]>
The Tcpdump Group git mirrors - libpcap/blob - pcap-linux.c
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.
28 static const char rcsid
[] =
29 "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.35 2000-10-20 06:55:28 guy Exp $ (LBL)";
34 * - setting promiscuous on loopback gives every packet twice
49 #include <sys/socket.h>
50 #include <sys/ioctl.h>
52 #include <netinet/in.h>
53 #include <linux/if_ether.h>
54 #include <netinet/if_ether.h>
56 #ifdef HAVE_NETPACKET_PACKET_H
57 #include <netpacket/packet.h>
59 #ifdef SO_ATTACH_FILTER
60 #include <linux/types.h>
61 #include <linux/filter.h>
65 typedef int socklen_t
;
72 #define MAX_LINKHEADER_SIZE 256
75 * When capturing on all interfaces we use this as the buffer size.
76 * Should be bigger then all MTUs that occur in real life.
77 * 64kB should be enough for now.
79 #define BIGGER_THAN_ALL_MTUS (64*1024)
82 * Prototypes for internal functions
84 static int map_arphrd_to_dlt(int arptype
);
85 static int live_open_old(pcap_t
*, char *, int, int, char *);
86 static int live_open_new(pcap_t
*, char *, int, int, char *);
87 static int pcap_read_packet(pcap_t
*, pcap_handler
, u_char
*);
90 * Wrap some ioctl calls
92 #ifdef HAVE_NETPACKET_PACKET_H
93 static int iface_get_id(int fd
, const char *device
, char *ebuf
);
95 static int iface_get_mtu(int fd
, const char *device
, char *ebuf
);
96 static int iface_get_arptype(int fd
, const char *device
, char *ebuf
);
97 #ifdef HAVE_NETPACKET_PACKET_H
98 static int iface_bind(int fd
, int ifindex
, char *ebuf
);
100 static int iface_bind_old(int fd
, const char *device
, char *ebuf
);
103 * Get a handle for a live capture from the given device. You can
104 * pass NULL as device to get all packages (without link level
105 * information of course). If you pass 1 as promisc the interface
106 * will be set to promiscous mode (XXX: I think this usage should
107 * be deprecated and functions be added to select that later allow
108 * modification of that values -- Torsten).
113 pcap_open_live(char *device
, int snaplen
, int promisc
, int to_ms
, char *ebuf
)
115 /* Allocate a handle for this session. */
117 pcap_t
*handle
= malloc(sizeof(*handle
));
118 if (handle
== NULL
) {
119 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "malloc: %s",
120 pcap_strerror(errno
));
124 /* Initialize some components of the pcap structure. */
126 memset(handle
, 0, sizeof(*handle
));
127 handle
->snapshot
= snaplen
;
128 handle
->md
.timeout
= to_ms
;
129 handle
->md
.promisc
= promisc
;
132 * NULL and "any" are special devices which give us the hint to
133 * monitor all devices.
135 if (!device
|| strcmp(device
, "any") == 0) {
137 handle
->md
.device
= strdup("any");
139 handle
->md
.device
= strdup(device
);
141 if (handle
->md
.device
== NULL
) {
142 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "strdup: %s",
143 pcap_strerror(errno
) );
149 * Current Linux kernels use the protocol family PF_PACKET to
150 * allow direct access to all packets on the network while
151 * older kernels had a special socket type SOCK_PACKET to
152 * implement this feature.
153 * While this old implementation is kind of obsolete we need
154 * to be compatible with older kernels for a while so we are
155 * trying both methods with the newer method preferred.
158 if (! (live_open_new(handle
, device
, promisc
, to_ms
, ebuf
) ||
159 live_open_old(handle
, device
, promisc
, to_ms
, ebuf
)) )
162 * Both methods to open the packet socket failed. Tidy
163 * up and report our failure (ebuf is expected to be
164 * set by the functions above).
167 free(handle
->md
.device
);
173 * Okay, now we have a packet stream open. Maybe we need to handle
174 * a timeout? In that case we set the filehandle to nonblocking
175 * so pcap_read can try reading the fd and call select if no data
176 * is available at first.
180 int flags
= fcntl(handle
->fd
, F_GETFL
);
183 flags
= fcntl(handle
->fd
, F_SETFL
, flags
);
186 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "fcntl: %s",
187 pcap_strerror(errno
));
197 * Read at most max_packets from the capture stream and call the callback
198 * for each of them. Returns the number of packets handled or -1 if an
201 * XXX: Can I rely on the Linux-specified behaviour of select (returning
202 * the time left in the timeval structure)? I really don't want to query
203 * the system time before each select call...
