X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanityn.sh;h=1e65fda2f8b7662d8c73311cbc9bbf7b19212abe;hp=aaf9035ba6fc4707528aa5f0aaae9f070180cb2f;hb=849d7d5b1b4cabb7578c3ab5aaf271e90dd33864;hpb=ad720ef6863538c83d4b74d6caef2d32e32a7f9b diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh index aaf9035..1e65fda 100644 --- 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} @@ -40,9 +32,12 @@ init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} init_logging -[ $(facet_fstype $SINGLEMDS) = "zfs" ] && -# bug number for skipped test: LU-2840 LU-2776 - ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21 51a" +if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then +# bug number for skipped test: LU-2189 LU-2776 + ALWAYS_EXCEPT="$ALWAYS_EXCEPT 36 51a" +# LU-2829 / LU-2887 - make allowances for ZFS slowness + TEST33_NFILES=${TEST33_NFILES:-1000} +fi [ "$SLOW" = "no" ] && EXCEPT_SLOW="33a" @@ -57,7 +52,7 @@ 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 +rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}* SAMPLE_FILE=$TMP/$(basename $0 .sh).junk dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=1 @@ -169,6 +164,18 @@ 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" @@ -489,28 +496,28 @@ test_21() { # Bug 5907 run_test 21 " Try to remove mountpoint on another dir ====" test_23() { # Bug 5972 - echo "others should see updated atime while another read" > $DIR1/f23 - + local at_diff=$(do_facet $SINGLEMDS \ + $LCTL get_param -n mdd.*MDT0000*.atime_diff | head -n1) + echo "atime should be updated while another read" > $DIR1/$tfile + # clear the lock(mode: LCK_PW) gotten from creating operation cancel_lru_locks osc - - time1=`date +%s` - #MAX_ATIME_DIFF 60, we update atime only if older than 60 seconds - sleep 61 - - multiop_bg_pause $DIR1/f23 or20_c || return 1 + time1=$(date +%s) + echo "now is $time1" + sleep $((at_diff + 1)) + + echo "starting reads" + multiop_bg_pause $DIR1/$tfile or20_c || return 1 # with SOM and opencache enabled, we need to close a file and cancel # open lock to get atime propogated to MDS - kill -USR1 $! + kill -USR1 $! || return 2 cancel_lru_locks mdc - time2=`stat -c "%X" $DIR2/f23` - - if (( $time2 <= $time1 )); then - error "atime doesn't update among nodes" - fi + time2=$(stat -c "%X" $DIR/$tfile) + echo "new atime is $time2" - rm -f $DIR1/f23 || error "rm -f $DIR1/f23 failed" + [ $time2 -gt $time1 ] || error "atime was not updated" + rm -f $DIR1/$tfile || error "rm -f $DIR1/$tfile failed" true } run_test 23 " others should see updated atime while another read====" @@ -523,7 +530,7 @@ test_24a() { lfs df -i $DIR2 || error "lfs df -i $DIR2 failed" lfs df $DIR1/$tfile || error "lfs df $DIR1/$tfile failed" lfs df -ih $DIR2/$tfile || error "lfs df -ih $DIR2/$tfile failed" - + OSC=`lctl dl | awk '/-osc-|OSC.*MNT/ {print $4}' | head -n 1` # OSC=`lctl dl | awk '/-osc-/ {print $4}' | head -n 1` lctl --device %$OSC deactivate @@ -624,11 +631,11 @@ test_27() { DD2_PID=$! usleep 50 log "dd 1 started" - + dd if=/dev/zero of=$DIR1/$tfile bs=$((16384-1024))k conv=notrunc count=1 seek=4 & DD1_PID=$! log "dd 2 started" - + sleep 1 dd if=/dev/zero of=$DIR1/$tfile bs=8k conv=notrunc count=1 seek=0 log "dd 3 finished" @@ -640,7 +647,7 @@ run_test 27 "align non-overlapping extent locks from request ===" test_28() { # bug 9977 ECHO_UUID="ECHO_osc1_UUID" - tOST=`$LCTL dl | | awk '/-osc-|OSC.*MNT/ { print $4 }' | head -1` + tOST=$($LCTL dl | awk '/-osc-|OSC.*MNT/ { print $4 }' | head -n1) $LFS setstripe $DIR1/$tfile -S 1048576 -i 0 -c 2 tOBJID=`$LFS getstripe $DIR1/$tfile | awk '$1 == 1 {print $2}'` @@ -742,8 +749,8 @@ enable_lockless_truncate() { } test_32a() { # bug 11270 - local p="$TMP/sanityN-$TESTNAME.parameters" - save_lustre_params $HOSTNAME osc.*.lockless_truncate > $p + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + save_lustre_params client "osc.