RFR: 8236859: WebSocket over authenticating proxy fails with NPE
Pavel Rappo
pavel.rappo at oracle.com
Tue Jan 14 14:39:46 UTC 2020
Daniel,
I imported the patch from the link you provided as follows:
hg import --no-commit open.patch
The patch applied successfully. I tried then to run the tests and saw that some
of them could not be compiled. For instance,
java/net/httpclient/websocket/BlowupOutputQueue.java
java/net/httpclient/websocket/PendingPingBinaryClose.java
...
etc.
> On 13 Jan 2020, at 13:06, Daniel Fuchs <daniel.fuchs at oracle.com> wrote:
>
> Hi,
>
> Please find below a fix for:
> 8236859: WebSocket over authenticating proxy fails with NPE
> https://bugs.openjdk.java.net/browse/JDK-8236859
>
>
> http://cr.openjdk.java.net/~dfuchs/webrev_8236859/webrev.00/
>
> It happens when we have a TLS tunnel and authentication with the server fails: in that case the TLS tunnel is not established. The WS logic tries to get the RawChannel to close it (as it should) but the debug trace gets in the way and causes a NPE.
> If the debug trace is fixed, the NPE occurs later (in connectFlows()).
> The solution is to avoid creating the rawChannel for the sole purpose of closing the connection. A new method added to RawChannel.Provider (internal API) does the trick.
>
> While investigating this issue I stumbled over several other problems:
> an assertion was occurring in the HTTP/1.1 connection pool, because
> the failed connection was wrongly returned to the pool instead of
> being closed. This will be fixed here two.
>
> Then I stumbled on another issue when both the proxy and the server
> required credentials: the proxy credentials were not added to the cache
> but requested every time because the 407 was followed by a 401, and
> that caused 407 to be returned again at the next round trip, causing
> the request to eventually fail with "too many retries".
> This is now fixed in AuthenticationFilter.
>
> The WebSocketProxyTest.java is now updated to test wss: and to test
> both proxy and server authentication in the same request.
>
> best regards,
>
> -- daniel
More information about the net-dev
mailing list