[
https://issues.apache.org/jira/browse/HTTPCORE-319?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13507741#comment-13507741
]
Oleg Kalnichevski commented on HTTPCORE-319:
--------------------------------------------
Hi Scott
I believe I have found and (hopefully) fixed the problem. It turned out to be
quite tricky. Several rare conditions have to coincide in order to leave the
affected SSL connection effectively locked in the CLOSING state being both
unable to decrypt incoming data and flush output data and close itself
gracefully. So, for the problem to occur the underlying connection has to be
abnormally terminated by the opposite end in the middle of a request execution.
There has to be a partial or corrupt incoming message stuck in the SSL input
buffer which cannot be decrypted without more data (leaving the SSL engine
permanently stuck in the UNWRAP state). There has to be some data still stuck
in the session output buffer (POST request body in your code) which needs to be
flushed in order to close the connection gracefully. The bug causes the SSL I/O
session to incorrectly set the i/o event interest mask (output interest does
not get activated while the SSL engine is in the UNWRAP state).
There are different ways of approaching the problem. I committed a rather
non-intrusive fix which I believe should fix the problem but I would need your
confirmation that this is indeed the case. I also added more SSL session
details that get printed to the log and a few defensive checks. Could you
please get the latest code from the 4.2.x branch (with this fix [1]) and try it
out in your environment? If the problem still persists, please attach a
complete wire log of the session to this ticket. I would have to try something
more invasive.
Oleg
[1] http://svn.apache.org/viewvc?view=revision&revision=1415878
> SSLIOSession goes into a loop if the server rejects an invalid certificate
> --------------------------------------------------------------------------
>
> Key: HTTPCORE-319
> URL: https://issues.apache.org/jira/browse/HTTPCORE-319
> Project: HttpComponents HttpCore
> Issue Type: Bug
> Components: HttpCore NIO
> Affects Versions: 4.2.2
> Reporter: Scott Stanton
> Assignee: Oleg Kalnichevski
> Priority: Critical
> Fix For: 4.2.3
>
> Attachments: log
>
>
> To reproduce:
> * Set up an SSL server that requests certificates from the client.
> * Set up a client with an expired SSL certificate.
> * Establish a connection from the client to the server using BaseNIOReactor
> and SSLIOSession.
> The server will proceed through the handshake until the client supplies its
> certificate in response to the CertificateRequest message. At this point,
> the server's certificate verification will fail and it will close the
> connection.
> The client socket will become readable due to the EOF and the
> SSLIOSession.isAppInputReady() method is called to handle the EOF. The
> bytesRead gets set to -1, which sets this.endOfStream = true. Nothing ever
> sets the session into the CLOSING or CLOSED state, so it keeps looping on the
> readable EOF event.
> I'm not sure what the best approach to fixing this should be. It appears
> that if I close the session manually with the debugger from inside
> isAppInputReady, the system proceeds normally from that point, however I
> don't know what the implications of doing that might be.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
