[Guy Harris <guy@xxxxxxxxxxxx>]

On Oct 16, 2009, at 6:10 PM, Rayne wrote:

> I noticed that every pcap file, even the empty ones without any  
> packets, contain a 24-byte "header" at the beginning of the file. At  
> least 3 of the bytes vary from file to file, and the rest appears to  
> be the same, at least from the files I've seen. If I were to omit  
> these 24 bytes from the file, Wireshark doesn't recognize the file  
> as a pcap anymore.
>
> So I guess these 24 bytes are to indicate that the file is of  
> libpcap format, but does anyone know what these 24 bytes are in  
> details, i.e. what they represent?

On a machine with an OS that includes libpcap 1.0.0 or later (OS X  
Snow Leopard, in this case, although recent versions of some Linux  
distributions might also have it now, and recent versions of some  
*BSDs might as well)

$ man 5 pcap-savefile
PCAP-SAVEFILE(5)                                              PCAP- 
SAVEFILE(5)



NAME
       pcap-savefile - libpcap savefile format

DESCRIPTION
       NOTE:  applications  and  libraries should, if possible, use  
libpcap to
       read savefiles, rather than having their own code  to  read   
savefiles.
       If,  in the future, a new file format is supported by libpcap,  
applica-
       tions and libraries using libpcap to read savefiles  will  be   
able  to
       read  the new format of savefiles, but applications and  
libraries using
       their own code to read savefiles will have to be changed to  
support the
       new file format.

       ``Savefiles''  read and written by libpcap and applications  
using libp-
       cap start with a per-file header.  The format of  the  per- 
file  header
       is:

              +------------------------------+
              |        Magic number          |
              +--------------+---------------+
              |Major version | Minor version |
              +--------------+---------------+
              |      Time zone offset        |
              +------------------------------+
              |     Time stamp accuracy      |
              +------------------------------+
              |       Snapshot length        |
              +------------------------------+
              |   Link-layer header type     |
              +------------------------------+
       All  fields  in  the  per-file header are in the byte order of  
the host
       writing the file.  The first field in the per-file header is   
a  4-byte
       magic  number,  with the value 0xa1b2c3d4.  The magic number,  
when read
       by a host with the same byte order as the host  that  wrote   
the  file,
       will have the value 0xa1b2c3d4, and, when read by a host with  
the oppo-
       site byte order as the host that wrote the file, will  have   
the  value
       0xd4c3b2a1.  That allows software reading the file to  
determine whether
       the byte order of the host that wrote the file is the same as  
the  byte
       order of the host on which the file is being read, and thus  
whether the
       values in the per-file and per-packet headers need to be  byte- 
swapped.

       Following this are:

              A  2-byte  file format major version number; the  
current version
              number is 2.

              A 2-byte file format minor version number; the   
current  version
              number is 4.

              A 4-byte time zone offset; this is always 0.

              A  4-byte number giving the accuracy of time stamps in  
the file;
              this is always 0.

              A 4-byte number giving the "snapshot  length"  of  the   
capture;
              packets  longer  than  the  snapshot length are  
truncated to the
              snapshot length, so that, if the snapshot length is N,  
only  the
              first  N  bytes of a packet longer than N bytes will be  
saved in
              the capture.

              a 4-byte number giving the link-layer header type for  
packets in
              the  capture; see pcap-linktype(7) for the LINKTYPE_  
values that
              can appear in this field.

       Following the per-file header are zero or  more  packets;   
each  packet
       begins  with  a per-packet header, which is immediately  
followed by the
       raw packet data.  The format of the per-packet header is:

              +---------------------------------------+
              |      Time stamp, seconds value        |
              +---------------------------------------+
              |    Time stamp, microseconds value     |
              +---------------------------------------+
              |    Length of captured packet data     |
              +---------------------------------------+
              |Un-truncated length of the packet data |
              +---------------------------------------+
       All fields in the per-packet header are in the byte order of   
the  host
       writing  the file.  The per-packet header begins with a time  
stamp giv-
       ing the approximate time the packet was captured; the time   
stamp  con-
       sists  of  a  4-byte value, giving the time in seconds since  
January 1,
       1970, 00:00:00 UTC, followed by a 4-byte  value,  giving  the   
time  in
       microseconds since that second.  Following that are a 4-byte  
value giv-
       ing the number of bytes of captured data  that  follow  the   
per-packet
       header  and  a  4-byte value giving the number of bytes that  
would have
       been present had the packet not been truncated by the  
snapshot  length.
       The two lengths will be equal if the number of bytes of packet  
data are
       less than or equal to the snapshot length.

SEE ALSO
       pcap(3PCAP), pcap-linktype(7)

(Note: before writing code to read this, PLEASE pay attention to the  
first paragraph on the man page.  Unless you have a compelling reason  
to read this file format yourself, rather than letting some existing  
code read it, and unless you either promise not to complain if tools  
start writing pcap-NG files or are willing to update your code to read  
pcap-NG files at some point, leave it up to libpcap/WinPcap or  
Wireshark's Wiretap library to do the reading.  If you're reading in a  
program written in Perl/Python/Ruby/Java/some .Net language/etc., see

	http://en.wikipedia.org/wiki/Pcap#Wrappers_for_use_of_libpcap.2FWinPcap_in_languages_other_than_C_and_C.2B.2B

and

	http://en.wikipedia.org/wiki/Pcap#External_links

for information on wrappers for libpcap/WinPcap for your favorite  
language.)