X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Flustre-rsync-test.sh;h=b16979ec34870cdc7083a9ec70fb3eb70af9b0e2;hp=49918ae0748153139729aea1123268704b93230e;hb=HEAD;hpb=2a7718f16d3eca0a0b4abff4c74d0fcb680df191

diff --git a/lustre/tests/lustre-rsync-test.sh b/lustre/tests/lustre-rsync-test.sh
index 49918ae..9e5720d 100644
--- a/lustre/tests/lustre-rsync-test.sh
+++ b/lustre/tests/lustre-rsync-test.sh
@@ -1,87 +1,83 @@
 #!/bin/bash
-# -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
-# vim:shiftwidth=4:softtabstop=4:tabstop=4:
 #
 # Run select tests by setting ONLY, or as arguments to the script.
 # Skip specific tests by setting EXCEPT.
 #
-# Run test by setting NOSETUP=true when ltest has setup env for us
 set -e
 
-SRCDIR=`dirname $0`
-export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/../utils:$PATH:/sbin
-
 ONLY=${ONLY:-"$*"}
-[ -n "$ONLY" ] && SLOW=yes
-ALWAYS_EXCEPT="$LRSYNC_EXCEPT"
-# bug number for skipped test:
-# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-
-[ "$ALWAYS_EXCEPT$EXCEPT" ] && \
-        echo "Skipping tests: `echo $ALWAYS_EXCEPT $EXCEPT`"
 
 KILL=/bin/kill
-
-TMP=${TMP:-/tmp}
 LREPL_LOG=$TMP/lustre_rsync.log
 ORIG_PWD=${PWD}
 
-LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
+LUSTRE=${LUSTRE:-$(dirname $0)/..}
 . $LUSTRE/tests/test-framework.sh
-init_test_env $@
-. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_test_env "$@"
 init_logging
 
+ALWAYS_EXCEPT="$LRSYNC_EXCEPT "
+# bug number for skipped test: LU-4256
+ALWAYS_EXCEPT+="               2b"
+# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
+
+build_test_filter
+
+[ -n "$FILESET" ] && skip "Not functional for FILESET set"
+
 check_and_setup_lustre
 
 DIR=${DIR:-$MOUNT}
 assert_DIR
 
-
-build_test_filter
+if getent group nobody; then
+	GROUP=nobody
+elif getent group nogroup; then
+	GROUP=nogroup
+else
+	error "No generic nobody group"
+fi
 
 export LRSYNC=${LRSYNC:-"$LUSTRE/utils/lustre_rsync"}
 [ ! -f "$LRSYNC" ] && export LRSYNC=$(which lustre_rsync)
 export LRSYNC="$LRSYNC -v -c no -d 2"
 
-# control the time of tests
-DBENCH_TIME=${DBENCH_TIME:-60}  # No of seconds to run dbench
+# Number of seconds to run dbench
+DBENCH_TIME=${DBENCH_TIME:-60}
 TGT=$TMP/target
 TGT2=$TMP/target2
-MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
-    awk '{gsub(/_UUID/,""); print $1}' | head -1)
+MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
+	awk '{ gsub(/_UUID/,""); print $1 }' | head -n1)
 
 init_changelog() {
-    CL_USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
-    echo $MDT0: Registered changelog user $CL_USER
-    CL_USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
-	mdd.$MDT0.changelog_users | wc -l) - 2 ))
-    [ $CL_USERS -ne 1 ] && \
-	echo "Other changelog users present ($CL_USERS)"
+	changelog_register || error "changelog_register failed"
+	CL_USER=(${CL_USERS[$SINGLEMDS]})
+	echo $MDT0: Registered changelog user $CL_USER
+	[ -z $CL_USER ] &&
+		echo "No changelog users present on $SINGLEMDS"
 }
 
-init_src() {
-    rm -rf $TGT/$tdir $TGT/d*.lustre_rsync-test 2> /dev/null
-    rm -rf $TGT2/$tdir $TGT2/d*.lustre_rsync-test 2> /dev/null
-    rm -rf ${DIR}/$tdir $DIR/d*.lustre_rsync-test ${DIR}/tgt 2> /dev/null
-    rm -f $LREPL_LOG
-    mkdir -p ${DIR}/$tdir
-    mkdir -p ${TGT}/$tdir
-    mkdir -p ${TGT2}/$tdir
-    if [ $? -ne 0 ]; then
-        error "Failed to create target: " $TGT
-    fi
+fini_changelog() {
+	changelog_clear
+	changelog_deregister
 }
 
