X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanityn.sh;h=0ab469be3e6f86166ef2a1daba5fe0a18c849f07;hp=42beac15832a54f477976e253b7ca5a371494b45;hb=cbc62b0b829afdceaa01820996e567b5bdeb281c;hpb=c50bd08b35757d20e606c10bb5b4474011dd07f6 diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh index 42beac1..0ab469b 100755 --- a/lustre/tests/sanityn.sh +++ b/lustre/tests/sanityn.sh @@ -3,45 +3,41 @@ set -e ONLY=${ONLY:-"$*"} -# bug number for skipped test: 9977/LU-7105 -# LU-7105 -ALWAYS_EXCEPT=" 28 $SANITYN_EXCEPT" -# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! - -SRCDIR=$(dirname $0) -PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH SIZE=${SIZE:-40960} -CHECKSTAT=${CHECKSTAT:-"checkstat -v"} OPENFILE=${OPENFILE:-openfile} OPENUNLINK=${OPENUNLINK:-openunlink} -export MULTIOP=${MULTIOP:-multiop} export TMP=${TMP:-/tmp} MOUNT_2=${MOUNT_2:-"yes"} CHECK_GRANT=${CHECK_GRANT:-"yes"} GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""} -SAVE_PWD=$PWD - -export NAME=${NAME:-local} - -LUSTRE=${LUSTRE:-`dirname $0`/..} +LUSTRE=${LUSTRE:-$(dirname $0)/..} . $LUSTRE/tests/test-framework.sh -CLEANUP=${CLEANUP:-:} -SETUP=${SETUP:-:} init_test_env $@ -. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} init_logging -if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then -# bug number for skipped test: - ALWAYS_EXCEPT="$ALWAYS_EXCEPT " -# LU-2829 / LU-2887 - make allowances for ZFS slowness +ALWAYS_EXCEPT="$SANITYN_EXCEPT " +# bug number for skipped test: LU-7105 +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 [ $mds1_FSTYPE = "zfs" ]; then + # LU-2829 / LU-2887 - make allowances for ZFS slowness TEST33_NFILES=${TEST33_NFILES:-1000} fi + # 23 (min)" [ "$SLOW" = "no" ] && EXCEPT_SLOW="33a" +build_test_filter + FAIL_ON_ERROR=false SETUP=${SETUP:-:} @@ -62,8 +58,6 @@ dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=1 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS -build_test_filter - test_1() { touch $DIR1/$tfile [ -f $DIR2/$tfile ] || error "Check create" @@ -316,65 +310,52 @@ test_14aa() { MULTIPID=$! $DIR2/$tdir/$tfile && error || true - kill -USR1 $MULTIPID - wait $MULTIPID || return 2 + kill $MULTIPID } run_test 14aa "execution of file open for write returns -ETXTBSY" test_14ab() { test_mkdir $DIR1/$tdir - cp -p $(which multiop) $DIR1/$tdir/multiop || error "cp failed" - MULTIOP_PROG=$DIR1/$tdir/multiop multiop_bg_pause $TMP/$tfile O_c || - return 1 - MULTIOP_PID=$! - $MULTIOP $DIR2/$tdir/multiop Oc && error "expected error, got success" - kill -USR1 $MULTIOP_PID || return 2 - wait $MULTIOP_PID || return 3 - rm $TMP/$tfile $DIR1/$tdir/multiop || error "removing multiop" + cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed" + $DIR1/$tdir/sleep 60 & + SLEEP_PID=$! + $MULTIOP $DIR2/$tdir/sleep Oc && error "expected error, got success" + kill $SLEEP_PID } run_test 14ab "open(RDWR) of executing file returns -ETXTBSY" test_14b() { # bug 3192, 7040 test_mkdir $DIR1/$tdir - cp -p $(which multiop) $DIR1/$tdir/multiop || error "cp failed" - MULTIOP_PROG=$DIR1/$tdir/multiop multiop_bg_pause $TMP/$tfile O_c || - return 1 - MULTIOP_PID=$! - $TRUNCATE $DIR2/$tdir/multiop 0 && kill -9 $MULTIOP_PID && \ + cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed" + $DIR1/$tdir/sleep 60 & + SLEEP_PID=$! + $TRUNCATE $DIR2/$tdir/sleep 60 && kill -9 $SLEEP_PID && \ error "expected truncate error, got success" - kill -USR1 $MULTIOP_PID || return 2 - wait $MULTIOP_PID || return 3 - cmp $(which multiop) $DIR1/$tdir/multiop || error "binary changed" - rm $TMP/$tfile $DIR1/$tdir/multiop || error "removing multiop" + kill $SLEEP_PID + cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed" } run_test 14b "truncate of executing file returns -ETXTBSY ======" test_14c() { # bug 3430, 7040 test_mkdir $DIR1/$tdir - cp -p $(which multiop) $DIR1/$tdir/multiop || error "cp failed" - MULTIOP_PROG=$DIR1/$tdir/multiop multiop_bg_pause $TMP/$tfile O_c || - return 1 - MULTIOP_PID=$! - cp /etc/hosts $DIR2/$tdir/multiop && error "expected error, got success" - kill -USR1 $MULTIOP_PID || return 2 - wait $MULTIOP_PID || return 3 - cmp $(which multiop) $DIR1/$tdir/multiop || error "binary changed" - rm $TMP/$tfile $DIR1/$tdir/multiop || error "removing multiop" + cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed" + $DIR1/$tdir/sleep 60 & + SLEEP_PID=$! + cp /etc/hosts $DIR2/$tdir/sleep && error "expected error, got success" + kill $SLEEP_PID + cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed" } run_test 14c "open(O_TRUNC) of executing file return -ETXTBSY ==" test_14d() { # bug 10921 test_mkdir $DIR1/$tdir - cp -p $(which multiop) $DIR1/$tdir/multiop || error "cp failed" - MULTIOP_PROG=$DIR1/$tdir/multiop multiop_bg_pause $TMP/$tfile O_c || - return 1 - MULTIOP_PID=$! + cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed" + $DIR1/$tdir/sleep 60 & + SLEEP_PID=$! log chmod - chmod 600 $DIR1/$tdir/multiop || error "chmod failed" - kill -USR1 $MULTIOP_PID || return 2 - wait $MULTIOP_PID || return 3 - cmp $(which multiop) $DIR1/$tdir/multiop || error "binary changed" - rm $TMP/$tfile $DIR1/$tdir/multiop || error "removing multiop" + chmod 600 $DIR1/$tdir/sleep || error "chmod failed" + kill $SLEEP_PID + cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed" } run_test 14d "chmod of executing file is still possible ========" @@ -382,7 +363,7 @@ test_15() { # bug 974 - ENOSPC echo "PATH=$PATH" sh oos2.sh $MOUNT1 $MOUNT2 wait_delete_completed - grant_error=`dmesg | grep "> available"` + grant_error=$(dmesg | grep "< tot_grant") [ -z "$grant_error" ] || error "$grant_error" } run_test 15 "test out-of-space with multiple writers ===========" @@ -391,7 +372,7 @@ COUNT=${COUNT:-2500} # The FSXNUM reduction for ZFS is needed until ORI-487 is fixed. # We don't want to skip it entirely, but ZFS is VERY slow and cannot # pass a 2500 operation dual-mount run within the time limit. -if [ "$(facet_fstype ost1)" = "zfs" ]; then +if [ "$ost1_FSTYPE" = "zfs" ]; then FSXNUM=$((COUNT / 5)) FSXP=1 elif [ "$SLOW" = "yes" ]; then @@ -408,6 +389,8 @@ test_16a() { local stripe_size=$(do_facet $SINGLEMDS \ "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize") + check_set_fallocate + # to allocate grant because it may run out due to test_15. $LFS setstripe -c -1 $file1 dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync @@ -415,12 +398,12 @@ test_16a() { rm -f $file1 $LFS setstripe -c -1 $file1 # b=10919 - fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 \ - || error "fsx failed" + $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 || + 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 \ + $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 -Z -r 4096 \ -w 4096 $file1 $file2 || error "fsx with O_DIRECT failed." } run_test 16a "$FSXNUM iterations of dual-mount fsx" @@ -429,18 +412,23 @@ run_test 16a "$FSXNUM iterations of dual-mount fsx" test_16b() { local file1=$DIR1/$tfile local file2=$DIR2/$tfile + local stripe_size=($($LFS getstripe -S $DIR)) + + check_set_fallocate # to allocate grant because it may run out due to test_15. lfs setstripe -c -1 $file1 - dd if=/dev/zero of=$file1 bs=$STRIPE_BYTES count=$OSTCOUNT oflag=sync - dd if=/dev/zero of=$file2 bs=$STRIPE_BYTES count=$OSTCOUNT oflag=sync + dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync || + error "dd failed writing to file=$file1" + dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync || + error "dd failed writing to file=$file2" rm -f $file1 lfs setstripe -c -1 $file1 # b=10919 # -o is set to 8192 because writes < 1 page and between 1 and 2 pages # create a mix of tiny writes & normal writes - fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -o 8192 -S 0 $file1 \ - $file2 + $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -o 8192 -S 0 \ + $file1 $file2 || error "fsx with tiny write failed." } run_test 16b "$FSXNUM iterations of dual-mount fsx at small size" @@ -450,7 +438,9 @@ test_16c() { local stripe_size=$(do_facet $SINGLEMDS \ "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize") - [ $(facet_fstype ost1) != ldiskfs ] && skip "dio on ldiskfs only" + [ "$ost1_FSTYPE" != ldiskfs ] && skip "dio on ldiskfs only" + + check_set_fallocate # to allocate grant because it may run out due to test_15. $LFS setstripe -c -1 $file1 @@ -468,8 +458,8 @@ test_16c() { set_osd_param $list '' writethrough_cache_enable 0 $LFS setstripe -c -1 $file1 # b=10919 - fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 \ - || error "fsx failed" + $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 || + error "fsx failed" rm -f $file1 set_osd_param $list '' read_cache_enable 1 @@ -479,6 +469,69 @@ test_16c() { } run_test 16c "verify data consistency on ldiskfs with cache disabled (b=17397)" +test_16d() { + local file1=$DIR1/$tfile + local file2=$DIR2/$tfile + local file3=$DIR1/file + local tmpfile=$(mktemp) + local stripe_size=$(do_facet $SINGLEMDS \ + "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize") + + # to allocate grant because it may run out due to test_15. + $LFS setstripe -c -1 $file1 + stack_trap "rm -f $file1 $file2 $file3 $tmpfile" + dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync + dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync + rm -f $file1 + + $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 + diff $file1 $tmpfile || error "file different(1)" + rm -f $file1 + + # buffer write on one client, but direct read from another + dd if=$tmpfile of=$file1 bs=1M count=100 + dd if=$file2 of=$file3 bs=1M iflag=direct count=100 + diff $file3 $tmpfile || error "file different(2)" + + rm -f $file3 $file2 $file1 + # direct write on one client + dd if=$tmpfile of=$file1 bs=1M count=100 oflag=direct + # buffer read from another client + dd if=$file2 of=$file3 bs=1M count=100 + diff $file3 $tmpfile || error "file different(3)" +} +run_test 16d "Verify DIO and buffer IO with two clients" + +test_16e() { # LU-13227 + # issue: LU-14314 + + (( "$MDS1_VERSION" >= $(version_code 2.13.53) )) || + skip "Need MDS version at least 2.13.53" + + local file1=$DIR1/$tfile + local file2=$DIR2/$tfile + + # client1 write 10M data + dd if=/dev/zero of=$file1 bs=1M count=10 + # drop locks + cancel_lru_locks osc > /dev/null + # use lockahead to generate one PW lock to keep LVB loaded. + $LFS ladvise -a lockahead --start 0 --length 1M \ + --mode WRITE $file1 + # direct write to extend file size on