X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=3f4a5cb4b2c6d56de7b41e08782cbfcdad71b74d;hp=aba5353305e1832fdd25bfad5154ae3599e6c704;hb=7136981bfaad8b95e15e9d2969b94c933491eb98;hpb=a4e6076ee269b54a381417fe9443ab8e9f6f80da

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index aba5353..3f4a5cb 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -2,8 +2,8 @@
 
 set -e
 
-#         bug  5494 5493  LU2034
-ALWAYS_EXCEPT="24   52    60      $RECOVERY_SMALL_EXCEPT"
+#         bug  5493  LU2034
+ALWAYS_EXCEPT="52    60      $RECOVERY_SMALL_EXCEPT"
 
 export MULTIOP=${MULTIOP:-multiop}
 PTLDEBUG=${PTLDEBUG:--1}
@@ -19,6 +19,10 @@ require_dsh_mds || exit 0
 #                                   1  2.5  2.5    4    4          (min)"
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="17  26a  26b    50   51     57"
 
+[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
+# bug number for skipped test:	      LU-2194
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 19b"
+
 build_test_filter
 
 # Allow us to override the setup if we already have a mounted system by
@@ -29,7 +33,7 @@ CLEANUP=${CLEANUP:-""}
 check_and_setup_lustre
 
 assert_DIR
-rm -rf $DIR/[df][0-9]*
+rm -rf $DIR/d[0-9]* $DIR/f.${TESTSUITE}*
 
 test_1() {
 	local f1="$DIR/$tfile"
@@ -146,56 +150,70 @@ run_test 9 "pause bulk on OST (bug 1420)"
 
 #bug 1521
 test_10() {
-    do_facet client mcreate $DIR/$tfile        || 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  || return 3
-    do_facet client "munlink $DIR/$tfile"
+	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
 }
 run_test 10 "finish request on server after client eviction (bug 1521)"
 
 #bug 2460
 # wake up a thread waiting for completion after eviction
 test_11(){
-    do_facet client $MULTIOP $DIR/$tfile Ow  || return 1
-    do_facet client $MULTIOP $DIR/$tfile or  || return 2
+	do_facet client $MULTIOP $DIR/$tfile Ow  ||
+		{ error "multiop write failed: $?"; return 1; }
+	do_facet client $MULTIOP $DIR/$tfile or  ||
+		{ error "multiop read failed: $?"; return 2; }
 
-    cancel_lru_locks osc
+	cancel_lru_locks osc
 
-    do_facet client $MULTIOP $DIR/$tfile or  || return 3
-    drop_bl_callback $MULTIOP $DIR/$tfile Ow || echo "evicted as expected"
+	do_facet client $MULTIOP $DIR/$tfile or  ||
+		{ error "multiop read failed: $?"; return 3; }
+	drop_bl_callback $MULTIOP $DIR/$tfile Ow || echo "evicted as expected"
 
-    do_facet client munlink $DIR/$tfile  || return 4
+	do_facet client munlink $DIR/$tfile ||
+		{ error "munlink failed: $?"; return 4; }
+	# allow recovery to complete
+	client_up || client_up || sleep $TIMEOUT
 }
 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
 
 #b=2494
 test_12(){
-    $LCTL mark $MULTIOP $DIR/$tfile OS_c
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
-    clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
-    multiop_bg_pause $DIR/$tfile OS_c || return 1
-    PID=$!
+	$LCTL mark $MULTIOP $DIR/$tfile OS_c
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
+	clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
+	multiop_bg_pause $DIR/$tfile OS_c ||
+		{ error "multiop failed: $?"; return 1; }
+	PID=$!
 #define OBD_FAIL_MDS_CLOSE_NET           0x115
-    kill -USR1 $PID
-    echo "waiting for multiop $PID"
-    wait $PID || return 2
-    do_facet client munlink $DIR/$tfile  || return 3
+	kill -USR1 $PID
+	echo "waiting for multiop $PID"
+	wait $PID || { error "wait for multiop faile: $?"; return 2; }
+	do_facet client munlink $DIR/$tfile ||
+		{ error "client munlink failed: $?"; return 3; }
+	# allow recovery to complete
+	client_up || client_up || sleep $TIMEOUT
 }
 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
 
 # Bug 113, check that readdir lost recv timeout works.
 test_13() {
-    mkdir -p $DIR/$tdir || return 1
-    touch $DIR/$tdir/newentry || return
+	mkdir -p $DIR/$tdir || { error "mkdir failed: $?"; return 1; }
+	touch $DIR/$tdir/newentry || { error "touch failed: $?"; return 2; }
 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
-    ls $DIR/$tdir || return 3
-    do_facet $SINGLEMDS "lctl set_param fail_loc=0"
-    rm -rf $DIR/$tdir || return 4
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
+	ls $DIR/$tdir || { error "ls failed: $?"; return 3; }
+	do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+	rm -rf $DIR/$tdir || { error "remove test dir failed: $?"; return 4; }
 }
 run_test 13 "mdc_readpage restart test (bug 1138)"
 
