Before this commit there was no way to access (boot into) specialisation of previous generations from grub,even tho they are there.
This commit will add grub submenu for each generation if the generation has any specialisation.
Which will allow you to boot into them.
Co-authored-by: Samuel Dionne-Riel <samuel@dionne-riel.com>
We should sometimes restart the units rather than reloading them so the
changes are actually applied. / and /nix are explicitly excluded because
there was some very old issue where these were unmounted. I don't think
this will affect many people since most people use fstab mounts instead
but I plan to adapt this behavior for fstab mounts as well in the future
(once I wrote a test for the fstab thingies).
When I boot there's a warning `stage-2-init: install: cannot change permissions of '/etc/nixos': No such file or directory`
because my /etc/nixos is a symlink to $HOME/dotfiles.
```
/etc/nixos -> /home/artturin/dotfiles
```
These lines were added in 56b4653904
This avoids creating a build-time reference on `boot.kernelParams` if
the configuration does not use a kernel, i.e., `boot.kernel.enable` is
set to `false`.
There is only other `with` with a somewhat broad scope, `with pkgs`, but
it's used in a place where it would become awkward to change out. And
anyway its scope is rather limited still.
With a limited testing of all packaged GRUB 2 themes (pkgs.nixos-grub2-theme)
this is tested to work.
Without this change, the theme loading will error out (waiting for a key press).
With this change, the theme loads and works as expected.
The intent was to not pass the flag when installing as removable. In
reality there is a third case, where you may not want to touch EFI
variables, and not want to install as removable.
In that case, it would install to the generic \EFI\grub\grubx64.efi,
which is not a good choice in any cases. The operating system should
"own" their path under \EFI\ to be a good citizen [citation needed].
With this change, there can be only two paths GRUB can be installed to:
- \EFI\NixOS-boot\grubx64.efi
- \EFI\BOOT\bootx64.efi
This removes the surprising behaviour where GRUB may be installed to a
different location only because we configured NixOS not to touch EFI
variables.
It may be necessary under some configurations to install GRUB without
touching EFI variables, but to the NixOS-owned location.
This commit updates the binfmt magic-patterns using
f5e6786de4/scripts/qemu-binfmt-conf.sh
The patterns prior to this commit did not understand the difference
between mips32-*-* (32-bit void*,int) and mips64-*-*abin32 (32-bit
void*, 64-bit int). This commit corrects that.
In some setups, and especially with sytemd-networkd becoming more widely
used, networking.useDHCP is set to false. Despite this, it may be useful
to have dhcp in the initramfs.
Build logs show:
> configure: WARNING: non-linux system; not building mount
> configure: WARNING: non-linux system; not building swapon
So skip these on non-Linux
Using getOutput prevents eval failures on other platforms.
Things should stay eval'able with NIXPKGS_ALLOW_UNSUPPORTED_SYSTEM=1
Co-authored-by: Artturin <Artturin@artturin.com>
got broken in 6ea1a2a1be which changed
runCommandCC to runCommand but was not
noticed because it was failing silently
runCommand doesn't include CC or bintools
Example 10. of man page of systemd.network(5) shows:
```
Example 10. MacVTap
This brings up a network interface "macvtap-test" and attaches it to "enp0s25".
# /usr/lib/systemd/network/25-macvtap.network
[Match]
Name=enp0s25
[Network]
MACVTAP=macvtap-test
```
Which is a MACVTAP example and is currently unsupported in NixOS.
This is useful for people using "modern" technologies with virtual machines.
The whole option set was recommended against since mid-2019, and never
worked with the Raspberry Pi 4 family of devices.
We should have deprecated it in early 2020 for removal by 2021. At the
time I did not feel confident in making such a decision, and never
ended-up getting around to it.
The ***only*** supported-by-NixOS boot methods for AArch64 are
standards-based boot methods, namely UEFI or the pragmatically
almost-standard extlinux-compatible for U-Boot.
You can quote me on that.
According to networkd netdev's manpage:
```
Independent=
Takes a boolean. When true, the vxlan interface is created without any underlying network interface. Defaults to false, which means that a .network
file that requests this VXLAN interface using VXLAN= is required for the VXLAN to be created.
```
is a valid option for [VXLAN] section.
According to systemd.netdev manpage:
```
MACAddress=
Specifies the MAC address to use for the device, or takes the special value "none". When "none", systemd-networkd does not request the MAC address for
the device, and the kernel will assign a random MAC address. For "tun", "tap", or "l2tp" devices, the MACAddress= setting in the [NetDev] section is
not supported and will be ignored. Please specify it in the [Link] section of the corresponding systemd.network(5) file. If this option is not set,
"vlan" device inherits the MAC address of the master interface. For other kind of netdevs, if this option is not set, then the MAC address is
generated based on the interface name and the machine-id(5).
