[linux-cifs-client] 2.6.31 kernel breaks CIFS mounts without warning

[Tom Chiverton]

I tried out an RC of the 2.6.31 kernel recently, and discovered that under 
that kernel the 'uid' option to mount.cifs has changed behaviour and the fix is 
to add some new other option. At a guess, in almost all cases of giving a uid= 
in fstab (different user UID locally and remotely) this change makes working 
things stop.

Based on discussion at http://bugzilla.kernel.org/show_bug.cgi?id=13861 
(initially, I thought this was a bug) it was suggested I bring the discussion 
here to see what a wider set of people thought.

Should we really not (at least) print a warning ('uid without forceuid, this 
might be wrong as behaviour has changed [url]') ? 
I spent a good while changing things on the server and client before I though 
to roll back the kernel, and there's nothing obvious in Google (or the Samba 
site) about this.

Also, why don't we just default the new option (forceuid) such that mounts 
keep working (i.e. the old behaviour), and then people who prefer the new 
behaviour can enable it if they want ? 

That seems to make more sense to me, rather than changing long established 
behaviour (it might well be for the better, but the side effects out weigh 
this, I feel). 
Swapping the default would create fstab 'spam' but that's still better than 
'printk spam', which in turn is better than having stuff break without warning 
or notice.

-- 
Tom
2 + 2 = 5 for extremely large values of 2.