*BSD News Article 16257


Return to BSD News archive

Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!agate!howland.reston.ans.net!newsserver.jvnc.net!yale.edu!yale!gumby!destroyer!news.itd.umich.edu!stimpy.css.itd.umich.edu!pauls
From: pauls@css.itd.umich.edu (Paul Southworth)
Newsgroups: comp.os.386bsd.bugs
Subject: NetBSD + 3com503 (8-bit) = death
Date: 18 May 1993 21:18:38 GMT
Organization: University of Michigan ITD Consulting and Support Services
Lines: 38
Distribution: world
Message-ID: <1tbjre$pqg@terminator.rs.itd.umich.edu>
NNTP-Posting-Host: stimpy.css.itd.umich.edu

So I finally sorted out my network problems.  As a person who is responsible
for maintaining a big ugly over-grown LAN, I have a big supply of ethernet
cards to play with.  As you may recall (or not) some of my previous posts
regarded network death during ftp sessions from my NetBSD 0.8 (full-install)
system to wherever.   This is what I found:

1. Generally the second retrieval of a file via ftp to my system would
   make the machine hang.  If I was doing it at console, there would be
   no way to break out of it.  If I was connected with a telnet session,
   the session would die but I could go login at console.  At console I
   would see messages like "acksend: sendto: no buffers available" and
   "inappropriate ioctl for device" with "no job control for this shell"
   when I logged in as root.  These could not be reproduced with absolute
   success, but it happened more than ten times and always while doing
   the same thing.

2. On examining the file that failed to transfer, it would always transfer
   a piece of it, and size would always be a multiple of 1024 bytes, most
   frequently 4096.  Very suspicious.  I also noted that after the network
   death, if I ifconfig'd the interface again, it came back to life, just
   as good as rebooting.

3. I swapped the 3com503 (8-bit, BNC/thinnet) for another one, configured
   identically, according to the docs (and both of these cards performed
   flawlessly with 386BSD 0.1 from the time it was released through three
   or four different patched kernels).  Problem persisted.

4. Finally I decided, after reviewing my configurations carefully, that
   there might be something wrong with the driver.  So I dropped in a
   Novell NE2000 (16-bit, BNC/thinnet-configured) and did:
   "mv /etc/hostname.ec0 /etc/hostname.ne0 ; sync ; shutdown -r now"
   (ie, absolutely no changes to configuration other than the card swap)
   and now have had no failures for 48 hours, whereas previously I was
   killing the net more than once a day on average.  I went nuts with 
   ftp sessions today and could not crash it.


Go figure.