Dear Maintainer, The SystemD 240 update has changed the handling of NOFILE for the init process and processes it directly spawns. See: https://github.com/systemd/systemd/pull/10244 Unfortunately, it seems that the patch above, which is forcing NOFILE to "infinity" (effectively 1G?) is now having a serious adverse effect on various processes that are spawned by SystemD directly, see: https://github.com/systemd/systemd/issues/10921 and a KDE init bug similarly. I can't find a bug reporting this to debian, even though the root cause seems to be this patch to force "infinity" onto PID 1. Hope this helps.
Hello, I am aware of https://github.com/systemd/systemd/issues/10921 and see that you commented on that PR before you filed this one here, but you have attributed the root cause to the wrong patch. Retitling this bug. This will take some time to fix correctly; I don't agree with systemd upstream's analysis that PAM should not reset limits, but clearly the limits we are ending up with have negative consequences at present. I think it is also unequivocally a bug in the other packages to perform any operations based on the NOFILE hard limit instead of the soft limit, and those packages should also be fixed regardless as this is a latent bug whether or not we change the default hard limit in Debian back down.
PAM Bug report https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=990412
On Thu, 7 Oct 2021 23:06:34 +0200 Chris Hofstaedtler <zeha@debian.org> wrote: outcome, it will package) so that "pam_limits.so". might be a and way if pid 1 set_all) pick it up In 2.38 util-linux started setting some defaults in su, so I don't think the original downstream change is needed anymore: https://github.com/util-linux/util-linux/commit/08273c672b105602e1a9031160ccefec171b02ed I am going to revert the change from #917167 that stopped the default fd limit from being bumped, sometimes next week. If changes are needed to deal with this in the pam/util-linux config/patches, I would appreciate if they could please be taken care of for Trixie. Thanks.