On Wed, 2008-08-06 at 15:56 -0400, J. Bruce Fields wrote:
> So the only solution that seems to really be holding up (at least for
> the near term) is the double-buffering "hack"; could we get a version
> of that with Neil's review comments addressed?
I inverted the logic so that the file system has to set a
FS_LOOKUP_IN_READDIR flag to avoid the hack, rather than setting a
FS_NO_LOOKUP_IN_READDIR flag to _use_ it. Renamed it so that it doesn't
say 'hack', took out the loop to allocate memory.
Tasks left for someone more familiar with the NFS code are:
- making it use one of the pages already allocated for the response.
- giving it a hint about how many dirents to read into the buffer.
David Woodhouse Open Source Technology Centre
[email protected] Intel Corporation