ostree/docs/composefs.md
Eric Curtin bc5c0717fc docs/atomic-rollbacks: Add a section on rollbacks
Describing how different types of rollbacks work.

Signed-off-by: Eric Curtin <ecurtin@redhat.com>
2024-02-13 17:07:17 +00:00

119 lines
4.6 KiB
Markdown

---
nav_order: 110
---
# Using composefs with OSTree
{: .no_toc }
1. TOC
{:toc}
## composefs
The [composefs](https://github.com/containers/composefs) project is a new
hybrid Linux stacking filesystem that provides many benefits when
used for bootable host systems, such as a strong story for integrity.
At the current time, integration of composefs and ostree is experimental.
[This issue](https://github.com/ostreedev/ostree/issues/2867) tracks the latest status.
### Enabling composefs (unsigned)
When building a disk image *or* to transition an existing system, run:
```
ostree config --repo=/ostree/repo set ex-integrity.composefs true
```
This will ensure that any future deployments (e.g. created by `ostree admin upgrade`)
have a `.ostree.cfs` file in the deployment directory which is a mountable
composefs metadata file, with a "backing store" directory that is
shared with the current `/ostree/repo/objects`.
### composefs configuration
The `ostree-prepare-root` binary will look for `ostree/prepare-root.conf` in `/etc` and
`/usr/lib` in the initramfs. Using that configuration file you can enable composefs,
and specify an Ed25519 public key to validate the booted commit.
See the manpage for `ostree-prepare-root` for details of how to configure it.
### Injecting composefs digests
When generating an OSTree commit, there is a CLI switch `--generate-composefs-metadata`
and a corresponding C API `ostree_repo_commit_add_composefs_metadata`. This will
inject the composefs digest as metadata into the ostree commit under a metadata
key `ostree.composefs.v0`. Because an OSTree commit can be signed, this allows
covering the composefs fsverity digest with a signature.
### Signatures
If a commit is signed with an Ed25519 private key (see `ostree
--sign`), and `composefs.keyfile` is specified in `prepare-root.conf`,
then the initrd will find the commit being booted in the system repo
and validate its signature against the public key. It will then ensure
that the composefs digest being booted has an fs-verity digest
matching the one in the commit. This allows a fully trusted read-only
/usr.
The exact usage of the signature is up to the user, but a common way
to use it with transient keys. This is done like this:
* Generate a new keypair before each build
* Embed the public key in the initrd that is part of the commit.
* Ensure the initrd has a `prepare-root.conf` with `[composefs] enabled=signed`, and either use `keypath` or inject `/etc/ostree/initramfs-root-binding.key`; for more see `man ostree-prepare-root`
* After committing, run `ostree --sign` with the private key.
* Throw away the private key.
When a transient key is used this way, that ties the initrd with the
userspace part from the commit. This means each initrd can only boot
the very same userspace it was made for. For example, if an older
version of the OS has a security flaw, you can't boot a new fixed
(signed) initrd and have it boot the older userspace with the flaw.
## Requirements
The current default composefs integration in ostree does not have any
requirements from the underlying kernel and filesystem other than
having the following kernel options set:
- `CONFIG_OVERLAY_FS`
- `CONFIG_BLK_DEV_LOOP`
- `CONFIG_EROFS_FS`
At the current time, there are no additional userspace runtime requirements.
## Status
**IMPORTANT** The integration with composefs is experimental and subject to change. Please
try it and report issues but do not deploy to production systems yet.
## Compatiblity
One issue that ostree users transitioning to composefs may hit is that it is no
longer possible to add new toplevel directories via the `chattr -i / && mkdir /somedir && chattr -i`
trick. A bit more on this in the following issues:
* <https://github.com/coreos/rpm-ostree/issues/337>
* <https://github.com/ostreedev/ostree/pull/2681>
However, users who were doing things like this probably want to enable the
`root.transient` option; see `man ostree-prepare-root` which will allow
this (but also change other behaviors too).
## Comparison with other approaches
There is also support for using [IMA](ima.md) with ostree. In short, composefs
provides much stronger and more efficient integrity:
- composefs validates an entire filesystem tree, not just individual files
- composefs makes files actually read-only, whereas IMA does not by default
- composefs uses fs-verity which does on-demand verification (IMA by default does a full readahead of every file accessed, though IMA can also use fs-verity as a backend)
## Further references
- https://github.com/containers/composefs
- https://www.kernel.org/doc/html/next/filesystems/fsverity.html
<!-- SPDX-License-Identifier: (CC-BY-SA-3.0 OR GFDL-1.3-or-later) -->