Correct SIGINT and SIGQUIT termination in debug mode

[EllezDi]

If server starts without -m option, then SIGINT and SIGQUIT signal handlers doesn't init.

The -m option for memory cleanup then correctly processed in sig_fatal function.

[alandekok]

Why is this "correct"? The SIGINT and SIGQUIT handlers shouldn't be set when -m isn't used.

[EllezDi]

On my debian 8.2 radiusd -XXX didn't terminate on Ctrl-C (terminates with radiusd -mXXX).
So, when I explicitly sets the fr_set_signal(SIGINT, sig_fatal); - it does.
My thoughts - this is because no one handles the SIGINT signal.
Maybe I'm wrong ?

[alandekok]

If there is no signal handler for CTRL-C, then the process should quit. It works fine on my system when using radiusd -X and CTRL-C.

[EllezDi]

I can't believe. Magic! After your post it works !! (recompiled distro 2 times). =) Thanks =)

[arr2036]

I've noticed some modules like rlm_ruby, register their own signal handlers which can cause this to happen.

[EllezDi]

I found it. When rlm_sql_oracle loads - it does not terminates on SIGINT. Only with SIGQUIT.
Without rlm_sql - works fine for both signals.

[arr2036]

We should probably try to remove any registered signal handlers (that we didn't add) after module instantiation. @alandekok thoughts?

[alandekok]

While I thing it's stupid for libraries to register signal handlers, I'm not sure it's worth adding more code to the server to work around stupid crap.

[arr2036]

@EllezDi can you confirm this fixes the issue? Then i'll merge to v3.0.x?

[EllezDi]

@arr2036 on 3.0.10 - your patch didn't fixes the issue. Oracle client 12.1 x64 (db 10 or 11)
export BEQUEATH_DETACH=YES - still didn't work
As I see in docs, bequeathed connections are all local. So for Net8 it may not work.