X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=sidebyside;f=lustre%2Ftests%2Fsanityn.sh;h=91f9f9a3aa17b45093c3f4a24ab9c2eb84b1949d;hb=HEAD;hp=ca804b64c343557fbc4651cef577dad78e1d813b;hpb=08827d622325db02072d3a9757475f0c416da329;p=fs%2Flustre-release.git diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh index ca804b6..896dce9 100755 --- a/lustre/tests/sanityn.sh +++ b/lustre/tests/sanityn.sh @@ -19,16 +19,12 @@ init_logging ALWAYS_EXCEPT="$SANITYN_EXCEPT " # bug number for skipped test: LU-7105 -ALWAYS_EXCEPT+=" 28 " +ALWAYS_EXCEPT+=" 28" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! -# skip tests for PPC until they are fixed -if [[ $(uname -m) = ppc64 ]]; then - # bug number: LU-11597 LU-11787 - ALWAYS_EXCEPT+=" 16a 71a" -fi - if [ $mds1_FSTYPE = "zfs" ]; then + # bug number: LU-15757 (test_102() causes crash in umount later) + ALWAYS_EXCEPT+=" 102" # LU-2829 / LU-2887 - make allowances for ZFS slowness TEST33_NFILES=${TEST33_NFILES:-1000} fi @@ -58,6 +54,34 @@ dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=1 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS +test_0() { + local client2=${CLIENT2:-$HOSTNAME} + local tmp=$(mktemp) + + printf 'a b\n' > $tmp + + do_node_vp "$HOSTNAME" printf 'a b\n' | + diff $tmp - || error "do_node_vp mismatch" + + do_node_vp "$client2" printf 'a b\n' | + diff $tmp - || error "do_node_vp mismatch" + + do_facet_vp mds1 printf 'a b\n' | + diff $tmp - || error "do_facet_vp mismatch" + + printf '%s' 1 2 3 4 5 6 7 8 \ 9 10 ' ' '"' "'" \! \' \( \) $'\n' > $tmp + + do_node_vp "$HOSTNAME" printf '%s' 1 2 3 4 5 6 7 8 \ 9 10 ' ' '"' "'" \! \' \( \) $'\n' | + diff $tmp - || error "do_node_vp mismatch" + + do_node_vp "$client2" printf '%s' 1 2 3 4 5 6 7 8 \ 9 10 ' ' '"' "'" \! \' \( \) $'\n' | + diff $tmp - || error "do_node_vp mismatch" + + do_facet_vp mds1 printf '%s' 1 2 3 4 5 6 7 8 \ 9 10 ' ' '"' "'" \! \' \( \) $'\n' | + diff $tmp - || error "do_facet_vp mismatch" +} +run_test 0 "do_node_vp() and do_facet_vp() do the right thing" + test_1() { touch $DIR1/$tfile [ -f $DIR2/$tfile ] || error "Check create" @@ -283,7 +307,7 @@ test_11() { run_test 11 "execution of file opened for write should return error ====" test_12() { - DIR=$DIR DIR2=$DIR2 sh lockorder.sh + DIR=$DIR DIR2=$DIR2 bash lockorder.sh } run_test 12 "test lock ordering (link, stat, unlink)" @@ -361,7 +385,7 @@ run_test 14d "chmod of executing file is still possible ========" test_15() { # bug 974 - ENOSPC echo "PATH=$PATH" - sh oos2.sh $MOUNT1 $MOUNT2 + bash oos2.sh $MOUNT1 $MOUNT2 wait_delete_completed grant_error=$(dmesg | grep "< tot_grant") [ -z "$grant_error" ] || error "$grant_error" @@ -402,9 +426,9 @@ test_16a() { error "fsx failed" rm -f $file1 - # O_DIRECT reads and writes must be aligned to the device block size. - $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 -Z -r 4096 \ - -w 4096 $file1 $file2 || error "fsx with O_DIRECT failed." + # O_DIRECT reads and writes must be aligned to the PAGE_SIZE. + $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 -Z -r $PAGE_SIZE \ + -w $PAGE_SIZE $file1 $file2 || error "fsx with O_DIRECT failed." } run_test 16a "$FSXNUM iterations of dual-mount fsx" @@ -486,7 +510,7 @@ test_16d() { $LFS setstripe -c -1 $file1 # b=10919 $LCTL set_param ldlm.namespaces.*.lru_size=clear - + # direct write on one client and direct read from another dd if=/dev/urandom of=$file1 bs=1M count=100 oflag=direct dd if=$file2 of=$tmpfile iflag=direct bs=1M @@ -533,13 +557,315 @@ test_16e() { # LU-13227 } run_test 16e "Verify size consistency for O_DIRECT write" +test_16f() { # LU-14541 + local file1=$DIR1/$tfile + local file2=$DIR2/$tfile + local duration=20 + local status + + (( $MDS1_VERSION > $(version_code 2.15.51) )) || + skip "Need MDS version at least 2.15.51" + + timeout --preserve-status --signal=USR1 $duration \ + rw_seq_cst_vs_drop_caches $file1 $file2 + status=$? + + case $((status & 0x7f)) in + 0) + echo OK # Computers must be fast now. + ;; + 6) # SIGABRT + error "sequential consistency violation detected" + ;; + 10) # SIGUSR1 + echo TIMEOUT # This is fine. + ;; + *) + error "strange status '$status'" + ;; + esac + + rm -f $file1 +} +run_test 16f "rw sequential consistency vs drop_caches" + +test_16g() { + local file1=$DIR1/$tfile + local file2=$DIR2/$tfile + local duration=20 + local status + + timeout --preserve-status --signal=USR1 $duration \ + rw_seq_cst_vs_drop_caches -m $file1 $file2 + status=$? + + case $((status & 0x7f)) in + 0) + echo OK # Computers must be fast now. + ;; + 6) # SIGABRT + error "sequential consistency violation detected" + ;; + 10) # SIGUSR1 + echo TIMEOUT # This is fine. + ;; + *) + error "strange status '$status'" + ;; + esac + + rm -f $file1 +} +run_test 16g "mmap rw sequential consistency vs drop_caches" + +test_16h() { + local tf=$DIR/$tdir/$tfile + local tf2=$DIR2/$tdir/$tfile + local cmd="$MMAP_CAT $tf | od -x | tail -q -n4" + local cmd2="$MMAP_CAT $tf2 | od -x | tail -q -n4" + + test_mkdir $DIR/$tdir + + # create file and populate data + cp /etc/passwd $tf || error "cp failed" + + local size=$(stat -c %s $tf) + c1=$(eval $cmd) + c2=$(eval $cmd2) + if [[ "$c1" != "$c2" ]]; then + echo " ------- mount 1 read --------" + echo $c1 + echo " ------- mount 2 read --------" + echo $c2 + error "content mismatch" + fi + + echo " ------- before truncate --------" + echo $c1 + + # truncate file + $TRUNCATE $tf $((size / 2)) || error "truncate file" + + #cancel_lru_locks + echo " ------- after truncate --------" + + # repeat the comparison + c1=$(eval $cmd) + c2=$(eval $cmd2) + if [[ "$c1" != "$c2" ]]; then + echo " ------- mount 1 read --------" + echo $c1 + echo " ------- mount 2 read --------" + echo $c2 + error "content mismatch after truncate" + fi + echo $c2 +} +run_test 16h "mmap read after truncate file" + +test_16i() { + local tf=$DIR/$tdir/$tfile + local tf2=$DIR2/$tdir/$tfile + + test_mkdir $DIR/$tdir + + # create file and populate data + cp /etc/passwd $tf || error "cp failed" + + local size=$(stat -c %s $tf) + + c1=$(dd if=$tf bs=1 2>/dev/null | od -x | tail -q -n4) + c2=$(dd if=$tf2 bs=1 2>/dev/null | od -x | tail -q -n4) + + if [[ "$c1" != "$c2" ]]; then + echo " ------- mount 1 read --------" + echo $c1 + echo " ------- mount 2 read --------" + echo $c2 + error "content mismatch" + fi + + echo " ------- before truncate --------" + echo $c1 + + # truncate file + $TRUNCATE $tf $((size / 2)) || error "truncate file" + + echo " ------- after truncate --------" + + # repeat the comparison + c1=$(dd if=$tf bs=1 2>/dev/null | od -x | tail -q -n4) + c2=$(dd if=$tf2 bs=1 2>/dev/null | od -x | tail -q -n4) + + if [[ "$c1" != "$c2" ]]; then + echo " ------- mount 1 read --------" + echo $c1 + echo " ------- mount 2 read --------" + echo $c2 + error "content mismatch after truncate" + fi + echo $c2 +} +run_test 16i "read after truncate file" + +test_16j() +{ + (( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs" + + local stripe_size=$((1024 * 1024)) #1 MiB + # Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os + local file_size=$((25 * stripe_size)) + local bsizes + + $LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1 + stack_trap "rm -f $DIR/$tfile.1" + + # Just a bit bigger than the largest size in the test set below + dd if=/dev/urandom bs=$file_size count=1 of=$DIR/$tfile.1 || + error "buffered i/o to create file failed" + + if zfs_or_rotational; then + # DIO on ZFS can take up to 2 seconds per IO + # rotational is better, but still slow. + # Limit testing on those media to larger sizes + bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \ + $((stripe_size + 1024))" + else + bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \ + $((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \ + $((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4)) \ + $((stripe_size - 1)) $stripe_size \ + $((stripe_size + 1)) $((stripe_size * 3/2)) \ + $((stripe_size * 4)) $((stripe_size * 4 + 1))" + fi + + # 1 process (BIO or DIO) on each client + for bs in $bsizes; do + $LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2 + echo "bs: $bs, file_size $file_size" + dd if=$DIR/$tfile.1 bs=$((bs *2 )) of=$DIR/tfile.2 \ + conv=notrunc oflag=direct iflag=direct & + pid_dio1=$! + # Buffered I/O with similar but not the same block size + dd if=$DIR2/$tfile.1 bs=$((bs * 2)) of=$DIR2/$tfile.2 \ + conv=notrunc & + pid_bio2=$! + wait $pid_dio1 + rc1=$? + wait $pid_bio2 + rc2=$? + if (( rc1 != 0 )); then + error "dio copy 1 w/bsize $bs failed: $rc1" + fi + if (( rc2 != 0 )); then + error "buffered copy 2 w/bsize $bs failed: $rc2" + fi + + $CHECKSTAT -t file -s $file_size $DIR/$tfile.2 || + error "size incorrect" + $CHECKSTAT -t file -s $file_size $DIR2/$tfile.2 || + error "size incorrect - mount 2" + cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 || + error "files differ, bsize $bs" + cmp --verbose $DIR2/$tfile.1 $DIR2/$tfile.2 || + error "files differ, bsize $bs - mount 2" + rm -f $DIR/$tfile.2 + done + + # 2 processes - both DIO and BIO - on each client + for bs in $bsizes; do + $LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2 + echo "bs: $bs, file_size $file_size" + # Client 1 + dd if=$DIR/$tfile.1 bs=$((bs *2 )) of=$DIR/tfile.2 \ + conv=notrunc oflag=direct iflag=direct & + pid_dio1=$! + # Buffered I/O with similar but not the same block size + dd if=$DIR/$tfile.1 bs=$((bs * 2)) of=$DIR/$tfile.2 \ + conv=notrunc & + pid_bio2=$! + # Client 2 + dd if=$DIR2/$tfile.1 bs=$((bs *2 )) of=$DIR2/tfile.2 \ + conv=notrunc oflag=direct iflag=direct & + pid_dio3=$! + # Buffered I/O with similar but not the same block size + dd if=$DIR2/$tfile.1 bs=$((bs * 2)) of=$DIR2/$tfile.2 \ + conv=notrunc & + pid_bio4=$! + wait $pid_dio1 + rc1=$? + wait $pid_bio2 + rc2=$? + wait $pid_dio3 + rc3=$? + wait $pid_bio4 + rc4=$? + + if (( rc1 != 0 )); then + error "dio copy 1 w/bsize $bs failed: $rc1" + fi + if (( rc2 != 0 )); then + error "buffered copy 2 w/bsize $bs failed: $rc2" + fi + if (( rc3 != 0 )); then + error "dio copy 3 w/bsize $bs failed: $rc1" + fi + if (( rc4 != 0 )); then + error "buffered copy 4 w/bsize $bs failed: $rc2" + fi + + $CHECKSTAT -t file -s $file_size $DIR/$tfile.2 || + error "size incorrect" + $CHECKSTAT -t file -s $file_size $DIR2/$tfile.2 || + error "size incorrect - mount 2" + cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 || + error "files differ, bsize $bs" + cmp --verbose $DIR2/$tfile.1 $DIR2/$tfile.2 || + error "files differ, bsize $bs - mount 2" + rm -f $DIR/$tfile.2 + done +} +run_test 16j "race dio with buffered i/o" + +test_16k() { + local fsxN=${FSX_NP:-5} + local fsxNops=${FSX_NOPS:-10000} + local fsxNparams=${FSXPARAMS_16k:-""} + local dropsleep=${DROP_SLEEP:-3} + local dpipd + local -a fsxpids + local cmd + + [ "$SLOW" = "no" ] && fsxNops=1000 + + while true; do + echo 3 > /proc/sys/vm/drop_caches + sleep $dropsleep + done & + dpipd=$! + stack_trap "kill -9 $dpipd" + + for ((i = 1; i <= fsxN; i++)); do + cmd="$FSX $fsxNparams -N $fsxNops $DIR/fsxfile.