client2 + dd if=/dev/zero of=$file2 bs=1M seek=20 count=1 \ + oflag=direct conv=notrunc + local filesize=$(stat -c %s $file2) + [ "$filesize" -eq 22020096 ] || + error "expected filesize 22020096 got $filesize" + rm -f $file1 +} +run_test 16e "Verify size consistency for O_DIRECT write" test_17() { # bug 3513, 3667 remote_ost_nodsh && skip "remote OST with nodsh" && return @@ -506,22 +559,27 @@ test_18() { excepts="$excepts -e $(($(printf %d \'$idx)-96))" done - $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2 $excepts + excepts="$excepts -e 7 -e 8 -e 9" + $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2 $excepts || + error "mmap_sanity test failed" sync; sleep 1; sync } run_test 18 "mmap sanity check =================================" test_19() { # bug3811 local node=$(facet_active_host ost1) + local device="$FSNAME-OST*" - [ "x$DOM" = "xyes" ] && node=$(facet_active_host $SINGLEMDS) + [ "x$DOM" = "xyes" ] && node=$(facet_active_host $SINGLEMDS) && + device="$FSNAME-MDT*" # check whether obdfilter is cache capable at all - get_osd_param $node '' read_cache_enable >/dev/null || + get_osd_param $node $device read_cache_enable >/dev/null || skip "not cache-capable obdfilter" - local MAX=$(get_osd_param $node '' readcache_max_filesize | head -n 1) - set_osd_param $node '' readcache_max_filesize 4096 + local max=$(get_osd_param $node $device readcache_max_filesize |\ + head -n 1) + set_osd_param $node $device readcache_max_filesize 4096 dd if=/dev/urandom of=$TMP/$tfile bs=512k count=32 local SUM=$(cksum $TMP/$tfile | cut -d" " -f 1,2) cp $TMP/$tfile $DIR1/$tfile @@ -536,22 +594,22 @@ test_19() { # bug3811 [ "$(cat $TMP/sum2)" = "$SUM" ] || \ error "$DIR2/$tfile $(cat $TMP/sum2) != $SUM" done - set_osd_param $node '' readcache_max_filesize $MAX + set_osd_param $node $device readcache_max_filesize $max rm $DIR1/$tfile } run_test 19 "test concurrent uncached read races ===============" test_20() { - test_mkdir $DIR1/d20 - cancel_lru_locks $OSC - CNT=$((`lctl get_param -n llite.*.dump_page_cache | wc -l`)) - $MULTIOP $DIR1/f20 Ow8190c - $MULTIOP $DIR2/f20 Oz8194w8190c - $MULTIOP $DIR1/f20 Oz0r8190c - cancel_lru_locks $OSC - CNTD=$((`lctl get_param -n llite.*.dump_page_cache | wc -l` - $CNT)) - [ $CNTD -gt 0 ] && \ - error $CNTD" page left in cache after lock cancel" || true + test_mkdir $DIR1/$tdir + cancel_lru_locks + CNT=$($LCTL get_param -n llite.*.dump_page_cache | wc -l) + $MULTIOP $DIR1/$tdir/$tfile Ow8190c + $MULTIOP $DIR2/$tdir/$tfile Oz8194w8190c + $MULTIOP $DIR1/$tdir/$tfile Oz0r8190c + cancel_lru_locks + CNT2=$($LCTL get_param -n llite.*.dump_page_cache | wc -l) + [[ $CNT2 == $CNT ]] || + error $((CNT2 - CNT))" page left in cache after lock cancel" } run_test 20 "test extra readahead page left in cache ====" @@ -707,7 +765,7 @@ test_27() { lctl clear dd if=/dev/zero of=$DIR2/$tfile bs=$((4096+4))k conv=notrunc count=4 seek=3 & DD2_PID=$! - usleep 50 + sleep 0.5 log "dd 1 started" dd if=/dev/zero of=$DIR1/$tfile bs=$((16384-1024))k conv=notrunc count=1 seek=4 & @@ -814,11 +872,13 @@ enable_lockless_truncate() { } test_32a() { # bug 11270 - local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + local save="$TMP/$TESTSUITE-$TESTNAME.parameters" local stripe_size=$(do_facet $SINGLEMDS \ "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize") - save_lustre_params client "$OSC.*.lockless_truncate" > $p + save_lustre_params client "$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 @@ -829,13 +889,15 @@ test_32a() { # bug 11270 log "checking cached lockless truncate" $TRUNCATE $DIR1/$tfile 8000000 - $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size" + $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 "wrong file size" + $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" @@ -843,13 +905,11 @@ test_32a() { # bug 11270 enable_lockless_truncate 0 clear_stats $OSC.*.${OSC}_stats $TRUNCATE $DIR2/$tfile 3000000 - $CHECKSTAT -s 3000000 $DIR1/$tfile || error "wrong file size" + $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 $DIR1/$tfile - # restore lockless_truncate default values - restore_lustre_params < $p - rm -f $p + rm -f $DIR1/$tfile } run_test 32a "lockless truncate" @@ -927,52 +987,52 @@ print_jbd_stat () { # commit on sharing tests test_33a() { - remote_mds_nodsh && skip "remote MDS with nodsh" && return + remote_mds_nodsh && skip "remote MDS with nodsh" && return - [ -z "$CLIENTS" ] && skip "Need two or more clients, have $CLIENTS" && return 0 - [ $CLIENTCOUNT -lt 2 ] && - skip "Need two or more clients, have $CLIENTCOUNT" && return 0 + [ -z "$CLIENTS" ] && skip "Need two or more clients, have $CLIENTS" + [ $CLIENTCOUNT -lt 2 ] && + skip "Need two or more clients, have $CLIENTCOUNT" - local nfiles=${TEST33_NFILES:-10000} - local param_file=$TMP/$tfile-params - local fstype=$(facet_fstype $SINGLEMDS) + local nfiles=${TEST33_NFILES:-10000} + local param_file=$TMP/$tfile-params + local COS + local jbdold="N/A" + local jbdnew="N/A" + local jbd save_lustre_params $(get_facets MDS) \ "mdt.*.commit_on_sharing" > $param_file - local COS - local jbdold="N/A" - local jbdnew="N/A" - local jbd - - for COS in 0 1; do - do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=$COS - avgjbd=0 - avgtime=0 - for i in 1 2 3; do - do_nodes $CLIENT1,$CLIENT2 "mkdir -p $DIR1/$tdir-\\\$(hostname)-$i" - - [ $fstype = ldiskfs ] && jbdold=$(print_jbd_stat) - echo "=== START createmany old: $jbdold transaction" - local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > /dev/null 2>&1") - [ $fstype = ldiskfs ] && jbdnew=$(print_jbd_stat) - [ $fstype = ldiskfs ] && jbd=$(( jbdnew - jbdold )) - echo "=== END createmany new: $jbdnew transaction : $jbd transactions nfiles $nfiles time $elapsed COS=$COS" - [ $fstype = ldiskfs ] && avgjbd=$(( avgjbd + jbd )) - avgtime=$(( avgtime + elapsed )) - done - eval cos${COS}_jbd=$((avgjbd / 3)) - eval cos${COS}_time=$((avgtime / 3)) - done + for COS in 0 1; do + do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=$COS + avgjbd=0 + avgtime=0 + for i in 1 2 3; do + do_nodes $CLIENT1,$CLIENT2 "mkdir -p $DIR1/$tdir-\\\$(hostname)-$i" + + [ "$mds1_FSTYPE" = ldiskfs ] && jbdold=$(print_jbd_stat) + echo "=== START createmany old: $jbdold transaction" + local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > /dev/null 2>&1") + [ "$mds1_FSTYPE" = ldiskfs ] && jbdnew=$(print_jbd_stat) + [ "$mds1_FSTYPE" = ldiskfs ] && jbd=$(( jbdnew - jbdold )) + echo "=== END createmany new: $jbdnew transaction : $jbd transactions nfiles $nfiles time $elapsed COS=$COS" + [ "$mds1_FSTYPE" = ldiskfs ] && avgjbd=$(( avgjbd + jbd )) + avgtime=$(( avgtime + elapsed )) + done + eval cos${COS}_jbd=$((avgjbd / 3)) + eval cos${COS}_time=$((avgtime / 3)) + done - echo "COS=0 transactions (avg): $cos0_jbd time (avg): $cos0_time" - echo "COS=1 transactions (avg): $cos1_jbd time (avg): $cos1_time" - [ "$cos0_jbd" != 0 ] && echo "COS=1 vs COS=0 jbd: $((((cos1_jbd/cos0_jbd - 1)) * 100 )) %" - [ "$cos0_time" != 0 ] && echo "COS=1 vs COS=0 time: $((((cos1_time/cos0_time - 1)) * 100 )) %" + echo "COS=0 transactions (avg): $cos0_jbd time (avg): $cos0_time" + echo "COS=1 transactions (avg): $cos1_jbd time (avg): $cos1_time" + [ "$cos0_jbd" != 0 ] && + echo "COS=1 vs COS=0 jbd: $((((cos1_jbd/cos0_jbd - 1)) * 100 )) %" + [ "$cos0_time" != 0 ] && + echo "COS=1 vs COS=0 time: $((((cos1_time/cos0_time - 1)) * 100 )) %" - restore_lustre_params < $param_file - rm -f $param_file - return 0 + restore_lustre_params < $param_file + rm -f $param_file + return 0 } run_test 33a "commit on sharing, cross crete/delete, 2 clients, benchmark" @@ -1037,13 +1097,17 @@ test_33b() { run_test 33b "COS: cross create/delete, 2 clients, benchmark under remote dir" test_33c() { - [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] && - skip "DNE CoS not supported" && return + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" + [ "$MDS1_VERSION" -lt $(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 @@ -1095,12 +1159,12 @@ op_trigger_cos() { } test_33d() { - [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] && - skip "DNE CoS not supported" && return + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" + [ "$MDS1_VERSION" -lt $(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" + op_trigger_cos "$LFS mkdir -i 0 $DIR/$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 @@ -1132,13 +1196,12 @@ test_33d() { run_test 33d "DNE distributed operation should trigger COS" test_33e() { - [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; } + [ -n "$CLIENTS" ] || skip "Need two or more clients" [ $CLIENTCOUNT -ge 2 ] || - { skip "Need two or more clients, have $CLIENTCOUNT" && - return 0; } - [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] && - skip "DNE CoS not supported" && return + skip "Need two or more clients, have $CLIENTCOUNT" + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" + [ "$MDS1_VERSION" -lt $(version_code 2.7.63) ] && + skip "DNE CoS not supported" local client2=${CLIENT2:-$(hostname)} @@ -1284,7 +1347,7 @@ test_35() { # bug 17645 for g in $gen; do if ! test "$g" -eq "${generation[count]}"; then list=$(lctl list_param mdc.$FSNAME-MDT*-mdc-*.import) - local c = 0 + local c=0 for imp in $list; do if [ $c = $count ]; then break @@ -1473,13 +1536,32 @@ test_39d() { # LU-7310 } run_test 39d "sync write should update mtime" +pdo_sched() { + # how long 40-47 take with specific delay + # sleep 0.1 # 78s + # sleep 0.2 # 103s + # sleep 0.3 # 124s + sleep 0.5 # 164s +} + +# for pdo testing, we must cancel MDT-MDT locks as well as client locks to +# avoid unexpected delays due to previous tests +pdo_lru_clear() { + cancel_lru_locks mdc + do_nodes $(comma_list $(mdts_nodes)) \ + $LCTL set_param -n ldlm.namespaces.*mdt*.lru_size=clear + do_nodes $(comma_list $(mdts_nodes)) \ + $LCTL get_param ldlm.namespaces.*mdt*.lock_unused_count \ + ldlm.namespaces.