X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity.sh;h=c50519fa569945a67cf2a26088103f379373db8f;hp=8f51dc58559d8cb9dc769e418e25273c22bee222;hb=4d905db121a5d1c3bb4962b66fc2d27d70412233;hpb=7a79e2b47b7d9078757c73a13578a3f728fb4ad6

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index 8f51dc5..c50519f 100644
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -72,7 +72,7 @@ LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
 init_test_env $@
 . ${CONFIG:=$LUSTRE/tests/cfg/${NAME}.sh}
 
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="24o 27m 36f 36g 51b 51c 60c 63 64b 68 71 73 77f 78 101 103 115 120g 124b"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="24o 24v 27m 36f 36g 51b 51c 60c 63 64b 68 71 73 77f 78 101 103 115 120g 124b"
 
 SANITYLOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh).log}
 FAIL_ON_ERROR=false
@@ -90,11 +90,12 @@ setup() {
 }
 
 check_kernel_version() {
-	VERSION_FILE=version
 	WANT_VER=$1
-	GOT_VER=$(lctl get_param -n $VERSION_FILE | awk '/kernel:/ {print $2}')
-	[ $GOT_VER == "patchless" ] && return 0
-	[ $GOT_VER -ge $WANT_VER ] && return 0
+	GOT_VER=$(lctl get_param -n version | awk '/kernel:/ {print $2}')
+	case $GOT_VER in
+	patchless|patchless_client) return 0;;
+	*) [ $GOT_VER -ge $WANT_VER ] && return 0 ;;
+	esac
 	log "test needs at least kernel version $WANT_VER, running $GOT_VER"
 	return 1
 }
@@ -473,6 +474,15 @@ test_17g() {
 }
 run_test 17g "symlinks: really long symlink name ==============================="
 
+test_17h() { #bug 17378
+        mkdir -p $DIR/$tdir
+        $SETSTRIPE $DIR/$tdir -c -1
+#define OBD_FAIL_MDS_LOV_PREP_CREATE 0x141
+        do_facet mds lctl set_param fail_loc=0x80000141
+        touch $DIR/$tdir/$tfile || true
+}
+run_test 17h "create objects: lov_free_memmd() doesn't lbug"
+
 test_18() {
 	touch $DIR/f
 	ls $DIR || error
@@ -541,7 +551,8 @@ test_22() {
 }
 run_test 22 "unpack tar archive as non-root user ==============="
 
-test_23() {
+# was test_23
+test_23a() {
 	mkdir -p $DIR/$tdir
 	local file=$DIR/$tdir/$tfile
 
@@ -549,7 +560,19 @@ test_23() {
 	openfile -f O_CREAT:O_EXCL $file &&
 		error "$file recreate succeeded" || true
 }
-run_test 23 "O_CREAT|O_EXCL in subdir =========================="
+run_test 23a "O_CREAT|O_EXCL in subdir =========================="
+
+test_23b() { # bug 18988
+	mkdir -p $DIR/$tdir
+	local file=$DIR/$tdir/$tfile
+
+        rm -f $file
+        echo foo > $file || error "write filed"
+        echo bar >> $file || error "append filed"
+        $CHECKSTAT -s 8 $file || error "wrong size"
+        rm $file
+}
+run_test 23b "O_APPEND check =========================="
 
 test_24a() {
 	echo '== rename sanity =============================================='
@@ -633,7 +656,7 @@ test_24i() {
 	mrename $DIR/R9/f $DIR/R9/a
 	$CHECKSTAT -t file $DIR/R9/f || error
 	$CHECKSTAT -t dir  $DIR/R9/a || error
-	$CHECKSTAT -a file $DIR/R9/a/f || error
+	$CHECKSTAT -a $DIR/R9/a/f || error
 }
 run_test 24i "rename file to dir error: touch f ; mkdir a ; rename f a"
 
@@ -746,6 +769,21 @@ test_24u() { # bug12192
 }
 run_test 24u "create stripe file"
 
+test_24v() {
+	local NRFILES=100000
+	local FREE_INODES=`lfs df -i|grep "filesystem summary" | awk '{print $5}'`
+	[ $FREE_INODES -lt $NRFILES ] && \
+		skip "not enough free inodes $FREE_INODES required $NRFILES" && \
+		return
+
+	mkdir -p $DIR/d24v
+	createmany -m $DIR/d24v/$tfile $NRFILES
+	ls $DIR/d24v >/dev/null || error "error in listing large dir"
+
+	rm $DIR/d24v -rf
+}
+run_test 24v "list directory with large files (handle hash collision, bug: 17560)"
+
 test_25a() {
 	echo '== symlink sanity ============================================='
 
@@ -819,6 +857,7 @@ run_test 26f "rm -r of a directory which has recursive symlink ="
 test_27a() {
 	echo '== stripe sanity =============================================='
 	mkdir -p $DIR/d27 || error "mkdir failed"
+	$GETSTRIPE $DIR/d27
 	$SETSTRIPE $DIR/d27/f0 -c 1 || error "lstripe failed"
 	$CHECKSTAT -t file $DIR/d27/f0 || error "checkstat failed"
 	pass
@@ -853,7 +892,7 @@ test_27e() {
 	$SETSTRIPE $DIR/d27/f12 -c 2 && error "lstripe succeeded twice"
 	$CHECKSTAT -t file $DIR/d27/f12 || error "checkstat failed"
 }
-run_test 27e "lstripe existing file (should return error) ======"
+run_test 27e "setstripe existing file (should return error) ======"
 
 test_27f() {
 	mkdir -p $DIR/d27
@@ -861,7 +900,7 @@ test_27f() {
 	dd if=/dev/zero of=$DIR/d27/f12 bs=4k count=4 || error "dd failed"
 	$GETSTRIPE $DIR/d27/fbad || error "lfs getstripe failed"
 }
-run_test 27f "lstripe with bad stripe size (should return error)"
+run_test 27f "setstripe with bad stripe size (should return error)"
 
 test_27g() {
 	mkdir -p $DIR/d27
@@ -880,7 +919,7 @@ test_27j() {
 	mkdir -p $DIR/d27
 	$SETSTRIPE $DIR/d27/f27j -i $OSTCOUNT && error "lstripe failed"||true
 }
-run_test 27j "lstripe with bad stripe offset (should return error)"
+run_test 27j "setstripe with bad stripe offset (should return error)"
 
 test_27k() { # bug 2844
 	mkdir -p $DIR/d27
@@ -934,10 +973,15 @@ run_test 27m "create file while OST0 was full =================="
 
