X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity.sh;h=924d05abab8f4756d5e0af02861eee6b1393305d;hb=4b47ec5a8e68950ced4895e1ccfdf2254c016194;hp=d168bd3e60b7861ab5a3ab5381fe964322e4b1a0;hpb=7e0208da21f9358d96feebce5124f8587778c318;p=fs%2Flustre-release.git

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index d168bd3..924d05a 100755
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -63,8 +63,8 @@ if [[ $(uname -m) = aarch64 ]]; then
 	ALWAYS_EXCEPT+=" 45	  317"
 fi
 
-# skip nfs tests on kernels >= 4.14.0 until they are fixed
-if [ $LINUX_VERSION_CODE -ge $(version_code 4.14.0) ]; then
+# skip nfs tests on kernels >= 4.12.0 until they are fixed
+if [ $LINUX_VERSION_CODE -ge $(version_code 4.12.0) ]; then
 	# bug number:	LU-12661
 	ALWAYS_EXCEPT+=" 817"
 fi
@@ -4880,6 +4880,8 @@ test_43A() { # was test_43
 
 	$DIR/$tdir/$tfile && error "execute $DIR/$tdir/$tfile succeeded" || true
 	kill -USR1 $pid
+	# Wait for multiop to exit
+	wait $pid
 }
 run_test 43A "execution of file opened for write should return -ETXTBSY"
 
@@ -5129,6 +5131,27 @@ test_48e() { # bug 4134
 }
 run_test 48e "Access to recreated parent subdir (should return errors)"
 
+test_48f() {
+	[[ $MDS1_VERSION -ge $(version_code 2.13.55) ]] ||
+		skip "need MDS >= 2.13.55"
+	[[ $MDSCOUNT -ge 2 ]] || skip "needs >= 2 MDTs"
+	[[ "$(facet_host mds1)" != "$(facet_host mds2)" ]] ||
+		skip "needs different host for mdt1 mdt2"
+	[[ $(facet_fstype mds1) == ldiskfs ]] || skip "ldiskfs only"
+
+	$LFS mkdir -i0 $DIR/$tdir
+	$LFS mkdir -i 1 $DIR/$tdir/sub1 $DIR/$tdir/sub2 $DIR/$tdir/sub3
+
+	for d in sub1 sub2 sub3; do
+		#define OBD_FAIL_OSD_REF_DEL	0x19c
+		do_facet mds1 $LCTL set_param fail_loc=0x8000019c
+		rm -rf $DIR/$tdir/$d && error "rm $d should fail"
+	done
+
+	rm -d --interactive=never $DIR/$tdir || error "rm $tdir fail"
+}
+run_test 48f "non-zero nlink dir unlink won't LBUG()"
+
 test_49() { # LU-1030
 	[ $PARALLEL == "yes" ] && skip "skip parallel run"
 	remote_ost_nodsh && skip "remote OST with nodsh"
@@ -7694,7 +7717,7 @@ test_60g() {
 
 		do_facet mds$index $LCTL set_param fail_loc=0x8000019a \
 			> /dev/null
-		usleep 100
+		sleep 0.01
 	done
 
 	kill -9 $pid
@@ -8607,36 +8630,45 @@ test_74c() {
 }
 run_test 74c "ldlm_lock_create error path, (shouldn't LBUG)"
 
-num_inodes() {
-	[ -f /sys/kernel/slab/lustre_inode_cache/shrink ] &&
-		echo 1 > /sys/kernel/slab/lustre_inode_cache/shrink
-	awk '/lustre_inode_cache/ {print $2; exit}' /proc/slabinfo
+slab_lic=/sys/kernel/slab/lustre_inode_cache
+num_objects() {
+	[ -f $slab_lic/shrink ] && echo 1 > $slab_lic/shrink
+	[ -f $slab_lic/objects ] && awk '{ print $1 }' $slab_lic/objects ||
+		awk '/lustre_inode_cache/ { print $2; exit }' /proc/slabinfo
 }
 
-test_76() { # Now for bug 20433, added originally in bug 1443
+test_76() { # Now for b=20433, added originally in b=1443
 	[ $PARALLEL == "yes" ] && skip "skip parallel run"
 
 	cancel_lru_locks osc
+	# there may be some slab objects cached per core
 	local cpus=$(getconf _NPROCESSORS_ONLN 2>/dev/null)
-	local before=$(num_inodes)
+	local before=$(num_objects)
 	local count=$((512 * cpus))
-	[ "$SLOW" = "no" ] && count=$((64 * cpus))
+	[ "$SLOW" = "no" ] && count=$((128 * cpus))
+	local margin=$((count / 10))
+	if [[ -f $slab_lic/aliases ]]; then
+		local aliases=$(cat $slab_lic/aliases)
+		(( aliases > 0 )) && margin=$((margin * aliases))
+	fi
 
