]>
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 1.33 2000-10-18 23:47:41 torsten 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 static int iface_get_id(int fd
, const char *device
, char *ebuf
);
93 static int iface_get_mtu(int fd
, const char *device
, char *ebuf
);
94 static int iface_get_arptype(int fd
, const char *device
, char *ebuf
);
95 static int iface_bind(int fd
, int ifindex
, char *ebuf
);
96 static int iface_bind_old(int fd
, const char *device
, char *ebuf
);
99 * Get a handle for a live capture from the given device. You can
100 * pass NULL as device to get all packages (without link level
101 * information of course). If you pass 1 as promisc the interface
102 * will be set to promiscous mode (XXX: I think this usage should
103 * be deprecated and functions be added to select that later allow
104 * modification of that values -- Torsten).
109 pcap_open_live(char *device
, int snaplen
, int promisc
, int to_ms
, char *ebuf
)
111 /* Allocate a handle for this session. */
113 pcap_t
*handle
= malloc(sizeof(*handle
));
114 if (handle
== NULL
) {
115 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "malloc: %s",
116 pcap_strerror(errno
));
120 /* Initialize some components of the pcap structure. */
122 memset(handle
, 0, sizeof(*handle
));
123 handle
->snapshot
= snaplen
;
124 handle
->md
.timeout
= to_ms
;
125 handle
->md
.promisc
= promisc
;
128 * NULL and "any" are special devices which give us the hint to
129 * monitor all devices.
131 if (!device
|| strcmp(device
, "any") == 0) {
133 handle
->md
.device
= strdup("any");
135 handle
->md
.device
= strdup(device
);
137 if (handle
->md
.device
== NULL
) {
138 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "strdup: %s",
139 pcap_strerror(errno
) );
145 * Current Linux kernels use the protocol family PF_PACKET to
146 * allow direct access to all packets on the network while
147 * older kernels had a special socket type SOCK_PACKET to
148 * implement this feature.
149 * While this old implementation is kind of obsolete we need
150 * to be compatible with older kernels for a while so we are
151 * trying both methods with the newer method preferred.
154 if (! (live_open_new(handle
, device
, promisc
, to_ms
, ebuf
) ||
155 live_open_old(handle
, device
, promisc
, to_ms
, ebuf
)) )
158 * Both methods to open the packet socket failed. Tidy
159 * up and report our failure (ebuf is expected to be
160 * set by the functions above).
163 free(handle
->md
.device
);
169 * Okay, now we have a packet stream open. Maybe we need to handle
170 * a timeout? In that case we set the filehandle to nonblocking
171 * so pcap_read can try reading the fd and call select if no data
172 * is available at first.
176 int flags
= fcntl(handle
->fd
, F_GETFL
);
179 flags
= fcntl(handle
->fd
, F_SETFL
, flags
);
182 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "fcntl: %s",
183 pcap_strerror(errno
));
193 * Read at most max_packets from the capture stream and call the callback
194 * for each of them. Returns the number of packets handled or -1 if an
197 * XXX: Can I rely on the Linux-specified behaviour of select (returning
198 * the time left in the timeval structure)? I really don't want to query
199 * the system time before each select call...
201 * pcap_read currently gets not only a packet from the kernel but also
202 * the sockaddr_ll returned as source of the packet. This way we can at
203 * some time extend tcpdump and libpcap to sniff on all devices at a time
204 * and find the right printing routine by using the information in the
205 * sockaddr_ll structure.
208 pcap_read(pcap_t
*handle
, int max_packets
, pcap_handler callback
, u_char
*user
)
215 * Fill in a timeval structure for select if we need to obeye a
218 if (handle
->md
.timeout
> 0) {
219 tv
.tv_usec
= (handle
->md
.timeout
% 1000) * 1000;
220 tv
.tv_sec
= (handle
->md
.timeout
/ 1000);
224 * Read packets until the packet limit has been reached or
225 * an error occured while reading. Call the user function
226 * for each received packet.
228 for (packets
= 0; max_packets
== -1 || packets
< max_packets
;)
230 status
= pcap_read_packet(handle
, callback
, user
);
235 } else if (status
== -1)
239 * If no packet is available we go to sleep. FIXME: This
240 * might be better implemented using poll(?)
