From espel Sat Jan 20 06:57:25 1996
Subject: fix for deop problems with TS
To: operlist@kei.com
Date: Sat, 20 Jan 1996 06:57:25 +0100 (MET)
X-Mailer: ELM [version 2.4 PL24 ME8a]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Content-Length: 3067      
Status: O

For all of those concerned about the deops happening with the TS 
servers, there's now an explanation and a fix.

After a bit of thinking and experimenting on a test net, slackie figured
out what situations could lead to servers deopping users who had been on
a channel for long, in a mixed TS/no-TS net.

Her explanation:

Assume we have 4 servers, A-D, with B-D being TS servers and A
being non-TS.
1. squit all servers
2. have users on servers A, C and D join channel #blah.  Have the
users on server A join it first, then D, then C. 
3. connect servers A and B.  Since A is a non-TS server, B will set
the TS for #blah to 0
4. connect servers B and C.  Since the TS for #blah on B is 0, #blah
will adopt the TS that server C has set.
5. connect server D to server B or C.  Since the TS on this server
is oldest, users on all other servers will be deopped. 

What happens is that users are joining TS leaf servers, causing a new TS 
to be set for #blah.  When the leaf with the newer TS connects to the 
server with the TS of 0 (which was set to 0 because the ops were 
propagated from non-TS servers), both servers accept the TS of the leaf.  
When the leaf with the older TS connects to the another TS server, all 
other ops on #blah are deopped.

This has nothing to do with clocks; if that was the problem it would be 
obvious b/c it would be one server in particular doing it over and over 
again, which it isn't. When TS's are being compared, they are typically 
much more different than whatever clock drift one could expect even 
counting with a few seconds of lag when synchronizing, so this shouldn't 
be a problem.

To fix this, servers need to be less aggresive in setting non-0 values 
for channels when part of the net doesn't do TS. The only way to 
completely prevent these deops is to make it so that when two TS 
servers, both with ops, and one with a non-0 TS and the other with 0 are 
exchanging information about a channel, they should set the TS to 0 on 
both servers rather than both taking the non-0 value.

This also means that on a mixed net channels will only gain and keep a 
non-0 TS when re-created; hacking ops from a TS server will no longer be 
enough.

I've written a patch that makes TSora servers behave this way; the patch 
can be applied directly to +CS and /digi servers too (I've mailed it to 
CoMsTuD and digital for their next versions). Mixing servers with and 
without the patch isn't a problem, but there will be deops as long 
as there are double splits involving servers without the patch.

The patch can be found at 
ftp://isr0954.urh.uiuc.edu/pub/orabidoo/channel.diff
as well as http://www.eleves.ens.fr:8080/home/espel/TS/index.html

If anyone has any other ideas or suggestions, plase mail me.

--
e-mail: roger.espel.llima@ens.fr                     IRC: orabidoo
<slackie> even celery has life, what some ppl lack is a brain