cli/cfsctl: Use proxy from composefs-rs

[Johan-Liebert1]

• Bump composefs-rs
  Also, update the git URL to point to
  https://github.com/composefs/composefs-rs

• cli/cfsctl: Use proxy from composefs-rs
  Now that we have cfsctl as a library, we simply use that to proxy all of
  our bootc internal cfs commands. Cleans up the codebase a bit, DRYs
  it and keeps things in sync

[cgwalters]

Nice!

[gemini-code-assist]

Code Review

The pull request successfully migrates the cfsctl functionality into an external composefs-rs library, which is a positive step towards code modularity and adherence to DRY principles. The dependency updates across Cargo.lock, Cargo.toml, and deny.toml are consistent with this change. The removal of the internal cfsctl.rs module and the corresponding updates in cli.rs and lib.rs correctly reflect the new architecture.

[Johan-Liebert1]

Ok wow

error: Installing to disk: Setting up composefs boot: Failed to mount composefs image: Mounting image '3ba387b6e47dacf983937fd9e3824cef20ebdce824199215dfa803fbe9a6e6b7379539fad23444667aeba209739b395a51345ba1f57dfc17e50a81da7fbf0074': Creating filesystem mount: Invalid argument (os error 22)

[Johan-Liebert1]

Looks like an issue with creating loop device

write(2, "[crates/composefs/src/mountcompa"..., 69[crates/composefs/src/mountcompat.rs:138:5] "make_erofs_mountable" = ) = 69
write(2, "\"", 1")                       = 1
write(2, "make_erofs_mountable", 20make_erofs_mountable)    = 20
write(2, "\"", 1")                       = 1
write(2, "\n", 1
)                       = 1
openat(AT_FDCWD, "/dev/loop-control", O_RDWR|O_CLOEXEC) = 12
ioctl(12, LOOP_CTL_GET_FREE)            = 1
openat(AT_FDCWD, "/dev/loop1", O_RDWR|O_CLOEXEC) = 13
ioctl(13, _IOC(_IOC_WRITE, 0x4c, 0xa, 0x130), 0x7ffe9710f9a8) = -1 EINVAL (Invalid argument)

Precisely the issue seems to be here https://github.com/composefs/composefs-rs/blob/022c9c62216615af471acaa78f486203a53f06e3/crates/composefs-ioctls/src/loop_device.rs#L154

Using raw ioctl fixes the issue

        if libc::ioctl(loop_dev.as_raw_fd(), 0x4C0A, &config) < 0 {
            println!("IOCTL RAW FAILED");
        }

which is really weird to me as I don't see any difference between the outputs for both produced by strace

[Johan-Liebert1]

We regressed with this commit composefs/composefs-rs@984ce03

But I don't see the issue immediately

[Johan-Liebert1]

Okay issues seems to be that _IOC(_IOC_WRITE, 0x4c, 0xa, 0x130) resolves to 0x41304c0a and not 0x4c0a which is what LOOP_CONFIGURE should be

[Johan-Liebert1]

Fix here composefs/composefs-rs#240
Replaces the _IO macros with hardcoded opcodes

[cgwalters]

I think this PR is trying to solve the same problem ("ro sysroot") in two different ways.

The simplest and IMO best way is what you've already done in the test code below: change the cfsctl invocation to operate on something in /var/tmp.

Changing this core storage code is I believe unnecessary and conceptually wrong because this particular module is really about the bootc-owned storage only.

[Johan-Liebert1]

Yes, in hindsight this makes sense. Reverted the change in get_ensure_composefs

[cgwalters]

BTW once this lands we should also be able to bump the bootc commit in composefs-rs and restore the revdep check.