${i} -l $((64 * 1024 * 1024))" + echo "+ $cmd" + eval $cmd & + fsxpids[$i]=$! + done + for ((i = 1; i <= fsxN; i++)); do + wait ${fsxpids[$i]} && echo "+ fsx $i done: rc=$?" || + error "- fsx $i FAILURE! rc=$?" + date + done +} +run_test 16k "Parallel FSX and drop caches should not panic" + test_17() { # bug 3513, 3667 remote_ost_nodsh && skip "remote OST with nodsh" && return lfs setstripe $DIR1/$tfile -i 0 -c 1 cp $SAMPLE_FILE $DIR1/$tfile cancel_lru_locks osc > /dev/null - #define OBD_FAIL_ONCE|OBD_FAIL_LDLM_CREATE_RESOURCE 0x30a + #define CFS_FAIL_ONCE|OBD_FAIL_LDLM_CREATE_RESOURCE 0x30a do_facet ost1 lctl set_param fail_loc=0x8000030a ls -ls $DIR1/$tfile | awk '{ print $1,$6 }' > $DIR1/$tfile-1 & \ ls -ls $DIR2/$tfile | awk '{ print $1,$6 }' > $DIR2/$tfile-2 @@ -760,6 +1086,31 @@ test_26b() { } run_test 26b "sync mtime between ost and mds" +test_26c() { + (( $MDS1_VERSION >= $(version_code 2.15.61) )) || + skip "Need MDS version at least 2.15.61" + + multiop_bg_pause $DIR1/$tfile O_c || error "multiop failed" + MULTIPID=$! + touch -am -d @978261179 $DIR2/$tfile + kill -USR1 $MULTIPID + wait $MULTIPID || error "wait for PID $MULTIPID failed" + sleep 1 + + echo + stat $DIR/$tfile + local times="$(stat -c"%X %Y" $DIR/$tfile)" + [[ "$times" == "978261179 978261179" ]] || + error "[am]times are not set in past on $DIR: $times" + echo + stat $DIR2/$tfile + local times="$(stat -c"%X %Y" $DIR2/$tfile)" + [[ "$times" == "978261179 978261179" ]] || + error "[am]times are not set in past on $DIR2: $times" + echo +} +run_test 26c "set-in-past on open file is not lost on close" + test_27() { cancel_lru_locks $OSC lctl clear @@ -782,47 +1133,14 @@ test_27() { run_test 27 "align non-overlapping extent locks from request ===" test_28() { # bug 9977 - ECHO_UUID="ECHO_osc1_UUID" - tOST=$($LCTL dl | awk '/-osc-|OSC.*MNT/ { print $4 }' | head -n1) - - $LFS setstripe $DIR1/$tfile -S 1048576 -i 0 -c 2 - tOBJID=`$LFS getstripe $DIR1/$tfile | awk '$1 == 1 {print $2}'` - dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 - - $LCTL <<-EOF - newdev - attach echo_client ECHO_osc1 $ECHO_UUID - setup $tOST - EOF - - tECHOID=`$LCTL dl | grep $ECHO_UUID | awk '{ print $1 }'` - $LCTL --device $tECHOID destroy "${tOBJID}:0" - - $LCTL <<-EOF - cfg_device ECHO_osc1 - cleanup - detach - EOF - - # reading of 1st stripe should pass - dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error "dd failed" - # reading of 2nd stripe should fail (this stripe was destroyed) - dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 && error - - # now, recreating test file - dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 || error "dd failed" - # reading of 1st stripe should pass - dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error "dd failed" - # reading of 2nd stripe should pass - dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 || - error "dd failed" + skip "echo_client on osc is no longer supported" } run_test 28 "read/write/truncate file with lost stripes" test_30() { #b=11110, LU-2523 test_mkdir $DIR1/$tdir cp -f /bin/bash $DIR1/$tdir/bash - /bin/sh -c 'sleep 1; rm -f $DIR2/$tdir/bash; cp /bin/bash $DIR2/$tdir' & + bash -c 'sleep 1; rm -f $DIR2/$tdir/bash; cp /bin/bash $DIR2/$tdir' & $DIR1/$tdir/bash -c 'sleep 2; openfile -f O_RDONLY /proc/$$/exe >& /dev/null; echo $?' wait @@ -866,52 +1184,29 @@ test_31b() { } run_test 31b "voluntary OST cancel / blocking ast race==============" -# enable/disable lockless truncate feature, depending on the arg 0/1 -enable_lockless_truncate() { - lctl set_param -n $OSC.*.lockless_truncate $1 -} +#LU-14949 - multi-client version of the test 31r in sanity. +test_31r() { + touch $DIR/$tfile.target + touch $DIR/$tfile.source -test_32a() { # bug 11270 - local save="$TMP/$TESTSUITE-$TESTNAME.parameters" - local stripe_size=$(do_facet $SINGLEMDS \ - "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize") + ls -l $DIR/$tfile.target # cache it for sure - save_lustre_params client "$OSC.*.lockless_truncate" > $save - # restore lockless_truncate default values on exit - stack_trap "restore_lustre_params < $save; rm -f $save" EXIT - cancel_lru_locks $OSC - enable_lockless_truncate 1 - rm -f $DIR1/$tfile - lfs setstripe -c -1 $DIR1/$tfile - dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$stripe_size > \ - /dev/null 2>&1 - clear_stats $OSC.*.${OSC}_stats + #OBD_FAIL_LLITE_OPEN_DELAY 0x1419 + $LCTL set_param fail_loc=0x1419 fail_val=3 + cat $DIR/$tfile.target & + CATPID=$! - log "checking cached lockless truncate" - $TRUNCATE $DIR1/$tfile 8000000 - $CHECKSTAT -s 8000000 $DIR2/$tfile || - error "cached truncate - wrong file size" - [ $(calc_stats $OSC.*.${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 "not cached truncate - wrong file size" - [ $(calc_stats $OSC.*.${OSC}_stats lockless_truncate) -ne 0 ] || - error "not cached truncate isn't lockless" - - log "disabled lockless truncate" - enable_lockless_truncate 0 - clear_stats $OSC.*.${OSC}_stats - $TRUNCATE $DIR2/$tfile 3000000 - $CHECKSTAT -s 3000000 $DIR1/$tfile || - error "lockless truncate disabled - wrong file size" - [ $(calc_stats $OSC.*.${OSC}_stats lockless_truncate) -eq 0 ] || - error "lockless truncate disabling failed" - rm -f $DIR1/$tfile + # Guarantee open is waiting before we get here + sleep 1 + mv $DIR2/$tfile.source $DIR2/$tfile.target + + wait $CATPID + RC=$? + if [[ $RC -ne 0 ]]; then + error "open with cat failed, rc=$RC" + fi } -run_test 32a "lockless truncate" +run_test 31r "open-rename(replace) race" test_32b() { # bug 11270 remote_ost_nodsh && skip "remote OST with nodsh" && return @@ -964,32 +1259,33 @@ test_32b() { # bug 11270 restore_lustre_params <$p rm -f $p } -run_test 32b "lockless i/o" +# Disable test 32b prior to full removal +#run_test 32b "lockless i/o" print_jbd_stat () { - local dev - local mdts=$(get_facets MDS) - local varcvs - local mds - - local stat=0 - for mds in ${mdts//,/ }; do - varsvc=${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 -n1") - val=${val%% *}; - stat=$(( stat + val)) - done - echo $stat + local mdts=$(get_facets MDS) + local stat=0 + local varsvc + local dev + local mds + + for mds in ${mdts//,/ }; do + varsvc=${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 -n1") + val=${val%% *}; + stat=$((stat + val)) + done + echo $stat } # commit on sharing tests test_33a() { remote_mds_nodsh && skip "remote MDS with nodsh" && return - [ -z "$CLIENTS" ] && skip "Need two or more clients, have $CLIENTS" [ $CLIENTCOUNT -lt 2 ] && skip "Need two or more clients, have $CLIENTCOUNT" @@ -1040,7 +1336,6 @@ run_test 33a "commit on sharing, cross crete/delete, 2 clients, benchmark" test_33b() { remote_mds_nodsh && skip "remote MDS with nodsh" && return - [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; } [ $CLIENTCOUNT -ge 2 ] || { skip "Need two or more clients, have $CLIENTCOUNT" && return 0; } @@ -1096,45 +1391,94 @@ test_33b() { } run_test 33b "COS: cross create/delete, 2 clients, benchmark under remote dir" +# arg1 is description, arg2 is operations before Sync-on-Lock-Cancel, arg3 is +# the operation that triggers SoLC +op_trigger_solc() { + local sync_count + local total=0 + local nodes=$(comma_list $(mdts_nodes)) + + sync_all_data + + # trigger CoS twice in case transaction commit before unlock + for i in 1 2; do + bash -c "$2" + do_nodes $nodes "$LCTL set_param -n mdt.*.sync_count=0" + bash -c "$3" + sync_count=$(do_nodes $nodes \ + "lctl get_param -n mdt.*MDT*.sync_count" | calc_sum) + total=$((total + sync_count)); + rm -rf $DIR/$tdir/* + sync_all_data + done + + echo $1 + echo " $2" + echo " $3" + echo " SoLC count $total" + (( total > 0 )) || error "$3 didn't trigger SoLC" +} + +test_33_run() { + echo $1 + echo " $2" + eval $2 +} + test_33c() { - [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" - [ "$MDS1_VERSION" -lt $(version_code 2.7.63) ] && + (( MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs" + (( MDS1_VERSION >= $(version_code 2.7.63) )) || skip "DNE CoS not supported" # LU-13522 stop mds1 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "start mds1 failed" - local sync_count - - mkdir $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir sync_all_data - do_facet mds1 "lctl set_param -n mdt.*.sync_count=0" - # do twice in case transaction is committed before unlock, see LU-8200 - for i in 1 2; do - # remote dir is created on MDT1, which enqueued lock of $tdir on - # MDT0 - $LFS mkdir -i 1 $DIR/$tdir/remote.$i - mkdir $DIR/$tdir/local.$i - done - sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count") - echo "sync_count $sync_count" - [ $sync_count -eq 0 ] && error "Sync-Lock-Cancel not triggered" + if (( MDS1_VERSION < $(version_code 2.15.55.204) )); then + op_trigger_solc "create remote dir and local dir" \ + "$LFS mkdir -i 1 $DIR/$tdir/remote" \ + "$LFS mkdir -i 0 $DIR/$tdir/local" + (( MDSCOUNT > 2 )) && + op_trigger_solc "create remote dirs on different MDTs" \ + "$LFS mkdir -i 1 $DIR/$tdir/remote.1" \ + "$LFS mkdir -i 2 $DIR/$tdir/remote.2" + fi + op_trigger_solc "create file on 2nd stripe under striped directory" \ + "$LFS mkdir -i 0 -c 2 $DIR/$tdir/striped" \ + "touch $DIR2/$tdir/striped/subfile" + + echo + echo "Below operations shouldn't trigger Solc:" + $LFS mkdir -i 0 -c 2 $DIR/$tdir/striped sync_all_data do_facet mds1 "lctl set_param -n mdt.*.sync_count=0" - $LFS mkdir -i 1 $DIR/$tdir/remote.3 - # during sleep remote mkdir should have been committed and canceled - # remote lock spontaneously, which shouldn't trigger sync - sleep 6 - mkdir $DIR/$tdir/local.3 + if (( MDS1_VERSION >= $(version_code 2.15.55.204) )); then + test_33_run "create file on 2nd stripe after setattr" \ + "chmod 777 $DIR/$tdir/striped; \ + touch $DIR2/$tdir/striped/subfile" + test_33_run "create remote dir and local dir" \ + "$LFS mkdir -i 1 $DIR/$tdir/remote" \ + "$LFS mkdir -i 0 $DIR/$tdir/local" + (( MDSCOUNT > 2 )) && + test_33_run "create remote dirs on different MDTs" \ + "$LFS mkdir -i 1 $DIR/$tdir/remote.1" \ + "$LFS mkdir -i 2 $DIR/$tdir/remote.2" + fi + test_33_run "create local dir after remote dir creation transaction commit" \ + "$LFS mkdir -i 1 $DIR/$tdir/remote.3; \ + do_facet mds2 $LCTL set_param -n osd*.