On Fri, Jul 22, 2005 at 10:04:34AM -0400, Brian A. Reiter said: > > I am experiencing a problem with clamd crashing shortly after > > a scan using the STREAM protocol. This affects recent > > versions of clamd including 0.86.1 and clamav-devel in CVS > > running on Interix. > > I have spent a little time looking into this. The segv happens when the > thread pool spins down the worker threads after the idle timeout for the > thread has elapsed. However it only segvs if all of the following are true: > > scanstream() was queued to the worker thread > cfgopt(copt, "TCPSocket"))->enabled) == TRUE > cfgopt(copt, "TCPAddr"))->enabled) == TRUE > > That pretty much says to me that the code that causes the instability is the > call to gethostbyname(3) to get the sin_addr for binding the STREAM socket > because the alternate execution path of binding to INADDR_ANY is > trouble-free. > > I'm trying to decide if I should replace gethostbyname(3) and the pthread > locks with gethostbyname_r/lwres_gethostbyname_r(3) (reentrant) or if there > is something in BIND9 on Interix that is broken. > > It would help if I knew if this is a problem that can be reproduced on other > platforms than Interix.
It does not appear to be reproducible on linux, either with a 2.4 or 2.6 kernel. I usually bind to 127.0.0.1, but have several machines that bind to one of the public interfaces. I have never experienced this crash on reload. It does sound like perhaps the implementation of gethostbyname is not reeentrant. Take care, -- -------------------------------------------------------------------------- | Stephen Gran | "Ask not what A Group of Employees can | | [EMAIL PROTECTED] | do for you. But ask what can All | | http://www.lobefin.net/~steve | Employees do for A Group of Employees." | | | -- Mike Dennison | --------------------------------------------------------------------------
signature.asc
Description: Digital signature
_______________________________________________ http://lurker.clamav.net/list/clamav-devel.html