*mdt*.lock_count | grep -v '=0' +} + # check that pid exists hence second operation wasn't blocked by first one # if it is so then there is no conflict, return 0 # else second operation is conflicting with first one, return 1 check_pdo_conflict() { local pid=$1 local conflict=0 - sleep 1 # to ensure OP1 is finished on client if OP2 is blocked by OP1 + pdo_sched # to ensure OP1 is finished on client if OP2 is blocked by OP1 if [[ `ps --pid $pid | wc -l` == 1 ]]; then conflict=1 echo "Conflict" @@ -1493,120 +1575,139 @@ check_pdo_conflict() { # test 40: check non-blocking operations test_40a() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + + mkdir $DIR2/$tdir + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - touch $DIR2 - mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - touch $DIR2/$tfile-2 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" + mkdir $DIR1/$tdir/$tfile & + PID1=$!; pdo_sched + 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 check_pdo_conflict $PID1 || error "parallel operation is blocked" + 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 $DIR2/$tdir + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - touch $DIR1/$tfile & - PID1=$! - sleep 1 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" + 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 check_pdo_conflict $PID1 || error "parallel operation is blocked" + 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 - touch $DIR1/$tfile + + mkdir $DIR2/$tdir + pdo_lru_clear + touch $DIR1/$tdir/$tfile #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - link $DIR1/$tfile $DIR1/$tfile-0 & - PID1=$! - sleep 1 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" + 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 check_pdo_conflict $PID1 || error "parallel operation is blocked" + 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 - touch $DIR1/$tfile + + mkdir $DIR2/$tdir + pdo_lru_clear + touch $DIR1/$tdir/$tfile #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - rm $DIR1/$tfile & - PID1=$! - sleep 1 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" + 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 check_pdo_conflict $PID1 || error "parallel operation is blocked" + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0 2>/dev/null || true" wait $PID1 return 0 } @@ -1614,357 +1715,523 @@ run_test 40d "pdirops: unlink and others ==============" test_40e() { remote_mds_nodsh && skip "remote MDS with nodsh" && return - touch $DIR1/$tfile + + mkdir $DIR2/$tdir + pdo_lru_clear + touch $DIR1/$tdir/$tfile #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - mv $DIR1/$tfile $DIR1/$tfile-0 & - PID1=$! - sleep 1 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" + 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 check_pdo_conflict $PID1 || error "parallel operation is blocked" + 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 ==============" # test 41: create blocking operations test_41a() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + 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 & - PID1=$! - sleep 1 - mkdir $DIR2/$tfile && error "mkdir must fail" + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + 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; echo "mkdir isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail" rm -rf $DIR/$tfile* return 0 } run_test 41a "pdirops: create vs mkdir ==============" test_41b() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + 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 & - PID1=$! - sleep 1 - $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + PID1=$! ; pdo_sched + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c & + 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 "create isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "create must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + 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 & - PID1=$! - sleep 1 - link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" + PID1=$! ; pdo_sched + link $DIR2/$tfile-2 $DIR2/$tfile & + 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 "link isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "link must fail" rm -rf $DIR/$tfile* return 0 } run_test 41c "pdirops: create vs link ==============" test_41d() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + 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 & - PID1=$! - sleep 1 - rm $DIR2/$tfile || error "unlink must succeed" + PID1=$! ; pdo_sched + rm $DIR2/$tfile & + 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 "unlink isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "unlink must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + 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 & - PID1=$! - sleep 1 - mv $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile-2 $DIR2/$tfile & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } run_test 41e "pdirops: create and rename (tgt) ==============" test_41f() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + 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 & - PID1=$! - sleep 1 - mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-2 & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } run_test 41f "pdirops: create and rename (src) ==============" test_41g() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + 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 & - PID1=$! - sleep 1 - stat $DIR2/$tfile > /dev/null || error "stat must succeed" + PID1=$! ; pdo_sched + stat $DIR2/$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 } run_test 41g "pdirops: create vs getattr ==============" test_41h() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + 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 & - PID1=$! - sleep 1 - ls -lia $DIR2/ > /dev/null + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + PID2=$! ; 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 "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } run_test 41h "pdirops: create vs readdir ==============" +sub_test_41i() { + local PID1 PID2 + local fail_loc="$1" + local ret=0 + + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=${fail_loc} || true" &>/dev/null + + $MULTIOP $DIR1/$tfile oO_CREAT:O_EXCL:c 2>/dev/null & + PID1=$! + sleep 0.2 + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c 2>/dev/null & + PID2=$! + + if ! wait $PID1 && ! wait $PID2; then + echo "Both creates failed (1 should fail, 1 should succeed)" + ret=1 + elif wait $PID1 && wait $PID2; then + echo "Both creates succeeded (1 should fail, 1 should succeed)" + ret=2 + fi + + #Clean + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x0 || true" &>/dev/null + rm -f $DIR/$tfile + + return $ret +} + +test_41i() { + [[ $MDS1_VERSION -le $(version_code 2.13.56) ]] || + skip "Need MDS version newer than 2.13.56" + local msg fail_loc + +#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 +#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a + for fail_loc in "0x80000169" "0x8000016a"; do + echo "Begin 100 tests with fail_loc=$fail_loc" + printf "Progress: " + for i in {1..100}; do + printf "*" + msg=$(sub_test_41i "$fail_loc") || + { echo; error "iter=$i : $msg"; } + done + echo + done +} +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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - mkdir $DIR2/$tfile && error "mkdir must fail" + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + 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 "mkdir isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail" rm -rf $DIR/$tfile* return 0 } run_test 42a "pdirops: mkdir vs mkdir ==============" test_42b() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + PID1=$! ; pdo_sched + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c & + 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 "create isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "create must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" + PID1=$! ; pdo_sched + link $DIR2/$tfile-2 $DIR2/$tfile & + 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 "link isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "link must fail" rm -rf $DIR/$tfile* return 0 } run_test 42c "pdirops: mkdir vs link ==============" test_42d() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - rmdir $DIR2/$tfile || error "unlink must succeed" + PID1=$! ; pdo_sched + rmdir $DIR2/$tfile & + 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 "unlink isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "unlink must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - mv -T $DIR2/$tfile-2 $DIR2/$tfile && error "rename must fail" + PID1=$! ; pdo_sched + mv -T $DIR2/$tfile-2 $DIR2/$tfile & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "rename must fail" rm -rf $DIR/$tfile* return 0 } run_test 42e "pdirops: mkdir and rename (tgt) ==============" test_42f() { + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-2 & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - stat $DIR2/$tfile > /dev/null || error "stat must succeed" + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" + mkdir $DIR1/$tdir/$tfile & + PID1=$! ; pdo_sched + 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"; } - rm -rf $DIR/$tfile* - return 0 + wait $PID2 ; [ $? -eq 0 ] || error "stat must succeed" + 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mkdir $DIR1/$tfile & - PID1=$! - sleep 1 - ls -lia $DIR2/ > /dev/null + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + PID2=$! ; 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 "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } run_test 42h "pdirops: mkdir vs readdir ==============" -# test 43: unlink and blocking operations +# test 43: rmdir,mkdir won't return -EEXIST test_43a() { - touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - rm $DIR1/$tfile & - PID1=$! - sleep 1 - mkdir $DIR2/$tfile || error "mkdir must succeed" - check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -rf $DIR/$tfile* + for i in {1..1000}; do + mkdir $DIR1/$tdir || error "mkdir $tdir failed" + rmdir $DIR2/$tdir || error "rmdir $tdir failed" + done return 0 } -run_test 43a "pdirops: unlink vs mkdir ==============" +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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & - PID1=$! - sleep 1 - $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" + PID1=$! ; pdo_sched + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c & + 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 "create isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "create must succeed" rm -rf $DIR/$tfile* return 0 } run_test 43b "pdirops: unlink vs create ==============" test_43c() { + pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & - PID1=$! - sleep 1 - link $DIR2/$tfile-2 $DIR2/$tfile || error "link must succeed" + PID1=$! ; pdo_sched + link $DIR2/$tfile-2 $DIR2/$tfile & + 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 "link isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "link must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & - PID1=$! - sleep 1 - rm $DIR2/$tfile && error "unlink must fail" + PID1=$! ; pdo_sched + rm $DIR2/$tfile & + 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 "unlink isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "unlink must fail" rm -rf $DIR/$tfile* return 0 } run_test 43d "pdirops: unlink vs unlink ==============" test_43e() { + pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & - PID1=$! - sleep 1 - mv -u $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed" + PID1=$! ; pdo_sched + mv -u $DIR2/$tfile-2 $DIR2/$tfile & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & - PID1=$! - sleep 1 - mv $DIR2/$tfile $DIR2/$tfile-2 && error "rename must fail" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-2 & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "rename must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & - PID1=$! - sleep 1 - stat $DIR2/$tfile > /dev/null && error "stat must fail" + PID1=$! ; pdo_sched + stat $DIR2/$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 ; [ $? -ne 0 ] || error "stat must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & - PID1=$! - sleep 1 - ls -lia $DIR2/ > /dev/null + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + PID2=$! ; 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 "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -1972,132 +2239,257 @@ run_test 43h "pdirops: unlink vs readdir ==============" 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" rm $DIR1/$tfile & - PID1=$! - sleep 1 - $LFS mkdir -i 1 $DIR2/$tfile || error "remote mkdir must succeed" + PID1=$! ; pdo_sched + $LFS mkdir -i 1 $DIR2/$tfile & + 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 "remote mkdir isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "remote mkdir must succeed" rm -rf $DIR/$tfile* return 0 } run_test 43i "pdirops: unlink vs remote mkdir" +test_43j() { + [[ $MDS1_VERSION -lt $(version_code 2.13.52) ]] && + skip "Need MDS version newer than 2.13.52" + + mkdir $DIR1/$tdir + for i in {1..100}; do +#define OBD_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/sub & + PID1=$! + mkdir $DIR2/$tdir/sub && ((OK++)) + wait $PID1 && ((OK++)) + (( OK == 1 )) || error "exactly one mkdir should succeed" + + rmdir $DIR1/$tdir/sub || error "rmdir failed" + done + return 0 +} +run_test 43j "racy mkdir return EEXIST ==============" + +sub_test_43k() { + local PID1 PID2 + local fail_loc="$1" + local ret=0 + + # We test in a separate directory to be able to unblock server thread in + # cfs_race() if LCK_PW is taken on the parent by mdt_reint_unlink. + test_mkdir $DIR2/$tdir + touch $DIR2/$tdir/$tfile + + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=${fail_loc} || true" &>/dev/null + echo content > $DIR1/$tdir/$tfile & PID1=$! + pdo_sched + multiop $DIR2/$tdir/$tfile u & PID2=$! + + wait $PID1 || + { ret=$?; \ + echo -n "overwriting $tfile should succeed (err=$ret); "; } + wait $PID2 || + { ret=$?; \ + echo -n "unlinking $tfile should succeed (err=$ret);"; } + + #Clean + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x0 || true" &>/dev/null + rm -rf $DIR/$tdir + + return $ret +} + +test_43k() { + [[ $MDS1_VERSION -le $(version_code 2.13.56) ]] || + skip "Need MDS version newer than 2.13.56" + local msg fail_loc + +#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 +#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a + for fail_loc in "0x80000169" "0x8000016a"; do + echo "Begin 100 tests with fail_loc=$fail_loc" + printf "Progress: " + for i in {1..100}; do + printf "*" + msg=$(sub_test_43k "$fail_loc") || + { echo; error "iter=$i : $msg"; } + done + echo + done + + #Clean + reset_fail_loc + + return 0 +} +run_test 43k "unlink vs create" + # test 44: rename tgt and blocking operations test_44a() { + pdo_lru_clear touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - mkdir $DIR2/$tfile && error "mkdir must fail" - check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + 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; date;error "mkdir isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail" + date rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + PID1=$! ; pdo_sched + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c & + 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 "create isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "create must fail" rm -rf $DIR/$tfile* return 0 } run_test 44b "pdirops: rename tgt vs create ==============" test_44c() { + pdo_lru_clear touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - link $DIR2/$tfile-3 $DIR2/$tfile && error "link must fail" + PID1=$! ; pdo_sched + link $DIR2/$tfile-3 $DIR2/$tfile & + 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 "link isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "link must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - rm $DIR2/$tfile || error "unlink must succeed" + PID1=$! ; pdo_sched + rm $DIR2/$tfile & + 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 "unlink isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "unlink must succeed" rm -rf $DIR/$tfile* return 0 } run_test 44d "pdirops: rename tgt vs unlink ==============" test_44e() { + pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile-3 $DIR2/$tfile & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } run_test 44e "pdirops: rename tgt and rename (tgt) ==============" test_44f() { + pdo_lru_clear touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - mv $DIR2/$tfile $DIR2/$tfile-3 || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-3 & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - stat $DIR2/$tfile > /dev/null || error "stat must succeed" + PID1=$! ; pdo_sched + stat $DIR2/$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 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - ls -lia $DIR2/ > /dev/null + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + PID2=$! ; 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 "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -2106,130 +2498,166 @@ run_test 44h "pdirops: rename tgt vs readdir ==============" # test 44: rename tgt and blocking operations 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true" mv $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail" + PID1=$! ; pdo_sched + $LFS mkdir -i 1 $DIR2/$tfile & + 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 "remote mkdir isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "remote mkdir must fail" rm -rf $DIR/$tfile* return 0 } run_test 44i "pdirops: rename tgt vs remote mkdir" -# test 45: rename src and blocking operations +# test 45: rename,mkdir doesn't fail with -EEXIST test_45a() { - touch $DIR1/$tfile -#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - mkdir $DIR2/$tfile || error "mkdir must succeed" - check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } - rm -rf $DIR/$tfile* + for i in {1..1000}; do + mkdir $DIR1/$tdir || error "mkdir $tdir failed" + mrename $DIR2/$tdir $DIR2/$tdir.$i > /dev/null || + error "mrename to $tdir.$i failed" + done + rm -rf $DIR/$tdir* return 0 } -run_test 45a "pdirops: rename src vs mkdir ==============" +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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" + PID1=$! ; pdo_sched + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c & + 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 "create isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "create must succeed" rm -rf $DIR/$tfile* return 0 } run_test 45b "pdirops: rename src vs create ==============" test_45c() { + pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-3 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - link $DIR2/$tfile-3 $DIR2/$tfile || error "link must succeed" + PID1=$! ; pdo_sched + link $DIR2/$tfile-3 $DIR2/$tfile & + 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 "link isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "link must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - rm $DIR2/$tfile && error "unlink must fail" + PID1=$! ; pdo_sched + rm $DIR2/$tfile & + 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 "unlink isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "unlink must fail" rm -rf $DIR/$tfile* return 0 } run_test 45d "pdirops: rename src vs unlink ==============" test_45e() { + pdo_lru_clear touch $DIR1/$tfile touch $DIR1/$tfile-3 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile-3 $DIR2/$tfile & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - mv $DIR2/$tfile $DIR2/$tfile-3 && error "rename must fail" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-3 & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "rename must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - stat $DIR2/$tfile > /dev/null && error "stat must fail" + PID1=$! ; pdo_sched + stat $DIR2/$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 ; [ $? -ne 0 ] || error "stat must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - ls -lia $DIR2/ > /dev/null + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0 2>/dev/null || true" check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -2237,131 +2665,229 @@ run_test 45h "pdirops: unlink vs readdir ==============" 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" mv $DIR1/$tfile $DIR1/$tfile-2 & - PID1=$! - sleep 1 - $LFS mkdir -i 1 $DIR2/$tfile || error "create remote dir must succeed" + PID1=$! ; pdo_sched + $LFS mkdir -i 1 $DIR2/$tfile & + 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 "create remote dir isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "create remote dir must succeed" rm -rf $DIR/$tfile* return 0 } run_test 45i "pdirops: rename src vs remote mkdir" +sub_test_45j() { + local PID1 PID2 + local fail_loc="$1" + local ret=0 + + # We test in a sparate directory to be able to unblock server thread in + # cfs_race if LCK_PW is taken on the parent by mdt_reint_rename. + test_mkdir $DIR2/$tdir + echo file1 > $DIR2/$tdir/$tfile + echo file2 > $DIR2/$tdir/$tfile-2 + + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=${fail_loc} || true" &>/dev/null + + cat $DIR1/$tdir/$tfile >/dev/null & + PID1=$! + pdo_sched + mrename $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile > /dev/null & + PID2=$! + + wait $PID1 || + { ret=$?; echo -n "cat $tfile should succeed (err=$ret); "; } + wait $PID2 || + { ret=$?; \ + echo -n "mrename $tfile-2 to $tfile failed (err=$ret);"; } + + #Clean + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x0 || true" &>/dev/null + rm -rf $DIR/$tdir + + return $ret +} + +test_45j() { + [[ $MDS1_VERSION -le $(version_code 2.13.56) ]] || + skip "Need MDS version newer than 2.13.56" + local msg fail_loc + +#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169 +#define OBD_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a + for fail_loc in "0x80000169" "0x8000016a"; do + echo "Begin 100 tests with fail_loc=$fail_loc" + printf "Progress: " + for i in {1..100}; do + printf "*" + msg=$(sub_test_45j "$fail_loc") || + { echo; error "iter=$i : $msg"; } + done + echo + done +} +run_test 45j "read vs rename ==============" + # test 46: link and blocking operations test_46a() { + pdo_lru_clear touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - mkdir $DIR2/$tfile && error "mkdir must fail" + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + 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 "mkdir isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + PID1=$! ; pdo_sched + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c & + 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 "create isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "create must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - link $DIR2/$tfile $DIR2/$tfile && error "link must fail" + PID1=$! ; pdo_sched + link $DIR2/$tfile $DIR2/$tfile & + 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 "link isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "link must fail" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 - link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - rm $DIR2/$tfile || error "unlink must succeed" + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" + link $DIR1/$tfile-2 $DIR1/$tfile & + PID1=$! ; pdo_sched + rm $DIR2/$tfile & + 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 "unlink isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "unlink must succeed" rm -rf $DIR/$tfile* return 0 } run_test 46d "pdirops: link vs unlink ==============" test_46e() { + pdo_lru_clear touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile-3 $DIR2/$tfile & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } run_test 46e "pdirops: link and rename (tgt) ==============" test_46f() { + pdo_lru_clear touch $DIR1/$tfile-2 touch $DIR1/$tfile-3 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - mv $DIR2/$tfile $DIR2/$tfile-3 || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-3 & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - stat $DIR2/$tfile > /dev/null || error "stat must succeed" + PID1=$! ; pdo_sched + stat $DIR2/$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 } 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - ls -lia $DIR2/ > /dev/null - check_pdo_conflict $PID1 && { wait $PID1; - error "readdir isn't blocked"; } + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + PID2=$! ; 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 "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -2369,15 +2895,20 @@ run_test 46h "pdirops: link vs readdir ==============" 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" link $DIR1/$tfile-2 $DIR1/$tfile & - PID1=$! - sleep 1 - $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail" + PID1=$! ; pdo_sched + $LFS mkdir -i 1 $DIR2/$tfile & + 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 "remote mkdir isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "remote mkdir must fail" rm -rf $DIR/$tfile* return 0 } @@ -2387,12 +2918,17 @@ run_test 46i "pdirops: link vs remote mkdir" test_47a() { #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + pdo_lru_clear + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & - PID1=$! - sleep 1 - mkdir $DIR2/$tfile && error "mkdir must fail" + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + 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 "mkdir isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail" rm -rf $DIR/$tfile* return 0 } @@ -2401,13 +2937,19 @@ run_test 47a "pdirops: remote mkdir vs mkdir" test_47b() { #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + pdo_lru_clear + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & - PID1=$! - sleep 1 - multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + PID1=$! ; pdo_sched + sleep 1 # please do not remove this sleep, see LU-10754 + multiop $DIR2/$tfile oO_CREAT:O_EXCL:c & + 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 "create isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "create must fail" rm -rf $DIR/$tfile* return 0 } @@ -2415,14 +2957,19 @@ run_test 47b "pdirops: remote mkdir vs create" 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & - PID1=$! - sleep 1 - link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" + PID1=$! ; pdo_sched + link $DIR2/$tfile-2 $DIR2/$tfile & + 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 "link isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "link must fail" rm -rf $DIR/$tfile* return 0 } @@ -2430,14 +2977,19 @@ 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & - PID1=$! - sleep 1 - rmdir $DIR2/$tfile || error "unlink must succeed" + PID1=$! ; pdo_sched + rmdir $DIR2/$tfile & + 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 "unlink isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rmdir must succeed" rm -rf $DIR/$tfile* return 0 } @@ -2445,15 +2997,20 @@ run_test 47d "pdirops: remote mkdir vs unlink" 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & - PID1=$! - sleep 1 - mv -T $DIR2/$tfile-2 $DIR2/$tfile && error "rename must fail" + PID1=$! ; pdo_sched + mv -T $DIR2/$tfile-2 $DIR2/$tfile & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -ne 0 ] || error "rename must fail" rm -rf $DIR/$tfile* return 0 } @@ -2461,14 +3018,19 @@ 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 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & - PID1=$! - sleep 1 - mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-2 & + 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 "rename isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed" rm -rf $DIR/$tfile* return 0 } @@ -2476,14 +3038,21 @@ run_test 47f "pdirops: remote mkdir and rename (src)" test_47g() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + sync + sync_all_data + pdo_lru_clear #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 - do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + do_nodes $(comma_list $(mdts_nodes)) \ + "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true" $LFS mkdir -i 1 $DIR1/$tfile & - PID1=$! - sleep 1 - stat $DIR2/$tfile > /dev/null || error "stat must succeed" + PID1=$! ; pdo_sched + stat $DIR2/$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 } @@ -2536,29 +3105,32 @@ test_51a() { run_test 51a "layout lock: refresh layout should work" test_51b() { - [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.59) ]] || - { skip "Need MDS version at least 2.3.59"; return 0; } + [[ "$MDS1_VERSION" -ge $(version_code 2.3.59) ]] || + skip "Need MDS version at least 2.3.59" local tmpfile=`mktemp` - # create an empty file - $MCREATE $DIR1/$tfile || error "mcreate $DIR1/$tfile failed" + $LFS setstripe -E 1M -c 1 -E -1 --extension-size 64M $DIR1/$tfile || + error "Create $DIR1/$tfile failed" + + dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc || + error "dd $DIR1/$tfile failed" # delay glimpse so that layout has changed when glimpse finish #define OBD_FAIL_GLIMPSE_DELAY 0x1404 - $LCTL set_param fail_loc=0x1404 + $LCTL set_param fail_loc=0x1404 fail_val=4 stat -c %s $DIR2/$tfile |tee $tmpfile & local pid=$! - sleep 1 + sleep 0.2 - # create layout of testing file - dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc >/dev/null || + # extend layout of testing file + dd if=/dev/zero of=$DIR1/$tfile bs=1M count=1 seek=2 conv=notrunc || error "dd $DIR1/$tfile failed" wait $pid local fsize=$(cat $tmpfile) - [ x$fsize = x1024 ] || error "file size is $fsize, should be 1024" + [ x$fsize = x3145728 ] || error "file size is $fsize, should be 3145728" rm -f $DIR1/$tfile $tmpfile } @@ -2625,6 +3197,30 @@ 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 & + pid=$! + sleep 1 + + $LFS getstripe $DIR2/$tfile + kill -USR1 $pid + wait $pid || error "multiop failed" + + $MULTIOP $DIR/$tfile oO_RDONLY:eR_E+eUc & + pid=$! + sleep 1 + + $LFS getstripe $DIR2/$tfile + kill -USR1 $pid + wait $pid || error "multiop failed" +} +run_test 51e "lfs getstripe does not break leases, part 2" + test_54_part1() { echo "==> rename vs getattr vs setxattr should not deadlock" @@ -2694,56 +3290,59 @@ 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 @@ -2752,35 +3351,36 @@ 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/ wait $PID1 && error "(2) mv succeeded" - rm -rf $DIR/f1 + rm -rf $DIR/$tdir } run_test 55d "rename file vs link" test_60() { - local MDSVER=$(lustre_build_version $SINGLEMDS) - [ $(version_code $MDSVER) -lt $(version_code 2.3.0) ] && - skip "MDS version $MDSVER must be >= 2.3.0" && return 0 + [ $MDS1_VERSION -lt $(version_code 2.3.0) ] && + skip "MDS version must be >= 2.3.0" # Create a file test_mkdir $DIR1/$tdir @@ -2866,22 +3466,19 @@ test_70b() { # LU-2781 run_test 70b "remove files after calling rm_entry" test_71a() { - local server_version=$(lustre_version_code $SINGLEMDS) - - [[ $server_version -lt $(version_code 2.1.6) ]] && - skip "Need MDS version at least 2.1.6" && return + [[ "$MDS1_VERSION" -lt $(version_code 2.1.6) ]] && + skip "Need MDS version at least 2.1.6" # Patch not applied to 2.2 and 2.3 branches - [[ $server_version -ge $(version_code 2.2.0) ]] && - [[ $server_version -lt $(version_code 2.4.0) ]] && - skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" && - return + [[ "$MDS1_VERSION" -ge $(version_code 2.2.0) ]] && + [[ "$MDS1_VERSION" -lt $(version_code 2.4.0) ]] && + skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" checkfiemap --test || - { skip "checkfiemap not runnable: $?" && return; } + skip "checkfiemap not runnable: $?" # write data this way: hole - data - hole - data dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1 - [ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tfile) + 