diff options
| -rwxr-xr-x | scripts/wol-f3s | 66 |
1 files changed, 65 insertions, 1 deletions
diff --git a/scripts/wol-f3s b/scripts/wol-f3s index c9563aa..1036394 100755 --- a/scripts/wol-f3s +++ b/scripts/wol-f3s @@ -2,6 +2,13 @@ # Wake-on-LAN and shutdown script for f3s hosts (f0, f1, f2, f3) # and optional shutdown for Raspberry Pi nodes (pi0–pi3) # +# Before any shutdown of f3s Beelink hosts (f0/f1/f2/f3), this script will +# attempt to umount all NFS filesystems currently mounted on this machine +# (earth). NFS mounts here are tunnelled via stunnel (127.0.0.1:2323 → +# 192.168.1.138:2323 CARP VIP on f0/f1). If any NFS mount is active and +# cannot be umounted, the shutdown is aborted to prevent data loss or a +# hung filesystem. +# # Usage: # wol-f3s # Wake f0, f1, and f2 # wol-f3s f0 # Wake only f0 @@ -50,6 +57,46 @@ wake() { wol -i "$BROADCAST" "$mac" } +# umount_nfs_mounts tries to umount all currently mounted NFS filesystems on +# this local machine (earth). It reads /proc/mounts to find active nfs/nfs4 +# mounts, attempts to umount each one, and returns 1 if any mount could not +# be umounted. This prevents shutting down f3s servers while NFS is still in +# use, which would leave the filesystem in a hung/stale state. +umount_nfs_mounts() { + # Collect all active NFS mount points from /proc/mounts (covers nfs and nfs4) + local nfs_mounts=() + while IFS= read -r mountpoint; do + nfs_mounts+=("$mountpoint") + done < <(awk '$3 ~ /^nfs/ { print $2 }' /proc/mounts) + + if [[ ${#nfs_mounts[@]} -eq 0 ]]; then + echo " No NFS filesystems currently mounted — nothing to umount." + return 0 + fi + + local failed=0 + for mp in "${nfs_mounts[@]}"; do + echo " Umounting NFS filesystem: $mp ..." + if umount "$mp" 2>/dev/null; then + echo " ✓ Umounted $mp" + else + echo " ✗ Failed to umount $mp (in use or already unmounted?)" + failed=1 + fi + done + + if [[ $failed -ne 0 ]]; then + echo "" + echo "✗ One or more NFS filesystems could not be umounted." + echo " Aborting shutdown to prevent data loss or a hung filesystem." + echo " Please check running processes that may have open files on NFS," + echo " then retry." + return 1 + fi + + return 0 +} + shutdown_host() { local name=$1 local ip=$2 @@ -81,7 +128,12 @@ case "$ACTION" in wake "f2" "$F2_MAC" ;; shutdown|poweroff|down) - # This is to mute Gogios alerts for a day + # Umount local NFS filesystems first; abort if any are stuck. + # The NFS mounts on earth tunnel through stunnel to the f3s CARP VIP, + # so they will hang or corrupt if the servers go down while mounted. + echo "Checking for locally mounted NFS filesystems..." + umount_nfs_mounts || exit 1 + # Mute Gogios monitoring alerts for a day on both OpenBSD gateways ssh rex@blowfish.buetow.org touch /tmp/f3s_taken_down ssh rex@fishfinger.buetow.org touch /tmp/f3s_taken_down shutdown_host "f0" "$F0_IP" @@ -92,12 +144,18 @@ case "$ACTION" in exit 0 ;; shutdown-f3|poweroff-f3|down-f3) + # Umount local NFS filesystems first; abort if any are stuck. + # f3 does not host the NFS CARP VIP (that is f0/f1), but umounting + # first is still the safe default to avoid stale state. + echo "Checking for locally mounted NFS filesystems..." + umount_nfs_mounts || exit 1 shutdown_host "f3" "$F3_IP" echo "" echo "✓ Shutdown command sent to f3." exit 0 ;; shutdown-pis) + # Raspberry Pis do not serve NFS, so no umount check is needed here. shutdown_host "pi0" "$PI0_IP" shutdown_host "pi1" "$PI1_IP" shutdown_host "pi2" "$PI2_IP" @@ -107,6 +165,12 @@ case "$ACTION" in exit 0 ;; shutdown-all) + # Umount local NFS filesystems first; abort if any are stuck. + # The NFS mounts on earth tunnel through stunnel to the f3s CARP VIP, + # so they will hang or corrupt if the servers go down while mounted. + echo "Checking for locally mounted NFS filesystems..." + umount_nfs_mounts || exit 1 + # Mute Gogios monitoring alerts for a day on both OpenBSD gateways ssh rex@blowfish.buetow.org touch /tmp/f3s_taken_down ssh rex@fishfinger.buetow.org touch /tmp/f3s_taken_down shutdown_host "f0" "$F0_IP" |