Re: locking patches for ufs_rename

[David Holland <dholland-tech%netbsd.org@localhost>]

On Thu, Oct 29, 2009 at 06:03:37PM +0100, Rhialto wrote:
 > > > Would it make sense to use a global lock only for cross-directory
 > > > renames, and not for in-directory renames?
 > > 
 > > Yes.
 > 
 > What if one rename renames (moves) across directories, into some other
 > directory where an in-directory rename is going on? Or away from such a
 > directory? The one move might take the global lock, but not the other.
 > Is that a problem?

I don't think so. As posted elsewhere the global lock serves two
purposes: first, to prevent deadlock if someone sets up a cycle of
renames among incommensurate portions of the tree, which a
same-directory rename inherently can't participate in; and second, to
make sure the parent check doesn't become invalid while the rename is
running. A same-directory rename can't give a subdirectory different
parents, so this should be ok also.

It might allow other more subtle inconsistencies in observed behavior;
however, there are already plenty of such cases and it's not at all
clear that they matter.

-- 
David A. Holland
dholland%netbsd.org@localhost