273861 - SSL_ForceHandshake a bit too forceful?

Status: NEW

(NSS :: Libraries, defect, P2)

Description

[Nelson Bolyard (seldom reads bugmail)]

After a first SSL3/TLS handshake has completed on an SSL socket, When you call
SSL_ForceHandshake to drive a SECOND handshake through to completion, it 
calls ssl3_GatherCompleteHandshake() without first checking to see if there 
is already previously-received and decrypted SSL appliation data still 
sitting there in the "gather state" buffer, still unconsumed by the application.
Consequently, any previously unconsumed appliation data sitting in that buffer
is immediately lost.  This function has always behaved this way.

In the past, we've told users to use SSL_DataPending to see if there is 
unconsumed buffered received data, and if so, to read it all out before calling
SSL_ForceHandshake (again).  

But I'm wondering (which is why I filed this bug as UNCONFIRMED) if the 
definition of SSL_ForceHandshake could be changed so that it checks for 
unconsumed read data and return immediately if such is found, and if such
a change would break any existing users of SSL_ForceHandshake.  

I've been trying to think why (or HOW) a program would be dependent on 
this data-discarding behavior of SSL_ForceHandshake, and it's not obvious.

Any thoughts?

Comment 1

[Nelson Bolyard (seldom reads bugmail)]

I think 
- this really is a bug
- it can be fixed without breaking any existing code that CORRECTLY uses 
  SSL_ForceHandshake
- I should fix it for 3.12 (if not sooner)

Comment 2

[Nelson Bolyard (seldom reads bugmail)]

Unsetting target milestone in unresolved bugs whose targets have passed.