[Freeipa-devel] #4389: DS deref broken after ACI refactoring

[Jakub Hrozek]

On Fri, Jun 20, 2014 at 04:45:45PM +0200, Martin Kosek wrote:
> On 06/20/2014 04:24 PM, Jakub Hrozek wrote:
> > On Fri, Jun 20, 2014 at 04:06:16PM +0200, Martin Kosek wrote:
> >> Hello all,
> >>
> >> I would like to discuss what should we do with the latest issue we found in
> >> SSSD-DS communication which is broken after the ACI refactoring.
> > 
> > It's not just SSSD-DS communication, any client, including ldapsearch
> > currently fails.
> 
> Sure.
> 
> >> I was working with Ludwig, there is a problem in the way how deref plugin
> >> checks the access to the referenced entry. Instead of checking the target entry
> >> itself, Ludwig found out that the deref plugin checks a dummy entry created
> >> from the dereferenced DN, not the real entry. Details in DS ticket:
> >> https://fedorahosted.org/389/ticket/47821
> >>
> >> Previously, we allowed read access globally so it worked fine. Now, when we
> >> have targeted ACIs using objectclass targetfilter, the access control goes
> >> wrong, deref plugin does not return all attributes and SSSD does not work (see
> >> 4389).
> >>
> >> Question is, what should we do in 4.0. We could have the DS team to fix the
> >> deref plugin, but this would break SSSD connected to old RHEL/CentOs 6.x
> >> replicas which would not have the fix. So we need to be cautious about this one.
> > 
> > Why? SSSD Simply uses a client control. See
> > http://tools.ietf.org/html/draft-masarati-ldap-deref-00 for all the
> > details.
> 
> Because this bug affects all client machines enrolled in FreeIPA.
> 
> >> I see couple ways:
> >> 1) Fix DS deref plugin in F20 and next RHEL 6.x and specify the RHEL-6.x as
> >> minimal version of FreeIPA/DS required when replicating with FreeIPA 4.0. This
> >> option is a bit clumsy.
> > 
> > The fact that this 'fix' seems to be backwards incompatible sounds
> > strange to me. How exactly did you intend to fix the plugin? If the fix
> > was about changing DS to check the target entry instead of some dummy
> > entry, then I see no impact on the client.
> 
> There is no impact on clients connected to the "fixed DS". This is the scenario
> I am concerned about:
> 
> User has RHEL/CentOS 6.x IPA server and wants to try the new nice and shiny
> FreeIPA 4.0. He installs the FreeIPA 4.0 replica (with fixed DS), the replica
> upgrades the ACIs to the new model. SSSD connected to FreeIPA 4.0 replica will
> work, SSSD connected to the old RHEL-6.x replica will not.

Ah, I didn't realize the bug could propagate to released versions via
replicas, sorry.

> 
> >> 2) Add temporary ACIs allowing access to attributes that SSSD needs for deref
> >> calls. I tested it with Jakub's example call and it fixed this query:
> >>
> >> # ipa permission-add --subtree
> >> cn=computers,cn=accounts,dc=idm,dc=lab,dc=eng,dc=brq,dc=redhat,dc=com
> >> --right={read,search,compare} --attrs={objectclass,memberof,managedby}
> >> --bindtype all deref_managedby
> >> # kinit -kt /etc/krb5.keytab
> >>
> >> # ldapsearch -Y GSSAPI -h vm-236.idm.lab.eng.brq.redhat.com -b
> >> fqdn=vm-086.idm.lab.bos.redhat.com,cn=computers,cn=accounts,dc=idm,dc=lab,dc=eng,dc=brq,dc=redhat,dc=com
> >> -E 'deref=managedBy:objectClass'
> >> ...
> >> dn: fqdn=vm-086.idm.lab.bos.redhat.com,cn=computers,cn=accounts,dc=idm,dc=lab,
> >>  dc=eng,dc=brq,dc=redhat,dc=com
> >> control: 1.3.6.1.4.1.4203.666.5.16 false MIQAAAEgMIQAAAEaBAltYW5hZ2VkQnkEaGZxZ
> >>  G49dm0tMDg2LmlkbS5sYWIuYm9zLnJlZGhhdC5jb20sY249Y29tcHV0ZXJzLGNuPWFjY291bnRzLG
> >>  RjPWlkbSxkYz1sYWIsZGM9ZW5nLGRjPWJycSxkYz1yZWRoYXQsZGM9Y29toIQAAACfMIQAAACZBAt
> >>  vYmplY3RDbGFzczGEAAAAhgQJaXBhb2JqZWN0BA1pZWVlODAyZGV2aWNlBAZuc2hvc3QECmlwYXNl
> >>  cnZpY2UEB3BraXVzZXIEB2lwYWhvc3QEDGtyYnByaW5jaXBhbAQPa3JicHJpbmNpcGFsYXV4BAppc
> >>  GFzc2hob3N0BAN0b3AEFGlwYVNzaEdyb3VwT2ZQdWJLZXlz
> >> # managedBy: <objectClass=ipaobject>;<objectClass=ieee802device>;<objectClass
> >>  =nshost>;<objectClass=ipaservice>;<objectClass=pkiuser>;<objectClass=ipahost
> >>  >;<objectClass=krbprincipal>;<objectClass=krbprincipalaux>;<objectClass=ipas
> >>  shhost>;<objectClass=top>;<objectClass=ipaSshGroupOfPubKeys>;fqdn=vm-086.idm
> >>  .lab.bos.redhat.com,cn=computers,cn=accounts,dc=idm,dc=lab,dc=eng,dc=brq,dc=
> >>  redhat,dc=com
> >>
> >> Jakub, what else would we need to allow? After this change, login/sudo seemed
> >> to work for me on F20.
> > 
> > Are you asking about the list of attributes we dereference or the list
> > of attributes we read from the dereferenced entries?
> 
> Actually both, so that we know for what type of objects and what attributes we
> would need to allow access.

I think we should just make a note to self to allow users to fix the
ACIs manually should they run into the problem while being unable to
upgrade for whatever reason.

So we only seem to dereference member and memberof. We dereference either user
groups to get users, host groups to get hosts. For hosts we are
interested about these attributes:
    "ipa_host_object_class"
    "ipa_host_name"
    "ipa_host_fqdn"
    "ipa_host_serverhostname"
    "ipa_host_member_of"
    "ipa_host_ssh_public_key"
    "ipa_host_uuid"

For users and groups, the list is longer and can be found here:
https://git.fedorahosted.org/cgit/sssd.git/tree/src/providers/ipa/ipa_opts.h#n166

Look for ipa_user_map and ipa_group_map.

But in general I agree with Simo that we shouldn't spend too much time
on this when the DS is fixed. 


> 
> > For IPA we only care about memberof, but keep in mind that attribute
> > maps in SSSD are configurable.
> 
> Hm, makes the option 2) even more challenging...
> 

But because the ACIs would only be applied on IPA servers, I think we
can limit ourselves to the IPA schema only for the workaround.