-cleanup_src_tgt() {
-    rm -rf $TGT/$tdir
-    rm -rf $DIR/$tdir
-    rm -rf $DIR/tgt
+init_src() {
+	rm -rf $TGT/$tdir $TGT/d*.lustre_rsync-test 2> /dev/null
+	rm -rf $TGT2/$tdir $TGT2/d*.lustre_rsync-test 2> /dev/null
+	rm -rf ${DIR}/$tdir $DIR/d*.lustre_rsync-test ${DIR}/tgt 2> /dev/null
+	rm -f $LREPL_LOG
+
+	mkdir_on_mdt0 $DIR/$tdir || error "Failed to create target: " $DIR/$tdir
+	mkdir -p ${TGT}/$tdir || error "Failed to create target: " $TGT/$tdir
+	mkdir -p ${TGT2}/$tdir || error "Failed to create target: " $TGT2/$tdir
 }
 
-fini_changelog() {
-    $LFS changelog_clear $MDT0 $CL_USER 0
-    do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $CL_USER
+cleanup_src_tgt() {
+	[ ! -d $TGT2/$tdir ] || rm -rf $TGT2/$tdir # /tmp/target2/...
+	[ ! -d $TGT/$tdir ] || rm -rf $TGT/$tdir # /tmp/target/...
+	rm -rf $DIR/$tdir  # /mnt/lustre/...
+	rm -rf $DIR/tgt    # /mnt/lustre/tgt
 }
 
 # Check whether the filesystem supports xattr or not.
@@ -90,21 +86,22 @@ fini_changelog() {
 # "small" - large xattr is unsupported but small xattr is supported
 # "no"    - xattr is unsupported
 check_xattr() {
-    local tgt=$1
-    local xattr="no"
+	local tgt=$1
+	local xattr="no"
 
-    touch $tgt
+	touch $tgt
 
-    local val="$(generate_string $(max_xattr_size))"
-    if large_xattr_enabled &&
-       setfattr -n user.foo -v $val $tgt 2>/dev/null; then
-            xattr="large"
-    else
-        setfattr -n user.foo -v bar $tgt 2>/dev/null && xattr="small"
-    fi
+	local val="$(generate_string $(max_xattr_size))"
+
+	if large_xattr_enabled &&
+		setfattr -n user.foo -v $val $tgt 2>/dev/null; then
+		xattr="large"
+	else
+		setfattr -n user.foo -v bar $tgt 2>/dev/null && xattr="small"
+	fi
 
