[Freeswitch-users] WSS SSL errors "decryption failed or bad record mac" under load

[Michael Jerris]

if you can reproduce this reliably, i’d try master as well.  Unless this is a bug in openssl, i can’t imagine how dtls would come into play in something like this.

> On May 11, 2017, at 5:48 PM, Luke Wahlmeier <lwahlmeier at gmail.com> wrote:
> 
> I keep semi-regularly running into issues using the wss transport when using dtls/strp/ice.  This is on the latest 1.6.17~34~0fc0946 on Debian jessie, but I am pretty sure it was happening on the last couple releases as well.
> 
> It seems like something bad/wrong happens to the encrypted data going over the websocket coming from freeswitch when more then 1 websocket connection are going and so far ice/srtp/dtls also seem to be needed in the invite to duplicate it.
> 
> I have tried many different languages and network/ssl stacks and keep running into this.  It is always on data coming in from freeswitch on the websocket connection, and its very very random.  Sometimes I will get it 20 times in a row, other times it takes thousands of connections/sessions before it happen.  It also, obviously, completely goes away if I use plain ws instead wss.
> 
> Here are the errors:
> python:
> SSLError: [SSL: DECRYPTION_FAILED_OR_BAD_RECORD_MAC] decryption failed or bad record mac (_ssl.c:1750)
> c/c++ (stunnel4):
> SSL_read: 1408F119: error:1408F119:SSL routines:SSL3_GET_RECORD:decryption failed or bad record mac
> Java:
> java.lang.IllegalArgumentException: Bad arguments
>     at javax.crypto.Mac.update(Mac.java:509)
>     at sun.security.ssl.MAC.compute(MAC.java:135)
>     at sun.security.ssl.InputRecord.checkMacTags(InputRecord.java:265)
>     at sun.security.ssl.InputRecord.decrypt(InputRecord.java:216)
>     at sun.security.ssl.EngineInputRecord.decrypt(EngineInputRecord.java:177)
>     at sun.security.ssl.SSLEngineImpl.readRecord(SSLEngineImpl.java:974)
>     at sun.security.ssl.SSLEngineImpl.readNetRecord(SSLEngineImpl.java:907)
>     at sun.security.ssl.SSLEngineImpl.unwrap(SSLEngineImpl.java:781)
>     at javax.net.ssl.SSLEngine.unwrap(SSLEngine.java:624)
> 
> Attached are a simple python script to do the load, my dialplan and sip_profile.  The python script can take a few runs before it see the error, and I know its not completing the sip or rtp, but even if it does this still happens.
> 
> I have also looked at libsofia-sip-ua/tport/ws.c and I dont see anything obvious.  I am getting setup to build v1.6 head and test this any guidance on ways I can trouble shoot this better or requests for more info are very welcome.
>