*MDT0001.force_sync 1; + mkdir $DIR/$tdir/local.3" sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count") - echo "sync_count $sync_count" + echo "Solc count $sync_count" [ $sync_count -eq 0 ] || error "Sync-Lock-Cancel triggered" } -run_test 33c "Cancel cross-MDT lock should trigger Sync-Lock-Cancel" +run_test 33c "Cancel cross-MDT lock should trigger Sync-on-Lock-Cancel" -# arg1 is operations done before CoS, arg2 is the operation that triggers CoS +# arg1 is description, arg2 is operations done before CoS, arg3 is the operation +# that triggers CoS op_trigger_cos() { local commit_nr local total=0 @@ -1144,9 +1488,9 @@ op_trigger_cos() { # trigger CoS twice in case transaction commit before unlock for i in 1 2; do - sh -c "$1" + bash -c "$2" do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0" - sh -c "$2" + bash -c "$3" commit_nr=$(do_nodes $nodes \ "lctl get_param -n mdt.*.async_commit_count" | calc_sum) total=$((total + commit_nr)); @@ -1154,75 +1498,101 @@ op_trigger_cos() { sync_all_data done - echo "CoS count $total" - [ $total -gt 0 ] || error "$2 didn't trigger CoS" + echo $1 + echo " $2" + echo " $3" + echo " CoS count $total" + (( total > 0 )) || error "$3 didn't trigger CoS" } test_33d() { - [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" - [ "$MDS1_VERSION" -lt $(version_code 2.7.63) ] && + (( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs" + (( MDS1_VERSION >= $(version_code 2.7.63) )) || skip "DNE CoS not supported" - # remote directory create - op_trigger_cos "mkdir $DIR/$tdir" "$LFS mkdir -i 1 $DIR/$tdir/subdir" - # remote directory unlink - op_trigger_cos "$LFS mkdir -i 1 $DIR/$tdir" "rmdir $DIR/$tdir" - # striped directory create - op_trigger_cos "mkdir $DIR/$tdir" "$LFS mkdir -c 2 $DIR/$tdir/subdir" - # striped directory setattr - op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; touch $DIR/$tdir" \ - "chmod 713 $DIR/$tdir" - # striped directory unlink - op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; touch $DIR/$tdir" \ - "rmdir $DIR/$tdir" - # cross-MDT link - op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; \ + if (( $MDS1_VERSION < $(version_code 2.15.55.133) )); then + op_trigger_cos "remote directory unlink" \ + "$LFS mkdir -i 1 $DIR/$tdir" "rmdir $DIR2/$tdir" + op_trigger_cos "striped directory create" "mkdir $DIR/$tdir" \ + "$LFS mkdir -c 2 $DIR2/$tdir/subdir" + op_trigger_cos "striped directory setattr" \ + "$LFS mkdir -c 2 $DIR/$tdir" "chmod 713 $DIR2/$tdir" + op_trigger_cos "striped directory unlink" \ + "$LFS mkdir -c 2 $DIR/$tdir" "rmdir $DIR2/$tdir" + op_trigger_cos "cross-MDT link" \ + "mkdir $DIR/$tdir; \ $LFS mkdir -i 0 $DIR/$tdir/d1; \ $LFS mkdir -i 1 $DIR/$tdir/d2; \ touch $DIR/$tdir/d1/tgt" \ - "ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src" - # cross-MDT rename - op_trigger_cos "$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" \ - "mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt" - # migrate - op_trigger_cos "$LFS mkdir -i 0 $DIR/$tdir" \ - "$LFS migrate -m 1 $DIR/$tdir" + "ln $DIR2/$tdir/d1/tgt $DIR2/$tdir/d2/src" + fi + + op_trigger_cos "remote directory create" "$LFS mkdir -i 0 $DIR/$tdir" \ + "$LFS mkdir -i 1 $DIR2/$tdir/subdir" + op_trigger_cos "cross-MDT rename" \ + "mkdir $DIR/$tdir; \ + $LFS mkdir -i 0 $DIR/$tdir/d1; \ + $LFS mkdir -i 1 $DIR/$tdir/d2; \ + touch $DIR/$tdir/d1/src" \ + "mv $DIR2/$tdir/d1/src $DIR2/$tdir/d2/tgt" + op_trigger_cos "migrate" \ + "$LFS mkdir -i 0 $DIR/$tdir" \ + "$LFS migrate -m 1 $DIR2/$tdir" return 0 } -run_test 33d "DNE distributed operation should trigger COS" +run_test 33d "dependent transactions should trigger COS" test_33e() { - [ -n "$CLIENTS" ] || skip "Need two or more clients" - [ $CLIENTCOUNT -ge 2 ] || - skip "Need two or more clients, have $CLIENTCOUNT" - [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" - [ "$MDS1_VERSION" -lt $(version_code 2.7.63) ] && + (( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs" + (( MDS1_VERSION >= $(version_code 2.7.63) )) || skip "DNE CoS not supported" - local client2=${CLIENT2:-$(hostname)} - - sync + $LFS mkdir -i 0 $DIR/$tdir + $LFS mkdir -i 0 $DIR/$tdir/d1 + $LFS mkdir -i 1 $DIR/$tdir/d2 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 + test_33_run "plain dir creation" "mkdir $DIR2/$tdir/plain" + test_33_run "open file and write" "echo abc > $DIR2/$tdir/$tfile" + test_33_run "append write" "echo dfg >> $DIR2/$tdir/$tfile" + test_33_run "setattr" "touch $DIR2/$tdir/$tfile" + test_33_run "file unlink" "rm $DIR2/$tdir/$tfile" + test_33_run "plain dir unlink" "rmdir $DIR2/$tdir/plain" + if (( MDS1_VERSION >= $(version_code 2.15.55.133) )); then + test_33_run "striped directory creation" \ + "$LFS mkdir -i 0 -c 2 $DIR2/$tdir/striped" + test_33_run "set default LMV to create striped subdir" \ + "$LFS setdirstripe -D -c 2 $DIR/$tdir" + test_33_run "striped subdir creation" \ + "createmany -d $DIR/$tdir/subdir 100" + test_33_run "sub file creation and write" \ + "createmany -o $DIR/$tdir/subfile 100; \ + echo abc > $DIR/$tdir/subfile1" + test_33_run "sub file append write" \ + "echo dfg >> $DIR2/$tdir/subfile2" + test_33_run "subdir setatttr" "touch $DIR2/$tdir/subdir1" + test_33_run "subdir unlink" \ + "unlinkmany -d $DIR/$tdir/subdir 100" + test_33_run "sub file unlink" \ + "unlinkmany $DIR2/$tdir/subfile 100" + test_33_run "sub file creation follows striped dir chmod" \ + "chmod 777 $DIR/$tdir/striped; \ + touch $DIR/$tdir/striped/subfile" + test_33_run "striped directory unlink" \ + "rm -rf $DIR2/$tdir/striped" + fi + + test_33_run "directory unlink" "rm -rf $DIR2/$tdir" 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 + echo "CoS count $async_commit_count" + (( async_commit_count == 0 )) || error "CoS triggerred" } -run_test 33e "DNE local operation shouldn't trigger COS" +run_test 33e "independent transactions shouldn't trigger COS" # End commit on sharing tests @@ -1452,7 +1822,7 @@ test_39a() { if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi done } -run_test 39a "test from 11063 ==================================" +run_test 39a "file mtime does not change after rename" test_39b() { local client1=${CLIENT1:-`hostname`} @@ -1479,7 +1849,7 @@ test_39b() { if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi done } -run_test 39b "11063 problem 1 ==================================" +run_test 39b "file mtime the same on clients with/out lock" test_39c() { local client1=${CLIENT1:-`hostname`} @@ -1575,25 +1945,26 @@ check_pdo_conflict() { # test 40: check non-blocking operations test_40a() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + + mkdir_on_mdt0 $DIR2/$tdir pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" - touch $DIR2 - mkdir $DIR1/$tfile & + mkdir $DIR1/$tdir/$tfile & PID1=$!; pdo_sched - touch $DIR2/$tfile-2 + touch $DIR2/$tdir/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" - mkdir $DIR2/$tfile-3 + mkdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "mkdir is blocked" - link $DIR2/$tfile-2 $DIR2/$tfile-4 + link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4 check_pdo_conflict $PID1 || error "link is blocked" - mv $DIR2/$tfile-2 $DIR2/$tfile-5 + mv $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-5 check_pdo_conflict $PID1 || error "rename is blocked" - stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null + stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null check_pdo_conflict $PID1 || error "getattr is blocked" - rm $DIR2/$tfile-4 $DIR2/$tfile-5 - rmdir $DIR2/$tfile-3 + rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-5 + rmdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" # all operations above shouldn't wait the first one @@ -1601,32 +1972,34 @@ test_40a() { do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0 2>/dev/null || true" wait $PID1 - rm -rf $DIR/$tfile* + rm -rf $DIR/$tdir return 0 } run_test 40a "pdirops: create vs others ==============" test_40b() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + + mkdir_on_mdt0 $DIR2/$tdir pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" - touch $DIR1/$tfile & + touch $DIR1/$tdir/$tfile & PID1=$!; pdo_sched # open|create - touch $DIR2/$tfile-2 + touch $DIR2/$tdir/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" - mkdir $DIR2/$tfile-3 + mkdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "mkdir is blocked" - link $DIR2/$tfile-2 $DIR2/$tfile-4 + link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4 check_pdo_conflict $PID1 || error "link is blocked" - mv $DIR2/$tfile-2 $DIR2/$tfile-5 + mv $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-5 check_pdo_conflict $PID1 || error "rename is blocked" - stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null + stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null check_pdo_conflict $PID1 || error "getattr is blocked" - rm $DIR2/$tfile-4 $DIR2/$tfile-5 - rmdir $DIR2/$tfile-3 + rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-5 + rmdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" # all operations above shouldn't wait the first one @@ -1634,33 +2007,35 @@ test_40b() { do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0 2>/dev/null || true" wait $PID1 - rm -rf $DIR/$tfile* + rm -rf $DIR/$tdir return 0 } run_test 40b "pdirops: open|create and others ==============" test_40c() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + + mkdir_on_mdt0 $DIR2/$tdir pdo_lru_clear - touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 + touch $DIR1/$tdir/$tfile +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" - link $DIR1/$tfile $DIR1/$tfile-0 & + link $DIR1/$tdir/$tfile $DIR1/$tdir/$tfile-0 & PID1=$!; pdo_sched # open|create - touch $DIR2/$tfile-2 + touch $DIR2/$tdir/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" - mkdir $DIR2/$tfile-3 + mkdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "mkdir is blocked" - link $DIR2/$tfile-2 $DIR2/$tfile-4 + link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4 check_pdo_conflict $PID1 || error "link is blocked" - mv $DIR2/$tfile-2 $DIR2/$tfile-5 + mv $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-5 check_pdo_conflict $PID1 || error "rename is blocked" - stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null + stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null check_pdo_conflict $PID1 || error "getattr is blocked" - rm $DIR2/$tfile-4 $DIR2/$tfile-5 - rmdir $DIR2/$tfile-3 + rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-5 + rmdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" # all operations above shouldn't wait the first one @@ -1668,33 +2043,35 @@ test_40c() { do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0 