-    rm -f $tgt
-    echo $xattr
+	rm -f $tgt
+	echo $xattr
 }
 
 check_diff() {
@@ -120,40 +117,72 @@ check_diff() {
 	fi
 }
 
-# Test 1 - test basic operations
-test_1() {
-    init_src
-    init_changelog
-    local xattr=$(check_xattr $TGT/foo)
-
-    # Directory create
-    mkdir $DIR/$tdir/d1
-    mkdir $DIR/$tdir/d2
-
-    # File create
-    touch $DIR/$tdir/file1
-    cp /etc/hosts  $DIR/$tdir/d1/
-    touch  $DIR/$tdir/d1/"space in filename"
-    touch  $DIR/$tdir/d1/file2
-
-    # File rename
-    mv $DIR/$tdir/d1/file2 $DIR/$tdir/d2/file3
-
-    # File and directory delete
-    touch $DIR/$tdir/d1/file4
-    mkdir $DIR/$tdir/d1/del
-    touch  $DIR/$tdir/d1/del/del1
-    touch  $DIR/$tdir/d1/del/del2
-    rm -rf $DIR/$tdir/d1/del
-    rm $DIR/$tdir/d1/file4
-
-    #hard and soft links
-    cat /etc/hosts > $DIR/$tdir/d1/link1
-    ln  $DIR/$tdir/d1/link1  $DIR/$tdir/d1/link2
-    ln -s $DIR/$tdir/d1/link1  $DIR/$tdir/d1/link3
-
-    # Device files
-    #mknod $DIR/$tdir/dev1 b 8 1
+procs_are_stopped() {
+	local pids="$*"
+	local state
+
+	for state in $(ps -p "$pids" -o state=); do
+		if [[ "$state" != T ]]; then
+			return 1
+		fi
+	done
+
+	return 0
+}
+
+# Send SIGSTOP to PIDs and wait up to 60 seconds for them to show a
+# stopped process state.
+stop_procs() {
+	local pids="$*"
+	local end
+
+	$KILL -SIGSTOP $pids
+	end=$((SECONDS + 60))
+	while ((SECONDS < end)); do
+		if procs_are_stopped $pids; then
+			return 0
+		fi
+
+		sleep 1
+	done
+
+	return 1
+}
+
+# Test 1A - test basic operations
+test_1A() { # was test_1
+	init_src
+	init_changelog
+	local xattr=$(check_xattr $TGT/foo)
+
+	# Directory create
+	mkdir $DIR/$tdir/d1
+	mkdir $DIR/$tdir/d2
+
+	# File create
+	touch $DIR/$tdir/file1
+	cp /etc/hosts  $DIR/$tdir/d1/
+	touch $DIR/$tdir/d1/"space in filename"
+	touch $DIR/$tdir/d1/file2
+
+	# File rename
+	mv $DIR/$tdir/d1/file2 $DIR/$tdir/d2/file3
+
+	# File and directory delete
+	touch $DIR/$tdir/d1/file4
+	mkdir $DIR/$tdir/d1/del
+	touch $DIR/$tdir/d1/del/del1
+	touch $DIR/$tdir/d1/del/del2
+	rm -rf $DIR/$tdir/d1/del
+	rm $DIR/$tdir/d1/file4
+
+	# Hard and soft links
+	cat /etc/hosts > $DIR/$tdir/d1/link1
+	ln $DIR/$tdir/d1/link1  $DIR/$tdir/d1/link2
+	ln -s $DIR/$tdir/d1/link1  $DIR/$tdir/d1/link3
+
+	# Device files
+	#mknod $DIR/$tdir/dev1 b 8 1
 
 	# Replicate
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
@@ -161,59 +190,94 @@ test_1() {
 	$LRSYNC -s $DIR -t $TGT -t $TGT2 -m $MDT0 -u $CL_USER -l $LREPL_LOG \
 		-D $LRSYNC_LOG
 
-    # Set attributes
-    chmod 000 $DIR/$tdir/d2/file3
-    chown nobody:nobody $DIR/$tdir/d2/file3
-
-    # Set xattrs
-    if [[ "$xattr" != "no" ]]; then
-        local value
-        touch $DIR/$tdir/file5
-        [[ "$xattr" = "large" ]] &&
-            value="$(generate_string $(max_xattr_size))" || value="bar"
-        setfattr -n user.foo -v $value $DIR/$tdir/file5
-    fi
+	# Set attributes
+	chmod 000 $DIR/$tdir/d2/file3
+	chown nobody:$GROUP $DIR/$tdir/d2/file3
+
+	# Set xattrs
+	if [[ "$xattr" != "no" ]]; then
+		local value
+		touch $DIR/$tdir/file5
+		[[ "$xattr" = "large" ]] &&
+			value="$(generate_string $(max_xattr_size))" || value="bar"
+		setfattr -n user.foo -v $value $DIR/$tdir/file5 ||
+			error "setfattr failed"
+	fi
 
 	echo "Replication #2"
 	$LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
 
-    if [[ "$xattr" != "no" ]]; then
-        local xval1=$(get_xattr_value user.foo $TGT/$tdir/file5)
-        local xval2=$(get_xattr_value user.foo $TGT2/$tdir/file5)
-    fi
-
-    RC=0
-
-    # fid2path and path2fid aren't implemented for block devices
-    #if [[ ! -b $TGT/$tdir/dev1 ]] || [[ ! -b $TGT2/$tdir/dev1 ]]; then
-    #   ls -l $DIR/$tdir/dev1 $TGT/$tdir/dev1 $TGT2/$tdir/dev1
-    #   error "Error replicating block devices"
-    #   RC=1
+	if [[ "$xattr" != "no" ]]; then
+		local xval1=$(get_xattr_value user.foo $TGT/$tdir/file5)
+		local xval2=$(get_xattr_value user.foo $TGT2/$tdir/file5)
 