*.lockless_truncate" > $p cancel_lru_locks osc enable_lockless_truncate 1 rm -f $DIR1/$tfile @@ -755,14 +762,14 @@ 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" + [ $(calc_osc_stats lockless_truncate) -ne 0 ] || + error "not cached truncate isn't lockless" log "disabled lockless truncate" enable_lockless_truncate 0 @@ -781,15 +788,19 @@ run_test 32a "lockless truncate" test_32b() { # bug 11270 remote_ost_nodsh && skip "remote OST with nodsh" && return - local node - local p="$TMP/sanityN-$TESTNAME.parameters" - save_lustre_params $HOSTNAME "osc.*.contention_seconds" > $p - for node in $(osts_nodes); do - save_lustre_params $node "ldlm.namespaces.filter-*.max_nolock_bytes" >> $p - save_lustre_params $node "ldlm.namespaces.filter-*.contended_locks" >> $p - save_lustre_params $node "ldlm.namespaces.filter-*.contention_seconds" >> $p - done - clear_osc_stats + local node + local facets=$(get_facets OST) + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + + save_lustre_params client "osc.*.contention_seconds" > $p + save_lustre_params $facets \ + "ldlm.namespaces.filter-*.max_nolock_bytes" >> $p + save_lustre_params $facets \ + "ldlm.namespaces.filter-*.contended_locks" >> $p + save_lustre_params $facets \ + "ldlm.namespaces.filter-*.contention_seconds" >> $p + clear_osc_stats + # agressive lockless i/o settings for node in $(osts_nodes); do do_node $node 'lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes 2000000; lctl set_param -n ldlm.namespaces.filter-*.contended_locks 0; lctl set_param -n ldlm.namespaces.filter-*.contention_seconds 60' @@ -831,8 +842,8 @@ print_jbd_stat () { varsvc=${mds}_svc dev=$(basename $(do_facet $mds "lctl get_param -n osd*.${!varsvc}.mntdev|\ xargs readlink -f" )) - val=$(do_facet $mds "cat /proc/fs/jbd*/${dev}{,:*,-*}/info 2>/dev/null|\ - head -1") + val=$(do_facet $mds "cat /proc/fs/jbd*/${dev}{,:*,-*}/info 2>/dev/null | + head -n1") val=${val%% *}; stat=$(( stat + val)) done @@ -851,7 +862,8 @@ test_33a() { local param_file=$TMP/$tfile-params local fstype=$(facet_fstype $SINGLEMDS) - save_lustre_params $(comma_list $(mdts_nodes)) "mdt.*.commit_on_sharing" > $param_file + save_lustre_params $(get_facets MDS) \ + "mdt.*.commit_on_sharing" > $param_file local COS local jbdold="N/A" @@ -902,8 +914,9 @@ test_33b() { local nfiles=${TEST33_NFILES:-10000} local param_file=$TMP/$tfile-params - save_lustre_params $(comma_list $(mdts_nodes)) \ - "mdt.*.commit_on_sharing" > $param_file + save_lustre_params $(get_facets MDS) \ + "mdt.*.commit_on_sharing" > $param_file + local COS local jbdold local jbdnew @@ -1019,8 +1032,9 @@ run_test 34 "no lock timeout under IO" test_35() { # bug 17645 local generation=[] local count=0 - for imp in /proc/fs/lustre/mdc/$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 @@ -1062,10 +1076,20 @@ 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 @@ -1082,33 +1106,29 @@ test_36() { #bug 16417 i=0 SIZE=50 let SIZE_B=SIZE*1024*1024 + sync; sleep 5; sync # wait for delete thread while [ $i -le 10 ]; do lctl mark "start test" - local before=$($LFS df | awk '{ if ($1 ~/^filesystem/) \ - { print $5; exit} }') + local before=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }') dd if=/dev/zero of=$DIR1/$tdir/$tfile bs=1M count=$SIZE || error "dd $DIR1/$tdir/$tfile ${SIZE}MB failed" sync # sync data from client cache sync_all_data # sync data from server cache (delayed allocation) - sleep 1 - local after_dd=$($LFS df | awk '{ if ($1 ~/^filesystem/) \ - { print $5; exit} }') + sleep 2 + local after_dd=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }') multiop_bg_pause $DIR2/$tdir/$tfile O_r${SIZE_B}c || return 3 read_pid=$! rm -f $DIR1/$tdir/$tfile kill -USR1 $read_pid wait $read_pid wait_delete_completed - local after=$($LFS df | awk '{ if ($1 ~/^filesystem/) \ - { print $5; exit} }') + local after=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }') echo "*** cycle($i) *** before($before) after_dd($after_dd)" \ "after($after)" # this free space! not used - if [ $after_dd -ge $after ]; then - error "space leaked" - return 1; - fi + (( $after_dd <= $after)) || + error "space leaked after_dd:$after_dd > after:$after" let i=i+1 done } @@ -1242,6 +1262,7 @@ check_pdo_conflict() { test_40a() { #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 @@ -1259,11 +1280,6 @@ test_40a() { rmdir $DIR2/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" - if [ $MDSCOUNT -ge 2 ]; then - $LFS mkdir -i 1 $DIR2/$tfile-6 - check_pdo_conflict $PID1 || error "remote mkdir is blocked" - fi - # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 @@ -1294,11 +1310,6 @@ test_40b() { check_pdo_conflict $PID1 || error "unlink is blocked" # all operations above shouldn't wait the first one - if [ $MDSCOUNT -ge 2 ]; then - $LFS mkdir -i 1 $DIR2/$tfile-6 - check_pdo_conflict $PID1 || error "remote mkdir is blocked" - fi - check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 rm -r $DIR1/* @@ -1328,11 +1339,6 @@ test_40c() { rmdir $DIR2/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" - if [ $MDSCOUNT -ge 2 ]; then - $LFS mkdir -i 1 $DIR2/$tfile-6 - check_pdo_conflict $PID1 || error "remote mkdir is blocked" - fi - # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 @@ -1363,11 +1369,6 @@ test_40d() { rmdir $DIR2/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" - if [ $MDSCOUNT -ge 2 ]; then - $LFS mkdir -i 1 $DIR2/$tfile-6 - check_pdo_conflict $PID1 || error "remote mkdir is blocked" - fi - # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 @@ -1395,11 +1396,6 @@ test_40e() { rmdir $DIR2/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" - if [ $MDSCOUNT -ge 2 ]; then - $LFS mkdir -i 1 $DIR2/$tfile-6 - check_pdo_conflict $PID1 || error "remote mkdir is blocked" - fi - # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 @@ -2299,6 +2295,9 @@ test_51a() { run_test 51a "layout lock: refresh layout should work" test_51b() { + [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.59) ]] || + { skip "Need MDS version at least 2.3.59"; return 0; } + local tmpfile=`mktemp` # create an empty file @@ -2324,29 +2323,216 @@ test_51b() { run_test 51b "layout lock: glimpse should be able to restart if layout changed" test_51c() { - # create an empty file - $MCREATE $DIR1/$tfile + [ $OSTCOUNT -ge 2 ] || { skip "need at least 2 osts"; return; } + + # set default layout to have 1 stripe + mkdir -p $DIR1/$tdir + $LFS setstripe -c 1 $DIR1/$tdir + + # create a file with empty layout + $MCREATE $DIR1/$tdir/$tfile #define OBD_FAIL_MDS_LL_BLOCK 0x172 - $LCTL set_param fail_loc=0x172 + do_facet $SINGLEMDS $LCTL set_param fail_loc=0x172 # change the layout of testing file - echo "Setting layout ..." - $LFS setstripe -c $OSTCOUNT $DIR1/$tfile & + echo "Setting layout to have $OSTCOUNT stripes ..." + $LFS setstripe -c $OSTCOUNT $DIR1/$tdir/$tfile & pid=$! sleep 1 - # get layout of this file should wait until dd is finished - local stripecnt=`$LFS getstripe -c $DIR2/$tfile` - [ $stripecnt -eq $OSTCOUNT ] || error "layout wrong" + # write something to the file, it should be blocked on fetching layout + dd if=/dev/zero of=$DIR2/$tdir/$tfile bs=1k count=1 conv=notrunc + local cnt=$($LFS getstripe -c $DIR2/$tdir/$tfile) + [ $cnt -eq $OSTCOUNT ] || error "have $cnt stripes, expected $OSTCOUNT" - rm -f $DIR1/$tfile + rm -fr $DIR1/$tdir } run_test 51c "layout lock: IT_LAYOUT blocked and correct layout can be returned" +test_51d() { + dd if=/dev/zero of=/$DIR1/$tfile bs=1M count=1 + cancel_lru_locks mdc + + # open should grant LAYOUT lock, mmap and read will install pages + $MULTIOP $DIR1/$tfile oO_RDWR:SMR_Uc & + local PID=$! + sleep 1 + + # rss before revoking + local br=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}') + echo "Before revoking layout lock: $br KB mapped" + + # delete the file will revoke layout lock + rm -f $DIR2/$tfile + + # rss after revoking + local ar=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}') + + kill -USR1 $PID + wait $PID || error + + [ $ar -eq 0 ] || error "rss before: $br, after $ar, some pages remained" +} +run_test 51d "layout lock: losing layout lock should clean up memory map region" + +test_54_part1() +{ + echo "==> rename vs getattr vs setxattr should not deadlock" + mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed" + + do_facet mds $LCTL set_param fail_loc=$1 + + mv -T $DIR/d1/d2/d3 $DIR/d1/d3 & + PID1=$! + sleep 1 + + stat $DIR/d1/d2 & + PID2=$! + sleep 1 + + setfattr -n user.attr1 -v value1 $DIR2/d1 || error "(2) setfattr failed" + wait $PID1 || error "(3) mv failed" + wait $PID2 || error "(4) stat failed" + echo + + rm -rf $DIR/d1 +} + +test_54_part2() { + echo "==> rename vs getattr vs open vs getattr should not deadlock" + mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed" + + do_facet mds $LCTL set_param fail_loc=$1 + + mv -T $DIR/d1/d2/d3 $DIR/d1/d3 & + PID1=$! + sleep 1 + + stat $DIR/d1/d2 & + PID2=$! + sleep 1 + + $MULTIOP $DIR2/d1/d2 Oc & + PID3=$! + sleep 1 + + stat $DIR/d1 || error "(2) stat failed" + + wait $PID1 || error "(3) mv failed" + wait $PID2 || error "(4) stat failed" + wait $PID3 && error "(5) multiop failed" + echo + rm -rf $DIR/d1 +} + +test_54() { + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + save_lustre_params client "llite.*.xattr_cache" > $p + lctl set_param llite.*.xattr_cache 1 || + { skip "xattr cache is not supported"; return 0; } + +#define OBD_FAIL_MDS_RENAME 0x153 +#define OBD_FAIL_MDS_RENAME2 0x154 + test_54_part1 0x80000153 || error 10 + test_54_part1 0x80000154 || error 11 + test_54_part2 0x80000153 || error 12 + test_54_part2 0x80000154 || error 13 + + restore_lustre_params < $p + rm -f $p +} +run_test 54 "rename locking" + +test_55a() { + 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=0x80000156 + + mv -T $DIR/d1/d2 $DIR/d3/d2 & + PID1=$! + sleep 1 + + rm -r $DIR2/d3 + wait $PID1 && error "(2) mv succeeded" + + rm -rf $DIR/d1 +} +run_test 55a "rename vs unlink target dir" + +test_55b() +{ + 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=0x80000156 + + mv -T $DIR/d1/d2 $DIR/d3/d2 & + PID1=$! + sleep 1 + + rm -r $DIR2/d1 + wait $PID1 && error "(2) mv succeeded" + + 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 @@ -2430,10 +2616,564 @@ test_70b() { # LU-2781 } run_test 70b "remove files after calling rm_entry" +test_71() { + local server_version=$(lustre_version_code $SINGLEMDS) + + [[ $server_version -lt $(version_code 2.1.6) ]] && + skip "Need MDS version at least 2.1.6" && return + + # Patch not applied to 2.2 and 2.3 branches + [[ $server_version -ge $(version_code 2.2.0) ]] && + [[ $server_version -lt $(version_code 2.4.0) ]] && + skip "Need MDS version at least 2.4.0" && return + + checkfiemap --test || + { skip "checkfiemap not runnable: $?" && return; } + # write data this way: hole - data - hole - data + dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1 + [ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tfile) + 1)))" = \ + "zfs" ] && + skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0 + dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1 + GET_STAT="lctl get_param -n ldlm.services.ldlm_cbd.stats" + stat $DIR2/$tfile + local can1=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') + echo $can1 + checkfiemap $DIR2/$tfile 81920 || + error "data is not flushed from client" + local can2=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') + echo $can2 + + # common case of "create file, copy file" on a single node + # should not flush data from ost + dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1 + dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1 + stat $DIR1/$tfile + local can3=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') + echo $can3 + checkfiemap $DIR1/$tfile 81920 || + error 4 + local can4=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') + echo $can2 + [ $can3 -eq $can4 ] || error $((can2-can1)) "cancel RPC occured." +} +run_test 71 "correct file map just after write operation is finished" + +test_72() { + local p="$TMP/sanityN-$TESTNAME.parameters" + local tlink1 + local tlink2 + save_lustre_params client "llite.*.xattr_cache" > $p + lctl set_param llite.*.xattr_cache 1 || + { skip "xattr cache is not supported"; return 0; } + + touch $DIR1/$tfile + setfattr -n user.attr1 -v value1 $DIR1/$tfile || + error "setfattr1 failed" + getfattr -n user.attr1 $DIR2/$tfile | grep value1 || + error "getfattr1 failed" + setfattr -n user.attr1 -v value2 $DIR2/$tfile || + error "setfattr2 failed" + getfattr -n user.attr1 $DIR1/$tfile | grep value2 || + error "getfattr2 failed" + + # check that trusted.link is consistent + tlink1=$(getfattr -n trusted.link $DIR1/$tfile | md5sum) + ln $DIR2/$tfile $DIR2/$tfile-2 || error "failed to link" + tlink2=$(getfattr -n trusted.link $DIR1/$tfile | md5sum) + echo "$tlink1 $tlink2" + [ "$tlink1" = "$tlink2" ] && error "trusted.link should have changed!" + + rm -f $DIR2/$tfile + + restore_lustre_params < $p + rm -f $p +} +run_test 72 "getxattr/setxattr cache should be consistent between nodes" + +test_73() { + local p="$TMP/sanityN-$TESTNAME.parameters" + save_lustre_params client "llite.*.xattr_cache" > $p + lctl set_param llite.*.xattr_cache 1 || + { skip "xattr cache is not supported"; return 0; } + + touch $DIR1/$tfile + setfattr -n user.attr1 -v value1 $DIR1/$tfile || + error "setfattr1 failed" + getfattr -n user.attr1 $DIR2/$tfile || error "getfattr1 failed" + getfattr -n user.attr1 $DIR1/$tfile || error "getfattr2 failed" + clear_llite_stats + # PR lock should be cached by now on both clients + getfattr -n user.attr1 $DIR1/$tfile || error "getfattr3 failed" + # 2 hits for getfattr(0)+getfattr(size) + [ $(calc_llite_stats getxattr_hits) -eq 2 ] || error "not cached in $DIR1" + getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed" + # 4 hits for more getfattr(0)+getfattr(size) + [ $(calc_llite_stats getxattr_hits) -eq 4 ] || error "not cached in $DIR2" + rm -f $DIR2/$tfile + + restore_lustre_params < $p + rm -f $p +} +run_test 73 "getxattr should not cause xattr lock cancellation" + +test_74() { + [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.93) ] && + skip "Need MDS version at least 2.4.93" && return + + dd if=/dev/zero of=$DIR1/$tfile-1 bs=1K count=1 + dd if=/dev/zero of=$DIR1/$tfile-2 bs=1K count=1 + flocks_test 4 $DIR1/$tfile-1 $DIR2/$tfile-2 +} +run_test 74 "flock deadlock: different mounts ==============" + +# LU-3889 +test_75() { + $LFS setstripe -c 2 -S 1m -i 0 $DIR1/$tfile + dd if=/dev/zero of=$DIR1/$tfile bs=1M count=2 + cancel_lru_locks osc + + dd of=$DIR1/$tfile if=/dev/zero bs=1M count=1 seek=1 conv=notrunc + sync + + # define OBD_FAIL_LDLM_ENQUEUE_HANG 0x31d + $LCTL set_param fail_loc=0x31d + stat -c %s $DIR1/$tfile & + local pid=$! + sleep 1 + kill -9 $pid + + # For bad lock error handler we should ASSERT and got kernel panic here + sleep 4 + $LCTL set_param fail_loc=0 +} +run_test 75 "osc: upcall after unuse lock===================" + +test_76() { #LU-946 + [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.53) ]] && + skip "Need MDS version at least 2.5.53" && return + + remote_mds_nodsh && skip "remote MDS with nodsh" && return + local fcount=2048 + declare -a fd_list + declare -a fid_list + + if remote_mds; then + nid=$($LCTL list_nids | sed "s/\./\\\./g") + else + nid="0@lo" + fi + + rm -rf $DIR/$tdir + test_mkdir -p $DIR/$tdir + + # 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 + touch $DIR/$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 + + 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 + # 2. Extra FIDs. Check 1 + # 3. Duplicated FID. Check 2 + # 4. Invalid FIDs. Check 2 + # 5. Valid FID, points to some other file. Check 3 + + # Check 1 + [ ${#fid_list[@]} -ne $fcount ] && + error "${#fid_list[@]} != $fcount open files" + + for (( i = 0; i < $fcount; i++ )) ; do + cmd="exec ${fd_list[i]} /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" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + # Only operate rules on ost0 since OSTs might run on the same OSS + # Add some rules + tbf_rule_operate ost0 "start\ localhost\ {0@lo}\ 1000" + local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)") + local client_nids=$(nids_list $address "\\") + tbf_rule_operate ost0 "start\ clients\ {$client_nids}\ 100" + tbf_rule_operate ost0 "start\ others\ {*.*.*.*@$NETTYPE}\ 50" + nrs_write_read + + # Change the rules + tbf_rule_operate ost0 "change\ localhost\ 1001" + tbf_rule_operate ost0 "change\ clients\ 101" + tbf_rule_operate ost0 "change\ others\ 51" + nrs_write_read + + # Stop the rules + tbf_rule_operate ost0 "stop\ localhost" + tbf_rule_operate ost0 "stop\ clients" + tbf_rule_operate ost0 "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" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + # Only operate rules on ost0 since OSTs might run on the same OSS + # Add some rules + tbf_rule_operate ost0 "start\ runas\ {iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ 1000" + tbf_rule_operate ost0 "start\ iozone_runas\ {iozone.$RUNAS_ID}\ 100" + tbf_rule_operate ost0 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50" + nrs_write_read "$RUNAS" + + # Change the rules + tbf_rule_operate ost0 "change\ runas\ 1001" + tbf_rule_operate ost0 "change\ iozone_runas\ 101" + tbf_rule_operate ost0 "change\ dd_runas\ 51" + nrs_write_read "$RUNAS" + + # Stop the rules + tbf_rule_operate ost0 "stop\ runas" + tbf_rule_operate ost0 "stop\ iozone_runas" + tbf_rule_operate ost0 "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" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + # Add a rule that only valid for Jobid TBF. If direct change between + # TBF types is not supported, this operation will fail. + tbf_rule_operate ost0 "start\ dd_runas\ {dd.$RUNAS_ID}\ 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_80() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + local MDTIDX=1 + local mdt_index + local i + local file + local pid + + mkdir -p $DIR1/$tdir/dir + createmany -o $DIR1/$tdir/dir/f 10 || + error "create files under remote dir failed $i" + + cp /etc/passwd $DIR1/$tdir/$tfile + + #migrate open file should fails + multiop_bg_pause $DIR2/$tdir/$tfile O_c || error "open $file failed" + pid=$! + # give multiop a chance to open + sleep 1 + + $LFS migrate -m $MDTIDX $DIR1/$tdir && + error "migrate open files should failed with open files" + + kill -USR1 $pid + + $LFS migrate -m $MDTIDX $DIR1/$tdir || + error "migrate remote dir error" + + echo "Finish migration, then checking.." + for file in $(find $DIR1/$tdir); do + mdt_index=$($LFS getstripe -M $file) + [ $mdt_index == $MDTIDX ] || + error "$file is not on MDT${MDTIDX}" + done + + diff /etc/passwd $DIR1/$tdir/$tfile || + error "file different after migration" + + rm -rf $DIR1/$tdir || error "rm dir failed after migration" +} +run_test 80 "migrate directory when some children is being opened" + +test_81() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + + rm -rf $DIR1/$tdir + + mkdir -p $DIR1/$tdir + + $LFS setdirstripe -i0 -c$MDSCOUNT $DIR1/$tdir/d0 + $LFS setdirstripe -i0 -c$MDSCOUNT $DIR1/$tdir/d1 + + cd $DIR1/$tdir + 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" + stat d0/0 || error "stat failed" + + local t=$(ls -ai $DIR1/$tdir/d0 | sort -u | wc -l) + + if [ $t -ne 3 ]; then + ls -ai $DIR1/$tdir/d0 + error "expect 3 get $t" + fi + + return 0 +} +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 $pid1 || error "multiop 1" +} +run_test 82 "fsetxattr and fgetxattr on orphan files" + log "cleanup: ======================================================" [ "$(mount | grep $MOUNT2)" ] && umount $MOUNT2 complete $SECONDS +rm -f $SAMPLE_FILE check_and_cleanup_lustre exit_status