This was a surprisingly involved change. Some of the difficulties included
java.util.Optional purposely not being Serializable (so I had to move a
few Optionals in mapreduce classes to @Nullable) and having to add the Truth
Java8 extension library for assertion support.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=171863777
We want to be safer and more explicit about the authentication needed by the many actions that exist.
As such, we make the 'auth' parameter required in @Action (so it's always clear who can run a specific action) and we replace the @Auth with an enum so that only pre-approved configurations that are aptly named and documented can be used.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=162210306
The affected actions have been changed to check that the user is logged in by [] so this attribute is no longer needed.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=159572365
The code to authenticate and authorize incoming requests (including via OAuth) has been in the system. This CL actually turns it on, since we are satisfied from logging information that it is not unjustly denying access.
Auth settings are also updated on a few commands missed earlier.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=152381820
[] changed RequestHandler to log a warning instead of returning an error to the user when authentication failed. Unfortunately, it didn't handle the resulting absent value of AuthResult. I don't yet know why the tests pass, and plan to add a test to check for this situation, but I wanted to start by fixing the problem as soon as possible, as it seems like this would result in exceptions any time a request was handled which didn't pass authentication.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=149136265
This is the first step in rolling out the changes so that we can check via logging whether turning on the logic would reject anything it should not.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=149050878
This follows up on Brian's work to transition not just to a new format
with an empty scope value, but instead to replace the existing format
entirely with a new one that:
1) includes a version number to support future format migrations
2) doesn't include a field for the scope at all, since scoping the
tokens adds no real security benefit and just makes verification
more difficult
3) replaces the raw SHA-256 hash with a SHA-256 HMAC instead, as a
best practice to avoid length-extension attacks [1], even though
in our particular case they would only be able to extend the
timestamp and would thus be relatively innocuous
The new format will be produced by calling generateToken(), and the
scope-accepting version is renamed to generateLegacyToken() in addition
to its existing deprecation, for maximum clarity.
I changed the validateToken() logic to stop accepting a scope entirely;
when validating a legacy-style token, we'll test it against the two
existing legacy scope values ("admin" and "console") and accept it if
it matches either one.
Note that this means the xsrfScope parameter in @Action is now wholly
obsolete; I'll remove it in a follow-up to avoid bringing extra files
into this CL.
After this CL hits production, the next one will replace all calls to
generateLegacyToken() with generateToken(). Once that CL is deployed,
the last step will be removing the legacy fallback in validateToken().
[1] See https://en.wikipedia.org/wiki/Length_extension_attack
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=148936805
It turns out this type parameter was never necessary. A builder only needs the reflexive second type parameter when you want to have a builder inheritance hierarchy where the descendant builders have methods that the ancestor builder doesn't. In that case, the type param enables the ancestor builder's setter methods to automatically return the correct derived type, so that if you start with a derived builder, you can call a setter method inherited from an ancestor and then continue the chain with setters from the derived builder (e.g. new ContactResource.Builder().setCreationTime(now).setContactId(), which otherwise would have returned an EppResource.Builder from setCreationTime(), at which point the call to setContactId() would not compile).
Even then, it's not strictly necessary to use the type parameter, since you could instead just have each derived type override every inherited method to specify itself as the return type. But that would be a lot of extra boilerplate and brittleness.
Anyway, in this case, there is a builder hierarchy, but RequestComponentBuilder specifies all the methods that we're ever going to want on our builders, so there's never any need to be able to call specific derived builder methods. We only even need the individual builder classes so that Dagger can generate them separately for each component.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=148269178
The one-day validity period is also moved from the caller into XsrfTokenManager.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=147857716
This refactors RequestHandler so that it handles the construction of the request
component itself, rather than being handed a pre-built request component
instance constructed by the invoking servlet.
The motivation for this change is so that RequestHandler can be extended in
future CLs to compute authentication results, and can provide those results as
an available binding in the constructed request component. An alternative
approach could have been to compute the authentication results within
RequestModule itself, but I think it's clearer to keep business logic like
that outside of Dagger providers.
This CL makes the following individual changes:
- Adds request component builders, which implement a RequestComponentBuilder
interface so they can all be manipulated by RequestHandler
- Instead of obtaining request components via factory methods on the global
components, one now can have global-scoped bindings just inject the request
component builders (which requires adding a module to each global component
declaring the subcomponent). This follows the recommended approach here:
http://google.github.io/dagger/subcomponents.html
- Instead of exposing request components on the global component interface,
we now expose module-specific subclasses of RequestHandler that @Inject the
appropriate request component builder's provider and pass it to the superclass
(note that inheritance isn't strictly necessary here but saves boilerplate)
- RequestHandler now takes the Provider<RequestComponentBuilder> and builds
the component itself using its own fresh RequestModule instance. This provides
some nice encapsulation but is mainly needed for adding a RequestAuthModule
in future work.
- RequestHandler also takes UserService now, which can be provided via Dagger
by the subclass. Longer-term that will go away in favor of instead providing
AuthStrategy instances, some of which will use UserService internally.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138815648
This change moves the reflective setAccessible() calls on the request component
methods (needed so that they can be invoked reflectively from RequestHandler)
to within Router itself, eliminating the need to manually call this from each
Servlet class and then pass in the resulting Method objects. Instead, we just
pass in the request component class and let Router do the rest.
Old comments say that cross-package reflection is not allowed on AppEngine, but
while it's quite possible this was once the case, I can't reproduce that
limitation, and the documentation seems to contradict any such restriction:
"""
An application is allowed full, unrestricted, reflective access to its own
classes. It can query any private members, call the method
java.lang.reflect.AccessibleObject.setAccessible(), and read/set private
members.
"""
https://cloud.google.com/appengine/docs/java/runtime#reflection
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138693006
The dark lord Gosling designed the Java package naming system so that
ownership flows from the DNS system. Since we own the domain name
registry.google, it seems only appropriate that we should use
google.registry as our package name.
This change renames directories in preparation for the great package
rename. The repository is now in a broken state because the code
itself hasn't been updated. However this should ensure that git
correctly preserves history for each file.
2016-05-13 18:55:08 -04:00
Renamed from java/com/google/domain/registry/request/RequestHandler.java (Browse further)