 # osc's keep a NOSPC stick flag that gets unset with rmdir
 reset_enospc() {
-	[ "$1" ] && FAIL_LOC=$1 || FAIL_LOC=0
+	local FAIL_LOC=${1:-0}
+	local OSTIDX=${2:-""}
+
 	mkdir -p $DIR/d27/nospc
 	rmdir $DIR/d27/nospc
-	do_nodes $(comma_list $(osts_nodes)) lctl set_param fail_loc=$FAIL_LOC
+	local list=$(comma_list $(osts_nodes))
+	[ "$OSTIDX" ] && list=$(facet_host ost$((OSTIDX + 1)))
+
+	do_nodes $list lctl set_param fail_loc=$FAIL_LOC
 }
 
 exhaust_precreations() {
@@ -945,7 +989,8 @@ exhaust_precreations() {
 	local MDSIDX=$(get_mds_dir "$DIR/d27")
 	echo OSTIDX=$OSTIDX MDSIDX=$MDSIDX
 
-	local OST=$(lfs osts | grep ${OSTIDX}": " | awk '{print $2}' | sed -e 's/_UUID$//')
+	local OST=$(lfs osts | grep ${OSTIDX}": " | \
+		awk '{print $2}' | sed -e 's/_UUID$//')
 	local MDT_INDEX=$(lfs df | grep "\[MDT:$((MDSIDX - 1))\]" | awk '{print $1}' | \
 			  sed -e 's/_UUID$//;s/^.*-//')
 
@@ -963,7 +1008,7 @@ exhaust_precreations() {
 	echo "Creating to objid $last_id on ost $OST..."
 	createmany -o $DIR/d27/${OST}-f $next_id $((last_id - next_id + 2))
 	do_facet mds${MDSIDX} lctl get_param osc.*${OST}-osc-${MDT_INDEX}.prealloc* | grep '[0-9]'
-	reset_enospc $2
+	reset_enospc $2 $OSTIDX
 }
 
 exhaust_all_precreations() {
@@ -982,9 +1027,9 @@ test_27n() {
 	reset_enospc
 	rm -f $DIR/d27/f27n
 	exhaust_precreations 0 0x80000215
-
+	$SETSTRIPE -c -1 $DIR/d27
 	touch $DIR/d27/f27n || error
-
+	$GETSTRIPE $DIR/d27/f27n
 	reset_enospc
 }
 run_test 27n "create file with some full OSTs =================="
@@ -1063,9 +1108,13 @@ test_27r() {
 run_test 27r "stripe file with some full OSTs (shouldn't LBUG) ="
 
 test_27s() { # bug 10725
-       mkdir -p $DIR/$tdir
-       $LSTRIPE $DIR/$tdir $((2048 * 1024 * 1024)) -1 2 && \
-               error "stripe width >= 2^32 succeeded" || true
+	mkdir -p $DIR/$tdir
+	local stripe_size=$((4096 * 1024 * 1024))	# 2^32
+	local stripe_count=0
+	[ $OSTCOUNT -eq 1 ] || stripe_count=2
+	$SETSTRIPE $DIR/$tdir -s $stripe_size -c $stripe_count && \
+		error "stripe width >= 2^32 succeeded" || true
+
 }
 run_test 27s "lsm_xfersize overflow (should error) (bug 10725)"
 
@@ -1110,15 +1159,17 @@ test_27v() { # bug 4900
 
         touch $DIR/$tdir/$tfile
         #define OBD_FAIL_TGT_DELAY_PRECREATE     0x705
-        lctl set_param fail_loc=0x705
-        START=`date +%s`
-        for F in `seq 1 32`; do
-                touch $DIR/$tdir/$tfile.$F
+        # all except ost1
+        for (( i=0; i < OSTCOUNT; i++ )) ; do
+                do_facet ost$i lctl set_param fail_loc=0x705
         done
-        lctl set_param fail_loc=0
+        local START=`date +%s`
+        createmany -o $DIR/$tdir/$tfile 32
+
+        reset_enospc
 
-        FINISH=`date +%s`
-        TIMEOUT=`lctl get_param -n timeout`
+        local FINISH=`date +%s`
+        local TIMEOUT=`lctl get_param -n timeout`
         [ $((FINISH - START)) -ge $((TIMEOUT / 2)) ] && \
                error "$FINISH - $START >= $TIMEOUT / 2"
 
@@ -1144,7 +1195,7 @@ test_27w() { # bug 10997
 }
 run_test 27w "check lfs setstripe -c -s -i options ============="
 
-# createtest also checks that device nodes are created and 
+# createtest also checks that device nodes are created and
 # then visible correctly (#2091)
 test_28() { # bug 2091
 	mkdir $DIR/d28
@@ -2080,7 +2131,7 @@ test_46() {
 }
 run_test 46 "dirtying a previously written page ================"
 
-# test_47 is removed "Device nodes check" is moved to test_28 
+# test_47 is removed "Device nodes check" is moved to test_28
 
 test_48a() { # bug 2399
 	check_kernel_version 34 || return 0
@@ -2276,7 +2327,7 @@ test_51bb() {
 		echo "mds $i: inodes count OLD ${OLDUSED[$i]} NEW ${NEWUSED[$i]}"
 		[ ${OLDUSED[$i]} -lt ${NEWUSED[$i]} ] || rc=$((rc + 1))
 	done
-	
+
 	lctl set_param -n lmv.*.placement=$savePOLICY
 
 	[ $rc -ne $MDSCOUNT ] || \
@@ -2750,7 +2801,7 @@ test_57b() {
 	mkdir -p $dir || error "creating $dir"
 	local num=$(get_mds_dir $dir)
 	local mymds=mds$num
-	
+
 	echo "mcreating $FILECOUNT files"
 	createmany -m $dir/f 1 $FILECOUNT || \
 		error "creating files in $dir"
@@ -3470,7 +3521,6 @@ test_77b() { # bug 10889
 		error "dd error: $?"
 	lctl set_param fail_loc=0
 	set_checksums 0
-	rm -f $DIR/f77b
 }
 run_test 77b "checksum error on client write ===================="
 
@@ -3488,6 +3538,7 @@ test_77c() { # bug 10889
 	done
 	set_checksums 0
 	set_checksum_type $ORIG_CSUM_TYPE
+	rm -f $DIR/f77b
 }
 run_test 77c "checksum error on client read ==================="
 
