Files
ProxMenux/scripts/global/disk_ops_helpers.sh
T
2026-04-12 20:32:34 +02:00

386 lines
15 KiB
Bash

#!/usr/bin/env bash
# ==========================================================
# ProxMenux - Disk Operations Helpers
# ==========================================================
# Author : MacRimi
# Copyright : (c) 2024 MacRimi
# License : MIT
# Version : 1.0
# Last Updated: 11/04/2026
# ==========================================================
# Shared low-level disk operations: wipe, partition, format.
# Consumed by format-disk.sh, disk_host.sh and future scripts.
#
# Output variables (set by helpers, read by callers):
# DOH_CREATED_PARTITION — partition path set by doh_create_partition()
# DOH_PARTITION_ERROR_DETAIL — error detail set by doh_create_partition()
# ==========================================================
if [[ -n "${__PROXMENUX_DISK_OPS_HELPERS__}" ]]; then
return 0
fi
__PROXMENUX_DISK_OPS_HELPERS__=1
# shellcheck disable=SC2034 # these are output variables read by callers (format-disk.sh, disk_host.sh)
DOH_CREATED_PARTITION=""
DOH_PARTITION_ERROR_DETAIL=""
DOH_FORMAT_ERROR_DETAIL=""
DOH_WIPE_ERROR_DETAIL=""
# Internal: print progress lines only when explicitly enabled by caller.
# Enabled with: export DOH_SHOW_PROGRESS=1
_doh_progress() {
[[ "${DOH_SHOW_PROGRESS:-0}" == "1" ]] || return 0
echo -e "${TAB}${YW}${HOLD}$*${CL}"
}
# Internal: collect command stdout with timeout protection (best-effort).
# Usage: _doh_collect_cmd <seconds> <cmd> [args...]
_doh_collect_cmd() {
local seconds="$1"
shift
if command -v timeout >/dev/null 2>&1; then
timeout --kill-after=2 "${seconds}s" "$@" 2>/dev/null || true
else
"$@" 2>/dev/null || true
fi
}
# Internal: run a command with a timeout, suppressing all output including
# the bash "Killed" job notification that leaks when --kill-after re-raises
# SIGKILL. Plain SIGTERM is not enough for processes stuck in kernel D-state
# (uninterruptible I/O wait on a busy ZFS/LVM disk), so --kill-after=2 is
# needed. The notification is suppressed by temporarily redirecting the
# current shell's stderr with exec before the call and restoring it after.
# Usage: _doh_run_quick_cmd <seconds> <cmd> [args...]
_doh_run_quick_cmd() {
local seconds="$1"
shift
if command -v timeout >/dev/null 2>&1; then
local _saved_stderr
exec {_saved_stderr}>&2 2>/dev/null
timeout --kill-after=2 "${seconds}s" "$@" >/dev/null 2>&1
local rc=$?
