rpm-ostree/docs/manual/treefile.md
Colin Walters f113fc5e27 Rework treecompose kernel processing
Prep for changing `boot_location: new` to use `/usr/lib/ostree-boot`
and `/usr/lib/modules`.  Rework our kernel postprocessing
so that we unify the `boot_location` handling with initramfs generation.

Instead of doing the initramfs first in postprocessing, we do it nearly last,
after e.g. `etc` is renamed to `usr/etc`. This has some consequences, such as
the fact that `run_bwrap_mutably()` is now called in both situations. In
general, our handling of `etc` is inconsistent, although understandably so.

As part of this, I finally got around to implementing the bit from
https://github.com/systemd/systemd/pull/4174 however suboptimal it is; need the
unified core so we can cleanly ignore the posttrans like we do others.  We
intentionally keep the file around in the generated tree so that installing a
kernel RPM per client doesn't try to do any of this either.

This all gets folded together so that the logic for handling the bootloader gets
simpler - in the Fedora case, we now know to find kernels in `/usr/lib/modules`
and can ignore `/boot`.

Closes: #959
Approved by: jlebon
2017-09-12 14:26:44 +00:00

9.5 KiB

Treefile

A "treefile" is a made up term for a JSON-formatted specification used as input to rpm-ostree compose tree to bind "set of RPMs with configuration" to "OSTree commit".

It's recommended to keep them in git, and set up a CI system like Jenkins to operate on them as it changes.

It supports the following parameters:

  • ref: string, mandatory: Holds a string which will be the name of the branch for the content.

  • gpg_key string, optional: Key ID for GPG signing; the secret key must be in the home directory of the building user. Defaults to none.

  • repos array of strings, mandatory: Names of yum repositories to use, from any files that end in .repo, in the same directory as the treefile. rpm-ostree compose tree does not use the system /etc/yum.repos.d, because it's common to want to compose a target system distinct from the one the host sytem is running.

  • selinux: boolean, optional: Defaults to true. If false, then no SELinux labeling will be performed on the server side.

  • boot_location: string, optional: Historically, ostree put bootloader data in /boot. However, this has a few flaws; it gets shadowed at boot time, and also makes dealing with Anaconda installation harder. There are 3 possible values:

    • "both": the default, kernel data goes in /boot and /usr/lib/ostree-boot
    • "legacy": Now an alias for "both"; historically meant just "boot"
    • "new": kernel data goes in /usr/lib/ostree-boot and /usr/lib/modules
  • etc-group-members: Array of strings, optional: Unix groups in this list will be stored in /etc/group instead of /usr/lib/group. Use this option for groups for which humans should be a member.

  • install-langs: Array of strings, optional. This sets the RPM _install_langs macro. Set this to e.g. ["en_US", "fr_FR"].

  • mutate-os-release: String, optional. This causes rpm-ostree to change the VERSION and PRETTY_NAME fields to include the ostree version, and adds a specific OSTREE_VERSION key that can be easier for processes to query than looking via ostree.

  • documentation: boolean, optional. If this is set to false it sets the RPM transaction flag "nodocs" which makes yum/rpm not install files marked as documentation. The default is true.

  • packages: Array of strings, mandatory: Set of installed packages. comps groups are currently not supported due to walters having issues with libcomp: 36d18b4905

  • packages-$basearch: Array of strings, optional: Set of installed packages, used only if $basearch matches the target architecture name.

  • bootstrap_packages: Array of strings, optional: Deprecated; you should now just include this set in the main packages array.

  • units: Array of strings, optional: Systemd units to enable by default

  • default_target: String, optional: Set the default systemd target

  • initramfs-args: Array of strings, optional. Passed to the initramfs generation program (presently dracut). An example use case for this with Dracut is --filesystems xfs,ext4 to ensure specific filesystem drivers are included.

  • remove-files: Array of files to delete from the generated tree.

  • remove-from-packages: Array, optional: Delete from specified packages files which match the provided array of regular expressions. This is safer than remove-files as it allows finer grained control with less risk of too-wide regular expressions.

    Each array element is an array, whose first member is a package name, and subsequent members are regular expressions (compatible with JavaScript).

    Example: remove-from-packages: [["cpio", "/usr/share/.*"], ["dhclient", "/usr/lib/.*", "/usr/share/.*"]]

    Note this does not alter the RPM database, so rpm -V will complain.

