On Tue, May 10, 2011 at 10:53:40 +0100, Daniel P. Berrange wrote: > On Mon, May 09, 2011 at 04:04:41PM +0200, Matthias Bolte wrote: > > 2011/5/9 Jiri Denemark <jdenemar@xxxxxxxxxx>: > > > virsh didn't call virInitialize(), which (among other things) > > > initializes virLastErr thread local variable. As a result of that, virsh > > > could just segfault in virEventRegisterDefaultImpl() since that is the > > > first call that touches (resets) virLastErr. > > > > > > I have no idea what lucky coincidence made this bug visible but I was > > > able to reproduce it in 100% cases but only in one specific environment > > > which included building in sandbox. > > > > Well, actually all public API that can be called as the first public > > API function ever called in a sound program calls virInitialize first > > internally. For example virConnectOpen. The documentation of > > virInitialize suggest to call it first, but doesn't require it. > > > > A fix in line with the current behavior would be to make > > virEventRegisterDefaultImpl and virEventRunDefaultImpl call > > virInitialize first too. > > Yep that's good. I looked at this and I'm not sure it's worth it :-) We would need to move the two functions from util/event.c to libvirt.c, which on one hand makes sense since they are public APIs but on the other hand it's kind of nice that we have event related functions in event.c. Moreover, I'm not a big fan of making more APIs special in that they automagically call virInitialize, esp. since one may ask why virEventRegisterDefaultImpl is special but virEventRegisterImpl is not? And why other APIs are not special? > > > > But this doesn't fix another problem, a program could call > > virDomainFree(NULL) for some reason as the first public API call ever > > made in the process and libvirt would try to report an error but would > > segfault because of calling virResetLastError before virInitialize was > > called. > > IMHO that would be a mis-use of the API because they should call > virInitialize. > > > So you found a much bigger problem and I'm not sure about the correct > > general solution. We could make calling virInitialize first mandatory > > but that might break existing applications. Or we make all public API > > functions call virInitialize first. > > I think we could just clarify the docs > > "Calling virInitialize is mandatory, unless your > first API call is one of > > virConnectOpen* > virEventRegisterDefaultImpl" I think leaving just virConnectOpen* APIs special should be enough. In more advanced code that doesn't just connect and do some stuff, there's no harm if we require users to call virInitialize first. So the following docs clarification would solve it: "Calling virInitialize is mandatory, unless your first API call is one of virConnectOpen*" Jirka -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list