pxe-server/playbook/blancco-init.sh

#!/bin/sh
# Blancco PXE Loader - init script for custom initramfs
# Boot chain: iPXE -> GRUB EFI -> Ubuntu kernel + this initramfs -> switch_root to Blancco
#
# Blancco's own kernel freezes / lacks NIC drivers for some Dell Precision
# hardware during PXE boot. Workaround: boot Ubuntu kernel (which has a wider
# NIC driver set), download Blancco rootfs (squashfs), overlay-mount, and
# switch_root into Blancco's userspace.
#
# Quiet by default: one line per step. Add blancco.debug=1 to the kernel
# cmdline for the full trace (set -x, modprobe -v, ip link/addr). Failures
# always print their diagnostics and drop to sh so the operator can
# investigate without re-imaging.

exec >/dev/console 2>&1

export PATH=/bin:/sbin
mount -t proc proc /proc
mount -t sysfs sysfs /sys
mount -t devtmpfs devtmpfs /dev 2>/dev/null || mount -t tmpfs tmpfs /dev

DEBUG=""
grep -q 'blancco.debug=1' /proc/cmdline && DEBUG=1
# dbg CMD...: run only in debug mode
dbg() { [ -n "$DEBUG" ] && "$@"; return 0; }
if [ -n "$DEBUG" ]; then
    set -x
else
    printf '\033c'   # clear the kernel messages off the screen
    dmesg -n 1 2>/dev/null || true
fi

echo ""
echo "  Blancco Drive Eraser - loading from $( [ -n "$DEBUG" ] && echo "PXE (debug)" || echo "PXE")"
echo ""
mkdir -p /tmp /run /run/lower /run/upper /run/work /run/newroot
KVER=$(uname -r)
dbg echo "Running kernel: $KVER"
dbg ls /lib/modules/

echo "  [1/5] Network drivers"
# Throw the full common-NIC driver list at the wall. modprobe resolves the
# deps from /lib/modules/$KVER/modules.dep (built by prepare-boot-tools).
# Anything missing is silently ignored; whatever matches PCI IDs will bind.
# USB host controllers first: xhci_pci is a module in the Ubuntu kernel
# (CONFIG_USB_XHCI_PCI=m) and there is no udev here, so without this no
# USB 3 / USB-C port works and USB-C to RJ45 adapters never appear.
for drv in \
    xhci_pci xhci_pci_renesas \
    mii libphy ptp \
    e1000 e1000e igb igc ixgbe ixgbevf i40e ice iavf \
    tg3 bnx2 bnx2x bnxt_en b44 \
    r8169 r8152 atlantic \
    vmxnet3 virtio_net virtio_pci \
    pcnet32 8139too 8139cp \
    sfc sfc_ef100 mlx4_en mlx5_core \
    alx atl1c atl1e atl2 \
    via_rhine via_velocity forcedeth \
    pegasus dm9601 asix ax88179_178a cdc_ether cdc_ncm rndis_host; do
    if [ -n "$DEBUG" ]; then
        modprobe -v "$drv" 2>/dev/null && echo "  OK $drv" || true
    else
        modprobe "$drv" >/dev/null 2>&1 || true
    fi
done
sleep 3

echo "  [2/5] Waiting for a network link"
dbg ls /sys/class/net/
dbg ip link

dbg echo "  Waiting up to 60s for non-lo interface..."
IFACE=""
for i in $(seq 1 60); do
    for n in /sys/class/net/*; do
        name="${n##*/}"
        [ "$name" = "lo" ] && continue
        [ -d "$n" ] && IFACE="$name" && break 2
    done
    sleep 1
    echo -n "."
done
echo ""

if [ -z "$IFACE" ]; then
    echo "ERROR: No network interface after 60s"
    echo "=== dmesg tail ==="; dmesg | tail -40
    echo "=== PCI devices (sysfs) ==="; ls /sys/bus/pci/devices/ 2>/dev/null
    echo "=== loaded modules ==="; cat /proc/modules
    echo "Dropping to shell - type 'exit' to reboot."
    exec /bin/sh
fi

