RFR: 8357708: com.sun.jndi.ldap.Connection ignores queued LDAP replies if Connection is subsequently closed

[Daniel Fuchs]

On Mon, 26 May 2025 12:28:16 GMT, Jaikiran Pai <jpai at openjdk.org> wrote:

> Can I please get a review of this change which proposes to address the issue noted in https://bugs.openjdk.org/browse/JDK-8357708?
> 
> As noted in the issue, the current code in `com.sun.jndi.ldap.Connection.readReply()` is susceptible to throwing a `ServiceUnavailableException` even when the LDAP replies have already been received and queued for processing. The JBS issue has additional details about how that can happen.
> 
> The commit in this PR simplifies the code in `com.sun.jndi.ldap.LdapRequest` to make sure it always gives out the replies that have been queued when the `LdapRequest.getReplyBer()` gets invoked. One of those queued values could be markers for a cancelled or closed request. In that case, the `getReplyBer()`, like previously, continues to throw the right exception. With this change, the call to `replies.take()` or `replies.poll()` (with an infinite timeout) is no longer expected to hang forever, if the `Connection` is closed (or the request cancelled). This then allows us to remove the connection closure (`sock == null`) check in `Connection.readReply()`.
> 
> A new jtreg test has been introduced to reproduce this issue and verify the fix. The test reproduces this issue consistently when the source fix isn't present. With the fix present, even after several thousand runs of this test, the issue no longer reproduces.
> 
> tier1, tier2 and tier3 tests continue to pass with this change. I've marked the fix version of this issue for 26 and I don't plan to push this for 25.

src/java.naming/share/classes/com/sun/jndi/ldap/LdapRequest.java line 100:

> 98:         // Add a new reply to the queue of unprocessed replies.
> 99:         try {
> 100:             replies.put(ber);

So theoretically this could get into the queue after one of the two markers has already been put in the queue, since we no longer use the lock. The question is: is this a problem? I'd be tempted to say no - except that getReplyBer() will take the markers out of the queue.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/25449#discussion_r2107600300