Re: NFS Client patch

[Andi Kleen]

On Wed, Jul 11, 2001 at 01:33:11AM +1200, Chris Wedgwood wrote:
> On Mon, Jul 09, 2001 at 08:33:31PM +0200, Andi Kleen wrote:
> 
>     Actually all the file systems who do that on Linux (JFS, XFS,
>     reiserfs) have fixed the issue properly server side, by adding a
>     layer that generates stable cookies. You should too.
> 
> I've always thought that was a stupid fix. Why not have the clients be
> smarted and make them responsible for getting a new cookie if the old
> one is hosed?

Because to get that new cookie you would need another cookie; otherwise
you could violate the readdir guarantee that it'll never return files
twice.

> For linux, with the dcache, I'm not even sure that this would be all
> the hard. Persumable Solaris could (does?) do the same?

dcache is not populated on readdir for good reasons (it would 
require reading the inodes and tie a of lot of memory) and you would need
to lock all the dcache entries belong to a directory while a nfs readdir;
tieing up even more memory. Also a readdir() is not bounded in time.

BTW; the cookie issue is not an NFS only problem. It occurs on local
IO as well. Just consider rm -rf - reading directories and in parallel
deleting them (the original poster's file system would have surely
gotten that wrong). Another tricky case is telldir().  

-Andi