Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!munnari.oz.au!csc.anu.oz.au!csc3.anu.oz.au!ccadfa!rodos2!wkt
From: wkt@rodos2.cs.adfa.oz.au (Warren Toomey)
Newsgroups: comp.os.minix
Subject: Re: TCP/IP for Minix
Message-ID: <2069@ccadfa.adfa.oz.au>
Date: 15 Nov 90 03:54:36 GMT
Sender: news@ccadfa.adfa.oz.au
Lines: 36

I've seen several people in the last few months asking `Is anybody
working on TCP/IP for Minix?'. Here's a sort of an answer...

Yes, and no. I started at the beginning of the year to implement
TCP/IP and sockets as a `Network manager' process under Minix.
I am using the TCP/IP/socket code directly from UCB, and after
battling my way to a comprehension of the data structures (thanks
to the black book - Leffner, Karels, McKusick and Quaterman), have
reached the point where I have IP & ICMP running thru a serial port
as a user process on a Sun 3/50. No, not under Minix yet, but the
mods should take a week or so. Don't expect me to have them done
next week.

The UCB code is efficient, speed and memory-wise. The big problem
is the socket code, which pervades the source. Adding sockets to
Minix would mean adding a number of new system calls, and doing
sendrec()s to the new `Network Manager' process. I've not felt
game enough to start this yet.

I think that the people who are thinking of, or doing, a TCP/IP
port should pool their resources. If sockets only can be added,
then adding the TCP/UDP and IP protocols would be easy.

More problems - how to minimise the effect of sockets on the
File Server (seeing as they both use file descriptors); minimising
the overloading of ioctl(), which is IMHO the most abused Unix
syscall - basically making the thing modular.

Email for specifics, not that there are any yet. General comments
here, with a reminder that Andy will never support this, so don't
ask him to.


   Warren Toomey VK1XWT, rescreened.
Deep in the bowels of ADFA Comp Science.
      `What the hell is SIGTTIN?'