# The first interface to appear is not necessarily the cabled one: a
# second onboard port, a dock or a USB NIC can come first, and a static
# IP on a dead port leaves wget at "Connecting" forever. Bring every
# wired interface up and use the first with carrier.
for n in /sys/class/net/*; do
    name="${n##*/}"
    case "$name" in lo|wl*) continue ;; esac
    ip link set "$name" up 2>/dev/null || ifconfig "$name" up 2>/dev/null || true
done
dbg echo "  Waiting up to 30s for a cabled interface..."
CARRIER=""
for i in $(seq 1 30); do
    for n in /sys/class/net/*; do
        name="${n##*/}"
        case "$name" in lo|wl*) continue ;; esac
        # USB NICs enumerate seconds after onboard ports: bring up late ones too
        [ "$(cat "$n/operstate" 2>/dev/null)" = "down" ] && { ip link set "$name" up 2>/dev/null || ifconfig "$name" up 2>/dev/null || true; }
        [ "$(cat "$n/carrier" 2>/dev/null)" = "1" ] && CARRIER="$name" && break 2
    done
    sleep 1
    echo -n "."
done
echo ""
if [ -z "$CARRIER" ]; then
    echo "ERROR: no wired interface has link. Interfaces seen:"
    for n in /sys/class/net/*; do
        drv=$(readlink "$n/device/driver" 2>/dev/null)
        echo "  ${n##*/} carrier=$(cat "$n/carrier" 2>/dev/null) driver=${drv##*/}"
    done
    echo "=== kernel messages (NIC / firmware / link) ==="
    dmesg | grep -i -E 'eth|enp|enx|r8152|r8169|e1000e|igc|cdc|firmware|link' | tail -20
    echo "Check the cable, then type 'exit' to reboot."
    exec /bin/sh
fi
IFACE="$CARRIER"

echo "        using $IFACE"
ip link set "$IFACE" up || ifconfig "$IFACE" up
sleep 2

SERVER=172.16.9.1

# Where the release files (rootfs, preferences, licence container,
# config.img) come from. Default /blancco, the active release. The webapp's
# test menu entry adds blancco.base=/blancco/staged/<version> to boot a
# staged release before it is activated. The kmod tarball, progress
# reporter and debug key are not part of a release and stay in /blancco.
BASE=/blancco
for arg in $(cat /proc/cmdline); do
    case "$arg" in
        blancco.base=/blancco/*) BASE="${arg#blancco.base=}" ;;
    esac
done
case "$BASE" in
    *..*|*[!A-Za-z0-9._/-]*) BASE=/blancco ;;
esac
BASE="${BASE%/}"
[ "$BASE" = /blancco ] || echo "        release under test: $BASE"

# Coarse progress for the webapp's /imaging dashboard (same endpoint the
# WinPE imaging clients use). Stage 1 also resets a finished tile for this
# serial. Best effort: a webapp outage must never stop a wipe.
post_status() {
    # sed, not tr: the initrd's busybox has no tr link
    serial=$(sed 's/["\\]//g' /sys/class/dmi/id/product_serial 2>/dev/null)
    [ -n "$serial" ] || return 0
    wget -q -T 5 -O /dev/null --header 'Content-Type: application/json' \
        --post-data "{\"serial\":\"$serial\",\"pctype\":\"blancco\",\"source\":\"blancco\",\"current_stage\":\"$2\",\"stage_index\":$1,\"stage_total\":7,\"status\":\"in_progress\",\"ip\":\"172.16.9.250\"}" \
        "http://$SERVER:9009/imaging/status" 2>/dev/null || true
}
ifconfig "$IFACE" 172.16.9.250 netmask 255.255.255.0 up
sleep 1
dbg echo "  IP: 172.16.9.250  SERVER: $SERVER"
dbg ip addr

echo "  [3/5] Downloading Blancco (~850 MB)"
post_status 1 "Downloading Blancco"
wget -T 30 -O /tmp/airootfs.sfs http://$SERVER$BASE/arch/x86_64/airootfs.sfs 2>&1
[ -s /tmp/airootfs.sfs ] || { echo "ERROR: download failed"; exec /bin/sh; }

