X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=cf7a5aef4447c698bd6e916233c99cca3be6e194;hp=803997aaae8e31542f4565d76115ed66dd206faf;hb=7936ec229c7f62cfbf5a0e4ff5933149967e7c8f;hpb=4cfcf91efa06decc155d8f12eafc8ede3e850fd5 diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh index 803997a..cf7a5ae 100755 --- a/lustre/tests/recovery-small.sh +++ b/lustre/tests/recovery-small.sh @@ -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 $SINGLEMDS) = "zfs" ] && +# bug number for skipped test: LU-2194 LU-2547 + ALWAYS_EXCEPT="$ALWAYS_EXCEPT 19b 24a 24b" + 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)" @@ -246,6 +264,7 @@ test_16() { 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)" @@ -314,7 +333,7 @@ test_18a() { 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" @@ -347,7 +366,7 @@ test_18b() { # 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)" @@ -386,7 +405,7 @@ test_18c() { # 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)" @@ -395,20 +414,26 @@ test_19a() { local BEFORE=`date +%s` local EVICT - mount_client $DIR2 + mount_client $DIR2 || error "failed to mount $DIR2" + + # cancel cached locks from OST to avoid eviction from it + cancel_lru_locks osc - 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 +441,61 @@ 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 + # cancel cached locks from MDT to avoid eviction from it + cancel_lru_locks mdc + + 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 $BEFORE ]] || error "eviction happened" +} +run_test 19c "check reconnect and lock resend do not trigger expired_lock_main" + test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup remote_ost_nodsh && skip "remote OST with nodsh" && return 0 @@ -704,7 +768,8 @@ test_24a() { # bug 11710 details correct fsync() behavior rc=$? lctl set_param fail_loc=0x0 client_reconnect - [ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true + [ $rc -eq 0 ] && + error_ignore bz5494 "multiop didn't fail fsync: rc $rc" || true } run_test 24a "fsync error (should return error)" @@ -721,7 +786,7 @@ wait_client_evicted () { test_24b() { remote_ost_nodsh && skip "remote OST with nodsh" && return 0 - dmesg -c + dmesg -c > /dev/null mkdir -p $DIR/$tdir lfs setstripe $DIR/$tdir -s 0 -i 0 -c 1 cancel_lru_locks osc @@ -744,9 +809,10 @@ test_24b() { lctl set_param fail_loc=0x0 client_reconnect [ $rc1 -eq 0 -o $rc2 -eq 0 ] && - error_ignore "multiop didn't fail fsync: $rc1 or close: $rc2" || true + error_ignore bz5494 "multiop didn't fail fsync: $rc1 or close: $rc2" || + true - dmesg | grep "dirty page discard:" || \ + dmesg | grep "dirty page discard:" || error "no discarded dirty page found!" } run_test 24b "test dirty page discard due to client eviction" @@ -859,6 +925,7 @@ 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 $SINGLEMDS ost FULL return 0 } run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)" @@ -894,7 +961,8 @@ test_50() { rc=$? echo writemany returned $rc #these may fail because of eviction due to slow AST response. - [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true + [ $rc -eq 0 ] || + error_ignore bz13652 "writemany returned rc $rc" || true } run_test 50 "failover MDS under load" @@ -913,19 +981,21 @@ 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 sleep $TIMEOUT kill -USR1 $CLIENT_PID - wait $CLIENT_PID + wait $CLIENT_PID rc=$? echo writemany returned $rc - [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true + [ $rc -eq 0 ] || + error_ignore bz13652 "writemany returned rc $rc" || true } run_test 51 "failover MDS during recovery" @@ -994,57 +1064,58 @@ test_55() { mkdir -p $DIR/$tdir + # Minimum pass speed is 2MBps + local ddtimeout=64 + # LU-2887/LU-3089 - set min pass speed to 500KBps + [ "$(facet_fstype ost1)" = "zfs" ] && ddtimeout=256 + # first dd should be finished quickly $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1 - dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 & + dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 & DDPID=$! count=0 echo "step1: testing ......" - while [ true ]; do - if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi - count=$[count+1] - if [ $count -gt 64 ]; then - error "dd should be finished!" - fi - sleep 1 - done + while kill -0 $DDPID 2> /dev/null; do + let count++ + if [ $count -gt $ddtimeout ]; then + error "dd should be finished!" + fi + sleep 1 + done echo "(dd_pid=$DDPID, time=$count)successful" $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2 #define OBD_FAIL_OST_DROP_REQ 0x21d do_facet ost1 lctl set_param fail_loc=0x0000021d # second dd will be never finished - dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 & + dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 & DDPID=$! count=0 echo "step2: testing ......" - while [ $count -le 64 ]; do - dd_name="`ps x | awk '$1 == '$DDPID' { print $5 }'`" - if [ -z $dd_name ]; then - ls -l $DIR/$tdir - echo "debug: (dd_name=$dd_name, dd_pid=$DDPID, time=$count)" - error "dd shouldn't be finished!" - fi - count=$[count+1] - sleep 1 - done + while [ $count -le $ddtimeout ]; do + if ! kill -0 $DDPID 2> /dev/null; then + ls -l $DIR/$tdir + error "dd shouldn't be finished! (time=$count)" + fi + let count++ + sleep 1 + done echo "(dd_pid=$DDPID, time=$count)successful" #Recover fail_loc and dd will finish soon do_facet ost1 lctl set_param fail_loc=0 count=0 echo "step3: testing ......" - while [ true ]; do - if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi - count=$[count+1] - if [ $count -gt 500 ]; then - error "dd should be finished!" - fi - sleep 1 - done + while kill -0 $DDPID 2> /dev/null; do + let count++ + if [ $count -gt $((ddtimeout + 440)) ]; then + error "dd should be finished!" + fi + sleep 1 + done echo "(dd_pid=$DDPID, time=$count)successful" - rm -rf $DIR/$tdir + rm -rf $DIR/$tdir } run_test 55 "ost_brw_read/write drops timed-out read/write request" @@ -1118,17 +1189,17 @@ test_59() { # bug 10589 run_test 59 "Read cancel race on client eviction" err17935 () { - # we assume that all md changes are in the MDT0 changelog - if [ $MDSCOUNT -gt 1 ]; then - error_ignore 17935 $* - else - error $* - fi + # we assume that all md changes are in the MDT0 changelog + if [ $MDSCOUNT -gt 1 ]; then + error_ignore bz17935 $* + else + error $* + fi } test_60() { - MDT0=$($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 @@ -1205,6 +1276,16 @@ 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:-$HOSTNAME} @@ -1304,8 +1385,8 @@ target_instance_match() local target=${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 ]) } @@ -1428,10 +1509,11 @@ test_103() stop mds1 # We need this test because mds is like a client in IR context. - start mds1 $MDSDEV1 || error "MDS should start w/o mgs" + start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || + error "MDS should start w/o mgs" # start mgs and remount mds w/ ir - start mgs $MGSDEV + start mgs $(mgsdevname) $MGS_MOUNT_OPTS clients_up # remount client so that fsdb will be created on the MGS @@ -1598,6 +1680,271 @@ test_107 () { 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 + 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 -c2 $remote_dir" || + error "lfs mkdir failed" + local diridx=$($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 -c2 $remote_dir" || + error "lfs mkdir failed" + + diridx=$($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 $MDTIDX "$LFS mkdir -i $MDTIDX -c2 $remote_dir" || + error "lfs mkdir failed" + + diridx=$($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 -c2 $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 -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" + + 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 -c2 $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" + +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 () +{ + [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] || + { skip "Need MDS version at least 2.3.62"; return 0; } + + local mdsdev=$(mdsdevname ${SINGLEMDS//mds/}) +#define OBD_FAIL_MDS_CHANGELOG_INIT 0x151 + do_facet $SINGLEMDS lctl set_param fail_loc=0x151 + stop $SINGLEMDS || error "stop MDS failed" + start $SINGLEMDS $mdsdev && error "start MDS should fail" + do_facet $SINGLEMDS lctl set_param fail_loc=0 + start $SINGLEMDS $mdsdev || error "start MDS failed" +} +run_test 111 "mdd setup fail should not cause umount oops" + +# LU-793 +test_112a() { + remote_ost_nodsh && skip "remote OST with nodsh" && return 0 + + do_facet_random_file client $TMP/$tfile 100K || + error_noexit "Create random file $TMP/$tfile" + + pause_bulk "cp $TMP/$tfile $DIR/$tfile" $TIMEOUT || + error_noexit "Can't pause_bulk copy" + + df $DIR + # expect cmp to succeed, client resent bulk + cmp $TMP/$tfile $DIR/$tfile || + error_noexit "Wrong data has been written" + rm $DIR/$tfile || + error_noexit "Can't remove file" + rm $TMP/$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" + complete $SECONDS check_and_cleanup_lustre exit_status