shiro-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alan D. Cabrera" <l...@toolazydogs.com>
Subject Re: Atlassian Crowd based realm
Date Sat, 27 Dec 2008 17:33:28 GMT

On Dec 16, 2008, at 6:22 AM, Les Hazlewood wrote:

> I'm slowly catching up on emails, so please forgive the delay ;)

Heh heh, please forgive my delay.  :)

> I'm not quite sure of the usage patterns of Crowd and its token  
> checking,
> but a clean solution might depend on the architecture.
>
> For example, if Crowd is anything like JSecurity, the 'token' is  
> sent over
> the wire attached to each RPC call and/or web request.  It is then  
> checked
> on the server-side for validity.  If not valid, it throws a specific
> exception indicating why it failed, preventing the invocation/ 
> request/etc.
>
> A more performant mechansim would probably be, on the client side (I'm
> calling the CrowdRealm the 'client side' and the Crowd Server the  
> 'server
> side' in this example), to just send the token across the wire each  
> time in
> conjunction with the intended request, without checking its validity  
> first.
>
> Then the filter or method interceptor on the client side would check  
> for an
> specific exception thrown by the server side related to token  
> invalidity and
> then show the user something that explains why the request/ 
> invocation failed
> (e.g. redirect to a nice page or show a popup dialog, etc).
>
> You'd have to show this information anyway if you proactively  
> checked the
> token.  Why go through the 'hit' of checking token validity before an
> invocation starts when that valididty will be checked by the server  
> anyway?
> Seems like one more round-trip hop that can be avoided.
>
> Of course, this assumes that checking validity requires a server- 
> side call.
> If not, that proactive checking might be viable, but I don't know  
> how Crowd
> works...

Your initial thoughts are along the same lines as mine.  The validity  
of the ticket is tested by a peer on first arrival.  This test  
involves communication with the Crowd server and can be considered  
heavy weight.  Usually peers will cache the result of the security  
check so that they will not have to subsequently check the ticket with  
the Crowd server.

Now, the Crowd server has a mechanism where a ticket can be  
invalidated.  Let's say a Wall Street trader has been let go and his  
privileges have to be removed immediately rather than wait until all  
his tickets expire.  Any subsequent security checks against the Crowd  
server will fail.  But remember that a peer may be caching the initial  
security check results.  What to do?  One simple solution is to have  
the peer do a security check w/ the Crowd server for every call it  
receives; obviously a resource intensive solution.  A different  
solution would be to have the peer check after some period, say five  
minutes.  This could be an acceptable compromise.

Jeremy's idea of using a CrowdFilter that would have its security  
check period configurable sounds like the way to go.

Hope this explains my current thinking.


Regards,
Alan



> On Mon, Nov 24, 2008 at 2:08 PM, Jeremy Haile <jhaile@fastmail.fm>  
> wrote:
>
>> Yeah - I think the mixup was largely on the user of the word  
>> "credential"
>> We use that word to mean the "secret" information used during
>> authentication, such as a password or private key.  In this case,  
>> in the
>> JSecurity parlance I would probably store the token as a  
>> "principal" via the
>> CrowdRealm.
>>
>> I agree that checking the validity of a token is a common security  
>> pattern,
>> especially for SSO setups.  I also like the idea of providing some  
>> easy hook
>> or way to do this in the JSecurity core.  It seems to me that this  
>> should
>> somehow fit in as a custom CrowdFilter that is plugged into the
>> JSecurityFilter and has the ability to route the user to a login  
>> URL if the
>> token is no longer valid - or perhaps to simply expire the users  
>> session if
>> the token is invalid, thus letting a later filter route the user to  
>> the
>> login URL based on the invalid (and now unauthenticated) session.
>>
>> Jeremy
>>
>>
>>
>>
>> On Nov 24, 2008, at 10:10 AM, Alan D. Cabrera wrote:
>>
>> Crowd calls its a token I would probably call it a ticket.  There's a
>>> method in the Crowd API that you can call to make sure the ticket  
>>> is still
>>> good.  The ticket is a handle to a "login session" at the Crowd  
>>> server.  You
>>> can invalidate that session using the Crowd console to the Crowd  
>>> server and,
>>> so, that the next time you check to see if the ticket is still  
>>> good the
>>> Crowd API will say "no".  So, you can have a filter that checks the
>>> token/ticket for every page hit/call or after a certain period  
>>> expires, say
>>> 5 minutes.
>>>
>>> This is a fairly common security pattern; at least one that I've  
>>> run into
>>> on a regular basis.  I think we should include it in JSecurity  
>>> proper
>>> instead of forcing people to make ad hoc solutions; not that  
>>> JSecurity
>>> doesn't provide adequate methods to do so.
>>>
>>>
>>> Regards,
>>> Alan
>>>
>>> On Nov 24, 2008, at 5:51 AM, Les Hazlewood wrote:
>>>
>>> Hi Alan,
>>>>
>>>> This is cool - I like seeing new realms :)
>>>>
>>>> What do you mean exactly when you say "one can invalidate the token
>>>> session from the Crowd console".  What is a 'token session'?
>>>>
>>>> It sounds as if this token would be a principal - an identifying
>>>> attribute of the current user.  A credential, strictly speaking, is
>>>> something that verifies a user identity by matching a known  
>>>> value.  Is
>>>> the token really a credential?
>>>>
>>>> Or perhaps it is just something attributed to the current user -  
>>>> not
>>>> an identifying attribute like a username or SSN - I dunno.  If  
>>>> that is
>>>> the case, there's nothing wrong with just sticking it in the  
>>>> session
>>>> either (Subject.getSession()).  The only benefit of including it  
>>>> as a
>>>> principal in the PrincipalCollection is that would be serialized  
>>>> as a
>>>> cookie if cookies are enabled, thereby not using a session at all  
>>>> (can
>>>> be a good thing depending on architecture, as I'm sure you know).
>>>>
>>>> In any case, I think checking it on a regular basis to make sure  
>>>> it is
>>>> still valid is a behavior specific to Crowd, not JSecurity, so I  
>>>> think
>>>> it makes sense to do this checking in a filter.  If it extends  
>>>> one of
>>>> the JSecurity Filter classes, then you can do anything they can do,
>>>> like redirect to the login page, etc.
>>>>
>>>> HTH,
>>>>
>>>> Les
>>>>
>>>> On Sun, Nov 23, 2008 at 4:00 PM, Alan D. Cabrera <list@toolazydogs.com

>>>> >
>>>> wrote:
>>>>
>>>>> So, I've checked in my first crack at a crowd based realm.  I  
>>>>> have a few
>>>>> questions.
>>>>>
>>>>> As you may have noticed, I used enums to direct the behavior of  
>>>>> the
>>>>> realm.
>>>>> It's not sure to me that this will work within the framework as  
>>>>> it now
>>>>> stands.
>>>>>
>>>>> When you log into Crowd you get a token.  This token can be  
>>>>> checked on a
>>>>> regular basis to make sure that it's still valid; one can  
>>>>> invalidate the
>>>>> token session from the Crowd console.  It's not clear to me  
>>>>> where I can
>>>>> place this check in JSecurity.  I'm thinking that I'll need to  
>>>>> write my
>>>>> own
>>>>> filter that gets the Subject and obtains the credentials object  
>>>>> that I
>>>>> returned during authentication.  In this credential is the token  
>>>>> and I
>>>>> can
>>>>> use that to check the validity of it.
>>>>>
>>>>> Thoughts?
>>>>>
>>>>>
>>>>> Regards,
>>>>> Alan
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>
>>


Mime
View raw message