Initramfs scripts for ZoL.

* Supports booting of a ZFS snapshot.
  Do this by cloning the snapshot into a dataset. If this, the resulting
  dataset, already exists, destroy it. Then mount it on root.
  * If snapshot does not exist, use base dataset (the part before '@')
    as boot filesystem instead.
  * If no snapshot is specified on the 'root=' kernel command line, but there
    is an '@', then get a list of snapshots below that filesystem and ask the
    user which to use.
  * Clone with 'mountpoint=none' and 'canmount=noauto' - we mount manually
    and explicitly.
    * For sub-filesystems, that doesn't have a mountpoint property set, we use
      the 'org.zol:mountpoint' to keep track of it's mountpoint.
  * Allow rollback of snapshots instead of clone it and boot from the clone.
* Allow mounting a root- and subfs with mountpoint=legacy set
* Allow mounting a filesystem which is using nativ encryption.
* Support all currently used kernel command line arguments
  All the different distributions have their own standard on what to specify
  on the kernel command line to boot of a ZFS filesystem.
  * Extra options:
    * zfsdebug=(on,yes,1)	Show extra debugging information
    * zfsforce=(on,yes,1)	Force import the pool
    * rollback=(on,yes,1)	Rollback (instead of clone) the snapshot
* Only try to import pool if it haven't already been imported
  * This will negate the need to force import a pool that have not been exported cleanly.
  * Support exclusion of pools to import by setting ZFS_POOL_EXCEPTIONS in /etc/default/zfs.
* Support additional configuration variable ZFS_INITRD_ADDITIONAL_DATASETS
  to mount additional filesystems not located under your root dataset.
* Include /etc/modprobe.d/{zfs,spl}.conf in the initrd if it/they exist.
* Include the udev rule to use by-vdev for pool imports.
* Include the /etc/default/zfs file to the initrd.
* Only try /dev/disk/by-* in the initrd if USE_DISK_BY_ID is set.
  * Use /dev/disk/by-vdev before anything.
  * Add /dev as a last ditch attempt.
  * Fallback to using the cache file if that exist if nothing else worked.
* Use /sbin/modprobe instead of built-in (BusyBox) modprobe.
  This gets rid of the message "modprobe: can't load module zcommon".
  Thanx to pcoultha for finding this.

Signed-off-by: Turbo Fredriksson <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #2116
Closes #2114

1 files changed, 962 insertions, 0 deletions

diff --git a/contrib/initramfs/scripts/zfs b/contrib/initramfs/scripts/zfs
new file mode 100644
index 000000000..43107c4dc
--- /dev/null
+++ b/contrib/initramfs/scripts/zfs
@@ -0,0 +1,962 @@
+# ZFS boot stub for initramfs-tools.
+#
+# In the initramfs environment, the /init script sources this stub to
+# override the default functions in the /scripts/local script.
+#
+# Enable this by passing boot=zfs on the kernel command line.
+#
+
+# Source the common init script
+. /etc/zfs/zfs-functions
+
+# Paths to what we need - in the initrd, these paths are hardcoded,
+# so override the defines in zfs-functions.
+ZFS="/sbin/zfs"
+ZPOOL="/sbin/zpool"
+ZPOOL_CACHE="/etc/zfs/zpool.cache"
+export ZFS ZPOOL ZPOOL_CACHE
+
+# This runs any scripts that should run before we start importing
+# pools and mounting any filesystems.
+pre_mountroot()
+{
+	if type run_scripts > /dev/null 2>&1 && \
+	    [ -f "/scripts/local-top" -o -d "/scripts/local-top" ]
+	then
+		[ "$quiet" != "y" ] && \
+		    zfs_log_begin_msg "Running /scripts/local-top"
+		run_scripts /scripts/local-top
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	if type run_scripts > /dev/null 2>&1 && \
+	    [ -f "/scripts/local-premount" -o -d "/scripts/local-premount" ]
+	then
+		[ "$quiet" != "y" ] && \
+		    zfs_log_begin_msg "Running /scripts/local-premount"
+		run_scripts /scripts/local-premount
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+}
+
+# If plymouth is availible, hide the splash image.
+disable_plymouth()
+{
+	if [ -x /bin/plymouth ] && /bin/plymouth --ping
+	then
+		/bin/plymouth hide-splash >/dev/null 2>&1
+	fi
+}
+
+# Get a ZFS filesystem property value.
+get_fs_value()
+{
+	local fs="$1"
+	local value=$2
+
+	"${ZFS}" get -H -ovalue $value "$fs" 2> /dev/null
+}
+
+# Find the 'bootfs' property on pool $1.
+# If the property does not contain '/', then ignore this
+# pool by exporting it again.
+find_rootfs()
+{
+	local pool="$1"
+
+	# If it's already specified, just keep it mounted and exit
+	# User (kernel command line) must be correct.