2>/dev/null || true" wait $PID1 - rm -rf $DIR/$tfile* + rm -rf $DIR/$tdir return 0 } run_test 40c "pdirops: link and others ==============" test_40d() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + + mkdir_on_mdt0 $DIR2/$tdir pdo_lru_clear - touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 + touch $DIR1/$tdir/$tfile +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" - rm $DIR1/$tfile & + rm $DIR1/$tdir/$tfile & PID1=$!; pdo_sched # open|create - touch $DIR2/$tfile-2 + touch $DIR2/$tdir/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" - mkdir $DIR2/$tfile-3 + mkdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "mkdir is blocked" - link $DIR2/$tfile-2 $DIR2/$tfile-4 + link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4 check_pdo_conflict $PID1 || error "link is blocked" - mv $DIR2/$tfile-2 $DIR2/$tfile-5 + mv $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-5 check_pdo_conflict $PID1 || error "rename is blocked" - stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null + stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null check_pdo_conflict $PID1 || error "getattr is blocked" - rm $DIR2/$tfile-4 $DIR2/$tfile-5 - rmdir $DIR2/$tfile-3 + rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-5 + rmdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" # all operations above shouldn't wait the first one @@ -1708,24 +2085,26 @@ run_test 40d "pdirops: unlink and others ==============" test_40e() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + + mkdir_on_mdt0 $DIR2/$tdir pdo_lru_clear - touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 + touch $DIR1/$tdir/$tfile +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" - mv $DIR1/$tfile $DIR1/$tfile-0 & + mv $DIR1/$tdir/$tfile $DIR1/$tdir/$tfile-0 & PID1=$!; pdo_sched # open|create - touch $DIR2/$tfile-2 + touch $DIR2/$tdir/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" - mkdir $DIR2/$tfile-3 + mkdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "mkdir is blocked" - link $DIR2/$tfile-2 $DIR2/$tfile-4 + link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4 check_pdo_conflict $PID1 || error "link is blocked" - stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null + stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null check_pdo_conflict $PID1 || error "getattr is blocked" - rm $DIR2/$tfile-4 $DIR2/$tfile-2 - rmdir $DIR2/$tfile-3 + rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-2 + rmdir $DIR2/$tdir/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" # all operations above shouldn't wait the first one @@ -1733,7 +2112,7 @@ test_40e() { do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0 2>/dev/null || true" wait $PID1 - rm -rf $DIR/$tfile* + rm -rf $DIR/$tdir return 0 } run_test 40e "pdirops: rename and others ==============" @@ -1741,7 +2120,7 @@ run_test 40e "pdirops: rename and others ==============" # test 41: create blocking operations test_41a() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & @@ -1759,7 +2138,7 @@ run_test 41a "pdirops: create vs mkdir ==============" test_41b() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & @@ -1778,7 +2157,7 @@ run_test 41b "pdirops: create vs create ==============" test_41c() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & @@ -1796,7 +2175,7 @@ run_test 41c "pdirops: create vs link ==============" test_41d() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & @@ -1815,7 +2194,7 @@ run_test 41d "pdirops: create vs unlink ==============" test_41e() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & @@ -1833,7 +2212,7 @@ run_test 41e "pdirops: create and rename (tgt) ==============" test_41f() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & @@ -1851,7 +2230,7 @@ run_test 41f "pdirops: create and rename (src) ==============" test_41g() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & @@ -1869,7 +2248,7 @@ run_test 41g "pdirops: create vs getattr ==============" test_41h() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & @@ -1916,12 +2295,12 @@ sub_test_41i() { } test_41i() { - [[ $MDS1_VERSION -le $(version_code 2.13.56) ]] || + (( $MDS1_VERSION >= $(version_code 2.13.56) )) || skip "Need MDS version newer than 2.13.56" local msg fail_loc -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a for fail_loc in "0x80000169" "0x8000016a"; do echo "Begin 100 tests with fail_loc=$fail_loc" printf "Progress: " @@ -1939,7 +2318,7 @@ run_test 41i "reint_open: create vs create" # test 42: unlink and blocking operations test_42a() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & @@ -1957,7 +2336,7 @@ run_test 42a "pdirops: mkdir vs mkdir ==============" test_42b() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & @@ -1976,7 +2355,7 @@ run_test 42b "pdirops: mkdir vs create ==============" test_42c() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & @@ -1994,7 +2373,7 @@ run_test 42c "pdirops: mkdir vs link ==============" test_42d() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & @@ -2013,7 +2392,7 @@ run_test 42d "pdirops: mkdir vs unlink ==============" test_42e() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & @@ -2031,7 +2410,7 @@ run_test 42e "pdirops: mkdir and rename (tgt) ==============" test_42f() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & @@ -2048,26 +2427,26 @@ test_42f() { run_test 42f "pdirops: mkdir and rename (src) ==============" test_42g() { + mkdir_on_mdt0 $DIR1/$tdir pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" - mkdir $DIR1/$tfile & + mkdir $DIR1/$tdir/$tfile & PID1=$! ; pdo_sched - stat $DIR2/$tfile > /dev/null & + stat $DIR2/$tdir/$tfile > /dev/null & PID2=$! ; pdo_sched do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0 2>/dev/null || true" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } wait $PID2 ; [ $? -eq 0 ] || error "stat must succeed" - rm -rf $DIR/$tfile* - return 0 + rm -rf $DIR/$tdir } run_test 42g "pdirops: mkdir vs getattr ==============" test_42h() { pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & @@ -2096,7 +2475,7 @@ run_test 43a "rmdir,mkdir doesn't return -EEXIST ==============" test_43b() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & @@ -2116,7 +2495,7 @@ test_43c() { pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & @@ -2135,7 +2514,7 @@ run_test 43c "pdirops: unlink vs link ==============" test_43d() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & @@ -2155,7 +2534,7 @@ test_43e() { pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & @@ -2174,7 +2553,7 @@ run_test 43e "pdirops: unlink and rename (tgt) ==============" test_43f() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & @@ -2193,7 +2572,7 @@ run_test 43f "pdirops: unlink and rename (src) ==============" test_43g() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & @@ -2212,7 +2591,7 @@ run_test 43g "pdirops: unlink vs getattr ==============" test_43h() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & @@ -2232,7 +2611,7 @@ test_43i() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & @@ -2253,19 +2632,20 @@ test_43j() { [[ $MDS1_VERSION -lt $(version_code 2.13.52) ]] && skip "Need MDS version newer than 2.13.52" + mkdir_on_mdt0 $DIR1/$tdir for i in {1..100}; do -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_CREATE_RACE 0x167 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_CREATE_RACE 0x167 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000167 2>/dev/null || true" OK=0 - mkdir $DIR1/$tdir & + mkdir $DIR1/$tdir/sub & PID1=$! - mkdir $DIR2/$tdir && ((OK++)) + mkdir $DIR2/$tdir/sub && ((OK++)) wait $PID1 && ((OK++)) (( OK == 1 )) || error "exactly one mkdir should succeed" - rmdir $DIR1/$tdir || error "rmdir failed" + rmdir $DIR1/$tdir/sub || error "rmdir failed" done return 0 } @@ -2280,6 +2660,7 @@ sub_test_43k() { # cfs_race() if LCK_PW is taken on the parent by mdt_reint_unlink. test_mkdir $DIR2/$tdir touch $DIR2/$tdir/$tfile + pdo_lru_clear do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=${fail_loc} || true" &>/dev/null @@ -2303,12 +2684,12 @@ sub_test_43k() { } test_43k() { - [[ $MDS1_VERSION -le $(version_code 2.13.56) ]] || + (( $MDS1_VERSION >= $(version_code 2.13.56) )) || skip "Need MDS version newer than 2.13.56" local msg fail_loc -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a for fail_loc in "0x80000169" "0x8000016a"; do echo "Begin 100 tests with fail_loc=$fail_loc" printf "Progress: " @@ -2331,7 +2712,7 @@ run_test 43k "unlink vs create" test_44a() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2351,7 +2732,7 @@ run_test 44a "pdirops: rename tgt vs mkdir ==============" test_44b() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2371,7 +2752,7 @@ test_44c() { pdo_lru_clear touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2390,7 +2771,7 @@ run_test 44c "pdirops: rename tgt vs link ==============" test_44d() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2411,7 +2792,7 @@ test_44e() { touch $DIR1/$tfile touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2431,7 +2812,7 @@ test_44f() { pdo_lru_clear touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2450,7 +2831,7 @@ run_test 44f "pdirops: rename tgt and rename (src) ==============" test_44g() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2469,7 +2850,7 @@ run_test 44g "pdirops: rename tgt vs getattr ==============" test_44h() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2490,7 +2871,7 @@ test_44i() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & @@ -2522,7 +2903,7 @@ run_test 45a "rename,mkdir doesn't return -EEXIST ==============" test_45b() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & @@ -2542,7 +2923,7 @@ test_45c() { pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-3 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & @@ -2561,7 +2942,7 @@ run_test 45c "pdirops: rename src vs link ==============" test_45d() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & @@ -2581,7 +2962,7 @@ test_45e() { pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-3 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & @@ -2600,7 +2981,7 @@ run_test 45e "pdirops: rename src and rename (tgt) ==============" test_45f() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & @@ -2619,7 +3000,7 @@ run_test 45f "pdirops: rename src and rename (src) ==============" test_45g() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & @@ -2638,7 +3019,7 @@ run_test 45g "pdirops: rename src vs getattr ==============" test_45h() { pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & @@ -2657,7 +3038,7 @@ test_45i() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & @@ -2684,6 +3065,7 @@ sub_test_45j() { test_mkdir $DIR2/$tdir echo file1 > $DIR2/$tdir/$tfile echo file2 > $DIR2/$tdir/$tfile-2 + pdo_lru_clear do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=${fail_loc} || true" &>/dev/null @@ -2709,12 +3091,12 @@ sub_test_45j() { } test_45j() { - [[ $MDS1_VERSION -le $(version_code 2.13.56) ]] || + (( $MDS1_VERSION >= $(version_code 2.13.56) )) || skip "Need MDS version newer than 2.13.56" local msg fail_loc -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a for fail_loc in "0x80000169" "0x8000016a"; do echo "Begin 100 tests with fail_loc=$fail_loc" printf "Progress: " @@ -2732,7 +3114,7 @@ run_test 45j "read vs rename ==============" test_46a() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2751,7 +3133,7 @@ run_test 46a "pdirops: link vs mkdir ==============" test_46b() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2770,7 +3152,7 @@ run_test 46b "pdirops: link vs create ==============" test_46c() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2789,7 +3171,7 @@ run_test 46c "pdirops: link vs link ==============" test_46d() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2809,7 +3191,7 @@ test_46e() { pdo_lru_clear touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2829,7 +3211,7 @@ test_46f() { pdo_lru_clear touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2848,7 +3230,7 @@ run_test 46f "pdirops: link and rename (src) ==============" test_46g() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2867,7 +3249,7 @@ run_test 46g "pdirops: link vs getattr ==============" test_46h() { pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2887,7 +3269,7 @@ test_46i() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & @@ -2906,7 +3288,7 @@ run_test 46i "pdirops: link vs remote mkdir" # test 47: remote mkdir and blocking operations test_47a() { -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear do_nodes $(comma_list $(mdts_nodes)) \ @@ -2925,7 +3307,7 @@ test_47a() { run_test 47a "pdirops: remote mkdir vs mkdir" test_47b() { -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear do_nodes $(comma_list $(mdts_nodes)) \ @@ -2949,7 +3331,7 @@ test_47c() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & @@ -2968,7 +3350,7 @@ run_test 47c "pdirops: remote mkdir vs link" test_47d() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & @@ -2989,7 +3371,7 @@ test_47e() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear touch $DIR1/$tfile-2 -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & @@ -3009,7 +3391,7 @@ run_test 47e "pdirops: remote mkdir and rename (tgt)" test_47f() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & @@ -3031,7 +3413,7 @@ test_47g() { sync sync_all_data pdo_lru_clear -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 +#define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_nodes $(comma_list $(mdts_nodes)) \ "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & @@ -3095,12 +3477,12 @@ test_51a() { run_test 51a "layout lock: refresh layout should work" test_51b() { - [[ "$MDS1_VERSION" -ge $(version_code 2.3.59) ]] || + (( $MDS1_VERSION >= $(version_code 2.3.59) )) || skip "Need MDS version at least 2.3.59" local tmpfile=`mktemp` - $LFS setstripe -E 1M -c 1 -E -1 --extension-size 64M $DIR1/$tfile || + $LFS setstripe -E 1m -S 1M -c 1 -E -1 -c 1 $DIR1/$tfile || error "Create $DIR1/$tfile failed" dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc || @@ -3188,6 +3570,9 @@ test_51d() { run_test 51d "layout lock: losing layout lock should clean up memory map region" test_51e() { + (( $MDS1_VERSION >= $(version_code 2.13.54.148) )) || + skip "MDS version must be at least 2.13.54.148" + local pid $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:eW_E+eUc & @@ -3277,56 +3662,59 @@ test_54() { run_test 54 "rename locking" test_55a() { - mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" + mkdir_on_mdt0 $DIR/$tdir + mkdir -p $DIR/$tdir/d1/d2 $DIR/$tdir/d3 || error "(1) mkdir failed" #define OBD_FAIL_MDS_RENAME4 0x156 do_facet mds1 $LCTL set_param fail_loc=0x80000156 - mv -T $DIR/d1/d2 $DIR/d3/d2 & + mv -T $DIR/$tdir/d1/d2 $DIR/$tdir/d3/d2 & PID1=$! sleep 1 - rm -r $DIR2/d3 + rm -r $DIR2/$tdir/d3 wait $PID1 && error "(2) mv succeeded" - rm -rf $DIR/d1 + rm -rf $DIR/$tdir } run_test 55a "rename vs unlink target dir" test_55b() { - mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" + mkdir_on_mdt0 $DIR/$tdir + mkdir -p $DIR/$tdir/d1/d2 $DIR/$tdir/d3 || error "(1) mkdir failed" #define OBD_FAIL_MDS_RENAME4 0x156 do_facet mds1 $LCTL set_param fail_loc=0x80000156 - mv -T $DIR/d1/d2 $DIR/d3/d2 & + mv -T $DIR/$tdir/d1/d2 $DIR/$tdir/d3/d2 & PID1=$! sleep 1 - rm -r $DIR2/d1 + rm -r $DIR2/$tdir/d1 wait $PID1 && error "(2) mv succeeded" - rm -rf $DIR/d3 + rm -rf $DIR/$tdir } run_test 55b "rename vs unlink source dir" test_55c() { - mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" + mkdir_on_mdt0 $DIR/$tdir + mkdir -p $DIR/$tdir/d1/d2 $DIR/$tdir/d3 || error "(1) mkdir failed" #define OBD_FAIL_MDS_RENAME4 0x156 do_facet mds1 $LCTL set_param fail_loc=0x156 - mv -T $DIR/d1/d2 $DIR/d3/d2 & + mv -T $DIR/$tdir/d1/d2 $DIR/$tdir/d3/d2 & PID1=$! sleep 1 # while rename is sleeping, open and remove d3 - $MULTIOP $DIR2/d3 D_c & + $MULTIOP $DIR2/$tdir/d3 D_c & PID2=$! sleep 1 - rm -rf $DIR2/d3 + rm -rf $DIR2/$tdir/d3 sleep 5 # while rename is sleeping 2nd time, close d3 @@ -3335,31 +3723,51 @@ test_55c() wait $PID1 && error "(2) mv succeeded" - rm -rf $DIR/d1 + rm -rf $DIR/$tdir } run_test 55c "rename vs unlink orphan target dir" test_55d() { - touch $DIR/f1 + mkdir_on_mdt0 $DIR/$tdir + + touch $DIR/$tdir/f1 #define OBD_FAIL_MDS_RENAME3 0x155 do_facet mds1 $LCTL set_param fail_loc=0x155 - mv $DIR/f1 $DIR/$tdir & + mv $DIR/$tdir/f1 $DIR/$tdir/$tdir & PID1=$! sleep 2 # while rename is sleeping, create $tdir, but as a directory - mkdir -p $DIR2/$tdir || error "(1) mkdir failed" + mkdir -p $DIR2/$tdir/$tdir || error "(1) mkdir failed" # link in reverse locking order - ln $DIR2/f1 $DIR2/$tdir/ + ln $DIR2/$tdir/f1 $DIR2/$tdir/$tdir/f1 || error "(2) ln failed" - wait $PID1 && error "(2) mv succeeded" - rm -rf $DIR/f1 + ! wait $PID1 || error "(3) mv succeeded" + rm -rf $DIR/$tdir } run_test 55d "rename file vs link" +test_56a() { + $LFS setstripe -c 1 $MOUNT/$tfile || error "creating $MOUNT/$tfile" + stack_trap "rm -f $MOUNT/$tfile" + echo "run_llverdev $MOUNT/$tfile -p -s $((16000000)) -c 4k" + run_llverdev $MOUNT/$tfile -p -s $((16000000)) -c 4k || + error "llverdev failed with rc=$?" +} +run_test 56a "test llverdev with single large stripe" + +test_56b() { + $LFS setstripe -C 2000 $MOUNT/$tfile || error "creating $MOUNT/$tfile" + stack_trap "rm -f $MOUNT/$tfile" + echo "run_llverdev $MOUNT/$tfile -p -s $((16000000 * OSTCOUNT)) -c 4k" + run_llverdev $MOUNT/$tfile -p -s $((16000000 * OSTCOUNT)) -c 4k || + error "llverdev failed with rc=$?" +} +run_test 56b "test llverdev and partial verify of wide stripe file" + test_60() { [ $MDS1_VERSION -lt $(version_code 2.3.0) ] && skip "MDS version must be >= 2.3.0" @@ -3459,28 +3867,28 @@ test_71a() { checkfiemap --test || skip "checkfiemap not runnable: $?" # write data this way: hole - data - hole - data - dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1 + dd if=/dev/urandom of=$DIR1/$tfile bs=64K seek=1 count=1 [ "$(facet_fstype ost$(($($LFS 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 + dd if=/dev/urandom of=$DIR1/$tfile bs=64K 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 || + checkfiemap $DIR2/$tfile 131072 || 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/$tfile bs=40K seek=3 count=1 + dd if=/dev/urandom of=$DIR1/$tfile bs=64K seek=1 count=1 + dd if=/dev/urandom of=$DIR1/$tfile bs=64K seek=3 count=1 stat $DIR1/$tfile local can3=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') echo $can3 - checkfiemap $DIR1/$tfile 81920 || + checkfiemap $DIR1/$tfile 131072 || error 4 local can4=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') echo $can2 @@ -3504,14 +3912,76 @@ test_71b() { mkdir -p $DIR1/$tdir $LFS setstripe -c -1 $DIR1/$tdir || error "setstripe failed" - dd if=/dev/urandom of=$DIR1/$tdir/$tfile bs=40K count=1 + dd if=/dev/urandom of=$DIR1/$tdir/$tfile bs=64K count=1 [ "$(facet_fstype ost$(($($LFS getstripe -i $DIR1/$tdir/$tfile) + 1)))" = \ "zfs" ] && skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0 - checkfiemap $DIR1/$tdir/$tfile 40960 || error "checkfiemap failed" + checkfiemap $DIR1/$tdir/$tfile 65536 || error "checkfiemap failed" } run_test 71b "check fiemap support for stripecount > 1" +_check_last_flag_with_filefrag() +{ + local file=$1 + local count=$2 + local i + local offset + + echo "check last flag for file with $count extents" + rm -f $file + for ((i=0; i<$count; i++)); do + offset=$((i * 256)) + dd if=/dev/zero of=$file bs=4K count=1 seek=$offset 2> /dev/null + done + + filefrag -s -v $file | grep "last" || + error "test file with $i extents failed" + + rm -f $file +} + +test_71c() { + local file="$DIR1/$tdir/$tfile" + + (( $CLIENT_VERSION >= $(version_code 2.15.57) )) || + skip "Need client version >= 2.15.57" + [ $(facet_fstype ost1) = "ldiskfs" ] || + skip "support only ldiskfs ost" + filefrag -V | grep wc || + skip "need whamcloud version of e2fsprogs" + + mkdir -p $DIR1/$tdir + + # filefrag uses u64[2028] buffer to fetch fiemap. The number of extents + # in the buffer is (8 * 2048 - 32) / 56 = 292. Test file with 291, 292 + # and 293 extents + _check_last_flag_with_filefrag $file 291 + _check_last_flag_with_filefrag $file 292 + _check_last_flag_with_filefrag $file 293 +} +run_test 71c "check FIEMAP_EXTENT_LAST flag with different extents number" + +test_71d() { #LU-17110 + checkfiemap --test || + skip "error $?: checkfiemap failed" + + local f=$DIR/$tfile + + # write data this way: hole - data - hole - data + dd if=/dev/urandom of=$f bs=64K count=1 + [[ "$(facet_fstype ost$(($($LFS getstripe -i $f) + 1)))" != "zfs" ]] || + skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" + dd if=/dev/urandom of=$f bs=64K seek=2 count=1 + dd if=/dev/urandom of=$f bs=64K seek=4 count=1 + dd if=/dev/urandom of=$f bs=64K seek=6 count=1 conv=fsync + echo "disk usage: $(du -B1 $f)" + echo "file size: $(du -b $f)" + + checkfiemap --corruption_test $f $((4 * 64 *1024)) || + error "checkfiemap failed" +} +run_test 71d "fiemap corruption test with fm_extent_count=0" + test_72() { local p="$TMP/sanityN-$TESTNAME.parameters" local tlink1 @@ -3558,12 +4028,15 @@ test_73() { clear_stats 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_stats llite.