Home-assistant through its `--runner` commandline flag supports sending
exit code 100 when the `homeassistant.restart` service is called.
With `RestartForceExitStatus` we can listen for that specific exit code
and restart the whole systemd unit, providing an actual clean restart
with fresh processes. Additional treat exit code 100 as a successful
termination.
Configures the emulated_hue component and expects CAP_NET_BIND_SERVICE
to be passed in order to be able to bind to 80/tcp.
Also print the systemd security analysis, so we can spot changes more
quickly.
This is what is still exposed, and it should still allow things to work
as usual.
✗ PrivateNetwork= Service has access to the host's … 0.5
✗ RestrictAddressFamilies=~AF_(INET… Service may allocate Internet soc… 0.3
✗ DeviceAllow= Service has a device ACL with som… 0.1
✗ IPAddressDeny= Service does not define an IP add… 0.2
✗ PrivateDevices= Service potentially has access to… 0.2
✗ PrivateUsers= Service has access to other users 0.2
✗ SystemCallFilter=~@resources System call allow list defined fo… 0.2
✗ RootDirectory=/RootImage= Service runs within the host's ro… 0.1
✗ SupplementaryGroups= Service runs with supplementary g… 0.1
✗ RestrictAddressFamilies=~AF_UNIX Service may allocate local sockets 0.1
→ Overall exposure level for home-assistant.service: 1.6 OK :-)
This can grow to as much as ~1.9 if you use one of the bluetooth or nmap
trackers or the emulated_hue component, all of which required elevated
permisssions.
git-daemon won't start up if its project directory (here /git) doesn't
exist. If we try to create it using the test harness, then we're racing
whether we manage to connect to the backdoor vs. the startup speed of
git-daemon.
Instead, use systemd-tmpfiles, which is guaranteed(?) to run before
network.target and thus before git-daemon.service starts.
rspamd seems to be consuming more memory now sometimes, causing OOMs in
the test.
Increase the memory given to these VMs to make the tests pass more
reliably.
Postfix has started outputting an error on startup that it can't parse
the compatibility level 9999.
Instead, just set the compatibility level to be identical to the current
version, which seems to be the (new) intent for the compatibility level.
This test was failing because Firefox was displaying a download prompt
rather than the page content, presumably because mumble mumble
content-type sniffing.
By explicitly setting a content-type, the test now passes.
Looks like GRUB has issues loading EFI binaries from (cd0), which is
what would be used in e.g. qemu with OVMF with `-cdrom`. Apparently also
what is used with AArch64 + U-Boot USB.
The serial output (but it's named console, not serial actually) causes
issues on U-Boot's EFI, at the very least.
This is inspired by OpenSUSE's approach:
* https://build.opensuse.org/package/view_file/Base:System/grub2/grub2-SUSE-Add-the-t-hotkey.patch
Where they add a hidden menu entry, which can be used to force the
console output.
The `echo` will be visible on the serial terminal (grub "console"),
while the graphical interface is shown. Note that input in the serial
terminal (grub "console") will continue controlling the graphical
interface. Useful if you have an SBC connectedinto an HDMI monitor, but
no keyboard connected to it.
- In order to use GIO/GVFS it is enough to enable the gvfs service.
- The module option services.gvfs.package can be used to choose a
variation of the gvfs package, if desired.
See [0]: "QEMU_OPTS is something that should be set by people running VM
tests interactively, to do port forwardings etc.
We really should not poke with it from the test script - that's what
virtualisation.qemu.options is for."
[0]: https://github.com/NixOS/nixpkgs/pull/119615#discussion_r624145020
Co-authored-by: Florian Klink <flokli@flokli.de>
Fixes these two deprecation warnings, by moving away from these options
towards a simple listener configuration.
> The 'bind_address' option is now deprecated and will be removed in a future version. The behaviour will default to true.
> The 'port' option is now deprecated and will be removed in a future version. Please use 'listener' instead.
Fixes: #120860
It can still network, it can only access the ssl related files if ssl is
enabled.
✗ PrivateNetwork= Service has access to the host's network 0.5
✗ RestrictAddressFamilies=~AF_(INET|INET6) Service may allocate Internet sockets 0.3
✗ DeviceAllow= Service has a device ACL with some special devices 0.1
✗ IPAddressDeny= Service does not define an IP address allow list 0.2
✗ RootDirectory=/RootImage= Service runs within the host's root directory 0.1
✗ RestrictAddressFamilies=~AF_UNIX Service may allocate local sockets 0.1
→ Overall exposure level for mosquitto.service: 1.1 OK 🙂
This allows us to continue to have the automatically sized image attempt
to build on Hydra, which should give us a good indication of when we've
got this fixed.
For reasons we haven't been able to work out, the aarch64 EC2 image now
regularly exceeds the output image size on hydra.nixos.org. As a
workaround, set this back to being statically sized again.
