[ "$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
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: $?"
}
-run_test 10 "finish request on server after client eviction (bug 1521)"
+run_test 10a "finish request on server after client eviction (bug 1521)"
+
+test_10b() {
+ local before=$(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: $?"
+}
+run_test 10b "re-send BL AST"
+
+test_10d() {
+ local before=$(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
+
+ $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_once "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")
+
+ rm $TMP/$tfile
+ umount_client $MOUNT2
+}
+run_test 10d "test failed blocking ast"
#bug 2460
# wake up a thread waiting for completion after eviction
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; }
sleep $TIMEOUT
do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 2
start_read_ahead
+ rm -f $TMP/$tfile
}
run_test 16 "timeout bulk put, don't evict client (2732)"
rc=0
pgcache_empty || rc=2
$LCTL --device $osc2dev activate
- rm -f $f
+ rm -f $f $TMP/$tfile
return $rc
}
run_test 18a "manual ost invalidate clears page cache immediately"
# cache after the client reconnects?
rc=0
pgcache_empty || rc=2
- rm -f $f
+ rm -f $f $TMP/$tfile
return $rc
}
run_test 18b "eviction and reconnect clears page cache (2766)"
# cache after the client reconnects?
rc=0
pgcache_empty || rc=2
- rm -f $f
+ rm -f $f $TMP/$tfile
return $rc
}
run_test 18c "Dropped connect reply after eviction handing (14755)"
# let the client reconnect
client_reconnect
- EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state | \
- awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+ EVICT=$(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;
# let the client reconnect
client_reconnect
- EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state | \
- awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
+ EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state |
+ awk -F"[ [,]" '/EVICTED ]$/ \
+ { if (mx < $5) {mx = $5;} } END { print mx }')
[ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
(do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
# 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"
}
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
#define OBD_FAIL_MDS_RESEND 0x136
touch $DIR/$tfile
do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
- stat $DIR/$tfile
+ stat $DIR/$tfile || error "stat failed"
do_facet $SINGLEMDS "lctl set_param fail_loc=0"
rm -f $DIR/$tfile
}
-run_test 56 "do not allow reconnect to busy exports"
+run_test 56 "do not fail on getattr resend"
test_57_helper() {
# no oscs means no client or mdt
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 $DIR"
}
}
test_60() {
- MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
- awk '{gsub(/_UUID/,""); print $1}' | head -1)
+ MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
+ awk '{ gsub(/_UUID/,""); print $1 }' | head -n1)
NUM_FILES=15000
mkdir -p $DIR/$tdir
local target=${srv}_svc
local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
- local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import | \
- awk '/instance/{ print $2 }' |head -1)
+ local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import |
+ awk '/instance/{ print $2 }' | head -n1)
return $([ $si -eq $ci ])
}
# get one of the clients from client list
local rcli=$(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)
[ $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"
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
local remote_dir=$DIR/$tdir/remote_dir
local MDTIDX=1
+ local num
+
+ #prepare for 110 test, which need set striped dir on remote MDT.
+ for num in $(seq $MDSCOUNT); do
+ do_facet mds$num \
+ lctl set_param -n mdt.${FSNAME}*.enable_remote_dir=1 \
+ 2>/dev/null
+ done
mkdir -p $DIR/$tdir
- drop_request "$LFS mkdir -i $MDTIDX $remote_dir" ||
+ drop_request "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
error "lfs mkdir failed"
local diridx=$($GETSTRIPE -M $remote_dir)
[ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
local MDTIDX=1
mkdir -p $DIR/$tdir
- drop_reint_reply "$LFS mkdir -i $MDTIDX $remote_dir" ||
+ drop_reint_reply "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
error "lfs mkdir failed"
diridx=$($GETSTRIPE -M $remote_dir)
local MDTIDX=1
mkdir -p $DIR/$tdir
- drop_update_reply $((MDTIDX + 1)) "$LFS mkdir -i $MDTIDX $remote_dir" ||
+ drop_update_reply $MDTIDX "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
error "lfs mkdir failed"
diridx=$($GETSTRIPE -M $remote_dir)
local MDTIDX=1
mkdir -p $DIR/$tdir
- $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+ $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
local MDTIDX=1
mkdir -p $DIR/$tdir
