Force expiring automounts breaks cvmfs namespaced mounts

Currently the eos-core package forces an automount expiry of all unused automounted filesystems.

Unfortunately this can cause CvmFS mounts which are only used within a namespace to break[4]. The only protection we have currently against this to run an external process that pings mounts in the main namespace that are mounted in a namespace only. [3]

However this forced expiry causes them to unmount anyway resulting in broken mounts.

This is fixed[1] in kernel 5.15 but that that is some time of from production of course.

Trying to reproduce this with EOS it behaves slightly differently.

server> ls /eos/user/u/user
server> podman run -it -v $XDG_RUNTIME_DIR/krb5cc:/tmp/krb5cc_0 -v /eos:/eos gitlab-registry.cern.ch/linuxsupport/c8-base /bin/bash
container> ls /eos/user/u/user
server> /usr/bin/killall --quiet --signal USR1 automount
container> mount | eos
/etc/auto.eos on /eos type autofs (rw,relatime,fd=24,pgrp=0,timeout=200,minproto=5,maxproto=5,indirect,pipe_ino=368651)
user on /eos/user type fuse (rw,nosuid,nodev,relatime,user_id=0,group_id=0,allow_other)
home-s on /eos/home-s type fuse (rw,nosuid,nodev,relatime,user_id=0,group_id=0,allow_other)
server>  mount | eos
/etc/auto.eos on /eos type autofs (rw,relatime,fd=24,pgrp=10607,timeout=200,minproto=5,maxproto=5,indirect,pipe_ino=368651)

It expires the mount outside but you can still mount again with a ls on the outside. With CvmFS the outside mount is broken. That is intresting for CvmFS but this forced expire should still go I believe. The timeout will do the job.

There's some background in CvmFS ticket[2]

cern-eos / eos

Force expiring automounts breaks cvmfs namespaced mounts #35