dhclient is no longer built by default in the dhcp package, so this
test has been broken since that change was made. To fix, switch to
dhcpcd. dhcpcd insists on writing into /var/run, so we need to ensure
that exists.
Fixes: a2c379d4b6 ("dhcp: make client and relay component optional")
This used to be StandardOutput=syslog, which was removed because
syslog is deprecated, but that caused the test to fail. So bring it
back, but set it to the non-deprecated "journal" value instead (which
is what systemd interprets "syslog" as now anyway).
Fixes: 962e15aebc ("nixos: remove StandardOutput=syslog, StandardError=syslog lines")
Otherwise, since the update to Virtualbox 6.1.22, the test would fail
due to the shared directory not existing.
Fixes: ba0da8a076 ("virtualbox: 6.1.18 -> 6.1.22")
Nested KVM has been enabled by default on Linux on Intel for a long
time now, and since Virtualbox 6.1.0, the test won't run without it
because Virtualbox now only supports running hardware-accelerated VMs.
Additionally, this means we can 64-bit guests by default. The 32-bit
guest additions don't currently build, so this is important to have
the tests work with the default options.
There were a bunch of unnecessary f-strings in there and I also removed
the "# fmt: on/off" comments, because we no longer use Black and thus
won't need those comments anymore.
Signed-off-by: aszlig <aszlig@nix.build>
Since systemd 243, docs were already steering users towards using
`journal`:
eedaf7f322
systemd 246 will go one step further, it shows warnings for these units
during bootup, and will [automatically convert these occurences to
`journal`](f3dc6af20f):
> [ 6.955976] systemd[1]: /nix/store/hwyfgbwg804vmr92fxc1vkmqfq2k9s17-unit-display-manager.service/display-manager.service:27: Standard output type syslog is obsolete, automatically updating to journal. Please update│······················
your unit file, and consider removing the setting altogether.
So there's no point of keeping `syslog` here, and it's probably a better
idea to just not set it, due to:
> This setting defaults to the value set with DefaultStandardOutput= in
> systemd-system.conf(5), which defaults to journal.
This module allows root autoLogin, so we would break that for users, but
they shouldn't be using it anyways. This gives the impression like auto
is some special display manager, when it's just lightdm and special pam
rules to allow root autoLogin. It was created for NixOS's testing
so I believe this is where it belongs.
Always enable the UART because the VirtualBug bug that required running without the UART was fixed in 6.0.10. Stop using an old kernel version because the tests work with the default kernel.
(cherry picked from commit ae93571e8d04cebd69491a789d902d6481e05d3f)
Quite some fixing was needed to get this to work.
Changes in VirtualBox and additions:
- VirtualBox is no longer officially supported on 32-bit hosts so i686-linux is removed from platforms
for VirtualBox and the extension pack. 32-bit additions still work.
- There was a refactoring of kernel module makefiles and two resulting bugs affected us which had to be patched.
These bugs were reported to the bug tracker (see comments near patches).
- The Qt5X11Extras makefile patch broke. Fixed it to apply again, making the libraries logic simpler
and more correct (it just uses a different base path instead of always linking to Qt5X11Extras).
- Added a patch to remove "test1" and "test2" kernel messages due to forgotten debugging code.
- virtualbox-host NixOS module: the VirtualBoxVM executable should be setuid not VirtualBox.
This matches how the official installer sets it up.
- Additions: replaced a for loop for installing kernel modules with just a "make install",
which seems to work without any of the things done in the previous code.
- Additions: The package defined buildCommand which resulted in phases not running, including RUNPATH
stripping in fixupPhase, and installPhase was defined which was not even run. Fixed this by
refactoring using phases. Had to set dontStrip otherwise binaries were broken by stripping.
The libdbus path had to be added later in fixupPhase because it is used via dlopen not directly linked.
- Additions: Added zlib and libc to patchelf, otherwise runtime library errors result from some binaries.
For some reason the missing libc only manifested itself for mount.vboxsf when included in the initrd.
