]>
The Tcpdump Group git mirrors - libpcap/blob - pcap-linux.c
4bb2496319ac2e71783a0e7fd516da828e199a84
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 1.29 2000-09-20 07:52:04 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 * Prototypes for internal functions
77 static int map_arphrd_to_dlt(int arptype
);
78 static int live_open_old(pcap_t
*, char *, int, int, char *);
79 static int live_open_new(pcap_t
*, char *, int, int, char *);
80 static int pcap_read_packet(pcap_t
*, pcap_handler
, u_char
*);
83 * Wrap some ioctl calls
85 static int iface_get_id(int fd
, const char *device
, char *ebuf
);
86 static int iface_get_mtu(int fd
, const char *device
, char *ebuf
);
87 static int iface_get_arptype(int fd
, const char *device
, char *ebuf
);
88 static int iface_bind(int fd
, int ifindex
, char *ebuf
);
89 static int iface_bind_old(int fd
, const char *device
, char *ebuf
);
92 * Get a handle for a live capture from the given device. You can
93 * pass NULL as device to get all packages (without link level
94 * information of course). If you pass 1 as promisc the interface
95 * will be set to promiscous mode (XXX: I think this usage should
96 * be deprecated and functions be added to select that later allow
97 * modification of that values -- Torsten).
102 pcap_open_live(char *device
, int snaplen
, int promisc
, int to_ms
, char *ebuf
)
104 /* Allocate a handle for this session. */
106 pcap_t
*handle
= malloc(sizeof(*handle
));
107 if (handle
== NULL
) {
108 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "malloc: %s",
109 pcap_strerror(errno
));
113 /* Initialize some components of the pcap structure. */
115 memset(handle
, 0, sizeof(*handle
));
116 handle
->snapshot
= snaplen
;
117 handle
->md
.timeout
= to_ms
;
118 handle
->md
.promisc
= promisc
;
119 handle
->md
.device
= strdup(device
);
120 if (handle
->md
.device
== NULL
) {
121 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "strdup: %s",
122 pcap_strerror(errno
) );
128 * Current Linux kernels use the protocol family PF_PACKET to
129 * allow direct access to all packets on the network while
130 * older kernels had a special socket type SOCK_PACKET to
131 * implement this feature.
132 * While this old implementation is kind of obsolete we need
133 * to be compatible with older kernels for a while so we are
134 * trying both methods with the newer method preferred.
137 if (! (live_open_new(handle
, device
, promisc
, to_ms
, ebuf
) ||
138 live_open_old(handle
, device
, promisc
, to_ms
, ebuf
)) )
141 * Both methods to open the packet socket failed. Tidy
142 * up and report our failure (ebuf is expected to be
143 * set by the functions above).
146 free(handle
->md
.device
);
152 * Okay, now we have a packet stream open. Maybe we need to handle
153 * a timeout? In that case we set the filehandle to nonblocking
154 * so pcap_read can try reading the fd and call select if no data
155 * is available at first.
159 int flags
= fcntl(handle
->fd
, F_GETFL
);
162 flags
= fcntl(handle
->fd
, F_SETFL
, flags
);
165 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "fcntl: %s",
166 pcap_strerror(errno
));
176 * Read at most max_packets from the capture stream and call the callback
177 * for each of them. Returns the number of packets handled or -1 if an
180 * XXX: Can I rely on the Linux-specified behaviour of select (returning
181 * the time left in the timeval structure)? I really don't want to query
182 * the system time before each select call...
184 * pcap_read currently gets not only a packet from the kernel but also
185 * the sockaddr_ll returned as source of the packet. This way we can at
186 * some time extend tcpdump and libpcap to sniff on all devices at a time
187 * and find the right printing routine by using the information in the
188 * sockaddr_ll structure.
191 pcap_read(pcap_t
*handle
, int max_packets
, pcap_handler callback
, u_char
*user
)
198 * Fill in a timeval structure for select if we need to obeye a
201 if (handle
->md
.timeout
> 0) {
202 tv
.tv_usec
= (handle
->md
.timeout
% 1000) * 1000;
203 tv
.tv_sec
= (handle
->md
.timeout
/ 1000);
207 * Read packets until the packet limit has been reached or
208 * an error occured while reading. Call the user function
209 * for each received packet.