243 FD_SET(handle
->fd
, &read_fds
);
245 status
= select(handle
->fd
+ 1,
246 &read_fds
, NULL
, NULL
, &tv
);
247 } while (status
== -1 && errno
== EINTR
);
250 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
251 "select: %s", pcap_strerror(errno
));
253 } else if (status
== 0 ||
254 (tv
.tv_usec
== 0 && tv
.tv_sec
== 0))
262 * Read a packet from the socket calling the handler provided by
263 * the user. Returns the number of packets received or -1 if an
267 pcap_read_packet(pcap_t
*handle
, pcap_handler callback
, u_char
*userdata
)
269 struct sockaddr from
;
271 int packet_len
, caplen
;
272 struct pcap_pkthdr pcap_header
;
275 * We don't currently use the from return value of recvfrom but
276 * this will probably be implemented in the future.
279 /* Receive a single packet from the kernel */
282 fromlen
= sizeof(from
);
283 packet_len
= recvfrom(
284 handle
->fd
, handle
->buffer
+ handle
->offset
,
285 handle
->snapshot
, MSG_TRUNC
,
286 (struct sockaddr
*) &from
, &fromlen
);
287 } while (packet_len
== -1 && errno
== EINTR
);
289 /* Check if an error occured */
291 if (packet_len
== -1) {
293 return 0; /* no packet there */
295 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
296 "recvfrom: %s", pcap_strerror(errno
));
302 * XXX: According to the kernel source we should get the real
303 * packet len if calling recvfrom with MSG_TRUNC set. It does
304 * not seem to work here :(, but it is supported by this code
306 * To be honest the code RELIES on that feature so this is really
307 * broken with 2.2.x kernels.
308 * I spend a day to figure out what's going on and I found out
309 * that the following is happening:
311 * The packet comes from a random interface and the packet_rcv
312 * hook is called with a clone of the packet. That code inserts
313 * the packet into the receive queue of the packet socket.
314 * If a filter is attached to that socket that filter is run
315 * first - and there lies the problem. The default filter always
316 * cuts the packet at the snaplen:
321 * So the packet filter cuts down the packet. The recvfrom call
322 * says "hey, it's only 68 bytes, it fits into the buffer" with
323 * the result that we don't get the real packet length. This
324 * is valid at least until kernel 2.2.17pre6.
326 * tcpdump is currently fixed by changing the BPF code generator
327 * to not truncate the received packet.
331 if (caplen
> handle
->snapshot
)
332 caplen
= handle
->snapshot
;
334 /* Run the packet filter if not using kernel filter */
335 if (!handle
->md
.use_bpf
&& handle
->fcode
.bf_insns
) {
336 if (bpf_filter(handle
->fcode
.bf_insns
, handle
->buffer
,
337 packet_len
, caplen
) == 0)
339 /* rejected by filter */
344 /* Fill in our own header data */
346 if (ioctl(handle
->fd
, SIOCGSTAMP
, &pcap_header
.ts
) == -1) {
347 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
348 "ioctl: %s", pcap_strerror(errno
));
351 pcap_header
.caplen
= caplen
;
352 pcap_header
.len
= packet_len
;
354 /* Call the user supplied callback function */
355 handle
->md
.stat
.ps_recv
++;
356 callback(userdata
, &pcap_header
, handle
->buffer
+ handle
->offset
);
362 * Get the statistics for the given packet capture handle.
363 * FIXME: Currently does not report the number of dropped packets.
366 pcap_stats(pcap_t
*handle
, struct pcap_stat
*stats
)
368 *stats
= handle
->md
.stat
;
373 * Attach the given BPF code to the packet capture device.
