A Handful of 2.6.21.5 bugs

[J. Bruce Fields]

Just cleaning out my inbox on a friday night....:

On Thu, Jun 14, 2007 at 08:00:10AM -0700, John McCorquodale wrote:
> One more, almost forgot:
> 
> My / is mounted nfs4, as I mentioned.  When it's freshly mounted, it doesn't
> seem to know that nods in /dev are nods.  E.g. I do
> 
>   $ echo hi > /dev/null
> 
> and get "file exists".  I do
> 
>   $ dd if=/realdev/zero of=/tmp/foo bs=1M count=1
>   dd: opening `zero': Invalid argument
> 
> and get "invalid argument".  Needless to say, a system can't boot if it
> can't open /dev/null for reading.  So, I copy the real contents of /dev from
> /realdev (in nfs4 /) with tar into a tmpfs /dev in early boot before invoking
> init.
> 
> I am _not_ mounting ENODEV (mount syscall flags is 0, nfs4 mountdata flags
> is NFS4_MOUNT_INTR).  Any clues what might be going on here?

I can't reproduce this on a recent kernel (client and server running
2.6.24-rc8), so I'm assuming that either this was fixed, or there was
something odd about this setup.

--b.