I added an unprotected flag to the proxy routes which is evaluated by
the authentication middleware. This way we won't have to maintain a
hardcoded list of unprotected paths and path prefixes and we will
hopefully reduce the times we encounter the basic auth prompt by web
browsers.
I refactored the proxy so that we execute the routing before the
authentication middleware. This is necessary so that we can determine
which routes are considered unprotected i.e. which routes don't need
authentication.
I admit it would be better to implement the tests but I tried and it is a bit tricky since we can't mock everything we would need to mock. I'll wan to get these changes in first and later in the near future we should revisit the auth middleware architecture and refactor it a bit more to be more testable and future proof.
The old approach of the authentication middlewares had the problem that when an authenticator could not authenticate a request it would still send it to the next handler, in case that the next one can authenticate it. But if no authenticator could successfully authenticate the request, it would still be handled, which leads to unauthorized access.
Allow to switch jwt access token verification and off. Many (most?) IDP
provide JWT encoded access tokens. If ocis is configure to assume jwt
access tokens (access_token_verify_method==jwt) we now properly verify
the tokens signature and a set of standard claims ("exp", "iat" and nbf"
by way of the jwt module's standard verification and "iss" explicitliy).
This change also allows for introduction of other access token verification
mechanism in the future (e.g. through introspection (RFC7662).
We try to parse the access token as a JWT now. Verifying the signature
using the keys downloaded from the jwks_uri of the IDP. Currently we
only use it to extract the expiry information from the JWT. This could
be reworked to extract other claims from the token for authorization
purposes.
Fixes: #3841
Up to now the builtin lico was using the "username" as the login
attribute, while the proxy (and to some extend the auth-basic) service
tried to uniquely identify users by mail address. This aligns the
default configuration of the services to use the username everywhere.
Fixes: #4039
The removed code was broken is serveral ways:
- The key use for verification was always empty
- Using a static key for access_token verification is incompatible
with key rotation
- The type assertion (which was never hit anyways) to convert
the return MapClaims to StandardClaims can't really work
So in the end we were always using the default ttl for caching
the userinfo.
David Christofas