Daan De Meyer [Fri, 13 Dec 2024 09:20:40 +0000 (09:20 +0000)]
tests: Build with --incremental=strict
Let's require cached images to be present before running the
integration tests. This makes sure the tests only need to build the
output that they're testing and it also opens up the road for running
tests in parallel in the future.
Daan De Meyer [Fri, 13 Dec 2024 10:23:39 +0000 (10:23 +0000)]
Remove package manager from cache manifest
We don't always have the tools tree available yet when generating the
cache manifest which can lead to cache manifest mismatches depending on
whether the tools tree is available or not. Since the package manager field
depends on whether the tools tree is available or not (dnf or dnf5, dnf or zypper),
let's remove it from the cache manifest.
Matteo Croce [Thu, 12 Dec 2024 16:47:44 +0000 (17:47 +0100)]
Use host model cpu
mkosi currently calls qemu with `-cpu max`. This enables all features
supported by the accelerator, which could be less than the ones
supported by the host.
Switch to the default `-cpu host` and keep max only when using binary
translation.
Daan De Meyer [Thu, 12 Dec 2024 10:08:36 +0000 (10:08 +0000)]
Rework crypto-policies handling slightly
Let's bind mount from the tools tree if it provides crypto policies
and only write our own for rpm-sequoia if the tools tree doesn't provide
its own crypto policies.
Brendan Jackman [Sun, 8 Dec 2024 12:03:57 +0000 (13:03 +0100)]
sandbox: Spit out some info when initial unshare gets EPERM
To try and minimise the pain of this issue
(https://github.com/systemd/mkosi/issues/3265), dump some info that might help
users resolve it.
I had a quick look around expecting to find a document from Red Hat discussing
this topic much like the Ubuntu one I've linked here, but I didn't find it.
Hopefully if it exists someone else can add it later.
I'm doing this via a direct write to stderr because of the comment at the top of
sandbox.py saying to avoid imports. If this is highly undesirable it looks like
log.log_notice would be the right choice here (then you don't need the
annoying ANSI codes).
Luca Boccassi [Sun, 8 Dec 2024 13:55:25 +0000 (13:55 +0000)]
mkosi-initrd: also take input from /etc/kernel/cmdline
Other kernel-install plugins (for Type 1) already take /etc/kernel/cmdline
into account when generating the local cmdline entry, so do the same
in mkosi-initrd for the UKI
Akihiko Odaki [Thu, 5 Dec 2024 04:51:21 +0000 (13:51 +0900)]
Define virt as default QEMU machine for riscv64
The documentation says:
> It is the recommended board type if you simply want to run a guest
> such as Linux and do not care about reproducing the idiosyncrasies and
> limitations of a particular bit of real-world hardware.
https://qemu.readthedocs.io/en/v9.1.0/system/riscv/virt.html
Daan De Meyer [Thu, 28 Nov 2024 21:43:03 +0000 (22:43 +0100)]
Fix Partition split_path calculation
Now that we use workdir() in make_image(), we have to change the
Partition initializer to remove the path added by workdir() again
to get the proper path on the host.
Daan De Meyer [Thu, 28 Nov 2024 15:12:03 +0000 (16:12 +0100)]
Don't resolve deps if we're reusing previous config
We ignore subimages if we're reusing the previous config so let's
make sure we ignore subimages as well after running configure scripts
if we're reusing previous config.
Daan De Meyer [Sun, 24 Nov 2024 20:34:57 +0000 (21:34 +0100)]
mkosi-initrd: Install btrfs-progs
Even if fsck.btrfs doesn't do anything particularly useful, given we
install the fs tools of all the other filesystems, let's install btrfs-progs
as well. This is also useful when using systemd-repart to create the
root filesystem on first boot.
Daan De Meyer [Sat, 23 Nov 2024 13:54:59 +0000 (14:54 +0100)]
Drop ForwardJournal= integration for coredumpctl and journalctl verbs
Now that we have the sandbox verb, this is trivial to achieve with the
sandbox verb so let's keep the coredumpctl and journalctl verbs focused
solely on reading the journal and coredumps from the image itself and leave
inspecting the forwarded journal to mkosi sandbox.