exec 2>&"${_saved_stderr}" {_saved_stderr}>&-
return $rc
fi
"$@" >/dev/null 2>&1
}
# Internal: unmount all ZFS datasets then export (or destroy) any ZFS pools
# whose vdevs live on <disk>. Called at the very start of doh_wipe_disk so
# ZFS fully releases the device before wipefs/sgdisk/partprobe touch it.
# If the pool is still held after export, processes on it will be in D-state
# and --kill-after in _doh_run_quick_cmd handles the force-kill.
_doh_release_zfs_pools() {
local disk="$1"
command -v zpool >/dev/null 2>&1 || return 0
local pool_name dev resolved base parent
while read -r pool_name; do
[[ -z "$pool_name" ]] && continue
local found=false
while read -r dev; do
[[ -z "$dev" ]] && continue
if [[ "$dev" == /dev/* ]]; then
resolved=$(readlink -f "$dev" 2>/dev/null)
elif [[ -e "/dev/disk/by-id/$dev" ]]; then
resolved=$(readlink -f "/dev/disk/by-id/$dev" 2>/dev/null)
elif [[ -e "/dev/$dev" ]]; then
resolved=$(readlink -f "/dev/$dev" 2>/dev/null)
else
continue
fi
[[ -z "$resolved" ]] && continue
base=$(lsblk -no PKNAME "$resolved" 2>/dev/null)
parent="${base:+/dev/$base}"
[[ -z "$parent" ]] && parent="$resolved"
if [[ "$parent" == "$disk" || "$resolved" == "$disk" ]]; then
found=true; break
fi
done < <(_doh_collect_cmd 12 zpool list -v -H "$pool_name" | awk '{print $1}' | \
grep -v '^-' | grep -v '^mirror' | grep -v '^raidz' | \
grep -v "^${pool_name}$")
if $found; then
_doh_progress "- Releasing active ZFS pool: $pool_name"
# Unmount all datasets (reverse order: deepest first)
if command -v zfs >/dev/null 2>&1; then
while read -r ds; do
[[ -z "$ds" ]] && continue
timeout 10s zfs unmount -f "$ds" >/dev/null 2>&1 || true
done < <(_doh_collect_cmd 10 zfs list -H -o name -r "$pool_name" | sort -r)
fi
# Export the pool so the kernel releases the block device
timeout 30s zpool export -f "$pool_name" >/dev/null 2>&1 || true
# Wait for udev to finish processing the device release
udevadm settle --timeout=5 >/dev/null 2>&1 || true
sleep 1
fi
done < <(_doh_collect_cmd 8 zpool list -H -o name)
}
# Internal: run a partitioning command with timeout, appending combined output to a file.
# Usage: _doh_part_cmd <seconds> <outfile> <cmd> [args...]
_doh_part_cmd() {
local secs="$1" outfile="$2"
shift 2
if command -v timeout >/dev/null 2>&1; then
timeout --kill-after=3 "${secs}s" "$@" >>"$outfile" 2>&1
else
"$@" >>"$outfile" 2>&1
fi
}
# doh_wipe_disk <disk>
# Unmounts all partitions, deactivates swap, wipes all filesystem metadata
# and partition tables (wipefs + sgdisk + dd first/last 16 MiB).
# Never fails — all sub-commands run with "|| true".
doh_wipe_disk() {
local disk="$1"
local node mountpoint total_sectors seek_sectors discard_max base
DOH_WIPE_ERROR_DETAIL=""
_doh_progress "[1/8] Preparing disk $disk"
# Optional heavy release flow (disabled by default to avoid hangs in busy hosts).
if [[ "${DOH_ENABLE_STACK_RELEASE:-0}" == "1" ]]; then
# Release any ZFS pools using this disk so the kernel lets go of it
_doh_release_zfs_pools "$disk"
# Deactivate any LVM VGs backed by this disk
if command -v vgchange >/dev/null 2>&1; then
local pv rp vg
while read -r pv; do
rp=$(readlink -f "$pv" 2>/dev/null)
base=$(lsblk -no PKNAME "${rp:-$pv}" 2>/dev/null)
if [[ "/dev/${base}" == "$disk" || "$rp" == "$disk" ]]; then
vg=$(_doh_collect_cmd 8 pvs --noheadings -o vg_name "${rp:-$pv}" | xargs)
[[ -n "$vg" ]] && _doh_run_quick_cmd 8 vgchange -an "$vg" || true
fi
done < <(_doh_collect_cmd 8 pvs --noheadings -o pv_name | xargs -r -n1)
fi
fi
# Unmount all partitions
_doh_progress "[2/8] Unmounting partitions"
while read -r node mountpoint; do
[[ -z "$node" || -z "$mountpoint" ]] && continue
_doh_run_quick_cmd 8 umount -f "$node" || true
done < <(lsblk -lnpo NAME,MOUNTPOINT "$disk" 2>/dev/null | awk 'NR>1 && $2!="" {print $1" "$2}')
# Deactivate swap
_doh_progress "[3/8] Disabling swap signatures"