  • preserve-passwd: boolean, optional: Defaults to true. If enabled, and check-passwd has a type other than file, copy the /etc/passwd (and /usr/lib/passwd) files from the previous commit if they exist. If check-passwd has the file type, then the data is preserved from that file to /usr/lib/passwd. This helps ensure consistent uid/gid allocations across builds. However, it does mean that removed users will exist in the passwd database forever.

  • check-passwd: Object, optional: Checks to run against the new passwd file before accepting the tree. All the entries specified should exist (unless ignored) and have the same values or the compose will fail. There are four types: none (for no checking), previous (to check against the passwd file in the previous commit), file (to check against another passwd file), and data to specify the relevant passwd data in the json itself. Note that if you choose file, and preserve-passwd is true then the data will be copied from the referenced file and not the previous commit.

    Example: check-passwd: { "type": "none" } Example: check-passwd: { "type": "previous" } Example: check-passwd: { "type": "file", "filename": "local-passwd" } Example: check-passwd: { "type": "data", "entries": { "bin": 1, "adm": [3, 4] } } See also: ignore-remove-users

  • check-groups: Object, optional: Checks to run against the new group file before accepting the tree. All the entries specified should exist (unless ignored) and have the same values or the compose will fail. There are four types: none (for no checking), previous (to check against the group file in the previous commit), file (to check against another group file), and data to specify the relevant group data in the json itself. Note that if you choose file, and preserve-passwd is true then the data will be copied from the referenced file and not the previous commit.

    Example: check-groups: { "type": "none" } Example: check-groups: { "type": "previous" } Example: check-groups: { "type": "file", "filename": "local-group" } Example: check-groups: { "type": "data", "entries": { "bin": 1, "adm": 4 } } See also: ignore-remove-groups

  • ignore-removed-users: Array, optional: Users to ignore if they are missing in the new passwd file. If an entry of * is specified then any user can be removed without failing the compose.

    Example: ignore-removed-users: ["avahi-autoipd", "tss"]

  • ignore-removed-groups: Array, optional: Groups to ignore if they are missing in the new group file. If an entry of * is specified then any group can be removed without failing the compose.

    Example: ignore-removed-groups: ["avahi"]

  • releasever: String, optional: Used to set the librepo $releasever variable, commonly used in yum repo files.

    Example: releasever: "26"

  • automatic_version_prefix: String, optional: Set the prefix for versions on the commits. The idea is that if the previous commit on the branch to the doesn't match the prefix, or doesn't have a version, then the new commit will have the version as specified. If the prefix matches exactly, then we append ".1". Otherwise we parse the number after the prefix and increment it by one and then append that to the prefix.

    This means that on an empty branch with an automatic_version_prefix of "22" the first three commits would get the versions: "22", "22.1", "22.2"

    Example: automatic_version_prefix: "22.0"

  • postprocess-script: String, optional: Full filesystem path to a script that will be executed in the context of the target tree. The script will be copied into the target into /tmp, and run as a container (a restricted chroot, with no network access). After execution is complete, it will be deleted.

    It is strongly recommended to avoid using this except as a last resort. Having the system generated through RPMs allows administrators to understand the inputs to the system. Any new files created through this mechanism will not have the versioning inherent in RPM.

    Only the script file will be copied in; thus if it has any dependencies, on data beyond what is in the target tree, you must embed them in the binary itself.

    An example use for this is working around bugs in the input RPMs that are hard to fix in stable releases.

    Note this does not alter the RPM database, so rpm -V will complain.

  • include: string, optional: Path to another treefile which will be used as an inheritance base. The semantics for inheritance are: Non-array values in child values override parent values. Array values are concatenated. Filenames will be resolved relative to the including treefile.

  • container: boolean, optional: Defaults to false. If true, then rpm-ostree will not do any special handling of kernel, initrd or the /boot directory. This is useful if the target for the tree is some kind of container which does not have its own kernel.

  • add-files: Array, optional: Copy external files to the rootfs.

    Each array element is an array, whose first member is the source file name, and the second element is the destination name. The source file must be in the same directory as the treefile.

    Example: "add-files": [["bar", "/bar"], ["foo", "/foo"]]

  • tmp-is-dir: boolean, optional: Defaults to false. By default, rpm-ostree creates symlink /tmp/sysroot/tmp. It's more flexible to leave it as a directory (systemd will mount it), and further, we don't want to encourage /sysroot to be writable. For host system composes, we recommend turning this on; it's left off by default to ease the transition.