205 * pcap_read currently gets not only a packet from the kernel but also
206 * the sockaddr_ll returned as source of the packet. This way we can at
207 * some time extend tcpdump and libpcap to sniff on all devices at a time
208 * and find the right printing routine by using the information in the
209 * sockaddr_ll structure.
212 pcap_read(pcap_t
*handle
, int max_packets
, pcap_handler callback
, u_char
*user
)
219 * Fill in a timeval structure for select if we need to obeye a
222 if (handle
->md
.timeout
> 0) {
223 tv
.tv_usec
= (handle
->md
.timeout
% 1000) * 1000;
224 tv
.tv_sec
= (handle
->md
.timeout
/ 1000);
228 * Read packets until the packet limit has been reached or
229 * an error occured while reading. Call the user function
230 * for each received packet.
232 for (packets
= 0; max_packets
== -1 || packets
< max_packets
;)
234 status
= pcap_read_packet(handle
, callback
, user
);
239 } else if (status
== -1)
243 * If no packet is available we go to sleep. FIXME: This
244 * might be better implemented using poll(?)
247 FD_SET(handle
->fd
, &read_fds
);
248 status
= select(handle
->fd
+ 1,
249 &read_fds
, NULL
, NULL
, &tv
);
254 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
255 "select: %s", pcap_strerror(errno
));
258 else if (status
== 0 ||
259 (tv
.tv_usec
== 0 && tv
.tv_sec
== 0))
267 * Read a packet from the socket calling the handler provided by
268 * the user. Returns the number of packets received or -1 if an
272 pcap_read_packet(pcap_t
*handle
, pcap_handler callback
, u_char
*userdata
)
274 struct sockaddr from
;
276 int packet_len
, caplen
;
277 struct pcap_pkthdr pcap_header
;
280 * We don't currently use the from return value of recvfrom but
281 * this will probably be implemented in the future.
284 /* Receive a single packet from the kernel */
287 fromlen
= sizeof(from
);
288 packet_len
= recvfrom(
289 handle
->fd
, handle
->buffer
+ handle
->offset
,
290 handle
->snapshot
, MSG_TRUNC
,
291 (struct sockaddr
*) &from
, &fromlen
);
292 } while (packet_len
== -1 && errno
== EINTR
);
294 /* Check if an error occured */
296 if (packet_len
== -1) {
298 return 0; /* no packet there */
300 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
301 "recvfrom: %s", pcap_strerror(errno
));
307 * XXX: According to the kernel source we should get the real
308 * packet len if calling recvfrom with MSG_TRUNC set. It does
309 * not seem to work here :(, but it is supported by this code
311 * To be honest the code RELIES on that feature so this is really
312 * broken with 2.2.x kernels.
313 * I spend a day to figure out what's going on and I found out
314 * that the following is happening:
316 * The packet comes from a random interface and the packet_rcv
317 * hook is called with a clone of the packet. That code inserts
318 * the packet into the receive queue of the packet socket.
319 * If a filter is attached to that socket that filter is run
320 * first - and there lies the problem. The default filter always
321 * cuts the packet at the snaplen:
326 * So the packet filter cuts down the packet. The recvfrom call
327 * says "hey, it's only 68 bytes, it fits into the buffer" with
328 * the result that we don't get the real packet length. This
329 * is valid at least until kernel 2.2.17pre6.
331 * tcpdump is currently fixed by changing the BPF code generator
332 * to not truncate the received packet.
336 if (caplen
> handle
->snapshot
)
337 caplen
= handle
->snapshot
;
339 /* Run the packet filter if not using kernel filter */
340 if (!handle
->md
.use_bpf
&& handle
->fcode
.bf_insns
) {
341 if (bpf_filter(handle
->fcode
.bf_insns
, handle
->buffer
,
342 packet_len
, caplen
) == 0)
344 /* rejected by filter */
349 /* Fill in our own header data */
351 if (ioctl(handle
->fd
, SIOCGSTAMP
, &pcap_header
.ts
) == -1) {
352 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
353 "ioctl: %s", pcap_strerror(errno
));
356 pcap_header
.caplen
= caplen
;
357 pcap_header
.len
= packet_len
;
359 /* Call the user supplied callback function */
360 handle
->md
.stat
.ps_recv
++;
361 callback(userdata
, &pcap_header
, handle
->buffer
+ handle
->offset
);
367 * Get the statistics for the given packet capture handle.
