[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: (ITS#4108) LWL causing core dumps in HEAD

hyc@symas.com wrote:
> So it turns out there were a few race conditions and cases of trying to 
> close the same connection twice. (Which was bad news, because the 
> descriptor could get reused immediately after the first close.)
> I believe this is now fixed in HEAD. Please test. Heavily.

One more footnote - testing this bug using tests/progs/slapd-bind 
freezes up my kernel when running slapd with debug output on the 
console. It seems to be safe when running from an ssh session. I also 
run out of socket resources on the server pretty quickly if the bind 
client is on the same machine as the server. Otherwise, running remote 
clients, it can run through hundreds of thousands of connections without 
any trouble. (Server was running Linux 2.6.13, now 2.6.14, same lockup 
occurs on both. The Magic SysRq doesn't work, the kernel is totally 
locked up, no way to get any diagnostics or recover at that point.)

  -- Howard Chu
  Chief Architect, Symas Corp.  http://www.symas.com
  Director, Highland Sun        http://highlandsun.com/hyc
  OpenLDAP Core Team            http://www.openldap.org/project/