zrep_snap

################ File: zrep_snap
# be sure to have included zrep_vars

# This file contains routines related to 
# "make new snapshot, using next sequence number".
# So it thus includes all snap sequence related routines
# It may contain "sync snapshot" related routines for now.
# It also is definitive for the format of snapshot names
# It also contains most "query status of snaps" type routines,
#  such as "getlastsnapsent"
#
# Normal style for making a snapshot and syncing it:
#   1.  create a snapshot.
#   2.  sync it over
#   3.  set "zrep:sent" on *snapshot*, with timestamp in seconds
# Old-nasty-zfs compat mode:
#   Step 3. Add/update "zrep:lastsent->snapname", and 
#           "zrep:lastsenttime->timestamp",  on *filesystem*
#
######################################################################


#pass in a zrep ZFS snapshot name. strip our our sequence number and echo in back
_getseqnum(){
	echo "$1" | sed 's/.*@'${ZREPTAG}'_\(......\).*/\1/'
}

# By observation, 'zfs list' shows snapshots order of creation.
# last listed, should be last in sequence.
# But, dont take chances!!
getlastsequence(){
	typeset lastval
	#remember, filesystems can have '_' in them
	_getseqnum `getlastsnap $1`
	
}

# prints out last snapshot zrep created, going purely by sequence.
# Note: "last created", which may or may NOT be "last successfully synced". 
# This is basically "getallsnaps |tail -1"
getlastsnap(){
	zfs list -t snapshot -H -o name $DEPTHCAP -r $1 |
	   sed -n "/@${ZREPTAG}_[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]/"p |
	   sort | tail -1
}

# Usage:   getlastsnapsent zpool/FSNAME
#   note to self: in modern zrep, ALL successfuly synced snapshots have
#   a timestamp value,   zrep:sent=TIMESTAMPHERE
#   Thats why the sort and tail is neccessary
getlastsnapsent(){
	# arg.  more efficient if we can just return value directly,
	# but i'm using backwards compat :(
	typeset lastsent
	lastsent=`zfs get  -H -o name -r -s local ${ZREPTAG}:sent $1 |
	   sort | tail -1`
	if [[ "$lastsent" != "" ]] ; then
		echo $lastsent
		return
	fi

	# Fallback method, for backwards compat with older ZFS code,
	# since it cant set properties on snapshots
	zfs get  -H -o value -s local ${ZREPTAG}:lastsent $1
}

# HORRIBLY DANGEROUS
# ONLY should be used by zrep_sentsync.
# Kind of the inverse to getlastsnapsent. But should NOT use "fallback method" in getlastsnapsent.
# This is only for zrep_sentsync, which only uses newer method.
_clearlast(){
	typeset lastsent
	lastsent=`zfs get  -H -o name -r -s local ${ZREPTAG}:sent $1|tail -1`

	while [[ "$lastsent" != "" ]] ; do
		_errprint WARNING: clearing sent value from  $lastsent
		zfs inherit ${ZREPTAG}:sent $lastsent
		lastsent=`zfs get  -H -o name -r -s local ${ZREPTAG}:sent $1|tail -1`
	done	

}

# return a number, which is the difference between the lastsnapshot counter, and
# the last successfully synced snapshot counter.
# In theory, can only be positive.
# 
getunsentcount(){
	typeset lastsynced lastsnap lastsyncedseq lastsnapseq
	lastsynced=`getlastsnapsent $1`
	lastsnap=`getlastsnap $1`

	if [[ "$lastsynced" == "$lastsnap" ]] ; then
		return 1
	fi

	lastsyncedseq=`getseqnum $lastsynced`
	lastsnapseq=`getseqnum $lastsnap`

	lastsyncedseq=$((16#$lastsyncedseq))
	lastsnapseq=$((16#$lastsnapseq))


	if [[ "$lastsyncedseq -lt 1 ]] ; then
		zrep_errquit Error: cannot parse $lastsynced
	fi
	if [[ "$lastsnapseq -lt 1 ]] ; then
		zrep_errquit Error: cannot parse $lastsnap
	fi

	echo $(( $lastsnapseq - $lastsyncedseq ))

}

# outputs time in seconds, of when the last successful sync for the
# filesystem was done. (in format compatible with zrep_gettimeinseconds() )
#  Note that this is time of actual sync, not snapshot creation time.
#
# This unfortunately needs to be compatible with both new way, and
# old-nasty-hack-way
# 
# In future, may take optional argument of which HOST to check 
# sync with. But since I currently only suport one host per fs... oh well.
# If never synced, will return 1, and echo ""
#
getlastsynctime(){
	typeset fs lastsent senttime

	if [[ "$1" == "" ]] ; then
		zrep_errquit Internal error: no arg to getlastsynctime
	fi
	fs="$1"