-	echo "before inodes: $before"
+	echo "before slab objects: $before"
 	for i in $(seq $count); do
 		touch $DIR/$tfile
 		rm -f $DIR/$tfile
 	done
 	cancel_lru_locks osc
-	local after=$(num_inodes)
-	echo "after inodes: $after"
-	while (( after > before + 8 * ${cpus:-1} )); do
+	local after=$(num_objects)
+	echo "created: $count, after slab objects: $after"
+	# shared slab counts are not very accurate, allow significant margin
+	# the main goal is that the cache growth is not permanently > $count
+	while (( after > before + margin )); do
 		sleep 1
-		after=$(num_inodes)
+		after=$(num_objects)
 		wait=$((wait + 1))
-		(( wait % 5 == 0 )) && echo "wait $wait seconds inodes: $after"
-		if (( wait > 30 )); then
-			error "inode slab grew from $before to $after"
+		(( wait % 5 == 0 )) && echo "wait $wait seconds objects: $after"
+		if (( wait > 60 )); then
+			error "inode slab grew from $before+$margin to $after"
 		fi
 	done
 }
@@ -9519,12 +9551,11 @@ test_101g_brw_size_test() {
 		sed -n '/pages per rpc/,/^$/p' |
 		awk '/'$pages':/ { reads += $2; writes += $6 }; \
 		END { print reads,writes }'))
-	[ ${rpcs[0]} -ne $count ] && error "${rpcs[0]} != $count read RPCs" &&
-		return 5
-	[ ${rpcs[1]} -ne $count ] && error "${rpcs[1]} != $count write RPCs" &&
-		return 6
-
-	return 0
+	# allow one extra full-sized read RPC for async readahead
+	[[ ${rpcs[0]} == $count || ${rpcs[0]} == $((count + 1)) ]] ||
+		{ error "${rpcs[0]} != $count read RPCs"; return 5; }
+	[[ ${rpcs[1]} == $count ]] ||
+		{ error "${rpcs[1]} != $count write RPCs"; return 6; }
 }
 
 test_101g() {
@@ -11782,18 +11813,18 @@ test_124a() {
 		skip "Limit is too small $LIMIT"
 	fi
 
-        # Make LVF so higher that sleeping for $SLEEP is enough to _start_
-        # killing locks. Some time was spent for creating locks. This means
-        # that up to the moment of sleep finish we must have killed some of
-        # them (10-100 locks). This depends on how fast ther were created.
-        # Many of them were touched in almost the same moment and thus will
-        # be killed in groups.
-        local LVF=$(($MAX_HRS * 60 * 60 / $SLEEP * $LIMIT / $LRU_SIZE))
-
-        # Use $LRU_SIZE_B here to take into account real number of locks
-        # created in the case of CMD, LRU_SIZE_B != $NR in most of cases
-        local LRU_SIZE_B=$LRU_SIZE
-        log "LVF=$LVF"
+	# Make LVF so higher that sleeping for $SLEEP is enough to _start_
+	# killing locks. Some time was spent for creating locks. This means
+	# that up to the moment of sleep finish we must have killed some of
+	# them (10-100 locks). This depends on how fast ther were created.
+	# Many of them were touched in almost the same moment and thus will
+	# be killed in groups.
+	local LVF=$(($MAX_HRS * 60 * 60 / $SLEEP * $LIMIT / $LRU_SIZE * 100))
+
+	# Use $LRU_SIZE_B here to take into account real number of locks
+	# created in the case of CMD, LRU_SIZE_B != $NR in most of cases
+	local LRU_SIZE_B=$LRU_SIZE
+	log "LVF=$LVF"
 	local OLD_LVF=$($LCTL get_param -n $NSDIR.pool.lock_volume_factor)
 	log "OLD_LVF=$OLD_LVF"
 	$LCTL set_param -n $NSDIR.pool.lock_volume_factor $LVF
@@ -14981,6 +15012,71 @@ test_160k() {
 }
 run_test 160k "Verify that changelog records are not lost"
 
