Now, /verify can return 401 or 403 depending on the user authentication.
Every public API endpoints and pages return 200 with error message in
JSON body or 401 if the user is not authorized.
This policy makes it complicated for an attacker to know what is the source of
the failure and hide server-side bugs (not returning 500), bugs being potential
threats.
One can now customize the default authentication method for all sub-domains,
i.e., either 'two_factor' or 'basic_auth' and define specific authentication
method per sub-domain.
For example, one can specify that every sub-domain must be authenticated with
two factor except one sub-domain that must be authenticated with basic auth.
Previously, logs were not very friendly and it was hard to track
a request because of the lack of request ID.
Now every log message comes with a header containing: method, path
request ID, session ID, IP of the user, date.
Moreover, the configurations displayed in the logs have their secrets
hidden from this commit.
Basically, the ACL configuration was very static and it was not allowed
to remove 'any', 'groups', 'users'. The application crashed when those
keys did not exist.
After this fix, every key is optional and replaced by a default value
for the app configuration to be complete and used by Authelia.
Later, a configuration validator will be implemented to detect issues
with configuration at startup.
Previously, string "{0}" was replaced by the user dn in the groups_filter
attributes of the LDAP configuration.
However, if the groups children only have a memberUid attribute, one would
like to use the username instead of the user dn.
Since the user dn can be built from the username, "{0}" is now replaced
by the username instead of the user dn so that an LDAP relying on attribute
'memberUid' can be used.
Client and server now have their own tsconfig so that the transpilation is only
done on the part that is being modified.
It also allows faster transpilation since tests are now excluded from tsconfig.
They are compiled by ts-node during unit tests execution.
Clément Michaud