unbound-1.19.0 alloc_reg_obtain() core dumps

[Sami Kerola]

On Tue, 20 Feb 2024 at 09:57, Yorgos Thessalonikefs <yorgos at nlnetlabs.nl> wrote:
> I believe this alloc_reg_release() needs to be there and I don't see it
> being called twice on failure, unless I miss something.

Hi Yorgos,

No problem, I just thought maybe the previous crash and this mystery case
have something in common.  Unfortunately I haven't had time to really start
understand unbound internals, so my 'fix might be needed here' guesses are
not very educated.

> As for the cores I started looking around; the latest ones seem really
> weird especially with pointers that seem to change between calls.

Would you like to receive my backtrace collection?  There are a lot of
similar looking backtraces, but maybe with enough samples something catches
eye.

https://github.com/NLnetLabs/unbound/issues/1017

Pretty much everything stops SIGSEGV, exceptions are marked in the files.

> Did you find out if there was a reason systemctl restarts would happen
> unexpectedly, so that we can rule that out?

Only thing I have seen systemd doing is start the service after crash.
Number of restarts is already fairly large, and I have not debugged each of
them what comes to logs so I cannot rule out there could confirm all cases
the same - just the ones I looked.

> An extra question while looking around; when reloading, do you maybe use
> the reload_keep_cache command?

The reload_keep_cache is not in use.

-- 
Sami Kerola
https://kerolasa.iki.fi/