[gpfsug-discuss] question about why unix extensions = no is recommended when using samba + gpfs?

[Jonathan Buzzard]

On 03/04/14 18:23, Sabuj Pattanayek wrote:

[SNIP]

>
> It doesn't work if the ACL on the file is POSIX or has been promoted to
> NFS4. In any case, we have need for the ACLs support to be both POSIX
> and NFS4 where needed for the window shares and we can't/won't split up
> the filesystem.
>

Client side Linux has support NFSv4 or more specifically NFSv4.1 for 
some time. The issue is the fact that the RichACL patches seem to have 
been stuck in limbo for years now.

Not that I have tried it even in development, but there is an 
alternative to the Linux kernel NFS server which sucks anyway as 
visibility into what is going on is near none existent.

It's the NFS-Ganesha server and according to the documentation it 
supports full NFSv4.1 and has GPFS support for a backend file system.

     https://github.com/nfs-ganesha/nfs-ganesha/wiki

That could potentially remove your "requirement" for Posix ACL's. That 
is you could go full NFSv4 ACL's in GPFS and an area could be accessed 
either by NFSv4 or SMB without the ACL's in GPFS getting mucked up. It 
is however not clear if NFSv4 ACL's are supported yet. I note that the 
latest versions of CTDB have support for NFS-Ganesha.

>
>     I suggest you use a development/test GPFS cluster to verify it.
>
> Yes, we're going to need a test/dev cluster for testing these for future
> upgrades, etc so I will try an FS that's pure nfs4 ACLs only and see if
> it behaves any differently.

I never cease to be amazed that people are doing large scale file 
servers with Linux/GPFS/Samba and don't have a test/dev cluster.

JAB.

-- 
Jonathan A. Buzzard                 Email: jonathan (at) buzzard.me.uk
Fife, United Kingdom.