	# Deal with possibly upgraded system;
	#   Check "lastsent", only as fallback.

	# copy from getlastsnapsent, but only using newest method
	lastsent=`zfs get  -H -o name -r -s local ${ZREPTAG}:sent $fs |
	   sort | tail -1`
	senttime=`zfs get  -H -o value ${ZREPTAG}:sent $lastsent`
	if [[ "$senttime" != "" ]] ; then echo $senttime ; return 0; fi

	# ooops. try fallback to nasty old zfs-compat style
	senttime=`zfs get -H -o value ${ZREPTAG}:lastsent $fs`
	echo $senttime
	if [[ "$senttime" != "" ]] ; then return 0; fi
	return 1
}

#This is for synctosnap, and also zrep_expire
getallsnaps(){
	zfs list -t snapshot -H -o name $DEPTHCAP -r $1 |
	   sed -n "/@${ZREPTAG}_[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]/"p |
	   sort
}

# list all snapshots of the given filesystem, that are made by this prog
# arg: fs
list_autosnaps(){
	if [[ "$1" == "" ]] ; then
		zrep_errquit "zrep internalerror: no arg for list_autosnaps"
	fi
 	zfs list $DEPTHCAP -r -H -o name -t snapshot $1 |
		grep "@${ZREPTAG}_[0-9a-f][0-9a-f]"
	# Make sure this format matches other routines in here
	# Okay to just check first few digits though
}

# User entrypoint. Part of pair: snaponly, sendonly
# Just makes snapshot. 
zrep_snaponly(){
	typeset srcfs

	if [[ "$1" == "all" ]] ; then
		set -- `zrep_list_master`
		if [[ "$1" == "" ]] ; then
			exit
		fi
	fi

	while [[ "$1" != "" ]] ; do
		srcfs="$1" ;shift
		zrep_lock_fs $srcfs
		if [[ $? -ne 0 ]] ; then
			# this function is supposed to be coordinated by user
			# therefore, if something else is competing,
			# coordination has failed. no retry.
			zrep_errquit zrep snaponly failed for $srcfs: cannot get lock
			
		fi

		makesnap $srcfs ||zrep_errquit snaponly for $srcfs failed

		zrep_unlock_fs $srcfs
		
	done	
}

# 
# creates next snapshot in sequence
# consider holding lock here
# Caller must have zrep lock on filesystem:
#   we verify with zrep_has_fs_lock
makesnap(){
	typeset  check oldseq newseq="" newseqX newsnap

	#sanity checks first!
	check="`$ZFSGETVAL ${ZREPTAG}:src-host $1`"
	if [[ "$check" != "$Z_LOCAL_HOST" ]] ; then
		_errprint ERROR: we are not master host for $1
		_errprint master is $check, we are $Z_LOCAL_HOST
		exit 1
	fi

	zrep_has_fs_lock $1
	if [[ $? -ne 0 ]] ; then
		_errprint Internal error:  makesnap fail, no lock on $1
		exit 1
	fi

	oldseq=`getlastsequence $1`
	# This means input is base 16
	newseq=$((16#$oldseq))

	newseqX=$(printf "%.6x" $(($newseq + 1))  )


	#_errprint DEBUG old=$oldseq new=$newseqX
	newsnap="$1@${ZREPTAG}_$newseqX"
	
	zfs snapshot $Z_SNAP_R $newsnap
	if [[ $? -eq 0 ]] ; then
		echo -n $newsnap; return 0
	else
		return 1
	fi
	
}

## This is the implentation for the "zrep clear" command
## Purpose is to remove all zrep related hooks from a local filesystem. 
##  (NOT delete it)
## Will remove zrep snapshots and zfs zrep: properties
zrep_clear(){

	if [[ "$1" == "" ]] ; then
		zrep_errquit clear command requires an argument
	fi

	echo "WARNING: Removing all zrep configs and snapshots from $1"
	echo " (for TAG=${ZREPTAG})"

	
	if [[ "$ZREP_FORCE" != "-f" ]] ; then
		echo Continuing in 10 seconds
		sleep 10
	fi

	_clearsnaps $1
	_clearvars $1
}

_clearsnaps(){
	echo Destroying any zrep-related snapshots from $1
	snaplist=`list_autosnaps $1`
	for snap in $snaplist ; do
		zfs destroy -r $snap
	done
}

_clearvars(){
	echo Removing zrep-related properties from $1
	proplist=`zfs get -H -o property all $1|grep ${ZREPTAG}:`
	for prop in $proplist ; do
		zfs inherit $prop $1
	done
}