*.stats getxattr_hits) -eq 2 ] || + # At least 2 hits for getfattr(0)+getfattr(size) + # There may be more if auditd has a filesystem-related rule enabled + (( $(calc_stats llite.*.stats getxattr_hits) >= 2 )) || error "not cached in $DIR1" + clear_stats llite.*.stats getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed" - # 4 hits for more getfattr(0)+getfattr(size) - [ $(calc_stats llite.*.stats getxattr_hits) -eq 4 ] || + # At least 2 hits for getfattr(0)+getfattr(size) + # There may be more if auditd has a filesystem-related rule enabled + (( $(calc_stats llite.*.stats getxattr_hits) >= 2 )) || error "not cached in $DIR2" rm -f $DIR2/$tfile @@ -3692,11 +4165,21 @@ nrs_write_read() { local n=16 local dir=$DIR/$tdir local myRUNAS="$1" + local create_as="$2" mkdir $dir || error "mkdir $dir failed" $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed" chmod 777 $dir + if [[ -n "$create_as" ]]; then + do_nodes $CLIENTS $create_as "touch $dir/nrs_r_\$HOSTNAME;" || + error "touch failed for $dir/nrs_r_*" + do_nodes $CLIENTS $create_as touch "$dir/nrs_w_\$HOSTNAME" || + error "touch failed for $dir/nrs_w_*" + do_nodes $CLIENTS $create_as "chmod 777 $dir/nrs_*_\$HOSTNAME;" || + error "chmod failed for $dir/nrs_*" + fi + do_nodes $CLIENTS $myRUNAS \ dd if=/dev/zero of="$dir/nrs_r_\$HOSTNAME" bs=1M count=$n || error "dd at 0 on client failed (1)" @@ -3862,6 +4345,7 @@ tbf_verify() { local dir=$DIR/$tdir local client1=${CLIENT1:-$(hostname)} local myRUNAS="$3" + local create_as="$4" local np=$(check_cpt_number ost1) [ $np -gt 0 ] || error "CPU partitions should not be $np." @@ -3871,6 +4355,11 @@ tbf_verify() { $LFS setstripe -c 1 -i 0 $dir || error "setstripe to $dir failed" chmod 777 $dir + if [[ -n "$create_as" ]]; then + $create_as touch $dir/tbf + chmod 777 $dir/tbf + fi + trap cleanup_tbf_verify EXIT echo "Limited write rate: $1, read rate: $2" echo "Verify the write rate is under TBF control" @@ -4187,6 +4676,8 @@ test_id() { local idstr="${1}id" local policy="${idstr}={$2}" local rate="rate=$3" + local runas_args="$4" + local createas_args="${5:-$runas_args}" do_nodes $(comma_list $(osts_nodes)) \ lctl set_param jobid_var=procname_uid \ @@ -4194,8 +4685,8 @@ test_id() { ost.OSS.ost_io.nrs_tbf_rule="start\ ost_${idstr}\ ${policy}\ ${rate}" [ $? -ne 0 ] && error "failed to set tbf ${idstr} policy" - nrs_write_read "runas $4" - tbf_verify $3 $3 "runas $4" + nrs_write_read "runas $runas_args" "runas $createas_args" + tbf_verify $3 $3 "runas $runas_args" "runas $createas_args" do_nodes $(comma_list $(osts_nodes)) \ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_${idstr}" \ @@ -4212,11 +4703,20 @@ test_77ja(){ skip "Need OST version at least 2.11.50" fi - test_id "u" "500" "5" "-u 500" - test_id "g" "500" "5" "-u 500 -g 500" + test_id "u" "$RUNAS_ID" "5" "-u $RUNAS_ID" + test_id "g" "$RUNAS_GID" "5" "-u $RUNAS_ID -g $RUNAS_GID" } run_test 77ja "check TBF-UID/GID NRS policy" +test_77jb() { # LU-16077 + (( "$OST1_VERSION" >= $(version_code 2.15.51) )) || + skip "Need OST version at least 2.15.51" + + test_id "u" "$RUNAS_ID" "5" "-u $RUNAS_ID" "-u 0 -g $RUNAS_GID" + test_id "g" "$RUNAS_GID" "5" "-u $RUNAS_ID -g $RUNAS_GID" "-u $RUNAS_ID -g 0" +} +run_test 77jb "check TBF-UID/GID NRS policy on files that don't belong to us" + cleanup_77k() { local rule_lists=$1 @@ -4284,29 +4784,29 @@ test_77k() { do_nodes $(comma_list $(osts_nodes)) \ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_a" \ ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_b" \ - ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ug\ uid={500}\&gid={1000}\ rate=5" - nrs_write_read "runas -u 500 -g 1000" - tbf_verify 5 5 "runas -u 500 -g 1000" + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ug\ uid={$RUNAS_ID}\&gid={$RUNAS_GID}\ rate=5" + nrs_write_read "runas -u $RUNAS_ID -g $RUNAS_GID" + tbf_verify 5 5 "runas -u $RUNAS_ID -g $RUNAS_GID" do_nodes $(comma_list $(osts_nodes)) \ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ug" \ - ost.OSS.ost_io.nrs_tbf_rule="start\ ext_uw\ uid={500}\&opcode={ost_write}\ rate=20" \ - ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ur\ uid={500}\&opcode={ost_read}\ rate=10" + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_uw\ uid={$RUNAS_ID}\&opcode={ost_write}\ rate=20" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ur\ uid={$RUNAS_ID}\&opcode={ost_read}\ rate=10" - nrs_write_read "runas -u 500" - tbf_verify 20 10 "runas -u 500" + nrs_write_read "runas -u $RUNAS_ID" + tbf_verify 20 10 "runas -u $RUNAS_ID" do_nodes $(comma_list $(osts_nodes)) \ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_uw" \ ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ur" \ - ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ uid={500},opcode={ost_write}\ rate=20" \ - ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ uid={500},opcode={ost_read}\ rate=10" - nrs_write_read "runas -u 500" - tbf_verify 10 10 "runas -u 500" - tbf_verify 20 10 "runas -u 500" + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ uid={$RUNAS_ID},opcode={ost_write}\ rate=20" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ uid={$RUNAS_ID},opcode={ost_read}\ rate=10" + nrs_write_read "runas -u $RUNAS_ID" + tbf_verify 10 10 "runas -u $RUNAS_ID" + tbf_verify 20 10 "runas -u $RUNAS_ID" cleanup_77k "ext_a ext_b" "fifo" } -run_test 77k "check TBF policy with NID/JobID/OPCode expression" +run_test 77k "check TBF policy with UID/GID/JobID/OPCode expression" test_77l() { [[ "$OST1_VERSION" -ge $(version_code 2.10.56) ]] || @@ -4420,6 +4920,203 @@ test_77n() { #LU-10802 } run_test 77n "check wildcard support for TBF JobID NRS policy" +test_77o() { + (( $OST1_VERSION > $(version_code 2.14.54) )) || + skip "need OST > 2.14.54" + + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies="tbf\ nid" + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ name\ nid={192.168.*.*@tcp}\ rate=10000" + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ name1\ nid={192.168.*.*@tcp}\ rate=10000" + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="change\ name1\ rank=name" + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="stop\ name" + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies="fifo" +} +run_test 77o "Changing rank should not panic" + +test_77q() { + local i + local gidlist="500 10 33 100 " + local uidlist=" 500 11 3" + + (( $MDS1_VERSION > $(version_code 2.14.54) )) || + skip "need MDS >= 2.14.54" + + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies="tbf" + stack_trap "do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies=fifo" + + # require 44cc782/LU-9859 to support list with unexpected spaces + (( MDS1_VERSION >= $(version_code 2.15.57) )) || + gidlist=$(echo $gidlist) uidlist=$(echo $uidlist) + + for i in {1..50}; do + local pid1 pid2 + + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'start rule77q_1 uid={$uidlist}&gid={$gidlist} rate=100'" & + pid1=$! + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'start rule77q_2 uid={1000}&gid={1000} rate=100'" & + pid2=$! + wait $pid1 || error "$i: Fail to start TBF rule 'rule77q_1'" + wait $pid2 || error "$i: Fail to start TBF rule 'rule77q_2'" + + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'stop rule77q_1'" & + pid1=$! + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'stop rule77q_2'" & + pid2=$! + wait $pid1 || error "$i: Fail to stop TBF rule 'rule77q_1'" + wait $pid2 || error "$i: Fail to stop TBF rule 'rule77q_2'" + done +} +run_test 77q "Parallel TBF rule definitions should not panic" + +test_77p() { + local c + local -a spec_chars=( + '@' '.' '~' '#' '/' '^' '%' '*' ';' ',' '?' '<' '>' ':' + '+' '=' ')' '(' '{' '}' '|' '[' ']' '!' '&' '\$' '\`' '\\') + + (( $MDS1_VERSION > $(version_code 2.14.54) )) || + skip "need MDS >= 2.14.54" + + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies="tbf" + stack_trap "do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies=fifo" + + # TBF rule name size is 16 bytes + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ test_77p_overflo\ uid={500}\ rate=500" && + error "The length of tbf rule name is not checked" || true + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ \ uid={500}\ rate=500" && + error "The server should not accept empty tbf rule name" || true + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ test_77p_empty" && + error "The server should not accept 'start ' without an expression" || true + + # Test with special chars + for c in "${spec_chars[@]}"; do + do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'start test77p${c}spec uid={500} rate=500'" && + error "Special char '${c}' should not be accepted in a tbf rule name" || true + done + +} +run_test 77p "Check validity of rule names for TBF policies" + +cleanup_77r() { + local pid=$1 + local saved_jobid=$2 + local current_jobid_var + + echo "cleanup 77r $pid" + + do_facet mds1 $LCTL set_param -n mds.MDS.mdt.nrs_policies=fifo + kill $pid || echo "fail to kill md thread" + + current_jobid_var=$($LCTL get_param -n jobid_var) + if [ $saved_jobid != $current_jobid_var ]; then + set_persistent_param_and_check client \ + "jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid + fi + + sleep 2 + rm -rf $DIR1/$tdir +} + +md_thread_run="true" +md_thread_77r() { + local pid + + while $md_thread_run; do + printf '%s\n' {$DIR1,$DIR2}/$tdir/${tfile}-{01..20} | + xargs -P20 -I{} $RUNAS bash -c 'touch {}; rm -f {}' \ + &> /dev/null & pid=$! + trap "echo kill md_thread xargs; md_thread_run=false; kill $pid" INT TERM + wait $pid + done +} + +wait_policy_state() { + local state="$1" + local policy="$2" + local change_pid="$3" + local time + + for time in {1..60}; do + local nbr_started + + nbr_started=$(do_facet mds1 $LCTL get_param mds.MDS.mdt.nrs_policies | + egrep -A2 "name: ${policy}$" | grep -c "state: $state") + + [[ "$nbr_started" != 2 ]] || return 0 + sleep 1 + done + + [[ -z "$change_pid" ]] || kill $change_pid || true + return 1 +} + +test_77r() { #LU-14976 + + (( MDS1_VERSION > $(version_code 2.15.56) )) || + skip "need MDS >= 2.15.56 c098c095 change nrs policies at run time" + + local pid + local -A rules + local -a policies + local saved_jobid_var + + rules["tbf uid"]="start md_rule uid={$RUNAS_ID} rate=1" + rules["tbf gid"]="start md_rule gid={$RUNAS_GID} rate=1" + rules["tbf jobid"]="start md_rule jobid={*.