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 @@ -2910,26 +3507,23 @@ test_71a() { run_test 71a "correct file map just after write operation is finished" test_71b() { - local server_version=$(lustre_version_code $SINGLEMDS) - - [[ $server_version -lt $(version_code 2.1.6) ]] && - skip "Need MDS version at least 2.1.6" && return + [[ "$MDS1_VERSION" -lt $(version_code 2.1.6) ]] && + skip "Need MDS version at least 2.1.6" # Patch not applied to 2.2 and 2.3 branches - [[ $server_version -ge $(version_code 2.2.0) ]] && - [[ $server_version -lt $(version_code 2.4.0) ]] && - skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" && - return - [[ $OSTCOUNT -ge 2 ]] || { skip "needs >= 2 OSTs"; return; } + [[ "$MDS1_VERSION" -ge $(version_code 2.2.0) ]] && + [[ "$MDS1_VERSION" -lt $(version_code 2.4.0) ]] && + skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" + [[ $OSTCOUNT -ge 2 ]] || skip "needs >= 2 OSTs" checkfiemap --test || - { skip "error $?: checkfiemap failed" && return; } + skip "error $?: checkfiemap failed" mkdir -p $DIR1/$tdir $LFS setstripe -c -1 $DIR1/$tdir || error "setstripe failed" dd if=/dev/urandom of=$DIR1/$tdir/$tfile bs=40K count=1 - [ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tdir/$tfile) + 1)))" = \ + [ "$(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" @@ -2997,8 +3591,8 @@ test_73() { run_test 73 "getxattr should not cause xattr lock cancellation" test_74() { - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.93) ] && - skip "Need MDS version at least 2.4.93" && return + [ "$MDS1_VERSION" -lt $(version_code 2.4.93) ] && + skip "Need MDS version at least 2.4.93" dd if=/dev/zero of=$DIR1/$tfile-1 bs=1K count=1 dd if=/dev/zero of=$DIR1/$tfile-2 bs=1K count=1 @@ -3029,10 +3623,10 @@ test_75() { run_test 75 "osc: upcall after unuse lock===================" test_76() { #LU-946 - [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.53) ]] && - skip "Need MDS version at least 2.5.53" && return + [[ "$MDS1_VERSION" -lt $(version_code 2.5.53) ]] && + skip "Need MDS version at least 2.5.53" - remote_mds_nodsh && skip "remote MDS with nodsh" && return + remote_mds_nodsh && skip "remote MDS with nodsh" local fcount=$((MDSCOUNT * 256)) declare -a fd_list declare -a fid_list @@ -3122,31 +3716,47 @@ nrs_write_read() { chmod 777 $dir do_nodes $CLIENTS $myRUNAS \ - dd if=/dev/zero of="$dir/nrs_r_$HOSTNAME" bs=1M count=$n || + dd if=/dev/zero of="$dir/nrs_r_\$HOSTNAME" bs=1M count=$n || error "dd at 0 on client failed (1)" - for ((i = 0; i < $n; i++)); do - do_nodes $CLIENTS $myRUNAS dd if=/dev/zero \ - of="$dir/nrs_w_$HOSTNAME" bs=1M seek=$i count=1 || - error "dd at ${i}MB on client failed (2)" & - local pids_w[$i]=$! - done + do_nodes $CLIENTS $myRUNAS \ + "declare -a pids_w; + for ((i = 0; i < $n; i++)); do + dd if=/dev/zero of=$dir/nrs_w_\$HOSTNAME bs=1M \ +seek=\\\$i count=1 conv=notrunc & + pids_w[\\\$i]=\\\$!; + done; + rc_w=0; + for ((i = 0; i < $n; i++)); do + wait \\\${pids_w[\\\$i]}; + newrc=\\\$?; + [ \\\$newrc -gt \\\$rc_w ] && rc_w=\\\$newrc; + done; + exit \\\$rc_w" & + local pid_w=$! do_nodes $CLIENTS sync; cancel_lru_locks osc - for ((i = 0; i < $n; i++)); do - do_nodes $CLIENTS $myRUNAS dd if="$dir/nrs_w_$HOSTNAME" \ - of=/dev/zero bs=1M seek=$i count=1 > /dev/null || - error "dd at ${i}MB on client failed (3)" & - local pids_r[$i]=$! - done + do_nodes $CLIENTS $myRUNAS \ + "declare -a pids_r; + for ((i = 0; i < $n; i++)); do + dd if=$dir/nrs_r_\$HOSTNAME bs=1M of=/dev/null \ +seek=\\\$i count=1 & + pids_r[\\\$i]=\\\$!; + done; + rc_r=0; + for ((i = 0; i < $n; i++)); do + wait \\\${pids_r[\\\$i]}; + newrc=\\\$?; + [ \\\$newrc -gt \\\$rc_r ] && rc_r=\\\$newrc; + done; + exit \\\$rc_r" & + local pid_r=$! cancel_lru_locks osc - for ((i = 0; i < $n; i++)); do - wait ${pids_w[$i]} - wait ${pids_r[$i]} - done - rm -rf $dir || error "rm -rf $dir failed" + wait $pid_w || error "dd (write) failed (2)" + wait $pid_r || error "dd (read) failed (3)" + rm -rvf $dir || error "rm -rf $dir failed" } test_77a() { #LU-3266 @@ -3323,7 +3933,7 @@ test_77e() { local idis local rateis - if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + if [ "$OST1_VERSION" -ge $(version_code 2.8.54) ]; then idis="nid=" rateis="rate=" fi @@ -3379,7 +3989,7 @@ test_77f() { local idis local rateis - if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + if [ "$OST1_VERSION" -ge $(version_code 2.8.54) ]; then idis="jobid=" rateis="rate=" fi @@ -3434,7 +4044,7 @@ test_77g() { local idis local rateis - if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then + if [ "$OST1_VERSION" -ge $(version_code 2.8.54) ]; then idis="jobid=" rateis="rate=" fi @@ -3451,8 +4061,8 @@ test_77g() { run_test 77g "Change TBF type directly" test_77h() { - [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] || - { skip "Need OST version at least 2.8.55"; return 0; } + [ "$OST1_VERSION" -ge $(version_code 2.8.55) ] || + skip "Need OST version at least 2.8.55" local old_policy=$(do_facet ost1 \ lctl get_param ost.OSS.ost_io.nrs_policies) @@ -3506,8 +4116,8 @@ tbf_rule_check() } test_77i() { - [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] || - { skip "Need OST version at least 2.8.55"; return 0; } + [ "$OST1_VERSION" -ge $(version_code 2.8.55) ] || + skip "Need OST version at least 2.8.55" for i in $(seq 1 $OSTCOUNT) do @@ -3561,11 +4171,10 @@ run_test 77i "Change rank of TBF rule" test_77j() { local idis local rateis - local ost_version=$(lustre_version_code ost1) - [ $ost_version -ge $(version_code 2.9.53) ] || - { skip "Need OST version at least 2.9.53"; return 0; } - if [ $ost_version -ge $(version_code 2.8.60) ]; then + [ "$OST1_VERSION" -ge $(version_code 2.9.53) ] || + skip "Need OST version at least 2.9.53" + if [ "$OST1_VERSION" -ge $(version_code 2.8.60) ]; then idis="opcode=" rateis="rate=" fi @@ -3617,10 +4226,10 @@ test_id() { } test_77ja(){ - if [ $(lustre_version_code ost1) -lt $(version_code 2.11.50) ]; then + if [ "$OST1_VERSION" -lt $(version_code 2.11.50) ]; then skip "Need OST version at least 2.11.50" - return 0 fi + test_id "u" "500" "5" "-u 500" test_id "g" "500" "5" "-u 500 -g 500" } @@ -3644,8 +4253,8 @@ cleanup_77k() } test_77k() { - [[ $(lustre_version_code ost1) -ge $(version_code 2.9.53) ]] || - { skip "Need OST version at least 2.9.53"; return 0; } + [[ "$OST1_VERSION" -ge $(version_code 2.9.53) ]] || + skip "Need OST version at least 2.9.53" do_nodes $(comma_list $(osts_nodes)) \ lctl set_param ost.OSS.ost_io.nrs_policies="tbf" \ @@ -3687,8 +4296,8 @@ test_77k() { trap "cleanup_77k \"ext_a ext_b\" \"fifo\"" EXIT - [[ $(lustre_version_code ost1) -ge $(version_code 2.10.58) ]] || - { skip "Need OST version at least 2.10.58"; return 0; } + [[ "$OST1_VERSION" -ge $(version_code 2.10.58) ]] || + skip "Need OST version at least 2.10.58" do_nodes $(comma_list $(osts_nodes)) \ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_a" \ @@ -3718,8 +4327,8 @@ test_77k() { run_test 77k "check TBF policy with NID/JobID/OPCode expression" test_77l() { - [[ $(lustre_version_code ost1) -ge $(version_code 2.10.56) ]] || - { skip "Need OST version at least 2.10.56"; return 0; } + [[ "$OST1_VERSION" -ge $(version_code 2.10.56) ]] || + skip "Need OST version at least 2.10.56" do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf" @@ -3738,9 +4347,8 @@ test_77l() { run_test 77l "check the output of NRS policies for generic TBF" test_77m() { - if [ $(lustre_version_code ost1) -lt $(version_code 2.9.54) ]; then + if [ "$OST1_VERSION" -lt $(version_code 2.9.54) ]; then skip "Need OST version at least 2.9.54" - return 0 fi local dir=$DIR/$tdir @@ -3790,9 +4398,8 @@ test_77m() { run_test 77m "check NRS Delay slows write RPC processing" test_77n() { #LU-10802 - if [ $(lustre_version_code ost1) -lt $(version_code 2.10.58) ]; then + if [ "$OST1_VERSION" -lt $(version_code 2.10.58) ]; then skip "Need OST version at least 2.10.58" - return 0 fi # Configure jobid_var @@ -4102,9 +4709,55 @@ test_81b() { } run_test 81b "rename under striped directory doesn't deadlock" +test_81c() { + [ $MDSCOUNT -lt 4 ] && skip_env "needs >= 4 MDTs" + [ $MDS1_VERSION -lt $(version_code 2.13.52) ] && + skip "Need MDS version at least 2.13.52" + + # source is local, source parent is remote + $LFS mkdir -i 0 $DIR1/${tdir}_src || error "mkdir ${tdir}_src" + $LFS mkdir -i 1 $DIR1/${tdir}_tgt || error "mkdir ${tdir}_tgt" + $LFS mkdir -i 3 $DIR1/${tdir}_src/sub || error "mkdir sub" + $LFS mkdir -i 3 $DIR1/${tdir}_tgt/sub || error "mkdir sub" + stat $DIR2/${tdir}_src/sub || error "stat sub failed" + mv $DIR1/${tdir}_src/sub $DIR1/${tdir}_tgt/ || error "mv failed" + [ -f $DIR2/${tdir}_src/sub ] && error "sub should be gone" + rm -rf $DIR1/${tdir}_src $DIR1/${tdir}_tgt + + # source is remote, source parent is local + $LFS mkdir -i 3 $DIR1/${tdir}_src || error "mkdir ${tdir}_src" + $LFS mkdir -i 1 $DIR1/${tdir}_tgt || error "mkdir ${tdir}_tgt" + $LFS mkdir -i 0 $DIR1/${tdir}_src/sub || error "mkdir sub" + $LFS mkdir -i 3 $DIR1/${tdir}_tgt/sub || error "mkdir sub" + stat $DIR2/${tdir}_src/sub || error "stat sub failed" + mv $DIR1/${tdir}_src/sub $DIR1/${tdir}_tgt/ || error "mv failed" + [ -f $DIR2/${tdir}_src/sub ] && error "sub should be gone" + rm -rf $DIR1/${tdir}_src $DIR1/${tdir}_tgt + + # source and source parent are remote + $LFS mkdir -i 0 $DIR1/${tdir}_src || error "mkdir ${tdir}_src" + $LFS mkdir -i 1 $DIR1/${tdir}_tgt || error "mkdir ${tdir}_tgt" + mkdir $DIR1/${tdir}_src/sub || error "mkdir sub" + $LFS mkdir -i 3 $DIR1/${tdir}_tgt/sub || error "mkdir sub" + stat $DIR2/${tdir}_src/sub || error "stat sub failed" + mv $DIR1/${tdir}_src/sub $DIR1/${tdir}_tgt/ || error "mv failed" + [ -f $DIR2/${tdir}_src/sub ] && error "sub should be gone" + rm -rf $DIR1/${tdir}_src $DIR1/${tdir}_tgt + + # source and source parent are remote, and source is remote object + $LFS mkdir -i 0 $DIR1/${tdir}_src || error "mkdir ${tdir}_src" + $LFS mkdir -i 1 $DIR1/${tdir}_tgt || error "mkdir ${tdir}_tgt" + $LFS mkdir -i 2 $DIR1/${tdir}_src/sub || error "mkdir sub" + $LFS mkdir -i 3 $DIR1/${tdir}_tgt/sub || error "mkdir sub" + stat $DIR2/${tdir}_src/sub || error "stat sub failed" + mv $DIR1/${tdir}_src/sub $DIR1/${tdir}_tgt/ || error "mv failed" + [ -f $DIR2/${tdir}_src/sub ] && error "sub should be gone" || true +} +run_test 81c "rename revoke LOOKUP lock for remote object" + test_82() { - [[ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.6.91) ]] || - { skip "Need MDS version at least 2.6.92"; return 0; } + [[ "$MDS1_VERSION" -gt $(version_code 2.6.91) ]] || + skip "Need MDS version at least 2.6.92" # Client 1 creates a file. multiop_bg_pause $DIR1/$tfile O_ac || error "multiop_bg_pause 1" @@ -4159,6 +4812,31 @@ test_83() { } run_test 83 "access striped directory while it is being created/unlinked" +test_84() { + [ $MDS1_VERSION -lt $(version_code 2.12.55) ] && + skip "lustre < 2.12.55 does not contain LU-12485 fix" + + local mtime + + $MULTIOP $DIR/$tfile oO_RDWR:O_CREAT:O_LOV_DELAY_CREATE:c || + error "create $tfile failed" + mtime=$(stat -c%Y $DIR/$tfile) + mtime=$((mtime + 200)) + + #define OBD_FAIL_OBD_0NLINK_RACE 0x60b + do_facet mds1 $LCTL set_param fail_loc=0x8000060b + + touch -c -m $mtime $DIR/$tfile & + setattr_pid=$! + # sleep a while to let 'touch' run first + sleep 5 + rm -f $DIR2/$tfile || error "unlink $tfile failed" + + # touch may fail + wait $setattr_pid || true +} +run_test 84 "0-nlink race in lu_object_find()" + test_90() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return local pid1 @@ -4263,29 +4941,29 @@ test_93() { mkdir -p $DIR1/$tfile-1/ mkdir -p $DIR2/$tfile-2/ - local old_rr=$(do_facet $SINGLEMDS lctl get_param -n \ - 'lod.lustre-MDT*/qos_threshold_rr' | sed -e 's/%//') - do_facet $SINGLEMDS lctl set_param -n \ - 'lod.lustre-MDT*/qos_threshold_rr' 100 + local old_rr=$(do_facet $SINGLEMDS "$LCTL get_param -n \ + lod.