Apparently the Apple macOS (and iOS?) ACME client seems to omit
leading null bytes from JWS signatures. The base64-url encoded
bytes decode to a shorter byte slice than what the JOSE library
expects (e.g. 63 bytes instead of 64 bytes for ES256), and then
results in a `jose.ErrCryptoFailure`.
This commit retries verification of the JWS in case the first
verification fails with `jose.ErrCryptoFailure`. The signatures are
checked to be of the correct length, and if not, null bytes are
prepended to the signature. Then verification is retried, which
might fail again, but for other reasons. On success, the payload
is returned.
Apple should fix this in their ACME client, but in the meantime
this commit prevents some "bad request" error cases from happening.
* Store and verify account location on acme requests
Co-authored-by: Herman Slatman <hslatman@users.noreply.github.com>
Co-authored-by: Mariano Cano <mariano@smallstep.com>
* api/render: initial implementation of the package
* acme/api: refactored to support api/render
* authority/admin: refactored to support api/render
* ca: refactored to support api/render
* api: refactored to support api/render
* api/render: implemented Error
* api: refactored to support api/render.Error
* acme/api: refactored to support api/render.Error
* authority/admin: refactored to support api/render.Error
* ca: refactored to support api/render.Error
* ca: fixed broken tests
* api/render, api/log: moved error logging to this package
* acme: refactored Error so that it implements render.RenderableError
* authority/admin: refactored Error so that it implements render.RenderableError
* api/render: implemented RenderableError
* api/render: added test coverage for Error
* api/render: implemented statusCodeFromError
* api: refactored RootsPEM to work with render.Error
* acme, authority/admin: fixed pointer receiver name for consistency
* api/render, errs: moved StatusCoder & StackTracer to the render package
The logic for both test cases is fairly similar, but with some
small differences. Made those clearer by means of some comments.
Also added some comments to the middleware logic that decided
whether to extract JWK or lookup by KID.
