[torquedev] Double free and touches of freed memory inside pbs_server
garrick at usc.edu
Thu Aug 5 23:14:06 MDT 2010
On Fri, Aug 06, 2010 at 07:48:47AM +0400, Eygene Ryabinkin alleged:
> Thu, Aug 05, 2010 at 01:23:29PM -0700, Garrick Staples wrote:
> > On Thu, Aug 05, 2010 at 11:39:47PM +0400, Eygene Ryabinkin alleged:
> > > It relays the request to the MOM and it is put to the working queue.
> > > Meanwhile, when the request flow continues, the passed request pointer
> > > is freed (before MOM's answer is worked out).
> > >
> > > So, not every server->mom request should fail, only those which resulted
> > > from the client-initiated request that involves server->mom secondary
> > > request(s).
> > So qrun, qdel, and qsig should have the same problem?
> Why? They are just user-space commands that initiate requests to the
> server from the client side -- they don't relay commands to the MOM,
> aren't they?
Yes, they do. Run qsig and pbs_server relays a signal request to MOM. Ditto for
qrun and qdel.
> They could initiate the problems within the pbs_server, but I am not so
> familiar with the workflow involved with these 3 commands. Though, I'd
> rather say that qalter should be the main headache, since it alters the
> attributes (resource limits) and this is the way how relay_to_mom() can
> be called inside req_modifyjob().
> Do you see some problems with the logics I had outlined in my initial
> posting and/or Bugzilla report?
Only that I haven't had a coredump in pbs_server or years and maui is modifing
jobs a thousand times a day.
Garrick Staples, GNU/Linux HPCC SysAdmin
University of Southern California
Life is Good!
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torquedev/attachments/20100805/1226cc47/attachment.bin
More information about the torquedev