Add support for mounting upper layers with images

[alexlarsson]

This is useful from the cli if you want to add a writable layer to a composefs mount. This is better than doing it in a separate overlayfs layer as that would increase the overlayfs stacking depth, etc.

[alexlarsson]

I was chatting a bit with some people at the embedded recipes conf about using composefs without bootc (for a bit custom stuff) and they would need this to be able to couple composefs images with one or more writable layers.

[Johan-Liebert1]

Some minor suggestions :)

[alexlarsson]

Also, I didn't look at it in detail, but I think composefs-setup-root could use this instead of manually doing the overlayfs mount.

[Johan-Liebert1]

I don't remember well, but I think the "non initramfs" part of composefs-setup-root doesn't work as intended. I could be wrong though

[alexlarsson]

We have a reverse dependency issue in bootc, where it needs to pass composefs::mount::NO_OVERLAY to mount. What is our current API stability guarantees here? Do i add a new mount_with_overlay() for this, or do we just fix bootc later?

[alexlarsson]

Seems easiest to go with mount_with_overlay(), so i added that.

[alexlarsson]

I have no idea what this failure is about:

    ---- privileged_test_digest_equivalence_debian_bootc ----
    command exited with non-zero code `bcvk ephemeral run-ssh --memory 10G localhost/composefs-rs-test-c9s:latest -- cfsctl-integration-tests --exact privileged_test_digest_equivalence_debian_bootc`: 255


    failures:
        privileged_test_digest_equivalence_debian_bootc

    test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 60 filtered out; finished in 265.11s

[cgwalters]

I was chatting a bit with some people at the embedded recipes conf about using composefs without bootc (for a bit custom stuff)

Would it help for us to support building bootc without the ostree dep? I would like to be able to do that.

[cgwalters]

We have a reverse dependency issue in bootc, where it needs to pass composefs::mount::NO_OVERLAY to mount. What is our current API stability guarantees here? Do i add a new mount_with_overlay() for this, or do we just fix bootc later?

I think We can break API at any point, but it's also useful to have a bit of backpressure to avoid doing so unnecessarily. On the counterpoint, #225 does break API and in that PR I pointed the bootc revdep CI at a patch to adapt bootc to it.

[alexlarsson]

I was chatting a bit with some people at the embedded recipes conf about using composefs without bootc (for a bit custom stuff)

Would it help for us to support building bootc without the ostree dep? I would like to be able to do that.

Nah, this was for "Flipper one" and the requirement they had was quite custom, focusing on allowing users to tweak and break their systems. I don't think it will ever be a good fit for bootc.

[cgwalters]

I have no idea what this failure is about:

It's likely bootc-dev/bcvk#153 - a rerun made it pass.

[cgwalters]

Looks sane to me, but I would really like to build up the pattern of having integration tests again because I think they're easy to add.

[cgwalters]

There's more possible options here - volatile is a big one!

I'm not saying we need to do that now, but we could make this #[non_exhaustive] to allow us to more easily add more options later.

Or we could just support pub options: Option<&'str>

[alexlarsson]

I made this a more generic non_exhaustive MountOptions.

[cgwalters]

One thing to bear in mind on this one too is the intersection with #304 - in the general case there what I think would work well is if we supported fd passing - for the mount operation we'd just hand back a detached fd by default.

[alexlarsson]

I updated this to use a more generic MountOption that we can later extend.

[alexlarsson]

I rebased on main and added support for varlink to this

[cgwalters]

OK but you weren't replying to my "needs tests" thread. I think the token cost of doing these things via LLMs is so cheap relative to the value they provide, so I spent some tokens and pushed a commit with a test case.

[alexlarsson]

lgtm