echo "  [4/5] Preparing system"
modprobe overlay 2>/dev/null || insmod /lib/modules/$KVER/kernel/fs/overlayfs/overlay.ko 2>/dev/null
modprobe squashfs 2>/dev/null || insmod /lib/modules/$KVER/kernel/fs/squashfs/squashfs.ko 2>/dev/null
modprobe loop 2>/dev/null
losetup /dev/loop0 /tmp/airootfs.sfs
mount -t squashfs -o ro /dev/loop0 /run/lower
mount -t tmpfs -o size=50% tmpfs /run/upper
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

echo "  [5/5] Loading drivers and settings"
# 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)

mkdir -p /run/newroot/albus

# preferences.xml REQUIRED. Without valid file, Blancco airootfs's
# /opt/scripts/validate_preferences.sh silently restores
# /albus/preferences.save (factory defaults, empty network_share) so
# erasure reports never reach SMB. Fail loud here instead.
PREFS=/run/newroot/albus/preferences.xml
if ! wget -q -T 30 -O "$PREFS" "http://$SERVER$BASE/preferences.xml"; then
    echo "ERROR: preferences.xml download failed from http://$SERVER$BASE/preferences.xml"
    echo "Reports would fall back to factory defaults (no SMB target)."
    echo "Dropping to shell - check Apache + network."
    exec /bin/sh
fi
# Busybox initramfs has no xmllint. Grep for required SMB markers instead;
# if either is missing, the prefs file in web_root is stale or corrupted.
if [ ! -s "$PREFS" ] || ! grep -q "<hostname>$SERVER</hostname>" "$PREFS" \
                     || ! grep -q '<path>blancco-reports</path>' "$PREFS"; then
    echo "ERROR: preferences.xml missing required network_share entries."
    echo "Expected <hostname>$SERVER</hostname> + <path>blancco-reports</path>."
    echo "=== first 60 lines of $PREFS ==="
    head -60 "$PREFS"
    exec /bin/sh
fi
# Clobber preferences.save too. validate_preferences.sh in airootfs falls
# back to preferences.save on any future xmllint failure; if that ever
# fires, we want the fallback to still have the right SMB target.
cp -f "$PREFS" /run/newroot/albus/preferences.save

if ! wget -q -T 30 -O /run/newroot/albus/config.xml "http://$SERVER$BASE/config-clean.xml"; then
    echo "ERROR: config.xml (license container) download failed."
    exec /bin/sh
fi

# On a native boot config.img is an initrd, its files land in the
# initramfs root and archiso keeps them in /run/initramfs, where Blancco's
# /opt/scripts/inc/* pick up report.key (report signing), certificate.ca,
# hybrid.ids and flexible_container. This path never had them, so
# 1_reportKey wrote an empty /albus/report.key. Recreate the layout on a
# tmpfs at the new root's /run; systemd leaves an existing /run mounted.
mount -t tmpfs -o mode=0755,nosuid,nodev tmpfs /run/newroot/run
mkdir -p /run/newroot/run/initramfs
if wget -q -T 30 -O /tmp/config.img "http://$SERVER$BASE/config.img" &&
   (cd /run/newroot/run/initramfs && cpio -id < /tmp/config.img 2>/dev/null) &&
   [ -s /run/newroot/run/initramfs/report.key ]; then
    dbg ls -la /run/newroot/run/initramfs
else
    echo "ERROR: config.img (report key, certificate) download or unpack failed."
    exec /bin/sh
fi
rm -f /tmp/config.img

mkdir -p /run/newroot/etc/X11/xorg.conf.d
# Don't pin a single Xorg driver - hardware varies per site (Intel iGPU,
# NVIDIA GK208, AMD, etc). Let Xorg auto-pick. With nomodeset + KMS
# blacklist on kernel cmdline, Xorg falls back to fbdev (uses
# kernel-provided framebuffer from vga=normal) which works on most boxes.
# Previous "modesetting" pin needed KMS we disabled; "vesa" pin also
# didn't drive NVIDIA cards. Removing the pin entirely.
rm -f /run/newroot/etc/X11/xorg.conf.d/20-failsafeDriver.conf 2>/dev/null || true