376 pcap_setfilter(pcap_t
*handle
, struct bpf_program
*filter
)
378 #ifdef SO_ATTACH_FILTER
379 struct sock_fprog fcode
;
385 strncpy(handle
->errbuf
, "setfilter: No filter specified",
386 sizeof(handle
->errbuf
));
390 /* Free old filter code if existing */
392 handle
->fcode
.bf_len
= 0;
393 if (handle
->fcode
.bf_insns
) {
394 free(handle
->fcode
.bf_insns
);
395 handle
->fcode
.bf_insns
= NULL
;
399 /* Make our private copy of the filter */
401 handle
->fcode
.bf_len
= filter
->bf_len
;
402 handle
->fcode
.bf_insns
=
403 malloc(filter
->bf_len
* sizeof(*filter
->bf_insns
));
404 if (handle
->fcode
.bf_insns
== NULL
) {
405 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
406 "malloc: %s", pcap_strerror(errno
));
409 memcpy(handle
->fcode
.bf_insns
, filter
->bf_insns
,
410 filter
->bf_len
* sizeof(*filter
->bf_insns
));
413 * Run user level packet filter by default. Will be overriden if
414 * installing a kernel filter succeeds.
416 handle
->md
.use_bpf
= 0;
418 /* Install kernel level filter if possible */
420 #ifdef SO_ATTACH_FILTER
422 * Oh joy, the Linux kernel uses struct sock_fprog instead of
423 * struct bpf_program and of course the length field is of
424 * different size. Pointed out by Sebastian
427 fcode
.filter
= (struct sock_filter
*) handle
->fcode
.bf_insns
;
428 fcode
.len
= filter
->bf_len
;
430 if (filter
->bf_len
> USHRT_MAX
) {
432 * fcode.len is an unsigned short for current kernel.
433 * I have yet to see BPF-Code with that much instructions
434 * but still it is possible. So for the sake of
435 * correctness I added this check.
437 fprintf(stderr
, "Warning: Filter to complex for kernel\n");
441 if (setsockopt(handle
->fd
, SOL_SOCKET
, SO_ATTACH_FILTER
,
442 &fcode
, sizeof(fcode
)) == 0)
444 /* Installation succeded - using kernel filter. */
445 handle
->md
.use_bpf
= 1;
450 * Print a warning if kernel filter available but a problem
453 if (errno
!= ENOPROTOOPT
&& errno
!= EOPNOTSUPP
) {
454 fprintf(stderr
, "Warning: Kernel filter failed: %s\n",
455 pcap_strerror(errno
));
464 * Linux uses the ARP hardware type to identify the type of an
465 * interface. pcap uses the DLT_xxx constants for this. This
466 * function maps the ARPHRD_xxx constant to an appropriate
469 * Returns -1 if unable to map the type.
471 static int map_arphrd_to_dlt(int arptype
)
475 case ARPHRD_METRICOM
:
476 case ARPHRD_LOOPBACK
: return DLT_EN10MB
;
477 case ARPHRD_EETHER
: return DLT_EN3MB
;
478 case ARPHRD_AX25
: return DLT_AX25
;
479 case ARPHRD_PRONET
: return DLT_PRONET
;
480 case ARPHRD_CHAOS
: return DLT_CHAOS
;
481 case ARPHRD_IEEE802
: return DLT_IEEE802
;
482 case ARPHRD_ARCNET
: return DLT_ARCNET
;
483 case ARPHRD_FDDI
: return DLT_FDDI
;
485 #ifndef ARPHRD_ATM /* FIXME: How to #include this? */
486 #define ARPHRD_ATM 19
488 case ARPHRD_ATM
: return DLT_ATM_CLIP
;
494 case ARPHRD_SLIP
: return DLT_RAW
;
500 /* ===== Functions to interface to the newer kernels ================== */
503 * Try to open a packet socket using the new kernel interface.
504 * Returns 0 on failure.
505 * FIXME: 0 uses to mean success (Sebastian)
508 live_open_new(pcap_t
*handle
, char *device
, int promisc
,
509 int to_ms
, char *ebuf
)
511 #ifdef HAVE_NETPACKET_PACKET_H
512 int sock_fd
= -1, device_id
, mtu
, arptype
;
513 struct packet_mreq mr
;
515 /* One shot loop used for error handling - bail out with break */
519 * Open a socket with protocol family packet. If a device is
520 * given we try to open it in raw mode otherwise we use
521 * the cooked interface.
