Arjan,
We are again at the point where you suggest that the system should create
another Credential and pass this to the IdentityStore. I don't like that
and it is also confusing.
Because you can end up that a certain IdentityStore receives another type
of Credentials in the case it is used alone or in combination with other
stores. Or depending the developer defines it as ValidationType.BOTH or
ValidationType.AUTHORIZATION.
This limits the reusability of them enormously.
I modeled the system according to the needs I had the last five years or
so. Maybe it is better that you create the code and that I verify how I can
use them in those situations.
Best Regards
Rudy
On 6 September 2016 at 17:18, arjan tijms <arjan.tijms_at_gmail.com> wrote:
> Hi,
>
> On Tue, Sep 6, 2016 at 4:31 PM, Rudy De Busscher <rdebusscher_at_gmail.com>
> wrote:
>
>> Hi Arjan,
>>
>> Thanks for your feedback.
>>
>
> Thank you too ;)
>
>
>>
>> A quick reply to your comments.
>>
>> I wonder, would it be possible to omit the CallerPrincipal argument here?
>>> There's already a caller in the Credential, and I'm afraid this would
>>> confuse people who have to implement the interface.
>>
>>
>> I don't like the caller in the Credential interface
>> - Credentials are user supplied and should be separated from the "System
>> determined values". The caller(Principal) is something which is determined
>> based on the credentials and thus I should not mix these 2 sources.
>>
>
> Perhaps, but if the store is asked for authorization only no actual user
> supplied credential is needed, just the caller name to get the groups by,
> which can be put in a dedicated system credential.
>
>
>
>> - The current implementation of UsernamePasswordCredential uses the
>> user-supplied username as value.
>>
>
>
> If the UsernamePasswordCredential is used for authentication, then this
> can be passed in as-is. If the caller is authenticated via an earlier call
> to an earlier identity store, then the handler knows this. If a followup
> store is configured for authorization only, the handler knows this, and can
> pass in the right system credential. This system credential can only
> contain the caller principal.
>
> So in these two situations only ever 1 caller name/principal is needed. By
> passing them in both at the same time, depending on the situation one or
> the other is used.
>
> The problem now is that a store that doesn't want to have anything to do
> with multi stores, sees two incoming principals (names). The one in the
> credential, and the one passed in as a separate argument. We have to tell
> users that unless they want to support authorise only and be multi store
> capable, they have to ignore the second parameter. IMHO this goes against
> the simplest possible design.
>
>
>
>> So there is no way to determine if the 'Caller' is already identified or
>> not. And this is a key principal when the Authorization and Authentication
>> is separated in different identityStores.
>>
>
> I think determining if the Caller is identified is a responsibility that
> should lie solely with the handler. The store just does what it's asked to
> do based on the incoming data and whether it's configured to do
> authentication, authorization or both. The handler on its turn knows
> everything, via the returned status in the CredentialValidationResult and
> the returned Principal of each store.
>
>
>
>> - The unique caller identification doesn't need to be the same as
>> user-supplied username (you can decide to use the internal id/sequence,
>> email address, LDAP node cn value, ...)
>>
>
> Indeed, and that's a key design element in the current store. The
> credential may contain a caller name (like e.g. in
> UsernamePasswordCredential), but the CredentialValidationResult contains
> both CallerPrincipal and CallerGroups return values. That CallerPrincipal
> is then taken as the one belonging to the authenticated identity.
>
> As an example: suppose we have the LDAPIdentityStore set to authenticate
> against and the DatabaseIdentityStore set to get the groups from. Now
> suppose we couple that to a mechanism that uses the
> UsernamePasswordCredential.
>
> Now let the caller authenticate with caller name "pete" and password
> "secret", then:
>
> 1. The handler constructs a UsernamePasswordCredential and passes that
> into the LDAPIdentityStore.
>
> 2. The LDAPIdentityStore validates that and returns a VALID result with
> CallerPrincipal(name="cn=pete, dn=peter parker") and no groups.