while read -r node; do
[[ -z "$node" ]] && continue
_doh_run_quick_cmd 8 swapoff "$node" || true
done < <(lsblk -lnpo NAME "$disk" 2>/dev/null | awk 'NR>1 {print $1}')
# Wipe filesystem signatures and RAID superblocks on every node
_doh_progress "[4/8] Removing filesystem/RAID signatures"
while read -r node; do
[[ -z "$node" ]] && continue
_doh_run_quick_cmd 10 wipefs -a -f "$node" || true
if command -v mdadm >/dev/null 2>&1; then
_doh_run_quick_cmd 8 mdadm --zero-superblock --force "$node" || true
fi
done < <(lsblk -lnpo NAME "$disk" 2>/dev/null)
# Zap partition table
_doh_progress "[5/8] Resetting partition table"
_doh_run_quick_cmd 12 sgdisk --zap-all "$disk" || true
# TRIM/discard if device supports it
_doh_progress "[6/8] Attempting discard/TRIM when supported"
discard_max=$(lsblk -dn -o DISC-MAX "$disk" 2>/dev/null | xargs)
if [[ -n "$discard_max" && "$discard_max" != "0B" && "$discard_max" != "0" ]]; then
_doh_run_quick_cmd 15 blkdiscard -f "$disk" || true
fi
# Zero first 16 MiB (destroys partition table / filesystem headers)
_doh_progress "[7/8] Zeroing first metadata region"
_doh_run_quick_cmd 20 dd if=/dev/zero of="$disk" bs=1M count=16 conv=fsync status=none || true
# Zero last 16 MiB (destroys backup GPT header)
_doh_progress "[8/8] Zeroing backup GPT region"
total_sectors=$(blockdev --getsz "$disk" 2>/dev/null || echo 0)
if [[ "$total_sectors" =~ ^[0-9]+$ ]] && (( total_sectors > 32768 )); then
seek_sectors=$(( total_sectors - 32768 ))
_doh_run_quick_cmd 20 dd if=/dev/zero of="$disk" bs=512 seek="$seek_sectors" count=32768 conv=fsync status=none || true
fi
udevadm settle --timeout=10 >/dev/null 2>&1 || true
_doh_run_quick_cmd 8 partprobe "$disk" || true
sleep 1
}
# doh_create_partition <disk>
# Creates a single GPT partition spanning the whole disk.
# Tries parted → sgdisk → sfdisk in order; stops at first success.
#
# On success: sets DOH_CREATED_PARTITION to the new partition path, returns 0.
# On failure: sets DOH_PARTITION_ERROR_DETAIL with tool diagnostics, returns 1.
doh_create_partition() {
local disk="$1"
local created=false tmp_out err_snippet
DOH_CREATED_PARTITION=""
DOH_PARTITION_ERROR_DETAIL=""
_doh_run_quick_cmd 5 blockdev --setrw "$disk" || true
# --- attempt 1: parted ---
if command -v parted >/dev/null 2>&1; then
tmp_out=$(mktemp)
if _doh_part_cmd 15 "$tmp_out" parted -s -f "$disk" mklabel gpt; then
if _doh_part_cmd 20 "$tmp_out" parted -s -f "$disk" mkpart primary 1MiB 100%; then
created=true
else
err_snippet=$(tr '\n' ' ' <"$tmp_out" | sed -E 's/[[:space:]]+/ /g; s/^ //; s/ $//')
DOH_PARTITION_ERROR_DETAIL+="parted mkpart: ${err_snippet:-no details}"$'\n'
fi
else
err_snippet=$(tr '\n' ' ' <"$tmp_out" | sed -E 's/[[:space:]]+/ /g; s/^ //; s/ $//')
DOH_PARTITION_ERROR_DETAIL+="parted mklabel: ${err_snippet:-no details}"$'\n'
fi
rm -f "$tmp_out"
else
DOH_PARTITION_ERROR_DETAIL+="parted command not found"$'\n'
fi
# --- attempt 2: sgdisk ---
if [[ "$created" != "true" ]] && command -v sgdisk >/dev/null 2>&1; then
tmp_out=$(mktemp)
_doh_run_quick_cmd 10 sgdisk --zap-all "$disk" || true
# sgdisk does not accept "1MiB" notation — use sector 2048 (= 1 MiB at 512 B/sector)
if _doh_part_cmd 20 "$tmp_out" sgdisk -o -n 1:2048:0 -t 1:8300 "$disk"; then
created=true
else
err_snippet=$(tr '\n' ' ' <"$tmp_out" | sed -E 's/[[:space:]]+/ /g; s/^ //; s/ $//')
DOH_PARTITION_ERROR_DETAIL+="sgdisk create: ${err_snippet:-no details}"$'\n'
fi
rm -f "$tmp_out"
elif [[ "$created" != "true" ]]; then
DOH_PARTITION_ERROR_DETAIL+="sgdisk command not found"$'\n'
fi
# --- attempt 3: sfdisk ---
if [[ "$created" != "true" ]] && command -v sfdisk >/dev/null 2>&1; then
tmp_out=$(mktemp)
local sfdisk_ok=1
if command -v timeout >/dev/null 2>&1; then
printf 'label: gpt\n,;\n' | timeout --kill-after=3 20s sfdisk --wipe always "$disk" >>"$tmp_out" 2>&1
sfdisk_ok=$?
