X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=03ce07c86e26fa7a002e1cc7eb843a6e420351c7;hp=62d20516cd8c49c6967329a581956ccdaded095e;hb=bebf1e3f09144ca59cc6b0c2b61c65290626ada8;hpb=aa6773d95b0212e5bfe2a187e679322c1ac9eb8e

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 62d2051..03ce07c 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -20,8 +20,8 @@ require_dsh_mds || exit 0
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="17  26a  26b    50   51     57"
 
 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test:	      LU-2194 LU-2547
-	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 19b     24a 24b"
+# bug number for skipped test:	      LU-2547
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 24a 24b"
 
 build_test_filter
 
@@ -149,21 +149,140 @@ test_9() {
 run_test 9 "pause bulk on OST (bug 1420)"
 
 #bug 1521
-test_10() {
-	do_facet client mcreate $DIR/$tfile ||
-		{ error "mcreate failed: $?"; return 1; }
-	drop_bl_callback "chmod 0777 $DIR/$tfile"  || echo "evicted as expected"
-	# wait for the mds to evict the client
-	#echo "sleep $(($TIMEOUT*2))"
-	#sleep $(($TIMEOUT*2))
-	do_facet client touch $DIR/$tfile || echo "touch failed, evicted"
-	do_facet client checkstat -v -p 0777 $DIR/$tfile ||
-		{ error "client checkstat failed: $?"; return 3; }
-	do_facet client "munlink $DIR/$tfile"
-	# allow recovery to complete
-	client_up || client_up || sleep $TIMEOUT
+test_10a() {
+	local before=$(date +%s)
+	local evict
+
+	do_facet client "stat $DIR > /dev/null"  ||
+		error "failed to stat $DIR: $?"
+	drop_bl_callback "chmod 0777 $DIR" ||
+		error "failed to chmod $DIR: $?"
+
+	# let the client reconnect
+	client_reconnect
+	evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
+	[ ! -z "$evict" ] && [[ $evict -gt $before ]] ||
+		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "no eviction: $evict before:$before")
+
+	do_facet client checkstat -v -p 0777 $DIR ||
+		error "client checkstat failed: $?"
+}
+run_test 10a "finish request on server after client eviction (bug 1521)"
+
+test_10b() {
+	local before=$(date +%s)
+	local evict
+
+	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.53) ]] &&
+		skip "Need MDS version at least 2.6.53" && return
+	do_facet client "stat $DIR > /dev/null"  ||
+		error "failed to stat $DIR: $?"
+	drop_bl_callback_once "chmod 0777 $DIR" ||
+		error "failed to chmod $DIR: $?"
+
+	# let the client reconnect
+	client_reconnect
+	evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
+
+	[ -z "$evict" ] || [[ $evict -le $before ]] ||
+		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "eviction happened: $evict before:$before")
+
+	do_facet client checkstat -v -p 0777 $DIR ||
+ 		error "client checkstat failed: $?"
+}
+run_test 10b "re-send BL AST"
+
+test_10c() {
+	local before=$(date +%s)
+	local evict
+	local mdccli
+	local mdcpath
+	local conn_uuid
+	local workdir
+	local pid
+	local rc
+
+	workdir="${DIR}/${tdir}"
+	mkdir -p ${workdir} || error "can't create workdir $?"
+	stat ${workdir} > /dev/null ||
+		error "failed to stat ${workdir}: $?"
+	mdtidx=$($LFS getdirstripe -i ${workdir})
+	mdtname=$($LFS mdts ${workdir} | grep -e "^$mdtidx:" |
+		  awk '{sub("_UUID", "", $2); print $2;}')
+	#assume one client
+	mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
+	conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
+	mdcpath="mdc.${mdccli}.import=connection=${conn_uuid}"
+
+	drop_bl_callback_once "chmod 0777 ${workdir}" &
+	pid=$!
+
+	# let chmod blocked
+	sleep 1
+	# force client reconnect
+	$LCTL set_param "${mdcpath}"
+
+	# wait client reconnect
+	client_reconnect
+	wait $pid
+	rc=$?
+	evict=$($LCTL get_param mdc.${mdccli}.state |
+	   awk -F"[ [,]" '/EVICTED]$/ { if (t<$4) {t=$4;} } END { print t }')
+
+	[[ $evict -le $before ]] ||
+		( $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "eviction happened: $EVICT before:$BEFORE" )
+
+	[ $rc -eq 0 ] || error "chmod must finished OK"
+	checkstat -v -p 0777 "${workdir}" ||
+		error "client checkstat failed: $?"
 }
-run_test 10 "finish request on server after client eviction (bug 1521)"
+run_test 10c "re-send BL AST vs reconnect race (LU-5569)"
+
+test_10d() {
+	local before=$(date +%s)
+	local evict
+
+	[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ]] &&
+		skip "Need MDS version at least 2.6.90" && return
+
+	# sleep 1 is to make sure that BEFORE is not equal to EVICTED below
+	sleep 1
+	rm -f $TMP/$tfile
+	echo -n ", world" | dd of=$TMP/$tfile bs=1c seek=5
+
+	mount_client $MOUNT2
+
+	cancel_lru_locks osc
+	$LFS setstripe -i 0 -c 1 $DIR1/$tfile
+	echo -n hello > $DIR1/$tfile
+
+	stat $DIR2/$tfile >& /dev/null
+	$LCTL set_param fail_err=71
+	drop_bl_callback "echo -n \\\", world\\\" >> $DIR2/$tfile"
+
+	client_reconnect
+
+	cmp $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
+	cmp $DIR1/$tfile $TMP/$tfile || error "wrong content found"
+
+	evict=$(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state | \
+		tr -d '\-\[\] ' | \
+	  awk -F"[ [,]" '/EVICTED$/ { if (mx<$1) {mx=$1;} } END { print mx }')
+
+	[[ $evict -gt $before ]] ||
+		(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state;
+		    error "no eviction: $evict before:$before")
+
+	$LCTL set_param fail_err=0
+	rm $TMP/$tfile
+	umount_client $MOUNT2
+}
+run_test 10d "test failed blocking ast"
 
