X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=sidebyside;f=lustre%2Ftests%2Fsanityn.sh;h=1170b2d8d7ea52a259a3c99ec14261b0758e43f5;hb=55c468ce5dcc2bd59b4452098918475161e8349f;hp=aaf9035ba6fc4707528aa5f0aaae9f070180cb2f;hpb=ad720ef6863538c83d4b74d6caef2d32e32a7f9b;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh
index aaf9035..1170b2d 100644
--- a/lustre/tests/sanityn.sh
+++ b/lustre/tests/sanityn.sh
@@ -41,8 +41,8 @@ init_test_env $@
 init_logging
 
 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test:        LU-2840 LU-2776
-	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21      51a"
+# bug number for skipped test:        LU-2840 LU-2189 LU-2776
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21      36      51a"
 
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="33a"
 
@@ -57,7 +57,7 @@ LOVNAME=$($LCTL get_param -n llite.*.lov.common_name | tail -n 1)
 OSTCOUNT=$($LCTL get_param -n lov.$LOVNAME.numobd)
 
 assert_DIR
-rm -rf $DIR1/[df][0-9]* $DIR1/lnk
+rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}*
 
 SAMPLE_FILE=$TMP/$(basename $0 .sh).junk
 dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=1
@@ -489,28 +489,28 @@ test_21() { # Bug 5907
 run_test 21 " Try to remove mountpoint on another dir ===="
 
 test_23() { # Bug 5972
-	echo "others should see updated atime while another read" > $DIR1/f23
-	
+	local at_diff=$(do_facet $SINGLEMDS \
+		$LCTL get_param -n mdd.*MDT0000*.atime_diff | head -1)
+	echo "atime should be updated while another read" > $DIR1/$tfile
+
 	# clear the lock(mode: LCK_PW) gotten from creating operation
 	cancel_lru_locks osc
-	
-	time1=`date +%s`	
-	#MAX_ATIME_DIFF 60, we update atime only if older than 60 seconds
-	sleep 61
-	
-	multiop_bg_pause $DIR1/f23 or20_c || return 1
+	time1=$(date +%s)
+	echo "now is $time1"
+	sleep $((at_diff + 1))
+
+	echo "starting reads"
+	multiop_bg_pause $DIR1/$tfile or20_c || return 1
         # with SOM and opencache enabled, we need to close a file and cancel
         # open lock to get atime propogated to MDS
-        kill -USR1 $!
+        kill -USR1 $! || return 2
         cancel_lru_locks mdc
 
-	time2=`stat -c "%X" $DIR2/f23`
+	time2=$(stat -c "%X" $DIR/$tfile)
+	echo "new atime is $time2"
 
-	if (( $time2 <= $time1 )); then
-		error "atime doesn't update among nodes"
-	fi
-
-	rm -f $DIR1/f23 || error "rm -f $DIR1/f23 failed"
+	[ $time2 -gt $time1 ] || error "atime was not updated"
+	rm -f $DIR1/$tfile || error "rm -f $DIR1/$tfile failed"
 	true
 }
 run_test 23 " others should see updated atime while another read===="
@@ -523,7 +523,7 @@ test_24a() {
 	lfs df -i $DIR2 || error "lfs df -i $DIR2 failed"
 	lfs df $DIR1/$tfile || error "lfs df $DIR1/$tfile failed"
 	lfs df -ih $DIR2/$tfile || error "lfs df -ih $DIR2/$tfile failed"
-	
+
 	OSC=`lctl dl | awk '/-osc-|OSC.*MNT/ {print $4}' | head -n 1`
 #	OSC=`lctl dl | awk '/-osc-/ {print $4}' | head -n 1`
 	lctl --device %$OSC deactivate
@@ -742,8 +742,8 @@ enable_lockless_truncate() {
 }
 
 test_32a() { # bug 11270
-        local p="$TMP/sanityN-$TESTNAME.parameters"
-        save_lustre_params $HOSTNAME osc.*.lockless_truncate > $p
+	local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
+	save_lustre_params client "osc.*.lockless_truncate" > $p
         cancel_lru_locks osc
         enable_lockless_truncate 1
         rm -f $DIR1/$tfile
@@ -781,15 +781,19 @@ run_test 32a "lockless truncate"
 test_32b() { # bug 11270
         remote_ost_nodsh && skip "remote OST with nodsh" && return
 