Note, even if "none" is specified, systemd-udevd will assign the persistent MAC address for the device, as 99-default.link has
MACAddressPolicy=persistent. So, it is also necessary to create a custom .link file for the device, if the MAC address assignment is not desired.
```
Therefore, `none` is an acceptable value.
These variables were previously used by the activation script
build commands, but are now embedded into those commands for
to improve reusability for an upcoming addition.
Without this change, GRUB installation on non-PC systems (such as
aarch64-linux) only works if boot.loader.grub.devices is set to exactly
`["nodev"]`. If boot.loader.grub.devices was any other value (including
the default `[]`), users got the error:
Died at /nix/store/an9ngv2vg95bdcy0ifsxlbkasprm4dcw-install-grub.pl line 586.
install-grub.pl verifies that if both $grub and $grubEfi are set, then
$grubTarget (e.g. i386-pc) and $grubTargetEfi (e.g. x86_64-efi) must
both be set, or the script will `die`. On non-PC systems, $grubTarget
is "".
When boot.loader.grub.devices is ["nodev"], $grub is set to null,
disabling non-EFI installation. But if a user has devices set for an
x86_64 config, or is using only mirroredBoots without setting devices,
they will hit this `die`.
This change sets $grub to "" if $grubTarget is "".
This helps with understanding the code.
We might make this not depend on environment variables later.
systemBuilderArgs is a form of global state, which isn't helpful.
This essentially backports
https://github.com/systemd/systemd/pull/27791. `systemd-networkd.service`
is sent the `SIGTERM` signal, but it is not required to be stopped
before `initrd-switch-root.target` is reached, despite the use of
`systemctl isolate initrd-switch-root.target`. This is because when
there is no ordering at all between two units, and a transaction stops
one and starts the other, the two operations can happen
simultaneously. This means the service could still be running when
`switch-root` actually occurs. Then, stage 2 systemd will see the
service still running and decide it doesn't need to add a start
operation for it to its initial transaction. Finally, the service
exits, but only after it's already too late. If, however, there is any
ordering at all between a stopping unit and a starting unit, then the
stop operation will be done first. This way, we ensure that the
service is properly exited before doing `switch-root`.
This is something to keep in mind going forward. There may be other
services that need this treatment. These `before` and `conflicts`
definitions are the correct way to ensure a unit is actually stopped
before you reach initrd-switch-root
This patch fixes "Argument list too long" build failure when passing a
list of store paths to system.extraDependencies that exceeds Linux'
MAX_ARG_STRLEN limit of 128 KiB. With the shortest possible derivation
names (one byte), the 128 KiB limit is equivalent to about 2850
derivations. With longer derivations names, the limit is hit earlier.
Fix this restriction.
it is now possible to supply a stratis pool uuid
for every filesystem, and if that filesystem
is required for boot, the relevant pool will be
started in the initramfs.
Enable using an erofs filesystem as one of the filesystems needed to
boot the system. This is useful for example in image based deployments
where the Nix store is mounted read only.
[erofs](https://docs.kernel.org/filesystems/erofs.html) offers multiple
benefits over older filesystems like squashfs. Skip fsck.erofs because
it is still experimental.
Note that this does not add to the `forbiddenDependenciesRegex`
code because that code check should be unaffected as it only checks
output dependencies, not build dependencies.
Build deps are added after that check, if those are enabled in the
first place.
Store the definition files in the initrd instead of reading them from
the Nix store in /sysroot.
This way, the initrd has to be re-generated every time the definition
files change. When the path to the defintion files instead of the
definition files themselves are embedded in the initrd, however, the
initrd also has to be re-generated every time. In this regard, this
change does not improve the status quo.
However, now systemd-repart also works reliable when the Nix store is
mounted separately from the root partition.
This change also enables new use-cases like creating partitions
necessary to boot the system. However, by default, the root partition
cannot be created on first boot because the systemd-repart service
requires a /sysroot to be mounted. Otherwise, systemd-repart cannot
determine the device to operate on.
Since v253, systemd-repart tries to create temporary directories in
/var/tmp. However, this directory doesn't exist in the initrd. This
commit adds an enviroment variable to re-use the existing /tmp directory
instead of /var/tmp.
When `nixpkgs.hostPlatform` != `nixpkgs.buildPlatform`, building the
top-level attribute fails since the bootspec portion of the system
builder tries to reference the host platform's `jq`. Change this to
reference the build platform's `jq`.
The `B` in bridge should be capitalized.