The other images do seem to build - it's just a case of the EC2 image
now being too large (occasionally non-determinstically).
This is supeer useful to allow the normal sd-image code to be used by
someone who wants to setup multiple partitions with a sd-image.
Currently I'm manually copying the sd-image file and modifying it
instead.
This ensures that newly created secrets will have the permissions
`0640`. With this change it's ensured that no sensitive information will
be word-readable at any time.
Related to #121293.
Strictly speaking this is a breaking change since each new directory
(including data-files) aren't world-readable anymore, but actually these
shouldn't be, unless there's a good reason for it.
This is what is still exposed, and it allows me to control my lamps from
within home-assistant.
✗ PrivateNetwork= Service has access to the host's network 0.5
✗ RestrictAddressFamilies=~AF_(INET|INET6) Service may allocate Internet sockets 0.3
✗ DeviceAllow= Service has a device ACL with some special devices 0.1
✗ IPAddressDeny= Service does not define an IP address allow list 0.2
✗ PrivateDevices= Service potentially has access to hardware devices 0.2
✗ RootDirectory=/RootImage= Service runs within the host's root directory 0.1
✗ SupplementaryGroups= Service runs with supplementary groups 0.1
✗ MemoryDenyWriteExecute= Service may create writable executable memory mappings 0.1
→ Overall exposure level for zigbee2mqtt.service: 1.3 OK 🙂
Until now, the `touch + chmod 600 + write` approach made it possible for
an unprivileged local user read the private key file, by opening
the file after the touch, before the read permissions are restricted.
This was only the case if `generatePrivateKeyFile = true` and the parent
directory of `privateKeyFile` already existed and was readable.
This commit fixes it by using `umask`, which ensures kernel-side that
the `touch` creates the file with the correct permissions atomically.
This commit also:
* Removes `mkdir --mode 0644 -p "${dirOf values.privateKeyFile}"`
because setting permissions `drw-r--r--` ("nobody can enter that dir")
is awkward. `drwx------` would perhaps make sense, like for `.ssh`.
However, setting the permissions on the private key file is enough,
and likely better, because `privateKeyFile` is about that file
specifically and no docs suggest that there's something special
about its parent dir.
* Removes the `chmod 0400 "${values.privateKeyFile}"`
because there isn't really a point in removing write access from
the owner of the private key.
- Set an explicit umask that allows u+rwx and g+r.
- Adds `ProtectControlGroups` and `ProtectKernelLogs`, there should be
no need to access either.
- Adds `ProtectClock` to prevent write-access to the system clock.
- `ProtectProc` hides processes from other users within the /proc
filesystem and `ProcSubSet` hides all files/directories unrelated to
the process management of the units process.
- Sets `RemoveIPC`, as there is no SysV or POSIX IPC within nginx that I
know of.
- Restricts the creation of arbitrary namespaces
- Adds a reasonable `SystemCallFilter` preventing calls to @privileged,
@obsolete and others.
And finally applies some sorting based on the order these options appear
in systemd.exec(5).
See #119615 for more details. The aarch64-linux test failed with
"qemu-system-aarch64: Virtio VGA not available" so I've restricted the
test to x86_64-linux (the virtio paravirtualized 3D graphics driver is
likely only available on very few platforms).
The result still looks far from ideal but at least it gets recognized
now. "-fa Monospace" is required to switch to a font from the FreeType
library so that "-fs 24" works.
Note: Using linuxPackages_latest is not required anymore.
On reboots and shutdowns promtail blocks for at least 90 seconds,
because it would still try to deliver log messages for loki, which isn't
possible when the network has already gone down.
Upstreams example unit also uses a ten seconds timeout, something which
has worked pretty well for me as well.
systemd-nspawn can react to SIGTERM and send a shutdown signal to the container
init process. use that instead of going through dbus and machined to request
nspawn sending the signal, since during host shutdown machined or dbus may have
gone away by the point a container unit is stopped.
to solve the issue that a container that is still starting cannot be stopped
cleanly we must also handle this signal in containerInit/stage-2.
This reverts commit d6e0d38b84.
We need shorter secrets to continue working, since the earlier
recommendation was too short and there's no way to rotate the them.
The upstream recommended minimum length for db_key_base is 30 bytes,
which our option descriptions repeated. Recently, however, upstream
has, in many places, moved to using aes-256-gcm, which requires a key
of exactly 32 bytes. To allow for shorter keys, the upstream code pads
the key in some places. However, in many others, it just truncates the
key if it's too long, leaving it too short if it was to begin
with. This adds a patch that fixes this and updates the descriptions
to recommend a key of at least 32 characters.
See https://gitlab.com/gitlab-org/gitlab/-/merge_requests/53602
This version contains a vulnerability[1], and isn't maintained. The
original reason to have two jellyfin versions was to allow end-users to
backup the database before the layout was upgraded, but these backups
should be done periodically.
