On Mon, Jan 27, 2014 at 01:05:51PM -0500, Stefan Berger wrote:
On 01/27/2014 12:15 PM, Daniel P. Berrange wrote:
On Thu, Jan 23, 2014 at 04:13:57PM -0500, Stefan Berger wrote:
This function here is called by src/nwfilter/nwfilter_learnip.c::learnIPAddressThread src/nwfilter/nwfilter_dhcpsnoop.c::virNWFilterSnoopIPLeaseInstallRule
They instantiate the filters once a VM's IP address has been detected. So this is where the *Late() comes from.
If you remove the locking from here, you have to lock it there. Considering what you do layer, I would keep the lock here and convert this into a reader lock layer on. Yes, now I'm squashing the read/write lock conversion in, I'll keep the locking in this location.
Is there interest in introducing an assert() in those functions expecting a lock to be held? I know it's not as simple as just putting an assert() into the code. Actually we would have to record in an array or linked list the threads holding a lock and provide a function to check whether pthread_self() is holding the lock. The latter function would be called by an assert(). I would give it a shot by introducing a boolean as parameter to the lock init function that activates the recording of which threads are currently holding a lock -- if this is thought to be useful.
Usage of assert() is forbiddenn in any libvirt code, and the complexity of tracking threads this really not something I'd want to persue. Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|