 #bug 2460
 # wake up a thread waiting for completion after eviction
@@ -177,7 +296,8 @@ test_11(){
 
 	do_facet client $MULTIOP $DIR/$tfile or  ||
 		{ error "multiop read failed: $?"; return 3; }
-	drop_bl_callback $MULTIOP $DIR/$tfile Ow || echo "evicted as expected"
+	drop_bl_callback_once $MULTIOP $DIR/$tfile Ow ||
+		echo "evicted as expected"
 
 	do_facet client munlink $DIR/$tfile ||
 		{ error "munlink failed: $?"; return 4; }
@@ -492,7 +612,7 @@ test_19c() {
 	# let the client reconnect
 	sleep 5
 	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
-	   awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
 
 	[ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
 }
@@ -911,7 +1031,8 @@ run_test 27 "fail LOV while using OSC's"
 
 test_28() {      # bug 6086 - error adding new clients
 	do_facet client mcreate $DIR/$tfile       || return 1
-	drop_bl_callback "chmod 0777 $DIR/$tfile" ||echo "evicted as expected"
+	drop_bl_callback_once "chmod 0777 $DIR/$tfile" ||
+		echo "evicted as expected"
 	#define OBD_FAIL_MDS_CLIENT_ADD 0x12f
 	do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
 	# fail once (evicted), reconnect fail (fail_loc), ok
@@ -1163,10 +1284,10 @@ test_58() { # bug 11546
         pid=$!
         sleep 1
         lctl set_param fail_loc=0
-        drop_bl_callback rm -f $DIR/$tfile
+        drop_bl_callback_once rm -f $DIR/$tfile
         wait $pid
         # the first 'df' could tigger the eviction caused by
-        # 'drop_bl_callback', and it's normal case.
+        # 'drop_bl_callback_once', and it's normal case.
         # but the next 'df' should return successfully.
         do_facet client "df $DIR" || do_facet client "df $DIR"
 }
@@ -1265,12 +1386,14 @@ test_61()
 
