[Freeipa-devel] More types of replica in FreeIPA

[Dmitri Pal]

On 03/12/2012 03:38 PM, Ondrej Hamada wrote:
> On 03/08/2012 04:54 PM, Dmitri Pal wrote:
>> On 03/06/2012 01:30 PM, Ondrej Hamada wrote:
>>> On 03/06/2012 05:47 PM, Dmitri Pal wrote:
>>>> On 03/06/2012 10:59 AM, Simo Sorce wrote:
>>>>> On Tue, 2012-03-06 at 10:56 -0500, Dmitri Pal wrote:
>>>>>> [...]
>>>>>>> For a read-only KDC we need to investigate what's the better
>>>>>> solution.
>>>>>>> There are many ways we can handle the issue, one of the simplest is
>>>>>>> probably to allow the RO KDC to use a special LDAP Extended
>>>>>> operation
>>>>>>> against a full R/W server to get the user keys to sign,
>>>>>> authenticating
>>>>>>> with a special R/O KDC principal. We can also investigate how MS
>>>>>> does
>>>>>>> internal forwarding and do something similar as I suspect that's
>>>>>>> something samba4-RODC will want to implement too, so we could share
>>>>>> some
>>>>>>> of the development burden there.
>>>>>>>
>>>>>>> Simo.
>>>>>>>
>>>>>> I do not think it is a good idea for the remote RO KDC to go back to
>>>>>> the main datacenter on every authentication without some sort of
>>>>>> caching. This is why I think that some kind of SSSD integration might
>>>>>> be due. If RO KDC would just pass the authentication to SSSD in some
>>>>>> way and SSSD would do the caching in case the office gets offline. I
>>>>>> understand that authhub as is will not work as the client sends time
>>>>>> stamp encrypted with password and SSSD needs plain text password as
>>>>>> credential. I do not know if there is a way to solve this without
>>>>>> actually sending the password in the tunnel. IMO it is more important
>>>>>> to make sure that remote office can have uninterrupted operation than
>>>>>> to worry about the password being sent inside the encrypted tunnel. It
>>>>>> is something that deployment should decide and weight risks against
>>>>>> convenience.
>>>>> This is why MS does partial replication, ie allows the RODC to have
>>>>> data
>>>>> about the office users. It's complex and there are many ways to handle
>>>>> it. We need to look at various options and see how they would work
>>>>> against uses cases we want to support.
>>>>> Simo.
>>>>>
>>>> Then may be Ondrej should start with formulating use cases and
>>>> requirements based on this discussion.
>>>>
>>> I see three possible use cases here, but only two should be considered
>>> when speaking about consumer node:
>>>
>>> 1) The office that should rely on that replica is quite a big one
>>> (hundreds of employees) or many different users are authenticating
>>> against its replica or there are located admins, who need to do a lot
>>> of write-operations. --> In this case I suppose the best solution is
>>> to deploy master replica there.
>>>
>>>
>>> 2) Office that doesn't fulfil the conditions in 1) - not a desperate
>>> need for write-operations on ipa-server, but the priority is to allow
>>> (some) clients to authenticate and use available services even when
>>> the network is down. --> We need a consumer with credentials caching,
>>> authentication requests for non-cached users or write operations must
>>> be forwarded to master.
>>>
>>> 3) Office that doesn't fulfil the conditions in 1), but the priority
>>> is security, so that the consumer is not allowed to store or cache any
>>> confidential data. --> We need a consumer, authentications and write
>>> operations must be forwarded to master.
>>>
>>> If we choose the second use case, both the caching and request
>>> forwarding must be implemented. I suppose that there shouldn't be big
>>> problem to decide during the installation to turn the caching off by
>>> some option like '-no-chaching' so that the consumer could be used for
>>> the third use case as well.
>>>
>> Can you please now create a set usage scenarios for the 2) and 3).
>> User logs in and he is in cache, he is not in cache, he is redirected
>> and data is cached, he failed and account lockout data is updated
>> locally or on the other server? Admin tries to perform and IPA command
>> or ldapmodify command - what happens?
>>
>> Can those work flows be spelled out in details for caching and non use
>> cases?
>>  
>>
>
> I'll start with usage scenario for 3), it's shorter:
> All write operations and authentication requests are forwarded to the
> master
>
> Operations when connection is OK:
> ----------------------------------------------
> read -- local
> write-forwarding to master
> authentication-forwarding to master
>
> Operations when connection is BROKEN:
> -----------------------------------------------------
> read-local (only until ticket expires)
> write-not available
> authentication-not available
>
>
> Usage scenario for 2):
>
> USER'S operations when connection is OK:
> -------------------------------------------------------
> read data -> local
> write data -> forwarding to master
> authentication:
> -credentials cached -- authenticate against credentials in local cache
>                         -on failure: log failure locally, update data
> about failures only on lock-down of account
>

I am not sure is this is always right. IMO we should make it
configurable and allow the admin to decide if in this case the
authentication should be performed against central server and then
cached credentials rather than always going to the cache first.

> -credentials not cached -- forward request to master, on success cache
> the credentials
>
> USER'S operations when connection is BROKEN:
> --------------------------------------------------------------
> read data -> local
> write data -> not available
> authentication:
> -credentials cached -- authenticate against credentials in local cache
>                             -on failure: log failure locally, on
> lock-down lock account locally and update information to master when
> the connection is back on
>

I am not sure we propagate the failures and account lockout. I think it
is per server and not replicated so when the connection is restored we
should just switch to authenticating against a central server.

> -credentials not cached -- deny access -- not possible to authenticate
> user
>
> ad ADMIN'S -- I suppose that admin's credentials should not be cached
> -- it doesn't make much sense because all the operations, he's
> permitted to do, could be executed on master server only
>
> ADMIN'S operations when connection is OK:
> ---------------------------------------------------------
> read - local
> write - forwarding to master
> authentication - forwarding to master
>
> ADMIN'S operations when connection is BROKEN:
> ----------------------------------------------------------------
> read-only operations until valid ticket expires
> write - not available
> authentication - not available
>

Now when we know the scenarios (assuming we agree on those after some
review) next would be to specify how you plan to implement the
forwarding and caching. Is auth forwarding would be implemented using
KDC or there will be some kind of referral/redirect issued to the
client. No you plan to implement write forwarding> I assume some kind of
the DS plugin? It might already be there but might require some tweaks
and custom configuration.

> -- 
> Regards,
>
> Ondrej Hamada
> FreeIPA team
> jabber: ohama at jabbim.cz
> IRC: ohamada


-- 
Thank you,
Dmitri Pal

Sr. Engineering Manager IPA project,
Red Hat Inc.


-------------------------------
Looking to carve out IT costs?
www.redhat.com/carveoutcosts/



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/freeipa-devel/attachments/20120312/52b41249/attachment.htm>