Bug 227432 – P-384 EC public JWK key in layout tests is wrong

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Yoshiaki Jitsukawa]

In https://github.com/WebKit/WebKit/blob/main/LayoutTests/crypto/subtle/ecdh-import-jwk-public-key-p384.html and https://github.com/WebKit/WebKit/blob/main/LayoutTests/crypto/subtle/ecdsa-import-jwk-public-key-alg-384.html, the JWK key is defined as:
var jwkKey = {
    kty: "EC",
    use: "enc",
    ext: true,
    crv: "P-384",
    x: "1bHwFrsaPRjYq-zFOyLXK8Ugv3EqbVF075ct7ouTl_pwyhjeBu03JHjKTsyVbNWK",
    y: "1bHwFrsaPRjYq-zFOyLXK8Ugv3EqbVF075ct7ouTl_pwyhjeBu03JHjKTsyVbNWK",
};
However x and y are the same value and the point (x, y) is not on the elliptic curve.

On windows Chrome, it fails to import the key, resulting in a DOM exception. Instead, using y = "OPQLbvRDpocNk4FRQLUVRBbqiauWcnUn7cAC2uSOH6GpSLoTo6B-4LtHq0ydd-iu" succeeds. 

The y value can be derived from x with openssl:
 1. With EC_POINT_set_compressed_coordinates_GFp(), set the coordinate on the curve, only specifying x (and the sign of y).
 2. Then retrieve both x and y with EC_POINT_get_affine_coordinates_GFp().

[Yoshiaki Jitsukawa]

Created attachment 432358 [details]
Patch

[Yoshiaki Jitsukawa]

Created attachment 432359 [details]
test html run on chrome

Attaching a test html I ran on windows Chrome.

[Yoshiaki Jitsukawa]

Created attachment 432361 [details]
Patch

[Yoshiaki Jitsukawa]

Created attachment 432439 [details]
Patch

[Alexey Proskuryakov]

Comment on attachment 432439 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=432439&action=review

> LayoutTests/ChangeLog:3
> +        P-384 EC public JWK key in layout tests is wrong

Is this a bad test that shouldn't be passing? trying to understand if there is a WebKit behavior bug, not just a test bug.

[Yoshiaki Jitsukawa]

> Is this a bad test that shouldn't be passing? trying to understand if there is a WebKit behavior bug, not just a test bug.

I believe these tests themselves are expected to be passed, and as you mentioned, there seems to be a bug that WebKit accepts a kind of invalid keys.

[Chris Dumez]

(In reply to Yoshiaki Jitsukawa from comment #6)
> > Is this a bad test that shouldn't be passing? trying to understand if there is a WebKit behavior bug, not just a test bug.
> 
> I believe these tests themselves are expected to be passed, and as you
> mentioned, there seems to be a bug that WebKit accepts a kind of invalid
> keys.

Seems we should align with Chrome and throw on bad input then?

[Yoshiaki Jitsukawa]

(In reply to Chris Dumez from comment #7)
> Seems we should align with Chrome and throw on bad input then?

I personally think so.

[Yoshiaki Jitsukawa]

I've filed bug 227480 - [WebCrypto] WebKit accepts elliptic curve key whose point is not on the curve

[EWS]

Committed r279374 (239240@main): <https://commits.webkit.org/239240@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 432439 [details].

[Radar WebKit Bug Importer]

<rdar://problem/79916328>