>
> 3. The handler then takes CallerPrincipal(name="cn=pete, dn=peter
> parker"), puts that into a suitable system Credential and passes that into
> the DatabaseIdentityStore.
>
> 4. The DatabaseIdentityStore then returns the groups associated with
> CallerPrincipal(name="cn=pete, dn=peter parker") say {"foo", "bar", "kaz"}
>
> 5. The handler now combines CallerPrincipal(name="cn=pete, dn=peter
> parker") and {"foo", "bar", "kaz"} puts that into the final
> CredentialValidationResult and returns it.
>
> Hope the above demonstrates that it should not be needed to have 2 caller
> principals passed in, and that the stores can do their work without having
> knowledge of what happened before.
>
>
>>
>> callerPrincipal.getName() would be null here
>>
>> This should never return null. callerPrincipal being not null means a
>> successful Authorization is carried out and thus the user should be
>> identifiable (and null is thus not an option)
>>
>
> Indeed, that's what I thought, so therefor if it can never be null perhaps
> the store implementations don't have to check against null?
>
> Kind regards,
> Arjan Tijms
>
>
>
>
>
>
>>
>> maybe it's better to have ValidationType directly on the
>>> EmbeddedIdentityStoreDefinition (and other definitions)?
>>
>>
>> That is indeed an option.
>>
>> Best regards
>> Rudy
>>
>> On 4 September 2016 at 21:28, arjan tijms <arjan.tijms_at_gmail.com> wrote:
>>
>>> Hi Rudy,
>>>
>>> Looks much improved, thanks!
>>>
>>> I have a few comments, but some are just cosmetic and could be looked at
>>> later.
>>>
>>> The most important one concerns the IdentityStore interface, it now has
>>> this method:
>>>
>>> public CredentialValidationResult validate(Credential credential,
>>> CallerPrincipal callerPrincipal) {
>>>
>>> I wonder, would it be possible to omit the CallerPrincipal argument
>>> here? There's already a caller in the Credential, and I'm afraid this would
>>> confuse people who have to implement the interface. I may also give the
>>> impression to implementors that they have to take the history of what
>>> happened before with other stores somehow into account. I think it would be
>>> best if the store itself is totally unaware of the concept of multiple
>>> stores, since that's what the handler should do. Since the handler is
>>> replaceable, a store implementation can't assume anything about it.
>>>
>>> In the implementations of the provided stores (e.g. the
>>> EmbeddedIdentityStore), there's this comment with some code below it.
>>>
>>> // We check also if caller != null to be sure the Authentication by
>>> another IdentityStore succeeded.
>>>
>>> If you look at the code flow that applies to it's:
>>>
>>> String caller = null;
>>> ...
>>> // We are Authorize Only mode, so get the caller determined
>>> previously.
>>> if (callerPrincipal != null) {
>>> caller = callerPrincipal.getName();
>>> }
>>>
>>> if (authenticated && caller != null) {
>>> ...
>>> }
>>>
>>> return INVALID_RESULT;
>>>
>>> So it callerPrincipal or callerPrincipal.getName() would be null here,
>>> the store would return INVALID_RESULT. But in that case, wouldn't it be
>>> better if the handler did not call the store at all? That would safe the
>>> store from having to check this case or even having to think of what other
>>> stores before it did. In other words, define in the contract and spec that
>>> if the store is used for authorization, the passed in caller is not null.
>>> Would that make sense?
>>>
>>> The idea here would be that a store is kept as simple as it can possibly
>>> be, as an often heard complaint is that all proprietary stores are quite
>>> complex. The multi-store algorithm should then 100% be with the handler.