@@ -3693,7 +3744,7 @@ test_99a() {
 	chown $RUNAS_ID $DIR/d99cvsroot
 	local oldPWD=$PWD	# bug 13584, use $TMP as working dir
 	cd $TMP
-	
+
 	$RUNAS cvs -d $DIR/d99cvsroot init || error
 	cd $oldPWD
 }
@@ -3768,10 +3819,10 @@ test_100() {
 		if [ $LPORT -ge 1024 ]; then
 			echo "bad: $PROT $SND $RCV $LOCAL $REMOTE $STAT"
 			netstat -tna
-			error "local: $LPORT > 1024, remote: $RPORT"
+			error_exit "local: $LPORT > 1024, remote: $RPORT"
 		fi
 	done
-	[ "$rc" = 0 ] || error "privileged port not found" )
+	[ "$rc" = 0 ] || error_exit "privileged port not found" )
 }
 run_test 100 "check local port using privileged port ==========="
 
@@ -3905,6 +3956,51 @@ test_101b() {
 }
 run_test 101b "check stride-io mode read-ahead ================="
 
+set_read_ahead() {
+   lctl get_param -n llite.*.max_read_ahead_mb | head -n 1
+   lctl set_param -n llite.*.max_read_ahead_mb $1 > /dev/null 2>&1
+}
+
+test_101d() {
+    local file=$DIR/$tfile
+    local size=${FILESIZE_101c:-500}
+    local ra_MB=${READAHEAD_MB:-40}
+
+    local space=$(df -P $DIR | tail -n 1 | awk '{ print $4 }')
+    [ $space -gt $((size / 1024)) ] ||
+        { skip "Need free space ${size}M, have $space" && return; }
+
+    echo Creating ${size}M test file $file
+    dd if=/dev/zero of=$file bs=1M count=$size
+    echo Cancel LRU locks on lustre client to flush the client cache
+    cancel_lru_locks osc
+
+    echo Disable read-ahead
+    local old_READAHEAD=$(set_read_ahead 0)
+
+    echo Reading the test file $file with read-ahead disabled
+    time_ra_OFF=$(do_and_time "dd if=$file of=/dev/null bs=1M count=$size")
+
+    echo Cancel LRU locks on lustre client to flush the client cache
+    cancel_lru_locks osc
+    echo Enable read-ahead with ${ra_MB}MB
+    set_read_ahead $ra_MB
+
+    echo Reading the test file $file with read-ahead enabled
+    time_ra_ON=$(do_and_time "dd if=$file of=/dev/null bs=1M count=$size")
+
+    echo read-ahead disabled time read $time_ra_OFF
+    echo read-ahead enabled  time read $time_ra_ON
+
+    set_read_ahead $old_READAHEAD
+    rm -f $file
+
+    [ $time_ra_ON -lt $time_ra_OFF ] ||
+        error "read-ahead enabled  time read (${time_ra_ON}s) is more than
+               read-ahead disabled time read (${time_ra_OFF}s) filesize ${size}M"
+}
+run_test 101d "file read with and without read-ahead enabled  ================="
+
 export SETUP_TEST102=no
 setup_test102() {
 	[ "$SETUP_TEST102" = "yes" ] && return
@@ -3998,16 +4094,16 @@ test_102b() {
 	echo "get/set/list trusted.lov xattr ..."
 	[ "$OSTCOUNT" -lt "2" ] && skip "skipping 2-stripe test" && return
 	local testfile=$DIR/$tfile
-	$SETSTRIPE $testfile -s 65536 -i 1 -c 2
+	$SETSTRIPE -s 65536 -i 1 -c 2 $testfile || error "setstripe failed"
 	getfattr -d -m "^trusted" $testfile 2> /dev/null | \
 	grep "trusted.lov" || error "can't get trusted.lov from $testfile"
 
 	local testfile2=${testfile}2
 	local value=`getfattr -n trusted.lov $testfile 2> /dev/null | \
 		     grep "trusted.lov" |sed -e 's/[^=]\+=//'`
-	
+
 	$MCREATE $testfile2
-	setfattr -n trusted.lov -v $value $testfile2 	
+	setfattr -n trusted.lov -v $value $testfile2
 	local tmp_file=${testfile}3
 	$GETSTRIPE -v $testfile2 > $tmp_file
 	local stripe_size=`grep "size"  $tmp_file| awk '{print $2}'`
@@ -4025,16 +4121,16 @@ test_102c() {
 	mkdir -p $DIR/$tdir
 	chown $RUNAS_ID $DIR/$tdir
 	local testfile=$DIR/$tdir/$tfile
-	$RUNAS $SETSTRIPE $testfile -s 65536 -i 1 -c 2
+	$RUNAS $SETSTRIPE -s 65536 -i 1 -c 2 $testfile||error "setstripe failed"
 	$RUNAS getfattr -d -m "^lustre" $testfile 2> /dev/null | \
 	grep "lustre.lov" || error "can't get lustre.lov from $testfile"
 
 	local testfile2=${testfile}2
 	local value=`getfattr -n lustre.lov $testfile 2> /dev/null | \
 		     grep "lustre.lov" |sed -e 's/[^=]\+=//'  `
-	
+
 	$RUNAS $MCREATE $testfile2
-	$RUNAS setfattr -n lustre.lov -v $value $testfile2 	
+	$RUNAS setfattr -n lustre.lov -v $value $testfile2
 	local tmp_file=${testfile}3
 	$RUNAS $GETSTRIPE -v $testfile2 > $tmp_file
 	local stripe_size=`grep "size"  $tmp_file| awk '{print $2}'`
@@ -4045,6 +4141,8 @@ test_102c() {
 run_test 102c "non-root getfattr/setfattr for lustre.lov EAs ==========="
 
 compare_stripe_info1() {
+	local stripe_index_all_zero=1
+
 	for num in 1 2 3 4
 	do
  		for count in `seq 1 $STRIPE_COUNT`
@@ -4061,11 +4159,13 @@ compare_stripe_info1() {
 					error "$file: different stripe count $stripe_count, expected $count" && return
 				fi
 				if [ $stripe_index -ne 0 ]; then
- 					error "$file: different stripe offset $stripe_index, expected 0" && return
+				       stripe_index_all_zero=0
 				fi
 			done
 		done
 	done
+	[ $stripe_index_all_zero -eq 1 ] && error "all files are being extracted starting from OST index 0"
+	return 0
 }
 
