Re: Possible permissions bug on NFSv3 kernel client

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

On Tue, 2004-05-04 at 05:55, Colin Paton wrote:
> As writing to a char/block device does not perform a write
> operation *on the server* then the client should not be asking the
> server for modify/extend permission in the case of char/block devices.

Sure it should: you are asking it for write permission! If you didn't
check the for modify/extend, then you be allowing world write permission
by default.

I just checked that Sun's NFS client indeed checks both "extend" and
"modify" when attempting to write to a character device.

> > The read-only mount option does *not apply* to char/block devices such
> > as /dev/hd[a-z]*, /dev/tty*. Permission checks on open() for those
> > devices are done on the server *only* via the ACCESS rpc call.
> 
> Should vfs_permission() (as called from nfs_permission) be sufficient to
> perform this check?

No! I repeat: vfs_permission() knows nothing about any uid/gid
mapping/squashing. You cannot rely exclusively on the mode bits in NFS.

> I don't believe that it is... it is possible to write to a block device
> on a filesystem that is mounted read-only, but not to write to a block
> device on an NFS filesystem that is *exported* read-only. 

That sounds like a server bug, then.

Cheers,
  Trond