apt-daily-upgrade.service uses `apt-helper wait-online` as an ExecStartPre
step. On multi-NIC systems with systemd-networkd this helper directly calls
systemd-networkd-wait-online in strict mode, waiting for all managed links
to become online.
On boards with multiple Ethernet ports where some interfaces are commonly
unplugged, this results in repeated timeouts and causes
apt-daily-upgrade / unattended-upgrades to abort, even when at least one
interface is already fully routable.
Replace the ExecStartPre step with a direct invocation of
systemd-networkd-wait-online using `--any`, allowing the service to proceed
as soon as one interface is online.
This preserves the intent of waiting for network availability while making
the behavior robust on multi-NIC and router-style systems.
Multi-port boards (e.g. NanoPi R6S) frequently have unused interfaces with
no-carrier. With the default systemd-networkd-wait-online behavior this can
result in "Online state: partial" and a timeout while waiting for all managed
links to become online.
That failure breaks units that depend on network-online.target such as
apt-daily-upgrade / unattended-upgrades.
Install a systemd drop-in overriding ExecStart to use:
systemd-networkd-wait-online --any --timeout=20
This keeps the intended "wait for networking" semantics while preventing
unplugged/unused ports from blocking boot and timers.
Add extension that enables 32-bit compat vDSO for arm64 kernels.
Without vDSO, 32-bit applications using gettimeofday() syscall heavily
suffer significant performance penalty on arm64 systems.
The extension:
- Adds gcc-arm-linux-gnueabi as build dependency
- Sets CROSS_COMPILE_COMPAT for kernel make
- Enables CONFIG_COMPAT_VDSO in kernel config
Enable with: ENABLE_EXTENSIONS=arm64-compat-vdso
Refs: https://github.com/armbian/build/issues/9216
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- similar to `rkdevflash`, but for Mediatek devices
- also simpler; lk.bin & fip.img are produced by image build and directly used
- requires Rust+Cargo, so add those to hostdeps
- since this is a core extension, those will be included in all Docker images too
- which was bound to happen anyway since Rust in Linux Kernel is no longer an experiment
- extensions/mtkflash: sha1-based bin path, use fork & add `--no-erase-boot1`
- example invocation:
- `BOARD=radxa-nio-12l BRANCH=collabora RELEASE=trixie EXT=ufs,mtkflash MTKFLASH_TTYACM_DEVICE=1`
Ubuntu 25.04+ replaced GNU coreutils with uutils coreutils, a Rust-based
reimplementation of Unix core utilities. These are different projects
with the same package name:
- GNU coreutils 9.x (C) - Ubuntu ≤24.04
- uutils coreutils 0.x (Rust) - Ubuntu ≥25.04
The uutils comm doesn't recognize sort output as sorted, causing
"comm: file is not in sorted order" errors.
Replace comm patterns with:
- grep -vxFf for set difference (lines in B but not in A)
- sort | uniq -d for finding duplicates
These alternatives don't depend on comm, ensuring compatibility
with both GNU and uutils coreutils.
- otherwise it is lost in a temp dir that is cleaned before the end of run
- same place as the `sample-extension.sh`
- keep in mind original limitations of this still apply:
- only extension methods _actually used_ by the specific run are included
- thus, it will _never_ be complete
- that was for my never-upstreamed cloud-init extension
- "/boot/efi" can't be used for cloud seed anymore since trixe
- as it's not going to be mounted by the time cloud-init runs
The upstream driver in radxa-pkg/aic8800 builds successfully against Linux 6.17. Relax the version guard so we only skip installing the aic8800 DKMS package on kernels >= 6.18.
make coderabbitai happy
Move the --[no-]install-recommends flag after the install subcommand and pass each package as its own argument to avoid them being interpreted as a single package name.
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
