Yes, I think I could restructure this to support that approach. You're
right in that FFDHE gets the short end of the stick in the current
scheme unless it's the only type in the namedGroups property and even
then it takes the longest in terms of time. I'll restructure this and
issue a new webrev.
--Jamil
On 7/27/2020 8:21 AM, Xuelei Fan wrote:
I was just wondering, could we just simplify the implementation by
using two named groups for the top two-preferred categories, without
limited to XDH and ECDHE? For example, if FFDHE is on the top 2,
FFDHE will be used as well. Normally, XDH and ECDHE will be used, but
the simplifying is a little bit more flexible if FFDHE is preferred.
What do you think?
Xuelei
On 7/8/2020 3:57 PM, Jamil Nimeh wrote:
Hello all,
This is a small enhancement to the TLS 1.3 client hello. With this
change JSSE will make a best-effort attempt to select the
most-preferred XDH and ECDHE named group. In most cases this will be
x25519 and secp256r1. This was done to help reduce the number of
times our clients get HelloRetryRequests due to servers wanting
secp256r1 when we would just assert x25519.
The asserted key shares can be affected by different settings for the
jdk.tls.namedGroups System property. If x25519 and/or secp256r1 are
missing, the next-most-preferred named group of that class will be
selected. If no other named group in that class is in the
supported_groups extension, then it will be skipped. In the unlikely
event that no XDH or ECDHE supported group is asserted, then the
client will assert one key share, the most preferred one out of the
supported groups.
This also fixes one minor spec compliance issue where we weren't
throwing an exception when a server returned an illegal
HelloRetryRequest with a named group that was in the original
key_shares extension from the client hello.
Webrev: https://cr.openjdk.java.net/~jnimeh/reviews/8247630/webrev.01/
JBS: https://bugs.openjdk.java.net/browse/JDK-8247630
--Jamil