$FSNAME-MDT0000-*/qos_threshold_rr" | sed -e 's/%//') + do_facet $SINGLEMDS "$LCTL set_param -n \ + lod.$FSNAME-MDT0000-*/qos_threshold_rr=100" #define OBD_FAIL_MDS_LOV_CREATE_RACE 0x163 - do_facet $SINGLEMDS "lctl set_param fail_loc=0x00000163" + do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x00000163" - $SETSTRIPE -c -1 $DIR1/$tfile-1/file1 & + $LFS setstripe -c -1 $DIR1/$tfile-1/file1 & local PID1=$! sleep 1 - $SETSTRIPE -c -1 $DIR2/$tfile-2/file2 & + $LFS setstripe -c -1 $DIR2/$tfile-2/file2 & local PID2=$! wait $PID2 wait $PID1 - do_facet $SINGLEMDS "lctl set_param fail_loc=0x0" - do_facet $SINGLEMDS "lctl set_param -n \ - 'lod.lustre-MDT*/qos_threshold_rr' $old_rr" + do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x0" + do_facet $SINGLEMDS "$LCTL set_param -n \ + lod.$FSNAME-MDT0000-*/qos_threshold_rr=$old_rr" - $GETSTRIPE $DIR1/$tfile-1/file1 - rc1=$($GETSTRIPE -q $DIR1/$tfile-1/file1 | + $LFS getstripe $DIR1/$tfile-1/file1 + rc1=$($LFS getstripe -q $DIR1/$tfile-1/file1 | awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l) - $GETSTRIPE $DIR2/$tfile-2/file2 - rc2=$($GETSTRIPE -q $DIR2/$tfile-2/file2 | + $LFS getstripe $DIR2/$tfile-2/file2 + rc2=$($LFS getstripe -q $DIR2/$tfile-2/file2 | awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l) echo "rc1=$rc1 and rc2=$rc2 " [ $rc1 -eq 0 ] && [ $rc2 -eq 0 ] || @@ -4293,11 +4971,47 @@ test_93() { } run_test 93 "alloc_rr should not allocate on same ost" +test_94() { + $LCTL set_param osc.*.idle_timeout=0 + dd if=/dev/zero of=$DIR2/$tfile bs=4k count=2 conv=fsync + + local before=$(date +%s) + local evict + + $LCTL mark write +#define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312 + $LCTL set_param fail_val=5 fail_loc=0x80000312 + dd if=/dev/zero of=$DIR/$tfile conv=notrunc oflag=append bs=4k count=1 & + local pid=$! + sleep 2 + +#define OBD_FAIL_LDLM_PAUSE_CANCEL_LOCAL 0x329 + $LCTL set_param fail_val=6 fail_loc=0x80000329 + $LCTL mark kill $pid + kill -ALRM $pid + + dd if=/dev/zero of=$DIR2/$tfile conv=notrunc oflag=append bs=4k count=1 + + wait $pid + dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 conv=fsync + + evict=$(do_facet client $LCTL get_param \ + osc.$FSNAME-OST*-osc-*/state | + awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }') + + [ -z "$evict" ] || [[ $evict -le $before ]] || + (do_facet client $LCTL get_param \ + osc.$FSNAME-OST*-osc-*/state; + error "eviction happened: $evict before:$before") + $LCTL set_param osc.*.idle_timeout=debug +} +run_test 94 "signal vs CP callback race" + # Data-on-MDT tests test_100a() { skip "Reserved for glimpse-ahead" && return - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] && - skip "Need MDS version at least 2.10.55" && return + [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] && + skip "Need MDS version at least 2.10.55" mkdir -p $DIR/$tdir @@ -4320,8 +5034,8 @@ test_100a() { run_test 100a "DoM: glimpse RPCs for stat without IO lock (DoM only file)" test_100b() { - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] && - skip "Need MDS version at least 2.10.55" && return + [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] && + skip "Need MDS version at least 2.10.55" mkdir -p $DIR/$tdir @@ -4343,8 +5057,8 @@ test_100b() { run_test 100b "DoM: no glimpse RPC for stat with IO lock (DoM only file)" test_100c() { - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] && - skip "Need MDS version at least 2.10.55" && return + [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] && + skip "Need MDS version at least 2.10.55" mkdir -p $DIR/$tdir @@ -4366,8 +5080,8 @@ test_100c() { run_test 100c "DoM: write vs stat without IO lock (combined file)" test_100d() { - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] && - skip "Need MDS version at least 2.10.55" && return + [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] && + skip "Need MDS version at least 2.10.55" mkdir -p $DIR/$tdir @@ -4390,9 +5104,31 @@ test_100d() { } run_test 100d "DoM: write+truncate vs stat without IO lock (combined file)" +test_100e() { + [ "$MDS1_VERSION" -lt $(version_code 2.11.50) ] && + skip "Need MDS version at least 2.11.50" + + local dom=$DIR/$tdir/dom + local dom2=$DIR2/$tdir/dom + mkdir -p $DIR/$tdir + + $LFS setstripe -E 1024K -L mdt $DIR/$tdir + + cancel_lru_locks mdc + dd if=/dev/urandom of=$dom bs=12000 count=1 + $TRUNCATE $dom2 6000 + cancel_lru_locks mdc + lctl set_param -n mdc.*.stats=clear + # expect read-on-open to return all data before write + cat /etc/hosts >> $dom + local read=$(lctl get_param -n mdc.*.stats | grep -c ost_read) + [[ $read -eq 0 ]] || error "Unexpected $read READ RPCs" +} +run_test 100e "DoM: read on open and file size" + test_101a() { - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] && - skip "Need MDS version at least 2.10.55" && return + [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] && + skip "Need MDS version at least 2.10.55" $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile # to get layout @@ -4416,8 +5152,8 @@ test_101a() { run_test 101a "Discard DoM data on unlink" test_101b() { - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] && - skip "Need MDS version at least 2.10.55" && return + [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] && + skip "Need MDS version at least 2.10.55" $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile touch $DIR1/${tfile}_2 @@ -4441,8 +5177,8 @@ test_101b() { run_test 101b "Discard DoM data on rename" test_101c() { - [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] && - skip "Need MDS version at least 2.10.55" && return + [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] && + skip "Need MDS version at least 2.10.55" $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile # to get layout @@ -4475,14 +5211,484 @@ 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) ] && + 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" +# Compare file size between first & second mount, ensuring the client correctly +# glimpses even with unused speculative locks - LU-11670 +test_103() { + [ $OST1_VERSION -lt $(version_code 2.10.50) ] && + skip "Lockahead needs OST version at least 2.10.50" + + local locktest=23 + + test_mkdir -p $DIR/$tdir + + # Force file on to OST0 + $LFS setstripe -i 0 $DIR/$tdir + + # Do not check multiple locks on glimpse + # OBD_FAIL_OSC_NO_SIZE_DATA 0x415 + $LCTL set_param fail_loc=0x415 + + # Delay write commit by 2 seconds to guarantee glimpse wins race + # The same fail_loc is used on client & server so it can work in the + # single node sanity setup + do_facet ost1 $LCTL set_param fail_loc=0x415 fail_val=2 + + echo "Incorrect size expected (no glimpse fix):" + lockahead_test -d $DIR/$tdir -D $DIR2/$tdir -t $locktest -f $tfile + rc=$? + if [ $rc -eq 0 ]; then + echo "This doesn't work 100%, but this is just reproducing the bug, not testing the fix, so OK to not fail test." + fi + + # guarantee write commit timeout has expired + sleep 2 + + # Clear fail_loc on client + $LCTL set_param fail_loc=0 + + # Delay write commit by 2 seconds to guarantee glimpse wins race + # OBD_FAIL_OST_BRW_PAUSE_BULK 0x214 + do_facet ost1 $LCTL set_param fail_loc=0x214 fail_val=2 + + # Write commit is still delayed by 2 seconds + lockahead_test -d $DIR/$tdir -D $DIR2/$tdir -t $locktest -f $tfile + rc=$? + [ $rc -eq 0 ] || error "Lockahead test$locktest failed, $rc" + + # guarantee write commit timeout has expired + sleep 2 + + rm -f $DIR/$tfile || error "unable to delete $DIR/$tfile" +} +run_test 103 "Test size correctness with lockahead" + +get_stat_xtimes() +{ + local xtimes + + xtimes=$(stat -c "%X %Y %Z" $DIR/$tfile) + + echo ${xtimes[*]} +} + +get_mdt_xtimes() +{ + local mdtdev=$1 + local output + local xtimes + + output=$(do_facet mds1 "$DEBUGFS -c -R 'stat ROOT/$tfile' $mdtdev") + ((xtimes[0]=$(awk -F ':' /atime/'{ print $2 }' <<< "$output"))) + ((xtimes[1]=$(awk -F ':' /mtime/'{ print $2 }' <<< "$output"))) + ((xtimes[2]=$(awk -F ':' /ctime/'{ print $2 }' <<< "$output"))) + + echo ${xtimes[*]} +} + +check_mdt_xtimes() +{ + local mdtdev=$1 + local xtimes=($(get_stat_xtimes)) + local mdt_xtimes=($(get_mdt_xtimes $mdtdev)) + + echo "STAT a|m|ctime ${xtimes[*]}" + echo "MDT a|m|ctime ${mdt_xtimes[*]}" + [[ ${xtimes[0]} == ${mdt_xtimes[0]} ]] || + error "$DIR/$tfile atime (${xtimes[0]}:${mdt_xtimes[0]}) diff" + [[ ${xtimes[1]} == ${mdt_xtimes[1]} ]] || + error "$DIR/$tfile mtime (${xtimes[1]}:${mdt_xtimes[1]}) diff" + [[ ${xtimes[2]} == ${mdt_xtimes[2]} ]] || + error "$DIR/$tfile ctime (${xtimes[2]}:${mdt_xtimes[2]}) diff" +} + +test_104() { + [ "$mds1_FSTYPE" == "ldiskfs" ] || skip_env "ldiskfs only test" + [ $MDS1_VERSION -lt $(version_code 2.12.4) ] && + skip "Need MDS version at least 2.12.4" + + local pid + local mdtdev=$(mdsdevname ${SINGLEMDS//mds/}) + local atime_diff=$(do_facet $SINGLEMDS \ + lctl get_param -n mdd.*MDT0000*.atime_diff) + + do_facet $SINGLEMDS \ + lctl set_param -n mdd.