From: The Wyvern (wyvern@gnu.ai.mit.edu)
Date: 02/21/93


From: wyvern@gnu.ai.mit.edu (The Wyvern)
Subject: Bug in 0.99.5 kernel loopback code
Date: 21 Feb 1993 20:57:53 GMT


Greetings - It seems I've been lucky enough to find a bug in the kernel
tcp code. This problem can be reproduced at will (by me, anyway) when
running term by tredir'ing a local port to some remote port on the
internet and then telnetting to the local port immediately.

% tredir 8000 255.255.255.255:23 (or whatever ip addr)
% telnet
telnet> open 0 8000

At this point Linux dies (and hard too - reset button time) and rebooting,
I find this in the logs:

Linux: EIP: 0008:0002B140
Linux: EFLAGS: 00013202
Linux: fs: 0017
Linux: base: 00000000, limit: C0000000
Linux: Pid: 33, process nr: 4
Linux: 8b 57 24 52 8d 47 38 50 e8 23

It may be worth noting that this has happened several times, but only once
has it been recorded in the syslog files. It confused me quite a bit
for a while.

Anyway, tracking the address down in System.map, I found these entries:

  _loopback_xmit: 0x2b114
  gcc2_compiled.: 0x2b114
  loopback.o: 0x2b114

0x2b114 was the closest address on the lowest side of 0x2b140 in System.map.

Here's my system configuration, for whatever enlightenment it may hold:
  Linux v0.99.5 (TCP/IP, -m486)
  i486 DX/33 system
  IDE HD
  8mb RAM
  No ethernet board