Jeff Layton writes via Kernel.org Bugzilla:
FYI:
521 == EBADHANDLE
10036 == NFS4ERR_BADXDR
3 == CB_GETATTR
The cb_status gets set via decode_cb_op_status() during the decode phase and is not cleared in nfsd4_cb_prepare.
Note that the code will call rpc_restart_call() in the NFS4ERR_DELAY case, which skips the rpc_call_prepare phase when the task is restarted. So, I think this may go something like this:
Get a NFS4ERR_DELAY in a callback operation, that restarts the call but doesn't clear cb_status. Then on the next attempt, it gets a non-zero tk_status (and probably skips the decode phase). That would cause this warning to pop, I think.
In practice, I'm warming up to the idea that this entire if statement is just bogus:
if (cb->cb_status) {
WARN_ONCE(task->tk_status,
"cb_status=%d tk_status=%d cb_opcode=%d",
cb->cb_status, task->tk_status, cb->cb_ops->opcode);
task->tk_status = cb->cb_status;
}
Consider the case where we successfully decode a cb_status value but then some later field in the reply was corrupt. We will have set the cb_status, but the decode will fail which will cause the tk_status to be non-zero, and this warning will pop.
I think we should probably just remove the entire if statement above.
View: https://bugzilla.kernel.org/show_bug.cgi?id=219737#c9
You can reply to this message to join the discussion.
--
Deet-doot-dot, I am a bot.
Kernel.org Bugzilla (bugspray 0.1-dev)