524 socket(PF_PACKET
, SOCK_RAW
, htons(ETH_P_ALL
))
525 : socket(PF_PACKET
, SOCK_DGRAM
, htons(ETH_P_ALL
));
528 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "socket: %s",
529 pcap_strerror(errno
) );
533 /* It seems the kernel supports the new interface. */
534 handle
->md
.sock_packet
= 0;
537 * What kind of frames do we have to deal with? Fall back
538 * to cooked mode if we have an unknown interface type.
542 arptype
= iface_get_arptype(sock_fd
, device
, ebuf
);
545 handle
->linktype
= map_arphrd_to_dlt(arptype
);
547 handle
->linktype
= DLT_RAW
;
549 if (handle
->linktype
== -1) {
550 /* Unknown interface type - reopen in cooked mode */
552 if (close(sock_fd
) == -1) {
553 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
554 "close: %s", pcap_strerror(errno
));
557 sock_fd
= socket(PF_PACKET
, SOCK_DGRAM
,
560 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
561 "socket: %s", pcap_strerror(errno
));
566 "Warning: Falling back to cooked socket\n");
567 handle
->linktype
= DLT_RAW
;
572 device_id
= iface_get_id(sock_fd
, device
, ebuf
);
576 if (iface_bind(sock_fd
, device_id
, ebuf
) == -1)
580 /* Select promiscous mode on/off */
584 * Hmm, how can we set promiscuous mode on all interfaces?
585 * I am not sure if that is possible at all.
589 memset(&mr
, 0, sizeof(mr
));
590 mr
.mr_ifindex
= device_id
;
591 mr
.mr_type
= promisc
?
592 PACKET_MR_PROMISC
: PACKET_MR_ALLMULTI
;
593 if (setsockopt(sock_fd
, SOL_PACKET
,
594 PACKET_ADD_MEMBERSHIP
, &mr
, sizeof(mr
)) == -1)
596 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
597 "setsockopt: %s", pcap_strerror(errno
));
603 /* Compute the buffersize */
605 mtu
= iface_get_mtu(sock_fd
, device
, ebuf
);
608 handle
->bufsize
= MAX_LINKHEADER_SIZE
+ mtu
;
610 /* Fill in the pcap structure */
612 handle
->fd
= sock_fd
;
615 handle
->buffer
= malloc(handle
->bufsize
);
616 if (!handle
->buffer
) {
617 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
618 "malloc: %s", pcap_strerror(errno
));
631 "New packet capturing interface not supported by build "
632 "environment", PCAP_ERRBUF_SIZE
);
637 #ifdef HAVE_NETPACKET_PACKET_H
639 * Return the index of the given device name. Fill ebuf and return
643 iface_get_id(int fd
, const char *device
, char *ebuf
)
647 memset(&ifr
, 0, sizeof(ifr
));
648 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
650 if (ioctl(fd
, SIOCGIFINDEX
, &ifr
) == -1) {
651 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
652 "ioctl: %s", pcap_strerror(errno
));
656 return ifr
.ifr_ifindex
;
660 * Bind the socket associated with FD to the given device.
663 iface_bind(int fd
, int ifindex
, char *ebuf
)
665 struct sockaddr_ll sll
;
667 memset(&sll
, 0, sizeof(sll
));
668 sll
.sll_family
= AF_PACKET
;
669 sll
.sll_ifindex
= ifindex
;
670 sll
.sll_protocol
= htons(ETH_P_ALL
);
672 if (bind(fd
, (struct sockaddr
*) &sll
, sizeof(sll
)) == -1) {
673 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
674 "bind: %s", pcap_strerror(errno
));
684 /* ===== Functions to interface to the older kernels ================== */
687 * With older kernels promiscuous mode is kind of interesting because we
688 * have to reset the interface before exiting. The problem can't really
689 * be solved without some daemon taking care of managing usage counts.
690 * We save the promiscuous state of the device when opening the capture
691 * stream and arrange for it to be reset on process exit.
693 * XXX: This solution is still not correct even for this case. The
694 * devices stay in promiscuous mode until the process exits. I need to
695 * modify pcap_close to solve this.
