ge-aerospace/pxe-server
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

blancco-init: per-kernel kmod tarball, lazy umount of the report share

Browse Source 
- Fetch kmod-$KVER.tar.gz first, falling back to kmod.tar.gz, so more
  than one boot kernel can share the web root. Live now boots the UEFI
  path on Ubuntu's 24.04 HWE 7.0.0-34-generic kernel: Panther Lake
  laptops (Dell Pro 14, AX211 on CNVi) need iwlwifi firmware API 100,
  which the 6.8 driver cannot load. 6.8 stays deployed as the rollback.
- The 10s DefaultTimeoutStopSec did not reach the /mnt/share mount: a
  mount unit's timeout is TimeoutSec=, defaulting to the start timeout.
  Add a mnt-share.mount drop-in with TimeoutSec=10s, LazyUnmount=yes and
  ForceUnmount=yes.

NOT in the playbook yet: it builds the initrd and kmod tarball from the
server's running kernel (6.8). The 7.0 artifacts on 172.16.9.1 were
built off-box; installing linux-generic-hwe-24.04 on the server is what
makes a playbook run reproduce them.
This commit is contained in:
cproudlock
2026-10-05 12:17:11 -04:00
parent 783f2fece0
commit 14eab13a1a
1 changed files with 16 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
playbook/blancco-init.sh
View File

@@ -152,7 +152,11 @@ mkdir -p /run/upper/upper /run/upper/work
mount -t overlay overlay -o lowerdir=/run/lower,upperdir=/run/upper/upper,workdir=/run/upper/work /run/newroot mount -t overlay overlay -o lowerdir=/run/lower,upperdir=/run/upper/upper,workdir=/run/upper/work /run/newroot
echo " [5/5] Loading drivers and settings" echo " [5/5] Loading drivers and settings"
wget -q -T 30 -O /tmp/kmod.tar.gz http://$SERVER/blancco/kmod.tar.gz 2>&1 # Modules (+ iwlwifi firmware) for the running kernel. kmod-$KVER.tar.gz
# lets several boot kernels share one web root; kmod.tar.gz is the
# original single-kernel name, kept as the fallback.
wget -q -T 30 -O /tmp/kmod.tar.gz "http://$SERVER/blancco/kmod-$KVER.tar.gz" 2>/dev/null ||
wget -q -T 30 -O /tmp/kmod.tar.gz http://$SERVER/blancco/kmod.tar.gz 2>&1
[ -s /tmp/kmod.tar.gz ] && (cd /run/newroot && gunzip -c /tmp/kmod.tar.gz | tar xf - && rm -f /tmp/kmod.tar.gz) [ -s /tmp/kmod.tar.gz ] && (cd /run/newroot && gunzip -c /tmp/kmod.tar.gz | tar xf - && rm -f /tmp/kmod.tar.gz)
mkdir -p /run/newroot/albus mkdir -p /run/newroot/albus
@@ -221,13 +225,21 @@ XEOF
# Restart after an erasure stalled on "A stop job is running for # Restart after an erasure stalled on "A stop job is running for
# /mnt/share": the network goes down before the SMB report share is # /mnt/share": the network goes down before the SMB report share is
# unmounted, and the CIFS umount waits out systemd's 90s default. Reports # unmounted, and the CIFS umount waits out the 90s timeout. A mount unit's
# are written long before shutdown, so cap stop jobs at 10s. # timeout is TimeoutSec= (default: the START timeout), so
mkdir -p /run/newroot/etc/systemd/system.conf.d # DefaultTimeoutStopSec alone does not reach it. Lazy + forced umount
# detaches the dead share at once; reports are written long before this.
mkdir -p /run/newroot/etc/systemd/system.conf.d /run/newroot/etc/systemd/system/mnt-share.mount.d
cat > /run/newroot/etc/systemd/system.conf.d/10-fast-stop.conf << XEOF cat > /run/newroot/etc/systemd/system.conf.d/10-fast-stop.conf << XEOF
[Manager] [Manager]
DefaultTimeoutStopSec=10s DefaultTimeoutStopSec=10s
XEOF XEOF
cat > /run/newroot/etc/systemd/system/mnt-share.mount.d/10-fast-umount.conf << XEOF
[Mount]
TimeoutSec=10s
LazyUnmount=yes
ForceUnmount=yes
XEOF
# Disable Xorg screensaver + DPMS so display stays on during long erasures. # Disable Xorg screensaver + DPMS so display stays on during long erasures.
# Numbered 00-* to run before vendor xinitrc.d hooks. # Numbered 00-* to run before vendor xinitrc.d hooks.