## This is a special internal routine, used only by zrep_init, 
## to reset target fs to pre-zrep state.
## call with "srcfs  errmsg1 errmsg2..."
## It will also REMOVE REMOTEFS if set in PROPERTIES!!
clearquit(){
	remhost=`$ZFSGETVAL ${ZREPTAG}:dest-host $1`
	remfs=`$ZFSGETVAL ${ZREPTAG}:dest-fs $1`

	if [[ $? -eq 0 ]] && [[ "$remhost" != "-" ]] && [[ "$remfs" != "-" ]]; then
		zrep_ssh $remhost zfs destroy -r $remfs
	fi
	_clearsnaps $1
	_clearvars $1

	shift
	zrep_errquit "$@"
}


# Shared internal routine.
#
# Set the to/from properties on a fs for zrep
# Called by zrep_init  and zrep_changeconfig
# Usage: 
#  setfsconfigs srcfs desthost destfs
#  setfsconfigs -d destfs srchost srcfs
#
setfsconfigs(){
	typeset srchost srcfs desthost destfs fsname

	if [[ "$1" == "-d" ]] ; then
		srchost="$3"
		srcfs="$4"
		desthost=$Z_LOCAL_HOST
		destfs="$2"
		fsname=${destfs}
	else
		srchost=$Z_LOCAL_HOST
		srcfs="$1"
		desthost="$2"
		destfs="$3"
		fsname=${srcfs}
	fi


	if [[ "$destfs" == "" ]] ; then
		zrep_errquit "zrep: no dest fs specified"
	fi
	zfs list $fsname >/dev/null ||zrep_errquit "filesystem $fsname must exist already"

	#
	# This is really only for when we are called from init I think?
	#
	case $destfs in
		# Originally, I had this passthrough only if fsname was at end
		# However,we must allow destfs to have different leaf name,
		# for circumstances such as replication to same host
		*/*)
			:
			;;
		*)
			# Only the pool name given.
			# With our curent workflow, that can ONLY WORK
			# if recent versions of ZFS are in use.
			if (( Z_HAS_REC_U )) ; then
				echo WARNING: forcing override sync to top level pool $desthost:$destfs
				echo Not using a usual sub-filesystem
				sleep 5
			else
				zrep_errquit older zfs version, cannot initialize top level pool
			fi
			;;
	esac

	zfs set ${ZREPTAG}:src-fs=$srcfs		$fsname
	zfs set ${ZREPTAG}:src-host=$srchost		$fsname
	zfs set ${ZREPTAG}:dest-fs=$destfs		$fsname
	zfs set ${ZREPTAG}:dest-host=$desthost		$fsname
	zfs set ${ZREPTAG}:savecount=$ZREP_SAVE_COUNT	$fsname
}


#
# Old way used to use recv -x, but not all systems have that. so
# preferred method is now to use recv -u instead.
# To init from existing snapshot instead, see documention at bolthole.com
# Note that remote fs must share same stem name as source. (for now?)
#
zrep_init(){
	typeset srcfs="$1" desthost="$2" destfs="$3" snap check vol=0
	typeset mountpoint vflags verbose
	typeset token

	if [[ "$ZREP_FORCE" != "-f" ]] ; then
		zrep_ssh $desthost zfs list $destfs 2>/dev/null &&
			zrep_errquit "$desthost:$destfs already exists! Will not overwrite without -f"
	fi

	if [[ "$ZREP_VERBOSE" != "" ]] ; then
		verbose=-v
	fi

	if [[ "$srcfs" == "" ]] ; then
		zrep_errquit "zrep: no fs specified"
	fi

	#sanity checks
	check="`$ZFSGETVAL ${ZREPTAG}:dest-fs $srcfs`"
	if [[ "$check" != "-" ]] ; then
		echo "$srcfs is at least partially configured by zrep"
		check="`$ZFSGETLVAL ${ZREPTAG}:master $srcfs`"
		if [[ "$check" != "" ]] ; then
			zrep_errquit "${ZREPTAG}:master detected!!"
		fi

		if [[ "$ZREP_RESUME" == "" ]] ; then
			zrep_errquit "To re-initialize, first use zrep clear $srcfs"
		fi

