[Freeipa-devel] [PATCH] 0197 client referral support for trusted domain principal

[Martin Basti]

On 10/08/2015 12:36 PM, Alexander Bokovoy wrote:
> On Mon, 05 Oct 2015, Sumit Bose wrote:
>> On Thu, Sep 03, 2015 at 06:22:05PM +0300, Alexander Bokovoy wrote:
>>> On Thu, 03 Sep 2015, Alexander Bokovoy wrote:
>>> >Hi,
>>> >
>>> >attached patch adds support for issuing client referrals when FreeIPA
>>> >KDC is asked to give a TGT for a principal from a trusted forest.
>>> >
>>> >We return a matching forest name as a realm and KDC then returns an
>>> >error pointing a client to a direction of that realm. You can see 
>>> how it
>>> >looks with http://fpaste.org/263064/14412849/ -- it shows behavior for
>>> >both 'kinit -E -C' and 'kinit -E'.
>>> >
>>> >Note that current MIT Kerberos KDC has a bug that prevents us from
>>> >responding with a correct client referral. A patched version for 
>>> Fedora
>>> >22 is available in COPR abbra/krb5-test, a fix to upstream krb5 is
>>> >https://github.com/krb5/krb5/pull/323/ and I'm working on filing 
>>> bugs to
>>> >Fedora and RHEL versions.
>>> >
>>> >With the version in my abbra/krb5-test COPR you can test the patch 
>>> with
>>> >the help of kinit like fpaste URL above shows.
>>> After discussing with Simo and Sumit, here is updated patch that
>>> operates directly on 'search_for' krb5_principal and avoids
>>> strchr()/strrchr() and additional memory allocations -- it uses
>>> memrchr() to find '@' in the last component of the search_for principal
>>> and considers the part of the component after '@' as an enterprise 
>>> realm
>>> to check.
>>
>> The patch looks good and works as advertised. I've tested in a IPA
>> domain which trusts two different forests. All requests to the forest
>> roots and child domains where properly redirected. I tested with your
>> krb5 test build and with MIT Kerberos 1.14 which contains the needed
>> fix.
>>
>> Nevertheless there are a view points I want to discuss:
>>
>> - missing support for AD's Alternative Domain Suffixes, this is
>>  important to allow AD users to login in with their "Email-Address"
>>  (which is the typical reference for a user name with an alternative
>>  domain suffix). I think this is not strictly related to the given
>>  ticket, so it can be solved in the context of a new ticket, do you
>>  agree?
> Yes, please add a separate ticket. We need to do a bit more here:
> - extend schema to allow adding the attribute for alternative domain
>   suffixes
> - switch to use different DCE RPC call to retrieve forest trust
>   information. We can do it now that we have a call-out mechanism and
>   can isolate access to TDO credentials (this is long standing issue
>   first identified by Metze as part of cross-forest trust support for
>   Samba 4.3)
> - Make possible to associate alternative domain suffixes with IPA
>   realm. We have support for realm domains already but we don't allow
>   to use them yet for the same call as in the above item.
>
>> - referrals from outside. If I call 'kinit -E admin at IPA.DOMAIN' from a
>>  client in a trusted AD forest I get a 'Client not found in database'
>>  error because AD tends to use lower case domain names in the referal
>>  response. The request is still properly send to the IPA KDC because
>>  DNS does not care about the case. The IPA KDC processes the request
>>  with the principal 'user\@IPA.DOMAIN at ipa.domain' until
>>  ipadb_is_princ_from_trusted_realm() returns KRB5_KDB_NOENTRY becasue
>>  it detects that the principal is from the local realm. I think it
>>  would be good to enhance your patch to handle this case.
> This is a separate bug too. Please file a ticket.
>
>
>> - S4U2Self. MIT Kerberos 1.14 can now properly handle S4U2Self across
>>  domain and forest boundaries (I tested this in a setup with 2 AD
>>  forests with request going from a child domain to a child domain in
>>  the other forest. Unfortunately it is currently not working with IPA
>>  in neither direction (I guess the case issue from above might be the
>>  reason for the incoming request to fail). Here I think a new ticket
>>  would to good as well because some research might be needed and the
>>  issue might even be in the MIT code. (If you want to run some tests I
>>  can give you access to my test environment.)
> I think we want to have this working, thus a ticket is due here. This is
> something we'll most likely require for some advanced 2FA operations for
> AD users.
>
>> Let me know if you prefer to handle the issues with other tickets, then
>> I would ACK the patch as it is.
> Please file separate tickets.
>

Summit, Alexander, is this patch ACKed or not?

Martin