Follow-up for ed024abac6
and 9fbe26cfa8
Also, let's not get too tangled up in the style of defining variables
in between. The functions are short enough, and vars involved are still
effectively at the beginning... Put differently, the separation from
'int r' is too deliberate and brings no actual value in my eyes.
If we have multiple trusted fs (i.e. luks or dm-verity) we generate via
repart at boot, we must make sure they cannot be "misappropriated", i.e.
used for a different mount they were intended for.
Hence, let's introduce "mount constraint" data (encoded in xattrs on the
root inode of the fs) that tells us where a file system has to be
mounted, and what the gpt partition metadata has to be for the fs to be
valid.
Inspired by this thread:
https://lists.freedesktop.org/archives/systemd-devel/2025-March/051244.html
- SO_PEERGROUPS is since kernel v4.13
(28b5ba2aa0f55d80adb2624564ed2b170c19519e),
- SO_BINDTOIFINDEX is since kernel v5.1
(f5dd3d0c9638a9d9a02b5964c4ad636f06cf7e2c).
If the target dir is tmpfs and we run on old kernels we cannot extract
xattrs and the extracting will fail if there are any. hence add
-no-xattrs to the two remaining unsquashfs invocations that don't have
it.
(Also all other invocations across our test tree spell "-dest" instead
of "-d", hence do so here too.)
Let's automatically generate validatefs xattrs by default, that encode
the intended use of partitions.
This defaults to on, since the structure of repart definition files
tells us enough on use for this to be safe. There's an option however,
to turn this off.
Let's turn on validatefs automatically for all auto-discovered
partitions.
Let's add an x-systemd.validatefs option to optionally turn this on for
fstab listed file systems.
This new tool looks for a three xattr on the root inode of a file system
that encode mount constraints of the file system. The tool is supposed
to be hooke into the mount logic and is supposed to protect against
misappropriating trusted file systems in unintended ways.
Consider the following scenario: we boot up on first boot and create a
tpm-locked pair of /var/ and /srv/ partitions via systemd-repart. An
attacker then offline modifies the partition table, exchanging the
metadata of the /var/ and /srv/ partition. So far we'd happily accept
that, honour the modified metadata and boot up. This could be used to
revert changes to /var/ or similar. And all that even though both
partitions are encrypted and locked to TPM!
With this new mechanism we can encode in the protected contents of the
file systems the ways it can be used: the partition type uuid, the
partition label and the intended mount point can be stored in xattrs,
and we can check them automatically on mount, and take action on
mismatch. (action would typically be immediate reboot).
PCR extensions are supposed to be useful for "destroying" the ability to
access TPM bound secrets. Hence, if for some reason we fail to extend a
PCR, it's safer to just reboot, instead of going on without the
extension, leaving secrets potentially accessible which should not be
accessible.
Note that the services exit gracefully if no TPM is found, hence this
should not be triggered on TPM-less systems. However, this enforces that
if there is a TPM that is accessible to Linux and that works properly,
the PCR measurement must complete too.
Inspired by this thread:
https://lists.freedesktop.org/archives/systemd-devel/2025-March/051244.html
Let's always prefer quotactl_fd() when it's available and use quotactl()
only as as a fallback on old kernels.
This way we can operate on the fds we typically already have open, or if
needed we can open a new one, and use for multiple fs operation.
In the long run we should really focus on operating exclusively by fd
instead of by path, by device nor or otherwise. This gets us a step
closer to that.
Instead of listing dependencies manually for the default tools tree,
let's reuse the prepare scripts from the build image. To make this work,
the sync script has to be configured for the tools tree as well so that
it's invoked both when building the tools tree and for the regular
image,
otherwise, when doing the first build in a fresh checkout, the sync
script
won't have executed yet as sync scripts for the regular images are
executed
after building the default tools tree.
With meson 1.7.0, meson won't rebuild the world anymore when running
meson test so since meson 1.7.0 is in tumbleweed, debian testing, fedora
rawhide and Arch Linux, drop --no-rebuild from the docs since all the default
tools trees built by mkosi in the systemd repo will have meson 1.7.0 or newer
installed.
We keep --no-rebuild in CI because in CI we run with the same tools tree
distribution as the target distribution we're building and in those it's not
guaranteed for meson 1.7.0 to be available yet.
The advantage of dropping --no-rebuild is that meson will now reconfigure itself
if needed, which is required if we change or move around meson files, as otherwise
meson tends to explode spectacularly in hard to debug ways.
https://mesonbuild.com/Release-notes-for-1-7-0.html#test-targets-no-longer-built-by-default
Instead of listing dependencies manually for the default tools tree,
let's reuse the prepare scripts from the build image. To make this work,
the sync script has to be configured for the tools tree as well so that
it's invoked both when building the tools tree and for the regular image,
otherwise, when doing the first build in a fresh checkout, the sync script
won't have executed yet as sync scripts for the regular images are executed
after building the default tools tree.
Instead of not creating the test at all, let's always create test but
simply mark them as skipped as this is more observable than simply not
creating the test at all.
This test only works if the image was built as root. Since that's
impossible to check as meson generally runs before we build the image,
let's use whether meson is run as root as a proxy.
If we don't need an initrd, let's not waste time in one. We have to
gate this by distributions that have the necessary kernel modules to
make this work as builtin modules, hopefully we can expand the list in
the future.
This is a required first step before testing and it's not specifically
mentioned in the doc. This tripped me up for a while, so let's save the
trouble for the next person.
Mike Yuan
Luca Boccassi
Yu Watanabe
Lennart Poettering
Busayo Dada
Luca Boccassi
Daan De Meyer
Allison Karlitskaya