-    if [[ "$xattr" != "no" ]] &&
-       [[ "$xval1" != "$value" || "$xval2" != "$value" ]]; then
-        error "Error in replicating xattrs."
-        RC=1
-    fi
+		if [[ "$xval1" != "$value" || "$xval2" != "$value" ]]; then
+			error "Error in replicating xattrs."
+		fi
+	fi
 
 	# Use diff to compare the source and the destination
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return $RC
+	fini_changelog
+	cleanup_src_tgt
+}
+run_test 1A "Simple Replication"
+
+# Test 1a - test create/delete operations in ROOT directory
+test_1a() { # LU-5005
+	rm -rf $TGT/root-* 2> /dev/null
+	rm -rf $DIR/root-* 2> /dev/null
+	init_src
+	init_changelog
+
+	# Directory create
+	mkdir_on_mdt0 $DIR/root-dir
+
+	# File create
+	touch $DIR/root-file
+	touch $DIR/root-file2
+
+	# File rename
+	mv $DIR/root-file2 $DIR/root-file3
+
+	# File and directory delete
+	touch $DIR/root-file4
+	mkdir_on_mdt0 $DIR/root-dir1
+	rm $DIR/root-file4
+	rm -rf $DIR/root-dir1
+
+	# Replicate
+	local LRSYNC_LOG=$(generate_logname "lrsync_log")
+	echo "Replication"
+	$LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
+		-D $LRSYNC_LOG
+
+	# Verify
+	stat $TGT/root-dir || error "Dir create not replicated"
+	stat $TGT/root-file || error "File create not replicated"
+	stat $TGT/root-file2 && error "Rename not replicated (src)"
+	stat $TGT/root-file3 || error "Rename not replicated (tgt)"
+	stat $TGT/root-dir1 && error "Dir delete not replicated"
+	stat $TGT/root-file4 && error "File delete not replicated"
+
+	cleanup_src_tgt
+	fini_changelog
+	rm -fr $TGT/root-*
+	rm -fr $DIR/root-*
+	return 0
 }
-run_test 1 "Simple Replication"
+run_test 1a "Replicate create/delete operations in ROOT directory"
 
 # Test 2a - Replicate files created by dbench
 test_2a() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
-    init_src
-    init_changelog
+	init_src
+	init_changelog
 
-    # Run dbench
-    sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME || error "dbench failed!"
+	# Run dbench
+	bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME ||
+		error "dbench failed"
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	# Replicate the changes to $TGT
@@ -224,28 +288,28 @@ test_2a() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 2a "Replicate files created by dbench."
 
-
 # Test 2b - Replicate files changed by dbench.
 test_2b() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
-
-    init_src
-    init_changelog
+	local child_pid
+	init_src
+	init_changelog
 
-    # Run dbench
-    sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
-    sleep 20
+	# Run dbench
+	bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
+	# wait for dbench to start
+	wait_for_function 'child_pid=$(pgrep dbench)' 360
+	# let dbench run for a bit
+	sleep 10
 
-    local child_pid=$(pgrep dbench)
-    echo PIDs: $child_pid
-    echo Stopping dbench
-    $KILL -SIGSTOP $child_pid
+	echo PIDs: $child_pid
+	echo Stopping dbench
+	stop_procs $child_pid
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	echo Starting replication
@@ -253,20 +317,20 @@ test_2b() {
 		-D $LRSYNC_LOG
 	check_diff $DIR/$tdir $TGT/$tdir
 
-    echo Resuming dbench
-    $KILL -SIGCONT $child_pid
-    sleep 10
+	echo Resuming dbench
+	$KILL -SIGCONT $child_pid
+	sleep 10
 
-    echo Stopping dbench
-    $KILL -SIGSTOP $child_pid
+	echo Stopping dbench
+	stop_procs $child_pid
 
 	echo Starting replication
 	$LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
 	check_diff $DIR/$tdir $TGT/$tdir
 
-    echo "Wait for dbench to finish"
-    $KILL -SIGCONT $child_pid
-    wait
+	echo "Wait for dbench to finish"
+	$KILL -SIGCONT $child_pid
+	wait
 
 	# Replicate the changes to $TGT
 	echo Starting replication
@@ -275,20 +339,19 @@ test_2b() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 2b "Replicate files changed by dbench."
 
