[PATCH] tdb: runtime check for robust mutexes may hang

[Ralph Böhme]

On Mon, Mar 13, 2017 at 06:49:00AM +1300, Andrew Bartlett wrote:
> On Sun, 2017-03-12 at 14:14 +0100, Ralph Böhme via samba-technical
> wrote:
> > 
> > As there's no POSIX function to change the signal mask of all threads
> > in a
> > process, this problem can't be solved by a change to
> > tdb_runtime_check_for_robust_mutexes().
> > 
> > Attached patch *moves* the runtime check to a library initializer
> > which
> > guarantees that only one thread exists in the process when the
> > function is run.
> 
> Why can we assert this is true?
> 
> Surely if tdb is loaded into a program by being linked to a plugin, eg
> dlopen(), the library initializer will run then, with threads already
> running?

golly, you're right. I missed this case.

Back to the drawing board I guess, unless we decide we don't care about this use
case. We probably should care, might actually be used by the Gnome VFS, who
knows.

Cheerio!
-slow