[SSSD] [PATCH] nested groups: fix group lookup hangs if member dn is incorrect

Jakub Hrozek jhrozek at redhat.com
Mon Jan 28 16:30:36 UTC 2013


On Mon, Jan 28, 2013 at 11:03:41AM +0100, Pavel Březina wrote:
> On 01/23/2013 03:51 PM, Pavel Březina wrote:
> >On 01/23/2013 03:46 PM, Pavel Březina wrote:
> >>On 01/23/2013 03:21 PM, Jakub Hrozek wrote:
> >>>On Wed, Jan 23, 2013 at 02:56:59PM +0100, Pavel Březina wrote:
> >>>>https://fedorahosted.org/sssd/ticket/1783
> >>>>
> >>>>The proper way to fix this would be to rewrite return codes of involved
> >>>>functions, so that we call tevent_req_done() and tevent_req_post() in
> >>>>_send(). However, rewriting return codes may be tricky and
> >>>>given the circumstances (blocker for 6.4 and deadline), I chose to call
> >>>>tevent_req_post() directly and open a new ticket for the return codes:
> >>>>
> >>>>https://fedorahosted.org/sssd/ticket/1784
> >>>
> >>>I'm not sure if this is the right approach if there already *is* a
> >>>callback, please check if this fix also works for scenarios where there
> >>>is both correct and incorrect DN.
> >>
> >>Right. Good catch. Tevent code says that the callback would be
> >>triggered twice. Tevent doesn't provider any function to determine
> >>whether there is a callback set or not. Possible hack:
> >>
> >>tevent_req_post() adds new immediate event, that fires the callback.
> >>
> >>             tevent_req_done(req);
> >>             tevent_req_set_callback(req, NULL, NULL);
> >>             tevent_req_post(req, state->ev);
> >>
> >>If a callback is set, it will be triggered with tevent_req_done() and
> >>then removed. Immediate event will be processed but basically as noop.
> >>
> >>If it is not set, the callback will be triggered in immediate event.
> >
> >I'm taking back my words. Callback usually frees the request so this
> >would likely end up with crash.
> 
> I'm attaching new patch. It is not pretty but it is the least
> invasive way that crossed my mind.
> 

Ack

While this is not the optimal solution, it is one that works and
refactoring the group processing to be more readable and maintanable
(https://fedorahosted.org/sssd/ticket/1784) is too invasive for 1.9.4 at
this point. We will get to the full refactor in 1.10 development phase.



More information about the sssd-devel mailing list