 # Test 2c - Replicate files while dbench is running
 test_2c() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
-    init_src
-    init_changelog
+	init_src
+	init_changelog
 
-    # Run dbench
-    sh rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
+	# Run dbench
+	bash rundbench -C -D $DIR/$tdir 2 -t $DBENCH_TIME &
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	# Replicate the changes to $TGT
@@ -310,21 +373,19 @@ test_2c() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 2c "Replicate files while dbench is running."
 
 # Test 3a - Replicate files created by createmany
 test_3a() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
-
-    init_src
-    init_changelog
+	init_src
+	init_changelog
 
-    local numfiles=1000
-    createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed!"
+	local numfiles=1000
+	createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed"
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	# Replicate the changes to $TGT
@@ -333,23 +394,21 @@ test_3a() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 3a "Replicate files created by createmany"
 
-
 # Test 3b - Replicate files created by writemany
 test_3b() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
-
-    init_src
-    init_changelog
+	init_src
+	init_changelog
 
-    local time=60
-    local threads=5
-    writemany -q -a $DIR/$tdir/$tfile $time $threads || error "writemany failed!"
+	local time=60
+	local threads=5
+	writemany -q -a $DIR/$tdir/$tfile $time $threads ||
+		error "writemany failed"
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	# Replicate the changes to $TGT
@@ -359,22 +418,20 @@ test_3b() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 3b "Replicate files created by writemany"
 
 # Test 3c - Replicate files created by createmany/unlinkmany
 test_3c() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
-
-    init_src
-    init_changelog
+	init_src
+	init_changelog
 
-    local numfiles=1000
-    createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed!"
-    unlinkmany $DIR/$tdir/$tfile $numfiles || error "unlinkmany failed!"
+	local numfiles=1000
+	createmany -o $DIR/$tdir/$tfile $numfiles || error "createmany failed"
+	unlinkmany $DIR/$tdir/$tfile $numfiles || error "unlinkmany failed"
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	# Replicate the changes to $TGT
@@ -383,30 +440,25 @@ test_3c() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 3c "Replicate files created by createmany/unlinkmany"
 
 # Test 4 - Replicate files created by iozone
 test_4() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
+	which iozone > /dev/null 2>&1
+	[ $? -ne 0 ] && skip "iozone not found"
 
-    which iozone > /dev/null 2>&1
-    if [ $? -ne 0 ]; then
-	skip "iozone not found. Skipping test"
-	return
-    fi
+	init_src
+	init_changelog
 
-    init_src
-    init_changelog
-
-    END_RUN_FILE=${DIR}/$tdir/run LOAD_PID_FILE=${DIR}/$tdir/pid \
-        MOUNT=${DIR}/$tdir run_iozone.sh &
-    sleep 30
-    child_pid=$(pgrep iozone)
-    $KILL -SIGSTOP $child_pid
+	END_RUN_FILE=${DIR}/$tdir/run LOAD_PID_FILE=${DIR}/$tdir/pid \
+		MOUNT=${DIR}/$tdir run_iozone.sh &
+	sleep 30
+	child_pid=$(pgrep iozone)
+	stop_procs $child_pid
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	# Replicate the changes to $TGT
@@ -415,42 +467,40 @@ test_4() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    $KILL -SIGCONT $child_pid
-    sleep 60
-    $KILL -SIGKILL $(pgrep run_iozone.sh)
-    $KILL -SIGKILL $(pgrep iozone)
-
-    # After killing 'run_iozone.sh', process 'iozone' becomes the
-    # child of PID 1. Hence 'wait' does not wait for it. Killing
-    # iozone first, means more iozone processes are spawned off which
-    # is not desirable. So, after sending a sigkill, the test goes
-    # into a wait loop for iozone to cleanup and exit.
-    wait
-    while [ "$(pgrep "iozone")" != "" ];
-    do
-      ps -ef | grep iozone | grep -v grep
-      sleep 1;
-    done
+	$KILL -SIGCONT $child_pid
+	sleep 60
+	$KILL -SIGKILL $(pgrep run_iozone.sh)
+	$KILL -SIGKILL $(pgrep iozone)
+
+	# After killing 'run_iozone.sh', process 'iozone' becomes the
+	# child of PID 1. Hence 'wait' does not wait for it. Killing
+	# iozone first, means more iozone processes are spawned off which
+	# is not desirable. So, after sending a sigkill, the test goes
+	# into a wait loop for iozone to cleanup and exit.
+	wait
+	while [ "$(pgrep "iozone")" != "" ];
+	do
+		ps -ef | grep iozone | grep -v grep
+		sleep 1;
+	done
 