# Hard-mask sleep/suspend/hibernate targets in the airootfs overlay. systemd
# kernel cmdline systemd.mask= only blocks systemd-side activation; userspace
# can still write /sys/power/state directly. /dev/null symlinks under
# /etc/systemd/system/ block ALL systemd-mediated paths AND stop logind from
# advertising sleep capability to userland. Combined with the logind drop-in
# below, /sys/power/state writes from non-root userland (Albus) also fail.
mkdir -p /run/newroot/etc/systemd/system
for tgt in sleep.target suspend.target hibernate.target hybrid-sleep.target suspend-then-hibernate.target; do
    ln -sf /dev/null "/run/newroot/etc/systemd/system/$tgt"
done

mkdir -p /run/newroot/etc/systemd/logind.conf.d
cat > /run/newroot/etc/systemd/logind.conf.d/no-suspend.conf << XEOF
[Login]
IdleAction=ignore
HandleSuspendKey=ignore
HandleHibernateKey=ignore
HandleLidSwitch=ignore
HandleLidSwitchDocked=ignore
HandlePowerKey=ignore
XEOF

# Restart after an erasure stalled on "A stop job is running for
# /mnt/share": the network goes down before the SMB report share is
# unmounted, and the CIFS umount waits out the 90s timeout. A mount unit's
# timeout is TimeoutSec= (default: the START timeout), so
# 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
[Manager]
DefaultTimeoutStopSec=10s
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.
# Numbered 00-* to run before vendor xinitrc.d hooks.
mkdir -p /run/newroot/etc/X11/xinit/xinitrc.d
cat > /run/newroot/etc/X11/xinit/xinitrc.d/00-no-screen-blank.sh << 'XEOF'
#!/bin/sh
xset s off -dpms 2>/dev/null || true
xset s noblank 2>/dev/null || true
setterm -blank 0 -powerdown 0 2>/dev/null || true
XEOF
chmod +x /run/newroot/etc/X11/xinit/xinitrc.d/00-no-screen-blank.sh

mkdir -p /run/newroot/proc /run/newroot/sys /run/newroot/dev /run/newroot/run /run/newroot/tmp
# Auto-save reports to the SMB share. process=auto never writes the share
# (Albus forces auto_backup off outside workflow/manual) but does keep each
# report in /reports; a small service copies them across.
mkdir -p /run/newroot/usr/local/sbin /run/newroot/etc/systemd/system/multi-user.target.wants
cat > /run/newroot/usr/local/sbin/blancco-report-sync << 'XEOF'
#!/bin/bash
# Copy Blancco reports to the SMB report share.
#
# In process=auto Albus saves each report to /reports/<document_id>.xml
# (and .pdf when pdf+xml is selected) and uploads it to the cloud, but never
# writes the share: auto_backup is forced off outside workflow/manual mode.
# Albus has the share mounted at /mnt/share already, so copy each finished
# file there, named the way a manual Save names it (UTC
# YYYYMMDD_HHMMSS_report.<ext>, the time taken from the XML so the PDF of
# the same report gets the same name).
# Usage: blancco-report-sync [--once]
DONE=/var/lib/blancco-report-sync
mkdir -p "$DONE"
while :; do
    for f in /reports/*.xml /reports/*.pdf; do
        [ -s "$f" ] || continue
        base=$(basename "$f")
        [ -e "$DONE/$base" ] && continue
        # Leave files Albus may still be writing.
        [ $(( $(date +%s) - $(stat -c %Y "$f") )) -ge 5 ] || continue
        mountpoint -q /mnt/share || continue
        id=${base%.*}
        ext=${base##*.}
        ref="/reports/$id.xml"
        [ -f "$ref" ] || ref="$f"
        name="$(date -u -r "$ref" +%Y%m%d_%H%M%S)_report.$ext"
        if cp "$f" "/mnt/share/$name.part" && mv -f "/mnt/share/$name.part" "/mnt/share/$name"; then
            touch "$DONE/$base"
            echo "copied $base to $name"
            serial=$(tr -d '"\\' 2>/dev/null < /sys/class/dmi/id/product_serial)
            [ -n "$serial" ] && curl -s -m 5 -o /dev/null -H 'Content-Type: application/json' \
                -d "{\"serial\":\"$serial\",\"log_lines\":[\"Report copied to share: $name\"]}" \
                http://172.16.9.1:9009/imaging/status || true
        else
            rm -f "/mnt/share/$name.part"
            echo "copy of $base failed, will retry"
        fi
    done
    [ "$1" = "--once" ] && exit 0
    sleep 10
done
XEOF
chmod 755 /run/newroot/usr/local/sbin/blancco-report-sync
cat > /run/newroot/etc/systemd/system/blancco-report-sync.service << 'XEOF'
[Unit]
Description=Copy Blancco reports to the SMB report share