-        local node
-        local p="$TMP/sanityN-$TESTNAME.parameters"
-        save_lustre_params $HOSTNAME "osc.*.contention_seconds" > $p
-        for node in $(osts_nodes); do
-                save_lustre_params $node "ldlm.namespaces.filter-*.max_nolock_bytes" >> $p
-                save_lustre_params $node "ldlm.namespaces.filter-*.contended_locks" >> $p
-                save_lustre_params $node "ldlm.namespaces.filter-*.contention_seconds" >> $p
-        done
-        clear_osc_stats
+	local node
+	local facets=$(get_facets OST)
+	local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
+
+	save_lustre_params client "osc.*.contention_seconds" > $p
+	save_lustre_params $facets \
+		"ldlm.namespaces.filter-*.max_nolock_bytes" >> $p
+	save_lustre_params $facets \
+		"ldlm.namespaces.filter-*.contended_locks" >> $p
+	save_lustre_params $facets \
+		"ldlm.namespaces.filter-*.contention_seconds" >> $p
+	clear_osc_stats
+
         # agressive lockless i/o settings
         for node in $(osts_nodes); do
                 do_node $node 'lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes 2000000; lctl set_param -n ldlm.namespaces.filter-*.contended_locks 0; lctl set_param -n ldlm.namespaces.filter-*.contention_seconds 60'
@@ -851,7 +855,8 @@ test_33a() {
     local param_file=$TMP/$tfile-params
     local fstype=$(facet_fstype $SINGLEMDS)
 
-    save_lustre_params $(comma_list $(mdts_nodes)) "mdt.*.commit_on_sharing" > $param_file
+	save_lustre_params $(get_facets MDS) \
+		"mdt.*.commit_on_sharing" > $param_file
 
     local COS
     local jbdold="N/A"
@@ -902,8 +907,9 @@ test_33b() {
 	local nfiles=${TEST33_NFILES:-10000}
 	local param_file=$TMP/$tfile-params
 
-	save_lustre_params $(comma_list $(mdts_nodes)) \
-				"mdt.*.commit_on_sharing" > $param_file
+	save_lustre_params $(get_facets MDS) \
+		"mdt.*.commit_on_sharing" > $param_file
+
 	local COS
 	local jbdold
 	local jbdnew
@@ -1082,33 +1088,29 @@ test_36() { #bug 16417
 	i=0
 	SIZE=50
 	let SIZE_B=SIZE*1024*1024
+	sync; sleep 5; sync # wait for delete thread
 
 	while [ $i -le 10 ]; do
 		lctl mark "start test"
-		local before=$($LFS df | awk '{ if ($1 ~/^filesystem/) \
-					      { print $5; exit} }')
+		local before=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
 		dd if=/dev/zero of=$DIR1/$tdir/$tfile bs=1M count=$SIZE ||
 			error "dd $DIR1/$tdir/$tfile ${SIZE}MB failed"
 		sync          # sync data from client cache
 		sync_all_data # sync data from server cache (delayed allocation)
-		sleep 1
-		local after_dd=$($LFS df | awk '{ if ($1 ~/^filesystem/) \
-						{ print $5; exit} }')
+		sleep 2
+		local after_dd=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
 		multiop_bg_pause $DIR2/$tdir/$tfile O_r${SIZE_B}c || return 3
 		read_pid=$!
 		rm -f $DIR1/$tdir/$tfile
 		kill -USR1 $read_pid
 		wait $read_pid
 		wait_delete_completed
-		local after=$($LFS df | awk '{ if ($1 ~/^filesystem/) \
-					     { print $5; exit} }')
+		local after=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
 		echo "*** cycle($i) *** before($before) after_dd($after_dd)" \
 			"after($after)"
 		# this free space! not used
-		if [ $after_dd -ge $after ]; then
-			error "space leaked"
-			return 1;
-		fi
+		(( $after_dd <= $after)) ||
+			error "space leaked after_dd:$after_dd > after:$after"
 		let i=i+1
 	done
 }
@@ -2299,6 +2301,9 @@ test_51a() {
 run_test 51a "layout lock: refresh layout should work"
 
 test_51b() {
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.59) ]] ||
+		{ skip "Need MDS version at least 2.3.59"; return 0; }
+
 	local tmpfile=`mktemp`
 
 	# create an empty file
@@ -2430,6 +2435,114 @@ test_70b() { # LU-2781
 }
 run_test 70b "remove files after calling rm_entry"
 