		# if it exists already, presume we need to resume an init.
		# if we cant, then its an error.

		
		
		token=`zrep_ssh $desthost $ZFSGETVAL receive_resume_token $destfs`
		if [[ "$token" == "-" ]] ; then token="" ; fi
		if [[ "$token" == "" ]] ; then
			zrep_errquit Partial init of $src detected but no resume token found. Suggest you zrep clear and start again
		fi

		echo "Partially complete init detected. Attempting to resume send"

		if [[ "$BBCP" != "" ]] ; then
			SENDCMD="zfs send -t $token ${ZREP_RESUME_FLAGS}"
			$BBCP -N io "$SENDCMD" \
			   "$desthost:zfs recv $destfs"

		else
			eval zfs send ${ZREP_RESUME_FLAGS} -t $token  ${Z_F_OUT} | 
			   zrep_ssh $desthost "${Z_F_IN} zfs recv $destfs"
		fi		
		if [[ $? -ne 0 ]] ; then
			zrep_errquit resume send of zrep init $srcfs failed
		fi
	fi


	check="`$ZFSGETVAL type $srcfs`"
	if [[ "$check" == "volume" ]] ; then
		vol=1
		if (( ! Z_HAS_REC_O )) ; then
			echo WARNING: no proper recv -o detected
			echo WARNING: extremely old versions of ZFS crash with volume init
			echo Continuing in 5 seconds....
			sleep 5
			vflags="`$ZFSGETVAL volsize $srcfs`"
			vflags="-V $vflags"


		fi
		# for details,see
		#  https://groups.google.com/forum/#!topic/comp.unix.solaris/-5bcZFInozk
		# subject:"solaris 11 crash when zfs send/receive of volume"
	fi
	

	#get this for later
	mountpoint=`$ZFSGETLVAL mountpoint $srcfs`

	#  Make this section conditional, so that we keep shared codepath for completion
	# of initial sync.
	# This is actually  the "normal" snapshot and sync path.
	# 
	if [[ "$token" == "" ]] ; then
	
		if [[ "$ZREP_RESUME" != "" ]] ; then
			recv_s=-s
		fi

		echo Setting zrep properties on $srcfs
		setfsconfigs $srcfs $desthost $destfs

		#setfsconfigs may do some "smarts" to adjust value, so get it again.
		# yes only check for LOCAL this time. Paranoia....
		destfs=`$ZFSGETLVAL ${ZREPTAG}:dest-fs $srcfs`

		if (( Z_HAS_REC_O )) ; then
			READONLYPROP="-o readonly=on"
		else
			READONLYPROP=""
			echo Warning: zfs recv lacking -o readonly
		fi
		if [[ "$ZREP_CREATE_FLAGS" != "" ]]  || [[ "$READONLYPROP" == "" ]] ; then
			echo Creating  destination filesystem as separate step
			# normally would want to use -o readonly here.
			# however, that breaks when -R is used.
			# set it after transfer instead
			zrep_ssh $desthost zfs create $ZREP_CREATE_FLAGS $vflags $destfs || zrep_errquit "Cannot create $desthost:$destfs"
		fi

		snap="${srcfs}@${ZREPTAG}_000000"
		echo Creating snapshot $snap
		zfs snapshot $Z_SNAP_R $snap || clearquit $srcfs "Cannot create initial snapshot $snap"

		# Note that we may not want to use -p for normal zrep syncs
		# We also should not use -F for normal recv. See workflow.txt
		# Note: we may have to reset readonly=on, if we used -p on send...
		#
		echo Sending initial replication stream to $desthost:$destfs
		if (( Z_HAS_REC_U )) ; then
			# This is the nice, clean, modern codepath, to send
			# zrep settings over automatically at first init.
			# Note that we use "zfs send -p" to preserve properties.

			if [[ "$BBCP" != "" ]] ; then
			$BBCP -N io "zfs send $verbose ${ZREP_R} ${ZREP_SEND_FLAGS} -p $snap" \
			    "$desthost:zfs recv -u  $READONLYPROP $recv_s -F $destfs"
			else
			eval zfs send $verbose ${ZREP_R} ${ZREP_SEND_FLAGS} -p $snap ${Z_F_OUT}|
				zrep_ssh $desthost "${Z_F_IN} zfs recv -u $READONLYPROP $recv_s -F $destfs"
			fi

		else
			## arg.. Update your systems!!
			# without -u, risky to use send -p if mountpoint set.
			#  (So we cant avoid it potentially WRONGLY setting mountpoint)
			# This means we have to manually set props lower down as well.
			# (yeah okay I could check if $mountpoint set, but
			#  there are too many code branches here already, so too bad!
			#  Update your system! :p )

