Library for supporting user authorisation on microservices.
Deprecated the 'name' Retrieval as at 04/2024 following a Tech Blog Post dated 2024-02-26.
Added 5 new allowlisted Retrievals; only useable by selected internal services:
- scpInformation
- trustId
- trustIdChangedAt
- trustIdChangedBy
- identityProviderType
Built for Scala 3 and drops Scala 2.12.
It uses http-verbs 15.0.0 and HttpClientV2. It is a major version since there are binary incompatibilities in this dependency.
The principalNino field in the TrustedHelper retrieval is now optional. This covers some edge cases where it could be None.
Adds support for Play 2.9 and Play 3.0 and drops support for Play 2.6 and 2.7. Versioning has changed to drop the play version - there will be an artefact for each play version supported.
Update to README and code changes to remove service manager plugin.
Auth Provider GDS/Verify is no longer supported. See 30th March 22 Tech Blog Post.
Scala 2.13 - minor version increment.
Removed dead OTAC & Passcode logic.
Removed retrieval support for the deprecated unread_message_count attribute.
Cross build for Scala 2.13.
Check for presence/absence of the bearer token in the session cookie before calling auth to create a session.
Presence of the Authorization header will now be checked before making the call to auth to check the session, so tests mocking a successful authorised() call will now need to be updated to set this header as well
Removed Joda time dependency.
Drop support for play 2.7 and play 2.6.
Updated to latest version of sbt-auto-build and dropped Artifactory.
Added full support for CL250, the Identity Verification (IV) Multiple Evidence Option (MEO).
For CLxxx meanings, see: https://confluence.tools.tax.service.gov.uk/display/VER/%27Confidence+Level%27+definitions
Earlier versions only supported CL50 or CL200, plus deprecated CL100.
See the May 2021 auth-client v5.x Tech Blog Post for further details inc. the full changelog under "Notes"
Added support for Play 2.8, plus CL250 in part.
Include the following dependency in your SBT build
resolvers += MavenRepository("HMRC-open-artefacts-maven2", "https://open.artefacts.tax.service.gov.uk/maven2")
libraryDependencies += "uk.gov.hmrc" %% "auth-client-play-xx" % "[INSERT-VERSION]"Where play-xx is your version of Play (e.g. play-29).
Note the version does not include the version of Play (as it did prior to version 7.0.0).
Note that this library is only available for Play 2.8.x and 2.9.x. No other play versions are supported.
Play < 2.8 is no longer supported. If you plan to use it in a microservice that is still using an older version of Play, then you will need to upgrade it first.
The overall approach is to add the library and take advantage of it by extending various classes. When a user first logs in they will not have an active session, and this must be handled by ensuring that your play global object extends uk.gov.hmrc.auth.frontend.Redirects and has overridden the 'resolveError' method that it provides. Your method should contain a NoActiveSession case and a default case as a bare minimum, see the Error handling / Redirects section below for more detail.
"auth-client is automatically included in your service, and is a transitive dependency provided by bootstrap-play which all microservices have within them. So there is no need to add auth-client as an explicit dependency. Additionally, keep the service's bootstrap-play library up to date so as to bring in the latest auth-client. Ideally, increment bootstrap-play to the latest version every time a microservice is updated. Same for the other library dependencies."
First, in any controller, service or connector where you want to protect any part of your logic, mix in the AuthorisedFunctions trait:
// AuthorisedFunctions mixin
class MyController @Inject() (val authConnector: AuthConnector) extends BaseController with AuthorisedFunctions {
}The AuthConnector instance itself is then usually defined somewhere in your wiring setup:
// AuthConnector Wiring
class ConcreteAuthConnector(val serviceUrl: String,
val httpClientV2: HttpClientV2) extends PlayAuthConnectorNote that you must enable the HttpClientV2Module in application.conf in order to use HttpClientV2:
play.modules.enabled += "uk.gov.hmrc.play.bootstrap.HttpClientV2Module"
Alternatively if using the HMRC bootstrap library, an AuthConnector instance can be made available by enabling the following module in application.conf:
# Provides an implementation of AuthConnector.
play.modules.enabled += "uk.gov.hmrc.play.bootstrap.AuthModule"
Depending on your requirements, you can define 0 to any number of predicates defining your authorisation logic, as well as 0 to any number of data retrievals from the current authority that you intend to use inside the function wrapper. In the simplest case (no authorisation predicates, no data retrieval) you can use an empty wrapper like this:
// Without Predicates
authorised() {
// your protected logic
}The minimal check that this code will still do is that the user is currently logged into the MDTP platform and has a bearer token in their session that is not yet expired. In case these conditions are not met, the function body will not execute. If you do have a few authorisation requirements, you can pass them to the authorise function:
// With Predicates
authorised(Relationship("TRUST",Set(BusinessKey("UTR","12345")))) {
// your protected logic
}In this example we require that the user has a relationship in the relationship establishment service, with the relationship name of TRUST and the business key of a UTR of 12345.
