fix bugs in disk-expand (#879)

7 files changed, 50 insertions, 35 deletions

diff --git a/packages/gce-disk-expand/packaging/debian/control b/packages/gce-disk-expand/packaging/debian/control
index ac38b5d..92f2e1b 100644
--- a/packages/gce-disk-expand/packaging/debian/control
+++ b/packages/gce-disk-expand/packaging/debian/control
@@ -9,6 +9,7 @@ Homepage: https://github.com/GoogleCloudPlatform/compute-image-packages
 Package: gce-disk-expand
 Architecture: all
 Depends: parted,
+         gdisk,
          ${misc:Depends}
 Description: Automatically resize the root partition on first boot.
  This package resizes the root partition on first boot using parted.
diff --git a/packages/gce-disk-expand/packaging/gce-disk-expand.spec b/packages/gce-disk-expand/packaging/gce-disk-expand.spec
index 409d898..3f6c0dc 100644
--- a/packages/gce-disk-expand/packaging/gce-disk-expand.spec
+++ b/packages/gce-disk-expand/packaging/gce-disk-expand.spec
@@ -20,9 +20,11 @@ License: Apache Software License
 Group: System Environment/Base
 URL: https://github.com/GoogleCloudPlatform/compute-image-packages
 Source0: %{name}_%{version}.orig.tar.gz
-Requires: e2fsprogs, dracut, grep, util-linux, parted
+Requires: e2fsprogs, dracut, grep, util-linux, parted, gdisk
 Conflicts: dracut-modules-growroot
 
+BuildRequires: rsync
+
 # Allow other files in the source that don't end up in the package.
 %define _unpackaged_files_terminate_build 0
 
diff --git a/packages/gce-disk-expand/src/expandfs-lib.sh b/packages/gce-disk-expand/src/expandfs-lib.sh
index 69f67ed..4a02b9f 100755
--- a/packages/gce-disk-expand/src/expandfs-lib.sh
+++ b/packages/gce-disk-expand/src/expandfs-lib.sh
@@ -29,8 +29,8 @@ resize_filesystem() {
       ;;
     ext*)
       if ! out=$(e2fsck -pf "$disk"); then
-        echo "Calling e2fsck \"${disk}\" failed: ${out}"
-        return 1
+        local ret=$?
+        echo "Calling e2fsck \"${disk}\" failed: ${out} exit code ${ret}"
       fi
       if ! out=$(resize2fs "$disk"); then
         echo "Calling resize2fs \"${disk}\" failed: ${out}"
@@ -60,32 +60,31 @@ blkid_get_fstype() (
     echo $ID_FS_TYPE
 )
 
+sgdisk_get_label() {
+    local root="$1"
+    [ -z "$root" ] && return 0
+
+    if sgdisk -p "$root" | grep -q "Found invalid GPT and valid MBR"; then
+        echo "mbr"
+    else
+        echo "gpt"
+    fi
+}
 
-# Checks for and corrects the end-of-disk GPT backup block in case of expanded
-# disk.
-parted_fix_gpt() {
+sgdisk_fix_gpt() {
   local disk="$1"
   [ -z "$disk" ] && return
 
-  if parted -sm "$disk" print 2>&1 | grep -q "fix the GPT"; then
-    # Running parted prompts the user to fix this condition, but only does so in
-    # the interactive exception handler. In order to pass input we must use the
-    # hidden triple-dash flag and pass both print and Fix arguments. `print`
-    # alone will not perform the fix, but `Fix` alone will fail the argument
-    # parser.
-    parted -m ---pretend-input-tty "$disk" print Fix >/dev/null 2>&1 </dev/null
-    parted -m ---pretend-input-tty "$disk" print Fix >/dev/null 2>&1 </dev/null
-    if parted -sm "$disk" print 2>&1 | grep -q "fix the GPT"; then
-      echo "Failed to fix the GPT."
-      return 1
-    fi
-  fi
+  local label=$(sgdisk_get_label "$disk")
+  [ "$label" != "gpt" ] && return
+
+  sgdisk --move-second-header "$disk"
 }
 
 # Returns "disk:partition", supporting multiple block types.
 split_partition() {
   local root="$1" disk="" partnum=""
-  [ -z "$root" ] && return
+  [ -z "$root" ] && return 0
 
   if [ -e /sys/block/${root##*/} ]; then
     echo "Root is not a partition, skipping partition resize."
@@ -104,7 +103,10 @@ split_partition() {
 # Checks if partition needs resizing.
 parted_needresize() {
   local disk="$1" partnum="$2" disksize="" partend=""
-  [ -z "$root" ] && return
+  if [ -z "$disk" ] || [ -z "$partnum" ]; then
+    echo "invalid args to parted_needresize"
+    return 1
+  fi
 
   if ! out=$(parted -sm "$disk" unit b print 2>&1); then
     echo "Failed to get disk details: ${out}"
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 3d12b35..62d7171 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
@@ -16,6 +16,12 @@
 # Contains dracut-specific logic for detecting disk, then calls appropriate
 # library functions.
 