 	replay_barrier $SINGLEMDS
 	createmany -o $DIR/$tdir/$tfile-%d 10 
-	local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
+	local oid=$(do_facet ost1 "lctl get_param -n \
+		obdfilter.${ost1_svc}.last_id" | sed -e 's/.*://')
 
 	fail_abort $SINGLEMDS
-	
+
 	touch $DIR/$tdir/$tfile
-	local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'`
+	local id=$($LFS getstripe $DIR/$tdir/$tfile |
+		awk '$1 == 0 { print $2 }')
 	[ $id -le $oid ] && error "the orphan objid was reused, failed"
 
 	# Cleanup
@@ -1288,6 +1411,72 @@ run_test 61 "Verify to not reuse orphan objects - bug 17025"
 #}
 #run_test 62 "Verify connection flags race - bug LU-1716"
 
+test_66()
+{
+	[[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] ||
+		{ skip "Need MDS version at least 2.7.51"; return 0; }
+
+	local list=$(comma_list $(osts_nodes))
+
+	# modify dir so that next revalidate would not obtain UPDATE lock
+	touch $DIR
+
+	# drop 1 reply with UPDATE lock
+	mcreate $DIR/$tfile || error "mcreate failed: $?"
+	drop_ldlm_reply_once "stat $DIR/$tfile" &
+	sleep 2
+
+	# make the re-sent lock to sleep
+#define OBD_FAIL_MDS_RESEND              0x136
+	do_nodes $list $LCTL set_param fail_loc=0x80000136
+
+	#initiate the re-connect & re-send
+	local mdccli=$($LCTL dl | awk '/-mdc-/ {print $4;}')
+	local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
+	$LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
+	sleep 2
+
+	#initiate the client eviction while enqueue re-send is in progress
+	mds_evict_client
+
+	client_reconnect
+	wait
+}
+run_test 66 "lock enqueue re-send vs client eviction"
+
+test_65() {
+	mount_client $DIR2
+
+	#grant lock1, export2
+	$SETSTRIPE -i -0 $DIR2/$tfile || return 1
+	$MULTIOP $DIR2/$tfile Ow  || return 2
+
+#define OBD_FAIL_LDLM_BL_EVICT            0x31e
+	do_facet ost $LCTL set_param fail_loc=0x31e
+	#get waiting lock2, export1
+	$MULTIOP $DIR/$tfile Ow &
+	PID1=$!
+	# let enqueue to get asleep
+	sleep 2
+
+	#get lock2 blocked
+	$MULTIOP $DIR2/$tfile Ow &
+	PID2=$!
+	sleep 2
+
+	#evict export1
+	ost_evict_client
+
+	sleep 2
+	do_facet ost $LCTL set_param fail_loc=0
+
+	wait $PID1
+	wait $PID2
+
+	umount_client $DIR2
+}
+run_test 65 "lock enqueue for destroyed export"
+
 check_cli_ir_state()
 {
         local NODE=${1:-$HOSTNAME}
@@ -1568,10 +1757,9 @@ test_105()
         # get one of the clients from client list
         local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
 
-        local old_MOUNTOPT=$MOUNTOPT
-        MOUNTOPT=${MOUNTOPT},noir
+        local mount_opts=${MOUNT_OPTS:+$MOUNT_OPTS,}noir
         zconf_umount $rcli $MOUNT || error "umount failed"
-        zconf_mount $rcli $MOUNT || error "mount failed"
+        zconf_mount $rcli $MOUNT $mount_opts || error "mount failed"
 