 	$LRSYNC -l $LREPL_LOG -D $LRSYNC_LOG
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 4 "Replicate files created by iozone"
 
 # Test 5a - Stop / start lustre_rsync
 test_5a() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
+	init_src
+	init_changelog
 
-    init_src
-    init_changelog
-
-    NUMTEST=2000
-    createmany -o $DIR/$tdir/$tfile $NUMTEST
+	NUMTEST=2000
+	createmany -o $DIR/$tdir/$tfile $NUMTEST
 
 	# Replicate the changes to $TGT
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
@@ -465,21 +515,19 @@ test_5a() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 5a "Stop / start lustre_rsync"
 
 # Test 5b - Kill / restart lustre_rsync
 test_5b() {
-    [ "$SLOW" = "no" ] && skip "Skipping slow test" && return
-
-    init_src
-    init_changelog
+	init_src
+	init_changelog
 
-    NUMTEST=2000
-    createmany -o $DIR/$tdir/$tfile $NUMTEST
+	NUMTEST=2000
+	createmany -o $DIR/$tdir/$tfile $NUMTEST
 
 	# Replicate the changes to $TGT
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
@@ -494,25 +542,26 @@ test_5b() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 5b "Kill / restart lustre_rsync"
 
 # Test 6 - lustre_rsync large no of hard links
 test_6() {
-    init_src
-    init_changelog
-
-    local NUMLINKS=128
-    touch $DIR/$tdir/link0
-    local i=1
-    while [ $i -lt $NUMLINKS ];
-    do
-      ln $DIR/$tdir/link0  $DIR/$tdir/link${i}
-      i=$(expr $i + 1)
-    done
+	init_src
+	init_changelog
+
+	local num_links=128
+	local i
+
+	touch $DIR/$tdir/link0
+	for ((i = 1; i < num_links - 1; i++)); do
+		ln $DIR/$tdir/link0 $DIR/$tdir/link$i
+	done
+	# create an extra hard link of src name ending with dest name
+	ln $DIR/$tdir/link0 $DIR/$tdir/ink0
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	# Replicate the changes to $TGT
@@ -521,27 +570,30 @@ test_6() {
 	check_diff $DIR/$tdir $TGT/$tdir
 	check_diff $DIR/$tdir $TGT2/$tdir
 
-    local count1=$(ls -l $TGT/$tdir/link0 | sed -r 's/ +/ /g' | cut -f 2 -d ' ')
-    local count2=$(ls -l $TGT/$tdir/link0 | sed -r 's/ +/ /g' | cut -f 2 -d ' ')
-    if [[ $count1 -ne $NUMLINKS ]] ||  [[ $count2 -ne $NUMLINKS ]]; then
-	ls -l $TGT/$tdir/link0 $TGT2/$tdir/link0
-	error "Incorrect no of hard links found $count1, $count2"
-    fi
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	local count1=$(stat --format=%h $TGT/$tdir/link0)
+	local count2=$(stat --format=%h $TGT2/$tdir/link0)
+	if ((count1 != num_links || count2 != num_links)); then
+		ls -l $TGT/$tdir/link0 $TGT2/$tdir/link0
+		error "Incorrect no of hard links found $count1, $count2"
+	fi
+
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 6 "lustre_rsync large no of hard links"
 
 # Test 7 - lustre_rsync stripesize
 test_7() {
-    init_src
-    mkdir -p ${DIR}/tgt/$tdir
-    init_changelog
+	local numfiles=100
+
+	init_src
+	mkdir -p ${DIR}/tgt/$tdir
+	init_changelog
 
