 reset_kernel_filter(pcap_t *handle)
6098 * setsockopt() barfs unless it get a dummy parameter.
6099 * valgrind whines unless the value is initialized,
6100 * as it has no idea that setsockopt() ignores its
6105 return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
6106 &dummy, sizeof(dummy));