         # make sure lustre mount at $rcli disabling IR
         local ir_state=$(check_cli_ir_state $rcli)
@@ -1593,8 +1781,7 @@ test_105()
 	[ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
 		error "IR status on ost1 should be DISABLED"
 
-        # restore it
-        MOUNTOPT=$old_MOUNTOPT
+        # remount with the default MOUNT_OPTS
         zconf_umount $rcli $MOUNT || error "umount failed"
         zconf_mount $rcli $MOUNT || error "mount failed"
 
@@ -1683,6 +1870,23 @@ test_107 () {
 }
 run_test 107 "drop reint reply, then restart MDT"
 
+test_108() {
+	mkdir -p $DIR/$tdir
+	$SETSTRIPE -c 1 -i 0 $DIR/$tdir
+
+	dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
+	local dd_pid=$!
+	sleep 0.1
+
+	ost_evict_client
+
+	wait $dd_pid
+
+	client_up || error "reconnect failed"
+	rm -f $DIR/$tdir/$tfile
+}
+run_test 108 "client eviction don't crash"
+
 test_110a () {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
 	local remote_dir=$DIR/$tdir/remote_dir
@@ -1792,9 +1996,9 @@ test_110g () {
 
 	createmany -o $remote_dir/f 100
 
-	#define OBD_FAIL_MIGRATE_NET_REP	0x1702
-	do_facet mds$MDTIDX lctl set_param fail_loc=0x1702
-	$LFS mv -M $MDTIDX $remote_dir || error "migrate failed"
+	#define OBD_FAIL_MIGRATE_NET_REP		0x1800
+	do_facet mds$MDTIDX lctl set_param fail_loc=0x1800
+	$LFS migrate -m $MDTIDX $remote_dir || error "migrate failed"
 	do_facet mds$MDTIDX lctl set_param fail_loc=0x0
 
 	for file in $(find $remote_dir); do
@@ -1807,6 +2011,77 @@ test_110g () {
 }
 run_test 110g "drop reply during migration"
 
+test_110h () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local src_dir=$DIR/$tdir/source_dir
+	local tgt_dir=$DIR/$tdir/target_dir
+	local MDTIDX=1
+
+	mkdir -p $src_dir
+	$LFS mkdir -i $MDTIDX $tgt_dir
+
+	dd if=/etc/hosts of=$src_dir/src_file
+	touch $tgt_dir/tgt_file
+	drop_update_reply $MDTIDX \
+		"mrename $src_dir/src_file $tgt_dir/tgt_file" ||
+		error "mrename failed"
+
+	$CHECKSTAT -t file $src_dir/src_file &&
+				error "src_file present after rename"
+
+	diff /etc/hosts $tgt_dir/tgt_file ||
+			error "file changed after rename"
+
+}
+run_test 110h "drop update reply during cross-MDT file rename"
+
+test_110i () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local src_dir=$DIR/$tdir/source_dir
+	local tgt_dir=$DIR/$tdir/target_dir
+	local MDTIDX=1
+
+	mkdir -p $src_dir
+	$LFS mkdir -i $MDTIDX $tgt_dir
+
+	mkdir $src_dir/src_dir
+	touch $src_dir/src_dir/a
+	mkdir $tgt_dir/tgt_dir
+	drop_update_reply $MDTIDX \
+		"mrename $src_dir/src_dir $tgt_dir/tgt_dir" ||
+		error "mrename failed"
+
+	$CHECKSTAT -t dir $src_dir/src_dir &&
+			error "src_dir present after rename"
+
+	$CHECKSTAT -t dir $tgt_dir/tgt_dir ||
+				error "tgt_dir not present after rename"
+
+	$CHECKSTAT -t file $tgt_dir/tgt_dir/a ||
+				error "a not present after rename"
+}
+run_test 110i "drop update reply during cross-MDT dir rename"
+
+test_110j () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local local_dir=$DIR/$tdir/local_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	mkdir $DIR/$tdir/local_dir
+	$LFS mkdir -i $MDTIDX $remote_dir
+
+	touch $local_dir/local_file
+	drop_update_reply $MDTIDX \
+		"ln $local_dir/local_file $remote_dir/remote_file" ||
+		error "ln failed"
+
+	$CHECKSTAT -t file $remote_dir/remote_file ||
+				error "remote not present after ln"
+}
+run_test 110j "drop update reply during cross-MDT ln"
+
 # LU-2844 mdt prepare fail should not cause umount oops
 test_111 ()
 {
@@ -1971,12 +2246,104 @@ test_113() {
 	# let the client reconnect
 	client_reconnect
 	EVICT=$($LCTL get_param mdc.$FSNAME-MDT*.state |
-	   awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+	  awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
 