 compare_stripe_info2() {
@@ -4251,7 +4351,7 @@ test_104() {
 	lfs df -i $DIR || error "lfs df -i $DIR failed"
 	lfs df $DIR/$tfile || error "lfs df $DIR/$tfile failed"
 	lfs df -ih $DIR/$tfile || error "lfs df -ih $DIR/$tfile failed"
-	
+
 	OSC=`lctl get_param -n devices | awk '/-osc-/ {print $4}' | head -n 1`
 	lctl --device %$OSC deactivate
 	lfs df || error "lfs df with deactivated OSC failed"
@@ -4423,7 +4523,7 @@ test_116() {
 	declare -i FILL
 	FILL=$(($MINV / 4))
 	echo "Filling 25% remaining space in OST${MINI} with ${FILL}Kb"
-	$SETSTRIPE $DIR/$tdir/OST${MINI} -i $MINI -c 1
+	$SETSTRIPE -i $MINI -c 1 $DIR/$tdir/OST${MINI}||error "setstripe failed"
 	i=0
 	while [ $FILL -gt 0 ]; do
 	    i=$(($i + 1))
@@ -4529,7 +4629,7 @@ reset_async() {
 test_118a() #bug 11710
 {
 	reset_async
-	
+
  	multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
 	DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
         WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache | grep -c writeback)
@@ -4573,7 +4673,7 @@ test_118b()
 	# until a subsequent RPC completes successfully without error.
 	multiop $DIR/$tfile Ow4096yc
 	rm -f $DIR/$tfile
-	
+
 	return 0
 }
 run_test 118b "Reclaim dirty pages on fatal error =========="
@@ -4615,7 +4715,7 @@ test_118c()
 	if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
 		error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
 	fi
-	
+
 	rm -f $DIR/$tfile
 	echo "Dirty pages flushed via fsync on EROFS"
 	return 0
@@ -4631,7 +4731,7 @@ test_118d()
 	#define OBD_FAIL_OST_BRW_PAUSE_BULK
 	set_nodes_failloc "$(osts_nodes)" 0x214
 	# multiop should block due to fsync until pages are written
-	multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &	
+	multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
 	MULTIPID=$!
 	sleep 1
 
@@ -4673,7 +4773,7 @@ test_118f() {
 	if [[ $RC -eq 0 ]]; then
 		error "Must return error due to dropped pages, rc=$RC"
 	fi
-	
+
         lctl set_param fail_loc=0x0
 
         LOCKED=$(lctl get_param -n llite.*.dump_page_cache | grep -c locked)
@@ -4705,7 +4805,7 @@ test_118g() {
 	# simulate local -ENOMEM
 	multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
 	RC=$?
-	
+
 	lctl set_param fail_loc=0
 	if [[ $RC -eq 0 ]]; then
 		error "Must return error due to dropped pages, rc=$RC"
@@ -4718,7 +4818,7 @@ test_118g() {
 	if [[ $LOCKED -ne 0 ]]; then
 		error "Locked pages remain in cache, locked=$LOCKED"
 	fi
-	
+
 	if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
 		error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
 	fi
@@ -4741,7 +4841,7 @@ test_118h() {
 	# Should simulate ENOMEM error which is recoverable and should be handled by timeout
         multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
         RC=$?
-	
+
         set_nodes_failloc "$(osts_nodes)" 0
 	if [[ $RC -eq 0 ]]; then
 		error "Must return error due to dropped pages, rc=$RC"
@@ -4754,7 +4854,7 @@ test_118h() {
 	if [[ $LOCKED -ne 0 ]]; then
 		error "Locked pages remain in cache, locked=$LOCKED"
 	fi
-	
+
 	if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
 		error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
 	fi
@@ -4773,13 +4873,13 @@ test_118i() {
 
 	#define OBD_FAIL_OST_BRW_WRITE_BULK      0x20e
         set_nodes_failloc "$(osts_nodes)" 0x20e
-	
+
 	# Should simulate ENOMEM error which is recoverable and should be handled by timeout
         multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
 	PID=$!
 	sleep 5
 	set_nodes_failloc "$(osts_nodes)" 0
-	
+
 	wait $PID
         RC=$?
 	if [[ $RC -ne 0 ]]; then
@@ -4792,7 +4892,7 @@ test_118i() {
 	if [[ $LOCKED -ne 0 ]]; then
 		error "Locked pages remain in cache, locked=$LOCKED"
 	fi
-	
+
 	if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
 		error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
 	fi
@@ -4826,7 +4926,7 @@ test_118j() {
 	if [[ $LOCKED -ne 0 ]]; then
 		error "Locked pages remain in cache, locked=$LOCKED"
 	fi
-	
+
 	# in recoverable error on OST we want resend and stay until it finished
 	if [[ $DIRTY -ne 0 || $WRITEBACK -ne 0 ]]; then
 		error "Dirty pages not flushed to disk, dirty=$DIRTY, writeback=$WRITEBACK"
@@ -4882,7 +4982,7 @@ test_119b() # bug 11737
 {
         [ "$OSTCOUNT" -lt "2" ] && skip "skipping 2-stripe test" && return
 
-        $SETSTRIPE $DIR/$tfile -c 2
+        $SETSTRIPE -c 2 $DIR/$tfile || error "setstripe failed"
         dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 seek=1 || error "dd failed"
         sync
         multiop $DIR/$tfile oO_RDONLY:O_DIRECT:r$((2048 * 1024)) || \
@@ -5161,37 +5261,38 @@ test_123a() { # was test 123, statahead(bug 11401)
                         log "statahead was stopped, maybe too many locks held!"
                 fi
 
-                [ $delta -eq 0 ] && continue
+                [ $delta -eq 0 -o $delta_sa -eq 0 ] && continue
 
-                if [ $((delta_sa * 100)) -gt $((delta * 105)) ]; then
+                if [ $((delta_sa * 100)) -gt $((delta * 105)) -a $delta_sa -gt $((delta + 2)) ]; then
                         if [  $SLOWOK -eq 0 ]; then
                                 error "ls $i files is slower with statahead!"
+                                debugsave
 
+                                lctl set_param debug=-1
                                 max=`lctl get_param -n llite.*.statahead_max | head -n 1`
                                 lctl set_param -n llite.*.statahead_max 0
                                 lctl get_param llite.*.statahead_max
                                 cancel_lru_locks mdc
                                 cancel_lru_locks osc
-                                $LCTL dk > /dev/null
+                                $LCTL clear
                                 stime=`date +%s`
                                 time ls -l $DIR/$tdir | wc -l
                                 etime=`date +%s`
-                                $LCTL dk > $TMP/sanity_test_123a_${i}_disable_${etime}.log
-                                delta=$((etime - stime))
-                                log "ls $i files without statahead: $delta sec, dump to $TMP/sanity_test_123a_${i}_disable_${etime}.log"
-                                lctl set_param llite.*.statahead_max=$max
+                                error "ls $i files (again) without statahead: $((etime - stime)) sec"
 