368 * FIXME: Currently does not report the number of dropped packets.
371 pcap_stats(pcap_t
*handle
, struct pcap_stat
*stats
)
373 *stats
= handle
->md
.stat
;
378 * Attach the given BPF code to the packet capture device.
381 pcap_setfilter(pcap_t
*handle
, struct bpf_program
*filter
)
383 #ifdef SO_ATTACH_FILTER
384 struct sock_fprog fcode
;
390 strncpy(handle
->errbuf
, "setfilter: No filter specified",
391 sizeof(handle
->errbuf
));
395 /* Free old filter code if existing */
397 handle
->fcode
.bf_len
= 0;
398 if (handle
->fcode
.bf_insns
) {
399 free(handle
->fcode
.bf_insns
);
400 handle
->fcode
.bf_insns
= NULL
;
404 /* Make our private copy of the filter */
406 handle
->fcode
.bf_len
= filter
->bf_len
;
407 handle
->fcode
.bf_insns
=
408 malloc(filter
->bf_len
* sizeof(*filter
->bf_insns
));
409 if (handle
->fcode
.bf_insns
== NULL
) {
410 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
411 "malloc: %s", pcap_strerror(errno
));
414 memcpy(handle
->fcode
.bf_insns
, filter
->bf_insns
,
415 filter
->bf_len
* sizeof(*filter
->bf_insns
));
418 * Run user level packet filter by default. Will be overriden if
419 * installing a kernel filter succeeds.
421 handle
->md
.use_bpf
= 0;
423 /* Install kernel level filter if possible */
425 #ifdef SO_ATTACH_FILTER
427 * Oh joy, the Linux kernel uses struct sock_fprog instead of
428 * struct bpf_program and of course the length field is of
429 * different size. Pointed out by Sebastian
432 fcode
.filter
= (struct sock_filter
*) handle
->fcode
.bf_insns
;
433 fcode
.len
= filter
->bf_len
;
435 if (filter
->bf_len
> USHRT_MAX
) {
437 * fcode.len is an unsigned short for current kernel.
438 * I have yet to see BPF-Code with that much instructions
439 * but still it is possible. So for the sake of
440 * correctness I added this check.
442 fprintf(stderr
, "Warning: Filter to complex for kernel\n");
446 if (setsockopt(handle
->fd
, SOL_SOCKET
, SO_ATTACH_FILTER
,
447 &fcode
, sizeof(fcode
)) == 0)
449 /* Installation succeded - using kernel filter. */
450 handle
->md
.use_bpf
= 1;
455 * Print a warning if kernel filter available but a problem
458 if (errno
!= ENOPROTOOPT
&& errno
!= EOPNOTSUPP
) {
459 fprintf(stderr
, "Warning: Kernel filter failed: %s\n",
460 pcap_strerror(errno
));
469 * Linux uses the ARP hardware type to identify the type of an
470 * interface. pcap uses the DLT_xxx constants for this. This
471 * function maps the ARPHRD_xxx constant to an appropriate
474 * Returns -1 if unable to map the type.
476 static int map_arphrd_to_dlt(int arptype
)
480 case ARPHRD_METRICOM
:
481 case ARPHRD_LOOPBACK
: return DLT_EN10MB
;
482 case ARPHRD_EETHER
: return DLT_EN3MB
;
483 case ARPHRD_AX25
: return DLT_AX25
;
484 case ARPHRD_PRONET
: return DLT_PRONET
;
485 case ARPHRD_CHAOS
: return DLT_CHAOS
;
486 case ARPHRD_IEEE802
: return DLT_IEEE802
;
487 case ARPHRD_ARCNET
: return DLT_ARCNET
;
488 case ARPHRD_FDDI
: return DLT_FDDI
;
490 #ifndef ARPHRD_ATM /* FIXME: How to #include this? */
491 #define ARPHRD_ATM 19
493 case ARPHRD_ATM
: return DLT_ATM_CLIP
;
499 case ARPHRD_SLIP
: return DLT_RAW
;
505 /* ===== Functions to interface to the newer kernels ================== */
508 * Try to open a packet socket using the new kernel interface.
509 * Returns 0 on failure.
