[SSSD] [REGRESSION] [PATCH] LDAP: Change defaults for ldap_user/group_objectsid

Jakub Hrozek jhrozek at redhat.com
Mon Nov 3 19:25:52 UTC 2014


On Mon, Nov 03, 2014 at 08:18:06PM +0100, Lukas Slebodnik wrote:
> On (03/11/14 19:51), Jakub Hrozek wrote:
> >On Mon, Nov 03, 2014 at 07:43:22PM +0100, Lukas Slebodnik wrote:
> >> On (03/11/14 17:10), Jakub Hrozek wrote:
> >> >On Sat, Oct 25, 2014 at 11:59:36PM +0200, Lukas Slebodnik wrote:
> >> >> On (24/10/14 21:14), Jakub Hrozek wrote:
> >> >> >On Fri, Oct 24, 2014 at 07:57:29PM +0200, Michal Židek wrote:
> >> >> >> On 10/24/2014 07:09 PM, Jakub Hrozek wrote:
> >> >> >> >On Fri, Oct 24, 2014 at 02:55:44PM +0200, Lukas Slebodnik wrote:
> >> >> >> >>>>-    { "ldap_group_objectsid", NULL, SYSDB_SID, NULL },
> >> >> >> >>>>+    { "ldap_group_objectsid", "objectSID", SYSDB_SID, NULL },
> >> >> >> >>>>     { "ldap_group_modify_timestamp", "modifyTimestamp", SYSDB_ORIG_MODSTAMP, NULL },
> >> >> >> >>>>     { "ldap_group_entry_usn", NULL, SYSDB_USN, NULL },
> >> >> >> >>>>     { "ldap_group_type", NULL, SYSDB_GROUP_TYPE, NULL },
> >> >> >> >>>These changes in rfc2307_user_map and rfc2307bis_user_map caused regression,
> >> >> >> >>>which was fixed by ticket https://fedorahosted.org/sssd/ticket/2383
> >> >> >> >>>
> >> >> >> >>>Dereference failed with openldapserver.
> >> >> >> >>>[sdap_nested_group_process_send] (0x2000): Dereferencing members of group [cn=topgrp,ou=Groups,dc=example,dc=com]
> >> >> >> >>>[sdap_deref_search_send] (0x2000): Server supports OpenLDAP deref
> >> >> >> >>>[sdap_x_deref_search_send] (0x0400): Dereferencing entry [cn=topgrp,ou=Groups,dc=example,dc=com] using OpenLDAP deref
> >> >> >> >>>[sdap_print_server] (0x2000): Searching 10.16.64.191
> >> >> >> >>>[sdap_get_generic_ext_step] (0x0400): calling ldap_search_ext with [no filter][cn=topgrp,ou=Groups,dc=example,dc=com].
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [objectClass]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [cn]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [userPassword]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [gidNumber]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [member]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [objectSID]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [modifyTimestamp]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [modifyTimestamp]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x1000): Requesting attrs: [uid]
> >> >> >> >>>[sdap_get_generic_ext_step] (0x2000): ldap_search_ext called, msgid = 3
> >> >> >> >>>[sdap_process_message] (0x4000): Message type: [LDAP_RES_SEARCH_RESULT]
> >> >> >> >>>[sdap_get_generic_op_finished] (0x0400): Search result: Protocol error(2), Dereference control: attribute decoding error
> >> >> >> >>>[sdap_get_generic_op_finished] (0x0040): Unexpected result from ldap: Protocol error(2), Dereference control: attribute decoding error
> >> >> >> >>>[generic_ext_search_handler] (0x0040): sdap_get_generic_ext_recv failed [5]: Input/output error
> >> >> >> >>>[sdap_deref_search_done] (0x0040): dereference processing failed [5]: Input/output error
> >> >> >> >>>[sdap_nested_group_deref_direct_done] (0x0020): Error processing direct membership [5]: Input/output error
> >> >> >> >>>[sdap_nested_done] (0x0020): Nested group processing failed: [5][Input/output error]
> >> >> >> >>>
> >> >> >> >>>
> >> >> >> >>>SSSD did equivalen of next command line search:
> >> >> >> >>>[root at host ~]# ldapsearch -x -LLL -h $SERVER -b 'dc=example,dc=com' \
> >> >> >> >>>    -E '!deref=member:objectClass,cn,userPassword,gidNumber,objectSID,modifyTimestamp,uid' \
> >> >> >> >>>    cn=topgrp
> >> >> >> >>>Protocol error (2)
> >> >> >> >>>Additional information: Dereference control: attribute decoding error
> >> >> >> >>>
> >> >> >> >>>Dereferencing failed because attribute objectSID isn't in any objectclass on
> >> >> >> >>>openldap server.
> >> >> >> >>>
> >> >> >> >>>Summary:
> >> >> >> >>>We should revert this patch. ObjectSID should not be set by default
> >> >> >> >>>in rfc2307 or rfc2307bis scheme.
> >> >> >> >>>
> >> >> >> >>BTW we had the same problem with nonstandard UUID attribute
> >> >> >> >>https://git.fedorahosted.org/cgit/sssd.git/commit/?id=dfb2960ab251f609466fa660449703835c97f99a
> >> >> >> >>
> >> >> >> >>LS
> >> >> >> >
> >> >> >> >OK, this is bad, thanks a lot for catching the problem.
> >> >> >> >
> >> >> >> >I agree that we should revert the patch as a first step. But at the same
> >> >> >> >time, the changes in 1.11 caused problems for several users running with
> >> >> >> >id_provider=ldap enrolled with an AD server.
> >> >> >> >
> >> >> >> >In retrospective, I wish we limited some of the AD-specific features to
> >> >> >> >the AD provider only, but in order to mitigate the problem, what about
> >> >> >> >setting the defaults if (and only if) we detect the server is an AD
> >> >> >> >server from rootDSE? We already do that for tokenGroups..
> >> >> >> 
> >> >> >> After discussion with Lukas I think we should not do this.
> >> >> >> Not even for tokenGroups. If user specifies his provider
> >> >> >> as ldap, we should not do background magic that changes
> >> >> >> defaults behind the scenes (based on rootDSE or anything
> >> >> >> else).
> >> >> >
> >> >> >I totally agree, see my previous mail -- "In retrospective, I wish we
> >> >> >limited some of the AD-specific features to the AD provider only".
> >> >> >
> >> >> >But I don't think we can do backwards-incompatible changes so easily
> >> >> >now. It's completely out of question for a minor update (1.12.x) and must
> >> >> >be handled with extreme care in major update (1.13).
> >> >> >
> >> >> >Additionally, we already detect search bases and other config from
> >> >> >rootDSE, what makes the default SID values any different?
> >> >> The problem was automatic enabling of features based on rootDSE.
> >> >> User used ldap provider with AD but without idmapping. The root of problem was
> >> >> that tokengroups were enabled by default (based on rootDSE) and SIDs could not
> >> >> be downloaded. Even if they were downloaded id mapping is disabled
> >> >> It would not spped upa anything.
> >> >
> >> >It wouldn't speed up anything for the first run, true, but there is a
> >> >big potential benefit for lookups of other users.
> >> >
> >> >Consider Alice and Bob are both members of two same groups. Alice logs
> >> >in first, the cache is empty, so we look up both sids. So we have a base
> >> >search for the user, a base search for the tokenGroups and then two searches
> >> >based on SID. Then Bob comes in, we check his tokenGroups attribute and
> >> >see that both SIDs are already cached, so we just link Bob's user entry
> >> >with the SIDs and we're done -- total of two lookups with scope=base.
> >> >
> >> >So this change /does/ have a nice speed benefit, however, see below..
> >> >
> there's nothing about POSIX lookups.