+                                lctl set_param debug=-1
+                                lctl set_param llite.*.statahead_max=$max
                                 lctl get_param -n llite.*.statahead_max | grep '[0-9]'
                                 cancel_lru_locks mdc
                                 cancel_lru_locks osc
-                                $LCTL dk > /dev/null
+                                $LCTL clear
                                 stime=`date +%s`
                                 time ls -l $DIR/$tdir | wc -l
                                 etime=`date +%s`
-                                $LCTL dk > $TMP/sanity_test_123a_${i}_enable_${etime}.log
-                                delta_sa=$((etime - stime))
-                                log "ls $i files with statahead: $delta_sa sec, dump to $TMP/sanity_test_123a_${i}_enable_${etime}.log"
+                                error "ls $i files (again) with statahead: $((etime - stime)) sec"
 		                lctl get_param -n llite.*.statahead_stats
+
+                                debugrestore
                         else
                                 log "ls $i files is slower with statahead!"
                         fi
@@ -5200,7 +5301,7 @@ test_123a() { # was test 123, statahead(bug 11401)
 
                 [ $delta -gt 20 ] && break
                 [ $delta -gt 8 ] && MULT=$((50 / delta))
-                [ "$SLOW" = "no" -a $delta -gt 3 ] && break
+                [ "$SLOW" = "no" -a $delta -gt 5 ] && break
         done
         log "ls done"
 
@@ -5218,7 +5319,7 @@ run_test 123a "verify statahead work"
 test_123b () { # statahead(bug 15027)
 	mkdir -p $DIR/$tdir
 	createmany -o $DIR/$tdir/$tfile-%d 1000
-	
+
         cancel_lru_locks mdc
         cancel_lru_locks osc
 
@@ -5286,7 +5387,7 @@ test_124a() {
         log "LVF=$LVF"
         local OLD_LVF=`lctl get_param -n $NSDIR.pool.lock_volume_factor`
         lctl set_param -n $NSDIR.pool.lock_volume_factor $LVF
-        
+
         # Let's make sure that we really have some margin. Client checks
         # cached locks every 10 sec.
         SLEEP=$((SLEEP+20))
@@ -5419,7 +5520,7 @@ test_126() { # bug 12829/13455
 run_test 126 "check that the fsgid provided by the client is taken into account"
 
 test_127() { # bug 15521
-        $LSTRIPE -i 0 -c 1 $DIR/$tfile
+        $SETSTRIPE -i 0 -c 1 $DIR/$tfile || error "setstripe failed"
         $LCTL set_param osc.*.stats=0
         FSIZE=$((2048 * 1024))
         dd if=/dev/zero of=$DIR/$tfile bs=$FSIZE count=1
@@ -5431,7 +5532,7 @@ test_127() { # bug 15521
                 echo "got $COUNT $NAME"
                 [ ! $MIN ] && error "Missing min value for $NAME proc entry"
                 eval $NAME=$COUNT || error "Wrong proc format"
-		
+
                 case $NAME in
                         read_bytes|write_bytes)
                         [ $MIN -lt 4096 ] && error "min is too small: $MIN"
@@ -5814,6 +5915,64 @@ test_131e() {
 }
 run_test 131e "test read hitting hole"
 
+get_ost_param() {
+        local token=$1
+        local gl_sum=0
+        for node in $(osts_nodes); do
+                gl=$(do_node $node "$LCTL get_param -n ost.OSS.ost.stats" | awk '/'$token'/ {print $2}' | head -n 1)
+                [ x$gl = x"" ] && gl=0
+                gl_sum=$((gl_sum + gl))
+        done
+        echo $gl
+}
+
+som_mode_switch() {
+        local som=$1
+        local gl1=$2
+        local gl2=$3
+
+        if [ x$som = x"enabled" ]; then
+                [ $((gl2 - gl1)) -gt 0 ] && error "no glimpse RPC is expected"
+                do_facet mgs "$LCTL conf_param $FSNAME.mdt.som=disabled"
+        else
+                [ $((gl2 - gl1)) -gt 0 ] || error "some glimpse RPC is expected"
+                do_facet mgs "$LCTL conf_param $FSNAME.mdt.som=enabled"
+        fi
+
+        # do remount to make new mount-conf parameters actual
+        echo remounting...
+        sync
+        stopall
+        setupall
+}
+
+test_132() { #1028, SOM
+	local num=$(get_mds_dir $DIR)
+	local mymds=mds${num}
+
+        dd if=/dev/zero of=$DIR/$tfile count=1 2>/dev/null
+        cancel_lru_locks osc
+
+        som1=$(do_facet $mymds "$LCTL get_param mdt.*.som" |  awk -F= ' {print $2}' | head -n 1)
+
+        gl1=$(get_ost_param "ldlm_glimpse_enqueue")
+        stat $DIR/$tfile >/dev/null
+        gl2=$(get_ost_param "ldlm_glimpse_enqueue")
+        echo "====> SOM is "$som1", "$((gl2 - gl1))" glimpse RPC occured"
+        cancel_lru_locks osc
+        som_mode_switch $som1 $gl1 $gl2
+
+        som2=$(do_facet $mymds "$LCTL get_param mdt.*.som" |  awk -F= ' {print $2}' | head -n 1)
+        [ $som1 != $som2 ] || error "som is still "$som2
+
+        gl1=$(get_ost_param "ldlm_glimpse_enqueue")
+        stat $DIR/$tfile >/dev/null
+        gl2=$(get_ost_param "ldlm_glimpse_enqueue")
+        echo "SOM is "$som2", "$((gl2 - gl1))" glimpse RPC occured"
+        som_mode_switch $som2 $gl1 $gl2
+}
+run_test 132 "som avoids glimpse rpc"
+
 test_140() { #bug-17379
         mkdir -p $DIR/$tdir || error "Creating dir $DIR/$tdir"
         cd $DIR/$tdir || error "Changing to $DIR/$tdir"
@@ -5875,76 +6034,81 @@ test_150() {
 run_test 150 "truncate/append tests"
 
 function roc_access() {
-	ACCNUM=`$LCTL get_param -n obdfilter.*.stats | \
-		grep 'cache_access'| awk '{print $2}' | \
-		awk '{sum=sum+$3} END{print sum}'`
-	echo $ACCNUM
+        local list=$(comma_list $(osts_nodes))
+        ACCNUM=`do_nodes $list $LCTL get_param -n obdfilter.*.stats | \
+                grep 'cache_access'| awk '{print $2}' | \
+                awk '{sum=sum+$3} END{print sum}'`
+        echo $ACCNUM
 }
 
 function roc_hit() {
-	ACCNUM=`$LCTL get_param -n obdfilter.*.stats | \
-		grep 'cache_hit'|awk '{print $2}' | \
-		awk '{sum=sum+$1} END{print sum}'`
-	echo $ACCNUM
+        local list=$(comma_list $(osts_nodes))
+        ACCNUM=`do_nodes $list $LCTL get_param -n obdfilter.*.stats | \
+                grep 'cache_hit'|awk '{print $2}' | \
+                awk '{sum=sum+$1} END{print sum}'`
+        echo $ACCNUM
 }
 
 test_151() {
-	local CPAGES=3
+        remote_ost_nodsh && skip "remote OST with nodsh" && return
 
