On 08/04/2011 07:10 PM, Alex Jia wrote:
We definitely have a bug here, but this is not the right fix. The bug is that the cleanup: label is trying to read from logfd if the vm crashed, without having opened logfd in the qemuProcessAttach case.
I think the more appropriate patch is this:
diff --git i/src/qemu/qemu_process.c w/src/qemu/qemu_process.c index 8508ff6..1eea45f 100644 --- i/src/qemu/qemu_process.c +++ w/src/qemu/qemu_process.c @@ -1214,7 +1214,7 @@ qemuProcessWaitForMonitor(struct qemud_driver* driver, cleanup: virHashFree(paths);
- if (kill(vm->pid, 0) == -1 && errno == ESRCH) { + if (pos != -1 && kill(vm->pid, 0) == -1 && errno == ESRCH) { /* VM is dead, any other error raised in the interim is probably * not as important as the qemu cmdline output */ qemuProcessReadLogFD(logfd, buf, buf_size, strlen(buf));
Agree, it indeed is a issue, I will check 'null pointer' issue with thiis fixed again, to avoid some warning from ccc-analyzer, if you set up this env, please also check it.
With that, I've gone ahead and pushed this, which also included fixing the double VIR_FREE. -- Eric Blake eblake@redhat.com +1-801-349-2682 Libvirt virtualization library http://libvirt.org