Given that we were overriding qemu_test to enable this anyway,
enabling this by default saves Hydra a QEMU build.
There's also clear demand from users[1] for this feature, so our
alternatives are:
- Offer a qemu-canokey attribute. I don't want to do this, because I
don't think there's any reason to make Hydra build an extra QEMU.
- Enable it only for qemu_test. I don't want to do this, because it
will lead to users using qemu_test without understanding its
subtleties.
- Force users to build from source. I don't think there's any reason
to do this when it's unlikely to hurt anybody having it enabled by
default. There's no reason to single out canokey to be disabled by
default in spite of users' needs given that we enable so many other
optional QEMU features.
[1]: https://github.com/canokeys/canokey-qemu/issues/6
This patch is about removing `wireguardPeerConfig`,
`dhcpServerStaticLeaseConfig` - a.k.a. the
AbstractSingletonProxyFactoryBean of nixpkgs - and friends.
As a former colleague said
> worst abstraction ever
I second that. I've written enough networkd config for NixOS systems so
far to have a strong dislike. In fact, these don't even make sense:
`netdevs.wireguardPeers._.wireguardPeerConfig` will be rendered into
the key `[WireGuardPeer]` and every key from `wireguardPeerConfig` is in
there. Since it's INI, there's no place where sections on the same level
as wireguardPeerConfig fit into. Hence, get rid of it all.
For the transition, using the old way is still allowed, but gives a
warning. I think we could drop this after one release.
The tests of rosenpass and systemd-networkd-dhcpserver-static-leases
were broken on the rev before, hence they were updated, but are still
not building.
Module has been fixed and now uses the maintenance service to cache
settings so as to not require environment files wherever possible.
The tests now test using mariadb and postgresql as well as sqlite to be
more complete. A test has been added for testing whether app.js has been
compiled successfully, as well as to check whether the cronjob fires
successfully.
This migrates the syntax for a few configuration values,
which now need to be quoted strings for user values.
This also disables the use of a public resolver,
which is not accessible in the sandbox.
This fixes build failures with the wayland dependency:
Quoting @nevivurn:
"guestadditions->...->wayland, but the test config pulls in
modules/profiles/minimal.nix which disables xlibs and thus it won't build"
Co-authored-by: Yongun Seong <nevivurn@nevi.dev>
Upstream (accidentally) broke support for postgresql 12 and 13 on
v0.2.1 by changing the signature of the `from_datum` function[^1].
This went unnoticed since the release branch `0.2` did not have CI.
Furthermore, they are removing support for these versions of postgresql
on v0.3.0[^2].
[^1]: 97e861d51d
[^2]: https://github.com/tensorchord/pgvecto.rs/issues/343
Our more thorough parametrised tests uncovered that with the changes for
supporting DynamicUser, we now have the situation that for static users
the root directory within the confined environment is now writable for
the user in question.
This is obviously not what we want and I'd consider that a regression.
However while discussing this with @ju1m and my suggestion being to
set TemporaryFileSystem to "/" (as we had previously), they had an even
better idea[1]:
> The goal is to deny write access to / to non-root users,
>
> * TemporaryFileSystem=/ gives us that through the ownership of / by
> root (instead of the service's user inherited from
> RuntimeDirectory=).
> * ProtectSystem=strict gives us that by mounting / read-only (while
> keeping its ownership to the service's user).
>
> To avoid the incompatibilities of TemporaryFileSystem=/ mentioned
> above, I suggest to mount / read-only in all cases with
> ReadOnlyPaths = [ "+/" ]:
>
> ...
>
> I guess this would require at least two changes to the current tests:
>
> 1. to no longer expect root to be able to write to some paths (like
> /bin) (at least not without first remounting / in read-write
> mode).
> 2. to no longer expect non-root users to fail to write to certain
> paths with a "permission denied" error code, but with a
> "read-only file system" error code.
I like the solution with ReadOnlyPaths even more because it further
reduces the attack surface if the user is root. In chroot-only mode this
is especially useful, since if there are no other bind-mounted paths
involved in the unit configuration, the whole file system within the
confined environment is read-only.
[1]: https://github.com/NixOS/nixpkgs/pull/289593#discussion_r1586794215
Signed-off-by: aszlig <aszlig@nix.build>
This is to make sure that we test all of the DynamicUser/User/Group and
PrivateTmp options in a uniform way. The reason why we need to do this
is because we recently introduced support for the DynamicUser option and
since there are some corner cases where we might end up with more
elevated privileges (eg. writable directories in some cases), we want to
make sure that the environment is as restrictive as with a static
User/Group assignment.
I also removed various checks that try to os.chown(), since with our new
recursive checker those are redundant.
Signed-off-by: aszlig <aszlig@nix.build>
So far the architecture for the tests was that we would use a systemd
socket unit using the Accept option to start a small shell process where
we can pipe commands into by connecting to the socket created by the
socket unit.
This is unnecessary since we can directly use the code snippets from the
individual subtests and systemd will take care of checking the return
code in case we get any assertions[^1].
Another advantage of this is that tests now run in parallel, so we can
do rather expensive things such as looking in /nix to see whether
anything is writable.
The new assert_permissions() function is the main driver behind this and
allows for a more fine-grained way to check whether we got the right
permissions whilst also ignoring irrelevant things such as read-only
empty directories.
Our previous approach also just did a read-only check, which might be
fine in full-apivfs mode where the attack surface already is large, but
in chroot-only mode we really want to make sure nothing is every
writable.
A downside of the new approach is that currently the unit names are
numbered via lib.imap1, which makes it annoying to track its definition.
[^1]: Speaking of assertions, I wrapped the code to be run with pytest's
assertion rewriting, so that we get more useful AssertionErrors.
Signed-off-by: aszlig <aszlig@nix.build>
When experimenting on ways how to refactor the test, I wrote a
significant enough amount of Python to warrant a dedicated Python file.
This commit is mainly to prepare for that and make it easier to track
renames.
Signed-off-by: aszlig <aszlig@nix.build>
The reason why I originally used the "description" attribute was that it
can be easily used to parametrise the tests so that we can specify
common constraints and apply it across a number of different
configurations.
When porting the tests to Python, the description attribute was replaced
by inlining it into the Python code, most probably because it was easier
to do in bulk since using Nix to generate the subtest parts would be
very complicated to do since we also had to please Black (a Python code
formatter that we no longer use in test scripts).
Since we now also want to support DynamicUser in systemd-confinement,
the need to parametrise the tests became apparent again because it's now
easier to refactor our subtests to run both with *and* without
DynamicUser set to true.
Signed-off-by: aszlig <aszlig@nix.build>
Eelco has made several early contributions to NixOS including writing
the samba module among other things, but is more or less inactive these
days.
By my brief inspection, he has not committed to the nixos/ tree since
releasing Nix 2.13 in early 2023 and merging a PR to networking tests
slightly before that. A lot of these tests/modules are actually
unmaintained in practice, so we should update the code to reflect the
practical reality so someone can consider picking them up.
Depending on the startup order of the two machines it might take a few
moments to get both zones transfered, which can lead to SERVFAIL
responses on busy machines.