-	# check whether obdfilter is cache capable at all
-	if ! $LCTL get_param -n obdfilter.*.read_cache_enable; then
-		echo "not cache-capable obdfilter"
-		return 0
-	fi
+        local CPAGES=3
+        local list=$(comma_list $(osts_nodes))
 
-	# check cache is enabled on all obdfilters
-	if $LCTL get_param -n obdfilter.*.read_cache_enable | grep 0 >&/dev/null; then
-		echo "oss cache is disabled"
-		return 0
-	fi
+        # check whether obdfilter is cache capable at all
+        if ! do_nodes $list $LCTL get_param -n obdfilter.*.read_cache_enable > /dev/null; then
+                echo "not cache-capable obdfilter"
+                return 0
+        fi
 
-	$LCTL set_param -n obdfilter.*.writethrough_cache_enable 1
+        # check cache is enabled on all obdfilters
+        if do_nodes $list $LCTL get_param -n obdfilter.*.read_cache_enable | grep 0 >&/dev/null; then
+                echo "oss cache is disabled"
+                return 0
+        fi
 
-	# pages should be in the case right after write
-	dd if=/dev/urandom of=$DIR/$tfile bs=4k count=$CPAGES || error "dd failed"
-	local BEFORE=`roc_hit`
-	cancel_lru_locks osc
-	cat $DIR/$tfile >/dev/null
-	local AFTER=`roc_hit`
-	if ! let "AFTER - BEFORE == CPAGES"; then
-		error "NOT IN CACHE: before: $BEFORE, after: $AFTER"
-	fi
+        do_nodes $list $LCTL set_param -n obdfilter.*.writethrough_cache_enable 1
 
-	# the following read invalidates the cache
-	cancel_lru_locks osc
-	$LCTL set_param -n obdfilter.*.read_cache_enable 0
-	cat $DIR/$tfile >/dev/null
+        # pages should be in the case right after write
+        dd if=/dev/urandom of=$DIR/$tfile bs=4k count=$CPAGES || error "dd failed"
+        local BEFORE=`roc_hit`
+        cancel_lru_locks osc
+        cat $DIR/$tfile >/dev/null
+        local AFTER=`roc_hit`
+        if ! let "AFTER - BEFORE == CPAGES"; then
+                error "NOT IN CACHE: before: $BEFORE, after: $AFTER"
+        fi
 
-	# now data shouldn't be found in the cache
-	BEFORE=`roc_hit`
-	cancel_lru_locks osc
-	cat $DIR/$tfile >/dev/null
-	AFTER=`roc_hit`
-	if let "AFTER - BEFORE != 0"; then
-		error "IN CACHE: before: $BEFORE, after: $AFTER"
-	fi
+        # the following read invalidates the cache
+        cancel_lru_locks osc
+        do_nodes $list $LCTL set_param -n obdfilter.*.read_cache_enable 0
+        cat $DIR/$tfile >/dev/null
 
-	$LCTL set_param -n obdfilter.*.read_cache_enable 1
-	rm -f $DIR/$tfile
+        # now data shouldn't be found in the cache
+        BEFORE=`roc_hit`
+        cancel_lru_locks osc
+        cat $DIR/$tfile >/dev/null
+        AFTER=`roc_hit`
+        if let "AFTER - BEFORE != 0"; then
+                error "IN CACHE: before: $BEFORE, after: $AFTER"
+        fi
+
+        do_nodes $list $LCTL set_param -n obdfilter.*.read_cache_enable 1
+        rm -f $DIR/$tfile
 }
 run_test 151 "test cache on oss and controls ==============================="
 
 test_152() {
         local TF="$TMP/$tfile"
 
-	# simulate ENOMEM during write
-#define OBD_FAIL_OST_NOMEM     	0x226
+        # simulate ENOMEM during write
+#define OBD_FAIL_OST_NOMEM      0x226
         lctl set_param fail_loc=0x80000226
         dd if=/dev/urandom of=$TF bs=6096 count=1 || error "dd failed"
         cp $TF $DIR/$tfile
         sync || error "sync failed"
         lctl set_param fail_loc=0
-	
+
         # discard client's cache
         cancel_lru_locks osc
 
@@ -5953,7 +6117,7 @@ test_152() {
         cmp $TF $DIR/$tfile || error "cmp failed"
         lctl set_param fail_loc=0
 
-	rm -f $TF
+        rm -f $TF
 }
 run_test 152 "test read/write with enomem ============================"
 
@@ -5985,7 +6149,6 @@ err17935 () {
     fi
 }
 test_160() {
-    do_facet $SINGLEMDS lctl set_param mdd.$MDT0.changelog on
     USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
     echo "Registered as changelog user $USER"
     do_facet $SINGLEMDS lctl get_param -n mdd.$MDT0.changelog_users | \
@@ -6000,8 +6163,17 @@ test_160() {
     ln -s $DIR/$tdir/pics/2008/portland.jpg $DIR/$tdir/pics/desktop.jpg
     rm $DIR/$tdir/pics/desktop.jpg
 
-    # verify contents
     $LFS changelog $MDT0 | tail -5
+
+    echo "verifying changelog mask"
+    do_facet $SINGLEMDS lctl set_param mdd.$MDT0.changelog_mask="-mkdir"
+    mkdir -p $DIR/$tdir/pics/2009/sofia
+    do_facet $SINGLEMDS lctl set_param mdd.$MDT0.changelog_mask="+mkdir"
+    mkdir $DIR/$tdir/pics/2009/zachary
+    DIRS=$($LFS changelog $MDT0 | tail -5 | grep -c MKDIR)
+    [ $DIRS -eq 1 ] || err17935 "changelog mask count $DIRS != 1"
+
+    # verify contents
     echo "verifying target fid"
     fidc=$($LFS changelog $MDT0 | grep timestamp | grep "CREAT" | \
 	tail -1 | awk '{print $5}')
@@ -6013,35 +6185,46 @@ test_160() {
 	tail -1 | awk '{print $6}')
     fidf=$($LFS path2fid $DIR/$tdir/pics/zach)
     [ "$fidc" == "p=$fidf" ] || \
-	err17935 "pfid in changelog $fidc != dir fid $fidf" 
+	err17935 "pfid in changelog $fidc != dir fid $fidf"
 
