On 05/30/2017 09:19 AM, Daniel P. Berrange wrote: > On Tue, May 30, 2017 at 03:03:14PM +0200, Pavel Hrdina wrote: >> On Tue, May 30, 2017 at 07:38:17AM -0400, John Ferlan wrote: >>> Introduce a recursive option for the lockable object which calls >>> virMutexInitRecursive instead of virMutexInit. >> >> We should avoid using recursive lock because they are dangerous. I >> would rather cleanup the nwfilter code than introducing more stuff for >> recursive locks. > > Last time I looked I came to the conclusion that it wasn't possible to > remove the recursive locking from nwfilter. > > I agree about /not/ introducing new usage of recursive locking though. > So this is a chicken/egg problem.... Cannot create a new API to use recursive locks and cannot remove the recursive lock without having adjustments to nwfilter code that would seemingly require self locking list objects to get hash table elements. If the comments in src/nwfilter/nwfilter_gentech_driver.c (search on XXX) could prove to be true, then it would be possible to not have a recursive lock. Perhaps that loop could be rewritten in virnwfilterobj to grab a reference to a hash table object while releasing the object lock that's causing the deadlock. I haven't yet circled back to determine whether this is possible yet, but since code in my branches is at least this far, I can look again. John -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list
