Inconsistent SSLEngine behavior for closing outbound while in handshake in 11ea22

Xue-Lei Fan xuelei.fan at oracle.com
Mon Aug 13 21:32:30 UTC 2018


On 8/13/2018 2:12 PM, Bradford Wetmore wrote:
> An update.
> 
> 1.  OK vs. BUFFER_UNDERFLOW may still be an issue.  It's not a big one, 
> but potentially confusing as to when you will get either.
> 
Did you figure out when OK is returned, and when BUFFER_UNDERFLOW is 
returned?  I think we have an improvement that if the buffer is not used 
currently, the BUFFER_UNDERFLOW may not be returned.  So OK may be used 
in some circumstances.

Xuelei

> 2.  I thought I saw this in the non-TLSv1.3 contexts also, but I think I 
> must have been reverted my test case back to "TLS", which uses the 
> TLS1.3 1/2 close.  I can't get it to duplicate in TLSv1.2/1.1/1 today. 
> Never mind.
> 
> Brad
> 
> 
> 
> On 8/10/2018 6:14 PM, Bradford Wetmore wrote:
>>
>>
>> On 8/9/2018 5:57 PM, Xuelei Fan wrote:
>>> Thank you, Tim!   Please feel free to submit bugs and comments.
>>
>> I went through a standard SSLEngineTemplate handshake/data/close with 
>> a fine-tooth comb (debugger), and found two bugs based on your -04 patch.
>>
>> 1.  Things may have changed since the original SSLEngine 
>> implementation, but unwrap() was supposed to return a BUFFER_UNDERFLOW 
>> if there is not enough (or any) inbound data to make a packet.  I'm 
>> seeing a mix of OK and BUFFER_UNDERFLOW.  I haven't determined the 
>> pattern, but you should be able to run my test (attached) and see 
>> pretty quickly.
>>
>> 2.   I haven't not had the cycles to follow the earlier discussion, so 
>> this may be a duplicate of something already discussed, but I got 
>> myself into an endless loop because the closes weren't happening as 
>> expected.
>>
>> In the attached file, after the handshaking and data exchange, close 
>> just one of the SSLEngines outbound side.  I'll use client for this 
>> example, but same error with server side:
>>
>> // clientEngine.closeOutbound();
>> clientEngine.getHandshakeStatus() == NEED_WRAP
>> client.isInboundDone(): false
>> client.isOutboundDone(): false
>>
>> // client.wrap() sends the Close Notify (CN).
>> client wrap: CLOSED/NEED_UNWRAP, 0/24 bytes
>> clientEngine.getHandshakeStatus() == NEED_UNWRAP
>> client.isInboundDone(): false
>> client.isOutboundDone(): true
>>
>> // server.unwrap() parses the CN, and closes the inbound side.
>> server unwrap: CLOSED/NEED_WRAP, 24/0 bytes
>> clientEngine.getHandshakeStatus() == NEED_WRAP
>> server.isInboundDone(): true
>> server.isOutboundDone(): false
>>
>> So far so good.
>>
>> // server.wrap() needs to send the corresponding close_notify.
>> server wrap: OK/NEED_WRAP, 0/0 bytes
>> clientEngine.getHandshakeStatus() == NEED_WRAP
>> server.isInboundDone(): true
>> server.isOutboundDone(): false
>>
>> Whoops, it didn't happen.  No outbound data was generated.  State is 
>> still in NEED_WRAP.  I just let the app run, it never closes because 
>> no CN is generated on the server side.
>>
>> Run the attached file for full details.
>>
>> Thanks,
>>
>> Brad
>>
>>> Xuelei
>>>
>>> On 8/9/2018 4:23 PM, Tim Brooks wrote:
>>>> Hi Xuelei,
>>>>
>>>> My test passed using that patch. I’ll continue to explore over the 
>>>> next few days. But that patch resolves the main issues that I had 
>>>> encountered.
>>>>
>>>> Thanks,
>>>>
>>>> - Tim
>>>>
>>>>> On Aug 7, 2018, at 8:54 AM, Xuelei Fan <xuelei.fan at oracle.com 
>>>>> <mailto:xuelei.fan at oracle.com>> wrote:
>>>>>
>>>>> Xuelei
>>>>



More information about the security-dev mailing list