On Tue, 2013-09-10 at 21:08 +0000, Adamson, Dros wrote:
> Yes, I thought about this approach, but backed away since we do get/put_rpccred on the cl_machine_cred elsewhere (like nfs4_get_machine_cred_locked).
>
> Shall I submit a new patch that gets rid of get/put_rpccred?
Please do.
> Do we also want to change other uses of cl_machine_cred?
Not for now; let's wait for another merge cycle before doing that.
> Thanks,
> -dros
>
> On Sep 10, 2013, at 3:31 PM, "Myklebust, Trond" <Trond.Myklebust@xxxxxxxxxx>
> wrote:
>
> > On Tue, 2013-09-10 at 14:55 -0400, Weston Andros Adamson wrote:
> >> Client calls get_rpccred when referencing the machine cred, so it must call
> >> put_rpccred after passing cred to rpc layer.
> >>
> >> This patch also moves nfs4_sp4_init (formerly nfs4_state_protect) to right
> >> before the call to rpc_run_task (and wrappers of rpc_run_task) to minimize
> >> error path cleanup code.
> >>
> >
> > Do we really need this? AFAICS once the clp->cl_machine_cred is set, it
> > isn't freed until the call to nfs_free_client(). In fact, we probably
> > _rely_ on this being the case for SP4_MACH_CRED, since if the machine
> > cred were to change, we'd no longer be able to change our session or
> > lease.
> >
> > IOW: instead of referencing the cred via get/put_rpccred(), maybe we can
> > just feed the current value of clp->cl_machine_cred into the
> > msg->rpc_cred.
> >
> > --
> > Trond Myklebust
> > Linux NFS client maintainer
> >
> > NetApp
> > Trond.Myklebust@xxxxxxxxxx
> > www.netapp.com
>
--
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