-    local NUMFILES=100
-    lfs setstripe -c $OSTCOUNT $DIR/$tdir
-    createmany -o $DIR/$tdir/$tfile $NUMFILES
+	$LFS setstripe -c $OSTCOUNT $DIR/$tdir ||
+		error "$LFS setstripe failed"
+	createmany -o $DIR/$tdir/$tfile $numfiles
 
 	# To simulate replication to another lustre filesystem, replicate
 	# the changes to $DIR/tgt. We can't turn off the changelogs
@@ -554,9 +606,9 @@ test_7() {
 	check_diff ${DIR}/$tdir $DIR/tgt/$tdir
 
 	local i=0
-	while [ $i -lt $NUMFILES ];
+	while [ $i -lt $numfiles ];
 	do
-		local count=$(lfs getstripe $DIR/tgt/$tdir/${tfile}$i | \
+		local count=$($LFS getstripe $DIR/tgt/$tdir/${tfile}$i |
 			      awk '/stripe_count/ {print $2}')
 		if [ $count -ne $OSTCOUNT ]; then
 			error "Stripe size not replicated"
@@ -571,22 +623,22 @@ run_test 7 "lustre_rsync stripesize"
 
 # Test 8 - Replicate multiple file/directory moves
 test_8() {
-    init_src
-    init_changelog
-
-    for i in 1 2 3 4 5 6 7 8 9; do
-	mkdir $DIR/$tdir/d$i
-	    for j in 1 2 3 4 5 6 7 8 9; do
-		mkdir $DIR/$tdir/d$i/d$i$j
-		createmany -o $DIR/$tdir/d$i/d$i$j/a 10 \
-		    > /dev/null
-		mv $DIR/$tdir/d$i/d$i$j $DIR/$tdir/d$i/d0$i$j
-		createmany -o $DIR/$tdir/d$i/d0$i$j/b 10 \
-	            > /dev/null
-		mv $DIR/$tdir/d$i/d0$i$j/a0 $DIR/$tdir/d$i/d0$i$j/c0
-	    done
-	    mv $DIR/$tdir/d$i $DIR/$tdir/d0$i
-    done
+	init_src
+	init_changelog
+
+	for i in 1 2 3 4 5 6 7 8 9; do
+		mkdir $DIR/$tdir/d$i
+		for j in 1 2 3 4 5 6 7 8 9; do
+			mkdir $DIR/$tdir/d$i/d$i$j
+			createmany -o $DIR/$tdir/d$i/d$i$j/a 10 \
+			    > /dev/null
+			mv $DIR/$tdir/d$i/d$i$j $DIR/$tdir/d$i/d0$i$j
+			createmany -o $DIR/$tdir/d$i/d0$i$j/b 10 \
+			    > /dev/null
+			mv $DIR/$tdir/d$i/d0$i$j/a0 $DIR/$tdir/d$i/d0$i$j/c0
+		done
+		mv $DIR/$tdir/d$i $DIR/$tdir/d0$i
+	done
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	$LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
@@ -594,18 +646,18 @@ test_8() {
 
 	check_diff ${DIR}/$tdir $TGT/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 8 "Replicate multiple file/directory moves"
 
 test_9() {
-    init_src
-    init_changelog
+	init_src
+	init_changelog
 
-    mkdir $DIR/$tdir/foo
-    touch $DIR/$tdir/foo/a1
+	mkdir $DIR/$tdir/foo
+	touch $DIR/$tdir/foo/a1
 
 	local LRSYNC_LOG=$(generate_logname "lrsync_log")
 	$LRSYNC -s $DIR -t $TGT -m $MDT0 -u $CL_USER -l $LREPL_LOG \
@@ -620,13 +672,13 @@ test_9() {
 
 	check_diff ${DIR}/$tdir $TGT/$tdir
 
-    fini_changelog
-    cleanup_src_tgt
-    return 0
+	fini_changelog
+	cleanup_src_tgt
+	return 0
 }
 run_test 9 "Replicate recursive directory removal"
 
 cd $ORIG_PWD
-complete $(basename $0) $SECONDS
+complete_test $SECONDS
 check_and_cleanup_lustre
 exit_status