After a couple years of experience, we (the identity community that supports Persona) have amassed a small collection of changes to the data formats which underly the BrowserID protocol. This post is a migration guide for maintainers of BrowserID client implementations.
(NOTE: This summary is based on the work of Dan Callahan, Dirkjan Ochtman, and others.)
The motivation for these changes include:
- Standards Conformance: We've been tracking the JOSE family of specs from the beginning. They've drifted. Tighter conformance suggests a greater chance of reusable cryptography code.
- Extensibility: We'd like to fully support experimentation and extension by allowing extended claims: identity providers should be able to include information in certificates, and RPs in assertions.
- Efficiency and Consistency: The initial implementation of BrowserID encodes binary data in multiple different ways, resulting in needless complexity and a bloated representation. Also, there is needless precision in time representation.
Key Representation
Currently, DSA keys are represented thus:
We will move to the following representation:
Specifically, the important changes are:
- instead of
algorithm
, we'll represent key type using thekty
key (see v19 of the JWK spec) - instead of abbreviated algorithm names, (DS), we'll use the full acronym (DSA).
- All binary data is to be represented using unpadded base64url encoding (NOTE: this is not base64);
Other than these changes, the property names of RSA or DSA keys is unchanged. There is no need to seperately cover RSA, because changes are completely analagous.
Certificate Representation
Recall all signed objects in browserid are represented thus:
<header>.<payload>.<signature>
The external representation of a signed object is, and will continue to be,
these three components base64url encoded, joined by periods (.
).
Header and signature segments will be unchanged, but there are a couple changes to the payload. Previously, the payload looked something like this:
We will move to a representation that looks like this:
Note that key representation is omitted in these examples as changes are outlined above.
The important changes are:
- rather than representing the user's identity under
principal.email
, we now use the subject field (with a key ofsub
from the JWT spec). - all times are now represented in POSIX time (seconds since the 1970 epoch).
- the user's public key is provided under a
pubkey
property.
Assertion Representation
What we typically refer to with the shorthand of "assertions" are encoded bundles, called Backed Identity Assertions, which look thus:
<cert header>.<cert payload>.<cert signature>~<assertion header>.<assertion payload>.<assertion signature>
That is a Backed Identity Assertion includes an Identity Certificate
whereby an identity authority vouches that a specific subject (typically a person)
controls a specific keypair - combined with an Identity Assertion assertion whereby the
subject claims she owns a specific identity and wishes to prove it to a specific "audience"
(typically a website) at a specific time. Note, Identity certificate and assertion are separated
with a tilde (~
- U+007E).
The changes to assertions include key representation and certificate representation as described above, as well as some minor changes to the assertion payload. Previously a payload may have looked something like this:
Which will be change to look like this:
The relevant changes in assertion payload include:
- All times are represented in POSIX time
- Issue time (
iat
) is included to make the validity period of an assertion explicit.
Further Reading
This post describes a subset of planned changes which affect two different groups: Client implementations of BrowserID (which generate keypairs, and create Backed Identity Assertions) will need to be updated, and Identity providers will need to update their support documents (with a new representation of a public key), and will need to generate Identity Certificates which conform to the new formats.
There is a final set of changes to support document representation that will allow more graceful key revocation and rotation by identity providers, that is not covered in this post.
For more in depth coverage of the updated BrowserID formats, you can refer to the in progress specification update. Finally, you can have a look at the new reference verification library that implements verification of assertions which conform to this new format, while retaining backwards compatibility.
Finally, this transition will begin slowly over the coming months. First a new verifier will be deployed, then we will migrate client implementations as we reach out to identity providers to migrate.