// With Predicates
authorised(Enrolment("SOME-ENROLMENT") and AuthProvider(GovernmentGateway)) {
// your protected logic
}In this example we require that the user has a SOME-ENROLMENT enrolment and only accept users logged in via Government Gateway.
There are many other predicates that can be freely combined with and and/or or.
If you also want load some data for the current authority, add a retrieval call after the authorisation logic:
// With Data Retrieval
import uk.gov.hmrc.auth.core.retrieve.Retrievals._
authorised(Enrolment("SOME-ENROLMENT)).retrieve(externalId) { externalId =>
// your protected logic
}Here we specify that we want to retrieve the externalId of the current user, which is a stable identifier that can be used by external systems to key data for that user. Note that the function block now takes a parameter, which gives you the requested data in a type-safe manner. For all available retrieval options see the section on Retrievals below. If you request more than one data item, they will be passed in a flexible wrapper that can be best accessed via a pattern match in the body:
// Multiple Data Retrievals
import uk.gov.hmrc.auth.core.retrieve.Retrievals._
authorised(Enrolment("SOME-ENROLMENT")).retrieve(internalId and userDetailsUri) {
case internalId ~ userDetailsUri => // your protected logic
}Note the ~ for chaining the values in the pattern. While this syntax is unusual, it is the only cheap way of giving you any number of type-safe results inside the function without introducing a rather heavy-weight library like shapeless. You can combine any number of retrievals with and. A logical or is not available here as it would not make any sense for retrievals. That's all there is to it. If you were one of the unlucky users who had to use one of our older auth libraries which forced you to implement an AuthProvider, a TaxRegime and maybe also a PageVisibilityPredicate and whatnot, I hope you appreciate the simplicity and consistency of the new library. We did not implement a single trait in the examples above, we merely used existing building blocks to express our requirements!
In order to check confidence level and retrieve its value, use the following snippet:
authorised(ConfidenceLevel.L200).retrieve(Retrievals.confidenceLevel) {
case confidenceLevel => // logic
}To check confidence level was obtained via multiple evidence sources, use CL250 instead as follows:
authorised(ConfidenceLevel.L250).retrieve(Retrievals.confidenceLevel) {
case confidenceLevel => // logic
}The config based approach should not be used for frontends. A frontend will generally require more control over the handling of authorisation errors than is available in the filter chain to avoid raw http errors surfacing in the user''s browser. Whilst this is not true of backend services which will generally just return any http error codes resulting from calls within the filter chain we believe that the wrapper function is a better approach being easier to understand, more explicit and more flexible.
The following shows an example for a configuration for two controllers:
controllers {
authorisation {
sa = {
patterns = [
"/my-service/enrolment1/:identifier"
"/other-service/enrolment1/:identifier/"
]
predicates = [{
enrolment = "ENROLMENT1"
identifiers = [{ key = "SOME-IDENTIFIER-KEY", value = "identifier" }]
}]
}
ct = {
patterns = [
"/my-service/enrolment2/:identifier"
"/other-service/enrolment2/:identifier/:rest"
]
predicates = [{
enrolment = "ENROLMENT2"
identifiers = [{ key = "SOME-IDENTIFIER-KEY", value = "identifier" }]
}]
}
}
foo.FooController = {
authorisedBy = ["enrolment1", "enrolment2"]
needsLogging = false
needsAuditing = false
}
bar.BarController = {
authorisedBy = ["enrolment1"]
needsLogging = false
needsAuditing = false
}
}The example shows all aspects you need to define for protecting your endpoints:
Inside the controllers section in the play configuration for your microservice, add an authorisation section, that contains one or more sections that define a concrete authorisation setup. In the example there are two sections named enrolment1 and enrolment2 respectively.
