X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanityn.sh;h=a4006aa95f9aad6c66cea701370e66e43de6e984;hp=249844ad9d281f3f64bc9066f54e19063825dab0;hb=85ac0da36748311528aa7b89f62679f087b7445c;hpb=3cbe63e72c83ba50c3553d28e2fe6bc9d18c347a diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh index 249844a..a4006aa 100755 --- a/lustre/tests/sanityn.sh +++ b/lustre/tests/sanityn.sh @@ -3,8 +3,9 @@ set -e ONLY=${ONLY:-"$*"} -# bug number for skipped test: 9977/LU-7105 LU-9452 -ALWAYS_EXCEPT=" 28 29 $SANITYN_EXCEPT" +# 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 $0) @@ -30,6 +31,7 @@ CLEANUP=${CLEANUP:-:} SETUP=${SETUP:-:} init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} +get_lustre_env init_logging if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then @@ -84,7 +86,8 @@ test_2a() { touch $DIR1/f2a ls -l $DIR2/f2a chmod 777 $DIR2/f2a - $CHECKSTAT -t file -p 0777 $DIR1/f2a || error + $CHECKSTAT -t file -p 0777 $DIR1/f2a || + error "Either not file type or perms not 0777" } run_test 2a "check cached attribute updates on 2 mtpt's ========" @@ -92,20 +95,23 @@ test_2b() { touch $DIR1/f2b ls -l $DIR2/f2b chmod 777 $DIR1/f2b - $CHECKSTAT -t file -p 0777 $DIR2/f2b || error + $CHECKSTAT -t file -p 0777 $DIR2/f2b || + error "Either not file type or perms not 0777" } run_test 2b "check cached attribute updates on 2 mtpt's ========" # NEED TO SAVE ROOT DIR MODE test_2c() { chmod 777 $DIR1 - $CHECKSTAT -t dir -p 0777 $DIR2 || error + $CHECKSTAT -t dir -p 0777 $DIR2 || + error "Either not dir type or perms not 0777" } run_test 2c "check cached attribute updates on 2 mtpt's root ===" test_2d() { chmod 755 $DIR1 - $CHECKSTAT -t dir -p 0755 $DIR2 || error + $CHECKSTAT -t dir -p 0755 $DIR2 || + error "Either not file type or perms not 0775" } run_test 2d "check cached attribute updates on 2 mtpt's root ===" @@ -114,7 +120,8 @@ test_2e() { ls -l $DIR1 ls -l $DIR2 chmod 777 $DIR1 - $RUNAS dd if=/dev/zero of=$DIR2/$tfile count=1 || error + $RUNAS dd if=/dev/zero of=$DIR2/$tfile count=1 || + error "dd failed" } run_test 2e "check chmod on root is propagated to others" @@ -161,7 +168,7 @@ test_2g() { local block2=$(stat $DIR2/$tfile | awk '/Blocks/ {print $4} ') echo "$DIR1/$tfile has $block1 blocks" echo "$DIR2/$tfile has $block2 blocks" - [ $block1 -eq $block2 ] || error + [ $block1 -eq $block2 ] || error "$block1 not equal to $block2" } run_test 2g "check blocks update on sync write" @@ -181,7 +188,8 @@ run_test 4 "fstat validation on multiple mount points ==========" test_5() { mcreate $DIR1/f5 $TRUNCATE $DIR2/f5 100 - $CHECKSTAT -t file -s 100 $DIR1/f5 || error + $CHECKSTAT -t file -s 100 $DIR1/f5 || + error "Either not file type or size not equal to 100 bytes" rm $DIR1/f5 } run_test 5 "create a file on one mount, truncate it on the other" @@ -205,7 +213,7 @@ test_8() { } run_test 8 "remove of open special file on other node ==========" -test_9() { +test_9a() { MTPT=1 local dir > $DIR2/f9 @@ -217,7 +225,23 @@ test_9() { [ "`cat $DIR1/f9`" = "abcdefghijkl" ] || \ error "`od -a $DIR1/f9` != abcdefghijkl" } -run_test 9 "append of file with sub-page size on multiple mounts" +run_test 9a "append of file with sub-page size on multiple mounts" + +#LU-10681 - tiny writes & appending to sparse striped file +test_9b() { + [[ $OSTCOUNT -ge 2 ]] || { skip "needs >= 2 OSTs"; return; } + + $LFS setstripe -c 2 -S 1M $DIR/$tfile + echo "foo" >> $DIR/$tfile + dd if=/dev/zero of=$DIR2/$tfile bs=1M count=1 seek=1 conv=notrunc || + error "sparse dd $DIR2/$tfile failed" + echo "foo" >> $DIR/$tfile + + data=$(dd if=$DIR2/$tfile bs=1 count=3 skip=$((2 * 1048576)) conv=notrunc) + echo "Data read (expecting 'foo')": $data + [ "$data" = "foo" ] || error "append to sparse striped file failed" +} +run_test 9b "append to striped sparse file" test_10a() { MTPT=1 @@ -260,7 +284,7 @@ test_11() { $DIR2/d11/f RC=$? kill -USR1 $MULTIPID - wait $MULTIPID || error + wait $MULTIPID || error "wait for PID $MULTIPID failed" [ $RC -eq 0 ] && error || true } run_test 11 "execution of file opened for write should return error ====" @@ -272,16 +296,16 @@ run_test 12 "test lock ordering (link, stat, unlink)" test_13() { # bug 2451 - directory coherency test_mkdir $DIR1/d13 - cd $DIR1/d13 || error + cd $DIR1/d13 || error "cd to $DIR1/d13 failed" ls ( touch $DIR1/d13/f13 ) # needs to be a separate shell ls - rm -f $DIR2/d13/f13 || error + rm -f $DIR2/d13/f13 || error "Cannot remove $DIR2/d13/f13" ls 2>&1 | grep f13 && error "f13 shouldn't return an error (1)" || true # need to run it twice ( touch $DIR1/d13/f13 ) # needs to be a separate shell ls - rm -f $DIR2/d13/f13 || error + rm -f $DIR2/d13/f13 || error "Cannot remove $DIR2/d13/f13" ls 2>&1 | grep f13 && error "f13 shouldn't return an error (2)" || true } run_test 13 "test directory page revocation" @@ -293,65 +317,52 @@ test_14aa() { MULTIPID=$! $DIR2/$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 ========" @@ -379,14 +390,16 @@ else FSXP=100 fi -test_16() { +test_16a() { local file1=$DIR1/$tfile local file2=$DIR2/$tfile + 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 $file1 - dd if=/dev/zero of=$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 + dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync rm -f $file1 $LFS setstripe -c -1 $file1 # b=10919 @@ -398,7 +411,62 @@ test_16() { fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 -Z -r 4096 \ -w 4096 $file1 $file2 || error "fsx with O_DIRECT failed." } -run_test 16 "$FSXNUM iterations of dual-mount fsx" +run_test 16a "$FSXNUM iterations of dual-mount fsx" + +# Consistency check for tiny writes, LU-9409 +test_16b() { + local file1=$DIR1/$tfile + local file2=$DIR2/$tfile + + # 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 + 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 \ + $file2 +} +run_test 16b "$FSXNUM iterations of dual-mount fsx at small size" + +test_16c() { + local file1=$DIR1/$tfile + local file2=$DIR2/$tfile + local stripe_size=$(do_facet $SINGLEMDS \ + "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize") + + [ $(facet_fstype ost1) != ldiskfs ] && skip "dio on ldiskfs only" + + # 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 + dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync + rm -f $file1 + wait_delete_completed + + local list=$(comma_list $(osts_nodes)) + if ! get_osd_param $list '' read_cache_enable >/dev/null; then + skip "not cache-capable obdfilter" + fi + + set_osd_param $list '' read_cache_enable 0 + 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" + rm -f $file1 + + set_osd_param $list '' read_cache_enable 1 + set_osd_param $list '' writethrough_cache_enable 1 + + return 0 +} +run_test 16c "verify data consistency on ldiskfs with cache disabled (b=17397)" + test_17() { # bug 3513, 3667 remote_ost_nodsh && skip "remote OST with nodsh" && return @@ -437,13 +505,10 @@ test_19() { # bug3811 [ "x$DOM" = "xyes" ] && node=$(facet_active_host $SINGLEMDS) # check whether obdfilter is cache capable at all - if ! get_osd_param $node '' read_cache_enable >/dev/null; then - echo "not cache-capable obdfilter" - return 0 - fi + get_osd_param $node '' read_cache_enable >/dev/null || + skip "not cache-capable obdfilter" - local MAX=$(get_osd_param $node '' readcache_max_filesize | \ - head -n 1) + local MAX=$(get_osd_param $node '' readcache_max_filesize | head -n 1) set_osd_param $node '' readcache_max_filesize 4096 dd if=/dev/urandom of=$TMP/$tfile bs=512k count=32 local SUM=$(cksum $TMP/$tfile | cut -d" " -f 1,2) @@ -606,7 +671,7 @@ test_25b() { run_test 25b "change ACL under remote dir on one mountpoint be seen on another" test_26a() { - utime $DIR1/f26a -s $DIR2/f26a || error + utime $DIR1/f26a -s $DIR2/f26a || error "utime failed for $DIR1/f26a" } run_test 26a "allow mtime to get older" @@ -670,16 +735,17 @@ test_28() { # bug 9977 EOF # reading of 1st stripe should pass - dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error + dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error "dd failed" # reading of 2nd stripe should fail (this stripe was destroyed) dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 && error # now, recreating test file - dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 || error + dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 || error "dd failed" # reading of 1st stripe should pass - dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error + dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error "dd failed" # reading of 2nd stripe should pass - dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 || error + dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 || + error "dd failed" } run_test 28 "read/write/truncate file with lost stripes" @@ -736,14 +802,18 @@ enable_lockless_truncate() { } test_32a() { # bug 11270 - local p="$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 lfs setstripe -c -1 $DIR1/$tfile - dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$STRIPE_BYTES > \ + dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$stripe_size > \ /dev/null 2>&1 clear_stats $OSC.