[1]: <https://nvd.nist.gov/vuln/detail/CVE-2021-21402>
Work around missing /dev files inside runInLinuxVM by creating a
symlink before calling nixos-enter.
This fixes https://github.com/NixOS/nixpkgs/issues/93381.
I ran into this issue when trying to create a VMware image that boots from EFI.
Thanks @colemickens for reporting this and @danielfullmer for fixing the same thing in in qemu-vm.nix (37676e77cb) and explaining what the issue was.
This ensures the following gptfdisk warning won't happen:
```
Warning: File size is not a multiple of 512 bytes! Misbehavior is likely!
```
Additionally, helps towards aligning the partition to be more optimal
for the underlying storage.
It is actually impossible to align for the actual underlying storage
optimally because we don't know what the block device will be!
But aligning on 1MiB should help.
An empty list results in no CapabilityBoundingSet at all, an empty
string however will set `CapabilityBoundingSet=`, which represents a
closed set.
Related: #120617
An empty list results in no CapabilityBoundingSet at all, an empty
string however will set `CapabilityBoundingSet=`, which represents a
closed set.
Related: #120617
The last bits to prevent babeld from running unprivileged was its
kernel_setup_interface routine, that wants to set per interface
rp_filter. This behaviour has been disabled in a patch that has been
submitted upstream at https://github.com/jech/babeld/pull/68 and reuses
the skip-kernel-setup config option.
→ Overall exposure level for babeld.service: 1.7 OK 🙂
This is a bit of a thorny issue. See, the actual `diskSize` variable is
for the *total* disk size, not for the filesystem!
The automatic numbers are meant to compute the *filesystem* required
space. So we have to add any other reserved space!
We have different requirements for reserved space. E.g. there could be
none (when it's actually a filesystem image). There could also be 1MiB
for alignment for an MBR image, legacy+gpt needs 2MiB, then GPT with an
ESP ("bootSize") needs to take the boot partition and GPT size into
account too!
Though luckily(?) for this latter situation we can cheat! As noted in the
change, `bootSize` is NOT the boot partition size. It is actually the
offset where the target filesystem starts.
Reserved space includes:
- inodes space in use (2 blocks per)
- about 5.2% of the space
The 5.2% reserved space was computed empirically when working on a
previous EXT4 image builder. It seems to stabilize around 5% even for
much larger filesystems.
On some filesystems, `du` without `--apparent-size` will not give the
actual size for a file. Using `--apparent-size` will give us the actual
file size.
Though, this is not actually correct still. 1000 × 1 bytes is not 1000
bytes. It is 1000 × ceil(filesize/blockSize)*blockSize.
So instead of adding up the actual file sizes. We are adding up the
block sizes.
Note that this also changes the builder to work with *bytes*, rather
than with any other units. Doing maths on bytes is less likely to go
awry than doing it on other units.
some ban actions need additional packages (eg ipset). since actions can be
provided by the user we need something general that's easy to configure.
we could also enable ipset regardless of the actual configuration of the system
if the iptables firewall is in use (like sshguard does), but that seems very
clumsy and wouldn't easily solve the binary-not-found problems other actions may
also have.
it's not possible to set a different default maxretry value in the DEFAULT jail
because the module already does so. expose the maxretry option to the
configuration to remedy this. (we can't really remove it entirely because
fail2ban defaults to 5)
As a temporary workaround for #120473 while the image builder is patched
to correctly look up disk sizes, partially revert
f3aa040bcb for EC2 disk images only.
We retain the type allowing "auto" but set the default back to the
previous value.
When performing OCR, some of the Tesseract settings perform better than
others on a variety of different workloads, but they mostly take
~negligible incremental time to run compared to the overhead of running
the ImageMagick filters.
After this commit, we try using all three of the current Tesseract
models (classic, LSTM, and classic+LSTM) to generate output text. This
fixes chromium-90's tests at release-20.09, and should make cases where
you're looking for *specific* text better, with the tradeoff of running
Tesseract multiple times.
To make it sensible to cherrypick this into release-20.09, this doesn't
change the existing API surface for the test driver. In particular,
get_screen_text continues to have the existing behaviour.
backends changing shouldn't be very likely, but services may well change. we
should restart sshguard from nixos-rebuild instead of merely plopping down a new
config file and waiting for the user to restart sshguard.
Rather than relying on carefully avoiding touching the 9P-mounted
/nix/store, we instead install a small NixOS system, similar to
the installer tests, and boot from that.
This avoids the various pitfalls associated with trying to unsuspend
properly and trades off a bunch of boilerplate for what will hopefully
be a more reliable test.
Additionally, this test now actually tests booting the system using a
bootloader, rather than the previous method of just booting the kernel
directly.
There is no need for a separate unit. Simplify the NixOS module by adding the shell code to preStart of the main unit, where the other initialization code already is.
