From: Mark (Mookie) <mark@zang.com>
Date: Sun, 14 Jan 1996 04:52:59 +0800
To: cypherpunks@toad.com
Subject: tsu's bpf
Message-ID: <199601122307.NAA09386@zang.com>
MIME-Version: 1.0
Content-Type: text


>>    Shimomura on BPF, NSA and Crypto:
>>
>>    One of the tools I modified for my work was a sophisticated
>>    piece of software called the Berkeley Packet Filter. ...
>>    Unlike the original BPF, my version was designed to bury
>	 ^^^^^^^^^^^^^^^^^^^^^^^
>>    itself inside the operating system of a computer and watch
>>    for certain information as it flowed through the computer
>>    from the Internet. When a packet from a certain address, or
>>    for that matter any other desired piece of information
>>    designated by the user flashed by, BPF would grab it and
>>    place it in a file where it could be kept for later
>>    viewing.
>
>This is *exactly* what BPF does, always did and was designed to do. As
>for writing the packets to a file, everything but opening and closing
>the file are described in the man page. You could code it in 10 lines.

Get off your high horse Julian, he means it's a modloadable version of bpf,
much like the modloadable NIT that is also available. There are at least
two sniffers that can use both the modloadable NIT and bpf packet interfaces,
maybe more. It certainly is easier than recompiling your kernel to include
the functionality which is generally the way things were done.

I prefer bpf as it is much more efficient, typically 10% of the impact that
NIT has on a machine. Some rough figures are a NIT might use one or two
minutes of CPU a day to monitor a reasonably quiet network, whilst bpf will
only use several seconds cpu time. (Most of the work is hidden in the kernel
anyway).

Mark