Bug#601659: Double free bug in libffado2

[Max Kellermann]

On 2010/10/28 11:18, Adrian Knoth <adi at drcomp.erfurt.thur.de> wrote:
> I bet it happens upon termination. That's
> 
>    http://subversion.ffado.org/ticket/306

Yes it looks similar, and actually I was able to reproduce the JACK
crash easily with MPD as a JACK client - after I closed qjackctl, JACK
also terminated the connection to MPD (why??), and shut itself down.
(Understand now why I would want to remove JACK from the equation?
It's yet another piece of software that can cause trouble.)

But this bug report was not upon termination.  MPD would have lived on
happily after, had it not been killed by libffado2.

> and is something weird on the new Juju firewire stack. It won't happen
> on the old ieee1394.
> 
> We know it since Linux-2.6.32-rcX, when we were first able to run FFADO
> on the new stack.

Which means this applies to the official Debian kernel.  Not good.

> I've spent many hours trying to find the culprit, and so did two kernel
> firewire developers and one FFADO dev.
> 
> I bet it's a race condition that could be fixed with a lock.
> 
> 
> If you like to debug this issue, many FFADO users would be glad. ;)

I would be glad if I didn't have to clean up yet another crappy
library.  And ffado doesn't exactly like a fun project - over complex
(creates 8 threads! wtf?), bad documentation, badly designed API, full
of invalid memory accesses and uninitialized variables (according to
valgrind).