-    echo "verifying user clear"
-    USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
-	mdd.$MDT0.changelog_users | wc -l) - 2 ))
-    FIRST_REC=$($LFS changelog $MDT0 | head -1 | awk '{print $1}')
-    $LFS changelog_clear $MDT0 $USER $(($FIRST_REC + 5))  
-    USER_REC=$(do_facet $SINGLEMDS lctl get_param -n \
+    USER_REC1=$(do_facet $SINGLEMDS lctl get_param -n \
 	mdd.$MDT0.changelog_users | grep $USER | awk '{print $2}')
-    [ $USER_REC == $(($FIRST_REC + 5)) ] || \
-	err17935 "user index should be $(($FIRST_REC + 5)); is $USER_REC"
-    CLEAR_REC=$($LFS changelog $MDT0 | head -1 | awk '{print $1}')
-    [ $CLEAR_REC == $(($FIRST_REC + 6)) -o $USERS -gt 1 ] || \
-	err17935 "first index should be $(($FIRST_REC + 6)); is $PURGE_REC"
+    $LFS changelog_clear $MDT0 $USER $(($USER_REC1 + 5))
+    USER_REC2=$(do_facet $SINGLEMDS lctl get_param -n \
+	mdd.$MDT0.changelog_users | grep $USER | awk '{print $2}')
+    echo "verifying user clear: $(( $USER_REC1 + 5 )) == $USER_REC2"
+    [ $USER_REC2 == $(($USER_REC1 + 5)) ] || \
+	err17935 "user index should be $(($USER_REC1 + 5)); is $USER_REC2"
+
+    MIN_REC=$(do_facet $SINGLEMDS lctl get_param mdd.$MDT0.changelog_users | \
+	awk 'min == "" || $2 < min {min = $2}; END {print min}')
+    FIRST_REC=$($LFS changelog $MDT0 | head -1 | awk '{print $1}')
+    echo "verifying min purge: $(( $MIN_REC + 1 )) == $FIRST_REC"
+    [ $FIRST_REC == $(($MIN_REC + 1)) ] || \
+	err17935 "first index should be $(($MIN_REC + 1)); is $FIRST_REC"
 
     echo "verifying user deregister"
     do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER
     do_facet $SINGLEMDS lctl get_param -n mdd.$MDT0.changelog_users | \
 	grep -q $USER && error "User $USER still found in changelog_users"
 
-    [ $USERS -eq 1 ] && \
-	do_facet $SINGLEMDS lctl set_param mdd.$MDT0.changelog off || true
+    USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
+	mdd.$MDT0.changelog_users | wc -l) - 2 ))
+    if [ $USERS -eq 0 ]; then
+	LAST_REC1=$(do_facet $SINGLEMDS lctl get_param -n \
+	    mdd.$MDT0.changelog_users | head -1 | awk '{print $3}')
+	touch $DIR/$tdir/chloe
+	LAST_REC2=$(do_facet $SINGLEMDS lctl get_param -n \
+	    mdd.$MDT0.changelog_users | head -1 | awk '{print $3}')
+	echo "verify changelogs are off if we were the only user: $LAST_REC1 == $LAST_REC2"
+	[ $LAST_REC1 == $LAST_REC2 ] || error "changelogs not off"
+    else
+	echo "$USERS other changelog users; can't verify off"
+    fi
 }
 run_test 160 "changelog sanity"
 
 test_161() {
-    # need local MDT for fid2path
-    remote_mds && skip "remote MDS" && return
-
     mkdir -p $DIR/$tdir
     cp /etc/hosts $DIR/$tdir/$tfile
     mkdir $DIR/$tdir/foo1
@@ -6051,8 +6234,8 @@ test_161() {
     ln $DIR/$tdir/$tfile $DIR/$tdir/foo1/luna
     ln $DIR/$tdir/$tfile $DIR/$tdir/foo2/thor
     local FID=$($LFS path2fid $DIR/$tdir/$tfile | tr -d '[')
-    if [ "$($LFS fid2path ${mds1_svc} $FID | wc -l)" != "5" ]; then
-	$LFS fid2path ${mds1_svc} $FID
+    if [ "$($LFS fid2path $DIR $FID | wc -l)" != "5" ]; then
+	$LFS fid2path $DIR $FID
 	err17935 "bad link ea"
     fi
     # middle
@@ -6063,9 +6246,9 @@ test_161() {
     rm $DIR/$tdir/$tfile
     # rename
     mv $DIR/$tdir/foo1/sofia $DIR/$tdir/foo2/maggie
-    if [ "$($LFS fid2path ${mds1_svc} --link 1 $FID)" != "/$tdir/foo2/maggie" ]
+    if [ "$($LFS fid2path $DIR --link 1 $FID)" != "/$tdir/foo2/maggie" ]
 	then
-	$LFS fid2path ${mds1_svc} $FID
+	$LFS fid2path $DIR $FID
 	err17935 "bad link rename"
     fi
     rm $DIR/$tdir/foo2/maggie
@@ -6074,11 +6257,11 @@ test_161() {
     local longname=filename_avg_len_is_thirty_two_
     createmany -l$DIR/$tdir/foo1/luna $DIR/$tdir/foo2/$longname 1000 || \
 	error "failed to hardlink many files"
-    links=$($LFS fid2path ${mds1_svc} $FID | wc -l)
+    links=$($LFS fid2path $DIR $FID | wc -l)
     echo -n "${links}/1000 links in link EA"
     [ ${links} -gt 60 ] || err17935 "expected at least 60 links in link EA"
     unlinkmany $DIR/$tdir/foo2/$longname 1000 || \
-	error "failed to unlink many hardlinks" 
+	error "failed to unlink many hardlinks"
 }
 run_test 161 "link ea sanity"
 