+	if [ -n "${ZFS_BOOTFS}" ]
+	then
+		POOL_IMPORTED=1
+		return 0
+	fi
+
+	# Not set, try to find it in the 'bootfs' property of the pool.
+	# NOTE: zpool does not support 'get -H -ovalue bootfs'...
+	ZFS_BOOTFS=$("${ZPOOL}" list -H -obootfs "$pool")
+
+	# Make sure it's not '-' and that it starts with /.
+	if [ "${ZFS_BOOTFS}" != "-" ] && \
+		$(get_fs_value "${ZFS_BOOTFS}" mountpoint | grep -q '^/$')
+	then
+		# Keep it mounted
+		POOL_IMPORTED=1
+		return 0
+	fi
+
+	# Not boot fs here, export it and later try again..
+	"${ZPOOL}" export "$pool"
+
+	return 1
+}
+
+# Support function to get a list of all pools, separated with ';'
+find_pools()
+{
+	local CMD="$*"
+	local pools
+
+	pools=$($CMD 2> /dev/null | \
+		grep -E "pool:|^[a-zA-Z0-9]" | \
+		sed 's@.*: @@' | \
+		sort | \
+		while read pool; do \
+		    echo -n "$pool;"
+		done)
+
+	echo "${pools%%;}" # Return without the last ';'.
+}
+
+# Get a list of all availible pools
+get_pools()
+{
+	local available_pools npools
+
+	# Get the base list of availible pools.
+	available_pools=$(find_pools "$ZPOOL" import)
+
+	# Just in case - seen it happen (that a pool isn't visable/found
+	# with a simple "zpool import" but only when using the "-d"
+	# option or setting ZPOOL_IMPORT_PATH).
+	if [ -d "/dev/disk/by-id" ]
+	then
+		npools=$(find_pools "$ZPOOL" import -d /dev/disk/by-id)
+		if [ -n "$npools" ]
+		then
+			# Because we have found extra pool(s) here, which wasn't
+			# found 'normaly', we need to force USE_DISK_BY_ID to
+			# make sure we're able to actually import it/them later.
+			USE_DISK_BY_ID='yes'
+
+			if [ -n "$available_pools" ]
+			then
+				# Filter out duplicates (pools found with the simple
+				# "zpool import" but which is also found with the
+				# "zpool import -d ...").
+				npools=$(echo "$npools" | sed "s,$available_pools,,")
+
+				# Add the list to the existing list of
+				# available pools
+				available_pools="$available_pools;$npools"
+			else
+				available_pools="$npools"
+			fi
+		fi
+	fi
+
+        # Filter out any exceptions...
+	if [ -n "$ZFS_POOL_EXCEPTIONS" ]
+	then
+		local found=""
+		local apools=""
+		local pool exception
+		OLD_IFS="$IFS" ; IFS=";"
+
+		for pool in $available_pools
+		do
+			for exception in $ZFS_POOL_EXCEPTIONS
+			do
+				[ "$pool" = "$exception" ] && continue 2
+				found="$pool"
+			done
+
+			if [ -n "$found" ]
+			then
+				if [ -n "$apools" ]
+				then
+					apools="$apools;$pool"
+				else
+					apools="$pool"
+				fi
+			fi
+		done
+
+		IFS="$OLD_IFS"
+		available_pools="$apools"
+	fi
+
+	# Return list of availible pools.
+	echo "$available_pools"
+}
+
+# Import given pool $1
+import_pool()
+{
+	local pool="$1"
+	local dirs dir ZFS_CMD ZFS_STDERR ZFS_ERROR
+
+	# Verify that the pool isn't already imported
+	# Make as sure as we can to not require '-f' to import.
+	"${ZPOOL}" status "$pool" > /dev/null 2>&1 && return 0
+
+	# For backwards compability, make sure that ZPOOL_IMPORT_PATH is set
+	# to something we can use later with the real import(s). We want to
+	# make sure we find all by* dirs, BUT by-vdev should be first (if it
+	# exists).
+	if [ -n "$USE_DISK_BY_ID" -a -z "$ZPOOL_IMPORT_PATH" ]
+	then
+		dirs="$(for dir in $(echo /dev/disk/by-*)
+		do
+			# Ignore by-vdev here - we want it first!
+			echo "$dir" | grep -q /by-vdev && continue
+			[ ! -d "$dir" ] && continue
+
+			echo -n "$dir:"
+		done | sed 's,:$,,g')"
+
+		if [ -d "/dev/disk/by-vdev" ]
+		then
+			# Add by-vdev at the beginning.
+			ZPOOL_IMPORT_PATH="/dev/disk/by-vdev:"
+		fi
+
+		# ... and /dev at the very end, just for good measure.
+		ZPOOL_IMPORT_PATH="$ZPOOL_IMPORT_PATH$dirs:/dev"
+	fi
+
+	# Needs to be exported for "zpool" to catch it.