[Service]
ExecStart=/usr/local/sbin/blancco-report-sync
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target
XEOF
ln -sf /etc/systemd/system/blancco-report-sync.service \
    /run/newroot/etc/systemd/system/multi-user.target.wants/blancco-report-sync.service

# Live erase progress for the /imaging dashboard: follows the AlbusSeven
# journal and posts stages and drive percentages. Fetched at boot so the
# server copy (playbook/blancco-progress.py) is the only one.
if wget -q -T 10 -O /run/newroot/usr/local/sbin/blancco-progress "http://$SERVER/blancco/blancco-progress.py"; then
    chmod 755 /run/newroot/usr/local/sbin/blancco-progress
    cat > /run/newroot/etc/systemd/system/blancco-progress.service << 'XEOF'
[Unit]
Description=Report Blancco erase progress to the PXE webapp

[Service]
ExecStart=/usr/bin/python3 /usr/local/sbin/blancco-progress
Restart=on-failure
RestartSec=10

[Install]
WantedBy=multi-user.target
XEOF
    ln -sf /etc/systemd/system/blancco-progress.service \
        /run/newroot/etc/systemd/system/multi-user.target.wants/blancco-progress.service
else
    echo "  (no blancco-progress.py on server: no live progress on the dashboard)"
fi

# Site timezone for Blancco's own clock (GUI, logs). Report timestamps keep
# an explicit UTC offset; report.utc_offset in the preferences stays 0
# because it is a fixed offset and cannot follow daylight saving.
if [ -e /run/newroot/usr/share/zoneinfo/America/New_York ]; then
    ln -sf /usr/share/zoneinfo/America/New_York /run/newroot/etc/localtime
fi

# Debug boots only: Blancco's own debug mode (/etc/debug.environment makes
# isDebugMode true, so 2_sshd starts sshd) plus an SSH key for root.
# Root is locked and sshd is key-only, so without the key there is no way
# in. The key is served from the PXE server and never baked in here.
if [ -n "$DEBUG" ]; then
    touch /run/newroot/etc/debug.environment
    mkdir -p /run/newroot/root/.ssh
    if wget -q -T 10 -O /run/newroot/root/.ssh/authorized_keys "http://$SERVER/blancco/debug_authorized_keys"; then
        chmod 700 /run/newroot/root/.ssh
        chmod 600 /run/newroot/root/.ssh/authorized_keys
        echo "  debug: sshd on, root key installed"
    else
        echo "  debug: no debug_authorized_keys on server, SSH stays closed"
    fi
fi

mount --move /proc /run/newroot/proc
mount --move /sys /run/newroot/sys
mount --move /dev /run/newroot/dev

# Hard-block kernel suspend/hibernate path. Bind-mount /dev/null over the
# sysfs power-control files so any userland write (Albus auto-suspend,
# pm-utils, etc) becomes a no-op. This is the LAST line of defense - kernel
# cmdline systemd.mask, /dev/null symlinks for sleep targets, and logind
# drop-ins were all bypassed by Albus writing /sys/power/state directly.
# Bind-mount works at the VFS layer below sysfs, so even kernel-side mount
# remounts wouldn't undo it.
mount --bind /dev/null /run/newroot/sys/power/state    2>/dev/null || true
mount --bind /dev/null /run/newroot/sys/power/disk     2>/dev/null || true
mount --bind /dev/null /run/newroot/sys/power/mem_sleep 2>/dev/null || true
mount --bind /dev/null /run/newroot/sys/power/autosleep 2>/dev/null || true

echo "  Starting Blancco..."
post_status 2 "Starting Blancco"
exec switch_root /run/newroot /sbin/init