699 * The device name and the interface flags to be restored at exit
701 struct ifreq restore_ifr
;
703 static void restore_interface( void )
705 int status
= socket(PF_INET
, SOCK_PACKET
, 0);
708 status
= ioctl(status
, SIOCSIFFLAGS
, &restore_ifr
);
712 "Can't restore interface flags. Please adjust manually. \n"
713 "Hint: This can't happen with Linux >= 2.2.0.\n");
718 * Try to open a packet socket using the old kernel interface.
719 * Returns 0 on failure.
720 * FIXME: 0 uses to mean success (Sebastian)
723 live_open_old(pcap_t
*handle
, char *device
, int promisc
,
724 int to_ms
, char *ebuf
)
726 int sock_fd
= -1, mtu
, arptype
;
730 /* Open the socket */
732 sock_fd
= socket(PF_INET
, SOCK_PACKET
, htons(ETH_P_ALL
));
734 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
735 "socket: %s", pcap_strerror(errno
));
739 /* It worked - we are using the old interface */
740 handle
->md
.sock_packet
= 1;
742 /* Bind to the given device */
745 strncpy(ebuf
, "pcap_open_live: No interface given",
749 if (iface_bind_old(sock_fd
, device
, ebuf
) == -1)
752 /* Go to promisc mode */
754 memset(&ifr
, 0, sizeof(ifr
));
755 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
756 if (ioctl(sock_fd
, SIOCGIFFLAGS
, &ifr
) == -1) {
757 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
758 "ioctl: %s", pcap_strerror(errno
));
761 if ((ifr
.ifr_flags
& IFF_PROMISC
) == 0) {
763 ifr
.ifr_flags
|= IFF_PROMISC
;
764 if (ioctl(sock_fd
, SIOCSIFFLAGS
, &ifr
) == -1) {
765 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
767 pcap_strerror(errno
));
770 if (atexit(restore_interface
) == -1) {
772 strncpy(ebuf
, "atexit failed",
780 /* Compute the buffersize */
782 mtu
= iface_get_mtu(sock_fd
, device
, ebuf
);
785 handle
->bufsize
= MAX_LINKHEADER_SIZE
+ mtu
;
787 /* All done - fill in the pcap handle */
789 arptype
= iface_get_arptype(sock_fd
, device
, ebuf
);
793 handle
->fd
= sock_fd
;
795 handle
->linktype
= map_arphrd_to_dlt(arptype
);
796 if (handle
->linktype
== -1) {
797 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
798 "interface type of %s not supported", device
);
801 handle
->buffer
= malloc(handle
->bufsize
);
802 if (!handle
->buffer
) {
803 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
804 "malloc: %s", pcap_strerror(errno
));
818 * Bind the socket associated with FD to the given device using the
819 * interface of the old kernels.
822 iface_bind_old(int fd
, const char *device
, char *ebuf
)
824 struct sockaddr saddr
;
826 memset(&saddr
, 0, sizeof(saddr
));
827 strncpy(saddr
.sa_data
, device
, sizeof(saddr
.sa_data
));
828 if (bind(fd
, &saddr
, sizeof(saddr
)) == -1) {
829 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
830 "bind: %s", pcap_strerror(errno
));
838 /* ===== System calls available on all supported kernels ============== */
841 * Query the kernel for the MTU of the given interface.
844 iface_get_mtu(int fd
, const char *device
, char *ebuf
)
849 return BIGGER_THAN_ALL_MTUS
;
851 memset(&ifr
, 0, sizeof(ifr
));
852 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
854 if (ioctl(fd
, SIOCGIFMTU
, &ifr
) == -1) {
855 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
856 "ioctl: %s", pcap_strerror(errno
));
864 * Get the hardware type of the given interface as ARPHRD_xxx constant.
867 iface_get_arptype(int fd
, const char *device
, char *ebuf
)
871 memset(&ifr
, 0, sizeof(ifr
));
872 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
874 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) == -1) {
875 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
876 "ioctl: %s", pcap_strerror(errno
));
880 return ifr
.ifr_hwaddr
.sa_family
;