			if [[ "$BBCP" != "" ]] ; then
			$BBCP -N io "zfs send ${ZREP_R} ${ZREP_SEND_FLAGS} $snap" \
			    "$desthost:zfs recv $READONLYPROP $recv_s -F $destfs"
			else
			eval zfs send ${ZREP_R} ${ZREP_SEND_FLAGS} $snap ${Z_F_OUT}|
			  zrep_ssh $desthost "${Z_F_IN} zfs recv $READONLYPROP $recv_s -F $destfs"
			fi
		fi
		if [[ $? -ne 0 ]] ; then
			if [[ "$ZREP_RESUME" == "" ]] ; then
				clearquit $srcfs "Error transferring $snap to $desthost:$destfs. Resetting"
			else
				zrep_errquit "Error transferring $snap to $desthost:$destfs. RESUME set. Not clearing $srcfs"
			fi
		fi

	fi   #resume token set


	# Successful initial sync! Woo! okay, record that fact and complete setting properties
	# ... after stupid old-zfs-compat junk, that is
	if (( ! Z_HAS_REC_U )) || [[ "$ZREP_CREATE_FLAGS" != "" ]] ; then
		_debugprint Because your zfs does not have recv -u,
		_debugprint   or maybe because ZREP_CREATE_FLAGS set,
		_debugprint setting remote properties by hand
		zrep_ssh $desthost zfs set readonly=on $destfs
		if [[ $? -ne 0 ]] ; then
			errquit Could not set readonly for $desthost:$destfs. However, snapshot has been synced. \
			Manual intervention required if you do not wish to simply clear and start again.
		fi

		zrep_ssh $desthost zfs set ${ZREPTAG}:src-fs=$srcfs $destfs
		zrep_ssh $desthost zfs set ${ZREPTAG}:src-host=$Z_LOCAL_HOST $destfs
		zrep_ssh $desthost zfs set ${ZREPTAG}:dest-fs=$destfs $destfs
		zrep_ssh $desthost zfs set ${ZREPTAG}:dest-host=$desthost $destfs
		zrep_ssh $desthost zfs set ${ZREPTAG}:savecount=$ZREP_SAVE_COUNT $destfs

	fi
	if [[ "$ZREP_CREATE_FLAGS" != "" ]]  || [[ "$READONLYPROP" == "" ]] ; then
		echo setting readonly on $desthost:$destfs manually
		zrep_ssh $desthost zfs set readonly=on $destfs
	fi

	# Extra, non-symmeric properties the user might want on the remote side.
	if [[ "$ZREP_INIT_REMOTE_PROPERTIES" != "" ]] ; then
		for prop in $ZREP_INIT_REMOTE_PROPERTIES ; do
			zrep_ssh $desthost zfs set $prop $destfs
		done
	fi


	# Success! So need to set success marker on remote side.
	# Make sure to set format to match what zrep_sync() looks for!
	if (( Z_HAS_SNAPPROPS )) ; then
		typeset sentprop="${ZREPTAG}:sent=`zrep_gettimeinseconds`"
		zfs set $sentprop ${snap}
	else
		# Arg stupidold stuff cant set props on a snapshot
		# So we have to manually set these on both sides also,
		# "Just in case"
		zfs set ${ZREPTAG}:lastsent=${snap} $srcfs
		zrep_ssh $desthost zfs set ${ZREPTAG}:lastsent=${snap} $destfs
	fi
	if [[ "$mountpoint" != "" ]] ; then
		echo "clearing mountpoint value for remote"
		zrep_ssh $desthost zfs inherit mountpoint $destfs
	fi

	# make sure the above ' set 's (sent, lastsent)
	# match what zrep_sync() does !!!


	# Note: we have to set master property NOW, not before,
	# because "recv -x zrep:master" Does Not Work properly
	# Also, it avoids things like "zrep sync all" from attempting
	# to sync it before initial sync has been done.
	# We don't even have to zrep_lock_fs until this is set
	#  Make sure value this matches zrep_sentsync
	zfs set ${ZREPTAG}:master=yes	$srcfs

	echo Initialization copy of $srcfs to $desthost:$destfs complete
	if (( Z_HAS_REC_U )) ; then
		echo Filesystem will not be mounted
	fi

}


zrep_changeconfig(){
	if [[ "$1" == "-f" ]] ; then
		# skip safety checks
		shift
		setfsconfigs $@
		return 
	fi
	typeset srcfs="$1" check
	
	if [[ "$srcfs" == "" ]] ; then
		zrep_errquit "zrep: no fs specified"
	fi

	check=`getlastsnap $srcfs`
	if [[ "$check" == "" ]] ; then
		_errprint "No pre-existing zrep snapshots found on $srcfs"
		_errprint $srcfs is not initialized for zrep. cannot change config.
		zrep_errquit Use zrep init on $srcfs instead
	fi

	setfsconfigs $@
	
}