211 for (packets
= 0; max_packets
== -1 || packets
< max_packets
;)
213 status
= pcap_read_packet(handle
, callback
, user
);
218 } else if (status
== -1)
222 * If no packet is available we go to sleep. FIXME: This
223 * might be better implemented using poll(?)
226 FD_SET(handle
->fd
, &read_fds
);
227 status
= select(handle
->fd
+ 1,
228 &read_fds
, NULL
, NULL
, &tv
);
230 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
231 "select: %s", pcap_strerror(errno
));
233 } else if (status
== 0 ||
234 (tv
.tv_usec
== 0 && tv
.tv_sec
== 0))
242 * Read a packet from the socket calling the handler provided by
243 * the user. Returns the number of packets received or -1 if an
247 pcap_read_packet(pcap_t
*handle
, pcap_handler callback
, u_char
*userdata
)
249 struct sockaddr from
;
251 int packet_len
, caplen
;
252 struct pcap_pkthdr pcap_header
;
255 * We don't currently use the from return value of recvfrom but
256 * this will probably be implemented in the future.
259 /* Receive a single packet from the kernel */
262 fromlen
= sizeof(from
);
263 packet_len
= recvfrom(
264 handle
->fd
, handle
->buffer
+ handle
->offset
,
265 handle
->snapshot
, MSG_TRUNC
,
266 (struct sockaddr
*) &from
, &fromlen
);
267 } while (packet_len
== -1 && errno
== EINTR
);
269 /* Check if an error occured */
271 if (packet_len
== -1) {
273 return 0; /* no packet there */
275 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
276 "recvfrom: %s", pcap_strerror(errno
));
282 * XXX: According to the kernel source we should get the real
283 * packet len if calling recvfrom with MSG_TRUNC set. It does
284 * not seem to work here :(, but it is supported by this code
286 * To be honest the code RELIES on that feature so this is really
287 * broken with 2.2.x kernels.
288 * I spend a day to figure out what's going on and I found out
289 * that the following is happening:
291 * The packet comes from a random interface and the packet_rcv
292 * hook is called with a clone of the packet. That code inserts
293 * the packet into the receive queue of the packet socket.
294 * If a filter is attached to that socket that filter is run
295 * first - and there lies the problem. The default filter always
296 * cuts the packet at the snaplen:
301 * So the packet filter cuts down the packet. The recvfrom call
302 * says "hey, it's only 68 bytes, it fits into the buffer" with
303 * the result that we don't get the real packet length. This
304 * is valid at least until kernel 2.2.17pre6.
306 * tcpdump is currently fixed by changing the BPF code generator
307 * to not truncate the received packet.
311 if (caplen
> handle
->snapshot
)
312 caplen
= handle
->snapshot
;
314 /* Run the packet filter if not using kernel filter */
315 if (!handle
->md
.use_bpf
&& handle
->fcode
.bf_insns
) {
316 if (bpf_filter(handle
->fcode
.bf_insns
, handle
->buffer
,
317 packet_len
, caplen
) == 0)
319 /* rejected by filter */
324 /* Fill in our own header data */
326 if (ioctl(handle
->fd
, SIOCGSTAMP
, &pcap_header
.ts
) == -1) {
327 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
328 "ioctl: %s", pcap_strerror(errno
));
331 pcap_header
.caplen
= caplen
;
332 pcap_header
.len
= packet_len
;
334 /* Call the user supplied callback function */
335 handle
->md
.stat
.ps_recv
++;
336 callback(userdata
, &pcap_header
, handle
->buffer
+ handle
->offset
);
342 * Get the statistics for the given packet capture handle.
343 * FIXME: Currently does not report the number of dropped packets.
346 pcap_stats(pcap_t
*handle
, struct pcap_stat
*stats
)
348 *stats
= handle
->md
.stat
;
353 * Attach the given BPF code to the packet capture device.
