On Mon, Jan 13, 2014 at 07:39:48AM +0000, David Holland wrote: > In the case of VOP_SYMLINK it definitely matters, as the new symlink > isn't filled until after VOP_SYMLINK returns. Since the locking > protocol for the containing directory is to unlock it *before* > returning, it then becomes possible to see the empty symlink. And that > isn't so good.
Er. This isn't actually true - I think I was thinking of some other older vfs interface. > It looks to me like it would definitely be better to first fix the > VOP_MKDIR interface to not release the directory lock within the call. > > Certainly for VOP_SYMLINK. > > Therefore, it's probably best to fix all the creation ops. Therefore, this is less critical; but I think the mkdir cases warrant doing it. -- David A. Holland dholl...@netbsd.org