- $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+ $LFS mkdir -i $MDTIDX -c2 $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"
local MDTIDX=1
mkdir -p $DIR/$tdir
- $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
+ $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
error "rm remote dir failed"
}
run_test 110f "remove remote directory: drop slave rep"
+test_110g () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ local remote_dir=$DIR/$tdir/remote_dir
+ local MDTIDX=1
+
+ mkdir -p $remote_dir
+
+ 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 $remote_dir || error "migrate failed"
+ do_facet mds$MDTIDX lctl set_param fail_loc=0x0
+
+ for file in $(find $remote_dir); do
+ mdt_index=$($LFS getstripe -M $file)
+ [ $mdt_index == $MDTIDX ] ||
+ error "$file is not on MDT${MDTIDX}"
+ done
+
+ rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 110g "drop reply during migration"
+
# LU-2844 mdt prepare fail should not cause umount oops
test_111 ()
{
test_112a() {
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
- [[ $(lustre_version_code ost1) -ge $(version_code 2.5.92) ]] ||
- { skip "Need OST version at least 2.5.92"; return 0; }
-
do_facet_random_file client $TMP/$tfile 100K ||
error_noexit "Create random file $TMP/$tfile"
}
run_test 112a "bulk resend while orignal request is in progress"
+# parameters: fail_loc CMD RC
+test_120_reply() {
+ local PID
+ local PID2
+ local rc=5
+ local fail
+
+ #define OBD_FAIL_LDLM_CP_CB_WAIT2 0x320
+ #define OBD_FAIL_LDLM_CP_CB_WAIT3 0x321
+ #define OBD_FAIL_LDLM_CP_CB_WAIT4 0x322
+ #define OBD_FAIL_LDLM_CP_CB_WAIT5 0x323
+
+ echo
+ echo -n "** FLOCK REPLY vs. EVICTION race, lock $2"
+ [ "$1" = "CLEANUP" ] &&
+ fail=0x80000320 && echo ", $1 cp first"
+ [ "$1" = "REPLY" ] &&
+ fail=0x80000321 && echo ", $1 cp first"
+ [ "$1" = "DEADLOCK CLEANUP" ] &&
+ fail=0x80000322 && echo " DEADLOCK, CLEANUP cp first"
+ [ "$1" = "DEADLOCK REPLY" ] &&
+ fail=0x80000323 && echo " DEADLOCK, REPLY cp first"
+
+ if [ x"$2" = x"get" ]; then
+ #for TEST lock, take a conflict in advance
+ # sleep longer than evictor to not confuse fail_loc: 2+2+4
+ echo "** Taking conflict **"
+ flocks_test 5 set read sleep 10 $DIR/$tfile &
+ PID2=$!
+
+ sleep 2
+ fi
+
+ $LCTL set_param fail_loc=$fail
+
+ flocks_test 5 $2 write $DIR/$tfile &
+ PID=$!
+
+ sleep 2
+ echo "** Evicting and re-connecting client **"
+ mds_evict_client
+
+ client_reconnect
+
+ if [ x"$2" = x"get" ]; then
+ wait $PID2
+ fi
+
+ wait $PID
+ rc=$?
+
+ # check if the return value is allowed
+ [ $rc -eq $3 ] && rc=0
+
+ $LCTL set_param fail_loc=0
+ return $rc
+}
+
+# a lock is taken, unlock vs. cleanup_resource() race for destroying
+# the ORIGINAL lock.
+test_120_destroy()
+{
+ local PID
+
+ flocks_test 5 set write sleep 4 $DIR/$tfile &
+ PID=$!
+ sleep 2
+
+ # let unlock to sleep in CP CB
+ $LCTL set_param fail_loc=$1
+ sleep 4
+
+ # let cleanup to cleep in CP CB
+ mds_evict_client
+
+ client_reconnect
+
+ wait $PID
+ rc=$?
+
+ $LCTL set_param fail_loc=0
+ return $rc
+}
+
+test_120() {
+ flock_is_enabled || { skip "mount w/o flock enabled" && return; }
+ touch $DIR/$tfile
+
+ test_120_reply "CLEANUP" set 5 || error "SET race failed"
+ test_120_reply "CLEANUP" get 5 || error "GET race failed"
+ test_120_reply "CLEANUP" unlock 5 || error "UNLOCK race failed"
+
+ test_120_reply "REPLY" set 5 || error "SET race failed"
+ test_120_reply "REPLY" get 5 || error "GET race failed"
+ test_120_reply "REPLY" unlock 5 || error "UNLOCK race failed"
+
+ # DEADLOCK tests
+ test_120_reply "DEADLOCK CLEANUP" set 5 || error "DEADLOCK race failed"
+ test_120_reply "DEADLOCK REPLY" set 35 || error "DEADLOCK race failed"
+
+ test_120_destroy 0x320 || error "unlock-cleanup race failed"
+}
+run_test 120 "flock race: completion vs. evict"
+
+test_113() {
+ local BEFORE=$(date +%s)
+ local EVICT
+
+ # modify dir so that next revalidate would not obtain UPDATE lock
+ touch $DIR
+
+ # drop 1 reply with UPDATE lock,
+ # resend should not create 2nd lock on server
+ mcreate $DIR/$tfile || error "mcreate failed: $?"
+ drop_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?"
+
+ # 2 BL AST will be sent to client, both must find the same lock,
+ # race them to not get EINVAL for 2nd BL AST
+ #define OBD_FAIL_LDLM_PAUSE_CANCEL2 0x31f
+ $LCTL set_param fail_loc=0x8000031f
+
+ $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0 > /dev/null
+ chmod 0777 $DIR/$tfile || error "chmod failed: $?"
+ $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1 > /dev/null
+
+ # let the client reconnect
+ client_reconnect
+ EVICT=$($LCTL get_param mdc.$FSNAME-MDT*.state |
+ 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"
+
complete $SECONDS
check_and_cleanup_lustre
exit_status
git://git.whamcloud.com / fs / lustre-release.git / blobdiff