Re: [libvirt] [PATCH] Release VM lock before acquiring virDomainObjListPtr lock

On Mon, Apr 08, 2013 at 09:15:28PM -0600, Eric Blake wrote: > On 02/11/2013 09:46 AM, Daniel P. Berrange wrote: >> From: "Daniel P. Berrange" <berrange(a)redhat.com&gt; >> >> When removing a VM from the virDomainObjListPtr, we must not >> be holding the VM lock while acquiring the list lock. Re-order >> code to ensure that we can release the VM lock early. >> --- >> src/conf/domain_conf.c | 3 +-- >> 1 file changed, 1 insertion(+), 2 deletions(-) >> >> diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c >> index 5e16ddf..d92e54a 100644 >> --- a/src/conf/domain_conf.c >> +++ b/src/conf/domain_conf.c >> @@ -2115,11 +2115,10 @@ void virDomainObjListRemove(virDomainObjListPtr doms, >> { >> char uuidstr[VIR_UUID_STRING_BUFLEN]; >> >> - virObjectLock(doms); >> virUUIDFormat(dom->def->uuid, uuidstr); >> - >> virObjectUnlock(dom); >> >> + virObjectLock(doms); > > This patch seems to be implicated in Peter's latest proof of a > use-after-free data race: > https://www.redhat.com/archives/libvir-list/2013-April/msg00674.html The caller of this method should own a reference on virDomainObjPtr. This method will result in that reference being released, as the dom is removed from the list. If any other thread experiances a use-after-free, then that thread must have been missing a reference.

> I'm trying to understand what the behavior was before this patch went in. Well this was just fixing a deadlock introduced in a previous patch. You need to look further back than just this patch. Originally the global QEMU driver lock would be held preventing any kind of concurrent execution. Regards, Daniel