>>>
>>> A somewhat cosmetic and implementation thing, but maybe still worth it
>>> to look at is this pattern in the stores:
>>>
>>> private EmbeddedIdentityStoreDefinition embeddedIdentityStoreDefinit
>>> ion;
>>>
>>> private ValidationType validationType;
>>>
>>> private void determineValidationType() {
>>> validationType = ValidationType.BOTH;
>>> if (embeddedIdentityStoreDefinition.authenticateOnly()) {
>>> validationType = ValidationType.AUTHENTICATION;
>>> } else {
>>> if (embeddedIdentityStoreDefinition.authorizeOnly()) {
>>> validationType = ValidationType.AUTHORIZATION;
>>> }
>>> }
>>> }
>>>
>>> public ValidationType validationType() {
>>> return validationType;
>>> }
>>>
>>> Seeing this code repeating, maybe it's better to have ValidationType
>>> directly on the EmbeddedIdentityStoreDefinition (and other definitions)?
>>>
>>> Then it could just be:
>>>
>>> public ValidationType validationType() {
>>> return embeddedIdentityStoreDefinition.validationType();
>>> }
>>>
>>> Another small thing, also cosmetic mostly, but there's now this kind of
>>> check often:
>>>
>>> if (validationType == ValidationType.AUTHENTICATION || validationType ==
>>> ValidationType.BOTH) {
>>> ...
>>> }
>>>
>>> What about making two helper methods on the ValidationType enum like e.g.
>>>
>>> public boolean isAuthentication() {
>>> return isOneOf(this, AUTHENTICATION, BOTH);
>>> }
>>>
>>> then the code using it could just do:
>>>
>>> if (validationType.isAuthentication()) {
>>> ...
>>> }
>>>
>>> Hope this feedback is useful ;)
>>>
>>> Kind regards,
>>> Arjan Tijms
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> On Sun, Sep 4, 2016 at 8:07 PM, Rudy De Busscher <rdebusscher_at_gmail.com>
>>> wrote:
>>>
>>>> Hi all,
>>>>
>>>> This is the updated multi IdentityStore based on earlier feedback on
>>>> this mailing list.
>>>>
>>>>
>>>> *Why multiple stores?*
>>>> There are various use cases where multiple identity stores can be
>>>> useful.
>>>> And it allows us to decouple the authentication and authorization parts
>>>> of the current mechanism. So that groups can be retrieved by another store
>>>> then the credential checking.
>>>>
>>>> *Examples*
>>>> * When your application should allow users which are defined in
>>>> multiple stores, like an LDAP for your internal users and a Database store
>>>> for the external ones.
>>>> * Users are authenticated by an OAuth2 provider and the groups are
>>>> retrieved from a database table.
>>>> * The first IdentityStore based on OAuth2 allows any Google account,
>>>> but the second store limits the access to a certain google domain.
>>>>
>>>> *IdentityStore interface*
>>>>
>>>> * CredentialValidationResult validate(Credential credential,
>>>> CallerPrincipal callerPrincipal);*
>>>>
>>>> * default int priority() {*
>>>> * return 100;*
>>>> * }*
>>>>
>>>> * default ValidationType validationType() {*
>>>> * return ValidationType.BOTH;*
>>>> * }*
>>>>
>>>> *validate method*
>>>>
>>>> the CallerPrincipal parameter contains previous validated Caller
>>>> information (authentication) if any. Value can be null.
>>>>
>>>> *ValidationType method*
>>>>
>>>> ValidationType contains meta information for the IdentityStore handler.
>>>> It defines if the store does authentication only
>>>> (ValidationType.AUTHENTICATION), authorization only
>>>> (ValidationType.AUTHORIZATION) or both; the default.
>>>> This information is used by the handler to define if the store should
>>>> be called. For example, the default IdentityStore handler does not call a
>>>> store who does Authorization only when no successful authentication took
>>>> place.
>>>>
>>>> *priority method *
>>>>
>>>> determines the order in the case when there are multiple ones. They are
>>>> ordered by ascending values, meaning that the IdentityStore with the lowest
>>>> value is processed first.
