On Fri, Sep 16, 2022 at 08:28:06AM +0200, Miklos Szeredi wrote: > This will break AT_SYMLINK_FOLLOW. Right you are. > And yes, we can add all the lookup logic to do_linkat() at which point > it will about 10x more complex than it was. Especially since you can't reject an apparent cross-fs link until you'v looked the fucker up, since it just might be a symlink to be followed. Which means it would have to be something like find parents again: if on different mounts if !follow fuck off lock old parent look the last component up if not an existing symlink fuck off unlock the parent and try to follow that symlink goto again lock parents look the last components up if symlink to be followed unlock parents try to follow symlink goto again proceed Not exactly fatal, but...