cvs commit: src/sys/nfsclient nfs_socket.c

[Robert Watson]

On Tue, 19 Jul 2005, Alfred Perlstein wrote:

>> I saw ESTALE errors on NFS clients after rebooting NFS server. 
>> umount/mount was the only way to fix it. Your commit should fix my 
>> problem? My machines are running 5-STABLE.
>
> Unlikely, the source of such a problem is probably that the order of 
> mounted filesystems has changed and a different fsid was given to your 
> exported fs.
>
> One way to fix it is to make sure that your fses are mounted in the 
> right order each time.  Or you can do a hack whereby exporting loads the 
> fsid from a persistent file in the filesystem.

I've been wondering for a while about the best way to address this, and a 
possible start is to, on mount, attempt to derive the fsid from the uuid 
of the file system (if present).  We would need to then detect collisions 
at run-time and fall back to an alternative fsid.  The chances of a 
collision are fairly low, so it might be an improvement over current 
behavior.

Dynamic minor numbers have further reduced the effectiveness of the fsid 
model, which had already been significant reduced by CAM.

Robert N M Watson