On 9/19/19 10:01 AM, Xu Yandong wrote:
We found a exception when libvirt occurrs segmentation fault. thread 1 is waiting object lock in qemuConnectMonitor, qemu process exits and sends EOF event as well, so thread 2 invokes qemuMonitorLastError but pointer mon is NULL.
Signed-off-by: Xu Yandong <xuyandong2@huawei.com> --- src/qemu/qemu_monitor.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/qemu/qemu_monitor.c b/src/qemu/qemu_monitor.c index aa230b3306..28f18391a5 100644 --- a/src/qemu/qemu_monitor.c +++ b/src/qemu/qemu_monitor.c @@ -1103,7 +1103,7 @@ qemuMonitorSend(qemuMonitorPtr mon, virErrorPtr qemuMonitorLastError(qemuMonitorPtr mon) { - if (mon->lastError.code == VIR_ERR_OK) + if (!mon || mon->lastError.code == VIR_ERR_OK) return NULL;
return virErrorCopyNew(&mon->lastError);
This works, but I think we need a better approach. The problem is, we register monitor FD into our event loop with unlocked @vm and only after that we try to lock @vm back and set priv->mon = mon (this is all done in qemuConnectMonitor()). I have better fix in mind but I have to run now, will post it tomorrow, if that's okay with you. Michal