+test_71() {
+	local server_version=$(lustre_version_code $SINGLEMDS)
+
+	[[ $server_version -lt $(version_code 2.1.6) ]] &&
+		skip "Need MDS version at least 2.1.6" && return
+
+	# Patch not applied to 2.2 and 2.3 branches
+	[[ $server_version -ge $(version_code 2.2.0) ]] &&
+	[[ $server_version -lt $(version_code 2.4.0) ]] &&
+		skip "Need MDS version at least 2.4.0" && return
+
+	checkfiemap --test ||
+		{ skip "checkfiemap not runnable: $?" && return; }
+	# write data this way: hole - data - hole - data
+	dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1
+	[ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tfile) + 1)))" = \
+		"zfs" ] &&
+		skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
+	dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1
+	GET_STAT="lctl get_param -n ldlm.services.ldlm_cbd.stats"
+	stat $DIR2/$tfile
+	local can1=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
+	echo $can1
+	checkfiemap $DIR2/$tfile 81920 ||
+		error "data is not flushed from client"
+	local can2=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
+	echo $can2
+
+	# common case of "create file, copy file" on a single node
+	# should not flush data from ost
+	dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1
+	dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1
+	stat $DIR1/$tfile
+	local can3=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
+	echo $can3
+	checkfiemap $DIR1/$tfile 81920 ||
+	error 4
+	local can4=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
+	echo $can2
+	[ $can3 -eq $can4 ] || error $((can2-can1)) "cancel RPC occured."
+}
+run_test 71 "correct file map just after write operation is finished"
+
+test_72() {
+	local p="$TMP/sanityN-$TESTNAME.parameters"
+	local tlink1
+	local tlink2
+	save_lustre_params client "llite.*.xattr_cache" > $p
+	lctl set_param llite.*.xattr_cache 1 ||
+		{ skip "xattr cache is not supported"; return 0; }
+
+	touch $DIR1/$tfile
+	setfattr -n user.attr1 -v value1 $DIR1/$tfile ||
+		error "setfattr1 failed"
+	getfattr -n user.attr1 $DIR2/$tfile | grep value1 ||
+		error "getfattr1 failed"
+	setfattr -n user.attr1 -v value2 $DIR2/$tfile ||
+		error "setfattr2 failed"
+	getfattr -n user.attr1 $DIR1/$tfile | grep value2 ||
+		error "getfattr2 failed"
+
+	# check that trusted.link is consistent
+	tlink1=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
+	ln $DIR2/$tfile $DIR2/$tfile-2 || error "failed to link"
+	tlink2=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
+	echo "$tlink1 $tlink2"
+	[ "$tlink1" = "$tlink2" ] && error "trusted.link should have changed!"
+
+	rm -f $DIR2/$tfile
+
+	restore_lustre_params < $p
+	rm -f $p
+}
+run_test 72 "getxattr/setxattr cache should be consistent between nodes"
+
+test_73() {
+	local p="$TMP/sanityN-$TESTNAME.parameters"
+	save_lustre_params client "llite.*.xattr_cache" > $p
+	lctl set_param llite.*.xattr_cache 1 ||
+		{ skip "xattr cache is not supported"; return 0; }
+
+	touch $DIR1/$tfile
+	setfattr -n user.attr1 -v value1 $DIR1/$tfile ||
+		error "setfattr1 failed"
+	getfattr -n user.attr1 $DIR2/$tfile || error "getfattr1 failed"
+	getfattr -n user.attr1 $DIR1/$tfile || error "getfattr2 failed"
+	clear_llite_stats
+	# PR lock should be cached by now on both clients
+	getfattr -n user.attr1 $DIR1/$tfile || error "getfattr3 failed"
+	# 2 hits for getfattr(0)+getfattr(size)
+	[ $(calc_llite_stats getxattr_hits) -eq 2 ] || error "not cached in $DIR1"
+	getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed"
+	# 4 hits for more getfattr(0)+getfattr(size)
+	[ $(calc_llite_stats getxattr_hits) -eq 4 ] || error "not cached in $DIR2"
+	rm -f $DIR2/$tfile
+
+	restore_lustre_params < $p
+	rm -f $p
+}
+run_test 73 "getxattr should not cause xattr lock cancellation"
+
+test_74() {
+	dd if=/dev/zero of=$DIR1/$tfile-1 bs=1K count=1
+	dd if=/dev/zero of=$DIR1/$tfile-2 bs=1K count=1
+	flocks_test 4 $DIR1/$tfile-1 $DIR2/$tfile-2
+}
+run_test 74 "flock deadlock: different mounts =============="
+
 log "cleanup: ======================================================"
 
 [ "$(mount | grep $MOUNT2)" ] && umount $MOUNT2