Chimera Kernel Module System
Version 0.1.0
This is a lightweight alternative to DKMS (https://github.com/dell/dkms). As DKMS is full of cruft and is essentially a massive bash script, I felt a change was needed. In general CKMS works more or less the same, and has the same filesystem layout, in order to make rewriting DKMS config files and scripts easy. It is, however, written entirely from scratch.
It is currently an incomplete work in progress.
See the examples/ directory for some module definitions.
Requirements
- Python 3.9 or newer
Usage
CKMS modules are collections of kernel modules, with ckms.ini defining
their metadata. They can have three states:
- added (registered with CKMS)
- built (done building but not installed yet)
- installed
- disabled
Each state is for a specific kernel version, except added, which is global.
To register a CKMS module, you can do something like this:
$ ckms add /usr/src/foo-1.0
This assumes ckms.ini exists in the directory. If it does not, you will
need to specify it manually via -c or --modconf.
Once done, the module will be added and you will no longer refer to it
by the full path. You can build it:
$ ckms build foo=1.0
It is still possible to refer to it by path too, this time in the database, e.g.
$ ckms build /var/lib/ckms/foo/1.0
That will build the module for the current kernel. If you want to build it
for another kernel, use the -k or --kernver parameter. The ckms.ini
is installed into the state directory with add, so you no longer have to
worry about it. You can still specify -c or --modconf manually if you
wish to override it for some reason.
It is possible to disable a module for a specific kernel version. A module is
disabled if the ckms-disable directory exists in the kernel module directory,
containing packagename, it itself containing packageversion. If this is done,
ckms will not allow you to build the module, and it will show as disabled in
status.
If disabled after it is built, it will show as built+disabled in status
and it will not be installable. If disabled after it is installed, it will
still show as installed in status and you will be able to uninstall it.
You will be able to clean it if built, regardless of installed status.
This functionality can be used e.g. by package managers to prevent CKMS from building modules that are managed through packaged binaries for specific kernels.
Once built, you can install it similarly, with
$ ckms install foo=1.0
Keep in mind that in order for the system to work, the CKMS state directory,
which is /var/lib/ckms by default, needs to exist. You can run most commands
as the user who owns /var/lib/ckms, which should in general not be root.
The only exception is install, which by default touches system locations
and therefore should be run as root unless you are installing into a special
destdir (which will also prevent depmod from running). Same goes for the
inverse of install, i.e. uninstall.
If you run non-install (or uninstall) steps as root, CKMS will drop
privileges to the owner of /var/lib/ckms.
Once installed, the modules are ready to be used. CKMS will run depmod if
the modules are installed in a real kernel, and will refresh your initramfs
if the module requires it and if there is an appropriate hook in place.
Unlike DKMS, CKMS is primarily designed to be integrated into package managers of distributions and avoids including any features that would overlap with that. Therefore, there is no support for e.g. distributing and managing tarballs, or binary modules, or so on.
Also, CKMS does not manage multiple kernels during one run. You have to run the build/install process for every kernel separately.
The install step will not run unless build has been run, and likewise
build will not run without add. The remove command (which unregisters
a CKMS module) will not run if a module is still built for some kernels. You
have to uninstall them before doing so.
TODO
- Fallback build helpers
- Shell expression option for boolean metadata
- Module signing
- More hooks
- More validation/sanity checking
- ...