Hi Eric, Thank you for your quick reply! This information might be interesting for the developers of mod_auth_openidc. I will show them your message and report back here if we manage to solve the issue. Best, Simon -----Ursprüngliche Nachricht----- Von: Eric Covener <covener@xxxxxxxxx> Gesendet: Montag, 2. März 2020 17:01 An: users@xxxxxxxxxxxxxxxx Betreff: Re: Segmentation faults on graceful reload of Apache httpd with APR 1.7.0 and mod_auth_openidc On Mon, Mar 2, 2020 at 10:59 AM Eric Covener <covener@xxxxxxxxx> wrote: > > On Mon, Mar 2, 2020 at 9:50 AM <simon.studer@xxxxxxx> wrote: > > > > Hi everyone, > > > > We have been having issues with segmentation faults on graceful restarts of Apache httpd in combination with the module mod_auth_openidc. > > > > After a back and forth with one of the developers over on Github, we came to the finding that the segmentation faults occur with version 1.7.0 of the APR but not version 1.6.5. The full discussion can be found in the Github issue at https://github.com/zmartzone/mod_auth_openidc/issues/458. > > > > To reliably reproduce the problem, my coworker created a Docker container and uploaded it to Docker Hub at https://hub.docker.com/r/ifrido/httpd-openidc-debug with detailed steps to produce the segmentation fault on the Overview page. The corresponding Dockerfile can be viewed at https://hub.docker.com/r/ifrido/httpd-openidc-debug/dockerfile. > > Hi Simon, re: your GH issue and specifically the snippet here: > https://github.com/zmartzone/mod_auth_openidc/issues/458 Argh, meant https://github.com/zmartzone/mod_auth_openidc/issues/458#issuecomment-581413230 > > To me it's a bit bizarre that this SHM cache does cleanups in the > child processes. > The SHM is created by the parent process and would normally be > destroyed in the parent process. This would be the usual handling of > shared memory AFAICT. > > Since "context" itself is not in shared memory, there is no way for > one child to tell another that it has destroyed the SHM segment. > Also, nothing modules should not really write to the pointers hung off > their per-module or per-directory apache configs. It is another red > flag in this same short snippet. > > Unless the maintainers have some trivia about why this cleanup of SHM > should run in the parent, I think it's the culprit. -- Eric Covener covener@xxxxxxxxx --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx
|