[Bacula-devel] Bacula Director reload race condition/crash

[Alexander W. <mai...@wi...>]

Hi Bacula developers,

We have stumbled on a race condition where two concurrent reload request
can cause bacula-dir to abort().

The following is enough to crash bacula-dir in most cases:

(echo reload | bconsole)&  (echo reload | bconsole)&

One of our developers Jeremy Peterson has been working on this problem and
has traced the abort statement to dird.c:

void reload_config(int sig)
    static bool already_here = false;
    if (already_here) {
       abort();                        /* Oops, recursion ->  die */
    }

Patching Bacula so that the abort() doesn't fire mitigates the race
condition and the director doesn't die. This piece of code has not changed
since since moving from cvs to git (in 2002).

I assume the abort statement is there for a reason, to stop recursion, but
we are not sure what would cause this recursion and therefore I'm asking
what the correct fix for this bug should be.


Sincerely,
Alexander Wigen