psuedofs changes for foreign distributions

[bladedvox]

fixes #414
fixes #288

tested on Void as well as Ubuntu; thanks to @imiric for actually finding the solution, i just tested it

[Duncaen]

I think both -R and -f flags for umount are just wrong. A plain umount should be able to remove just the bind mountpoint. no lazy needed.

Not sure whether mount --make-rslave is required either.

[bladedvox]

Not really sure why but without these umount_pseudofs fails at the beginning of generate_squashfs on Ubuntu, where it works fine in Void.

[Duncaen]

Because -R tries to unmount sub mounts of /sys like /sys/fs/cgroup first, and mounts are by default shared systemd systems, which results in the busy error, because the same cgroupfs is still in use on the host system.

The change works around it by not propagating mount events (--make-rslave).

But AFAIU this is a non issue if its changed to not recursively umount a bind mount and instead just umount the actual bind mount.

[bladedvox]

Ok good to know. I can give that a test and report back

[Duncaen]

After a bit of discussion on irc, I think mount --make-rslave "$ROOTFS"/$f is fine in combination with just umount -R "$ROOTFS/$f".

This should be the most "correct" solution.

[bladedvox]

just tested that and it doesn't work 😿

[bladedvox]

@Duncaen ran a few more rounds of tests just to be sure; mount --make-rslave and umount -R -l are both necessary; umount -f not strictly necessary but i see why it's there for the edge cases where it makes a difference

[Duncaen]

I don't see why --lazy should be necessary, if that is the case there is a bigger problem.

[bladedvox]

I mean after reading the docs I agree, which makes me think there's something more insidious going on in the way the cgroup mount is done in the first place. But also based on the docs, the lazy unmount does seem to be the safe way to handle the symptom as observed.