@@ -6101,9 +6284,6 @@ check_path() {
 }
 
 test_162() {
-    # need local MDT for fid2path
-    remote_mds && skip "remote MDS" && return
-
     # Make changes to filesystem
     mkdir -p $DIR/$tdir/d2
     touch $DIR/$tdir/d2/$tfile
@@ -6112,24 +6292,43 @@ test_162() {
     mkdir -p $DIR/$tdir/d2/a/b/c
     mkdir -p $DIR/$tdir/d2/p/q/r
     FID=$($LFS path2fid $DIR/$tdir/d2/$tfile | tr -d '[')
-    check_path "/$tdir/d2/$tfile" ${mds1_svc} $FID --link 0
+    check_path "/$tdir/d2/$tfile" $DIR $FID --link 0
     ln $DIR/$tdir/d2/$tfile $DIR/$tdir/d2/p/q/r/hlink
     mv $DIR/$tdir/d2/$tfile $DIR/$tdir/d2/a/b/c/new_file
     FID=$($LFS path2fid $DIR/$tdir/d2/a/b/c/new_file | tr -d '[')
-    check_path "/$tdir/d2/a/b/c/new_file" ${mds1_svc} $FID --link 1
-    check_path "/$tdir/d2/p/q/r/hlink" ${mds1_svc} $FID --link 0
+    # fid2path dir/fsname should both work
+    check_path "/$tdir/d2/a/b/c/new_file" $FSNAME $FID --link 1
+    check_path "/$tdir/d2/p/q/r/hlink" $DIR $FID --link 0
     # check that there are 2 links
-    ${LFS} fid2path ${mds1_svc} $FID | wc -l | grep -q 2 || \
-	err17935 "expected 2 links" 
+    ${LFS} fid2path $DIR $FID | wc -l | grep -q 2 || \
+	err17935 "expected 2 links"
 
     rm $DIR/$tdir/d2/p/q/r/hlink
-    check_path "/$tdir/d2/a/b/c/new_file" ${mds1_svc} $FID --link 0
-    # Doesnt work with CMD yet: 17935 
+    check_path "/$tdir/d2/a/b/c/new_file" $DIR $FID --link 0
+    # Doesnt work with CMD yet: 17935
     return 0
 }
 run_test 162 "path lookup sanity"
 
+test_169() {
+	# do directio so as not to populate the page cache
+	log "creating a 10 Mb file"
+	multiop $DIR/$tfile oO_CREAT:O_DIRECT:O_RDWR:w$((10*1048576))c || error "multiop failed while creating a file"
+	log "starting reads"
+	dd if=$DIR/$tfile of=/dev/null bs=4096 &
+	log "truncating the file"
+	multiop $DIR/$tfile oO_TRUNC:c || error "multiop failed while truncating the file"
+	log "killing dd"
+	kill %+ || true # reads might have finished
+	echo "wait until dd is finished"
+	wait
+	log "removing the temporary file"
+	rm -rf $DIR/$tfile || error "tmp file removal failed"
+}
+run_test 169 "parallel read and truncate should not deadlock ==="
+
 test_170() {
+        $LCTL clear	# bug 18514
         $LCTL debug_daemon start $TMP/${tfile}_log_good
         touch $DIR/$tfile
         $LCTL debug_daemon stop
@@ -6140,40 +6339,40 @@ test_170() {
         rm -rf $DIR/$tfile
         $LCTL debug_daemon stop
 
-        $LCTL df $TMP/${tfile}_log_bad 2&> $TMP/${tfile}_log_bad.out ||
+        $LCTL df $TMP/${tfile}_log_bad > $TMP/${tfile}_log_bad.out 2>&1 ||
                error "lctl df log_bad failed"
 
         local bad_line=$(tail -n 1 $TMP/${tfile}_log_bad.out | awk '{print $9}')
         local good_line1=$(tail -n 1 $TMP/${tfile}_log_bad.out | awk '{print $5}')
 
-        $LCTL df $TMP/${tfile}_log_good 2&>$TMP/${tfile}_log_good.out 
+        $LCTL df $TMP/${tfile}_log_good > $TMP/${tfile}_log_good.out 2>&1
         local good_line2=$(tail -n 1 $TMP/${tfile}_log_good.out | awk '{print $5}')
 
-	[ "$bad_line" ] && [ "$good_line1" ] && [ "$good_line2" ] || 
+	[ "$bad_line" ] && [ "$good_line1" ] && [ "$good_line2" ] ||
 		error "bad_line good_line1 good_line2 are empty"
- 
+
+        cat $TMP/${tfile}_log_good >> $TMP/${tfile}_logs_corrupt
+        cat $TMP/${tfile}_log_bad >> $TMP/${tfile}_logs_corrupt
         cat $TMP/${tfile}_log_good >> $TMP/${tfile}_logs_corrupt
-        cat $TMP/${tfile}_log_bad >> $TMP/${tfile}_logs_corrupt 
-        cat $TMP/${tfile}_log_good >> $TMP/${tfile}_logs_corrupt           
 
-        $LCTL df $TMP/${tfile}_logs_corrupt 2&> $TMP/${tfile}_log_bad.out
+        $LCTL df $TMP/${tfile}_logs_corrupt > $TMP/${tfile}_log_bad.out 2>&1
         local bad_line_new=$(tail -n 1 $TMP/${tfile}_log_bad.out | awk '{print $9}')
         local good_line_new=$(tail -n 1 $TMP/${tfile}_log_bad.out | awk '{print $5}')
 
-	[ "$bad_line_new" ] && [ "$good_line_new" ] || 
+	[ "$bad_line_new" ] && [ "$good_line_new" ] ||
 		error "bad_line_new good_line_new are empty"
- 
+
         local expected_good=$((good_line1 + good_line2*2))
 
-        rm -rf $TMP/${tfile}*
+        rm -f $TMP/${tfile}*
         if [ $bad_line -ne $bad_line_new ]; then
                 error "expected $bad_line bad lines, but got $bad_line_new"
-                return 1 
+                return 1
         fi
 
         if [ $expected_good -ne $good_line_new ]; then
                 error "expected $expected_good good lines, but got $good_line_new"
-                return 2 
+                return 2
         fi
         true
 }
@@ -6314,6 +6513,17 @@ test_212() {
 }
 run_test 212 "Sendfile test ============================================"
 
+test_213() {
+	dd if=/dev/zero of=$DIR/$tfile bs=4k count=4
+	cancel_lru_locks osc
+	lctl set_param fail_loc=0x8000040f
+	# generate a read lock
+	cat $DIR/$tfile > /dev/null
+	# write to the file, it will try to cancel the above read lock.
+	cat /etc/hosts >> $DIR/$tfile
+}
+run_test 213 "OSC lock completion and cancel race don't crash - bug 18829"
+
 #
 # tests that do cleanup/setup should be run at the end
 #