Don't let plugins fail due to read-only fs

[dmnks]

Read-only file systems fall into the same category as unsupported ones (i.e. writing the desired file metadata to them is meaningless) so handle those gracefully in the plugin hooks, too.

This is useful e.g. when preparing a system for IMA by reinstalling the desired packages with the IMA plugin enabled, including ones that own paths such as /mnt which are meant to be used as mount points, often read-only (e.g. when mounting an ISO).

Fixes: RHEL-33726

[dmnks]

One scenario just came to mind:

1. You mount a directory (as read-only) over another directory that's owned by an installed package
2. The plugin skips the directory silently
3. You unmount the directory
4. The original directory does not have the file attributes applied (!)

In a way, this is a security risk, too.

Maybe we should make this into a warning after all...

[pmatilai]

These should probably be handled the same as fsm does (see commit 09d554d): if there's an error applying, skip if the attributes match, otherwise it's an error.

[dmnks]

Ack, I'll look into that.

[pmatilai]

Of course, the issue of underlying directory not getting updated goes for any (temporary) mounts we happen to touch during a transaction. This may call for stepping back a bit for an overall strategy (fancy expression for "I don't know" 😆 )

We know what mounts we cross paths with, and we technically even know which ones are readonly mounts. We just don't make a good use of that information: if we try to install something on such a filesystem we fail out with "not enough space" which is a bit confusing. Maybe we could automatically add readonly mounts to %_netsharedpath so they get skipped, but we shouldn't do that for an fs where we're actually trying to install to, only for the mountpoints.

And, that still leaves the question of what to do with any temporary mounts.

[dmnks]

Indeed, this needs a step back, failure to write to a file system is not a plugin specific thing, unlike e.g. EOPNOTSUPP that we already handle in plugins. I'll close this PR since it obviously isn't the right solution, and will open a discussion item instead.