$RUNAS_ID} rate=1" + rules["tbf"]="start md_rule uid={$RUNAS_ID} rate=1" + policies=( + "tbf uid" + "tbf gid" + "tbf jobid" + "tbf" + "fifo" + ) + + test_mkdir -i 0 -c 1 $DIR1/$tdir + chmod 777 $DIR1/$tdir + + # Configure jobid_var + saved_jobid_var=$($LCTL get_param -n jobid_var) + if [ $saved_jobid_var != procname_uid ]; then + set_persistent_param_and_check client \ + "jobid_var" "$FSNAME.sys.jobid_var" procname_uid + fi + + # start md thread + md_thread_77r & pid=$! + stack_trap "cleanup_77r $pid '$saved_jobid_var'" + + local policy + for policy in "${policies[@]}"; do + local change_pid + + # wait to queue requests + sleep 5 + + do_facet mds1 "$LCTL set_param mds.MDS.mdt.nrs_policies='$policy'" & + change_pid=$! + + wait_policy_state "started" "$policy" "$change_pid" || + error "timeout to start '$policy' policy" + + [[ -n "${rules[$policy]}" ]] || continue + + do_facet mds1 "$LCTL set_param mds.MDS.mdt.nrs_tbf_rule='${rules[$policy]}'" || + error "fail to set rule '${rules[$policy]}' to '$policy'" + done + + wait_policy_state "stopped" "tbf" || + error "fail to stop tbf policy" + + echo "check the number of requests in queue:" + local awkcmd='/name: / {last = $3} ' + awkcmd+='/queued: / {printf " %s: %d\n", last, $2;' + awkcmd+=' if (last == "tbf" && $2 > 0) exit 1;}' + do_facet mds1 $LCTL get_param mds.MDS.mdt.nrs_policies | awk "$awkcmd" || + error "request leak in tbf policies" +} +run_test 77r "Change type of tbf policy at run time" + test_78() { #LU-6673 local rc @@ -4478,6 +5175,7 @@ test_80a() { local file local pid + mkdir_on_mdt0 $DIR1/$tdir mkdir -p $DIR1/$tdir/dir createmany -o $DIR1/$tdir/dir/f 10 || error "create files under remote dir failed $i" @@ -4517,121 +5215,79 @@ cleanup_80b() { kill -9 $migrate_pid } +success_count=0 +failure_count=0 + +run_and_count() +{ + eval $@ &>/dev/null && success_count=$((success_count + 1)) || + failure_count=$((failure_count + 1)) +} + 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 migdir1=$DIR1/$tdir/migrate_dir + local migdir2=$DIR2/$tdir/migrate_dir 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 + # prepare migrate directory + mkdir -p $migdir1 for F in {1,2,3,4,5}; do - echo "$F$F$F$F$F" > $migrate_dir1/file$F + echo "$F$F$F$F$F" > $migdir1/file$F echo "$F$F$F$F$F" > $DIR/$tdir/file$F done - #migrate the directories among MDTs + # migrate the directories among MDTs ( while true; do mdt_idx=$((RANDOM % MDSCOUNT)) - $LFS migrate -m $mdt_idx $migrate_dir1 &>/dev/null || - rc=$? - [ $rc -ne 0 -o $rc -ne 16 ] || break + # migrate may fail + $LFS migrate -m $mdt_idx $migdir1 &>/dev/null done ) & migrate_pid=$! 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 $migrate_dir2/link_file \ - &>/dev/null || rc1=$? - if [ -e $migrate_dir2/link_file ]; then - rm -rf $migrate_dir2/link_file - fi - - mrename $migrate_dir2/source_file \ - $migrate_dir2/target_file &>/dev/null || rc1=$? - [ $rc1 -ne 0 -o $rc1 -ne 1 ] || { - echo "rename failed with $rc1" - break - } - - if [ -e $migrate_dir2/target_file ]; then - rm -rf $migrate_dir2/target_file &>/dev/null || - rc1=$? - else - rm -rf $migrate_dir2/source_file &>/dev/null || - rc1=$? - 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 + # access the files at the same time + start_time=$SECONDS + echo "accessing the migrating directory for 1 minute..." + while ((SECONDS - start_time < 60)); do + run_and_count ls $migdir2 + run_and_count diff -u $DIR2/$tdir/file1 $migdir2/file1 + run_and_count "cat $migdir2/file2 > $migdir2/file3" + run_and_count "echo "aaaaa" > $migdir2/file4" + run_and_count stat $migdir2/file5 + run_and_count touch $migdir2/source_file + if [ -e $migdir2/source_file ]; then + run_and_count ln $migdir2/source_file \ + $migdir2/link_file + + [ -e $migdir2/link_file ] && + rm -rf $migdir2/link_file + + run_and_count mrename $migdir2/source_file \ + $migdir2/target_file + + [ -e $migdir2/target_file ] && + run_and_count rm -rf $migdir2/target_file || + run_and_count rm -rf $migdir2/source_file fi kill -0 $migrate_pid || { echo "migration stopped 1" break } - - [ $duration -ge 300 ] && break done - #check migration are still there + # check migration are still there kill -0 $migrate_pid || error "migration stopped 2" cleanup_80b + # access during migration may fail + echo "concurrent access $failure_count failures, $success_count successes" } run_test 80b "Accessing directory during migration" @@ -4667,13 +5323,10 @@ test_81b() { [ $MDSCOUNT -lt 2 ] && skip "We need at least 2 MDTs for this test" - local total local setattr_pid - total=1000 - $LFS mkdir -c $MDSCOUNT $DIR1/$tdir || error "$LFS mkdir" - createmany -o $DIR1/$tdir/$tfile. $total || error "createmany" + createmany -o $DIR1/$tdir/$tfile. $COUNT || error "createmany" ( while true; do @@ -4682,7 +5335,7 @@ test_81b() { ) & setattr_pid=$! - for i in $(seq $total); do + for ((i = 0; i < COUNT; i++)); do mrename $DIR2/$tdir/$tfile.$i $DIR2/$tdir/$tfile-new.$i \ > /dev/null done @@ -4737,6 +5390,66 @@ test_81c() { } run_test 81c "rename revoke LOOKUP lock for remote object" +cleanup_81d() { + for ((mds = 0; mds < $MDSCOUNT; mds++)); do + local d2=$DIR2/$tdir-$mds + + rm -rf $d2 & + done + wait || error "rm failed" +} + +test_81d() { + local setattr_pid + local mdts=$(comma_list $(mdts_nodes)) + + do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear > /dev/null" + + stack_trap cleanup_81d + for ((mds = 0; mds < $MDSCOUNT; mds++)); do + local d1=$DIR1/$tdir-$mds + + $LFS mkdir -i $mds $d1 $d1/_temporary || error "mkdir failed" + createmany -o $d1/_temporary/$tfile. $COUNT || + error "createmany failed for $d1/_temporary" + done + + for ((mds = 0; mds < $MDSCOUNT; mds++)); do + local d1=$DIR1/$tdir-$mds + local d2=$DIR2/$tdir-$mds + + for ((i = 0; i < COUNT; i++)); do + mrename $d1/_temporary/$tfile.$i $d1/$tfile.$i & + ((i++)) + mrename $d2/_temporary/$tfile.$i $d2/$tfile.$i & + done + done + wait || error "rename failed" + + cleanup_81d + local stats=$DIR1/md_stats + local total=$((MDSCOUNT * COUNT)) + + do_nodes $mdts "$LCTL get_param -n mdt.*.md_stats" > $stats + cat $stats + crossdir=$(awk '/crossdir_rename/ {sum+=$2} END {print sum}' $stats) + (( crossdir == total )) || + error "not crossdir: $crossdir != $total" + samedir=$(awk '/samedir_rename/ {sum+=$2} END {print sum}' $stats) + (( samedir == 0 )) || error "considered samedir: $samedir" + pardir=$(awk '/parallel_rename_dir/ {sum+=$2} END {print sum}' $stats) + (( pardir == 0 )) || error "considered directory: $pardir" + + + (( MDS1_VERSION >= $(version_code 2.15.60) )) || + { echo "need MDS >= 2.15.60 for parallel cross-dir"; return 0; } + + parfile=$(awk '/parallel_rename_file/ {sum+=$2} END {print sum}' $stats) + (( parfile == total )) || + error "not considered file: $parfile != $total" +} +run_test 81d "parallel rename file cross-dir on same MDT" + test_82() { [[ "$MDS1_VERSION" -gt $(version_code 2.6.91) ]] || skip "Need MDS version at least 2.6.92" @@ -4819,6 +5532,11 @@ test_84() { } run_test 84 "0-nlink race in lu_object_find()" +test_85() { + llapi_root_test $DIR/$tfile $DIR2/$tfile +} +run_test 85 "Lustre API root cache race" + test_90() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return local pid1 @@ -4989,6 +5707,69 @@ test_94() { } run_test 94 "signal vs CP callback race" +test_95a() { + local file=$DIR/$tfile + local file2=$DIR2/$tfile + local fast_read_save + local pid + + fast_read_save=$($LCTL get_param -n llite.*.fast_read | head -n 1) + [ -z "$fast_read_save" ] && skip "no fast read support" + + stack_trap "$LCTL set_param llite.*.fast_read=$fast_read_save" EXIT + $LCTL set_param llite.*.fast_read=0 + + $LFS setstripe -c $OSTCOUNT $file || error "failed to setstripe $file" + dd if=/dev/zero of=$file bs=1M count=2 || error "failed to write $file" + cancel_lru_locks $OSC + $MULTIOP $file Oz1048576w4096c || error "failed to write $file" + $MULTIOP $file oz1044480r4096c || error "failed to read $file" + + # OBD_FAIL_LLITE_PAGE_INVALIDATE_PAUSE 0x1421 + $LCTL set_param fail_loc=0x80001421 fail_val=7 + $MULTIOP $file2 Oz1048576w4096_c & + pid=$! + + sleep 2 + # OBD_FAIL_LLITE_READPAGE_PAUSE 0x1422 + $LCTL set_param fail_loc=0x80001422 fail_val=10 + $MULTIOP $file oz1044480r4096c || error "failed to read $file" + + kill -USR1 $pid && wait $pid || error "wait for PID $pid failed" +} +run_test 95a "Check readpage() on a page that was removed from page cache" + +test_95b() { + local file=$DIR/$tfile + local file2=$DIR2/$tfile + local fast_read_save + local pid + + fast_read_save=$($LCTL get_param -n llite.*.fast_read | head -n 1) + [ -z "$fast_read_save" ] && skip "no fast read support" + + stack_trap "$LCTL set_param llite.*.fast_read=$fast_read_save" EXIT + $LCTL set_param llite.*.fast_read=0 + + $LFS setstripe -c $OSTCOUNT $file || error "failed to setstripe $file" + dd if=/dev/zero of=$file bs=$((PAGE_SIZE * 3)) count=1 || + error "failed to write $file" + + # This does the read from the second mount, so this flushes the pages + # the first mount and creates new ones on the second mount + # OBD_FAIL_LLITE_READPAGE_PAUSE2 0x1424 + $LCTL set_param fail_loc=0x80001424 fail_val=5 + $MULTIOP $file2 or${PAGE_SIZE}c & + pid=$! + + sleep 2 + fadvise_dontneed_helper $file2 + $LCTL set_param fail_loc=0 + sleep 4 + wait $pid || error "failed to read file" +} +run_test 95b "Check readpage() on a page that is no longer uptodate" + # Data-on-MDT tests test_100a() { skip "Reserved for glimpse-ahead" && return @@ -5193,14 +5974,34 @@ run_test 101c "Discard DoM data on close-unlink" # This test opens the file normally on $DIR1, which is on one mount, and then # opens it by handle on $DIR2, which is on a different mount. test_102() { - [ "$MDS1_VERSION" -lt $(version_code 2.11.57) ] && + (( "$MDS1_VERSION" >= $(version_code 2.11.57) )) || skip "Needs MDS version 2.11.57 or later" echo "Test file_handle syscalls" > $DIR/$tfile || error "write failed" check_fhandle_syscalls $DIR/$tfile $DIR2 || - error "check_fhandle_syscalls failed" - rm -f $DIR2/$tfile + error "check_fhandle_syscalls $tfile failed" + + # test this is working on DNE directories also + if (( MDSCOUNT > 1 && MDS1_VERSION >= $(version_code 2.14.52) )); then + $LFS mkdir -i 1 $DIR/$tdir.remote + cancel_lru_locks mdc + check_fhandle_syscalls $DIR/$tdir.remote $DIR2 || + error "check_fhandle_syscalls $tdir.remote failed" + $LFS mkdir -c -1 $DIR/$tdir.remote/subdir + cancel_lru_locks mdc + check_fhandle_syscalls $DIR/$tdir.remote/subdir $DIR2 || + error "check_fhandle_syscalls $tdir.remote/subdir fail" + + $LFS mkdir -c -1 $DIR/$tdir.stripe + cancel_lru_locks mdc + check_fhandle_syscalls $DIR/$tdir.stripe $DIR2 || + error "check_fhandle_syscalls $tdir.stripe failed" + $LFS mkdir -c -1 $DIR/$tdir.stripe/subdir + cancel_lru_locks mdc + check_fhandle_syscalls $DIR/$tdir.stripe/subdir $DIR2 || + error "check_fhandle_syscalls $tdir.stripe/subdir fail" + fi } run_test 102 "Test open by handle of unlinked file" @@ -5447,8 +6248,8 @@ test_106c() { # Mask supported in stx_attributes by Lustre is # STATX_ATTR_IMMUTABLE(0x10) | STATX_ATTR_APPEND(0x20) : (0x30). mask=$($STATX -c %p $DIR/$tfile) - [[ $mask == "30" ]] || - error "supported stx_attributes: got '$mask', expected '30'" + (( (0x$mask & 0x30) == 0x30 )) || + error "supported stx_attributes: got '0x$mask', expected '0x30' at least" chattr +i $DIR/$tfile || error "chattr +i $DIR/$tfile failed" mask=$($STATX -c %r $DIR/$tfile) [[ $mask == "10" ]] || @@ -5585,7 +6386,7 @@ test_109() { for i in {1..50}; do log "Iteration $i" -#define OBD_FAIL_ONCE|OBD_FAIL_LLITE_RACE_MOUNT 0x80001417 +#define CFS_FAIL_ONCE|OBD_FAIL_LLITE_RACE_MOUNT 0x80001417 $LCTL set_param -n fail_loc=0x80001417 mount_client $MOUNT & pid1=$! @@ -5651,6 +6452,294 @@ test_110() { } run_test 110 "do not grant another lock on resend" +test_111() { + [ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs" + [[ $(facet_active_host mds1) = $(facet_active_host mds2) ]] || + skip "MDT0 and MDT1 should be on the same node" + + mkdir $DIR1/$tdir + $LFS mkdir -i 0 $DIR1/$tdir/mdt0dir + $LFS mkdir -i 1 $DIR1/$tdir/mdt1dir + + mkdir $DIR1/$tdir/mdt0dir/foodir + touch $DIR1/$tdir/mdt0dir/foodir/{file1,file2} + + $MULTIOP $DIR2/$tdir/mdt0dir/foodir/file2 Ow4096_c & + MULTIOP_PID=$! + ln $DIR1/$tdir/mdt0dir/foodir/file2 $DIR1/$tdir/mdt1dir/file2 + + #define OBD_FAIL_MDS_LINK_RENAME_RACE 0x18a + do_facet mds1 $LCTL set_param fail_loc=0x8000018a + + ln $DIR1/$tdir/mdt0dir/foodir/file2 $DIR1/$tdir/mdt1dir/file2x & + sleep 1 + + rm $DIR2/$tdir/mdt1dir/file2 + sleep 1 + + mv $DIR2/$tdir/mdt0dir/foodir/file1 $DIR2/$tdir/mdt0dir/foodir/file2 + sleep 1 + + kill $MULTIOP_PID + wait + rm -r $DIR1/$tdir || error "Removing test dir failed" +} +run_test 111 "A racy rename/link an open file should not cause fs corruption" + +test_112() { + (( MDSCOUNT >= 2 )) || + skip "We need at least 2 MDTs for this test" + + (( MDS1_VERSION >= $(version_code 2.14.54) )) || + skip "Need server version at least 2.14.54" + + local rr + local count + + rr=$($LCTL get_param -n lmv.*.qos_threshold_rr | head -n1) + rr=${rr%%%} + stack_trap "$LCTL set_param lmv.*.qos_threshold_rr=$rr > /dev/null" + + mkdir -p $DIR1/$tdir/s1/s2 || error "mkdir s2 failed" + $LFS mkdir -i 0 $DIR1/$tdir/s1/s2/s3 || error "mkdir s3 failed" + $LFS setdirstripe -D -i -1 --max-inherit-rr=0 $DIR1/$tdir/s1/s2/s3 || + error "setdirstripe s3 failed" + $LCTL set_param lmv.*.qos_threshold_rr=90 + mkdir $DIR2/$tdir/s1/s2/s3/d{1..64} + count=$($LFS getstripe -m $DIR2/$tdir/s1/s2/s3/d* | grep ^0 | wc -l) + (( count == 64 )) || error "only $count subdirs created on MDT0" + + $LFS setdirstripe -D -i -1 --max-inherit-rr=3 $DIR1/$tdir/s1/s2/s3 || + error "setdirstripe s3 failed" + mkdir $DIR2/$tdir/s1/s2/s3/s{1..64} + count=$($LFS getstripe -m $DIR2/$tdir/s1/s2/s3/s* | grep ^0 | wc -l) + (( count == 64 / MDSCOUNT )) || error "$count subdirs created on MDT0" +} +run_test 112 "update max-inherit in default LMV" + +test_113 () { + (( MDS1_VERSION >= $(version_code 2.15.50) )) || + skip "Need server version at least 2.15.50" + + local instance + local nid + + instance=$($LFS getname -i $DIR1) || + error "cannot get instance of $DIR1" + + $LFS check osts $DIR1 | grep $instance || + error "cannot find OSTs of instance $instance" + + $LFS check osts $DIR1 | grep -v $instance + if (( $? == 0 )); then + error "find OSTs other than instance $instance" + fi + + $LFS check osts | grep $instance || + error "cannot find other OSTs" + + nid=$(df $DIR2 | tail -1 | sed 's%:/.*%%') || + error "cannot parse nid for $DIR2" + + $LFS check mgts $DIR2 | grep MGC$nid || + error "cannot find mgc of $nid" + + $LFS check mgts $DIR2 | grep -v MGC$nid + if (( $? == 0 )); then + error "find MGTs other than nid $nid" + fi +} +run_test 113 "check servers of specified fs" + +check_default_lmv() { + local dir=$1 + + local enabled + local dmv + local index + local count + local inherit + local inherit_rr + local raw + + enabled=$(do_facet mds1 \ + $LCTL get_param -n mdt.*-MDT0000*.enable_dmv_implicit_inherit) + + dmv=$($LFS getdirstripe -D $dir) + echo $dir $dmv + index=$(echo $dmv | awk '{ print $4 }') + (( index == $2 )) || error "$dir default stripe index $index != $2" + + count=$(echo $dmv | awk '{ print $2 }') + (( count == $3 )) || error "$dir default stripe count $count != $3" + + inherit=$(echo $dmv | awk '{ print $8 }') + (( inherit == $4 )) || error "$dir default max-inherit $inherit != $4" + + if [ $index -eq -1 ]; then + inherit_rr=$(echo $dmv | awk '{ print $10 }') + (( inherit_rr == $5 )) || + error "$dir default max-inherit-rr $inherit_rr != $5" + fi + + # with --raw, print default LMV stored in inode, otherwise print nothing + raw=$($LFS getdirstripe -D --raw $dir) + if (( enabled == 1 )); then + [ -z $raw ] || + error "implicit inherited DMV is printed with --raw" + else + # if disabled, dmv is stored in inode, which will always + # print max-inherit-rr + echo $dir $raw + [[ $raw =~ $dmv.* ]] || error "$dir raw $raw != dmv $dmv" + fi +} + +test_dmv_imp_inherit() { + local dmv + local raw + local index + local count + local inherit + local inherit_rr + + rm -rf $DIR/$tdir || error "rm $tdir failed" + mkdir -p $DIR/$tdir || error "mkdir $tdir failed" + + # set dir default LMV + $LFS setdirstripe -D -c1 -X4 --max-inherit-rr 2 $DIR/$tdir || + error "setdirstripe -D $tdir failed" + dmv=$($LFS getdirstripe -D $DIR/$tdir) + raw=$($LFS getdirstripe -D --raw $DIR/$tdir) + [ "$dmv" == "$raw" ] || error "$dmv != $raw" + + mkdir -p $DIR/$tdir/l1/l2/l3 || error "mkdir $DIR/$tdir/l1/l2/l3 failed" + check_default_lmv $DIR/$tdir/l1/l2/l3 -1 1 1 0 + check_default_lmv $DIR2/$tdir/l1/l2/l3 -1 1 1 0 + + # below tests are valid only when this feature is enabled + local enabled=$(do_facet mds1 \ + $LCTL get_param -n mdt.*-MDT0000*.enable_dmv_implicit_inherit) + + (( enabled == 1 )) || return 0 + + # set l2 default LMV, dmv of l3 should change immediately + $LFS setdirstripe -D -i1 -c2 -X4 $DIR/$tdir/l1/l2 || + error "setdirstripe -D $tdir/l1/l2 failed" + + check_default_lmv $DIR/$tdir/l1/l2/l3 1 2 3 + check_default_lmv $DIR2/$tdir/l1/l2/l3 1 2 3 + + # change tdir default LMV, dmv of l3 should be unchanged because dmv + # of l2 is explicitly set + $LFS setdirstripe -D -i2 -c2 -X3 $DIR/$tdir || + error "setdirstripe -D $tdir failed" + + check_default_lmv $DIR/$tdir/l1 2 2 2 + check_default_lmv $DIR2/$tdir/l1 2 2 2 + check_default_lmv $DIR/$tdir/l1/l2/l3 1 2 3 + check_default_lmv $DIR2/$tdir/l1/l2/l3 1 2 3 +} + +test_114() { + (( MDSCOUNT >= 2 )) || + skip "We need at least 2 MDTs for this test" + + (( MDS1_VERSION >= $(version_code 2.15.55.45) )) || + skip "Need server version at least 2.15.54.45" + + test_dmv_imp_inherit + + # disable dmv_imp_inherit to simulate old client + local mdts=$(comma_list $(mdts_nodes)) + + do_nodes $mdts $LCTL set_param -n \ + mdt.*MDT*.enable_dmv_implicit_inherit=0 + test_dmv_imp_inherit + do_nodes $mdts $LCTL set_param -n \ + mdt.*MDT*.enable_dmv_implicit_inherit=1 +} +run_test 114 "implicit default LMV inherit" + +test_115() { + local td=$DIR/$tdir + + [ "$mds1_FSTYPE" == "ldiskfs" ] || skip_env "ldiskfs only test" + + mkdir_on_mdt0 $td || error "can't mkdir" + # turn it htree (don't really needed) + createmany -m $td/f 3000 || error "can't createmany" + + # here is an example of debugfs output for htree command: + # Entry #0: Hash 0x00000000, block 27 + # Reading directory block 27, phys 16760 + # 938 0x0016fb58-7f3d21f5 (32) f775 834 0x001db8c8-d31a4e0e (32) f671 + # 1085 0x0040cb70-4498abd4 (32) f922 1850 0x0066a1e6-f6f0dc69 (32) f1687 + # 2005 0x006c1a46-ef466058 (32) f1842 2025 0x007e64d4-8b28b734 (32) f1862 + # 642 0x008b53a0-77adc601 (32) f479 447 0x009ec152-af54eea3 (32) f284 + # 1740 0x00c38f56-ed310e61 (32) f1577 2165 0x00cdfd66-f429a93f (32) f2002 + # 930 0x00d7ada4-b80421c9 (32) f767 1946 0x00da6a7a-e8080600 (32) f1783 + # 273 0x00f8ea00-760bf97c (32) f110 1589 0x0103c4ee-94fad5dd (32) f1426 + # 1383 0x01193516-83120b48 (32) f1220 2379 0x01431e3c-e85b5bd9 (32) f2216 + # + # find couple names in a same htree block of the same size + mdt_dev=$(facet_device $SINGLEMDS) + de=( $(do_facet $SINGLEMDS "debugfs -c -R 'htree /ROOT/$tdir' $mdt_dev" | + awk '/Reading directory block/ { getline; print $4,$8; exit; }' )) + local de1=${de[0]} + local de2=${de[1]} + [[ $de1 == "" || $de2 == "" ]] && error "de1=$de1 de2=$de2" + echo "USE: $de1 $de2" + # release one mkdir will lookup + rm $DIR/$tdir/$de2 +#define OBD_FAIL_MDS_PAUSE_CREATE_AFTER_LOOKUP 0x2401 + do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80002401 fail_val=5 + mkdir $DIR/$tdir/$de2 & + sleep 0.3 + local PID1=$! + # recreate $de2 + mkdir $DIR2/$tdir/$de2 + # release space $de1 (should be enough to save $de2) + rm $DIR2/$tdir/$de1 + # ready to create a dup of $de2 + wait $PID1 + local found=$(ls $DIR/$tdir/|grep "^$de2\$"|wc -l) + (( $found == 1 )) || error "found $found" +} +run_test 115 "ldiskfs doesn't check direntry for uniqueness" + +test_116() { + (( $MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs" + (( $MDS1_VERSION >= $(version_code 2.15.61) )) || + skip "Need MDS version at least 2.15.61 for intent mkdir" + + local mdt_idx + local save="$TMP/$TESTSUITE-$TESTNAME.parameters" + + save_lustre_params client "llite.*.intent_mkdir" > $save + stack_trap "restore_lustre_params < $save; rm -f $save" EXIT + $LCTL set_param llite.*.intent_mkdir=1 + + $LFS mkdir -c$MDSCOUNT -i0 $DIR/$tdir || + error "$LFS mkdir $DIR/$tdir failed" + echo "MD layout $DIR/$tdir:" + $LFS getdirstripe $DIR/$tdir + echo "mkdir $DIR/$tdir/tdir0" + mkdir $DIR/$tdir/tdir0 || error "mkdir tdir0 failed" + echo "setdirstripe -D -i1 $DIR2/$tdir/tdir0" + $LFS setdirstripe -D -i1 $DIR2/$tdir/tdir0 || + error "$LFS setdirstripe $DIR2/$tdir/tdir0 failed" + echo "mkdir $DIR/$tdir/tdir0/tdir11" + mkdir $DIR/$tdir/tdir0/tdir11 || error "mkdir tdir0/tdir11 failed" + $LFS getdirstripe $DIR/$tdir/tdir0 + $LFS getdirstripe $DIR/$tdir/tdir0/tdir11 + + mdt_idx=$($LFS getstripe -m $DIR/$tdir/tdir0/tdir11) + [ $mdt_idx == 1 ] || + error "$DIR/$tdir/tdir0/tdir11 on wrong MDT $mdt_idx" +} +run_test 116 "DNE: Set default LMV layout from a remote client" + log "cleanup: ======================================================" # kill and wait in each test only guarentee script finish, but command in script @@ -5659,7 +6748,7 @@ log "cleanup: ======================================================" [ "$(mount | grep $MOUNT2)" ] && wait_update $HOSTNAME "fuser -m $MOUNT2" "" || true -complete $SECONDS +complete_test $SECONDS rm -f $SAMPLE_FILE check_and_cleanup_lustre exit_status