1312976 - Add support for exporters on SHA 384 cipher suites

Status: ASSIGNED

(NSS :: Libraries, defect, P3)

Description

[Martin Thomson [:mt:]]

We do absolutely the wrong thing for exporting when the PRF hash is SHA-384.

See bug 1310061 for details.

Comment 1

[Martin Thomson [:mt:]]

Attachment: bug1312976.patch

Bob, we recently discovered that Elio's work on implementing AES-256-GCM and SHA-384 was incomplete.  This makes the TLS 1.2 PRF more generic (it takes a hash argument).  Can you check that my pk11 code isn't completely bonkers?

Comment 2

[Martin Thomson [:mt:]]

Oh, to be clear here, I am treating mechanisms as requiring ABI compatibility.  On that basis, none of the existing mechanisms are usable here.  The only one to take a variable hash function also limits the size of its output (to the size of the TLS Finished).

Comment 3

[Robert Relyea]

if we have problems with the mechansims as defined, we should submit patches back to the OASIS spec rather than define yet another NSS specific version.

Comment 4

[Martin Thomson [:mt:]]

Note that the existing mechanisms I'm talking about are also NSS internal mechanisms.

Comment 5

[Martin Thomson [:mt:]]

Oh, and I think that we should ask OASIS for help here, but TLS 1.3 is a moving target, and clearly TLS 1.2 never got the proper treatment.

Comment 6

[Martin Thomson [:mt:]]

OK, this is just a safeguard for the moment.

https://hg.mozilla.org/projects/nss/rev/5a2b0e8da4e1b59c64f5bd8b9256116a753aff70

Comment 7

[Martin Thomson [:mt:]]

Bob, see https://docs.google.com/document/d/1vNkQcZ_yAVtzh1JZPULVqtQbqVYHKWkZf2-WCBHvsv8/edit?usp=sharing

Comment 8

[Hubert Kario]

side question: is there any command line tool that can print and/or derive the TLS Exporter values? I don't see anything in the tstclnt help meesage in 3.39.0...

Comment 9

[Martin Thomson [:mt:]]

Writing an addition to tstclnt or selfserv (or both) would be relatively straightforward.  I would suggest that you open a new bug for that though.

Comment 10

[Hubert Kario]

(In reply to Martin Thomson [:mt:] from comment #9)
> Writing an addition to tstclnt or selfserv (or both) would be relatively
> straightforward.  I would suggest that you open a new bug for that though.

done, bug 1494063