nfsd4: don't pin clientids to pseudoflavors
authorJ. Bruce Fields <bfields@redhat.com>
Tue, 21 Aug 2012 16:48:30 +0000 (12:48 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 21 Oct 2012 16:32:38 +0000 (09:32 -0700)
commit55db1c6363a84f7dcf9479ec791e390c0758aa05
tree34b78180ba40a3ee36da6f01dbfab6f995b2d7ab
parent9fa2b82e5592a7aa7a63b7f6a32c5aa9e580643a
nfsd4: don't pin clientids to pseudoflavors

commit 68eb35081e297b37db49d854cda144c6a3397699 upstream.

I added cr_flavor to the data compared in same_creds without any
justification, in d5497fc693a446ce9100fcf4117c3f795ddfd0d2 "nfsd4: move
rq_flavor into svc_cred".

Recent client changes then started making

mount -osec=krb5 server:/export /mnt/
echo "hello" >/mnt/TMP
umount /mnt/
mount -osec=krb5i server:/export /mnt/
echo "hello" >/mnt/TMP

to fail due to a clid_inuse on the second open.

Mounting sequentially like this with different flavors probably isn't
that common outside artificial tests.  Also, the real bug here may be
that the server isn't just destroying the former clientid in this case
(because it isn't good enough at recognizing when the old state is
gone).  But it prompted some discussion and a look back at the spec, and
I think the check was probably wrong.  Fix and document.

Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/nfsd/nfs4state.c