+	[ -n "$ZPOOL_IMPORT_PATH" ] && export ZPOOL_IMPORT_PATH
+
+
+	[ "$quiet" != "y" ] && zfs_log_begin_msg \
+		"Importing pool '${pool}' using defaults"
+
+	ZFS_CMD="${ZPOOL} import -N ${ZPOOL_FORCE}"
+	ZFS_STDERR="$($ZFS_CMD "$pool" 2>&1)"
+	ZFS_ERROR="$?"
+	if [ "${ZFS_ERROR}" != 0 ]
+	then
+		[ "$quiet" != "y" ] && zfs_log_failure_msg "${ZFS_ERROR}"
+
+		[ "$quiet" != "y" ] && zfs_log_begin_msg \
+			"Importing pool '${pool}' using cachefile."
+
+		ZFS_CMD="${ZPOOL} import -c ${ZPOOL_CACHE} -N ${ZPOOL_FORCE}"
+		ZFS_STDERR="$($ZFS_CMD "$pool" 2>&1)"
+		ZFS_ERROR="$?"
+		if [ "${ZFS_ERROR}" != 0 ]
+		then
+			[ "$quiet" != "y" ] && zfs_log_failure_msg "${ZFS_ERROR}"
+
+			disable_plymouth
+			echo ""
+			echo "Command: ${ZFS_CMD} '$pool'"
+			echo "Message: $ZFS_STDERR"
+			echo "Error: $ZFS_ERROR"
+			echo ""
+			echo "Failed to import pool '$pool'."
+			echo "Manually import the pool and exit."
+			/bin/sh
+		fi
+	fi
+
+	[ "$quiet" != "y" ] && zfs_log_end_msg
+	return 0
+}
+
+# Load ZFS modules
+# Loading a module in a initrd require a slightly different approach,
+# with more logging etc.
+load_module_initrd()
+{
+	if [ "$ZFS_INITRD_PRE_MOUNTROOT_SLEEP" > 0 ]
+	then
+		if [ "$quiet" != "y" ]; then
+			zfs_log_begin_msg "Sleeping for" \
+				"$ZFS_INITRD_PRE_MOUNTROOT_SLEEP seconds..."
+		fi
+		sleep "$ZFS_INITRD_PRE_MOUNTROOT_SLEEP"
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	# Wait for all of the /dev/{hd,sd}[a-z] device nodes to appear.
+	if type wait_for_udev > /dev/null 2>&1 ; then
+		wait_for_udev 10
+	elif type wait_for_dev > /dev/null 2>&1 ; then
+		wait_for_dev
+	fi
+
+	# zpool import refuse to import without a valid mtab
+	[ ! -f /proc/mounts ] && mount proc /proc
+	[ ! -f /etc/mtab ] && cat /proc/mounts > /etc/mtab
+
+	# Load the module
+	load_module || return 1
+
+	if [ "$ZFS_INITRD_POST_MODPROBE_SLEEP" > 0 ]
+	then
+		if [ "$quiet" != "y" ]; then
+			zfs_log_begin_msg "Sleeping for" \
+				"$ZFS_INITRD_POST_MODPROBE_SLEEP seconds..."
+		fi
+		sleep "$ZFS_INITRD_POST_MODPROBE_SLEEP"
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	return 0
+}
+
+# Mount a given filesystem
+mount_fs()
+{
+	local fs="$1"
+	local mountpoint ZFS_CMD ZFS_STDERR ZFS_ERROR
+
+	# Need the _original_ datasets mountpoint!
+	mountpoint=$(get_fs_value "$fs" mountpoint)
+	if [ "$mountpoint" = "legacy" -o "$mountpoint" = "none" ]; then
+		# Can't use the mountpoint property. Might be one of our
+		# clones. Check the 'org.zol:mountpoint' property set in
+		# clone_snap() if that's usable.
+		mountpoint=$(get_fs_value "$fs" org.zol:mountpoint)
+		if [ "$mountpoint" = "legacy" -o \
+		    "$mountpoint" = "none" -o \
+		    "$mountpoint" = "-" ]
+		then
+			if [ "$fs" != "${ZFS_BOOTFS}" ]; then
+				# We don't have a proper mountpoint, this
+				# isn't the root fs. So extract the root fs
+				# value from the filesystem, and we should
+				# (hopefully!) have a mountpoint we can use.
+				mountpoint="${fs##$ZFS_BOOTFS}"
+			else
+				# Last hail-mary: Hope 'rootmnt' is set!
+				mountpoint=""
+			fi
+		fi
+
+		if [ "$mountpoint" = "legacy" ]; then
+			ZFS_CMD="mount -t zfs"
+		else
+			# If it's not a legacy filesystem, it can only be a
+			# native one...
+			ZFS_CMD="mount -o zfsutil -t zfs"
+		fi
+	else
+		ZFS_CMD="mount -o zfsutil -t zfs"
+	fi
+
+	# Possibly decrypt a filesystem using native encryption.
