[PATCH] Improve code in ctdb lock helper

[Volker Lendecke]

On Fri, Dec 02, 2016 at 04:19:58PM +1100, Amitay Isaacs wrote:
> Volker, I would really appreciate if you can verify that I am doing testing
> correctly.
> 
> Usually it takes few runs to reproduce the problem.  So I run this as
> follows:
> 
>   # while true ; do ./bin/test_mutex_raw /tmp/foobar 10 0 ; done

Thanks for that test program! It works fine on FreeBSD but stalls on
a virtual Debian Jessie x86_64 and on a Debian testing on a Raspi2 for
me. So it seems at least CPU architecture independent.

What this program does exactly matches what I have expected from robust
mutexes. I have been proven wrong just a few days ago by the FreeBSD
developers, so I would not count on this too much :-)

I think we need to raise this with glibc and/or kernel people to explain
to us what we are doing wrong this time. Question is -- who can we
ask? Do you want me to look around for appropriate mailing lists?

Volker