On Wed, Oct 01, 2025 at 02:25:55PM -0000, Denis Rastyogin wrote:
Indeed, when virDomainCreate() < 0, virDispatchError() should be invoked, but since we do not check its return value, there is no absolute guarantee that the error message will always be set successfully.
virDispatchError() contains the following code:
virErrorPtr err = virLastErrorObject();
/* Can only happen on OOM. */ if (!err) return;
The comment clearly indicates that virLastErrorObject() may return NULL in case of OOM.
That's just an outdated comment from before we switched all allocations to g_new0 to eliminate OOM handling.
Moreover, even if the call err = g_new0(virError, 1); succeeds, this does not guarantee that virThreadLocalSet(&virLastErr, err) < 0 will always false.
For example, for large values of l->key, an additional memory allocation may occur, and if that allocation fails, g_clear_pointer(&err, g_free) will be executed, causing virLastErrorObject() to return NULL.
Although this situation is extremely unlikely, it is better to add an explicit NULL check.
No, it really isn't useful to do that. If pthread_setspecific is failing with OOM then there is nothing useful the program can still do. The only sane thing is to assert & dump core. With regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|