Matthew Dillon dillon at apollo.backplane.com
Thu Jan 5 09:26:37 PST 2006

:Maybe a doubly linked list of/in the namecache entries.  Locking would 
:always lock either just the first entry in the list (I think that's 
:sufficient) or at least always obey the same order to prevent deadlocks. 
:  This way renames to self (in another incarnation) can be detected as 
:well (by running through the list).
:   simon

    I don't think it's going to be that simple.  We are going to need a
    cache coherency mechanism for namespace operations between machines just like
    we are going to need it for file read/write ops.   And, just like as I
    have described for file I/O, whatever mechanism we come up with will work
    just as well for competing entities on the same host as they will for
    competing entities on different hosts.

    In anycase, as I have said, I do NOT think this is an issue that nullfs itself
    has to deal with.  It is definitely an issue that we want the namecache layer
    to deal with, or more particularly that we want a generalized cache coherency
    layer to deal with.

					Matthew Dillon 
					<dillon at xxxxxxxxxxxxx>

More information about the Kernel mailing list