wireshark/docbook/wsdg_src/WSDG_chapter_capture.xml

83 lines
3.7 KiB
XML

<!-- WSDG Chapter Capture -->
<!-- $Id$ -->
<chapter id="ChapterCapture">
<title>Packet capturing</title>
<para>
XXX - this chapter has to be reviewed and extended!
</para>
<section id="ChCaptureAddLibpcap">
<title>How to add a new capture type to libpcap</title>
<para>
The following is an excerpt from a developer mailing list mail, about
adding ISO 9141 and 14230 (simple serial line card diagnostics) to
Wireshark:
</para>
<para>
For libpcap, the first thing you'd need to do would be to get DLT_ values
for all the link-layer protocols you'd need. If ISO 9141 and 14230 use
the same link-layer protocol, they might be able to share a DLT_ value,
unless the only way to know what protocols are running above the link
layer is to know which link-layer protocol is being used, in which case
you might want separate DLT_ values.
</para>
<para>
For the rest of the libpcap discussion, I'll assume you're working with
the current top-of-tree CVS version of libpcap, and that this is on a
UN*X platform. You probably don't want to work with a version older than
0.8, even if whatever OS you're using happens to include libpcap - older
versions are not as friendly towards adding support for devices other than
standard network interfaces.
</para>
<para>
Then you'd probably add to the "pcap_open_live()" routine, for whatever
platform or platforms this code should work, something such as a check
for device names that look like serial port names and, if the check
succeeds, a call to a routine to open the serial port.
</para>
<para>
See, for example, the "#ifdef HAVE_DAG_API" code in pcap-linux.c and
pcap-bpf.c.
</para>
<para>
The serial port open routine would open the serial port device, set the
baud rate and do anything else needed to open the device. It'd allocate
a pcap_t, set its "fd" member to the file descriptor for the serial
device, set the "snapshot" member to the argument passed to the open
routine, set the "linktype" member to one of the DLT_ values, and set the
"selectable_fd" member to the same value as the "fd" member. It should
also set the "dlt_count" member to the number of DLT_ values to support,
and allocate an array of "dlt_count" "u_int"s, assign it to the "dlt_list"
member, and fill in that list with all the DLT_ values.
</para>
<para>
You'd then set the various _op fields to routines to handle the
operations in question. read_op is the routine that'd read packets from
the device. inject_op would be for sending packets; if you don't care
about that, you'd set it to a routine that returns an error indication.
setfilter_op can probably just be set to install_bpf_program. set_datalink
would just set the "linktype" member to the specified value if it's one
of the values for OBD, otherwise it should return an error. getnonblock_op
can probably be set to pcap_getnonblock_fd; setnonblock_op can probably be
set to pcap_setnonblock_fd. stats_op would be set to a routine that
reports statistics. close_op can probably be set to pcap_close_common.
</para>
<para>
If there's more than one DLT_ value, you definitely want a set_datalink
routine, so that the user can select the appropriate link-layer type.
</para>
<para>
For Wireshark, you'd add support for those DLT_ values to wiretap/libpcap.c,
which might mean adding one or more WTAP_ENCAP types to wtap.h and to the
encap_table[] table in wiretap/wtap.c. You'd then have to write a
dissector or dissectors for the link-layer protocols or protocols and have
them register themselves with the "wtap_encap" dissector table, with the
appropriate WTAP_ENCAP values, by calling "dissector_add_uint()".
</para>
</section>
</chapter>
<!-- End of EDG Chapter Dissection -->