@@ -395,20 +413,23 @@ test_19a() {
 	local BEFORE=`date +%s`
 	local EVICT
 
-	mount_client $DIR2
+	mount_client $DIR2 || error "failed to mount $DIR2"
 
-	do_facet client mcreate $DIR/$tfile        || return 1
-	drop_ldlm_cancel "chmod 0777 $DIR2"
+	do_facet client "stat $DIR > /dev/null"  ||
+		error "failed to stat $DIR: $?"
+	drop_ldlm_cancel "chmod 0777 $DIR2" ||
+		error "failed to chmod $DIR2"
 
 	umount_client $DIR2
-	do_facet client "munlink $DIR/$tfile"
 
 	# let the client reconnect
-	sleep 5
+	client_reconnect
 	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state | \
 	    awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
 
-	[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] || error "no eviction"
+	[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
+		(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
+		    error "no eviction: $EVICT before:$BEFORE")
 }
 run_test 19a "test expired_lock_main on mds (2867)"
 
@@ -416,22 +437,58 @@ test_19b() {
 	local BEFORE=`date +%s`
 	local EVICT
 
-	mount_client $DIR2
+	mount_client $DIR2 || error "failed to mount $DIR2: $?"
 
-	do_facet client $MULTIOP $DIR/$tfile Ow  || return 1
-	drop_ldlm_cancel $MULTIOP $DIR2/$tfile Ow
-	umount_client $DIR2
-	do_facet client munlink $DIR/$tfile
+	do_facet client $MULTIOP $DIR/$tfile Ow ||
+		error "failed to run multiop: $?"
+	drop_ldlm_cancel $MULTIOP $DIR2/$tfile Ow ||
+		error "failed to ldlm_cancel: $?"
+
+	umount_client $DIR2 || error "failed to unmount $DIR2: $?"
+	do_facet client munlink $DIR/$tfile ||
+		error "failed to unlink $DIR/$tfile: $?"
 
 	# let the client reconnect
-	sleep 5
+	client_reconnect
 	EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state | \
 	    awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
 
-	[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] || error "no eviction"
+	[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
+		(do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
+		    error "no eviction: $EVICT before:$BEFORE")
 }
 run_test 19b "test expired_lock_main on ost (2867)"
 
+test_19c() {
+	local BEFORE=`date +%s`
+
+	mount_client $DIR2
+	$LCTL set_param ldlm.namespaces.*.early_lock_cancel=0
+
+	mkdir -p $DIR1/$tfile
+	stat $DIR1/$tfile
+
+#define OBD_FAIL_PTLRPC_CANCEL_RESEND 0x516
+	do_facet mds $LCTL set_param fail_loc=0x80000516
+
+	touch $DIR2/$tfile/file1 &
+	PID1=$!
+	# let touch to get blocked on the server
+	sleep 2
+
+	wait $PID1
+	$LCTL set_param ldlm.namespaces.*.early_lock_cancel=1
+	umount_client $DIR2
+
+	# let the client reconnect
+	sleep 5
+	EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
+	   awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+
+	[ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
+}
+run_test 19c "check reconnect and lock resend do not trigger expired_lock_main"
+
 test_20a() {	# bug 2983 - ldlm_handle_enqueue cleanup
 	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
 
@@ -690,7 +747,7 @@ test_23() { #b=4561
 }
 run_test 23 "client hang when close a file after mds crash"
 
-test_24() { # bug 11710 details correct fsync() behavior
+test_24a() { # bug 11710 details correct fsync() behavior
 	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
 
 	mkdir -p $DIR/$tdir
@@ -706,7 +763,7 @@ test_24() { # bug 11710 details correct fsync() behavior
 	client_reconnect
 	[ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
 }
-run_test 24 "fsync error (should return error)"
+run_test 24a "fsync error (should return error)"
 
 wait_client_evicted () {
 	local facet=$1
@@ -714,9 +771,43 @@ wait_client_evicted () {
 	local varsvc=${facet}_svc
 
