Chimera at FOSDEM 2023 and the path towards alpha

It has been a while without an update post, so perhaps it’s time to refresh things a little.

No news does not mean progress hasn’t been happening; there has been a continuous stream of commits in cports as well as other parts of the project, so I will do my best to summarize it as well as provide an updated overview of what’s going to happen.

FOSDEM 2023

FOSDEM 2023 is happening once again with an in-person format, as usual in Brussels, on the first weekend of February, which is the 4th and 5th this time. I will be giving a talk about Chimera, this time in the BSD devroom (huge thanks to the organizers for letting me have a slot, despite this project being a Linux system).

I will give a general overview of the project, our progress since last FOSDEM, as well as what’s planned for the future, and perhaps more, in the form of a full length talk (we have a 50 minute slot). The devroom changes into the LLVM devroom right afterwards, which is fitting considering we are also using the LLVM toolchain.

Cports progress since last post

The previous post was at the beginning of November, which is two and a half months ago. Since then, there has been a lot of updates in the project. Here are the main highlights, in chronological order.

1. A general refresh of packaging templates, with everything being updated to its most recent version.
2. Our suite of Dinit services, dinit-chimera has received a complete overhaul. Besides being more fine-grained, it also provides a cleaned up targets system, better thought-out configuration, and better integration.
3. Full-disk encryption is now supported, besides a variety of other initramfs improvements, which includes better support for LVM, root on ZFS and others.
4. CKMS, our kernel module source build system that replaces DKMS, got an initial release, and no longer conflicts with binary modules (so you can have binary ZFS for some kernels while letting CKMS manage it for others without interfering).
5. We now use a custom version of the musl libc, which uses Scudo (a part of LLVM and default in Android/Fuchsia) as the system allocator (malloc implementation). This brings significantly better performance in multithreaded scenarios.
6. A big overhaul of kernel packaging, alongside Linux 6.1, which is the new baseline version. The new packaging brings support for kernel backups on upgrades besides other things.
7. Cbuild hardening overhaul, with significantly expanded list of hardening types, and new defaults. Now, templates are built with UBSan integer overflow checks by default, as well as hidden visibility and CFI (Control Flow Integrity) by default. Enabling templates to properly use it is still a work in progress. There is also initial infrastructure for other hardening including Intel CET and ARM BTI (which will both need support in musl to be useful) as well as Clang SafeStack. All ELF files are now also checked for executable stack in the build system.
8. Cbuild now supports locking, preventing race conditions when building multiple things in parallel. The sources are properly locked, as are the repositories when generating packages.
9. Cbuild no longer requires fakeroot in the host system.
10. New policy packages base-devel and base-devel-static. These provide a way for users to declare that they want development packages to be automatically installed alongside runtime packages. This allows users to choose whether they wish to save space not installing development files (default) or whether they want the convenience of having development files for everything (similarly to e.g. Arch Linux).

This list is not exhaustive, but includes most major things.

The Chimera handbook

The documentation for the project has undergone significant expansion, now containing detailed installation instructions including how to deal with things like disk encryption and root oN ZFS, and various configuration tasks.

The FAQ is now a part of the handbook and has been expanded as well.

Preparing for alpha

We still have plans to release an alpha as soon as possible. This will be the point where the distro is ready for early adopters. The following needs finishing:

1. The hardening overhaul fallout. Since we have enabled the UBSan checks as well as CFI by default, this exposes all sorts of bugs in libraries and applications, turning them into crashes. Therefore we are rebuilding and testing things as necessary, trying to iron out most issues to have a stable experience before the alpha launches.
2. Packages will need updating to their latest version at the time of the alpha.
3. Automated build system for packages still needs launching. This is experiencing delays, but we plan to have that up as soon as possible.
4. There will be a world rebuild before the alpha happens, on all 4 architectures that are currently supported in repositories. This is needed in order to accommodate the various cbuild updates that have happened in the meantime.

Since these are still pretty significant tasks, it will take some time to get them done. Therefore, the alpha will not come out before the FOSDEM talk. Right now, the idea is to make it coincide with one of the beta releases of FreeBSD 13.2, to get a chance to rebase the userland. That means mid February to early March most likely.

There will be a new set of ISO images before the alpha comes out, to give people a chance to test and expose various issues. Another set will then be made for the alpha release.

After the alpha

The alpha cycle is planned for 6 months to 1 year. Once it is over and the project is ready to be declared beta quality, another world rebuild will be done.

Summary

I am hoping there will be no more significant delays. Right now, it is very near, with only a small number of tasks remaining to do. Those tasks however cover a lot of ground, so they take time.