-#/bin/bash
-
-set -x
+#!/bin/bash
+# Author: Carl Baldwin & Alan Pippin
+# Description: This script replicates a remote zfs filesystem to a local zfs pool.
+# This script will keep all snapshots in sync, removing the ones
+# that have been deleted since the last replicate was performed.
+# This script will only send the new, or missing, snapshots since
+# the last replicate was performed.
# Usage: replicate <hostname> <zfs filesystem>
+
+# source our configuration
+config="${0%/*}/zfs-scripts.conf"
+[ -e "${config}.dist" ] && . ${config}.dist
+[ -e "${config}" ] && . ${config}
+
+# command line arg parsing
remote=$1
remote_fs=$2
+remote_pool=${2%%/*}
+
+# Setup our cleanup and exit trap
+cleanup() {
+ if [[ -e "$local_list" ]]; then
+ rm -f $local_list
+ fi
+ if [[ -e "$remote_list" ]]; then
+ rm -f $remote_list
+ fi
+ $ssh $remote ls -d "$lockdir" > /dev/null 2>&1
+ if [[ $? == 0 ]]; then
+ $ssh $remote rm -rf "$lockdir"
+ fi
+}
+fatal_and_exit() {
+ echo -e 2>&1 "$1"
+ # Destroy the current backup markers from the local backup_pool and remote_pool if they exist
+ if [[ -n "$current_backup_marker" ]]; then
+ # Local backup pool current backup marker
+ $zfs list -t snapshot ${backup_pool}/${current_backup_marker} > /dev/null 2>&1
+ if [ $? == 0 ]; then
+ $zfs destroy ${backup_pool}/${current_backup_marker}
+ fi
+ # Remote pool current backup marker
+ $ssh $remote zfs list -t snapshot ${current_backup_marker} > /dev/null 2>&1
+ if [ $? == 0 ]; then
+ $ssh $remote $zfs destroy ${current_backup_marker}
+ fi
+ fi
+ # send email notification
+ if [[ -n "$2" ]]; then
+ echo -e "$1" | $mailx -s "zfs replicate on $hostname failed" "$2"
+ fi
+ # exit
+ exit 1
+}
+trap fatal_and_exit INT
+trap cleanup EXIT
-# change to match the name of the local backup pool
-local_pool=backpool
+# Declare a function to handle the replicate operation
+replicate() {
+ zfs_send="$1"
+ zfs_recv="$zfs receive -vF -d ${backup_pool}/${remote_pool}"
+ glue="$throttle $throttle_opt"
+ if [[ $throttle_enable == 1 && -e $throttle ]]; then
+ # handle using the glue in the local and remote host case properly
+ if [[ -z "$ssh" ]]; then
+ # local host glue case
+ $zfs_send | $glue | $zfs_recv
+ else
+ # remote host glue case
+ $ssh $remote "$zfs_send | $glue" | $zfs_recv
+ fi
+ else
+ # no glue case - works for both the local and remote host cases
+ $ssh $remote $zfs_send | $zfs_recv
+ fi
+ # The return code of the zfs_send | zfs_recv operation will be returned to the caller
+}
-# The ssh connection doesn't find zfs without this.
-zfs=/usr/sbin/zfs
+# Make sure we have valid arguments
+if [[ -z "$remote" ]] || [[ -z "$remote_fs" ]]; then
+ fatal_and_exit "Usage: $0 <hostname> <zfs filesystem>"
+fi
+
+# check for localhost
+if [[ $remote = "localhost" ]]; then
+ remote=""
+ ssh=""
+fi
+
+# Make sure the local backup pool and local receiving filesystem exist, or print some errors
+zpool list -H "$backup_pool" >/dev/null 2>&1
+if [ $? != 0 ]; then
+ fatal_and_exit "-E- The local backup pool on $hostname, '$backup_pool' doesn't seem to exist." $mailto
+fi
+zfs list "$backup_pool/$remote_pool" >/dev/null 2>&1
+if [ $? != 0 ]; then
+ echo >&2 "-I- The local filesystem for the remote pool, '$backup_pool/$remote_pool' doesn't seem to exist."