+	decrypt_fs "$fs"
+
+	[ "$quiet" != "y" ] && \
+	    zfs_log_begin_msg "Mounting '${fs}' on '${rootmnt}/${mountpoint}'"
+	[ -n "${ZFS_DEBUG}" ] && \
+	    zfs_log_begin_msg "CMD: '$ZFS_CMD ${fs} ${rootmnt}/${mountpoint}'"
+
+	ZFS_STDERR=$(${ZFS_CMD} "${fs}" "${rootmnt}/${mountpoint}" 2>&1)
+	ZFS_ERROR=$?
+	if [ "${ZFS_ERROR}" != 0 ]
+	then
+		[ "$quiet" != "y" ] && zfs_log_failure_msg "${ZFS_ERROR}"
+
+		disable_plymouth
+		echo ""
+		echo "Command: ${ZFS_CMD} ${fs} ${rootmnt}/${mountpoint}"
+		echo "Message: $ZFS_STDERR"
+		echo "Error: $ZFS_ERROR"
+		echo ""
+		echo "Failed to mount ${fs} on ${rootmnt}/${mountpoint}."
+		echo "Manually mount the filesystem and exit."
+		/bin/sh
+	else
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	return 0
+}
+
+# Unlock a ZFS native crypted filesystem.
+decrypt_fs()
+{
+	local fs="$1"
+	local ZFS_CMD ZFS_STDERR ZFS_ERROR
+
+	# If the 'zfs key' command isn't availible, exit right here.
+	"${ZFS}" 2>&1 | grep -q 'key -l ' || return 0
+
+	# Check if filesystem is encrypted. If not, exit right here.
+	[ "$(get_fs_value "$fs" encryption)" != "off" ] || return 0
+
+	[ "$quiet" != "y" ] && \
+	    zfs_log_begin_msg "Loading crypto wrapper key for $fs"
+
+	# Just make sure that ALL crypto modules module is loaded.
+	# Simplest just to load all...
+	for mod in sun-ccm sun-gcm sun-ctr
+	do
+		[ "$quiet" != "y" ] && zfs_log_progress_msg "${mod} "
+
+		ZFS_CMD="/sbin/modprobe $mod"
+		ZFS_STDERR="$(${ZFS_CMD} 2>&1)"
+		ZFS_ERROR="$?"
+
+		if [ "${ZFS_ERROR}" != 0 ]
+		then
+			[ "$quiet" != "y" ] && zfs_log_failure_msg "${ZFS_ERROR}"
+
+			disable_plymouth
+			echo ""
+			echo "Command: $ZFS_CMD"
+			echo "Message: $ZFS_STDERR"
+			echo "Error: $ZFS_ERROR"
+			echo ""
+			echo "Failed to load $mod module."
+			echo "Please verify that it is availible on the initrd image"
+			echo "(without it it won't be possible to unlock the filesystem)"
+			echo "and rerun:  $ZFS_CMD"
+			/bin/sh
+		else
+			[ "$quiet" != "y" ] && zfs_log_end_msg
+		fi
+	done
+
+	# If the key isn't availible, then this will fail!
+	ZFS_CMD="${ZFS} key -l -r $fs"
+	ZFS_STDERR="$(${ZFS_CMD} 2>&1)"
+	ZFS_ERROR="$?"
+
+	if [ "${ZFS_ERROR}" != 0 ]
+	then
+		[ "$quiet" != "y" ] && zfs_log_failure_msg "${ZFS_ERROR}"
+
+		disable_plymouth
+		echo ""
+		echo "Command: $ZFS_CMD"
+		echo "Message: $ZFS_STDERR"
+		echo "Error: $ZFS_ERROR"
+		echo ""
+		echo "Failed to load zfs encryption wrapper key (s)."
+		echo "Please verify dataset property 'keysource' for datasets"
+		echo "and rerun:  $ZFS_CMD"
+		/bin/sh
+	else
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	return 0
+}
+
+# Destroy a given filesystem.
+destroy_fs()
+{
+	local fs="$1"
+	local ZFS_CMD ZFS_STDERR ZFS_ERROR
+
+	[ "$quiet" != "y" ] && \
+	    zfs_log_begin_msg "Destroying '$fs'"
+
+	ZFS_CMD="${ZFS} destroy $fs"
+	ZFS_STDERR="$(${ZFS_CMD} 2>&1)"
+	ZFS_ERROR="$?"
+	if [ "${ZFS_ERROR}" != 0 ]
+	then
+		[ "$quiet" != "y" ] && zfs_log_failure_msg "${ZFS_ERROR}"
+
+		disable_plymouth
+		echo ""
+		echo "Command: $ZFS_CMD"
+		echo "Message: $ZFS_STDERR"
+		echo "Error: $ZFS_ERROR"
+		echo ""
+		echo "Failed to destroy '$fs'. Please make sure that '$fs' is not availible."