else
printf 'label: gpt\n,;\n' | sfdisk --wipe always "$disk" >>"$tmp_out" 2>&1
sfdisk_ok=$?
fi
if [[ $sfdisk_ok -eq 0 ]]; then
created=true
else
err_snippet=$(tr '\n' ' ' <"$tmp_out" | sed -E 's/[[:space:]]+/ /g; s/^ //; s/ $//')
DOH_PARTITION_ERROR_DETAIL+="sfdisk create: ${err_snippet:-no details}"$'\n'
fi
rm -f "$tmp_out"
elif [[ "$created" != "true" ]]; then
DOH_PARTITION_ERROR_DETAIL+="sfdisk command not found"$'\n'
fi
[[ "$created" == "true" ]] || return 1
# Wait for the kernel to expose the new partition node
udevadm settle --timeout=10 >/dev/null 2>&1 || true
_doh_run_quick_cmd 8 partprobe "$disk" || true
local part
for _ in {1..15}; do
sleep 0.3
part=$(lsblk -lnpo NAME "$disk" 2>/dev/null | awk 'NR==2{print; exit}')
if [[ -n "$part" && -b "$part" ]]; then
DOH_CREATED_PARTITION="$part"
return 0
fi
done
# Fallback: derive partition name from disk path (handles NVMe p-suffix)
local fallback
if [[ "$disk" =~ [0-9]$ ]]; then
fallback="${disk}p1"
else
fallback="${disk}1"
fi
if [[ -b "$fallback" ]]; then
DOH_CREATED_PARTITION="$fallback"
return 0
fi
DOH_PARTITION_ERROR_DETAIL+="partition node not detected after table refresh"$'\n'
return 1
}
# doh_format_partition <partition> <filesystem> [label] [zfs_pool_name] [zfs_mountpoint]
#
# Formats <partition> with <filesystem>.
# label : optional FS label for ext4/xfs/btrfs (ignored for ZFS)
# zfs_pool_name : required when filesystem=zfs; defaults to label if empty
# zfs_mountpoint : ZFS pool mountpoint (default: "none" — no automatic mount)
#
# On failure: sets DOH_FORMAT_ERROR_DETAIL with tool diagnostics.
# Returns 0 on success, 1 on failure.
doh_format_partition() {
local partition="$1"
local filesystem="$2"
local label="${3:-}"
local zfs_pool="${4:-}"
local zfs_mountpoint="${5:-none}"
local tmp_out rc=1
DOH_FORMAT_ERROR_DETAIL=""
tmp_out=$(mktemp)
case "$filesystem" in
ext4)
if [[ -n "$label" ]]; then
mkfs.ext4 -F -L "$label" "$partition" >"$tmp_out" 2>&1; rc=$?
else
mkfs.ext4 -F "$partition" >"$tmp_out" 2>&1; rc=$?
fi
;;
xfs)
if [[ -n "$label" ]]; then
mkfs.xfs -f -L "$label" "$partition" >"$tmp_out" 2>&1; rc=$?
else
mkfs.xfs -f "$partition" >"$tmp_out" 2>&1; rc=$?
fi
;;
exfat)
mkfs.exfat "$partition" >"$tmp_out" 2>&1; rc=$?
;;
btrfs)
if [[ -n "$label" ]]; then
mkfs.btrfs -f -L "$label" "$partition" >"$tmp_out" 2>&1; rc=$?
else
mkfs.btrfs -f "$partition" >"$tmp_out" 2>&1; rc=$?
fi
;;
zfs)
[[ -z "$zfs_pool" ]] && zfs_pool="${label:-pool}"
zpool labelclear -f "$partition" >/dev/null 2>&1 || true
zpool create -f -o ashift=12 \
-O compression=lz4 -O atime=off -O xattr=sa -O acltype=posixacl \
-m "$zfs_mountpoint" "$zfs_pool" "$partition" >"$tmp_out" 2>&1
rc=$?
;;
*)
echo "Unknown filesystem: $filesystem" >"$tmp_out"
rc=1
;;
esac
if [[ $rc -ne 0 ]]; then
DOH_FORMAT_ERROR_DETAIL=$(tr '\n' ' ' <"$tmp_out" | sed -E 's/[[:space:]]+/ /g; s/^ //; s/ $//')
fi
rm -f "$tmp_out"
return $rc
}