RFR: 8273553: sun.security.ssl.SSLEngineImpl.closeInbound also has similar error of JDK-8253368

[Xue-Lei Andrew Fan]

On Wed, 23 Mar 2022 11:34:33 GMT, Daniel Fuchs <dfuchs at openjdk.org> wrote:

>> JDK-8253368 changed the behavior of SSLSocket to no longer throw a fatal internal_error (80) and invalidate existing sessions (either completed or under construction) as described in (RFC 4346/TLSv1.1+) if a connection was closed without receiving a close_notify alert from the peer.  
>> 
>> This change introduces similar behavior to SSLEngine.
>> 
>> The unit test checks that closing the read(input) sides of the SSLSocket/SSLEngine throws an SSLException, but doesn't invalidate their respective sessions.
>> 
>> Tier1/2 mach5 tests have been successfully run.
>
> src/java.base/share/classes/sun/security/ssl/SSLEngineImpl.java line 799:
> 
>> 797:         } finally {
>> 798:                 conContext.closeInbound();
>> 799:                 engineLock.unlock();
> 
> I see that `onContext.closeInbound()` might throw, which would leave the `engineLock` locked and could cause deadlocks down the road. So maybe you should have a nested `try { } finally { }` here to make sure the lock is properly unlocked.

+1.

-------------

PR: https://git.openjdk.java.net/jdk/pull/7796