Yes, I describe the scenario with POSIX attributes enabled here. Without
POSIX attributes, initgroups is much simpler because you don't have to
look up the groups, ever. So with ID mapping, all you have to do is to
search the entry, then do a base search for TG and then derive GIDs from
the SIDs. So the number of searches when ID mapping is always constant -
two.

The optimization when it comes to POSIX attributes with tokenGroups is
that you can check what SIDs the user is a member of and only fetch
those SIDs that are not cached yet. The cached entries can be just
linked with.

The difference between using and not using TGs in this case is mainly
that without TGs you need to do several layers of searches. You'd start
with searching for member=user_dn first, that would yield some groups
which you need to recurse into. With TGs the group memberships are
unrolled already.

> 
> >> >> 
> >> >> We tried to solve it with this patch, but there is another solution.
> >> >> To disable tokengroups for ldap provider (rfc2307, rfc2307bis).
> >> >
> >> >I spent pretty much the whole day today triaging issues after users
> >> >upgraded to RHEL-6.6. And I tend to agree we should revert this change
> >> >for id_provider=ldap && ldap_id_mapping=False.
> >> >
> >> >Setting the SID attribute values to the AD defaults as I proposed earlier
> >> >is still something that might make sense and would solve maybe 50% of
> >> >issues. But the other 50% are strange errors like looking
> >> >up SIDs from another subdomain, where the LDAP provider has no access etc.
> >> >
> there's nothing about POSIX lookups either.
> 
> >> >> 
> >> >> If someone wants to use AD as any other ldap server we should not enable ad
> >> >> features by default.
> >> >  ^^^^^^^^
> >> >I disagree, we should take advantage of whatever features we have at
> >> >our disposal. But tokenGroups along with the LDAP provider proved to be
> >> >too fragile..
> >> >
> >> >So after experiencing the breakage from downstream, I also vote for
> >> >enabling tokenGroups if:
> >> >    * id_provider = ad, with or without ID mapping
> >> >    * ldap_provider = ldap, without ID mapping
> >> >
> >> >To make sure id_provider=ldap && mapping=true we should also set the user
> >> >and group SID attribute defaults.
> >> >
> >> It is clear that tokengroups and id mapping are related options.
> >
> >They are not. As I said above, tokengroups can be used to speed up POSIX
> >lookups, too.
> >
> I read mails few times, but I didn't find anything about POSIX lookups.
> Did I overlook something?
> 
> LS
> _______________________________________________
> sssd-devel mailing list
> sssd-devel at lists.fedorahosted.org
> https://lists.fedorahosted.org/mailman/listinfo/sssd-devel



More information about the sssd-devel mailing list