Within these sections, define one or more patterns for URIs that should be affected by this configuration. In the example both sections contain two path patterns each:
patterns = [
"/my-service/enrolment1/:identifier"
"/other-service/enrolment1/:identifier/"
]Note that the pattern syntax is close to the one used for Play routes definition, it is not a regular expression like in our older library. Both paths above have several literal path elements like enrolment1, and one dynamic path element each (:identifier). These dynamic path elements match any string and the actual concrete value can be referred to in the predicate definitions as shown further below. For each incoming request the paths will be tried against the URI of the request in the order they are defined until one of them matches.
Below the path patterns, define zero or any number of predicates that define the authorisation requirements for each incoming request:
predicates = [{
enrolment = "ENROLMENT1"
identifiers = [{ key = "SOME-IDENTIFIER-KEY", value = "identifier" }]
}]In this example there is just one predicate. It specifies that the current user has to have an ENROLMENT-1 enrolment with a specific SOME-IDENTIFIER-KEY extracted from the URI. This ensures that all calls happen in this specific context. But as you see the predicates property is an array, so you can specify any number of predicates. They are combined with an implicit AND, if you need an OR combinator, you can use a mongo-style $or property:
predicates = [{
$or = [
{ affinityGroup = "Organisation" }
{ affinityGroup = "Individual" }
]
}]Here the endpoint is blocked for all agents, as we only allow either Organisations or Individuals in. Finally, like with the Scala API you can omit predicates altogether with an empty array:
predicates = []In this case the filter will only check that the user is currently logged into the platform and has a bearer token in her session that is not yet expired. For all the available predicates you can use see the Predicate Reference page.
In our example one controller is linked to 2 authorisation presets:
foo.FooController = {
authorisedBy = ["enrolment1", "enrolment2"]
needsLogging = false
needsAuditing = false
}Here it will first try the 2 patterns defined in the "enrolment1" preset, and if any of the 2 matches, apply the predicates defined in there. If none of them matches, it tries the 2 patterns defined in the enrolment2 preset, and again, if any of the 2 matches, applies the predicates in that section. If no pattern matches it is considered an insecure call which will lead to a 401 response. This is to avoid that a new endpoint that got added to an existing controller with a pattern that does not match does not accidentally become unprotected. As a consequence this means that within a single controller all endpoints are either protected or not. If you need more flexibility, use the Scala API inside your controller implementations. If you want to leave a controller unprotected, simply omit the authorisedBy predicate.
For backend services you usually do not want to explicitly handle unauthorised calls. It is usually the task of a frontend to determine whether a user is allowed to perform a certain action and then the frontend implementation ensures that all calls to backend services happen in an authorised context. If this is not the case, it is best to let the 401 response bubble back up to the frontend.
In frontends you would usually want to avoid letting 401 responses bubble up to the user, so you have two options to recover from such a failure:
Centralised Recovery in the Global object
In many frontends, the recovery from certain types of authorisation failures will probably be similar. So keeping the logic in the Global object is convenient.
import import play.api.mvc.Results._
object MyGlobal extends DefaultFrontendGlobal {
override def resolveError(rh: RequestHeader, ex: Throwable): Result = ex match {
case ex: InsufficientEnrolments("HMRC-SA") => // your custom recovery logic, usually redirects
case _ => super.resolveError(rh, ex)
}
}Individual Recoveries in Controllers
In rare cases where a single endpoint would require custom recovery logic, you can add it immediately after the authorised block in your controller logic:
authorised(Enrolment("SOME-ENROLMENT")) {
// your protected logic
} recoverWith {
case ex: InsufficientEnrolments("SOME-ENROLMENT") => // your custom recovery logic, usually redirects
}Whenever the auth microservice returns a 401 response to the library, it will contain a WWW-Authenticate header with details about the failure which will then be translated to a custom exception. They are all a subtype of AuthorisationException. This is the complete list of possible exceptions:
- InsufficientEnrolments(enrolmentName): One or more of the requested enrolments were not present in the authority, the first failing enrolment name is returned in the
enrolmentfield - InsufficientConfidenceLevel: The confidence level requested for one of the enrolments was too low
- UnsupportedAuthProvider: the provider for the current authority is not one of the providers your auth predicates listed as accepted
- UnsupportedAffinityGroup: the requested affinityGroup did not match the one in the authority or the user is not a GG user
- UnsupportedCredentialRole: the requested credentialRole did not match the one in the authority or the user is not a GG user
- NoActiveSession: The user does not have an active session. This is an abstract type with 4 concrete subtypes, but you would usually handle all 4 types in the same way (most often by redirecting to the login page). The 4 subtypes are: MissingBearerToken (the user was probably not logged in), BearerTokenExpired (the user was logged in, but the session is expired), and two types which should never occur as they would hint at an internal error: InvalidBearerToken and SessionRecordNotFound.