356 pcap_setfilter(pcap_t
*handle
, struct bpf_program
*filter
)
358 #ifdef SO_ATTACH_FILTER
359 struct sock_fprog fcode
;
365 strncpy(handle
->errbuf
, "setfilter: No filter specified",
366 sizeof(handle
->errbuf
));
370 /* Free old filter code if existing */
372 handle
->fcode
.bf_len
= 0;
373 if (handle
->fcode
.bf_insns
) {
374 free(handle
->fcode
.bf_insns
);
375 handle
->fcode
.bf_insns
= NULL
;
379 /* Make our private copy of the filter */
381 handle
->fcode
.bf_len
= filter
->bf_len
;
382 handle
->fcode
.bf_insns
=
383 malloc(filter
->bf_len
* sizeof(*filter
->bf_insns
));
384 if (handle
->fcode
.bf_insns
== NULL
) {
385 snprintf(handle
->errbuf
, sizeof(handle
->errbuf
),
386 "malloc: %s", pcap_strerror(errno
));
389 memcpy(handle
->fcode
.bf_insns
, filter
->bf_insns
,
390 filter
->bf_len
* sizeof(*filter
->bf_insns
));
393 * Run user level packet filter by default. Will be overriden if
394 * installing a kernel filter succeeds.
396 handle
->md
.use_bpf
= 0;
398 /* Install kernel level filter if possible */
400 #ifdef SO_ATTACH_FILTER
402 * Oh joy, the Linux kernel uses struct sock_fprog instead of
403 * struct bpf_program and of course the length field is of
404 * different size. Pointed out by Sebastian
407 fcode
.filter
= (struct sock_filter
*) handle
->fcode
.bf_insns
;
408 fcode
.len
= filter
->bf_len
;
410 if (filter
->bf_len
> USHRT_MAX
) {
412 * fcode.len is an unsigned short for current kernel.
413 * I have yet to see BPF-Code with that much instructions
414 * but still it is possible. So for the sake of
415 * correctness I added this check.
417 fprintf(stderr
, "Warning: Filter to complex for kernel\n");
421 if (setsockopt(handle
->fd
, SOL_SOCKET
, SO_ATTACH_FILTER
,
422 &fcode
, sizeof(fcode
)) == 0)
424 /* Installation succeded - using kernel filter. */
425 handle
->md
.use_bpf
= 1;
430 * Print a warning if kernel filter available but a problem
433 if (errno
!= ENOPROTOOPT
&& errno
!= EOPNOTSUPP
) {
434 fprintf(stderr
, "Warning: Kernel filter failed: %s\n",
435 pcap_strerror(errno
));
444 * Linux uses the ARP hardware type to identify the type of an
445 * interface. pcap uses the PCAP_ENCAP_xxx constants for this. This
446 * function maps the ARPHRD_xxx constant to an appropriate
447 * PCAP_ENCAP__xxx constant.
449 * Returns -1 if unable to map the type.
451 static int map_arphrd_to_dlt(int arptype
)
455 case ARPHRD_METRICOM
:
456 case ARPHRD_LOOPBACK
:
457 return PCAP_ENCAP_ETHERNET
;
460 return PCAP_ENCAP_EXP_ETHERNET
;
463 return PCAP_ENCAP_AX25
;
466 return PCAP_ENCAP_PRONET
;
469 return PCAP_ENCAP_CHAOS
;
472 return PCAP_ENCAP_TOKEN_RING
;
475 return PCAP_ENCAP_ARCNET
;
478 return PCAP_ENCAP_FDDI
;
480 #ifndef ARPHRD_ATM /* FIXME: How to #include this? */
481 #define ARPHRD_ATM 19
484 return PCAP_ENCAP_ATM_CLIP
;
491 return PCAP_ENCAP_RAW
;
497 /* ===== Functions to interface to the newer kernels ================== */
500 * Try to open a packet socket using the new kernel interface.
501 * Returns 0 on failure.
502 * FIXME: 0 uses to mean success (Sebastian)
505 live_open_new(pcap_t
*handle
, char *device
, int promisc
,
506 int to_ms
, char *ebuf
)
508 #ifdef HAVE_NETPACKET_PACKET_H
509 int sock_fd
= -1, device_id
, mtu
, arptype
;
510 struct packet_mreq mr
;
512 /* One shot loop used for error handling - bail out with break */
515 /* Open a socket with protocol family packet. */
516 sock_fd
= socket( PF_PACKET
, SOCK_RAW
, htons(ETH_P_ALL
) );
518 snprintf(ebuf
, PCAP_ERRBUF_SIZE
, "socket: %s",
519 pcap_strerror(errno
) );
523 /* It seems the kernel supports the new interface. */
524 handle
->md
.sock_packet
= 0;
527 * Currently we only support monitoring a single interface.
