prepare-disk: check state of disk detection when an instance sees another

1 files changed, 123 insertions, 22 deletions

diff --git a/testModule/scripts/prepare-disks b/testModule/scripts/prepare-disks
index 2e7dd673..1992a262 100755
--- a/testModule/scripts/prepare-disks
+++ b/testModule/scripts/prepare-disks
@@ -3,7 +3,15 @@
 # GLOBALS
 #
 
-declare -rg FLAG="/tmp/openslx.disk"
+# flag file containing pids of running instances for concurrency checks
+declare -rg OPENSLX_DISK_FLAG="/tmp/openslx.disk"
+# file that will contain the name of the device used for the /tmp partition
+# - label 'OPENSLX_TMP' in GPT / type '0x44' in MBR
+declare -rg OPENSLX_TMP_MARKER="/tmp/openslx.disk.tmp"
+# file that will contain the name of the device used for storing qcow2
+# - label 'OPENSLX_SYS' in GPT / type '0x46' in MBR
+declare -rg OPENSLX_SYS_MARKER="/tmp/openslx.disk.sys"
+# mount point for system partition
 declare -rg OPENSLX_SYS_MOUNT="/opt/openslx/system"
 
 #
@@ -17,8 +25,7 @@ declare -rg OPENSLX_SYS_MOUNT="/opt/openslx/system"
 # Usage: mount_sys_part <path_to_sys_partition>
 mount_sys_part() {
 	if [ ! -b "$1" ]; then
-		warn "'$1' is not a block device?"
-		echo "!block: $1" >> /tmp/openslx.disk.log
+		warn "Refusing to mount '$1' as its not a block device!"
 		return 1
 	fi
 
@@ -26,13 +33,42 @@ mount_sys_part() {
 	mkdir -p ${OPENSLX_SYS_MOUNT}
 	if ! mount -t auto "${OPENSLX_SYS_DEVICE}" "${OPENSLX_SYS_MOUNT}"; then
 		warn "Mounting '${OPENSLX_SYS_DEVICE}' to '${OPENSLX_SYS_MOUNT}' failed."
-		echo "mount phail" >> /tmp/openslx.disk.log
 		return 1
 	fi
 	return 0
 
 }
 #
+# generic helper to format the given partition with the given filesystem or
+# from the prefdefined list of xfs, ext4, ...
+# Usage: format_disk <dev> <fs>
+#  e.g.  format_disk /dev/sda1 xfs
+format_disk () {
+	local TARGET_DEVICE="$1"
+	local fslist="xfs ext4"
+	# if we have a second arguments, its the filesystem of choice
+	local fs
+	[ $# -ge 2 ] && fslist="$2"
+	for fs in $fslist ; do
+		unset found
+		local MKFS="$(busybox which mkfs.$fs)"
+		if [ -n $MKFS ]; then
+			found=yes
+			case "mkfs.$fs" in
+				mkfs.xfs)   fopt="-fq"  ;;
+				mkfs.ext4)  fopt="-Fq"  ;;
+			esac
+			info "Formatting $TARGET_DEVICE as $fs"
+			return $(${MKFS} ${fopt} "${TARGET_DEVICE}")
+		fi
+		[ -n "$found" ] && break
+	done
+	# still here? then we didn't find a proper formatter...
+	warn "Could not format $PART_DEV as $fs."
+	return 1
+}
+
+#
 # END FUNCTION DEFINITIONS
 ###############################################################################
 
@@ -45,51 +81,116 @@ command -v warn >/dev/null || . /lib/dracut-lib.sh
 # let check the arguments
 if [ "$#" -ne 2 ]; then
 	warn "'$0' need 2 arguments: '$0 [OPENSLX_SYS|OPENSLX_TMP] <dev_path>'"
-	echo "Not enough args!" >> /tmp/openslx.disk.log
 	exit 1
 fi
 # $1 sane?
 if [ "x$1" != "xOPENSLX_SYS" ] && [ "x$1" != "xOPENSLX_TMP" ]; then
-	warn "First arg needs to be either 'OPENSLX_SYS' or 'OPENSLX_TMP'."
-	warn "Given: $1"
-	echo "OPENSLX? $1" >> /tmp/openslx.disk.log
+	warn "First arg needs to be either 'OPENSLX_SYS' or 'OPENSLX_TMP', given: $1"
 	exit 1
 fi
 # $2 sane?
 if [ ! -b "/dev/$2" ]; then
 	warn "Second arg appears not to be a block device!"
-	echo "block? $2" >> /tmp/openslx.disk.log
 	exit 1
 fi
 
-# ok all seems well
+# ok all seems well, set the arguments
 PART_TYPE="$1"
 PART_DEV="/dev/$2"
 
