Re: [CFT][PATCH] ramfs/tmpfs readdir()

[Trond Myklebust]

>>>>> " " == Manfred Spraul <[e-mail]> writes:

     > Content-Type: text/plain; charset=us-ascii
     > Content-Transfer-Encoding: 7bit

    >>
    >> Note that other filesystems would already enjoy having a
    >> d_offset in the dentry: it allows for various other
    >> optimizations (ie making "unlink()" a O(1) operation, by not
    >> having to search the directory).
    >>
     > The dentry structure already contains 2 members for filesystem
     > use (d_time and d_fsdata), is a third member really required?

It could be useful. For NFS, we're also looking for 64-bits in which
to store the parent directory's mtime. This is needed in order to
improve the dentry revalidation heuristics.

Ultimately, though, allowing the filesystem to allocate dentries
itself on its own slab might be a preferable manner to accomodate the
need for these private fields. IOW to allow one to allocate

   struct nfs_dentry { 
        struct dentry dentry;
        ... private fields ....
   };

Cheers,
  Trond