528 * While the kernel can do more I want to reimplement the
529 * old features first before adding new.
533 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
534 "pcap_open_live: No device given");
539 * What kind of frames do we have to deal with? Fall back
540 * to cooked mode if we have an unknown interface type.
543 arptype
= iface_get_arptype(sock_fd
, device
, ebuf
);
546 handle
->linktype
= map_arphrd_to_dlt(arptype
);
547 if (handle
->linktype
== -1) {
548 /* Unknown interface type - reopen in cooked mode */
550 if (close(sock_fd
) == -1) {
551 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
552 "close: %s", pcap_strerror(errno
));
555 sock_fd
= socket(PF_PACKET
, SOCK_DGRAM
,
558 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
559 "socket: %s", pcap_strerror(errno
));
564 "Warning: Falling back to cooked socket\n");
565 handle
->linktype
= PCAP_ENCAP_RAW
;
569 device_id
= iface_get_id(sock_fd
, device
, ebuf
);
573 if (iface_bind(sock_fd
, device_id
, ebuf
) == -1)
576 /* Select promiscous mode on/off */
580 * XXX: We got reports that this does not work in 2.3.99.
581 * Need to investigate. Using ioctl to switch the promisc
582 * mode at device level costs us most of the benefits of
583 * using the new kernel interface.
584 * UPDATE: I found the bug. The kernel checks mr_alen
585 * even if it is of zero interest for the request. A
586 * random value there made the kernel return EINVAL.
587 * Probably the right solution is to memset the whole
591 memset(&mr
, 0, sizeof(mr
));
592 mr
.mr_ifindex
= device_id
;
593 mr
.mr_type
= promisc
?
594 PACKET_MR_PROMISC
: PACKET_MR_ALLMULTI
;
595 if (setsockopt(sock_fd
, SOL_PACKET
, PACKET_ADD_MEMBERSHIP
,
596 &mr
, sizeof(mr
)) == -1)
598 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
599 "setsockopt: %s", pcap_strerror(errno
));
604 /* Compute the buffersize */
606 mtu
= iface_get_mtu(sock_fd
, device
, ebuf
);
609 handle
->bufsize
= MAX_LINKHEADER_SIZE
+ mtu
;
611 /* Fill in the pcap structure */
613 handle
->fd
= sock_fd
;
616 handle
->buffer
= malloc(handle
->bufsize
);
617 if (!handle
->buffer
) {
618 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
619 "malloc: %s", pcap_strerror(errno
));
632 "New packet capturing interface not supported by build "
633 "environment", PCAP_ERRBUF_SIZE
);
638 #ifdef HAVE_NETPACKET_PACKET_H
640 * Return the index of the given device name. Fill ebuf and return
644 iface_get_id(int fd
, const char *device
, char *ebuf
)
648 memset(&ifr
, 0, sizeof(ifr
));
649 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
651 if (ioctl(fd
, SIOCGIFINDEX
, &ifr
) == -1) {
652 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
653 "ioctl: %s", pcap_strerror(errno
));
657 return ifr
.ifr_ifindex
;
661 * Bind the socket associated with FD to the given device.
664 iface_bind(int fd
, int ifindex
, char *ebuf
)
666 struct sockaddr_ll sll
;
668 memset(&sll
, 0, sizeof(sll
));
669 sll
.sll_family
= AF_PACKET
;
670 sll
.sll_ifindex
= ifindex
;
671 sll
.sll_protocol
= htons(ETH_P_ALL
);
673 if (bind(fd
, (struct sockaddr
*) &sll
, sizeof(sll
)) == -1) {
674 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
675 "bind: %s", pcap_strerror(errno
));
685 /* ===== Functions to interface to the older kernels ================== */
688 * With older kernels promiscuous mode is kind of interesting because we
689 * have to reset the interface before exiting. The problem can't really
690 * be solved without some daemon taking care of managing usage counts.
691 * We save the promiscuous state of the device when opening the capture
692 * stream and arrange for it to be reset on process exit.
694 * XXX: This solution is still not correct even for this case. The
695 * devices stay in promiscuous mode until the process exits. I need to
696 * modify pcap_close to solve this.