 	[ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
 }
 run_test 113 "ldlm enqueue dropped reply should not cause deadlocks"
 
+T130_PID=0
+test_130_base() {
+	test_mkdir -p $DIR/$tdir
+
+	# Prevent interference from layout intent RPCs due to
+	# asynchronous writeback. These will be tested in 130c below.
+	do_nodes ${CLIENTS:-$HOSTNAME} sync
+
+	# get only LOOKUP lock on $tdir
+	cancel_lru_locks mdc
+	ls $DIR/$tdir/$tfile 2>/dev/null
+
+	# get getattr by fid on $tdir
+	#
+	# we need to race with unlink, unlink must complete before we will
+	# take a DLM lock, otherwise unlink will wait until getattr will
+	# complete; but later than getattr starts so that getattr found
+	# the object
+#define OBD_FAIL_MDS_INTENT_DELAY		0x160
+	set_nodes_failloc "$(mdts_nodes)" 0x80000160
+	stat $DIR/$tdir &
+	T130_PID=$!
+	sleep 2
+
+	rm -rf $DIR/$tdir
+
+	# drop the reply so that resend happens on an unlinked file.
+#define OBD_FAIL_MDS_LDLM_REPLY_NET	 0x157
+	set_nodes_failloc "$(mdts_nodes)" 0x80000157
+}
+
+test_130a() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	test_130_base
+
+	wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
+	return 0
+}
+run_test 130a "enqueue resend on not existing file"
+
+test_130b() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+	test_130_base
+	# let the reply to be dropped
+	sleep 10
+
+#define OBD_FAIL_SRV_ENOENT              0x217
+	set_nodes_failloc "$(mdts_nodes)" 0x80000217
+
+	wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
+	return 0
+}
+run_test 130b "enqueue resend on a stale inode"
+
+test_130c() {
+	remote_mds_nodsh && skip "remote MDS with nodsh" && return
+
+	do_nodes ${CLIENTS:-$HOSTNAME} sync
+	echo XXX > $DIR/$tfile
+
+	cancel_lru_locks mdc
+
+	# Trigger writeback on $tfile.
+	#
+	# we need to race with unlink, unlink must complete before we will
+	# take a DLM lock, otherwise unlink will wait until intent will
+	# complete; but later than intent starts so that intent found
+	# the object
+#define OBD_FAIL_MDS_INTENT_DELAY		0x160
+	set_nodes_failloc "$(mdts_nodes)" 0x80000160
+	sync &
+	T130_PID=$!
+	sleep 2
+
+	rm $DIR/$tfile
+
+	# drop the reply so that resend happens on an unlinked file.
+#define OBD_FAIL_MDS_LDLM_REPLY_NET	 0x157
+	set_nodes_failloc "$(mdts_nodes)" 0x80000157
+
+	# let the reply to be dropped
+	sleep 10
+
+#define OBD_FAIL_SRV_ENOENT              0x217
+	set_nodes_failloc "$(mdts_nodes)" 0x80000217
+
+	wait $T130_PID
+
+	return 0
+}
+run_test 130c "layout intent resend on a stale inode"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status