pthreads leaks handles and threads when threads use sockets

[Mark Pizzolato]

Hi Reini,

> Reini Urban wrote:
> Mark Pizzolato schrieb:
> > I've been using clamav's clamd under cygwin and noticed that over time
> > the handle count as viewed with TaskManager seems to grow to arbitrary
> > values. I used clamd's option IdleTimeout set to 600 seconds which
> > dramatically reduced the growth rate of the Handle Count. Of course
> > clamd has many things going on that could contribute to handle leakags,
> > so I tried to write a simple program to demonstrate the problem.
>
> Thanks a lot! Maybe we should restart the two daemons daily or weekly?
>
> I will change the default IdleTimeout to 600 secs with the upcoming
> clamav-0.81 release. Which fixes the freshclam proxy problem and some
> OLE issues.

Merely setting IdleTimeout to 600 is currently insufficient due to a bug 
which only uses the IdleTimeout parameter for the Initial value used.  After 
the AV Database is reloaded, the idle timeout is hardset to 30 seconds.  The 
attached patch (to 0.80 or 0.81) fixes this issue.  This patch has been 
submitted on the clamav-devel list.

The right choice for the IdleTimeout is a value which is larger than the 
largest gap between messages that arrive on your system.  This is somewhat 
complicated by multiple connections arriving concurrently which is handled 
by MaxThreads.  MaxThreads has a default value of 10.  This would be fine 
for most systems, however libclamav uses tmpfile() internally which is NOT 
threadsafe (using newlib's tmpfile()) for any system which returns the same 
value for getpid() for each thread in a process (i.e. it works fine on Linux 
since getpid() on Linux returns a unique value for all threads on the 
system).  I've submitted changes which address this to the clamav 
folks(avoiding tempfile()), but they have not been accepted as yet.  To 
avoid this issue (and avoid clamd producing "ERROR: ScanStream: Can't create 
temporary file." messages), setting MaxThreads to 1 should work, but will 
probably affect the behavior of client software that talks to it (possibly 
causing deadlocks).

Do you have any insight to help address the underlying socket issues in 
threaded programs would clearly help with clamd and every other multi 
threaded program which may not even know it has these issues.

- Mark Pizzolato 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: idletimeout.patch
Type: application/octet-stream
Size: 460 bytes
Desc: not available
URL: <http://cygwin.com/pipermail/cygwin/attachments/20050129/7323ddf5/attachment.obj>
-------------- next part --------------
--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/