700 * The device name and the interface flags to be restored at exit
702 struct ifreq restore_ifr
;
704 static void restore_interface( void )
706 int status
= socket(PF_INET
, SOCK_PACKET
, 0);
709 status
= ioctl(status
, SIOCSIFFLAGS
, &restore_ifr
);
713 "Can't restore interface flags. Please adjust manually. \n"
714 "Hint: This can't happen with Linux >= 2.2.0.\n");
719 * Try to open a packet socket using the old kernel interface.
720 * Returns 0 on failure.
721 * FIXME: 0 uses to mean success (Sebastian)
724 live_open_old(pcap_t
*handle
, char *device
, int promisc
,
725 int to_ms
, char *ebuf
)
727 int sock_fd
= -1, mtu
, arptype
;
731 /* Open the socket */
733 sock_fd
= socket(PF_INET
, SOCK_PACKET
, htons(ETH_P_ALL
));
735 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
736 "socket: %s", pcap_strerror(errno
));
740 /* It worked - we are using the old interface */
741 handle
->md
.sock_packet
= 1;
743 /* Bind to the given device */
746 strncpy(ebuf
, "pcap_open_live: No interface given",
750 if (iface_bind_old(sock_fd
, device
, ebuf
) == -1)
753 /* Go to promisc mode */
755 memset(&ifr
, 0, sizeof(ifr
));
756 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
757 if (ioctl(sock_fd
, SIOCGIFFLAGS
, &ifr
) == -1) {
758 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
759 "ioctl: %s", pcap_strerror(errno
));
762 if ((ifr
.ifr_flags
& IFF_PROMISC
) == 0) {
764 ifr
.ifr_flags
|= IFF_PROMISC
;
765 if (ioctl(sock_fd
, SIOCSIFFLAGS
, &ifr
) == -1) {
766 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
768 pcap_strerror(errno
));
771 if (atexit(restore_interface
) == -1) {
773 strncpy(ebuf
, "atexit failed",
781 /* Compute the buffersize */
783 mtu
= iface_get_mtu(sock_fd
, device
, ebuf
);
786 handle
->bufsize
= MAX_LINKHEADER_SIZE
+ mtu
;
788 /* All done - fill in the pcap handle */
790 arptype
= iface_get_arptype(sock_fd
, device
, ebuf
);
794 handle
->fd
= sock_fd
;
796 handle
->linktype
= map_arphrd_to_dlt(arptype
);
797 if(handle
->linktype
== -1) {
798 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
799 "interface type of %s not supported", device
);
802 handle
->buffer
= malloc(handle
->bufsize
);
803 if (!handle
->buffer
) {
804 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
805 "malloc: %s", pcap_strerror(errno
));
819 * Bind the socket associated with FD to the given device using the
820 * interface of the old kernels.
823 iface_bind_old(int fd
, const char *device
, char *ebuf
)
825 struct sockaddr saddr
;
827 memset(&saddr
, 0, sizeof(saddr
));
828 strncpy(saddr
.sa_data
, device
, sizeof(saddr
.sa_data
));
829 if (bind(fd
, &saddr
, sizeof(saddr
)) == -1) {
830 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
831 "bind: %s", pcap_strerror(errno
));
839 /* ===== System calls available on all supported kernels ============== */
842 * Query the kernel for the MTU of the given interface.
845 iface_get_mtu(int fd
, const char *device
, char *ebuf
)
849 memset(&ifr
, 0, sizeof(ifr
));
850 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
852 if (ioctl(fd
, SIOCGIFMTU
, &ifr
) == -1) {
853 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
854 "ioctl: %s", pcap_strerror(errno
));
862 * Get the hardware type of the given interface as ARPHRD_xxx constant.
865 iface_get_arptype(int fd
, const char *device
, char *ebuf
)
869 memset(&ifr
, 0, sizeof(ifr
));
870 strncpy(ifr
.ifr_name
, device
, sizeof(ifr
.ifr_name
));
872 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) == -1) {
873 snprintf(ebuf
, PCAP_ERRBUF_SIZE
,
874 "ioctl: %s", pcap_strerror(errno
));
878 return ifr
.ifr_hwaddr
.sa_family
;