+		echo "Hint: Try:  zfs destroy -Rfn $fs"
+		echo "If this dryrun looks good, then remove the 'n' from '-Rfn' and try again."
+		/bin/sh
+	else
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	return 0
+}
+
+# Clone snapshot $1 to destination filesystem $2
+# Set 'canmount=noauto' and 'mountpoint=none' so that we get to keep
+# manual controll over it's mounting (i.e., make sure it's not automatically
+# mounted with a 'zfs mount -a' in the init/systemd scripts).
+clone_snap()
+{
+	local snap="$1"
+	local destfs="$2"
+	local mountpoint="$3"
+	local ZFS_CMD ZFS_STDERR ZFS_ERROR
+
+	[ "$quiet" != "y" ] && zfs_log_begin_msg "Cloning '$snap' to '$destfs'"
+
+	# Clone the snapshot into a dataset we can boot from
+	# + We don't want this filesystem to be automatically mounted, we
+	#   want controll over this here and nowhere else.
+	# + We don't need any mountpoint set for the same reason.
+	# We use the 'org.zol:mountpoint' property to remember the mountpoint.
+	ZFS_CMD="${ZFS} clone -o canmount=noauto -o mountpoint=none"
+	ZFS_CMD="${ZFS_CMD} -o org.zol:mountpoint=${mountpoint}"
+	ZFS_CMD="${ZFS_CMD} $snap $destfs"
+	ZFS_STDERR="$(${ZFS_CMD} 2>&1)"
+	ZFS_ERROR="$?"
+	if [ "${ZFS_ERROR}" != 0 ]
+	then
+		[ "$quiet" != "y" ] && zfs_log_failure_msg "${ZFS_ERROR}"
+
+		disable_plymouth
+		echo ""
+		echo "Command: $ZFS_CMD"
+		echo "Message: $ZFS_STDERR"
+		echo "Error: $ZFS_ERROR"
+		echo ""
+		echo "Failed to clone snapshot."
+		echo "Make sure that the any problems are corrected and then make sure"
+		echo "that the dataset '$destfs' exists and is bootable."
+		/bin/sh
+	else
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	return 0
+}
+
+# Rollback a given snapshot.
+rollback_snap()
+{
+	local snap="$1"
+	local ZFS_CMD ZFS_STDERR ZFS_ERROR
+
+	[ "$quiet" != "y" ] && zfs_log_begin_msg "Rollback $snap"
+
+	ZFS_CMD="${ZFS} rollback -Rf $snap"
+	ZFS_STDERR="$(${ZFS_CMD} 2>&1)"
+	ZFS_ERROR="$?"
+	if [ "${ZFS_ERROR}" != 0 ]
+	then
+		[ "$quiet" != "y" ] && zfs_log_failure_msg "${ZFS_ERROR}"
+
+		disable_plymouth
+		echo ""
+		echo "Command: $ZFS_CMD"
+		echo "Message: $ZFS_STDERR"
+		echo "Error: $ZFS_ERROR"
+		echo ""
+		echo "Failed to rollback snapshot."
+		/bin/sh
+	else
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	return 0
+}
+
+# Get a list of snapshots, give them as a numbered list
+# to the user to choose from.
+ask_user_snap()
+{
+	local fs="$1"
+	local i=1
+	local SNAP snapnr snap debug
+
+	# We need to temporarily disable debugging. Set 'debug' so we
+	# remember to enabled it again.
+	if [ -n "${ZFS_DEBUG}" ]; then
+		unset ZFS_DEBUG
+		set +x
+		debug=1
+	fi
+
+	# Because we need the resulting snapshot, which is sent on
+	# stdout to the caller, we use stderr for our questions.
+	echo "What snapshot do you want to boot from?" > /dev/stderr
+	while read snap; do
+	    echo "  $i: ${snap}" > /dev/stderr
+	    eval `echo SNAP_$i=$snap`
+	    i=$((i + 1))
+	done <<EOT
+$("${ZFS}" list -H -oname -tsnapshot "${fs}")
+EOT
+
+	echo -n "  Snap nr [0-$((i-1))]? " > /dev/stderr
+	read snapnr
+
+	# Reenable debugging.
+	if [ -n "${debug}" ]; then
+		ZFS_DEBUG=1
+		set -x
+	fi
+
+	echo "$(eval echo "$"SNAP_$snapnr)"
+}
+
+setup_snapshot_booting()
+{
+	local snap="$1"
+	local s destfs subfs mountpoint retval=0 filesystems fs
+
+	# Make sure that the snapshot specified actually exist.
+	if [ ! $(get_fs_value "${snap}" type) ]
+	then
+		# Snapshot does not exist (...@<null> ?)
+		# ask the user for a snapshot to use.
+		snap="$(ask_user_snap "${snap%%@*}")"
+	fi
+
+	# Separate the full snapshot ('$snap') into it's filesystem and
+	# snapshot names. Would have been nice with a split() function..
