Topic : Raw IP Networking FAQ
Author : Thamer Al-Harbash
Page : << Previous 3  Next >>
Go to page :

        exceeds a certain limit (resources are finite), the packets
        are dropped and are not placed in the buffer.

        Packet filters allow a process to dictate which packets it's
        interested in. The usual way is to have a set of opcodes for
        routines to perform on the packet, reading values off it, and
        deciding whether or not it's wanted. These opcodes usually
        perform very simple operations, allowing powerful filters to
        be constructed.

        BPF filters and then buffers; this is optimal since the
        buffer only contains packets that are interesting to the
        process. It's hoped that the filter cuts down the amount of
        packets buffered to stop overflowing the buffer, which leads
        to packet loss.

        NIT, unfortunately, does not do this; it applies the filter
        after buffering, when the user process starts to read from
        the buffered data.

        According to route <> Linux' SOCK_PACKET
        does not do any buffering and has no kernel filtering.

        Your mileage may vary with other packet capturing facilities.

        1.5) How do I limit packet loss when sniffing a network?

        If you're experiencing a lot of packet loss, you may want to
        limit the scope of the packets read by using filters. This
        will only work if the filtering is done before any buffering.
        If this still doesn't work because your packet capturing
        facility is broken like NIT, you'll have to read the packets
        faster in a user process and send them to another process --
        basically attempt to do additional buffering in user space.

        Another way of improving performance, is by using a larger
        buffer. On Irix using SNOOP, the man page recommends using
        SO_RCVBUF. On BSD with BPF one can use the BIOCSBLEN ioctl
        call to increase the buffer size. On Solaris bufmod and pfmod
        can be used for altering buffer size and filters

        Remember, the longer your process is busy and not attending the incoming packets, the quicker they'll be dropped by the kernel.

        1.6) What is packet capturing usually used for?

        (Question suggested by Michael T. Stolarchuk <>
        along with some suggestions for the answer.)

            Network diagnostics such as the verification of a
            network's setup, examples are tools like arp, that report
            the ARP messages sent from hosts.

            Reconstruction of end to end sessions. tcpshow attempts
            to do this, but more sophisticated examples are the array
            of security tools which try to keep tabs on network

            Monitoring network load. Probably one of the most
            practical uses, a lot of commercial products usually use
            specialized hardware to accomplish this.

        1.7) Will I have to replace any packets captured off the


        No, the packet capturing facilities mentioned make copies of
        the packets, and do not remove them from the system's TCP/IP
        stack. If you wish to prevent packets from reaching the
        TCP/IP stack you need to use a firewall, (which should be
        able to do packet filtering). Don't confuse the packet
        filtering done by packet capturing facilities with those done
        by firewalls. They serve different purposes.

        1.8) Is there a portable API to send raw packets into a


        Yes, route <> maintains Libnet, a library
        that provides an API for low level packet writing and
        handling. It serves as a good compliment for libpcap, if you
        wish to read and write packets. The project's webpage can be
        found at:

        1.9) Are there any high level language APIs (Not C) for raw
        IP access?


        A PERL module that gives access to raw sockets is available

        A Python library "py-libpap" can be found at:

    2) RAW socket questions:

        2.1) What is a RAW socket?

        The BSD socket API allows one to open a raw socket and bypass
        layers in the TCP/IP stack. Be warned that if an OS doesn't
        support correct BSD semantics (correct is used loosely here),
        you're going to have a hard time making it work. Below, an
        attempt is made to address some of the bugs or surprises
        you're in store for. On almost all sane systems only root
        (superuser) can open a raw socket.

        2.2) How do I use a raw socket?

            2.2.1) How do I send a TCP/IP packet through a raw


Page : << Previous 3  Next >>