[
https://issues.apache.org/jira/browse/GEODE-4086?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16290091#comment-16290091
]
Bruce Schuchardt commented on GEODE-4086:
-----------------------------------------
[~WireBaron] and I created a test to reproduce this issue and it shows that
even though the ServerConnection has removed the client's credentials from its
map the credentials are still in the Shiro ThreadContext and in the
ServerConnection's clientUserAuths object and are used to authorize a message
received during the shutdown of the connection.
We paused the health monitor thread just before closing the client's connection
and then had the client send another message. The message went through the
proper authorization checks using established credentials even though we
verified that ServerConnection's map no longer had the credentials.
We added more pause points to make the health monitor clear the clientUserAuths
just before the ServerConnection reads the security Subject and observed that
the Subject was still available.
We altered the test to close the socket before having the client send another
message. This resulted in a new ServerConnection that went through
authentication and established its own credentials.
> ClientHealthMonitor removing client - socket should be closed before cleanup
> is done
> ------------------------------------------------------------------------------------
>
> Key: GEODE-4086
> URL: https://issues.apache.org/jira/browse/GEODE-4086
> Project: Geode
> Issue Type: Bug
> Components: client/server
> Reporter: Brian Baynes
> Fix For: 1.4.0
>
>
> When ClientHealthMonitor decides to remove a client, it appears to initiate
> cleanup (including removing the client's unique ID/auth token) before
> ensuring the socket is closed, creating a race condition where the server may
> accept additional requests from the client before the connection is closed.
> This results in an auth exception because the auth token for the client was
> already removed.
> Instead, the socket should be closed before the cleanup is done.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
