[DIRMINA-1029] The sent buffer is reset to its original position when using the SSL Filter after a session.write()

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0.13
• Fix Version/s: 2.0.14
• Assignee: Unassigned
• Reporter: Emmanuel Lécharny
• Created: 16/Feb/16 14:07
• Updated: 16/Feb/16 14:41
• Resolved: 16/Feb/16 14:41

Description

Last night, I played around one of our test, adding a SSL version of it. It's basically a TcpServer and a TcpCLient, which sends a message N times.

The TCP version uses a buffer :

    /** The buffer containing the message to send */
    private IoBuffer buffer = IoBuffer.allocate(8);

The main class contains (with commented the Buffer content) :

        // HeapBuffer[pos=0 lim=8 cap=8: 00 00 00 00 00 00 00 00]
        client.buffer.putLong(client.counter.getCount());
        // HeapBuffer[pos=8 lim=8 cap=8: empty]
        client.buffer.flip();
        // HeapBuffer[pos=0 lim=8 cap=8: 00 00 00 00 00 01 86 A0]
        session.write(client.buffer);
        // HeapBuffer[pos=8 lim=8 cap=8: empty]

The same code, when ran on a SSL connection :

        // HeapBuffer[pos=0 lim=8 cap=8: 00 00 00 00 00 00 00 00]
        client.buffer.putLong(client.counter.getCount());
        // HeapBuffer[pos=8 lim=8 cap=8: empty]
        client.buffer.flip();
        // HeapBuffer[pos=0 lim=8 cap=8: 00 00 00 00 00 01 86 A0]
        session.write(client.buffer);
        //HeapBuffer[pos=0 lim=8 cap=8: 00 00 00 00 00 01 86 A0]

As you can see, in one case, the buffer is reset ({(SSL}}) while it is consumed on the other side(non SSL).

It seems to be a clear bug, because we would expect the buffer to have the same value in both case, and I would expect the SSL use case to behave as the non-SSL use case.

The fact is that the SSLFilter.filterWrite() method does :

    ...
                    } else if (sslHandler.isHandshakeComplete()) {
                        // SSL encrypt
                        int pos = buf.position();
                        sslHandler.encrypt(buf.buf());
                        buf.position(pos);                <-------------------- Wrong !
    ...

This buffer reset has been done a very long time ago (10 years and 9 months) to fix a problem :
http://svn.apache.org/viewvc/directory/network/branches/api_integration/src/java/org/apache/mina/filter/SSLFilter.java?r1=169302&r2=169301&pathrev=169302

IMO, the fix is just plain wrong. What should be done is to send back the original buffer which is always in the WriteRequest to the IoHandler.