Re: If not readdir() then what?

[Trond Myklebust <trond.myklebust@fys.uio.no>]

On Wed, 2007-04-11 at 08:33 +1000, Neil Brown wrote:

>   A READDIR (aka getdents2) should take a directory handle, a cookie,
>   and a filename, and should return filenames and cookies.  The
>   cookies may all be identical or may not.  The filename might be used
>   by the filesystem, or it might not.
> 
>   Filesystems that require a cursor in the 'struct file' to support
>   (local) getdents cannot be used with NFS.
> 
> While it doesn't make it possible to support all conceivable
> filesystems, it should make it easier for some filesystems to support
> the demands of NFS.

In order to be useful, I think you need to add a demand that the READDIR
call cannot loop back on itself for the case of a series of sequential
reads.

IOW: if a client attempts to step sequentially through the directory,
and is supplying valid filenames+cookies from the preceding READDIR
call, then the next READDIR call should be guaranteed never to loop back
to an earlier entry. Alternatively, if there is a danger that it might
due to some sudden and radical change in the directory layout, then it
should notify the client by returning something like a BAD_COOKIE error.

Otherwise, the client would have to cache _all_ previous READDIR results
since the last opendir()/rewinddir() in order to be able to do its own
loop detection and that will obviously never scale for large directories
or for directories that change frequently...

Cheers
  Trond