unionfs port/update

Matthew Dillon dillon at apollo.backplane.com
Tue Feb 23 14:27:07 PST 2010


:...
:> I haven't yet read the unionfs papers so the theory is still very
:> generalized in my head.  Basically as I understand it all writes
:> should go to an upper level vnode and reads should try the upper level
:> first and fall through to the lower level if there is no upper shadow
:> copy.  Removes should always operate on the upper level as well as
:> creates and renames.
:>
:> Please let me know your thoughts and most importantly any tips and
:
:FWIW LWN ran a serie of articles explaining various unionfs approaches:
:
:http://lwn.net/Articles/327738/

    If we had data-dedup for HAMMER we wouldn't even have to worry about
    the complexities of unionfs.  cpdup would work fine.

    One thing I've noticed in DragonFly is that there is much less of a
    case of needing unionfs when one has snapshots and when nullfs mounts
    work well.

    In thinking about how snapshots help here it occurs to me that we might
    want to add a snapshot locking feature whereby a snapshot can be locked
    (preventing hammer cleanup, prune, or prune-everything from destroying
    it).

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>





More information about the Kernel mailing list