On Mon, 2012-03-19 at 21:32 +0000, Adamson, Andy wrote:
> On Mar 16, 2012, at 11:12 AM, Andy Adamson wrote:
>
> > On Thu, Mar 15, 2012 at 8:27 PM, Myklebust, Trond
> > <Trond.Myklebust@xxxxxxxxxx> wrote:
> >> On Thu, 2012-03-15 at 14:40 -0400, andros@xxxxxxxxxx wrote:
> >>> From: Andy Adamson <andros@xxxxxxxxxx>
> >>>
> >>> Register a new filelayout DS rpc_action callback for sleeping on the fore
> >>> channel slot table waitq. Avoid any additional RPC FSM states
> >>> (such as timeout) when waking up to an invalid deviceid and reset
> >>> the task for io to the MDS.
> >>
> >> Why can't you simply put this call to filelayout_write_sleepon_cb in
> >> filelayout_write_prepare (before calling nfs41_setup_sequence())?
> >
> > I guess I can. Will do.
>
> Actually, I that won't work.
>
> >
> > -->Andy
> >
> >>
> >> Since nothing is going to change the task->tk_action if
> >> nfs41_setup_sequence() puts you to sleep, what value does the callback
> >> add?
>
> The rpc_action remains rpc_call_prepare. But! We want the tasks coming off the failed DS fore channel slot table queue to be redirected to the MDS nfs_XXX_prepare, not the filelayout_xxx_prepare.
So? Calling rpc_restart_call_prepare() will have exactly the same effect
if you do it within filelayout_xxx_prepare as it will within a callback.
> Note that filelayout_reset_read and filelayout_reset_write set the data->ds_clp to NULL which makes the call to nfs41_setup_sequence() in filelayout_write/read_prepare Oops…..
>
> So, I'll keep this patch as is.
NACK. Adding a parameter to the sequence arguments is way too ugly a
hack. This has nothing to do with the sequence op.
--
Trond Myklebust
Linux NFS client maintainer
NetApp
Trond.Myklebust@xxxxxxxxxx
www.netapp.com
��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