- FailedRelationship: The requested relationship in the authority did not match the one in the Relationship Establishment.
In most cases testing a controller that uses auth-client is a matter of overriding the AuthConnector by providing a stubbed implementation.
Here you can find an example which retrieves credentials and enrolments:
private class TestSetupAuthorisationDemoController(stubbedRetrievalResult: Future[_]) extends AuthorisationDemoController {
override val authConnector: AuthConnector = new AuthConnector {
def authorise[A](predicate: Predicate, retrieval: Retrieval[A])(implicit hc: HeaderCarrier, ec: ExecutionContext): Future[A] = {
stubbedRetrievalResult.map(_.asInstanceOf[A])
}
}
}
"calling retrieveProviderIdAndAuthorisedEnrolments" should {
"return 200 for successful retrieval of 'authProviderId' and 'authorisedEnrolments' from auth-client" in {
val retrievalResult: Future[~[LegacyCredentials, Enrolments]] = Future.successful(new ~(GGCredId("cred-1234"),
Enrolments(Set(Enrolment("enrolment-value")))))
val controller = new TestSetupAuthorisationDemoController(retrievalResult)
val result = controller.retrieveProviderIdAndAuthorisedEnrolments()(fakeRequest)
status(result) shouldBe Status.OK
}
"return 401 for unsuccessful retrieval of 'authProviderId' and 'authorisedEnrolments' from auth-client" in {
val controller = new TestSetupAuthorisationDemoController(Future.failed(InsufficientEnrolments()))
val result = controller.retrieveProviderIdAndAuthorisedEnrolments()(fakeRequest)
status(result) shouldBe Status.UNAUTHORIZED
}
}
Another possibility is using AuthConnector as a class dependency. In that case you can rely on AuthConnector#authorise() being mocked.
"allow mocking of response" in {
implicit val hc = HeaderCarrier(authorization = Some(Authorization("Bearer 123")))
val mockAuthConnector: AuthConnector = mock[AuthConnector]
val retrievalResult: Future[~[Credentials, Enrolments]] = Future.successful(new ~(Credentials("gg", "cred-1234"),
Enrolments(Set(Enrolment("enrolment-value")))))
Mockito.when(mockAuthConnector.authorise[~[Credentials, Enrolments]](any(), any())(any(), any()))
.thenReturn(retrievalResult)
mockAuthConnector.authorise(EmptyPredicate, Retrievals.credentials and Retrievals.authorisedEnrolments)
Mockito.verify(mockAuthConnector).authorise(equalTo(EmptyPredicate),
equalTo(Retrievals.credentials and Retrievals.authorisedEnrolments))(any(), any())
}
When using Wiremock to test your controller's endpoint, you may get an error Bearer token not supplied.
To fix this, you can either provide an HTTP Authorization header
wsClient.url("/baseUrl/endpoint")
.withHttpHeaders("Authorization" -> "Bearer123")
.get()or provide a mocking session cookie as demonstrated below
wsClient.url("/baseUrl/endpoint")
.withCookies(mockSessionCookie)
.get()
// mocking session cookie
def mockSessionCookie = {
def makeSessionCookie(session:Session): Cookie = {
val cookieCrypto = inject[SessionCookieCrypto]
val cookieBaker = inject[SessionCookieBaker]
val sessionCookie = cookieBaker.encodeAsCookie(session)
val encryptedValue = cookieCrypto.crypto.encrypt(PlainText(sessionCookie.value))
sessionCookie.copy(value = encryptedValue.value)
}
val mockSession = Session(Map(
SessionKeys.lastRequestTimestamp -> System.currentTimeMillis().toString,
SessionKeys.authToken -> "mock-bearer-token",
SessionKeys.sessionId -> "mock-sessionid"
))
val cookie = makeSessionCookie(mockSession)
new WSCookie() {
override def name: String = cookie.name
override def value: String = cookie.value
override def domain: Option[String] = cookie.domain
override def path: Option[String] = Some(cookie.path)
override def maxAge: Option[Long] = cookie.maxAge.map(_.toLong)
override def secure: Boolean = cookie.secure
override def httpOnly: Boolean = cookie.httpOnly
}
}
For more testing examples head to uk.gov.hmrc.auth.core.AuthConnectorSpec or Auth-Demo-Frontend
This code is open source software licensed under the Apache 2.0 License.