On Tue, Jun 5, 2018 at 7:19 PM, shankarapailoor <shankarapailoor@xxxxxxxxx> wrote: > Hi Cong, > > I added that check and it seems to stop the crash. Like you said, I > don't see where the reference count for the file is increased. The > inode lock also seems to be held during this call. I know inode lock is held for ->setattr(), but not for ->release(), this is why I suspect sock_close() could still race with sockfs_setattr() after my patch. I am not sure if it is crazy to just hold fd refcnt for fchmodat() too..
