c561e6179e
During the early design of FCOS and RHCOS, we chose a value of 384M for the boot partition. This turned out to be too small: some arches other than x86_64 have larger initrds, kernel binaries, or additional artifacts (like device tree blobs). We'll likely bump the boot partition size in the future, but we don't want to abandon all the nodes deployed with the current size.[[1]] Because stale entries in `/boot` are cleaned up after new entries are written, there is a window in the update process during which the bootfs temporarily must host all the `(kernel, initrd)` pairs for the union of current and new deployments. This patch determines if the bootfs is capable of holding all the pairs. If it can't but it could hold all the pairs from just the new deployments, the outgoing deployments (e.g. rollbacks) are deleted *before* new deployments are written. This is done by updating the bootloader in two steps to maintain atomicity. Since this is a lot of new logic in an important section of the code, this feature is gated for now behind an environment variable (`OSTREE_ENABLE_AUTO_EARLY_PRUNE`). Once we gain more experience with it, we can consider turning it on by default. This strategy increases the fallibility of the update system since one would no longer be able to rollback to the previous deployment if a bug is present in the bootloader update logic after auto-pruning (see [[2]] and following). This is however mitigated by the fact that the heuristic is opportunistic: the rollback is pruned *only if* it's the only way for the system to update. [1]: https://github.com/coreos/fedora-coreos-tracker/issues/1247 [2]: https://github.com/ostreedev/ostree/issues/2670#issuecomment-1179341883 Closes: #2670
libostree
This project is now known as "libostree", though it is still appropriate to use the previous name: "OSTree" (or "ostree"). The focus is on projects which use libostree's shared library, rather than users directly invoking the command line tools (except for build systems). However, in most of the rest of the documentation, we will use the term "OSTree", since it's slightly shorter, and changing all documentation at once is impractical. We expect to transition to the new name over time.
As implied above, libostree is both a shared library and suite of command line tools that combines a "git-like" model for committing and downloading bootable filesystem trees, along with a layer for deploying them and managing the bootloader configuration.
The core OSTree model is like git in that it checksums individual files and has a content-addressed-object store. It's unlike git in that it "checks out" the files via hardlinks, and they thus need to be immutable to prevent corruption. Therefore, another way to think of OSTree is that it's just a more polished version of Linux VServer hardlinks.
Features:
- Transactional upgrades and rollback for the system
- Replicating content incrementally over HTTP via GPG signatures and "pinned TLS" support
- Support for parallel installing more than just 2 bootable roots
- Binary history on the server side (and client)
- Introspectable shared library API for build and deployment systems
- Flexible support for multiple branches and repositories, supporting projects like Flatpak which use libostree for applications, rather than hosts.
Documentation
For more information, see the project documentation or the project documentation website.
Operating systems and distributions using OSTree
Apertis uses libostree for their host system as well as Flatpak. See update documentation and apertis-update-manager
Endless OS uses libostree for their host system as well as Flatpak. See their eos-updater and deb-ostree-builder projects.
For Debian/apt, see also https://github.com/stb-tester/apt2ostree and the LWN article Merkle trees and build systems.
Fedora derivatives use rpm-ostree (noted below); there are 4 variants using OSTree:
Red Hat Enterprise Linux CoreOS is a derivative of Fedora CoreOS, used in OpenShift 4. The machine-config-operator manages upgrades. RHEL CoreOS is also the successor to RHEL Atomic Host, which uses rpm-ostree as well.
GNOME Continuous is where OSTree was born - as a high performance continuous delivery/testing system for GNOME.
GNOME OS is a testing OS that uses libostree for their host system as well as Flatpak.
Liri OS has the option to install their distribution using ostree.
TorizonCore is a Linux distribution for embedded systems that updates via OSTree images delivered via Uptane and aktualizr.
Distribution build tools
meta-updater is a layer available for OpenEmbedded systems.
QtOTA is Qt's over-the-air update framework which uses libostree.
The BuildStream build and integration tool supports importing and exporting from libostree repos.
fedora-iot/otto is a tool that helps ship ostree commits inside Docker/OCI containers and run a webserver to serve the commits.
Fedora coreos-assembler is the build tool used to generate Fedora CoreOS derivatives.
debos is a tool-chain for simplifying the process of building a Debian-based OS image.
Projects linking to libostree
rpm-ostree is used by the Fedora-derived operating systems listed above. It is a full hybrid image/package system. By default it uses libostree to atomically replicate a base OS (all dependency resolution is done on the server), but it supports "package layering", where additional RPMs can be layered on top of the base. This brings a "best of both worlds"" model for image and package systems.
eos-updater is a daemon that implements updates on EndlessOS.
Flatpak uses libostree for desktop application containers. Unlike most of the other systems here, Flatpak does not use the "libostree host system" aspects (e.g. bootloader management), just the "git-like hardlink dedup". For example, Flatpak supports a per-user OSTree repository.
aktualizr is an Uptane-conformant software update client library intended for use in automotive and other security-sensitive embedded devices. It uses OSTree to manage the OS of the host device by default.
Language bindings
libostree is accessible via GObject Introspection; any language which has implemented the GI binding model should work. For example, Both pygobject and gjs are known to work and further are actually used in libostree's test suite today.
Some bindings take the approach of using GI as a lower level and write higher level manual bindings on top; this is more common for statically compiled languages. Here's a list of such bindings:
Building
Releases are available as GPG signed git tags, and most recent versions support extended validation using git-evtag.
However, in order to build from a git clone, you must update the submodules. If you're packaging OSTree and want a tarball, I recommend using a "recursive git archive" script. There are several available online; this code in OSTree is an example.
Once you have a git clone or recursive archive, building is the same as almost every autotools project:
git submodule update --init
env NOCONFIGURE=1 ./autogen.sh
./configure --prefix=...
make
make install DESTDIR=/path/to/dest
Contact and discussion forums
OSTree has a mailing list and
there is also an #ostree channel on Libera.Chat. However, asynchronous+logged
communication is preferred for nontrivial questions.
Contributing
See Contributing.
Licensing
The licensing for the code of libostree can be canonically found in the individual files; and the overall status in the COPYING file in the source. Currently, that's LGPLv2+. This also covers the man pages and API docs.
The license for the manual documentation in the doc/ directory is:
SPDX-License-Identifier: (CC-BY-SA-3.0 OR GFDL-1.3-or-later)
This is intended to allow use by Wikipedia and other projects.
In general, files should have a SPDX-License-Identifier and that is canonical.