[asterisk-dev] How to fix a crashing spandsp

[Matthew Jordan]

On Tue, Nov 12, 2013 at 3:20 AM, Torrey Searle <tsearle at gmail.com> wrote:

> Hello all,
>
> I have a scenario which can cause asterisk to crash in SpanDSP to crash.
>
> Basically res_fax.c has a 5 second timeout that, if triggered disables
> AST_FAX_TECH_T38.   This causes the T.38 stack not to be initialized and
> the asterisk will crash if any UDPTL packet arrives after that point.
>
> I have already identified two similar call flows that can trigger this.
>
> 1.  We are slow to receive the ACK message from the calling party.  The
> T.38 switchover attempt of ReceiveFax will instead just set the
> NEEDREINVITE flag.  In this case, a T.38 RE-invite can be sent (and
> succeed!) after res_fax has timed out and disabled T.38 => crash
>
>
> 2. We have sent the T.38 re-invite, but are slow to get the 200 OK from
> the carrier, we time out, disable T.38, and the 200ok arrives after we give
> up => crash
>
>
> Problem is, I'm not sure whats the best way of resolving this.  I have
> been able to resolve case 1 by not setting the NEEDREINVITE flag, and
> directly failing to g711 if the ACK hasn't arrived yet, but that still
> leaves me crashing in case 2.
>
> Other ideas I had were to remove the timeout (trust the channel driver to
> report failure on re-invite transaction timeout)
>
> Alternatively I thought it might be good to always initialize the T.38
> stack of span dsp just in case we find out later that we actually need to
> use it.
>
> I would like to get your feed back on what's the best solution for this
> issue
>
>
>
Hey Torrey -

This sounds a lot like ASTERISK-21242 (
https://issues.asterisk.org/jira/browse/ASTERISK-21242), which I was
looking at this week as well. Ashley's patch is to just always initialize
the T38 stack to avoid those kinds of problems. It appears to have resolved
the crashes he saw; it may be worthwhile trying that patch to see if it
resolves the two scenarios you're seeing as well.

The part I wanted to look into still was what impact the initialization has
if we do choose to always initialize it. The other option would be to try
to handle the various off nominal cases in the T.38 state machine handling;
I have a feeling, however, that to do that would be rather difficult.

Matt

-- 
Matthew Jordan
Digium, Inc. | Engineering Manager
445 Jan Davis Drive NW - Huntsville, AL 35806 - USA
Check us out at: http://digium.com & http://asterisk.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.digium.com/pipermail/asterisk-dev/attachments/20131112/722e3ad1/attachment.html>