+# Verifies that a file passed as a parameter has recently had an operation
+# performed on it that has generated an MTIME changelog which contains the
+# correct parent FID. As files might reside on a different MDT from the
+# parent directory in DNE configurations, the FIDs are translated to paths
+# before being compared, which should be identical
+compare_mtime_changelog() {
+	local file="${1}"
+	local mdtidx
+	local mtime
+	local cl_fid
+	local pdir
+	local dir
+
+	mdtidx=$($LFS getstripe --mdt-index $file)
+	mdtidx=$(printf "%04x" $mdtidx)
+
+	# Obtain the parent FID from the MTIME changelog
+	mtime=$($LFS changelog $FSNAME-MDT$mdtidx | tail -n 1 | grep MTIME)
+	[ -z "$mtime" ] && error "MTIME changelog not recorded"
+
+	cl_fid=$(sed -e 's/.* p=//' -e 's/ .*//' <<<$mtime)
+	[ -z "$cl_fid" ] && error "parent FID not present"
+
+	# Verify that the path for the parent FID is the same as the path for
+	# the test directory
+	pdir=$($LFS fid2path $MOUNT "$cl_fid")
+
+	dir=$(dirname $1)
+
+	[[ "${pdir%/}" == "$dir" ]] ||
+		error "MTIME changelog parent FID is wrong, expected $dir, got $pdir"
+}
+
+test_160l() {
+	[ $PARALLEL == "yes" ] && skip "skip parallel run"
+
+	remote_mds_nodsh && skip "remote MDS with nodsh"
+	[[ $MDS1_VERSION -ge $(version_code 2.13.55) ]] ||
+		skip "Need MDS version at least 2.13.55"
+
+	local cl_user
+
+	changelog_register || error "changelog_register failed"
+	cl_user="${CL_USERS[$SINGLEMDS]%% *}"
+
+	changelog_users $SINGLEMDS | grep -q $cl_user ||
+		error "User '$cl_user' not found in changelog_users"
+
+	# Clear some types so that MTIME changelogs are generated
+	changelog_chmask "-CREAT"
+	changelog_chmask "-CLOSE"
+
+	test_mkdir $DIR/$tdir || error "failed to mkdir $DIR/$tdir"
+
+	# Test CL_MTIME during setattr
+	touch $DIR/$tdir/$tfile
+	compare_mtime_changelog $DIR/$tdir/$tfile
+
+	# Test CL_MTIME during close
+	dd if=/dev/urandom of=$DIR/$tdir/${tfile}_2 bs=1M count=64 ||
+		error "cannot create file $DIR/$tdir/${tfile}_2"
+	compare_mtime_changelog $DIR/$tdir/${tfile}_2
+}
+run_test 160l "Verify that MTIME changelog records contain the parent FID"
+
 test_161a() {
 	[ $PARALLEL == "yes" ] && skip "skip parallel run"
 
@@ -15690,7 +15786,8 @@ obdecho_test() {
 test_180a() {
 	[ $PARALLEL == "yes" ] && skip "skip parallel run"
 
-	if ! module_loaded obdecho; then
+	if ! [ -d /sys/fs/lustre/echo_client ] &&
+	   ! module_loaded obdecho; then
 		load_module obdecho/obdecho &&
 			stack_trap "rmmod obdecho" EXIT ||
 			error "unable to load obdecho on client"
@@ -17214,6 +17311,34 @@ test_226b () {
 }
 run_test 226b "call path2fid and fid2path on files of all type under remote dir"
 
+test_226c () {
+	[ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs"
+	[[ $MDS1_VERSION -ge $(version_code 2.13.55) ]] ||
+		skip "Need MDS version at least 2.13.55"
+
+	local submnt=/mnt/submnt
+	local srcfile=/etc/passwd
+	local dstfile=$submnt/passwd
+	local path
+	local fid
+
+	rm -rf $DIR/$tdir
+	rm -rf $submnt
+	$LFS setdirstripe -c -1 -i 1 $DIR/$tdir ||
+		error "create remote directory failed"
+	mkdir -p $submnt || error "create $submnt failed"
+	$MOUNT_CMD $MGSNID:/$FSNAME/$tdir $submnt ||
+		error "mount $submnt failed"
+	stack_trap "umount $submnt" EXIT
+
+	cp $srcfile $dstfile
+	fid=$($LFS path2fid $dstfile)
+	path=$($LFS fid2path $submnt "$fid")
+	[ "$path" = "$dstfile" ] ||
+		error "fid2path $submnt $fid failed ($path != $dstfile)"
+}
+run_test 226c "call path2fid and fid2path under remote dir with subdir mount"
+
 # LU-1299 Executing or running ldd on a truncated executable does not
 # cause an out-of-memory condition.
 test_227() {
@@ -18214,6 +18339,41 @@ test_230q() {
 }
 run_test 230q "dir auto split"
 
