On Mon, May 02, 2016 at 10:47:50AM +0200, Pavel Březina wrote:
On 05/02/2016 08:53 AM, Petr Cech wrote:
> On 04/28/2016 01:41 PM, Pavel Březina wrote:
> > On 04/26/2016 09:38 AM, Petr Cech wrote:
> > > Hi list,
> > >
> > > this simple patch fixes talloc hierarchy in initializing negative caches
> > > in responders.
> > >
> > > This patch is applicable after [1].
> > >
> > > [1]
> > >
https://www.mail-archive.com/sssd-devel@lists.fedorahosted.org/msg26515.html
> > >
> > >
> >
> > The patch seems sane, but I think we should rather move ncache to rctx
> > and its initialization to common responder initialization since we use
> > it basically everywhere and when we convert responders to cache_req we
> > will use it even in those responders that don't implement it yet.
>
> Hello Pavel,
>
> thank you for respond.
>
> I think, it is possible. There are two responders which don't use
> ncache, but it is not obstacle.
Those are ssh and autofs responders. SSH looks up user and will use
cache_req which can take advantage of negative cache for free thus autofs
will be the only one where it is not used. I think it is worth it, but if
anyone disagrees I can ack this patch since it is definitely correct.
I agree it would be nice to enable them with negative cache, but
wouldn't it be more systematic to convert the ssh responder to cache_req
instead? For autofs we can file a ticket, I'm not sure we will convert
that one to cache_req at all (and the cache logic there is a bit
different from the other responders anyway IIRC..)