510 * FIXME: 0 uses to mean success (Sebastian)
513 live_open_new(pcap_t
*handle
, char *device
, int promisc
,
514 int to_ms
, char *ebuf
)
516 #ifdef HAVE_NETPACKET_PACKET_H
517 int sock_fd
= -1, device_id
, mtu
, arptype
;
518 struct packet_mreq mr
;
520 /* One shot loop used for error handling - bail out with break */
524 * Open a socket with protocol family packet. If a device is
525 * given we try to open it in raw mode otherwise we use
526 * the cooked interface.
529 socket(PF_PACKET
, SOCK_RAW
, htons(ETH_P_ALL
))
530 : socket(PF_PACKET
, SOCK_DGRAM
, htons(ETH_P_ALL
));
533 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "socket: %s",
534 pcap_strerror(errno
) );
538 /* It seems the kernel supports the new interface. */
539 handle
->md
.sock_packet
= 0;
542 * What kind of frames do we have to deal with? Fall back
543 * to cooked mode if we have an unknown interface type.
547 arptype
= iface_get_arptype(sock_fd
, device
, ebuf
);
550 handle
->linktype
= map_arphrd_to_dlt(arptype
);
552 handle
->linktype
= DLT_RAW
;
554 if (handle
->linktype
== -1) {
555 /* Unknown interface type - reopen in cooked mode */
557 if (close(sock_fd
) == -1) {
558 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
559 "close: %s", pcap_strerror(errno
));
562 sock_fd
= socket(PF_PACKET
, SOCK_DGRAM
,
565 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
566 "socket: %s", pcap_strerror(errno
));
571 "Warning: Falling back to cooked socket\n");
572 handle
->linktype
= DLT_RAW
;
577 device_id
= iface_get_id(sock_fd
, device
, ebuf
);
581 if (iface_bind(sock_fd
, device_id
, ebuf
) == -1)
585 /* Select promiscous mode on/off */
589 * Hmm, how can we set promiscuous mode on all interfaces?
590 * I am not sure if that is possible at all.
594 memset(&mr
, 0, sizeof(mr
));
595 mr
.mr_ifindex
= device_id
;
596 mr
.mr_type
= promisc
?
597 PACKET_MR_PROMISC
: PACKET_MR_ALLMULTI
;
598 if (setsockopt(sock_fd
, SOL_PACKET
,
599 PACKET_ADD_MEMBERSHIP
, &mr
, sizeof(mr
)) == -1)
601 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
602 "setsockopt: %s", pcap_strerror(errno
));
608 /* Compute the buffersize */
610 mtu
= iface_get_mtu(sock_fd
, device
, ebuf
);
613 handle
->bufsize
= MAX_LINKHEADER_SIZE
+ mtu
;
614 if (handle
->bufsize
< handle
->snapshot
)
615 handle
->bufsize
= handle
->snapshot
;
617 /* Fill in the pcap structure */
619 handle
->fd
= sock_fd
;
622 handle
->buffer
= malloc(handle
->bufsize
);
623 if (!handle
->buffer
) {
624 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
625 "malloc: %s", pcap_strerror(errno
));
638 "New packet capturing interface not supported by build "
639 "environment", PCAP_ERRBUF_SIZE
);
644 #ifdef HAVE_NETPACKET_PACKET_H
646 * Return the index of the given device name. Fill ebuf and return
650 iface_get_id(int fd
, const char *device
, char *ebuf
)
654 memset(&ifr
, 0, sizeof(ifr
));
655 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
657 if (ioctl(fd
, SIOCGIFINDEX
, &ifr
) == -1) {
658 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
659 "ioctl: %s", pcap_strerror(errno
));
663 return ifr
.ifr_ifindex
;
667 * Bind the socket associated with FD to the given device.
670 iface_bind(int fd
, int ifindex
, char *ebuf
)
672 struct sockaddr_ll sll
;
674 memset(&sll
, 0, sizeof(sll
));
675 sll
.sll_family
= AF_PACKET
;
676 sll
.sll_ifindex
= ifindex
;
677 sll
.sll_protocol
= htons(ETH_P_ALL
);
679 if (bind(fd
, (struct sockaddr
*) &sll
, sizeof(sll
)) == -1) {
680 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
681 "bind: %s", pcap_strerror(errno
));
691 /* ===== Functions to interface to the older kernels ================== */
694 * With older kernels promiscuous mode is kind of interesting because we
695 * have to reset the interface before exiting. The problem can't really
696 * be solved without some daemon taking care of managing usage counts.