>>>>
>>>> *Status rules*
>>>>
>>>> Stores should return the following Status (contained in the
>>>> CredentialValidationResult result parameter)
>>>> *NOT_VALIDATED*: Type of Credential isn't compatible with the ones
>>>> expected by the store.
>>>> *INVALID* : In case the store does Authentication, the validation of
>>>> the credential failed.
>>>> *AUTHENTICATED* : In case the store does Authentication only and the
>>>> validation succeeded.
>>>> *VALID* : In case the store does Authorization and groups are supplied
>>>> for the Caller (can be an empty list indicating no groups assigned to
>>>> caller)
>>>>
>>>> So When Store does handle both (Authentication and Authorization) it
>>>> should return NOT_VALIDATED, INVALID or VALID
>>>> In case store does Authentication only; it is NOT_VALIDATED, INVALID or
>>>> AUTHENTICATED
>>>> In case store does Authorization only; it is NOT_VALIDATED or VALID
>>>> (the assumption is that when store can handle the type of Credentials, the
>>>>
>>>> *IdentityStore Handler *
>>>>
>>>> Default IdentityStore Handler uses the following algorithm
>>>> 1) Order all identityStores based on the priority value
>>>> 2) Loop over all IdentityStores unless one of the stores return the
>>>> status INVALID
>>>> 3) Call a certain IdentityStore when
>>>> * The Store does Authentication, regardless of the result of the
>>>> calls to previous IdentityStores.
>>>> * The Store does Authorization only and another Store already
>>>> successful validated the Credentials (Result is AUTHENTICATED or VALID)
>>>> 4) The result of the call to the IdentityStore is combined with
>>>> 'previous' result of the there calls to the IdentityStores.
>>>> * Groups are added to the already defined groups of the caller.
>>>>
>>>> *Issues*
>>>>
>>>> The current default IdentityStore Handler handler can't handle this use
>>>> case
>>>> - Caller can be authenticated by one IdentityStore (for example LDAP or
>>>> Database)
>>>>
>>>> This is because the loop stops when an INVALID result is returned by a
>>>> IdentityStore. This is required to veto another successful authentication
>>>> like in this use case
>>>> - The first IdentityStore based on OAuth2 allows any Google account,
>>>> but the second store limits the access to a certain google domain.
>>>>
>>>> So probably we need some configuration to be able to combine both use
>>>> cases in one algorithm.
>>>>
>>>> *Code status*
>>>>
>>>> The current code needs probably some more tests and verification. I try
>>>> to write them in the following weeks.
>>>>
>>>> Changes for this updated proposal can be viewed in this commit
>>>> https://github.com/rdebusscher/soteria/commit/afa4cecf626266
>>>> d0ae400f806b0358ed7b07bc61
>>>> And are added to the first pull request to the Soteria repository.
>>>>
>>>> Best regards
>>>> Rudy
>>>>
>>>>
>>>> --
>>>> You received this message because you are subscribed to the Google
>>>> Groups "Java EE Security API - JSR 375 - Experts" group.
>>>> To unsubscribe from this group and stop receiving emails from it, send
>>>> an email to jsr375-experts+unsubscribe_at_googlegroups.com.
>>>> To post to this group, send email to jsr375-experts_at_googlegroups.com.
>>>> To view this discussion on the web visit https://groups.google.com/d/ms
>>>> gid/jsr375-experts/CAL%2Bwt-5QmUqZiZ9V2uTYXAE-A%2B%2BFrzNwYK
>>>> kYxmZBxaHyYLq15g%40mail.gmail.com
>>>> <https://groups.google.com/d/msgid/jsr375-experts/CAL%2Bwt-5QmUqZiZ9V2uTYXAE-A%2B%2BFrzNwYKkYxmZBxaHyYLq15g%40mail.gmail.com?utm_medium=email&utm_source=footer>
>>>> .
>>>> For more options, visit https://groups.google.com/d/optout.
>>>>
>>>
>>>
>>
>