use flock in EL disk expand (#892)

2 files changed, 36 insertions, 36 deletions

diff --git a/packages/gce-disk-expand/src/expandfs-lib.sh b/packages/gce-disk-expand/src/expandfs-lib.sh
index 7a26330..657dcfa 100755
--- a/packages/gce-disk-expand/src/expandfs-lib.sh
+++ b/packages/gce-disk-expand/src/expandfs-lib.sh
@@ -84,15 +84,8 @@ sgdisk_fix_gpt() {
   local label=$(sgdisk_get_label "$disk")
   [ "$label" != "gpt" ] && return
 
-  # TODO Find a better solution than sleep.
-  # Add sleeps to allow this operation to fully complete. On some systems, other
-  # operations such as systemd-fsck tasks can collide and fail.
-  udevadm settle
-  sleep 1
   kmsg "Moving GPT header for $disk with sgdisk."
   sgdisk --move-second-header "$disk"
-  udevadm settle
-  sleep 1
 }
 
 # Returns "disk:partition", supporting multiple block types.
@@ -128,8 +121,6 @@ parted_needresize() {
     return 1
   fi
 
-  udevadm settle
-
   if ! printf "$out" | sed '$!d' | grep -q "^${partnum}:"; then
     kmsg "Root partition is not final partition on disk. Not resizing."
     return 1
@@ -157,5 +148,4 @@ parted_resizepart() {
     kmsg "Unable to resize ${disk}${partnum}: ${out}"
     return 1
   fi
-  udevadm settle
 }
diff --git a/packages/gce-disk-expand/src/usr/share/dracut/modules.d/50expand_rootfs/expand_rootfs.sh b/packages/gce-disk-expand/src/usr/share/dracut/modules.d/50expand_rootfs/expand_rootfs.sh
index 6e1be71..41cc6da 100755
--- a/packages/gce-disk-expand/src/usr/share/dracut/modules.d/50expand_rootfs/expand_rootfs.sh
+++ b/packages/gce-disk-expand/src/usr/share/dracut/modules.d/50expand_rootfs/expand_rootfs.sh
@@ -22,9 +22,8 @@
 # logged, rather than causing end of execution. Note that error handling in the
 # main() function always calls return 0
 
-
 main() {
-  local disk="" partnum="" fs_type="" rootdev=""
+  local rootdev="" disk="" partnum=""
 
   # Remove 'block:' prefix and find the root device.
   if ! rootdev=$(readlink -f "${root#block:}") || [ -z "${rootdev}" ]; then
@@ -40,36 +39,47 @@ main() {
   disk=${out%:*}
   partnum=${out#*:}
 
-  if ! parted_needresize "$disk" "$partnum"; then
-    kmsg "Disk ${rootdev} doesn't need resizing"
-    return
-  fi
+  (
+    # If we can't obtain an exclusive lock on FD 9 (which is associated in this
+    # subshell with the root device we're modifying), then exit. This is needed
+    # to prevent systemd from issuing udev re-enumerations and fsck calls before
+    # we're done. See https://systemd.io/BLOCK_DEVICE_LOCKING/
 
-  if ! parted --help | grep -q 'resizepart'; then
-    kmsg "No 'resizepart' command in this parted"
-    return
-  fi
+    if ! flock -n 9; then
+      kmsg "couldn't obtain lock on ${rootdev}"
+      exit
+    fi
 
-  kmsg "Resizing disk ${rootdev}"
-  # Reset the counter for failed job starts to prevent overrunning the default
-  # start burst limits for systemd-fsck-root.service.
-  systemctl reset-failed
+    if ! parted_needresize "$disk" "$partnum"; then
+      kmsg "Disk ${rootdev} doesn't need resizing"
+      exit
+    fi
 
-  # First, move the secondary GPT to the end.
-  if ! out=$(sgdisk_fix_gpt "$disk"); then
-    kmsg "Failed to fix GPT: ${out}"
-  fi
+    if ! parted --help | grep -q 'resizepart'; then
+      kmsg "No 'resizepart' command in this parted"
+      exit
+    fi
 
-  if ! out=$(parted_resizepart "$disk" "$partnum"); then
-    kmsg "Failed to resize partition: ${out}"
-    return
-  fi
+    kmsg "Resizing disk ${rootdev}"
 
-  if ! out=$(resize_filesystem "$rootdev"); then
-    kmsg "Not resizing filesystem: ${out}"
-    return
-  fi
+    # First, move the secondary GPT to the end.
+    if ! out=$(sgdisk_fix_gpt "$disk"); then
+      kmsg "Failed to fix GPT: ${out}"
+    fi
+
+    if ! out=$(parted_resizepart "$disk" "$partnum"); then
+      kmsg "Failed to resize partition: ${out}"
+      exit
+    fi
+
+    if ! out=$(resize_filesystem "$rootdev"); then
+      kmsg "Failed to resize filesystem: ${out}"
+      exit
+    fi
+  ) 9<$rootdev
 }
 
 . /lib/expandfs-lib.sh
+udevadm settle
 main
+udevadm settle