+	rootfs="${snap%%@*}"
+	snapname="${snap##*@}"	
+	ZFS_BOOTFS="${rootfs}_${snapname}"
+
+	if ! grep -qiE '(^|[^\\](\\\\)* )(rollback)=(on|yes|1)( |$)' /proc/cmdline
+	then
+		# If the destination dataset for the clone
+		# already exists, destroy it. Recursivly
+		if [ $(get_fs_value "${rootfs}_${snapname}" type) ]; then
+			filesystems=$("${ZFS}" list -oname -tfilesystem -H \
+			    -r "${ZFS_BOOTFS}" | sort -r)
+			for fs in $filesystems; do
+				destroy_fs "${fs}"
+			done
+		fi
+	fi
+
+	# Get all snapshots, recursivly (might need to clone /usr, /var etc
+	# as well).
+	for s in $("${ZFS}" list -H -oname -tsnapshot -r "${rootfs}" | \
+	    grep "${snapname}")
+	do
+		if grep -qiE '(^|[^\\](\\\\)* )(rollback)=(on|yes|1)( |$)' /proc/cmdline
+		then
+			# Rollback snapshot
+			rollback_snap "$s" || retval=$((retval + 1))
+		else
+			# Setup a destination filesystem name.
+			# Ex: Called with 'rpool/ROOT/debian@snap2'
+			#       rpool/ROOT/debian@snap2		=> rpool/ROOT/debian_snap2
+			#       rpool/ROOT/debian/boot@snap2	=> rpool/ROOT/debian_snap2/boot
+			#       rpool/ROOT/debian/usr@snap2	=> rpool/ROOT/debian_snap2/usr
+			#       rpool/ROOT/debian/var@snap2	=> rpool/ROOT/debian_snap2/var
+			subfs="${s##$rootfs}"
+			subfs="${subfs%%@$snapname}"
+
+			destfs="${rootfs}_${snapname}" # base fs.
+			[ -n "$subfs" ] && destfs="${destfs}$subfs" # + sub fs.
+
+			# Get the mountpoint of the filesystem, to be used
+			# with clone_snap(). If legacy or none, then use
+			# the sub fs value.
+			mountpoint=$(get_fs_value "${s%%@*}" mountpoint)
+			if [ "$mountpoint" = "legacy" -o \
+			    "$mountpoint" = "none" ]
+			then
+				if [ -n "${subfs}" ]; then
+					mountpoint="${subfs}"
+				else
+					mountpoint="/"
+				fi
+			fi
+
+			# Clone the snapshot into its own
+			# filesystem
+			clone_snap "$s" "${destfs}" "${mountpoint}" || \
+			    retval=$((retval + 1))
+		fi
+	done
+
+	# If we haven't return yet, we have a problem...
+	return "${retval}"
+}
+
+# ================================================================
+
+# This is the main function.
+mountroot()
+{
+	local snaporig
+	local snapsub
+	local destfs
+
+	# ----------------------------------------------------------------
+	# I N I T I A L   S E T U P
+
+	# ------------
+	# Run the pre-mount scripts from /scripts/local-top.
+	pre_mountroot
+
+	# ------------
+	# Source the default setup variables.
+	[ -r '/etc/default/zfs' ] && . /etc/default/zfs
+
+	# ------------
+	# Support debug option
+	if grep -qiE '(^|[^\\](\\\\)* )(zfs_debug|zfs\.debug|zfsdebug)=(on|yes|1)( |$)' /proc/cmdline
+	then
+		ZFS_DEBUG=1
+		mkdir /var/log
+		#exec 2> /var/log/boot.debug
+		set -x
+	fi
+
+	# ------------
+	# Load ZFS module etc.
+	if ! load_module_initrd; then
+		disable_plymouth
+		echo ""
+		echo "Failed to load ZFS modules."
+		echo "Manually load the modules and exit."
+		/bin/sh
+	fi
+
+	# ------------
+	# Look for the cache file (if any).
+	[ ! -f ${ZPOOL_CACHE} ] && unset ZPOOL_CACHE
+
+	# ------------
+	# Compatibility: 'ROOT' is for Debian GNU/Linux (etc),
+	#		 'root' is for Redhat/Fedora (etc),
+	#		 'REAL_ROOT' is for Gentoo
+	if [ -z "$ROOT" ]
+	then
+		[ -n "$root" ] && ROOT=${root}
+
+		[ -n "$REAL_ROOT" ] && ROOT=${REAL_ROOT}
+	fi
+
+	# ------------
+	# Where to mount the root fs in the initrd - set outside this script
+	# Compatibility: 'rootmnt' is for Debian GNU/Linux (etc),
+	#		 'NEWROOT' is for RedHat/Fedora (etc),
+	#		 'NEW_ROOT' is for Gentoo
+	if [ -z "$rootmnt" ]
+	then
+		[ -n "$NEWROOT" ] && rootmnt=${NEWROOT}
+
+		[ -n "$NEW_ROOT" ] && rootmnt=${NEW_ROOT}
+	fi
+
+	# ------------
+	# No longer set in the defaults file, but it could have been set in
+	# get_pools() in some circumstances. If it's something, but not 'yes',
+	# it's no good to us.
