On Fri, Mar 12, 2021 at 09:44:09 +0100, Peter Krempa wrote: > On Thu, Mar 11, 2021 at 16:47:54 -0700, Jim Fehlig wrote: > > On 3/10/21 9:37 AM, Peter Krempa wrote: > > > Commit 94e45d1042e broke exec-restart of virtlogd and virtlockd as the > > > code waiting for the daemon shutdown closed the daemons before > > > exec-restarting. > > > > This reminds me of an odd issue we encountered three years ago, fixed by Daniel > > > > https://listman.redhat.com/archives/libvir-list/2018-March/msg00298.html > > > > I tested your patches but notice locks are still lost on re-exec. > > > > qemu.conf: > > lock_manager = "lockd" > > > > qemu-lockd.conf: > > file_lockspace_dir = "/var/lib/libvirt/lockspace" > > > > /var/lib/libvirt/lockspace is nothing special, xfs on a local disk. After > > starting a VM > > > > # ls /var/lib/libvirt/lockspace/ > > a89872e150e6b9e4cbd59ef2bd289bc6cd0a8fa6fbf533c41957f77a90381e9c > > # lslocks | grep lockd > > virtlockd 95009 POSIX WRITE 0 0 0 /var/lib/libvirt/lockspace/a89872e150e6b9e4cbd59ef2bd289bc6cd0a8fa6fbf533c41957f77a90381e9c > > virtlockd 95009 POSIX 5B WRITE 0 0 0 /run/virtlockd.pid > > # systemctl reload virtlockd > > Could you make sure that the virtlockd process before and after has the > same pid, so that it wasn't actually restarted by systemct? > > I'm asking because in my current test I've encountered another crash > when exec-restarting: > > 2021-03-12 08:41:31.649+0000: 2765718: error : virJSONValueToBuffer:1946 : internal error: failed to convert virJSONValue to yajl data > double free or corruption (fasttop) > > > Looks like a double free. I'll post patches later for this. https://listman.redhat.com/archives/libvir-list/2021-March/msg00569.html