Changes in nixos/tests/virtualbox:
- Update the simple-gui test to send the right keys to start the VM. With VirtualBox 5
it was enough to just send "return", but with 6 the Tools thing may be selected by
default. Send "home" to reliably select Tools, "down" to move to the VM and "return"
to start it.
- Disable the VirtualBox UART by default because it causes a crash due to a regression
in VirtualBox (specific to software virtualization and serial port usage). It can
still be enabled using an option but there is an assert that KVM nested virtualization
is enabled, which works around the problem (see below).
- Add an option to enable nested KVM virtualization, allowing VirtualBox to use hardware
virtualization. This works around the UART problem and also allows using 64-bit
guests, but requires a kernel module parameter.
- Add an option to run 64-bit guests. Tested that the tests pass with that. As mentioned
this requires KVM nested virtualization.
He prefers to contribute to his own nixpkgs fork triton.
Since he is still marked as maintainer in many packages
this leaves the wrong impression he still maintains those.
I've started digging into the actual cause of the problem a week ago but
didn't continue fixing this.
The reason why the tests are failing is because
torvalds/linux/commit/72f5e08dbba2d01aa90b592cf76c378ea233b00b has
remapped the location of the TSS into the CPU entry area and we did
update our default kernel to version 4.14 in NixOS/nixpkgs@88530e02b6.
Back to VirtualBox: The guru meditation happens in
selmRCGuestTssPostWriteCheck, which I think is only a followup error. I
believe the right location couldn't be determined by VirtualBox and thus
the write check function triggers that panic because it's reading from
the wrong location.
So the actual problem *only* surfaces whenever we use software
virtualization, which we do for our tests because we don't have nested
virtualization available.
Our tests are also for testing the functionality of VirtualBox itself
and not certain kernel versions or kernel features, so for the time
being and until this is fixed, let's actually use kernel version 4.9 for
the guests within the VM tests. Kernel 4.9 didn't have the mentioned
change of the TSS location and thus the tests succeed.
Signed-off-by: aszlig <aszlig@nix.build>
Cc: @dtzWill
-s, --script: never prompts for user intervention
Sometimes the NixOS installer tests fail when they invoke parted, e.g.
https://hydra.nixos.org/build/62513826/nixlog/1. But instead of exiting
right there, the tests hang until the Nix builder times out (and kills
the build). With this change the tests would instead fail immediately,
which is preferred.
While at it, use "parted --script" treewide, so nobody gets build
timeout due to parted error (or misuse). (Only nixos/ use it, and only
non-interactive.)
A few instances already use the short option "-s", convert them to long
option "--short".
This has surfaced since f803270b7e.
The commit bumped bash to version 4.4, which caused to change the order
of --subst-var flags in substituteAll, which this test was relying on,
because it added a @shell@ to boot.initrd.postMountCommands.
Our substituter is currently working a bit like this:
original.replace('@var1@', 'val1').replace('@var2@', 'val2')...
Unfortunately, this means that if @var2@ occurs within @var1@ it is
replaced by the new value, so the order of the substvars actually
matter. I highly doubt that we want a behaviour like this and I'm
wondering why it didn't occur to me as a problem while writing the
initial implementation of the VirtualBox tests.
Whether to get rid of this and disallowing substitution of substvars
within substvars is another topic which I think needs discussion in a
different place.
As for now, I'm using stdenv.shell, because the closure size of this
should fit within the initrd, so it's fine especially because it's just
a test.
Tested with the net-hostonlyif and systemd-detect-virt tests and they
both succeed with this change.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Reported-by: @globin on IRC
The motivation for this change is the following: As gnu-netcat,
e. g. does not support ipv6, it is not suitable as default netcat.
This commit also fixes all obvious build issues caused by this change.
