`extra-utils` composes the set of programs and libraries needed by
1. copying over all programs
2. copying over all libraries any program directly links against
3. set the runtime path for every program to the library directory
It seems that this approach misses the case where a library itself links
against another library. That is to say, `extra-utils` assumes that
either only progams link against libraries or that every library linked
to by a library is already linked to by a program.
`mount.zfs` linking against `libcrypto`, in turn linking against `libdl`
shows how the current approach falls short:
```
$ objdump -p $(which mount.zfs) | grep NEEDED | grep -e libdl -e libcrypto
NEEDED libcrypto.so.1.1
$ ldd (which mount.zfs) | grep libdl
libdl.so.2 => /nix/store/ybkkrhdwdj227kr20vk8qnzqnmj7a06x-glibc-2.34-115/lib/libdl.so.2 (0x00007f9967a9a000
```
Using `mount.zfs` directly in stage 1 init still works since
`LD_LIBRARY_PATH` overrides this (as intended).
util-linux's `mount` however executes `mount.zfs` with LD_LIBRARY_PATH
removed from its environment as can be seen with strace(1) in an
interactive stage 1 init shell (`boot.shell_on_fail` kernel parameter):
```
# env -i LD_LIBRARY_PATH=$LD_LIBRARY_PATH $(which strace) -ff -e trace=/exec -v -qqq $(which mount) /mnt-root
execve("/nix/store/3gqbb3swgiy749fxd5a4k6kirkr2jr9n-extra-utils/bin/mount", ["/nix/store/3gqbb3swgiy749fxd5a4k"..., "/mnt-root"], ["LD_LIBRARY_PATH=/nix/store/3gqbb"...]) = 0
[pid 1026] execve("/sbin/mount.zfs", ["/sbin/mount.zfs", "<redacted>", "/mnt-root", "-o", "rw,zfsutil"], []) = 0
/sbin/mount.zfs: error while loading shared libraries: libdl.so.2: cannot open shared object file: No such file or directory
--- SIGCHLD {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=1026, si_uid=0, si_status=127, si_utime=0, si_stime=0} ---
```
env(1) is used for clarity (hence subshells for absoloute paths).
While `mount` uses the right library path, `mount.zfs` is stripped of
it, so ld.so(8) fails resolve `libdl` (as required by `libcrypto`).
To fix this and not rely on `LD_LIBRARY_PATH` to be set, fix the library
path inside libraries as well.
This finally mounts all ZFS filesystems using `zfsutil` with correct and
intended mount options.
At least pkgs/os-specific/linux/util-linux/default.nix uses
```
"--enable-fs-paths-default=/run/wrappers/bin:/run/current-system/sw/bin:/sbin"
```
which does not cover stage 1 init's PATH as all executables are put
under /bin/.
Fix util-linux's `mount` usage by symlinking /sbin to it.
Consider ZFS filesystems meant to be mounted with zfs.mount(8), e.g.
```
config.fileSystems."/media".options = [ "zfsutil" ];
config.fileSystems."/nix".options = [ "zfsutil" ];
```
`zfsutil` uses dataset properties as mount options such that zfsprops(7)
do not have to be duplicated in fstab(5) entries or manual mount(8)
invocations.
Given the example configuation above, /media is correctly mounted with
`setuid=off` translated into `nosuid`:
```
$ zfs get -Ho value setuid /media
off
$ findmnt -t zfs -no options /media
rw,nosuid,nodev,noexec,noatime,xattr,posixacl
```
/nix however was mounted with default mount(8) options:
```
$ zfs get -Ho value setuid /nix
off
$ findmnt -t zfs -no options /nix
rw,relatime,xattr,noacl
```
This holds true for all other ZFS properties/mount options, including
`exec/[no]exec`, `devices/[no]dev`, `atime/[no]atime`, etc.
/nix is mounted using BusyBox's `mount` during stage 1 init while /media
is mounted later using proper systemd and/or util-linux's `mount`.
Tracing stage 1 init showed that BusyBox never tried to execute
mount.zfs(8) as intended by `zfsutil`.
Replacing it with util-linux's `mount` and adding the mount helper
showed attempts to execute mount.zfs(8).
Ensure ZFS filesystems are mounted with correct options iff `zfsutil` is
used.
Account for all `with*` options causing their respective unit files to
not be built, just like the current code `withCryptsetup` already does.
This fixes build errors like the following:
```
missing /nix/store/5fafsfms64fn3ywv274ky7arhm9yq2if-systemd-250.4/example/systemd/system/systemd-importd.service
error: builder for '/nix/store/67rdli5q5akzwmqgf8q0a1yp76jgr0px-system-units.drv' failed with exit code 1
```
Found by using a customised systemd package as follows:
```
systemd.package = pkgs.systemd-small;
nixpkgs.config.packageOverrides = pkgs: {
"systemd-small" = pkgs.systemd.override {
withImportd = false;
withMachined = false;
...