It currently leads to an evuluation error:
```
error: attribute 'sectionbridgeVLAN' missing
at /nix/store/7wmrwj0sgwg1iivxk43lpkqjhji57mq7-source/nixos/modules/system/boot/networkd.nix:2386:56:
2385| example = { VLAN = "10-20"; };
2386| type = types.addCheck (types.attrsOf unitOption) check.network.sectionbridgeVLAN;
| ^
2387| description = lib.mdDoc ''
Did you mean sectionBridgeVLAN?
```
This removes the feature preview warning, enable by default bootspec,
adds a validation flag to prevent Go to go into build-time closure.
This will break all downstream users of bootspec as those changes are
not backward-compatible.
This query yielded no results on search.nixos.org.
I don't think I can make all options magically appear, but you can
the other options by reading the text.
In order to fix
https://github.com/NixOS/nixpkgs/issues/114552 (profile name with
special characters), all OSError have been ignored while only the OSError
with errno 22 (invalid argument) could has been ignored.
The drawback of ignoring all OSError is that the "No space left on
device" error is also ignored. When the /boot doesn't have enough
available disk space, the switch-to-configuration script succeeds
while the boot menu has not been updated: the user thinks it's system
has been updated, but on the next reboot it is actually rollbacked.
This option allows adding the build closure of the system to its
runtime closure, enabling fully-offline rebuilds (as long as no new
packages are added).
Extension based matching for Windows targets define it '.exe' but kernel
documentation explicitly states it should be passed "without the .".
From https://docs.kernel.org/admin-guide/binfmt-misc.html
> * magic:
> [...] If you chose filename extension matching, this is the
> extension to be recognised (without the ., the \x0a specials are not
> allowed).
`systemd-growfs@.service` and `systemd-growfs-root.service` became real units since:
50072ccf1b
we need to add them to the nixos module so growfs works again
systemd now requires the /tmp mount point in the initrd cpio archive
since https://github.com/systemd/systemd/pull/25723
setting `"/tmp/.keep".text` will create the directory.
this fixes a boot failure:
```
(sd-gens): Failed to overmount /tmp/: No such file or directory
```
This reverts commits f5483464d5 and
6b9583e5e1.
Ideally, we shouldn't cause friction for users that bump `stateVersion`,
and I'd consider having to switch and/or manually hardcode a UID/GID
to supress the warning friction. I think it'd be more beneficial to, in
this rare case of an ID being missed, just let it be until more
discussion happens surrounding this overall issue.
See https://github.com/NixOS/nixpkgs/pull/217785 for more context.
Being able to override `boot.initrd.systemd.initrdBin` with `boot.initrd.systemd.extraBin` is a desirable behavior, so this PR changes the `ln -s` command to `ln -sf` to force link even if the file already exists.
{manpage} already exapnds to a link but akkoma wants to link to
a specific setting. split the mention for clarity.
networkd just straight up duplicated what {manpage} generates anyway, so
that link can go away completely.
Since this feature's inception, we've compiled a binary that uses OpenSSL
to perform PBKDF-SHA512 during the extra-utils build. In addition to this
being inefficient, it broke as of 6ea1a2a1be,
which switched the extra-utils derivation to use stdenvNoCC.
For now, I think the path of least resistence is to move the pbkdf-sha512
build to its own derivation, to fix the breakage, as well as improving
the efficiency of the extra-utils build.
(I do believe that at some point, we should revisit this binary -- perhaps
rewriting it -- as Clang even just on its default settings emits more
warnings than you'd want to see in a security-related codebase when
compiling it.)
We shouldn't be creating a systemd.services.systemd-binfmt value when
the upstream unit isn't being pulled in, because it results in a
service unit file with no ExecStart line
this converts meta.doc into an md pointer, not an xml pointer. since we
no longer need xml for manual chapters we can also remove support for
manual chapters from md-to-db.sh
since pandoc converts smart quotes to docbook quote elements and our
nixos-render-docs does not we lose this distinction in the rendered
output. that's probably not that bad, our stylesheet didn't make use of
this anyway (and pre-23.05 versions of the chapters didn't use quote
elements either).
also updates the nixpkgs manual to clarify that option docs support all
extensions (although it doesn't support headings at all, so heading
anchors don't work by extension).
Previously, secrets were named according to the initrd they were
associated with. This created a problem: If secrets were changed whilst
the initrd remained the same, there were two versions of the secrets
with one initrd. The result was that only one version of the secrets would
by recorded into the /boot partition and get used. AFAICT this would
only be the oldest version of the secrets for the given initrd version.
This manifests as #114594, which I found frustrating while trying to use
initrd secrets for the first time. While developing the secrets I found
I could not get new versions of the secrets to take effect.