+unset OPENSLX_TMP_DEVICE
+unset OPENSLX_SYS_DEVICE
+
 # lets check if we are already running
-INSTANCES="$(grep "$PART_TYPE" "$FLAG" | busybox wc -l)"
+INSTANCES="$(grep "$PART_TYPE" "$OPENSLX_DISK_FLAG" | busybox wc -l)"
 if [ "$INSTANCES" -ge 1 ]; then
-	# uhoh we are not alone!
-	warn "'$0' already running for $PART_TYPE ... ignoring."
-	echo "already running for $PART_TYPE" >> /tmp/openslx.disk.log
-	exit 1
+	# uhoh we are not alone! Need to check
+	# if the other instance actually did its job
+	warn "'$0' already running for $PART_TYPE ... checking state."
+	# here two/three cases depending on which PART_TYPE we are
+	# currently processing.
+	for timeout in 1 1 2; do
+		# always give the other instance time to finish
+		# but only check 3 times overall
+		sleep $timeout
+		case "$PART_TYPE" in
+			OPENSLX_TMP)
+				# was the tmp partition marker created with a device?
+				[ -f "${OPENSLX_TMP_MARKER}" ] || continue
+				# it was, is it a valid block device?
+				OPENSLX_TMP_DEVICE="$(cat ${OPENSLX_TMP_MARKER})"
+				[ -b $OPENSLX_TMP_DEVICE ] || continue
+				# its detected, its a block device and as mounting is
+				# done later, we can not check for more at this point
+				info "Valid state for $OPENSLX_TMP_DEVICE"
+				exit 0
+				;;
+			OPENSLX_SYS)
+				# was the system partition marker created with a device?
+				[ -f "${OPENSLX_SYS_MARKER}" ] || continue
+				# it was, is it a valid block device?
+				OPENSLX_SYS_DEVICE="$(cat ${OPENSLX_SYS_MARKER})"
+				[ -b $OPENSLX_SYS_DEVICE ] || continue
+				# its detected, its a block device, is it mounted?
+				if mount | grep -qE "^$OPENSLX_SYS_DEVICE\ on $OPENSLX_SYS_MOUNT"; then
+					info "Valid state for $OPENSLX_SYS_DEVICE"
+					exit 0
+				fi
+				# if its not mounted, we want to keep on, so no exit!
+				;;
+			*)
+				# weird case which should never happen
+				warn "If you see this, then $0 was called with a bad PART_TYPE: $@"
+				continue
+				;;
+		esac
+	done
+	warn "'$PART_TYPE' was found but not device was associated to it!"
 fi
 
-# now comes the funny part. We write our pid to $FLAG in order to make sure
+# We write our pid to $OPENSLX_DISK_FLAG in order to make sure
 # we are the only instance of this script running.
-echo "$PART_TYPE.$$" >> "$FLAG"
+info "Processing: $PART_TYPE -> $PART_DEV ($$)"
+echo "$PART_TYPE.$$" >> "$OPENSLX_DISK_FLAG"
 
 # if we are still here, then we can go on and process the partition
-echo "Processing: $PART_TYPE -> $PART_DEV" >> /tmp/openslx.disk.log
-
 if [ "$PART_TYPE" = "OPENSLX_TMP" ]; then
-	# mark it for later mounting!
-	echo "$PART_DEV" > /tmp/openslx.disk.tmp
+	# always format /tmp partition
+	if ! format_disk "${PART_DEV}" xfs; then
+		# error while formatting, cleanup
+		warn "Error formatting $PART_DEV ... removing $$"
+		sed -i "/^${PART_TYPE}\.$$/d" "${OPENSLX_DISK_FLAG}"
+		exit 1
+	fi
+	info "Using '$PART_DEV' as '$PART_TYPE'"
+	# mark it for later: in pre-pivot we will check this file
+	# and mount it as $NEWROOT/tmp
+	echo "$PART_DEV" > "$OPENSLX_TMP_MARKER"
+	# remove our pid from OPENSLX_DISK_FLAG
+	sed -i "/^${PART_TYPE}\.$$/d" "${OPENSLX_DISK_FLAG}"
 	exit 0
 fi
 if [ "$PART_TYPE" = "OPENSLX_SYS" ]; then
+	# TODO make the formatting of the system partition configurable
+	if ! format_disk "${PART_DEV}" xfs; then
+		# error while formatting, cleanup
+		warn "Error formatting $PART_DEV ... removing $$"
+		sed -i "/^${PART_TYPE}\.$$/d" "${OPENSLX_DISK_FLAG}"
+		exit 1
+	fi
+	# mark it
 	# mount it now, since qemu-nbd needs it asap!
-	exit $(mount_sys_part "$PART_DEV")
+	if mount_sys_part "$PART_DEV"; then
+		# mount worked, mark it as done
+		info "Using '$PART_DEV' as '$PART_TYPE'"
+		echo "$PART_DEV" > "$OPENSLX_SYS_MARKER"
+		sed -i "/^${PART_TYPE}\.$$/d" "${OPENSLX_DISK_FLAG}"
+		exit 0
+	else
+		warn "'mount_sys_part' failed in $0"
+		sed -i "/^${PART_TYPE}\.$$/d" "${OPENSLX_DISK_FLAG}"
+		exit 1
+	fi
 fi
 
 #