Description of problem:

At work, in upgrading our e-mail system, we were forcing stronger passwords 
for all accounts. More than one sysadmin had the GUI running at once. Later in 
the day, as we stopped and started the GUI, we would discover that some people 
that were getting on could no longer get on. I'd saved a copy of shadow so we 
could guarantee that we got everyone changed, and as I did a diff between it 
and the latest shadow, it was apparent that changes were dropped. The only 
explanation I see is that the GUI doesn't do any locking for row updates to 
the text file and perhaps retains too much in memory to be written when it 
quits, thus wiping out changes made by anyone else in the interim. This may be 
off base, but seems to be the only explanation that fits what we observed.

Version-Release number of selected component (if applicable):
system-config-users-1.2.72-1.fc8.noarch.rpm - all other RPMs were at the level 
of Feb 13, 2008. The error was observed Feb 20 due to changes on the 19th and 
20th.

How reproducible:
Apparently always.

Steps to Reproduce:
1. Log on in two different spots via ssh and run system-config-users (one was 
at system console and one was remote, although I doubt it matters).

2. Update passwords in both places for different users.

3. Save at different times.
  
Actual results:
Whoever leaves first gets their changes lost. We actually had people going in 
and out of the config-users program, so when a new one started that was the 
snapshot it appeared to work with.

Expected results:
Anything dealing with files like passwd, shadow, and group should be doing 
locking on write and only updating the row of information that was changed via 
the properties button. No other accounts should be affected. If two people are 
editing the same row (which wasn't the case here), then whoever writes last 
should win. But you should never assume that other rows hadn't changed in the 
interim and write everything out as if you were the only one in control.


Additional info: