X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanityn.sh;h=334168c36fa50b019b3098dd378befca231463d6;hp=a3bf7caeb5f52cd8919cb1e255effae6d7540713;hb=3ecdedd620703d6ff7af0e2cbb2bb67fe19f2eb7;hpb=29bd1e0228c0dee31bf876548c7621b51d8a09af diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh old mode 100644 new mode 100755 index a3bf7ca..334168c --- a/lustre/tests/sanityn.sh +++ b/lustre/tests/sanityn.sh @@ -3,23 +3,15 @@ set -e ONLY=${ONLY:-"$*"} -# bug number for skipped test: 3192 LU-1205 15528/3811 16929 9977 15528/11549 18080 -ALWAYS_EXCEPT=" 14b 18c 19 22 28 29 35 $SANITYN_EXCEPT" +# bug number for skipped test: 3192 LU-1205 15528/3811 9977 15528/11549 18080 +ALWAYS_EXCEPT=" 14b 18c 19 28 29 35 $SANITYN_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! -# bug number for skipped test: 12652 12652 -grep -q 'Enterprise Server 10' /etc/SuSE-release 2> /dev/null && - ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 14" || true - -# It will be ported soon. -EXCEPT="$EXCEPT 22" - -SRCDIR=`dirname $0` +SRCDIR=$(dirname $0) PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH SIZE=${SIZE:-40960} CHECKSTAT=${CHECKSTAT:-"checkstat -v"} -MCREATE=${MCREATE:-mcreate} OPENFILE=${OPENFILE:-openfile} OPENUNLINK=${OPENUNLINK:-openunlink} export MULTIOP=${MULTIOP:-multiop} @@ -41,12 +33,12 @@ init_test_env $@ init_logging if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then -# bug number for skipped test: LU-2840 LU-2189 LU-2776 - ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21 36 51a" +# bug number for skipped test: LU-2189 LU-2776 + ALWAYS_EXCEPT="$ALWAYS_EXCEPT 36 51a" # LU-2829 / LU-2887 - make allowances for ZFS slowness TEST33_NFILES=${TEST33_NFILES:-1000} fi - +# 23 (min)" [ "$SLOW" = "no" ] && EXCEPT_SLOW="33a" FAIL_ON_ERROR=false @@ -56,9 +48,6 @@ TRACE=${TRACE:-""} check_and_setup_lustre -LOVNAME=$($LCTL get_param -n llite.*.lov.common_name | tail -n 1) -OSTCOUNT=$($LCTL get_param -n lov.$LOVNAME.numobd) - assert_DIR rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}* @@ -172,6 +161,18 @@ test_2f() { } run_test 2f "check attr/owner updates on DNE with 2 mtpt's" +test_2g() { + dd if=/dev/zero of=$DIR1/$tfile oflag=sync bs=1M count=2 + + local block1=$(stat $DIR1/$tfile | awk '/Blocks/ {print $4} ') + cancel_lru_locks osc + local block2=$(stat $DIR2/$tfile | awk '/Blocks/ {print $4} ') + echo "$DIR1/$tfile has $block1 blocks" + echo "$DIR2/$tfile has $block2 blocks" + [ $block1 -eq $block2 ] || error +} +run_test 2g "check blocks update on sync write" + test_3() { local target="this/is/good" ln -s $target $DIR1/$tfile || error "ln -s $target $DIR1/$tfile failed" @@ -293,7 +294,7 @@ test_13() { # bug 2451 - directory coherency } run_test 13 "test directory page revocation ====================" -test_14() { +test_14aa() { test_mkdir -p $DIR1/$tdir cp -p /bin/ls $DIR1/$tdir/$tfile multiop_bg_pause $DIR1/$tdir/$tfile Ow_c || return 1 @@ -303,9 +304,9 @@ test_14() { kill -USR1 $MULTIPID wait $MULTIPID || return 2 } -run_test 14 "execution of file open for write returns -ETXTBSY =" +run_test 14aa "execution of file open for write returns -ETXTBSY" -test_14a() { +test_14ab() { test_mkdir -p $DIR1/d14 cp -p `which multiop` $DIR1/d14/multiop || error "cp failed" MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1 @@ -315,7 +316,7 @@ test_14a() { wait $MULTIOP_PID || return 3 rm $TMP/test14.junk $DIR1/d14/multiop || error "removing multiop" } -run_test 14a "open(RDWR) of executing file returns -ETXTBSY ====" +run_test 14ab "open(RDWR) of executing file returns -ETXTBSY" test_14b() { # bug 3192, 7040 test_mkdir -p $DIR1/d14 @@ -358,16 +359,6 @@ test_14d() { # bug 10921 } run_test 14d "chmod of executing file is still possible ========" -test_14e() { # LU-4398 - test_mkdir -p $DIR1/$tdir - rm -f $DIR1/$tdir/echo - cp /bin/echo $DIR1/$tdir/echo - $DIR1/$tdir/echo Hi - $DIR1/$tdir/echo Hi - echo Bye > $DIR2/$tdir/echo -} -run_test 14e "conflicting locks are flushed on open" - test_15() { # bug 974 - ENOSPC echo "PATH=$PATH" sh oos2.sh $MOUNT1 $MOUNT2 @@ -768,14 +759,14 @@ test_32a() { # bug 11270 log "checking cached lockless truncate" $TRUNCATE $DIR1/$tfile 8000000 $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size" - [ $(calc_osc_stats lockless_truncate) -eq 0 ] || - error "lockless truncate doesn't use cached locks" + [ $(calc_osc_stats lockless_truncate) -ne 0 ] || + error "cached truncate isn't lockless" log "checking not cached lockless truncate" $TRUNCATE $DIR2/$tfile 5000000 $CHECKSTAT -s 5000000 $DIR1/$tfile || error "wrong file size" - [ $(calc_osc_stats lockless_truncate) -ne 0 ] || - error "not cached trancate isn't lockless" + [ $(calc_osc_stats lockless_truncate) -ne 0 ] || + error "not cached truncate isn't lockless" log "disabled lockless truncate" enable_lockless_truncate 0 @@ -934,7 +925,7 @@ test_33b() { avgjbd=0 avgtime=0 for i in 1 2 3; do - do_node $CLIENT1 "$LFS mkdir -i $MDTIDX -p \ + do_node $CLIENT1 "$LFS mkdir -i $MDTIDX \ $DIR1/$tdir-\\\$(hostname)-$i" jbdold=$(print_jbd_stat) @@ -967,6 +958,118 @@ test_33b() { } run_test 33b "COS: cross create/delete, 2 clients, benchmark under remote dir" +test_33c() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] && + skip "DNE CoS not supported" && return + + sync + + mkdir $DIR/$tdir + # remote mkdir is done on MDT2, which enqueued lock of $tdir on MDT1 + $LFS mkdir -i 1 $DIR/$tdir/d1 + do_facet mds1 "lctl set_param -n mdt.*.sync_count=0" + mkdir $DIR/$tdir/d2 + local sync_count=$(do_facet mds1 \ + "lctl get_param -n mdt.*MDT0000.sync_count") + [ $sync_count -eq 1 ] || error "Sync-Lock-Cancel not triggered" + + $LFS mkdir -i 1 $DIR/$tdir/d3 + do_facet mds1 "lctl set_param -n mdt.*.sync_count=0" + # during sleep remote mkdir should have been committed and canceled + # remote lock spontaneously, which shouldn't trigger sync + sleep 6 + mkdir $DIR/$tdir/d4 + local sync_count=$(do_facet mds1 \ + "lctl get_param -n mdt.*MDT0000.sync_count") + [ $sync_count -eq 0 ] || error "Sync-Lock-Cancel triggered" +} +run_test 33c "Cancel cross-MDT lock should trigger Sync-Lock-Cancel" + +ops_do_cos() { + local nodes=$(comma_list $(mdts_nodes)) + do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0" + sh -c "$@" + local async_commit_count=$(do_nodes $nodes \ + "lctl get_param -n mdt.*.async_commit_count" | calc_sum) + [ $async_commit_count -gt 0 ] || error "CoS not triggerred" + + rm -rf $DIR/$tdir + sync +} + +test_33d() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] && + skip "DNE CoS not supported" && return + + sync + # remote directory create + mkdir $DIR/$tdir + ops_do_cos "$LFS mkdir -i 1 $DIR/$tdir/subdir" + # remote directory unlink + $LFS mkdir -i 1 $DIR/$tdir + ops_do_cos "rmdir $DIR/$tdir" + # striped directory create + mkdir $DIR/$tdir + ops_do_cos "$LFS mkdir -c 2 $DIR/$tdir/subdir" + # striped directory setattr + $LFS mkdir -c 2 $DIR/$tdir + touch $DIR/$tdir + ops_do_cos "chmod 713 $DIR/$tdir" + # striped directory unlink + $LFS mkdir -c 2 $DIR/$tdir + touch $DIR/$tdir + ops_do_cos "rmdir $DIR/$tdir" + # cross-MDT link + $LFS mkdir -c 2 $DIR/$tdir + $LFS mkdir -i 0 $DIR/$tdir/d1 + $LFS mkdir -i 1 $DIR/$tdir/d2 + touch $DIR/$tdir/d1/tgt + ops_do_cos "ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src" + # cross-MDT rename + $LFS mkdir -c 2 $DIR/$tdir + $LFS mkdir -i 0 $DIR/$tdir/d1 + $LFS mkdir -i 1 $DIR/$tdir/d2 + touch $DIR/$tdir/d1/src + ops_do_cos "mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt" + # migrate + $LFS mkdir -i 0 $DIR/$tdir + ops_do_cos "$LFS migrate -m 1 $DIR/$tdir" + return 0 +} +run_test 33d "DNE distributed operation should trigger COS" + +test_33e() { + [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; } + [ $CLIENTCOUNT -ge 2 ] || + { skip "Need two or more clients, have $CLIENTCOUNT" && + return 0; } + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] && + skip "DNE CoS not supported" && return + + local client2=${CLIENT2:-$(hostname)} + + sync + + local nodes=$(comma_list $(mdts_nodes)) + do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0" + + $LFS mkdir -c 2 $DIR/$tdir + mkdir $DIR/$tdir/subdir + echo abc > $DIR/$tdir/$tfile + do_node $client2 echo dfg >> $DIR/$tdir/$tfile + do_node $client2 touch $DIR/$tdir/subdir + + local async_commit_count=$(do_nodes $nodes \ + "lctl get_param -n mdt.*.async_commit_count" | calc_sum) + [ $async_commit_count -gt 0 ] && error "CoS triggerred" + + return 0 +} +run_test 33e "DNE local operation shouldn't trigger COS" + # End commit on sharing tests get_ost_lock_timeouts() { @@ -989,6 +1092,7 @@ cleanup_34() { } test_34() { #16129 + remote_ost_nodsh && skip "remote OST with nodsh" && return local OPER local lock_in local lock_out @@ -1038,8 +1142,9 @@ run_test 34 "no lock timeout under IO" test_35() { # bug 17645 local generation=[] local count=0 - for imp in /proc/fs/lustre/mdc/$FSNAME-MDT*-mdc-*; do - g=$(awk '/generation/{print $2}' $imp/import) + gen=$(lctl get_param mdc.$FSNAME-MDT*-mdc-*.import | grep generation | + awk '/generation/{print $2}') + for g in $gen; do generation[count]=$g let count=count+1 done @@ -1081,10 +1186,20 @@ test_35() { # bug 17645 do_facet client "lctl set_param fail_loc=0x0" df -h $MOUNT1 $MOUNT2 count=0 - for imp in /proc/fs/lustre/mdc/$FSNAME-MDT*-mdc-*; do - g=$(awk '/generation/{print $2}' $imp/import) - if ! test "$g" -eq "${generation[count]}"; then - error "Eviction happened on import $(basename $imp)" + gen=$(lctl get_param mdc.$FSNAME-MDT*-mdc-*.import | grep generation | + awk '/generation/{print $2}') + for g in $gen; do + if ! test "$g" -eq "${generation[count]}"; then + list=$(lctl list_param mdc.$FSNAME-MDT*-mdc-*.import) + local c = 0 + for imp in $list; do + if [ $c = $count ]; then + break + fi + c=c+1 + done + imp=$(echo "$imp" | awk -F"." '{print $2}') + error "Eviction happened on import $imp" fi let count=count+1 done @@ -1255,8 +1370,10 @@ check_pdo_conflict() { # pdirop tests # test 40: check non-blocking operations test_40a() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + touch $DIR2 mkdir $DIR1/$tfile & PID1=$! sleep 1 @@ -1274,15 +1391,16 @@ test_40a() { rmdir $DIR2/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" - # all operations above shouldn't wait the first one + # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 40a "pdirops: create vs others ==============" test_40b() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 touch $DIR1/$tfile & @@ -1306,12 +1424,13 @@ test_40b() { check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 40b "pdirops: open|create and others ==============" test_40c() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return touch $DIR1/$tfile #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 @@ -1336,12 +1455,13 @@ test_40c() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 40c "pdirops: link and others ==============" test_40d() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return touch $DIR1/$tfile #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 @@ -1371,6 +1491,7 @@ test_40d() { run_test 40d "pdirops: unlink and others ==============" test_40e() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return touch $DIR1/$tfile #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 @@ -1393,7 +1514,7 @@ test_40e() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 40e "pdirops: rename and others ==============" @@ -1407,7 +1528,7 @@ test_41a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; echo "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41a "pdirops: create vs mkdir ==============" @@ -1420,7 +1541,7 @@ test_41b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41b "pdirops: create vs create ==============" @@ -1434,7 +1555,7 @@ test_41c() { sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41c "pdirops: create vs link ==============" @@ -1447,7 +1568,7 @@ test_41d() { sleep 1 rm $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41d "pdirops: create vs unlink ==============" @@ -1461,7 +1582,7 @@ test_41e() { sleep 1 mv $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41e "pdirops: create and rename (tgt) ==============" @@ -1474,7 +1595,7 @@ test_41f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41f "pdirops: create and rename (src) ==============" @@ -1487,7 +1608,7 @@ test_41g() { sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41g "pdirops: create vs getattr ==============" @@ -1500,7 +1621,7 @@ test_41h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 41h "pdirops: create vs readdir ==============" @@ -1514,7 +1635,7 @@ test_42a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42a "pdirops: mkdir vs mkdir ==============" @@ -1527,7 +1648,7 @@ test_42b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42b "pdirops: mkdir vs create ==============" @@ -1541,7 +1662,7 @@ test_42c() { sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42c "pdirops: mkdir vs link ==============" @@ -1554,7 +1675,7 @@ test_42d() { sleep 1 rmdir $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42d "pdirops: mkdir vs unlink ==============" @@ -1568,7 +1689,7 @@ test_42e() { sleep 1 mv -T $DIR2/$tfile-2 $DIR2/$tfile && error "rename must fail" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42e "pdirops: mkdir and rename (tgt) ==============" @@ -1581,7 +1702,7 @@ test_42f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42f "pdirops: mkdir and rename (src) ==============" @@ -1594,7 +1715,7 @@ test_42g() { sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42g "pdirops: mkdir vs getattr ==============" @@ -1607,7 +1728,7 @@ test_42h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 42h "pdirops: mkdir vs readdir ==============" @@ -1622,7 +1743,7 @@ test_43a() { sleep 1 mkdir $DIR2/$tfile || error "mkdir must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43a "pdirops: unlink vs mkdir ==============" @@ -1636,7 +1757,7 @@ test_43b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43b "pdirops: unlink vs create ==============" @@ -1651,7 +1772,7 @@ test_43c() { sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile || error "link must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43c "pdirops: unlink vs link ==============" @@ -1665,7 +1786,7 @@ test_43d() { sleep 1 rm $DIR2/$tfile && error "unlink must fail" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43d "pdirops: unlink vs unlink ==============" @@ -1680,7 +1801,7 @@ test_43e() { sleep 1 mv -u $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43e "pdirops: unlink and rename (tgt) ==============" @@ -1694,7 +1815,7 @@ test_43f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-2 && error "rename must fail" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43f "pdirops: unlink and rename (src) ==============" @@ -1708,7 +1829,7 @@ test_43g() { sleep 1 stat $DIR2/$tfile > /dev/null && error "stat must fail" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43g "pdirops: unlink vs getattr ==============" @@ -1722,7 +1843,7 @@ test_43h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43h "pdirops: unlink vs readdir ==============" @@ -1738,7 +1859,7 @@ test_43i() { $LFS mkdir -i 1 $DIR2/$tfile || error "remote mkdir must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "remote mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 43i "pdirops: unlink vs remote mkdir" @@ -1753,7 +1874,7 @@ test_44a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44a "pdirops: rename tgt vs mkdir ==============" @@ -1767,7 +1888,7 @@ test_44b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44b "pdirops: rename tgt vs create ==============" @@ -1782,7 +1903,7 @@ test_44c() { sleep 1 link $DIR2/$tfile-3 $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44c "pdirops: rename tgt vs link ==============" @@ -1796,7 +1917,7 @@ test_44d() { sleep 1 rm $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44d "pdirops: rename tgt vs unlink ==============" @@ -1812,7 +1933,7 @@ test_44e() { sleep 1 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44e "pdirops: rename tgt and rename (tgt) ==============" @@ -1827,7 +1948,7 @@ test_44f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-3 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44f "pdirops: rename tgt and rename (src) ==============" @@ -1841,7 +1962,7 @@ test_44g() { sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44g "pdirops: rename tgt vs getattr ==============" @@ -1855,7 +1976,7 @@ test_44h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44h "pdirops: rename tgt vs readdir ==============" @@ -1872,7 +1993,7 @@ test_44i() { $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "remote mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 44i "pdirops: rename tgt vs remote mkdir" @@ -1887,7 +2008,7 @@ test_45a() { sleep 1 mkdir $DIR2/$tfile || error "mkdir must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45a "pdirops: rename src vs mkdir ==============" @@ -1901,7 +2022,7 @@ test_45b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45b "pdirops: rename src vs create ==============" @@ -1916,7 +2037,7 @@ test_45c() { sleep 1 link $DIR2/$tfile-3 $DIR2/$tfile || error "link must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45c "pdirops: rename src vs link ==============" @@ -1930,7 +2051,7 @@ test_45d() { sleep 1 rm $DIR2/$tfile && error "unlink must fail" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45d "pdirops: rename src vs unlink ==============" @@ -1945,7 +2066,7 @@ test_45e() { sleep 1 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45e "pdirops: rename src and rename (tgt) ==============" @@ -1959,7 +2080,7 @@ test_45f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-3 && error "rename must fail" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45f "pdirops: rename src and rename (src) ==============" @@ -1973,7 +2094,7 @@ test_45g() { sleep 1 stat $DIR2/$tfile > /dev/null && error "stat must fail" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45g "pdirops: rename src vs getattr ==============" @@ -1987,7 +2108,7 @@ test_45h() { sleep 1 ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45h "pdirops: unlink vs readdir ==============" @@ -2003,7 +2124,7 @@ test_45i() { $LFS mkdir -i 1 $DIR2/$tfile || error "create remote dir must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "create remote dir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 45i "pdirops: rename src vs remote mkdir" @@ -2018,7 +2139,7 @@ test_46a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46a "pdirops: link vs mkdir ==============" @@ -2032,7 +2153,7 @@ test_46b() { sleep 1 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46b "pdirops: link vs create ==============" @@ -2046,7 +2167,7 @@ test_46c() { sleep 1 link $DIR2/$tfile $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46c "pdirops: link vs link ==============" @@ -2060,7 +2181,7 @@ test_46d() { sleep 1 rm $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46d "pdirops: link vs unlink ==============" @@ -2075,7 +2196,7 @@ test_46e() { sleep 1 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46e "pdirops: link and rename (tgt) ==============" @@ -2090,7 +2211,7 @@ test_46f() { sleep 1 mv $DIR2/$tfile $DIR2/$tfile-3 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46f "pdirops: link and rename (src) ==============" @@ -2104,7 +2225,7 @@ test_46g() { sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46g "pdirops: link vs getattr ==============" @@ -2119,7 +2240,7 @@ test_46h() { ls -lia $DIR2/ > /dev/null check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46h "pdirops: link vs readdir ==============" @@ -2135,7 +2256,7 @@ test_46i() { $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "remote mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 46i "pdirops: link vs remote mkdir" @@ -2150,7 +2271,7 @@ test_47a() { sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47a "pdirops: remote mkdir vs mkdir" @@ -2165,7 +2286,7 @@ test_47b() { multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47b "pdirops: remote mkdir vs create" @@ -2180,7 +2301,7 @@ test_47c() { sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47c "pdirops: remote mkdir vs link" @@ -2195,7 +2316,7 @@ test_47d() { rmdir $DIR2/$tfile || error "unlink must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47d "pdirops: remote mkdir vs unlink" @@ -2211,7 +2332,7 @@ test_47e() { mv -T $DIR2/$tfile-2 $DIR2/$tfile && error "rename must fail" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47e "pdirops: remote mkdir and rename (tgt)" @@ -2226,7 +2347,7 @@ test_47f() { mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47f "pdirops: remote mkdir and rename (src)" @@ -2241,7 +2362,7 @@ test_47g() { stat $DIR2/$tfile > /dev/null || error "stat must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } - rm -r $DIR1/* + rm -rf $DIR/$tfile* return 0 } run_test 47g "pdirops: remote mkdir vs getattr" @@ -2441,8 +2562,8 @@ run_test 54 "rename locking" test_55a() { mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" -#define OBD_FAIL_MDS_RENAME 0x153 - do_facet mds $LCTL set_param fail_loc=0x80000153 +#define OBD_FAIL_MDS_RENAME4 0x156 + do_facet mds $LCTL set_param fail_loc=0x80000156 mv -T $DIR/d1/d2 $DIR/d3/d2 & PID1=$! @@ -2459,8 +2580,8 @@ test_55b() { mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" -#define OBD_FAIL_MDS_RENAME 0x155 - do_facet mds $LCTL set_param fail_loc=0x80000155 +#define OBD_FAIL_MDS_RENAME4 0x156 + do_facet mds $LCTL set_param fail_loc=0x80000156 mv -T $DIR/d1/d2 $DIR/d3/d2 & PID1=$! @@ -2469,13 +2590,64 @@ test_55b() rm -r $DIR2/d1 wait $PID1 && error "(2) mv succeeded" - rm -rf $DIR/d1 + rm -rf $DIR/d3 } run_test 55b "rename vs unlink source dir" +test_55c() +{ + mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" + +#define OBD_FAIL_MDS_RENAME4 0x156 + do_facet mds $LCTL set_param fail_loc=0x156 + + mv -T $DIR/d1/d2 $DIR/d3/d2 & + PID1=$! + sleep 1 + + # while rename is sleeping, open and remove d3 + $MULTIOP $DIR2/d3 D_c & + PID2=$! + sleep 1 + rm -rf $DIR2/d3 + sleep 5 + + # while rename is sleeping 2nd time, close d3 + kill -USR1 $PID2 + wait $PID2 || error "(3) multiop failed" + + wait $PID1 && error "(2) mv succeeded" + + rm -rf $DIR/d1 +} +run_test 55c "rename vs unlink orphan target dir" + +test_55d() +{ + touch $DIR/f1 + +#define OBD_FAIL_MDS_RENAME3 0x155 + do_facet mds $LCTL set_param fail_loc=0x155 + mv $DIR/f1 $DIR/$tdir & + PID1=$! + sleep 2 + + # while rename is sleeping, create $tdir, but as a directory + mkdir -p $DIR2/$tdir || error "(1) mkdir failed" + + # link in reverse locking order + ln $DIR2/f1 $DIR2/$tdir/ + + wait $PID1 && error "(2) mv succeeded" + rm -rf $DIR/f1 +} +run_test 55d "rename file vs link" + test_60() { - [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] || - { skip "Need MDS version at least 2.3.0"; return; } + local MDSVER=$(lustre_build_version $SINGLEMDS) + [ $(version_code $MDSVER) -lt $(version_code 2.3.0) ] && + skip "MDS version $MDSVER must be >= 2.3.0" && return 0 + # Create a file test_mkdir -p $DIR1/$tdir file1=$DIR1/$tdir/file @@ -2695,14 +2867,9 @@ run_test 75 "osc: upcall after unuse lock===================" test_76() { #LU-946 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.53) ]] && skip "Need MDS version at least 2.5.53" && return - [ $MDSCOUNT -ge 2 ] && skip "skip now for LU-4573" && return #LU-4573 remote_mds_nodsh && skip "remote MDS with nodsh" && return local fcount=2048 - local fd - local cmd - local mdt_idx - local mds_idx declare -a fd_list declare -a fid_list @@ -2714,29 +2881,24 @@ test_76() { #LU-946 rm -rf $DIR/$tdir test_mkdir -p $DIR/$tdir - if [ $MDSCOUNT -gt 1 ]; then - mdt_idx=$($LFS getdirstripe -i $DIR/$tdir) - else - mdt_idx=0 - fi - mds_idx=$((mdt_idx + 1)) - proc_ofile="mdt.*$mdt_idx.exports.'$nid'.open_files" + # drop all open locks and close any cached "open" files on the client cancel_lru_locks mdc echo -n "open files " ulimit -n 8096 - for (( i = 0; i < $fcount; i++ )) ; do + for ((i = 0; i < $fcount; i++)); do touch $DIR/$tdir/f_$i - fd=$(free_fd) - cmd="exec $fd<$DIR/$tdir/f_$i" + local fd=$(free_fd) + local cmd="exec $fd<$DIR/$tdir/f_$i" eval $cmd fd_list[i]=$fd echo -n "." done echo - fid_list=($(do_facet mds$mds_idx $LCTL get_param -n $proc_ofile)) + local get_open_fids="$LCTL get_param -n mdt.*.exports.'$nid'.open_files" + local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $get_open_fids)) # Possible errors in openfiles FID list. # 1. Missing FIDs. Check 1 @@ -2769,7 +2931,354 @@ test_76() { #LU-946 } run_test 76 "Verify open file for 2048 files" -test_80() { +nrs_write_read() { + local n=16 + local dir=$DIR/$tdir + local myRUNAS="$1" + + mkdir $dir || error "mkdir $dir failed" + $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed" + chmod 777 $dir + + do_nodes $CLIENTS $myRUNAS dd if=/dev/zero of="$dir/nrs_r_$HOSTNAME"\ + bs=1M count=$n > /dev/null 2>&1 + + for ((i = 0; i < $n; i++)); do + do_nodes $CLIENTS $myRUNAS dd if=/dev/zero\ + of="$dir/nrs_w_$HOSTNAME" bs=1M seek=$i count=1\ + > /dev/null 2>&1 & + local pids_w[$i]=$! + done + do_nodes $CLIENTS sync; + cancel_lru_locks osc + + for ((i = 0; i < $n; i++)); do + do_nodes $CLIENTS $myRUNAS dd if="$dir/nrs_w_$HOSTNAME"\ + of=/dev/zero bs=1M seek=$i count=1 > /dev/null 2>&1 & + local pids_r[$i]=$! + done + cancel_lru_locks osc + + for ((i = 0; i < $n; i++)); do + wait ${pids_w[$i]} + wait ${pids_r[$i]} + done + rm -rf $dir || error "rm -rf $dir failed" +} + +test_77a() { #LU-3266 + do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_policies="fifo" + nrs_write_read + + return 0 +} +run_test 77a "check FIFO NRS policy" + + +test_77b() { #LU-3266 + do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_policies="crrn" + do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_crrn_quantum=1 + + echo "policy: crr-n, crrn_quantum 1" + nrs_write_read + + do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_crrn_quantum=64 + + echo "policy: crr-n, crrn_quantum 64" + nrs_write_read + + return 0 +} +run_test 77b "check CRR-N NRS policy" + +orr_trr() { + local policy=$1 + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies=$policy + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_"$policy"_quantum=1 + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_"$policy"_offset_type="physical" + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_"$policy"_supported="reads" + done + + echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type \ + physical, ${policy}_supported reads" + nrs_write_read + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_${policy}_supported="writes" + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_${policy}_quantum=64 + done + echo "policy: $policy, ${policy}_quantum 64, \ + ${policy}_offset_type physical, ${policy}_supported writes" + nrs_write_read + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_${policy}_supported="reads_and_writes" + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_${policy}_offset_type="logical" + done + echo "policy: $policy, ${policy}_quantum 64, \ + ${policy}_offset_type logical, ${policy}_supported reads_and_writes" + nrs_write_read + + return 0 +} + +test_77c() { #LU-3266 + orr_trr "orr" + return 0 +} +run_test 77c "check ORR NRS policy" + +test_77d() { #LU-3266 + orr_trr "trr" + return 0 +} +run_test 77d "check TRR nrs policy" + +tbf_rule_operate() +{ + local facet=$1 + shift 1 + + do_facet $facet lctl set_param \ + ost.OSS.ost_io.nrs_tbf_rule="$*" + [ $? -ne 0 ] && + error "failed to operate on TBF rules" +} + +test_77e() { + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ nid" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + local idis + local rateis + if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + idis="nid=" + rateis="rate=" + fi + + # Only operate rules on ost1 since OSTs might run on the same OSS + # Add some rules + tbf_rule_operate ost1 "start\ localhost\ ${idis}{0@lo}\ ${rateis}1000" + local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)") + local client_nids=$(nids_list $address "\\") + tbf_rule_operate ost1 "start\ clients\ ${idis}{$client_nids}\ ${rateis}100" + tbf_rule_operate ost1 "start\ others\ ${idis}{*.*.*.*@$NETTYPE}\ ${rateis}50" + nrs_write_read + + # Change the rules + tbf_rule_operate ost1 "change\ localhost\ ${rateis}1001" + tbf_rule_operate ost1 "change\ clients\ ${rateis}101" + tbf_rule_operate ost1 "change\ others\ ${rateis}51" + nrs_write_read + + # Stop the rules + tbf_rule_operate ost1 "stop\ localhost" + tbf_rule_operate ost1 "stop\ clients" + tbf_rule_operate ost1 "stop\ others" + nrs_write_read + + # Cleanup the TBF policy + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && + error "failed to set policy back to fifo" + done + nrs_write_read + return 0 +} +run_test 77e "check TBF NID nrs policy" + +test_77f() { + # Configure jobid_var + local saved_jobid_var=$($LCTL get_param -n jobid_var) + if [ $saved_jobid_var != procname_uid ]; then + set_conf_param_and_check client \ + "$LCTL get_param -n jobid_var" \ + "$FSNAME.sys.jobid_var" procname_uid + fi + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ jobid" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + local idis + local rateis + if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + idis="jobid=" + rateis="rate=" + fi + + # Only operate rules on ost1 since OSTs might run on the same OSS + # Add some rules + tbf_rule_operate ost1 "start\ runas\ ${idis}{iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ ${rateis}1000" + tbf_rule_operate ost1 "start\ iozone_runas\ ${idis}{iozone.$RUNAS_ID}\ ${rateis}100" + tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50" + nrs_write_read "$RUNAS" + + # Change the rules + tbf_rule_operate ost1 "change\ runas\ ${rateis}1001" + tbf_rule_operate ost1 "change\ iozone_runas\ ${rateis}101" + tbf_rule_operate ost1 "change\ dd_runas\ ${rateis}51" + nrs_write_read "$RUNAS" + + # Stop the rules + tbf_rule_operate ost1 "stop\ runas" + tbf_rule_operate ost1 "stop\ iozone_runas" + tbf_rule_operate ost1 "stop\ dd_runas" + nrs_write_read "$RUNAS" + + # Cleanup the TBF policy + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && + error "failed to set policy back to fifo" + done + nrs_write_read "$RUNAS" + + local current_jobid_var=$($LCTL get_param -n jobid_var) + if [ $saved_jobid_var != $current_jobid_var ]; then + set_conf_param_and_check client \ + "$LCTL get_param -n jobid_var" \ + "$FSNAME.sys.jobid_var" $saved_jobid_var + fi + return 0 +} +run_test 77f "check TBF JobID nrs policy" + +test_77g() { + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ nid" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ jobid" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + local idis + local rateis + if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + idis="jobid=" + rateis="rate=" + fi + + # Add a rule that only valid for Jobid TBF. If direct change between + # TBF types is not supported, this operation will fail. + tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50" + + # Cleanup the TBF policy + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && + error "failed to set policy back to fifo" + done + return 0 +} +run_test 77g "Change TBF type directly" + +test_77h() { + [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] || + { skip "Need OST version at least 2.8.55"; return 0; } + + local old_policy=$(do_facet ost1 \ + lctl get_param ost.OSS.ost_io.nrs_policies) + local new_policy + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="abc" + [ $? -eq 0 ] && error "should return error" + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ abc" + [ $? -eq 0 ] && error "should return error" + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ reg" + [ $? -eq 0 ] && error "should return error" + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ reg\ abc" + [ $? -eq 0 ] && error "should return error" + + do_facet ost1 lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ abc\ efg" + [ $? -eq 0 ] && error "should return error" + + new_policy=$(do_facet ost1 lctl get_param ost.OSS.ost_io.nrs_policies) + [ $? -eq 0 ] || error "shouldn't LBUG" + + [ "$old_policy" = "$new_policy" ] || error "NRS policy should be same" + + return 0 +} +run_test 77h "Wrong policy name should report error, not LBUG" + +test_78() { #LU-6673 + local rc + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="orr" & + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_orr_quantum=1 + rc=$? + # Valid return codes are: + # 0: Tuning succeeded + # ENODEV: Policy is still stopped + # EAGAIN: Policy is being initialized + [ $rc -eq 0 -o $rc -eq 19 -o $rc -eq 11 ] || + error "Expected set_param to return 0|ENODEV|EAGAIN" + done + + # Cleanup the ORR policy + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && + error "failed to set policy back to fifo" + done + return 0 +} +run_test 78 "Enable policy and specify tunings right away" + +test_80a() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return local MDTIDX=1 local mdt_index @@ -2789,12 +3298,12 @@ test_80() { # give multiop a chance to open sleep 1 - $LFS mv -M $MDTIDX $DIR1/$tdir && + $LFS migrate -m $MDTIDX $DIR1/$tdir && error "migrate open files should failed with open files" kill -USR1 $pid - $LFS mv -M $MDTIDX $DIR1/$tdir || + $LFS migrate -m $MDTIDX $DIR1/$tdir || error "migrate remote dir error" echo "Finish migration, then checking.." @@ -2809,7 +3318,130 @@ test_80() { rm -rf $DIR1/$tdir || error "rm dir failed after migration" } -run_test 80 "migrate directory when some children is being opened" +run_test 80a "migrate directory when some children is being opened" + +cleanup_80b() { + trap 0 + kill -9 $migrate_pid +} + +test_80b() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + local migrate_dir1=$DIR1/$tdir/migrate_dir + local migrate_dir2=$DIR2/$tdir/migrate_dir + local migrate_run=$LUSTRE/tests/migrate.sh + local start_time + local end_time + local show_time=1 + local mdt_idx + local rc=0 + local rc1=0 + + trap cleanup_80b EXIT + #prepare migrate directory + mkdir -p $migrate_dir1 + for F in {1,2,3,4,5}; do + echo "$F$F$F$F$F" > $migrate_dir1/file$F + echo "$F$F$F$F$F" > $DIR/$tdir/file$F + done + + #migrate the directories among MDTs + ( + while true; do + mdt_idx=$((RANDOM % MDSCOUNT)) + $LFS migrate -m $mdt_idx $migrate_dir1 2&>/dev/null || + rc=$? + [ $rc -ne 0 -o $rc -ne 16 ] || break + done + ) & + migrate_pid=$! + + echo "start migration thread $migrate_pid" + #Access the files at the same time + start_time=$(date +%s) + echo "accessing the migrating directory for 5 minutes..." + while true; do + ls $migrate_dir2 > /dev/null || { + echo "read dir fails" + break + } + diff -u $DIR2/$tdir/file1 $migrate_dir2/file1 || { + echo "access file1 fails" + break + } + + cat $migrate_dir2/file2 > $migrate_dir2/file3 || { + echo "access file2/3 fails" + break + } + + echo "aaaaa" > $migrate_dir2/file4 > /dev/null || { + echo "access file4 fails" + break + } + + stat $migrate_dir2/file5 > /dev/null || { + echo "stat file5 fails" + break + } + + touch $migrate_dir2/source_file > /dev/null || rc1=$? + [ $rc1 -ne 0 -o $rc1 -ne 1 ] || { + echo "touch file failed with $rc1" + break; + } + + if [ -e $migrate_dir2/source_file ]; then + ln $migrate_dir2/source_file $migrate_dir2/link_file \ + 2&>/dev/null || rc1=$? + if [ -e $migrate_dir2/link_file ]; then + rm -rf $migrate_dir2/link_file + fi + + mrename $migrate_dir2/source_file \ + $migrate_dir2/target_file 2&>/dev/null || rc1=$? + [ $rc1 -ne 0 -o $rc1 -ne 1 ] || { + echo "rename failed with $rc1" + break + } + + if [ -e $migrate_dir2/target_file ]; then + rm -rf $migrate_dir2/target_file 2&>/dev/null || + rc1=$? + else + rm -rf $migrate_dir2/source_file 2&>/dev/null || + rc1=$? + fi + [ $rc1 -ne 0 -o $rc1 -ne 1 ] || { + echo "unlink failed with $rc1" + break + } + fi + + end_time=$(date +%s) + duration=$((end_time - start_time)) + if [ $((duration % 10)) -eq 0 ]; then + if [ $show_time -eq 1 ]; then + echo "...$duration seconds" + show_time=0 + fi + else + show_time=1 + fi + + kill -0 $migrate_pid || { + echo "migration stopped 1" + break + } + + [ $duration -ge 300 ] && break + done + + #check migration are still there + kill -0 $migrate_pid || error "migration stopped 2" + cleanup_80b +} +run_test 80b "Accessing directory during migration" test_81() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return @@ -2825,7 +3457,7 @@ test_81() { touch d0/0 || error "create 0 failed" mv d0/0 d1/0 || error "rename d0/0 d1/0 failed" stat d0/0 && error "stat mv filed succeed" - mv $DIR2/$tdir/d1/0 $DIR2/$tdir/d0/0 || "rename d1/0 d0/0 failed" + mv $DIR2/$tdir/d1/0 $DIR2/$tdir/d0/0 || error "rename d1/0 d0/0 failed" stat d0/0 || error "stat failed" local t=$(ls -ai $DIR1/$tdir/d0 | sort -u | wc -l) @@ -2839,8 +3471,167 @@ test_81() { } run_test 81 "rename and stat under striped directory" +test_82() { + [[ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.6.91) ]] || + { skip "Need MDS version at least 2.6.92"; return 0; } + + # Client 1 creates a file. + multiop_bg_pause $DIR1/$tfile O_ac || error "multiop_bg_pause 1" + pid1=$! + # Client 2 opens the file. + multiop_bg_pause $DIR2/$tfile o_Ac || error "multiop_bg_pause 2" + pid2=$! + # Client 1 makes the file an orphan. + rm $DIR1/$tfile || error "rm" + # Client 2 sets EA "user.multiop". + kill -s USR1 $pid2 + wait $pid2 || error "multiop 2" + # Client 1 gets EA "user.multiop". This used to fail because the EA + # cache refill would get "trusted.link" from mdd_xattr_list() but + # -ENOENT when trying to get "trusted.link"'s value. See also sanity + # 102q. + kill -s USR1 $pid1 + wait $pid1 || error "multiop 1" +} +run_test 82 "fsetxattr and fgetxattr on orphan files" + +test_83() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + local pid1 + local pid2 + + ( + cd $DIR1 + while true; do + $LFS mkdir -i1 -c2 $tdir + rmdir $tdir + done + ) & + pid1=$! + echo "start pid $pid1 to create/unlink striped directory" + + # Access the directory at the same time + ( + cd $DIR2 + while true; do + stat $tdir > /dev/null 2>&1 + done + ) & + pid2=$! + echo "start pid $pid2 to stat striped directory" + + sleep 120 + kill $pid1 $pid2 + wait $pid1 $pid2 + + return 0 +} +run_test 83 "access striped directory while it is being created/unlinked" + +test_90() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + local pid1 + local pid2 + local duration=180 + + [ "$SLOW" = "yes" ] && duration=600 + # Open/Create under striped directory + ( + cd $DIR1 + while true; do + $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1 + touch $tdir/f{0..3} > /dev/null 2>&1 + done + ) & + pid1=$! + echo "start pid $pid1 to open/create under striped directory" + + # unlink the striped directory at the same time + ( + cd $DIR2 + while true; do + rm -rf $tdir > /dev/null 2>&1 + done + ) & + pid2=$! + echo "start pid $pid2 to unlink striped directory" + + sleep $duration + + kill $pid1 $pid2 + wait $pid1 $pid2 + + return 0 +} +run_test 90 "open/create and unlink striped directory" + +test_91() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + local pid1 + local pid2 + local duration=180 + + [ "$SLOW" = "yes" ] && duration=600 + # chmod striped directory + ( + cd $DIR1 + while true; do + $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1 + chmod go+w $tdir > /dev/null 2>&1 + done + ) & + pid1=$! + echo "start pid $pid1 to chmod striped directory" + + # unlink the striped directory at the same time + ( + cd $DIR2 + while true; do + rm -rf $tdir > /dev/null 2>&1 + done + ) & + pid2=$! + echo "start pid $pid2 to unlink striped directory" + + sleep $duration + + kill $pid1 $pid2 + wait $pid1 $pid2 + + return 0 +} +run_test 91 "chmod and unlink striped directory" + +test_92() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + + local fd=$(free_fd) + local cmd="exec $fd<$DIR1/$tdir" + $LFS setdirstripe -c$MDSCOUNT $DIR1/$tdir || error "mkdir $tdir fails" + eval $cmd + cmd="exec $fd<&-" + trap "eval $cmd" EXIT + cd $DIR1/$tdir || error "cd $DIR1/$tdir fails" + rmdir ../$tdir || error "rmdir ../$tdir fails" + + #define OBD_FAIL_LLITE_NO_CHECK_DEAD 0x1408 + $LCTL set_param fail_loc=0x1408 + mkdir $DIR2/$tdir/dir && error "create dir succeeds" + $LFS setdirstripe -i1 $DIR2/$tdir/remote_dir && + error "create remote dir succeeds" + $LCTL set_param fail_loc=0 + eval $cmd + return 0 +} +run_test 92 "create remote directory under orphan directory" + log "cleanup: ======================================================" +# kill and wait in each test only guarentee script finish, but command in script +# like 'rm' 'chmod' may still be running, wait for all commands to finish +# otherwise umount below will fail +wait_update $HOSTNAME "fuser -m $MOUNT2" "" || true + [ "$(mount | grep $MOUNT2)" ] && umount $MOUNT2 complete $SECONDS