+# Notes for developing dracut modules: this module must never exit with anything
+# other than a 0 exit code. That means no use of set -e or traps on err, and
+# every command must be defensively written so that errors are caught and
+# logged, rather than causing end of execution. Note that error handling in the
+# main() function always calls return 0
+
 kmsg() {
   echo "expand_rootfs: $@" > /dev/kmsg
 }
@@ -49,13 +55,14 @@ main() {
 
   kmsg "Resizing disk ${rootdev}"
 
+  # 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
-    # Try fixing the GPT and try resizing again.
-    parted_fix_gpt "$disk"
-    if ! out=$(parted_resizepart "$disk" "$partnum"); then
-      kmsg "Failed to resize partition: ${out}"
-      return
-    fi
+    kmsg "Failed to resize partition: ${out}"
+    return
   fi
 
   if ! out=$(resize_filesystem "$rootdev"); then
diff --git a/packages/gce-disk-expand/src/usr/share/dracut/modules.d/50expand_rootfs/install b/packages/gce-disk-expand/src/usr/share/dracut/modules.d/50expand_rootfs/install
index cdd021d..4fa6a5b 100755
--- a/packages/gce-disk-expand/src/usr/share/dracut/modules.d/50expand_rootfs/install
+++ b/packages/gce-disk-expand/src/usr/share/dracut/modules.d/50expand_rootfs/install
@@ -21,6 +21,7 @@ inst_hook pre-mount 50 "$moddir/expand_rootfs.sh"
 inst_hook pre-pivot 99 "$moddir/xfs_growfs.sh"
 
 dracut_install parted
+dracut_install sgdisk
 dracut_install cut
 dracut_install sed
 dracut_install grep
diff --git a/packages/gce-disk-expand/src/usr/share/initramfs-tools/hooks/expand-rootfs b/packages/gce-disk-expand/src/usr/share/initramfs-tools/hooks/expand-rootfs
index dae461a..a463ea5 100755
--- a/packages/gce-disk-expand/src/usr/share/initramfs-tools/hooks/expand-rootfs
+++ b/packages/gce-disk-expand/src/usr/share/initramfs-tools/hooks/expand-rootfs
@@ -36,3 +36,5 @@ copy_exec /sbin/e2fsck /bin
 copy_exec /sbin/blkid /bin
 copy_exec /sbin/resize2fs /bin
 copy_exec /bin/udevadm /bin
+copy_exec /usr/sbin/dmidecode /bin
+copy_exec /sbin/sgdisk /bin
diff --git a/packages/gce-disk-expand/src/usr/share/initramfs-tools/scripts/local-premount/expand_rootfs b/packages/gce-disk-expand/src/usr/share/initramfs-tools/scripts/local-premount/expand_rootfs
index a380bc2..9b59b20 100755
--- a/packages/gce-disk-expand/src/usr/share/initramfs-tools/scripts/local-premount/expand_rootfs
+++ b/packages/gce-disk-expand/src/usr/share/initramfs-tools/scripts/local-premount/expand_rootfs
@@ -29,15 +29,15 @@ esac
 . /scripts/expandfs-lib.sh
 
 if ! rootdev=$(resolve_device "${ROOT}"); then
-  log_failure_message "Failed to resolve root device for \"${ROOT}\""
+  log_failure_msg "Failed to resolve root device for \"${ROOT}\""
 fi
 
 if ! fs_type=$(get_fstype "${rootdev}"); then
-  log_failure_message "Failed to determine fstype for \"${rootdev}\""
+  log_failure_msg "Failed to determine fstype for \"${rootdev}\""
 fi
 
 if ! out=$(split_partition "$rootdev"); then
-  log_failure_message "Failed to detect disk and partition info: ${out}"
+  log_failure_msg "Failed to detect disk and partition info: ${out}"
   exit 0
 fi
 
@@ -45,23 +45,23 @@ disk=${out%:*}
 partnum=${out#*:}
 
 if ! parted_needresize "${disk}" "${partnum}"; then
-  log_success_message "Disk ${rootdev} doesn't need resizing."
+  log_success_msg "Disk ${rootdev} doesn't need resizing."
   exit 0
 fi
 
-if ! out=$(parted_fix_gpt "$disk"); then
-  log_failure_message "Failed to fix GPT: ${out}"
+if ! out=$(sgdisk_fix_gpt "$disk"); then
+  log_failure_msg "Failed to fix GPT: ${out}"
   exit 0
 fi
 
 echo "Resizing partition on ${rootdev}"
 if ! out=$(parted_resizepart "${disk}" "${partnum}"); then
-  log_failure_message "Failed to resize partition: ${out}"
+  log_failure_msg "Failed to resize partition: ${out}"
   exit 0
 fi
 
 echo "Resizing filesystem on ${rootdev}"
 if ! out=$(resize_filesystem "${rootdev}"); then
-  log_failure_message "Failed to resize filesystem: ${out}"
+  log_failure_msg "Failed to resize filesystem: ${out}"
   exit 0
 fi