NFSv4.1: Fix the callback 'highest_used_slotid' behaviour
authorTrond Myklebust <Trond.Myklebust@netapp.com>
Tue, 2 Aug 2011 18:46:29 +0000 (14:46 -0400)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Thu, 4 Aug 2011 15:55:35 +0000 (11:55 -0400)
commit55a673990ec04cf63005318bcf08c2b0046e5778
tree71e4fb3292060553063042c6190031d104d9c1f5
parent9af7db3228acc286c50e3a0f054ec982efdbc6c6
NFSv4.1: Fix the callback 'highest_used_slotid' behaviour

Currently, there is no guarantee that we will call nfs4_cb_take_slot() even
though nfs4_callback_compound() will consistently call
nfs4_cb_free_slot() provided the cb_process_state has set the 'clp' field.
The result is that we can trigger the BUG_ON() upon the next call to
nfs4_cb_take_slot().

This patch fixes the above problem by using the slot id that was taken in
the CB_SEQUENCE operation as a flag for whether or not we need to call
nfs4_cb_free_slot().
It also fixes an atomicity problem: we need to set tbl->highest_used_slotid
atomically with the check for NFS4_SESSION_DRAINING, otherwise we end up
racing with the various tests in nfs4_begin_drain_session().

Cc: stable@kernel.org [2.6.38+]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
fs/nfs/callback.h
fs/nfs/callback_proc.c
fs/nfs/callback_xdr.c