*MDT0000*.atime_diff=0 + + stack_trap "do_facet $SINGLEMDS \ + lctl set_param -n mdd.*MDT0000*.atime_diff=$atime_diff" EXIT + + dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc + check_mdt_xtimes $mdtdev + sleep 2 + + dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc + check_mdt_xtimes $mdtdev + sleep 2 + $MULTIOP $DIR2/$tfile Oz8192w8192_c & + pid=$! + sleep 2 + dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc + sleep 2 + kill -USR1 $pid && wait $pid || error "multiop failure" + check_mdt_xtimes $mdtdev + + local xtimes + local mdt_xtimes + + # Verify mtime/ctime is NOT upated on MDS when there is no modification + # on the client side + xtimes=($(get_stat_xtimes)) + $MULTIOP $DIR/$tfile O_c & + pid=$! + sleep 2 + kill -USR1 $pid && wait $pid || error "multiop failure" + mdt_xtimes=($(get_mdt_xtimes $mdtdev)) + [[ ${xtimes[1]} == ${mdt_xtimes[1]} ]] || + error "$DIR/$tfile mtime (${xtimes[1]}:${mdt_xtimes[1]}) diff" + [[ ${xtimes[2]} == ${mdt_xtimes[2]} ]] || + error "$DIR/$tfile ctime (${xtimes[2]}:${mdt_xtimes[2]}) diff" + check_mdt_xtimes $mdtdev + + sleep 2 + # Change ctime via chmod + $MULTIOP $DIR/$tfile o_tc & + pid=$! + sleep 2 + kill -USR1 $pid && wait $pid || error "multiop failure" + check_mdt_xtimes $mdtdev +} +run_test 104 "Verify that MDS stores atime/mtime/ctime during close" + +test_105() { + test_mkdir -p $DIR/$tdir + echo test > $DIR/$tdir/$tfile + $LCTL set_param fail_loc=0x416 + cancel_lru_locks osc & sleep 1 + fsize1=$(stat -c %s $DIR2/$tdir/$tfile) + wait + [[ $fsize1 = 5 ]] || error "Glimpse returned wrong file size $fsize1" +} +run_test 105 "Glimpse and lock cancel race" + +test_106a() { + [ "$mds1_FSTYPE" == "ldiskfs" ] && statx_supported || + skip_env "Test only for ldiskfs and statx() supported" + + local btime + local mdt_btime + local output + local mdtdev=$(mdsdevname ${SINGLEMDS//mds/}) + + dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc + btime=$($STATX -c %W $DIR/$tfile) + output=$(do_facet mds1 "$DEBUGFS -c -R 'stat ROOT/$tfile' $mdtdev") + echo $output + ((mdt_btime=$(awk -F ':' /crtime/'{ print $2 }' <<< "$output"))) + [[ $btime == $mdt_btime ]] || + error "$DIR/$tfile btime ($btime:$mdt_btime) diff" + +} +run_test 106a "Verify the btime via statx()" + +test_106b() { + statx_supported || skip_env "statx() only test" + + local rpcs_before + local rpcs_after + + $LFS setstripe -c 1 $DIR/$tfile || error "$DIR/$tfile setstripe failed" + dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc + cancel_lru_locks $OSC + rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + $STATX $DIR/$tfile + rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + [ $rpcs_after -eq $((rpcs_before + 1)) ] || + error "$STATX should send 1 glimpse RPC to $OSC" + + cancel_lru_locks $OSC + rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + # %n: FILENAME; %i: STATX_INO; %A STATX_MODE; %h STATX_NLINK; + # %u: STATX_UID; %g: STATX_GID; %W STATX_BTIME; %X STATX_ATIME; + # %Z: STATX_CTIME + $STATX -c "%n %i %A %h %u %g %W %X %Z" $DIR/$tfile + rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + [ $rpcs_after -eq $rpcs_before ] || + error "$STATX should not send glimpse RPCs to $OSC" + + cancel_lru_locks $OSC + rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + $STATX --cached=always $DIR/$tfile + rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + [ $rpcs_after -eq $rpcs_before ] || + error "$STATX should not send glimpse RPCs to $OSC" + + cancel_lru_locks $OSC + rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + $STATX -c %Y $DIR/$tfile + rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + [ $rpcs_after -eq $((rpcs_before + 1)) ] || + error "$STATX -c %Y should send 1 glimpse RPC to $OSC" + + cancel_lru_locks $OSC + rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + $STATX -c %s $DIR/$tfile + rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + [ $rpcs_after -eq $((rpcs_before + 1)) ] || + error "$STATX -c %s should send 1 glimpse RPC to $OSC" + + cancel_lru_locks $OSC + rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + $STATX -c %b $DIR/$tfile + rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + [ $rpcs_after -eq $((rpcs_before + 1)) ] || + error "$STATX -c %b should send 1 glimpse RPC to $OSC" +} +run_test 106b "Glimpse RPCs test for statx" + +test_106c() { + statx_supported || skip_env "statx() only test" + + local mask + + touch $DIR/$tfile + # 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'" + chattr +i $DIR/$tfile || error "chattr +i $DIR/$tfile failed" + mask=$($STATX -c %r $DIR/$tfile) + [[ $mask == "10" ]] || + error "got immutable flags '$mask', expected '10'" + chattr -i $DIR/$tfile || error "chattr -i $DIR/$tfile failed" + mask=$($STATX -c %r $DIR/$tfile) + [[ $mask == "0" ]] || error "got flags '$mask', expected '0'" + chattr +a $DIR/$tfile || error "chattr +a $DIR/$tfile failed" + mask=$($STATX -c %r $DIR/$tfile) + [[ $mask == "20" ]] || error "got flags '$mask', expected '20'" + chattr -a $DIR/$tfile || error "chattr -a $DIR/$tfile failed" + mask=$($STATX -c %r $DIR/$tfile) + [[ $mask == "0" ]] || error "got flags '$mask', expected '0'" + chattr +ia $DIR/$tfile || error "chattr +ia $DIR/$tfile failed" + mask=$($STATX -c %r $DIR/$tfile) + [[ $mask == "30" ]] || error "got flags '$mask', expected '30'" + chattr -ia $DIR/$tfile || error "chattr -ia $DIR/$tfile failed" + mask=$($STATX -c %r $DIR/$tfile) + [[ $mask == "0" ]] || error "got flags '$mask', expected '0'" +} +run_test 106c "Verify statx attributes mask" + +test_107a() { # LU-1031 + dd if=/dev/zero of=$DIR1/$tfile bs=1M count=10 + local gid1=14091995 + local gid2=16022000 + + $LFS getstripe $DIR1/$tfile + + multiop_bg_pause $DIR1/$tfile OG${gid1}_g${gid1}c || return 1 + local MULTIPID1=$! + multiop_bg_pause $DIR2/$tfile O_G${gid2}r10g${gid2}c || return 2 + local MULTIPID2=$! + kill -USR1 $MULTIPID2 + sleep 2 + if [[ $(ps h -o comm -p $MULTIPID2) == "" ]]; then + error "First grouplock does not block second one" + else + echo "First grouplock blocks second one" + fi + kill -USR1 $MULTIPID1 + wait $MULTIPID1 + wait $MULTIPID2 +} +run_test 107a "Basic grouplock conflict" + +test_107b() { + dd if=/dev/zero of=$DIR1/$tfile bs=1M count=10 + local gid1=14091995 + local gid2=16022000 + + $LFS getstripe $DIR1/$tfile + + multiop_bg_pause $DIR1/$tfile OG${gid1}_g${gid1}c || return 1 + local MULTIPID1=$! + multiop $DIR2/$tfile Or10c & + local MULTIPID2=$! + sleep 2 + + if [[ $(ps h -o comm -p $MULTIPID2) == "" ]]; then + error "Grouplock does not block IO" + else + echo "Grouplock blocks IO" + fi + + multiop $DIR2/$tfile OG${gid2}_g${gid2}c & + local MULTIPID3=$! + sleep 2 + if [[ $(ps h -o comm -p $MULTIPID3) == "" ]]; then + error "First grouplock does not block second one" + else + echo "First grouplock blocks second one" + fi + + kill -USR1 $MULTIPID1 + sleep 2 + + if [[ $(ps h -o comm -p $MULTIPID3) == "" ]]; then + error "Second grouplock thread disappeared" + fi + + if [[ $(ps h -o comm -p $MULTIPID2) == "" ]]; then + error "Second grouplock does not block IO" + else + echo "Second grouplock blocks IO" + fi + + kill -USR1 $MULTIPID3 + wait $MULTIPID1 + wait $MULTIPID2 + wait $MULTIPID3 +} +run_test 107b "Grouplock is added to the head of waiting list" + +test_108a() { + local offset + + $LFS setstripe -E 1M -c 1 -E -1 $DIR1/$tfile || + error "Create $DIR1/$tfile failed" + + dd if=/dev/zero of=$DIR1/$tfile bs=10000 count=1 || + error "dd $DIR1/$tfile failed" + offset=$(lseek_test -d 5000 $DIR2/$tfile) + [[ $offset == 5000 ]] || error "offset $offset != 5000" + + $TRUNCATE $DIR1/$tfile 2000 + offset=$(lseek_test -l 1000 $DIR2/$tfile) + [[ $offset == 2000 ]] || error "offset $offset != 2000" + + #define OBD_FAIL_OSC_DELAY_IO 0x414 + $LCTL set_param fail_val=4 fail_loc=0x80000414 + dd if=/dev/zero of=$DIR1/$tfile count=1 bs=8M conv=notrunc oflag=dsync & + local pid=$! + sleep 2 + + offset=$(lseek_test -l 8000 $DIR2/$tfile) + wait $pid + [[ $offset == 8388608 ]] || error "offset $offset != 8388608" +} +run_test 108a "lseek: parallel updates" + +# LU-14110 +test_109() { + local i + local pid1 pid2 + + ! local_mode || + skip "Clients need to be on different nodes than the servers" + + umount_client $MOUNT + umount_client $MOUNT2 + + echo "Starting race between client mount instances (50 iterations):" + for i in {1..50}; do + log "Iteration $i" + +#define OBD_FAIL_ONCE|OBD_FAIL_LLITE_RACE_MOUNT 0x80001417 + $LCTL set_param -n fail_loc=0x80001417 + + mount_client $MOUNT & pid1=$! + mount_client $MOUNT2 & pid2=$! + wait $pid1 || error "Mount $MOUNT fails with $?" + wait $pid2 || error "Mount $MOUNT2 fails with $?" + + umount_client $MOUNT & pid1=$! + umount_client $MOUNT2 & pid2=$! + wait $pid1 || error "Umount $MOUNT fails with $?" + wait $pid2 || error "Umount $MOUNT2 fails with $?" + + $LUSTRE_RMMOD || error "Fail to remove lustre modules" + load_modules + echo + done + + mount_client $MOUNT + mount_client $MOUNT2 +} + +run_test 109 "Race with several mount instances on 1 node" + +test_110() { + local before=$(date +%s) + local evict + + mkdir -p $DIR/$tdir + touch $DIR/$tdir/f1 + touch $DIR/$tfile + + #define OBD_FAIL_PTLRPC_RESEND_RACE 0x525 + do_facet mds1 lctl set_param fail_loc=0x525 fail_val=3 + + # disable last_xid logic by dropping link reply + ln $DIR/$tdir/f1 $DIR/$tdir/f2 & + sleep 1 + + #define OBD_FAIL_PTLRPC_ENQ_RESEND 0x534 + do_facet mds1 lctl set_param fail_loc=0x534 + + # RPC will race with its Resend and the Resend will sleep to let + # the original lock to get granted & cancelled. + # + # AST_SENT is set artificially, so an explicit conflict is not needed + # + # The woken up Resend gets a new lock, but client does not wait for it + stat $DIR/$tfile + sleep $TIMEOUT + do_facet mds1 lctl set_param fail_loc=0 fail_val=0 + + # Take a conflict to wait long enough to see the eviction + touch $DIR2/$tfile + + # let the client reconnect + client_reconnect + evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state | + awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }') + + [ -z "$evict" ] || [[ $evict -le $before ]] || + (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state; + error "eviction happened: $evict before:$before") +} +run_test 110 "do not grant another lock on resend" + log "cleanup: ======================================================" # kill and wait in each test only guarentee script finish, but command in script