+	[ -n "$USE_DISK_BY_ID" -a "$USE_DISK_BY_ID" != 'yes' ] && \
+	    unset USE_DISK_BY_ID
+
+	# ----------------------------------------------------------------
+	# P A R S E   C O M M A N D   L I N E   O P T I O N S
+
+	# This part is the really ugly part - there's so many options and permutations
+	# 'out there', and if we should make this the 'primary' source for ZFS initrd
+	# scripting, we need/should support them all.
+	#
+	# Supports the following kernel command line argument combinations
+	# (in this order - first match win):
+	#
+	#	rpool=<pool>			(tries to finds bootfs automatically)
+	#	bootfs=<pool>/<dataset>		(uses this for rpool - first part)
+	#	rpool=<pool> bootfs=<pool>/<dataset>
+	#	-B zfs-bootfs=<pool>/<fs>	(uses this for rpool - first part)
+	#	rpool=rpool			(default if none of the above is used)
+	#	root=<pool>/<dataset>		(uses this for rpool - first part)
+	#	root=ZFS=<pool>/<dataset>	(uses this for rpool - first part, without 'ZFS=')
+	#	root=zfs:AUTO			(tries to detect both pool and rootfs
+	#	root=zfs:<pool>/<dataset>	(uses this for rpool - first part, without 'zfs:')
+	#
+	# Option <dataset> could also be <snapshot>
+
+	# ------------
+	# Support force option
+	# In addition, setting one of zfs_force, zfs.force or zfsforce to
+	# 'yes', 'on' or '1' will make sure we force import the pool.
+	# This should (almost) never be needed, but it's here for
+	# completeness.
+	ZPOOL_FORCE=""
+	if grep -qiE '(^|[^\\](\\\\)* )(zfs_force|zfs\.force|zfsforce)=(on|yes|1)( |$)' /proc/cmdline
+	then
+		ZPOOL_FORCE="-f"
+	fi
+
+	# ------------
+	# Look for 'rpool' and 'bootfs' parameter
+	[ -n "$rpool" ] && ZFS_RPOOL="${rpool#rpool=}"
+	[ -n "$bootfs" ] && ZFS_BOOTFS="${bootfs#bootfs=}"
+
+	# ------------
+	# If we have 'ROOT' (see above), but not 'ZFS_BOOTFS', then use
+	# 'ROOT'
+	[ -n "$ROOT" -a -z "${ZFS_BOOTFS}" ] && ZFS_BOOTFS="$ROOT"
+
+	# ------------
+	# Check for the `-B zfs-bootfs=%s/%u,...` kind of parameter.
+	# NOTE: Only use the pool name and dataset. The rest is not
+	#       supported by ZoL (whatever it's for).
+	if [ -z "$ZFS_RPOOL" ]
+	then
+		# The ${zfs-bootfs} variable is set at the kernel commmand
+		# line, usually by GRUB, but it cannot be referenced here
+		# directly because bourne variable names cannot contain a
+		# hyphen.
+		#
+		# Reassign the variable by dumping the environment and
+		# stripping the zfs-bootfs= prefix.  Let the shell handle
+		# quoting through the eval command.
+		eval ZFS_RPOOL=$(set | sed -n -e 's,^zfs-bootfs=,,p')
+	fi
+
+	# ------------
+	# No root fs or pool specified - do auto detect.
+	if [ -z "$ZFS_RPOOL" -a -z "${ZFS_BOOTFS}" ]
+	then
+		# Do auto detect. Do this by 'cheating' - set 'root=zfs:AUTO'
+		# which will be caught later
+		ROOT=zfs:AUTO
+	fi
+
+	# ----------------------------------------------------------------
+	# F I N D   A N D   I M P O R T   C O R R E C T   P O O L
+
+	# ------------
+	if [ "$ROOT" = "zfs:AUTO" ]
+	then
+		# Try to detect both pool and root fs.
+
+		[ "$quiet" != "y" ] && \
+		    zfs_log_begin_msg "Attempting to import additional pools."
+
+		# Get a list of pools available for import
+		if [ -n "$ZFS_RPOOL" ]
+		then
+			# We've specified a pool - check only that
+			POOLS=$ZFS_RPOOL
+		else
+			POOLS=$(get_pools)
+		fi
+
+		OLD_IFS="$IFS" ; IFS=";"
+		for pool in $POOLS
+		do
+			[ -z "$pool" ] && continue
+
+			import_pool "$pool"
+			find_rootfs "$pool"
+		done
+		IFS="$OLD_IFS"
+
+		[ "$quiet" != "y" ] && zfs_log_end_msg $ZFS_ERROR
+	else
+		# No auto - use value from the command line option.
