51475 - SSL doesn't invalidate Sessions when connections abort

Status: RESOLVED WONTFIX

(NSS :: Libraries, defect, P3)

Description

[Nelson Bolyard (seldom reads bugmail)]

The SSL spec (and the TLS spec) say that if an SSL connection terminates
without a "close notify" alert being sent, then SSL should invalidate
the SSL session so that it cannot be restarted thereafter.  

NSS's SSL code doesn't do that, but it should.

We might discover, once this is implemented, that lots of other 
clients or servers don't close connections properly, and that this
change results in a huge performance degradation, due to many more
full handshakes being done.  So, when implementing this, it might be
a good idea to implement an option to disable this feature, restoring
compatibility with the old behavior.

Comment 1

[Thomas]

Hey Nelson, can I help you with this one?

Comment 2

[Nelson Bolyard (seldom reads bugmail)]

This bug/RFE is very low priority.  I expect that implementing it will result
in a large amount of time being spent understanding its effects fully and 
deciding whether the benefits of conforming to the spec in this area are 
worth the costs incurred by doing so.  Based on our TLS experience, I expect
that after implementing this, we will find many SSL clients and/or servers 
that do not send close_notify alerts when they should, that that those 
products will begin to experience a much higher frequency of full (RSA) SSL 
handshakes than before.  I expect that the impact of implementing this will 
be a great increase in the frequency of FULL (RSA) handshakes due to the 
inability to restart invalidated sessions.  I expect many people (especially 
server operators) will find that quite objectionable.  I predict the problem 
will be bad enough that we will ultimately decided not to conform to the 
standards in this one respect.

But someone who is willing to take the time to study that and understand it
would be welcome to try it!  

It should be tried on both the client and server ends.  

Comment 3

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

From RFC 4346 (TLS 1.1): "Premature closes no longer cause a session to be nonresumable."

As Nelson noted, the behavior requested in this bug is very bad for performance and there is no security benefit to it.

I recommend closing the bug as WONTFIX.

Comment 4

[Nelson Bolyard (seldom reads bugmail)]

Capital suggestion.