The test complains[1][2] that
Failed to start message bus: Failed to bind socket "/run/dbus/system_bus_socket": No such file or directory
In 639e5401ff, the dbus socket dir is set
to `/run/dbus`; in the test vm `/var/run/dbus` is used, but the standard
`/run -> /var/run` link is typically not created until stage 2 init, not
in the minimal init used here. Thus, dbus fails to run within the test
environment . Fix by changing `/var/run/dbus` to simply `/run/dbus`.
[1]: https://hydra.nixos.org/build/42534725
[2]: https://hydra.nixos.org/build/42523834
One reason why it took me so long for debugging the test failure with
systemd-detect-virt was that simple-cli has succeeded while the former
has not.
This now makes sure we have consistency accross all the subtests and if
problems like the one in the previos commit ever show up again, we will
have just the headless test succeeding and it's more obvious where the
actual problem resides.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
We don't have (simulated) sound hardware within the qemu VM, neither do
we have it available within VirtualBox that's running within the qemu
VMs.
With sound hardware the VirtualBox UI displays an error dialog, which in
turn causes the VM process to hang on unregister. This in turn has
caused the tests to fail because of the following error:
Cannot unregister the machine '...' while it is locked
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Using waitUntilSucceeds for testing whether the shutdown signalling
files have vanished is quite noisy because it prints two lines for every
try. This is now fixed with a while loop on the guest VM which does the
same check but with only one output for the command that's executed and
another one when the conditions are met.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
VBoxService needs dbus in order to work properly, which failed to start
up so far, because it was searching in /run/current-system/sw for its
configuration files.
We now no longer run with the --system flag but specify the
configuration file directly instead.
This fixes at least the "simple-gui" test and probably the others as
well, which I haven't tested yet.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
We can't use waitForWindow here because it runs xwininfo as user root,
who in turn is not authorized to connect to the X server running as
alice.
So instead, we use xprop from user alice which should fix waiting for
the VirtualBox manager window.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
The VirtualBox tests so far ran the X server as root instead of user
"alice" and it did work, because we had access control turned off by
default.
Fortunately, it was changed in 1541fa351b.
As a side effect, it caused all the VirtualBox tests to fail because
they now can't connect to the X server, which is a good thing because
it's a bug of the VirtualBox tests.
So to fix it, let's just start the X server as user alice.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Now subtests are separate derivations, because the individual tests do
not depend on state from previous test runs.
This has the advantage that it's easier to run individiual tests and
it's also easier to pinpoint individual tests that randomly fail.
I ran all of these tests locally and they still succeed.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
I'm not quite sure why the official Hydra gets a kernel panic in one of
two VMs using the exact same kernels:
https://hydra.nixos.org/build/26339384
Because the kernel panic happens before stage 1, let's wait for the
first VM to boot up and after the bootup is done, start the second one
in hope that it won't trigger the panic.
Oddly enough, whenever I run the test on my own Hydra and on my local
machines, I don't get anything like that.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
I forgot to do this in da0e642. It shouldn't be a big problem but it's
more clean to destroy the VM once we're done testing.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
We previously had 1024 MB of memory to fit a VirtualBox VM with 512 MB
plus the memory needed of the VirtualBox host VM. That obviously won't
work for two VirtualBox VMs, which are used for testing networking
between two VirtualBox guests.
Now, we have 2048 MB on the qemu guest (the VirtualBox host) and 768 MB
for each VirtualBox guest. That should be enough to fit in two
VirtualBox guests (I hope).
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Unfortunately, we can't test whether USB is really working, but we can
make sure that VirtualBox has access to the USB devices.
This is essentially testing #9736, which I haven't yet been able to
reproduce though, but it makes sense to test it so it won't happen in
future releases.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Addresses #9876 in the way that we want to make sure that VirtualBox 5.x
is going to be properly detected. Right now the result is "kvm", so the
subtest fails as expected with:
error: systemd-detect-virt returned "kvm" instead of "oracle" at (eval
14) line 414, <__ANONIO__> line 92.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Makes it easier to debug and find out for which machine a certain log
socket has been started or stopped.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