+
+		# Strip 'zfs:' and 'ZFS='.
+		ZFS_BOOTFS="${ROOT#*[:=]}"
+
+		# Stip everything after the first slash.
+		ZFS_RPOOL="${ZFS_BOOTFS%%/*}"
+	fi
+
+	# Import the pool (if not already done so in the AUTO check above).
+	if [ -n "$ZFS_RPOOL" -a -z "${POOL_IMPORTED}" ]
+	then
+		[ "$quiet" != "y" ] && \
+		    zfs_log_begin_msg "Importing ZFS root pool '$ZFS_RPOOL'"
+
+		import_pool "${ZFS_RPOOL}"
+		find_rootfs "${ZFS_RPOOL}"
+
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+
+	if [ -z "${POOL_IMPORTED}" ]
+	then
+		# No pool imported, this is serious!
+		disable_plymouth
+		echo ""
+		echo "Command: $ZFS_CMD"
+		echo "Message: $ZFS_STDERR"
+		echo "Error: $ZFS_ERROR"
+		echo ""
+		echo "No pool imported. Manually import the root pool"
+		echo "at the command prompt and then exit."
+		echo "Hint: Try:  zpool import -R ${rootmnt} -N ${ZFS_RPOOL}"
+		/bin/sh
+	fi
+
+	# ----------------------------------------------------------------
+	# P R E P A R E   R O O T   F I L E S Y S T E M
+
+	if [ -n "${ZFS_BOOTFS}" ]
+	then
+		# Booting from a snapshot?
+		# Will overwrite the ZFS_BOOTFS variable like so:
+		#   rpool/ROOT/debian@snap2 => rpool/ROOT/debian_snap2
+		echo "${ZFS_BOOTFS}" | grep -q '@' && \
+		    setup_snapshot_booting "${ZFS_BOOTFS}"
+	fi
+
+	if [ -z "${ZFS_BOOTFS}" ]
+	then
+		# Still nothing! Let the user sort this out.
+		disable_plymouth
+		echo ""
+		echo "Error: Unknown root filesystem - no 'bootfs' pool property and"
+		echo "       not specified on the kernel command line."
+		echo ""
+		echo "Manually mount the root filesystem on $rootmnt and then exit."
+		echo "Hint: Try:  mount -o zfsutil -t zfs ${ZFS_RPOOL-rpool}/ROOT/system $rootmnt"
+		/bin/sh
+	fi
+
+	# ----------------------------------------------------------------
+	# M O U N T   F I L E S Y S T E M S
+
+	# * Ideally, the root filesystem would be mounted like this:
+	#
+	#     zpool import -R "$rootmnt" -N "$ZFS_RPOOL"
+	#     zfs mount -o mountpoint=/ "${ZFS_BOOTFS}"
+	#
+	#   but the MOUNTPOINT prefix is preserved on descendent filesystem
+	#   after the pivot into the regular root, which later breaks things
+	#   like `zfs mount -a` and the /etc/mtab refresh.
+	#
+	# * Mount additional filesystems required
+	#   Such as /usr, /var, /usr/local etc.
+	#   NOTE: Mounted in the order specified in the
+	#         ZFS_INITRD_ADDITIONAL_DATASETS variable so take care!
+
+	# Go through the complete list (recursivly) of all filesystems below
+	# the real root dataset
+	filesystems=$("${ZFS}" list -oname -tfilesystem -H -r "${ZFS_BOOTFS}")
+	for fs in $filesystems $ZFS_INITRD_ADDITIONAL_DATASETS
+	do
+		mount_fs "$fs"
+	done
+
+	# ------------
+	# Debugging information
+	if [ -n "${ZFS_DEBUG}" ]
+	then
+		#exec 2>&1-
+
+		echo "DEBUG: imported pools:"
+		"${ZPOOL}" list -H
+		echo
+
+		echo "DEBUG: mounted ZFS filesystems:"
+		mount | grep zfs
+		echo
+
+		echo "=> waiting for ENTER before continuing because of 'zfsdebug=1'. "
+		echo -n "   'c' for shell, 'r' for reboot, 'ENTER' to continue. "
+		read b
+
+		[ "$b" = "c" ] && /bin/sh
+		[ "$b" = "r" ] && reboot -f
+
+		set +x
+	fi
+
+	# ------------
+	# Run local bottom script
+	if type run_scripts > /dev/null 2>&1 && \
+	    [ -f "/scripts/local-bottom" -o -d "/scripts/local-bottom" ]
+	then
+		[ "$quiet" != "y" ] && \
+		    zfs_log_begin_msg "Running /scripts/local-bottom"
+		run_scripts /scripts/local-bottom
+		[ "$quiet" != "y" ] && zfs_log_end_msg
+	fi
+}