697 * We save the promiscuous state of the device when opening the capture
698 * stream and arrange for it to be reset on process exit.
700 * XXX: This solution is still not correct even for this case. The
701 * devices stay in promiscuous mode until the process exits. I need to
702 * modify pcap_close to solve this.
706 * The device name and the interface flags to be restored at exit
708 struct ifreq restore_ifr
;
710 static void restore_interface( void )
712 int status
= socket(PF_INET
, SOCK_PACKET
, 0);
715 status
= ioctl(status
, SIOCSIFFLAGS
, &restore_ifr
);
719 "Can't restore interface flags. Please adjust manually. \n"
720 "Hint: This can't happen with Linux >= 2.2.0.\n");
725 * Try to open a packet socket using the old kernel interface.
726 * Returns 0 on failure.
727 * FIXME: 0 uses to mean success (Sebastian)
730 live_open_old(pcap_t
*handle
, char *device
, int promisc
,
731 int to_ms
, char *ebuf
)
733 int sock_fd
= -1, mtu
, arptype
;
737 /* Open the socket */
739 sock_fd
= socket(PF_INET
, SOCK_PACKET
, htons(ETH_P_ALL
));
741 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
742 "socket: %s", pcap_strerror(errno
));
746 /* It worked - we are using the old interface */
747 handle
->md
.sock_packet
= 1;
749 /* Bind to the given device */
752 strncpy(ebuf
, "pcap_open_live: No interface given",
756 if (iface_bind_old(sock_fd
, device
, ebuf
) == -1)
759 /* Go to promisc mode */
761 memset(&ifr
, 0, sizeof(ifr
));
762 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
763 if (ioctl(sock_fd
, SIOCGIFFLAGS
, &ifr
) == -1) {
764 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
765 "ioctl: %s", pcap_strerror(errno
));
768 if ((ifr
.ifr_flags
& IFF_PROMISC
) == 0) {
770 ifr
.ifr_flags
|= IFF_PROMISC
;
771 if (ioctl(sock_fd
, SIOCSIFFLAGS
, &ifr
) == -1) {
772 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
774 pcap_strerror(errno
));
777 if (atexit(restore_interface
) == -1) {
779 strncpy(ebuf
, "atexit failed",
787 /* Compute the buffersize */
789 mtu
= iface_get_mtu(sock_fd
, device
, ebuf
);
792 handle
->bufsize
= MAX_LINKHEADER_SIZE
+ mtu
;
793 if (handle
->bufsize
< handle
->snapshot
)
794 handle
->bufsize
= handle
->snapshot
;
796 /* All done - fill in the pcap handle */
798 arptype
= iface_get_arptype(sock_fd
, device
, ebuf
);
802 handle
->fd
= sock_fd
;
804 handle
->linktype
= map_arphrd_to_dlt(arptype
);
805 if (handle
->linktype
== -1) {
806 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
807 "interface type of %s not supported", device
);
810 handle
->buffer
= malloc(handle
->bufsize
);
811 if (!handle
->buffer
) {
812 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
813 "malloc: %s", pcap_strerror(errno
));
827 * Bind the socket associated with FD to the given device using the
828 * interface of the old kernels.
831 iface_bind_old(int fd
, const char *device
, char *ebuf
)
833 struct sockaddr saddr
;
835 memset(&saddr
, 0, sizeof(saddr
));
836 strncpy(saddr
.sa_data
, device
, sizeof(saddr
.sa_data
));
837 if (bind(fd
, &saddr
, sizeof(saddr
)) == -1) {
838 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
839 "bind: %s", pcap_strerror(errno
));
847 /* ===== System calls available on all supported kernels ============== */
850 * Query the kernel for the MTU of the given interface.
853 iface_get_mtu(int fd
, const char *device
, char *ebuf
)
858 return BIGGER_THAN_ALL_MTUS
;
860 memset(&ifr
, 0, sizeof(ifr
));
861 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
863 if (ioctl(fd
, SIOCGIFMTU
, &ifr
) == -1) {
864 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
865 "ioctl: %s", pcap_strerror(errno
));
873 * Get the hardware type of the given interface as ARPHRD_xxx constant.
876 iface_get_arptype(int fd
, const char *device
, char *ebuf
)
880 memset(&ifr
, 0, sizeof(ifr
));
881 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
883 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) == -1) {
884 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
885 "ioctl: %s", pcap_strerror(errno
));
889 return ifr
.ifr_hwaddr
.sa_family
;