Matt Pavlovich created AMQ-8288:
-----------------------------------
Summary: Review NIOSSLTransport thread safety
Key: AMQ-8288
URL: https://issues.apache.org/jira/browse/AMQ-8288
Project: ActiveMQ
Issue Type: Task
Affects Versions: 5.17.0
Reporter: Matt Pavlovich
Moving the synchronized down to secureRead() removes a pseudo-de-facto
synchronization on doHandshake() which is the only place handshakeInProgress =
true (L424) appears to be assigned. Then finishHandshake() assigns the
sslSession (L242).
There is a -minor- tiny risk that the sslSession is swapped out from underneath
doHandshake. If the getApplicationBufferSize() changes to result in a
BUFFER_OVERFLOW.
Investigate the need to improve the concurrency of the sslSession pointer
assignment b/w doHandshake() and finishHandshake(), given secureRead() has a
condition where it can invoke finishHandshake()
ref:
[SSLSesssionImpl.getApplicationBufferSize()](https://github.com/openjdk/jdk/blob/476775808f82a4b0d42ac58fdb801812b54e01a1/src/java.base/share/classes/sun/security/ssl/SSLSessionImpl.java#L1399)
has deterministic logic of the size based on various factors.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