+ echo >&2 " Creating the local filesystem to receive the remote pool into: $backup_pool/$remote_pool"
+ $zfs create $backup_pool/$remote_pool
+ if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote $zfs on $hostname create command failed" $mailto
+ fi
+fi
+
+# Obtain the zpool guid for the local backup pool
+backup_pool_guid=`zpool get guid $backup_pool 2>&1 | grep $backup_pool | awk '{ print $3 }'`
+zpool get guid $backup_pool > /dev/null 2>&1
+if [ $? != 0 ]; then
+ fatal_and_exit "-E- Unable to extract the guid for the local backup pool on $hostname: $backup_pool" $mailto
+fi
+
+# Turn on shell verbosity
+set -x
+
+# Create the remote lockdir before continuing with the replicate
+# Spinlock on creating the lock
+maxsleeptime=60
+maxattempts=500
+attempts=0
+while true; do
+ $ssh $remote mkdir "$lockdir" >/dev/null 2>&1
+ if [ $? != 0 ]; then
+ # Another zfs admin tool is running.
+ # Wait a random amount of time and try again
+ ransleep=$(($RANDOM % $maxsleeptime))
+ sleep $ransleep
+ ((attempts=attempts+1))
+ else
+ # No other zfs admin tool is running, we can now.
+ break
+ fi
+ if [[ $attempts -gt $maxattempts ]]; then
+ # We've exceeded our maximum while loop count
+ echo "-E- The zfs filesystem has been locked down. Skipping replicate operation."
+ fail_msg=`$ssh $remote ls -ld $lockdir 2>&1`
+ fatal_and_exit "zfs-replicate-all on $hostname unable to obtain zfs admin lock:\n$fail_msg" $mailto
+ fi
+done
+
+# Setup our backup marker names
+current_backup_marker=${remote_fs}@current-backup-${backup_pool_guid}
+previous_backup_marker=${remote_fs}@previous-backup-${backup_pool_guid}
# List the snapshots on the remote machine.
remote_list=$(mktemp /tmp/replicate.XXXXXX)
-ssh $remote \
+$ssh $remote \
$zfs list -H -t snapshot |
grep ^${remote_fs}@ |
awk '{print$1}' > $remote_list
+if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote $zfs list on $hostname command failed" $mailto
+fi
# List the snapshots on the local machine.
+# Don't list the current backup marker if it exists on the local side.
+# If you do, it can mess up the common finding algorithm below.
local_list=$(mktemp /tmp/replicate.XXXXXX)
$zfs list -H -t snapshot |
- grep ^${local_pool}/${remote_fs}@ |
- awk '{gsub(/^${local_pool}./,"",$1); print$1}' > $local_list
+ grep ^${backup_pool}/${remote_fs}@ |
+ grep -v ^${backup_pool}/${current_backup_marker} |
+ awk "{gsub(/^$backup_pool./,\"\",\$1); print\$1}" > $local_list
+if [ $? != 0 ]; then
+ fatal_and_exit "-E- local $zfs list on $hostname command failed" $mailto
+fi
+
+# Destroy the current backup marker snapshot on the remote system if it exists
+grep -q ${current_backup_marker} $remote_list
+if [ $? == 0 ]; then
+ $ssh $remote $zfs destroy ${current_backup_marker}
+ if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote $zfs destroy on $hostname command failed" $mailto
+ fi
+fi
+
+# Create the current backup marker snapshot on the remote system
+$ssh $remote $zfs snapshot ${current_backup_marker}
+if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote $zfs snapshot on $hostname command failed" $mailto
+fi
+
+# Check to see if the previous backup marker exists in the remote snapshot list.
+# Check to see if the previous backup marker exists in the local snapshot list.
+# If the previous backup markers exists, perform an incremental replicate. Else:
+# 1) check to see if a common snapshot exists, and perform an incremental replicate.
+# 2) if no common snapshot exists, destroy the local filesystem, and perform a full replicate.
+grep -q ${previous_backup_marker} $remote_list
+no_markers=$?
