[OpenAFS-devel] pthreaded fileserver deadlocks on shutdown/restart

[Rainer Toebbicke]

Nickolai Zeldovich wrote:
> Rainer Toebbicke <rtb@pclella.cern.ch> wrote:
> 
> 
>>The FSYNC_sync thread runs with no signals masked.
> 
> 
> This is actually a problem for all threads, not just FSYNC_sync.
> There's already a (more general) patch to fix this problem, by
> redirecting all signals to a special thread, in cvs.  I'm not
> sure if it's stable enough to pull up to 1.2.9; that would be
> Derrick's call.
> 

Well, it's wrong, hung us twice and we simply fixed it. When the fix (or any 
enhancement) gets pulled up into the standard OpenAFS is not my immediate 
concern - we're running 1.2.6 anyway.

I agree it would be less error-prone if the long list of invariants that 
surround the various pthread_create()s were all done in a single library 
routine. All the other threads except the initial one mask signals so things 
work fine - FSYNC_sync was probably simply forgotten.


-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Rainer Toebbicke        http://cern.ch/~rtb         rtb@mail.cern.ch  O__
European Laboratory for Particle Physics(CERN) - Geneva, Switzerland   > |
Phone: +41 22 767 8985       Fax: +41 22 767 7155                     ( )\( )