Technically this breaks compat but I'm convinced nobody's actually using this
specific bit of integration so it shouldn't matter if we remove it.
Daan De Meyer [Sat, 23 Nov 2024 13:36:19 +0000 (14:36 +0100)]
Use tools tree again for ssh, coredumpctl and journalctl
In effect these verbs are very similar to the sandbox verb except
they run a predefined command line instead of an arbitrary one. So
let's make them behave the same as the sandbox verb when it comes to
the tools tree.
Brendan Jackman [Fri, 22 Nov 2024 14:14:58 +0000 (14:14 +0000)]
Clarify shell vs boot in man page
In retrospect it seems like this should have been obvious but it
confused me. Thanks to Daan for pointing it out on Matrix. Reword to
help out people like me in the future.
Daan De Meyer [Fri, 22 Nov 2024 09:37:06 +0000 (10:37 +0100)]
Include machine name in virtiofsd unit name
If the same directory is shared across multiple guests, we end up
with duplicate unit names even though we need separate virtiofsd
instances. Let's fix this by including the machine name in the
virtiofsd unit name so that the unit names will be unique across
guests.
Daan De Meyer [Thu, 21 Nov 2024 20:51:43 +0000 (21:51 +0100)]
sysupdate: Make it work in combination with a tools tree
- We have to make sure systemd-sysupdate looks at the os-release
from the host even when using a tools tree.
- systemd-sysupdate can't detect the root block device when running
with a tools tree. Let's abuse /run/systemd/volatile-root to shortcut
the detection logic instead.
Septatrix [Wed, 20 Nov 2024 22:10:45 +0000 (23:10 +0100)]
Keep track of collection resets from CLI arguments (fixes #3208)
Previously it was only possible to completely reset but not append,
or only append but not reset to collection/list settings
via command line arguments.
Now we track if a setting has ever been set to None (i.e. reset)
during command line parsing.
This information is used during value finalization
to decide whether to merge both collections or only keep the CLI value.
Martin Hundebøll [Tue, 19 Nov 2024 11:11:25 +0000 (12:11 +0100)]
Propagate --tools-tree-certificates to initrd build
The initrd build uses a config created from a "hand-crafted" command
line passed to `parse_config()`. This command line includes the relevant
settings (i.e. those deemed relevant for build the initrd) from the
"proper" config object.
Since the --tools-tree-certificates setting wasn't added to that command
line, it defaults to true regardless of any setting configured when
invoking mkosi. When building behind a cooperate (transparent) proxy
using a self-signed certificate, the initrd build fails when trying to
download packages (unless the needed packages happened to be downloaded
as part of a previous image build).
Make sure the --tools-tree-certificates setting applies to the initrd
build also, by explicitly including it in the hand-crafted command line.
The setting can be added unconditionally because of the default value
specified in config.py.
Daan De Meyer [Mon, 18 Nov 2024 20:06:17 +0000 (21:06 +0100)]
Also fail early if default tools tree is out of date without --force
We fail early if the tools tree does not exist and build or --force
was not specified, let's do the same if the tools tree is incremental
and the cache is out-of-date.
Daan De Meyer [Sat, 16 Nov 2024 14:47:43 +0000 (15:47 +0100)]
Add ToolsTreePackageDirectories=
Sometimes, we want to add locally built rpm packages to the default
tools tree. For example, systemd-repart depends on mkfs binaries that
might not be available on the host system, so the only way to run it
is from within the tools tree, which means we need a way to install
systemd-repart built from source within the tools tree.
Daan De Meyer [Fri, 15 Nov 2024 15:24:18 +0000 (16:24 +0100)]
Use passwd symlinks instead of bind mounts
Bind mounts don't reflect changes to the original files if they're
replaced instead of modified. Let's use symlinks instead so that
changes to the original files are always reflected.
Daan De Meyer [Fri, 15 Nov 2024 11:51:19 +0000 (12:51 +0100)]
Add --debug-sandbox
This will help in debugging sandbox related issues. We run the sandbox
with strace and detach on execve() so we don't strace the command that
we're running.