+grep -q ${previous_backup_marker} $local_list
+no_markers=$(($no_markers || $?))
+
+if [ $no_markers == 0 ]; then
+ # We found backup markers, incrementally send the new snaps
+
+ # First, rollback the local backup pool to the previous backup marker in case the previous
+ # backup was interrupted for some reason. If we don't do this, the zfs send -R command
+ # below may complain about snaps already existing as it tries to resend from the
+ # previous backup marker again from a previously interrupted replicate.
+ $zfs rollback -r ${backup_pool}/${previous_backup_marker}
+ if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote incremental $zfs rollback command failed on $hostname" $mailto
+ fi
+ # Now it should be safe to send the snaps
+ replicate "$zfs send -R -I${previous_backup_marker} ${current_backup_marker}"
+ if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote incremental $zfs send command failed on $hostname" $mailto
+ fi
+else
+ # We didn't find any backup markers, next check to see if we have a common snapshot.
-# See what the most recent snapshot on the remote end is.
-latest=$(tail -n 1 $remote_list)
+ # See what the most recent snapshot on the remote end is.
+ latest=$(tail -n 1 $remote_list)
-# I did this to make sure that diff would always display the most recent common
-echo bogus.remote >> $remote_list
-echo bogus.local >> $local_list
-common=$(diff -u $remote_list $local_list | grep '^ ' | tail -n 1)
+ # I did this to make sure that diff would always display the most recent common
+ # Since we're keying off the context of the diff, we need to ensure we will get context
+ # by injecting a known difference in case no others exist in the lists.
+ echo bogus.remote >> $remote_list
+ echo bogus.local >> $local_list
+ common=$(diff -u $remote_list $local_list | grep '^ ' | tail -n 1)
-ssh $remote $zfs send -R -I${common/*@/@} $latest |
- $zfs receive -vF -d ${local_pool}/${remote_fs%/*}
+ if [[ -n "$common" ]]; then
+ # We found a common snapshot, incrementally send the new snaps
+ replicate "$zfs send -R -I${common/*@/@} ${current_backup_marker}"
+ if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote incremental $zfs send command failed on $hostname" $mailto
+ fi
+ else
+ # We did not find any markers or a common snapshot
+ # At this point, we'll have to send the entire filesystem
+ # Destroy the local filesystem if it exists before receving the full replicate
+ zfs list ${backup_pool}/${remote_fs} > /dev/null 2>&1
+ if [ $? == 0 ]; then
+ if [[ $destroy_local_filesystem_on_full_replicate == 1 ]]; then
+ $zfs destroy -r ${backup_pool}/${remote_fs}
+ if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote full $zfs destroy command failed on $hostname" $mailto
+ fi
+ else
+ echo "-W- We need to destroy a local filesystem before receiving a full stream."
+ echo " However, since the option is set to prevent this, skipping replicate operation."
+ fatal_and_exit "unable to destroy local filesystem:\n$zfs destroy -r ${backup_pool}/${remote_fs} not able to run on $hostname" $mailto
+ fi
+ fi
+ # Send the full filesystem
+ replicate "$zfs send -R ${current_backup_marker}"
+ if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote full $zfs send command failed on $hostname" $mailto
+ fi
+ fi
+fi
+
+# destroy the previous backup markers now that we've replicated past them
+# don't check the return codes here because these may not exist, and that is ok
+$zfs destroy ${backup_pool}/${previous_backup_marker} > /dev/null 2>&1
+$ssh $remote $zfs destroy ${previous_backup_marker} > /dev/null 2>&1
-rm -f $local_list $remote_list
+# Rename the current backup marker to be the previous backup marker
+$zfs rename ${backup_pool}/${current_backup_marker} ${backup_pool}/${previous_backup_marker}
+if [ $? != 0 ]; then
+ fatal_and_exit "-E- local $zfs rename command failed on $hostname" $mailto
+fi
+$ssh $remote $zfs rename ${current_backup_marker} ${previous_backup_marker}
+if [ $? != 0 ]; then
+ fatal_and_exit "-E- remote $zfs rename command failed on $hostname" $mailto
+fi