};
};
```
These two packages don't have a lib/firmware directory, so putting
them in hardware.firmware has no effect. This will become a hard
error once firmware compression is implemented.
(In the case of Linux, the firmware was all moved to linux-firmware.)
This special case for Btrfs was added in 51bc82960a. One year later beddd36c95 added code to skip the fsck entirely if the filesystem is Btrfs. This made the `if` statement unnecessary.
People running nixos-install in non-NixOS environments
occasionally run into the mktemp builtin not being loaded
into bash (yes, even NixOS' bash). Rather than try and
figure out why exactly that is happening, just use a known
good mktemp from coreutils.
We can make the growfs and makefs binaries conditional because we know
if we'll need them. Also move the cryptsetup generator to the luksroot
so it's not included when not needed.
We drop some generators altogether: systemd-getty-generator because we
don't have getty anyway in stage 1, systemd-system-update-generator
because we don't use that logic in NixOS and
systemd-veritysetup-generator because stage 1 has no veritysetup support
(yet) and if it had, we still wouldn't want to include the generator
unconditionally.
cpio includes the number of directory hard links in archives it creates.
Some filesystems, like btrfs, do not count directory hard links the same
way as more common filesystems like ext4 or tmpfs, so archives built
when /tmp is on such a filesystem do not reproduce. This patch replaces
cpio with bsdtar, which does not have this issue. The specific
invocation is from this page:
https://reproducible-builds.org/docs/archives/
It's already defined in `systemd/user.nix`.
This is a leftover from commit b6d50528dd
where all `systemd.user` settings were moved to `systemd/user.nix`.
- Fix the name of the env
- Add the correct kmod to the initrd
- Add `less` to make journalctl usable
- Fix SYSTEMD_SULOGIN_FORCe for rescue.target
- Add some missing binaries
The networkd.conf file controls a variety of interesting settings
which don't seem to be configurable at the moment, including
adding names to route tables (for networkd only, although this commit
also exports them into iproute2 for convenience's sake), and
the speed metering functionality built into networkd.
Importantly, however, this also allows disabling the systemd
functionality where it likes to delete all the routes and routing rules
that haven't been configured through networkd whenever something causes
it to perform a reconfiguration.
As requested by @roberth, we now have an option similar to
environment.etc. There's also extra store paths to copy and a way to
suppress store paths to make customizations possible.
We also link mount and umount to /bin to make recovery easier when
something fails
using freeform is the new standard way of using modules and should replace
extraConfig.
In particular, this will allow us to place a condition on mails
This accomplishes multiple things:
- Allows us to start systemd without stage-2-init.sh. This was not
possible before because the environment would have been wrong
- `systemctl daemon-reexec` also changes the environment, giving us
newer tools for the fs packages
- Starts systemd in a fully clean environment, making everything more
consistent and pure
At some point, I'd like to make another attempt at
71f1f4884b ("openssl: stop static binaries referencing libs"), which
was reverted in 195c7da07d. One problem with my previous attempt is
that I moved OpenSSL's libraries to a lib output, but many dependent
packages were hardcoding the out output as the location of the
libraries. This patch fixes every such case I could find in the tree.
It won't have any effect immediately, but will mean these packages
will automatically use an OpenSSL lib output if it is reintroduced in
future.
This patch should cause very few rebuilds, because it shouldn't make
any change at all to most packages I'm touching. The few rebuilds
that are introduced come from when I've changed a package builder not
to use variable names like openssl.out in scripts / substitution
patterns, which would be confusing since they don't hardcode the
output any more.
I started by making the following global replacements:
${pkgs.openssl.out}/lib -> ${lib.getLib pkgs.openssl}/lib
${openssl.out}/lib -> ${lib.getLib openssl}/lib
Then I removed the ".out" suffix when part of the argument to
lib.makeLibraryPath, since that function uses lib.getLib internally.
Then I fixed up cases where openssl was part of the -L flag to the
compiler/linker, since that unambigously is referring to libraries.
Then I manually investigated and fixed the following packages:
- pycurl
- citrix-workspace
- ppp
- wraith
- unbound
- gambit
- acl2
I'm reasonably confindent in my fixes for all of them.
For acl2, since the openssl library paths are manually provided above
anyway, I don't think openssl is required separately as a build input
at all. Removing it doesn't make a difference to the output size, the
file list, or the closure.
I've tested evaluation with the OfBorg meta checks, to protect against
introducing evaluation failures.
We can perform most of the mkdir/ln/rm using systemd-tmpfiles
instead which cleans up the script.
/bin and /home are created by their activation script snippets
usbfs is deprecated and unused.
hwclock seems to be automatically executed by systemd on startup.
The mkswap to prevent hibernation cycles seems to be executed by systemd
as well since the provided regression tests succeeds.
