[SSSD] [PATCHES] backend: initialize sudo only when it is enabled in services

Sumit Bose sbose at redhat.com
Tue Sep 11 16:54:34 UTC 2012


On Mon, Sep 10, 2012 at 01:52:53PM +0200, Pavel Březina wrote:
> On 09/10/2012 10:48 AM, Sumit Bose wrote:
> >On Mon, Sep 10, 2012 at 09:35:25AM +0200, Pavel Březina wrote:
> >>On 09/07/2012 11:31 AM, Sumit Bose wrote:
> >>>On Wed, Sep 05, 2012 at 03:05:49PM +0200, Pavel Březina wrote:
> >>>>0001 fixes https://fedorahosted.org/sssd/ticket/1458
> >>>>0002 fixes a memory leak when be_process_init() fails. I think it
> >>>>should be fixed, even though the backend is in this case immediately
> >>>>terminated
> >>>
> >>>Hi,
> >>>
> >>>the patches compile without errors and work as expected. Nevertheless I
> >>>have a few comments:
> >>>
> >>>0001:
> >>>  - I think it make sense to skip the initialization of the sudo target
> >>>    if there is no responder configured. But to avoid surprises on the
> >>>    user side I would recommend to do the initialization if it is
> >>>    configured explicitly in sssd.conf, i.e sudo_provider = ldap, and
> >>>    give a warning in the logs that the corresponding responder will not
> >>>    be running.
> >>
> >>I discussed this with Jakub before writing the patch. sudo_provider
> >>currently defaults to id_provider. If I get it right, you are
> >>suggesting to make the default value "none".
> >
> >No, sorry for not being clear here. I think the id_provider default is
> >right. I just wanted to say that if for whatever reasons the
> >sudo_provider parameter is used explicitly in sssd.conf the
> >sudo_provider should be initialization even if sudo is not listed in
> >services.
> >
> >bye,
> >Sumit
> >
> >>
> >>We agreed with Jakub that this makes more pressure on the user to
> >>understand the configuration so we chose not to do it. Although the
> >>configurations I've been seeing so far are mostly generated by
> >>authconfig or ipa-client-install (where are all providers set
> >>explicitly) so I don't think it is a problem after all. What do you say
> >>Jakub?
> >>
> >>If we decide to do it this way, it will also require changes in man page.
> >>
> >>>>+    ret = confdb_get_string(ctx->cdb, ctx, CONFDB_MONITOR_CONF_ENTRY,
> >>>>+                            CONFDB_MONITOR_ACTIVE_SERVICES, NULL, &services);
> >>>>+    if (ret != EOK) {
> >>>>+        DEBUG(SSSDBG_FATAL_FAILURE,
> >>>>+              ("Unable to read from confdb [%d]: %s\n", ret, strerror(ret)));
> >>>>+        return ret;
> >>>>+    }
> >>>>+
> >>>
> >>>  - I would prefer to use confdb_get_string_as_list() and use strcmp()
> >>>    later on, to be on the safe side.
> >>>
> >>>0002: ACK
> >>>
> >>>bye,
> >>>Sumit
> 
> OK, that makes sense. Patches are attached.
> 
> I moved the code into separate function.

Thank you for the new version. I only have a few comments left:
 - provider_enabled is not used
 - in the (!responder_enabled && provider != NULL && strcmp(provider, NO_PROVIDER)
   case, please just print the debug message and call load_backend_module()
   afterwards. So that the provider is loaded as requested in the config
   file. This would also avoid the strange debug message
(Tue Sep 11 18:46:38 2012) [sssd[be[ipa17.devel]]] [be_process_init_sudo] (0x0080): SUDO provider is set, but it is not listed in active services. SUDO support will not work!
(Tue Sep 11 18:46:38 2012) [sssd[be[ipa17.devel]]] [be_process_init] (0x2000): SUDO backend target successfully loaded from provider [(null)].


bye,
Sumit




More information about the sssd-devel mailing list