 	wait_update $(facet_active_host $facet) \
-                "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
-                $((exports - 1)) $3
+		"lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
+		$((exports - 1)) $3
+}
+
+test_24b() {
+	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+	dmesg -c > /dev/null
+	mkdir -p $DIR/$tdir
+	lfs setstripe $DIR/$tdir -s 0 -i 0 -c 1
+	cancel_lru_locks osc
+	multiop_bg_pause $DIR/$tdir/$tfile-1 Ow8192_yc ||
+		error "mulitop Ow8192_yc failed"
+
+	MULTI_PID1=$!
+	multiop_bg_pause $DIR/$tdir/$tfile-2 Ow8192_c ||
+		error "mulitop Ow8192_c failed"
+
+	MULTI_PID2=$!
+	ost_evict_client
+
+	kill -USR1 $MULTI_PID1
+	wait $MULTI_PID1
+	rc1=$?
+	kill -USR1 $MULTI_PID2
+	wait $MULTI_PID2
+	rc2=$?
+	lctl set_param fail_loc=0x0
+	client_reconnect
+	[ $rc1 -eq 0 -o $rc2 -eq 0 ] &&
+	error_ignore 5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
+		true
+
+	dmesg | grep "dirty page discard:" || \
+		error "no discarded dirty page found!"
 }
+run_test 24b "test dirty page discard due to client eviction"
 
 test_26a() {      # was test_26 bug 5921 - evict dead exports by pinger
 # this test can only run from a client on a separate node.
@@ -826,6 +917,7 @@ test_29a() { # bug 22273 - error adding new clients
 	# fail abort so client will be new again
 	fail_abort $SINGLEMDS
 	client_up || error "reconnect failed"
+	wait_osc_import_state mds ost FULL
 	return 0
 }
 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
@@ -880,10 +972,11 @@ test_51() {
 	SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
         echo will failover at $SEQ
         for i in $SEQ
-          do
-          echo failover in $i sec
-          sleep $i
-          facet_failover $SINGLEMDS
+        do
+		#echo failover in $i sec
+		log "test_$testnum: failover in $i sec"
+		sleep $i
+		facet_failover $SINGLEMDS
         done
 	# client process should see no problems even though MDS went down
 	# and recovery was interrupted
@@ -1172,6 +1265,16 @@ test_61()
 }
 run_test 61 "Verify to not reuse orphan objects - bug 17025"
 
+# test_62 as seen it b2_1 please do not reuse test_62
+#test_62()
+#{
+#	zconf_umount `hostname` $DIR
+#	#define OBD_FAIL_PTLRPC_DELAY_IMP_FULL   0x516
+#	lctl set_param fail_loc=0x516
+#	mount_client $DIR
+#}
+#run_test 62 "Verify connection flags race - bug LU-1716"
+
 check_cli_ir_state()
 {
         local NODE=${1:-$HOSTNAME}
@@ -1542,6 +1645,122 @@ test_106() { # LU-1789
 }
 run_test 106 "lightweight connection support"
 
+test_107 () {
+	local CLIENT_PID
+	local close_pid
+
+	mkdir -p $DIR/$tdir
+	# OBD_FAIL_MDS_REINT_NET_REP   0x119
+	do_facet $SINGLEMDS lctl set_param fail_loc=0x119
+	multiop $DIR/$tdir D_c &
+	close_pid=$!
+	mkdir $DIR/$tdir/dir_106 &
+	CLIENT_PID=$!
+	do_facet $SINGLEMDS lctl set_param fail_loc=0
+	fail $SINGLEMDS
+
+	wait $CLIENT_PID || rc=$?
+	checkstat -t dir $DIR/$tdir/dir_106 || return 1
+
+	kill -USR1 $close_pid
+	wait $close_pid || return 2
+
+	return $rc
+}
+run_test 107 "drop reint reply, then restart MDT"
+
+test_110a () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	drop_request "$LFS mkdir -i $MDTIDX $remote_dir" ||
+					error "lfs mkdir failed"
+	local diridx=$($GETSTRIPE -M $remote_dir)
+	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 110a "create remote directory: drop client req"
+
+test_110b () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	drop_reint_reply "$LFS mkdir -i $MDTIDX $remote_dir" ||
+					error "lfs mkdir failed"
+
+	diridx=$($GETSTRIPE -M $remote_dir)
+	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 110b "create remote directory: drop Master rep"
+
+test_110c () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	drop_update_reply $((MDTIDX + 1)) "$LFS mkdir -i $MDTIDX $remote_dir" ||
+						error "lfs mkdir failed"
+
+	diridx=$($GETSTRIPE -M $remote_dir)
+	[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 110c "create remote directory: drop update rep on slave MDT"
+
+test_110d () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+
+	drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110d "remove remote directory: drop client req"
+
+test_110e () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir  || error "lfs mkdir failed"
+	drop_reint_reply "rm -rf $remote_dir" || error "rm remote dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+
+	return 0
+}
+run_test 110e "remove remote directory: drop master rep"
+
+test_110f () {
+	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+	local remote_dir=$DIR/$tdir/remote_dir
+	local MDTIDX=1
+
+	mkdir -p $DIR/$tdir
+	$LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+	drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
+					error "rm remote dir failed"
+
+	rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 110f "remove remote directory: drop slave rep"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status