Additionally, it's a nasty issue to run into if you had cause to change
the initrd secrets for credential rotation, etc, if you change them and
discover you cannot, or alternatively that you can't roll back as you
would expect.
Additional changes in this patch.
* Add a regression test that switching to another grub configuration
with the alternate secrets works. This test relies on the fact that it
is not changing the initrd. I have checked that the test fails if I
undo my change.
* Persist the useBootLoader disk state, similarly to other boot state.
* I had to do this, otherwise I could not find a route to testing the
alternate boot configuration. I did attempt a few different ways of
testing this, including directly running install-grub.pl, but what
I've settled on is most like what a user would do and avoids
depending on lots of internal details.
* Making tests that test the boot are a bit tricky (see hibernate.nix
and installer.nix for inspiration), I found that in addition to
having to copy quite a bit of code I still couldn't get things to
work as desired since the bootloader state was being clobbered.
My change to persist the useBootLoader state could break things,
conceptually. I need some help here discovering if that is the case,
possibly by letting this run through a staging CI if there is one.
Fix#114594.
cc potential reviewers:
@lopsided98 (original implementer) @joachifm (original reviewer),
@wkennington (numerous fixes to grub-install.pl), @lheckemann (wrote
original secrets test).
On some systems, EFI variables are not supported or otherwise wonky.
bootctl attempting to access them causes failures during bootloader
installations and updates. For such systems, NixOS provides the options
`boot.loader.efi.canTouchEfiVariables` and
`boot.loader.systemd-boot.graceful` which pass flags to bootctl that
change whether and how EFI variables are accessed.
Previously, these flags were only passed to bootctl during an install
operation. However, they also apply during an update operation, which
can cause the same sorts of errors. This change passes the flags during
update operations as well to prevent those errors.
Fixes https://github.com/NixOS/nixpkgs/issues/151336
The build of initrd-secrets can routinely fail for old boot entries
if the secrets have been removed or renamed in a later generation.
This always happens for generation 1, because it's built from the
NixOS installer and the paths differs by the mount point (i.e. /mnt).
The error is very confusing because it fails to mention it's about
an older generation and that it's somewhat harmless.
This commit turns the error into a warning for all generations but the
current, adds the name of the failed entry to the message and a note
explaining why it can happen.
apparently pandoc has changed behavior over the past releases, so the
files are no longer in sync. occasionally this requires edits
to the markdown source to not remove an anchor that was there
before (albeit wth a very questionable id), or where things were simply
being misrendered due to syntax errors.
This reverts commit da905d4cf9.
See the commit linked above for further information on why this was
needed. Apparently this is not needed anymore because the need for
LD_LIBRARY_PATH (which is needed for `modprobe(8)` to find
`libpthread.so.0`) doesn't exist anymore.
Since d33e52b253 the library path of each
binary in extra-utils is patched correctly.
nixos-enter sets up /etc/resolv.conf as a bind mount from the host
system, so trying to activate a system that sets
`environment.etc."resolv.conf"` (e.g. with systemd-resolved enabled)
results in an unhelpful warning.
Skip linking /etc/resolv.conf if we're in a nixos-enter environment, as
determined by the IN_NIXOS_ENTER environment variable.
Make the warnings more helpful, indicating which file we failed to link.
Unlink temporary files in case of failure.
That version has a regression that leaves some machines unbootable.
While we wait for the fix (252.2) to land in master, this is a workaround that
should save people some pain.
To reduce size, stage 1 (the initrd) is populated by copying specific
binaries in, then copying the libraries specifically needed by those
binaries. `patchelf` is then used to make the binaries search in the
directory where these libraries are copied to instead of their original
store paths.
Some filesystems (e.g. ZFS) do not guarantee that copying the same files
in the same order into a given directory will result in `find` returning
them in any particular order (though the order appears consistent so
long as the directory is not modified).
Therefore, when the binaries are scanned for libraries to copy in, they
might be scanned in a different order each time the derivation is built.
If two binaries need two different libraries with the same name, then a
different instance of the library might be copied in first, changing the
derivation contents and breaking reproducibility.
This turns out to be the case with `libudev.so.1` from both `systemd`
(needed by e.g. `mdadm`) and `systemdMinimal` (needed by e.g.
`dmsetup`). This issue is fixed by sorting the list of binaries to be
scanned instead of relying on filesystem order so that the same instance
always gets seen and copied first.
Both before this change (at least on ext4) and after this change
(without any options that affect stage 1), this is the `libudev.so.1`
from `systemdMinimal` by way of `dmsetup`. Whether this is appropriate
and how much the two different systemd configurations and udev libraries
need to be involved is a topic left for future work.
We separate the different steps (injecting the toplevel and injecting
the specialisations) so that it's easy to document what each snippet is
actually doing.
