hc-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Paul Donohue (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HTTPCORE-322) SSLIOSession infinite loop after alert during handshake
Date Sat, 17 Nov 2012 17:12:12 GMT

    [ https://issues.apache.org/jira/browse/HTTPCORE-322?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13499465#comment-13499465

Paul Donohue commented on HTTPCORE-322:

I suddenly had an epiphany ... I was wrong before - timing doesn't actually have anything
to do with this issue, so I was only occasionally accidentally generating the proper conditions
when I was messing around with timing.

Here's what actually happened:
- Connection was established and client sent the ClientHello and CloseNotify.  Exact timing
doesn't matter except that the CloseNotify must be received before the server processes the
ClientHello and queues up the ServerHello, ChangeCipherSpec, and Finished messages.
- isAppInputReady() called doHandshake(), which got NEED_UNWRAP and called doUnwrap(), which
processed the ClientHello (and queued the ServerHello, ChangeCipherSpec, and Finished messages)
but left the CloseNotify in the buffer
- The next loop in doHandshake() got NEED_WRAP, and called doWrap() which processed the ServerHello
but left the ChangeCipherSpec and Finished messages in the queue.
- The next loop in doHandshake() got NEED_WRAP, and called doWrap() again
- SSLIOSession allocates SSLEngine.getSession().getPacketBufferSize() bytes for outEncrypted
and SSLEngine.wrap() returns BUFFER_OVERFLOW if outEncrypted.remaining() < outputRecord.maxRecordSize,
which basically means that BUFFER_OVERFLOW is always returned if there is any data in outEncrypted
and any SSL records queued for wrapping
- So, doWrap() returned BUFFER_OVERFLOW and doHandshake() returned
- isAppInputReady() called decryptData()
- doUnwrap() calls SSLEngine.unwrap() calls SSLEngineImpl.readNetRecord() calls SSLEngineImpl.getHSStatus(),
which returns NEED_WRAP due to the pending messages that need to be written out, which causes
doUnwrap() to return with Status.OK
- decryptData() loops forever calling doUnwrap(), which keeps returning NEED_WRAP

I previously thought that HTTPCORE-319 was a separate (though similar) issue, but I now think
it is probably the same issue.  I'm guessing that the server writes out some messages as soon
as it detects a bad client certificate and before it finishes processing the client handshake
messages, which leads to the exact same infinite loop.

I'm attaching a test program which fakes out the SSLEngine and network Channel to induce the
conditions which led to the infinite loop.
> SSLIOSession infinite loop after alert during handshake
> -------------------------------------------------------
>                 Key: HTTPCORE-322
>                 URL: https://issues.apache.org/jira/browse/HTTPCORE-322
>             Project: HttpComponents HttpCore
>          Issue Type: Bug
>          Components: HttpCore NIO
>            Reporter: Paul Donohue
>            Priority: Critical
>             Fix For: 4.2.3
> I have an application using httpcore-nio that occasionally gets stuck in an infinite
loop in SSLIOSession:
> "I/O dispatcher 1" prio=10 tid=0x00002aaab82da000 nid=0x5448 runnable [0x000000004271c000]
>    java.lang.Thread.State: RUNNABLE
>         at sun.security.ssl.SSLEngineImpl.readNetRecord(SSLEngineImpl.java:767)
>         - locked <0x00000000a4f03420> (a sun.security.ssl.SSLEngineImpl)
>         at sun.security.ssl.SSLEngineImpl.unwrap(SSLEngineImpl.java:719)
>         - locked <0x00000000a51faaa0> (a java.lang.Object)
>         at javax.net.ssl.SSLEngine.unwrap(SSLEngine.java:624)
>         at org.apache.http.nio.reactor.ssl.SSLIOSession.doUnwrap(SSLIOSession.java:228)
>         at org.apache.http.nio.reactor.ssl.SSLIOSession.decryptData(SSLIOSession.java:348)
>         at org.apache.http.nio.reactor.ssl.SSLIOSession.isAppInputReady(SSLIOSession.java:371)
>         - locked <0x00000000a4f03498> (a org.apache.http.nio.reactor.ssl.SSLIOSession)
>         at org.apache.http.impl.nio.reactor.AbstractIODispatch.inputReady(AbstractIODispatch.java:118)
>         at org.apache.http.impl.nio.reactor.BaseIOReactor.readable(BaseIOReactor.java:160)
>         at org.apache.http.impl.nio.reactor.AbstractIOReactor.processEvent(AbstractIOReactor.java:342)
>         at org.apache.http.impl.nio.reactor.AbstractIOReactor.processEvents(AbstractIOReactor.java:320)
>         at org.apache.http.impl.nio.reactor.AbstractIOReactor.execute(AbstractIOReactor.java:280)
>         at org.apache.http.impl.nio.reactor.BaseIOReactor.execute(BaseIOReactor.java:106)
>         at org.apache.http.impl.nio.reactor.AbstractMultiworkerIOReactor$Worker.run(AbstractMultiworkerIOReactor.java:604)
>         at java.lang.Thread.run(Thread.java:636)
> In the above example, I was able to determine that this behavior was caused by a client
sending a ClientHello that resumed an established session, immediately followed by a CloseNotify
Alert.  This sequence only sometimes triggers the infinite loop, so the problem appears to
be timing related.  I get the feeling that there are other sequences that may trigger this,
but I don't have any other concrete examples at the moment.

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: dev-unsubscribe@hc.apache.org
For additional commands, e-mail: dev-help@hc.apache.org

View raw message