+test_230r() {
+	[[ $PARALLEL != "yes" ]] || skip "skip parallel run"
+	[[ $MDSCOUNT -ge 2 ]] || skip_env "needs >= 2 MDTs"
+	[[ $MDS1_VERSION -ge $(version_code 2.13.54) ]] ||
+		skip "Need MDS version at least 2.13.54"
+
+	# maximum amount of local locks:
+	# parent striped dir - 2 locks
+	# new stripe in parent to migrate to - 1 lock
+	# source and target - 2 locks
+	# Total 5 locks for regular file
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i1 -c2 $DIR/$tdir/dir1
+	touch $DIR/$tdir/dir1/eee
+
+	# create 4 hardlink for 4 more locks
+	# Total: 9 locks > RS_MAX_LOCKS (8)
+	$LFS mkdir -i1 -c1 $DIR/$tdir/dir2
+	$LFS mkdir -i1 -c1 $DIR/$tdir/dir3
+	$LFS mkdir -i1 -c1 $DIR/$tdir/dir4
+	$LFS mkdir -i1 -c1 $DIR/$tdir/dir5
+	ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir2/eee
+	ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir3/eee
+	ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir4/eee
+	ln $DIR/$tdir/dir1/eee $DIR/$tdir/dir5/eee
+
+	cancel_lru_locks mdc
+
+	$LFS migrate -m1 -c1 $DIR/$tdir/dir1 ||
+		error "migrate dir fails"
+
+	rm -rf $DIR/$tdir || error "rm dir failed after migration"
+}
+run_test 230r "migrate with too many local locks"
+
 test_231a()
 {
 	# For simplicity this test assumes that max_pages_per_rpc
@@ -21887,6 +22047,11 @@ test_398d() { #  LU-13846
 	aiocp -a $PAGE_SIZE -b 64M -s 64M -f O_DIRECT $DIR/$tfile $aio_file
 
 	diff $DIR/$tfile $aio_file || "file diff after aiocp"
+
+	# make sure we don't crash and fail properly
+	aiocp -a 512 -b 64M -s 64M -f O_DIRECT $DIR/$tfile $aio_file &&
+		error "aio not aligned with PAGE SIZE should fail"
+
 	rm -rf $DIR/$tfile $aio_file
 }
 run_test 398d "run aiocp to verify block size > stripe size"
@@ -22356,6 +22521,8 @@ test_410()
 {
 	[[ $CLIENT_VERSION -lt $(version_code 2.9.59) ]] &&
 		skip "Need client version at least 2.9.59"
+	[ -f $LUSTRE/tests/kernel/kinode.ko ] ||
+		skip "Need MODULES build"
 
 	# Create a file, and stat it from the kernel
 	local testfile=$DIR/$tfile
@@ -23203,6 +23370,41 @@ test_424() {
 }
 run_test 424 "simulate ENOMEM in ptl_send_rpc bulk reply ME attach"
 
+test_425() {
+	test_mkdir -c -1 $DIR/$tdir
+	$LFS setstripe -c -1 $DIR/$tdir
+
+	lru_resize_disable "" 100
+	stack_trap "lru_resize_enable" EXIT
+
+	sleep 5
+
+	for i in $(seq $((MDSCOUNT * 125))); do
+		local t=$DIR/$tdir/$tfile_$i
+
+		dd if=/dev/zero of=$t bs=4K count=1 > /dev/null 2>&1 ||
+			error_noexit "Create file $t"
+	done
+	stack_trap "rm -rf $DIR/$tdir" EXIT
+
+	for oscparam in $($LCTL list_param ldlm.namespaces.*osc-[-0-9a-f]*); do
+		local lru_size=$($LCTL get_param -n $oscparam.lru_size)
+		local lock_count=$($LCTL get_param -n $oscparam.lock_count)
+
+		[ $lock_count -le $lru_size ] ||
+			error "osc lock count $lock_count > lru size $lru_size"
+	done
+
+	for mdcparam in $($LCTL list_param ldlm.namespaces.*mdc-*); do
+		local lru_size=$($LCTL get_param -n $mdcparam.lru_size)
+		local lock_count=$($LCTL get_param -n $mdcparam.lock_count)
+
+		[ $lock_count -le $lru_size ] ||
+			error "mdc lock count $lock_count > lru size $lru_size"
+	done
+}
+run_test 425 "lock count should not exceed lru size"
+
 prep_801() {
 	[[ $MDS1_VERSION -lt $(version_code 2.9.55) ]] ||
 	[[ $OST1_VERSION -lt $(version_code 2.9.55) ]] &&