*.${OSC}_stats @@ -766,10 +836,7 @@ test_32a() { # bug 11270 $CHECKSTAT -s 3000000 $DIR1/$tfile || error "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" @@ -961,39 +1028,57 @@ test_33c() { [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] && skip "DNE CoS not supported" && return - sync + local sync_count mkdir $DIR/$tdir - # remote mkdir is done on MDT2, which enqueued lock of $tdir on MDT1 - $LFS mkdir -i 1 $DIR/$tdir/d1 + sync_all_data do_facet mds1 "lctl set_param -n mdt.*.sync_count=0" - mkdir $DIR/$tdir/d2 - local sync_count=$(do_facet mds1 \ - "lctl get_param -n mdt.*MDT0000.sync_count") - [ $sync_count -eq 1 ] || error "Sync-Lock-Cancel not triggered" + # do twice in case transaction is committed before unlock, see LU-8200 + for i in 1 2; do + # remote dir is created on MDT1, which enqueued lock of $tdir on + # MDT0 + $LFS mkdir -i 1 $DIR/$tdir/remote.$i + mkdir $DIR/$tdir/local.$i + done + sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count") + echo "sync_count $sync_count" + [ $sync_count -eq 0 ] && error "Sync-Lock-Cancel not triggered" - $LFS mkdir -i 1 $DIR/$tdir/d3 + sync_all_data do_facet mds1 "lctl set_param -n mdt.*.sync_count=0" + $LFS mkdir -i 1 $DIR/$tdir/remote.3 # during sleep remote mkdir should have been committed and canceled # remote lock spontaneously, which shouldn't trigger sync sleep 6 - mkdir $DIR/$tdir/d4 - local sync_count=$(do_facet mds1 \ - "lctl get_param -n mdt.*MDT0000.sync_count") + mkdir $DIR/$tdir/local.3 + sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count") + echo "sync_count $sync_count" [ $sync_count -eq 0 ] || error "Sync-Lock-Cancel triggered" } run_test 33c "Cancel cross-MDT lock should trigger Sync-Lock-Cancel" -ops_do_cos() { +# arg1 is operations done before CoS, arg2 is the operation that triggers CoS +op_trigger_cos() { + local commit_nr + local total=0 local nodes=$(comma_list $(mdts_nodes)) - do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0" - sh -c "$@" - local async_commit_count=$(do_nodes $nodes \ - "lctl get_param -n mdt.*.async_commit_count" | calc_sum) - [ $async_commit_count -gt 0 ] || error "CoS not triggerred" - rm -rf $DIR/$tdir - sync + sync_all_data + + # trigger CoS twice in case transaction commit before unlock + for i in 1 2; do + sh -c "$1" + do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0" + sh -c "$2" + commit_nr=$(do_nodes $nodes \ + "lctl get_param -n mdt.*.async_commit_count" | calc_sum) + total=$((total + commit_nr)); + rm -rf $DIR/$tdir + sync_all_data + done + + echo "CoS count $total" + [ $total -gt 0 ] || error "$2 didn't trigger CoS" } test_33d() { @@ -1001,39 +1086,34 @@ test_33d() { [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] && skip "DNE CoS not supported" && return - sync # remote directory create - mkdir $DIR/$tdir - ops_do_cos "$LFS mkdir -i 1 $DIR/$tdir/subdir" + op_trigger_cos "mkdir $DIR/$tdir" "$LFS mkdir -i 1 $DIR/$tdir/subdir" # remote directory unlink - $LFS mkdir -i 1 $DIR/$tdir - ops_do_cos "rmdir $DIR/$tdir" + op_trigger_cos "$LFS mkdir -i 1 $DIR/$tdir" "rmdir $DIR/$tdir" # striped directory create - mkdir $DIR/$tdir - ops_do_cos "$LFS mkdir -c 2 $DIR/$tdir/subdir" + op_trigger_cos "mkdir $DIR/$tdir" "$LFS mkdir -c 2 $DIR/$tdir/subdir" # striped directory setattr - $LFS mkdir -c 2 $DIR/$tdir - touch $DIR/$tdir - ops_do_cos "chmod 713 $DIR/$tdir" + op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; touch $DIR/$tdir" \ + "chmod 713 $DIR/$tdir" # striped directory unlink - $LFS mkdir -c 2 $DIR/$tdir - touch $DIR/$tdir - ops_do_cos "rmdir $DIR/$tdir" + op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; touch $DIR/$tdir" \ + "rmdir $DIR/$tdir" # cross-MDT link - $LFS mkdir -c 2 $DIR/$tdir - $LFS mkdir -i 0 $DIR/$tdir/d1 - $LFS mkdir -i 1 $DIR/$tdir/d2 - touch $DIR/$tdir/d1/tgt - ops_do_cos "ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src" + op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; \ + $LFS mkdir -i 0 $DIR/$tdir/d1; \ + $LFS mkdir -i 1 $DIR/$tdir/d2; \ + touch $DIR/$tdir/d1/tgt" \ + "ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src" # cross-MDT rename - $LFS mkdir -c 2 $DIR/$tdir - $LFS mkdir -i 0 $DIR/$tdir/d1 - $LFS mkdir -i 1 $DIR/$tdir/d2 - touch $DIR/$tdir/d1/src - ops_do_cos "mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt" + op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; \ + $LFS mkdir -i 0 $DIR/$tdir/d1; \ + $LFS mkdir -i 1 $DIR/$tdir/d2; \ + touch $DIR/$tdir/d1/src" \ + "mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt" # migrate - $LFS mkdir -i 0 $DIR/$tdir - ops_do_cos "$LFS migrate -m 1 $DIR/$tdir" + op_trigger_cos "$LFS mkdir -i 0 $DIR/$tdir" \ + "$LFS migrate -m 1 $DIR/$tdir" + return 0 } run_test 33d "DNE distributed operation should trigger COS" @@ -1085,7 +1165,7 @@ cleanup_34() { do_nodes $(comma_list $(osts_nodes)) \ "lctl set_param -n fail_loc=0 2>/dev/null || true" for i in $(seq $OSTCOUNT); do - wait_osc_import_state client ost$i FULL + wait_osc_import_ready client ost$i done } @@ -1191,7 +1271,7 @@ 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 @@ -1380,13 +1460,21 @@ 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 +} + # 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" @@ -1400,12 +1488,12 @@ check_pdo_conflict() { # test 40: check non-blocking operations test_40a() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + cancel_lru_locks mdc #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=$! - sleep 1 + PID1=$!; pdo_sched touch $DIR2/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" mkdir $DIR2/$tfile-3 @@ -1422,6 +1510,7 @@ test_40a() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" + do_facet $SINGLEMDS lctl set_param fail_loc=0 wait $PID1 rm -rf $DIR/$tfile* return 0 @@ -1430,11 +1519,11 @@ run_test 40a "pdirops: create vs others ==============" test_40b() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 touch $DIR1/$tfile & - PID1=$! - sleep 1 + PID1=$!; pdo_sched # open|create touch $DIR2/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" @@ -1452,6 +1541,7 @@ test_40b() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" + do_facet $SINGLEMDS lctl set_param fail_loc=0 wait $PID1 rm -rf $DIR/$tfile* return 0 @@ -1460,12 +1550,12 @@ run_test 40b "pdirops: open|create and others ==============" test_40c() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + cancel_lru_locks mdc touch $DIR1/$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=$! - sleep 1 + PID1=$!; pdo_sched # open|create touch $DIR2/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" @@ -1483,6 +1573,7 @@ test_40c() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" + do_facet $SINGLEMDS lctl set_param fail_loc=0 wait $PID1 rm -rf $DIR/$tfile* return 0 @@ -1491,12 +1582,12 @@ run_test 40c "pdirops: link and others ==============" test_40d() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + cancel_lru_locks mdc 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=$! - sleep 1 + PID1=$!; pdo_sched # open|create touch $DIR2/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" @@ -1514,6 +1605,7 @@ test_40d() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" + do_facet $SINGLEMDS lctl set_param fail_loc=0 wait $PID1 return 0 } @@ -1521,12 +1613,12 @@ run_test 40d "pdirops: unlink and others ==============" test_40e() { remote_mds_nodsh && skip "remote MDS with nodsh" && return + cancel_lru_locks mdc 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-0 & - PID1=$! - sleep 1 + PID1=$!; pdo_sched # open|create touch $DIR2/$tfile-2 check_pdo_conflict $PID1 || error "create is blocked" @@ -1542,6 +1634,7 @@ test_40e() { # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" + do_facet $SINGLEMDS lctl set_param fail_loc=0 wait $PID1 rm -rf $DIR/$tfile* return 0 @@ -1550,106 +1643,130 @@ run_test 40e "pdirops: rename and others ==============" # test 41: create blocking operations test_41a() { + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -1657,106 +1774,130 @@ run_test 41h "pdirops: create vs readdir ==============" # test 42: unlink and blocking operations test_42a() { + cancel_lru_locks mdc #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 - mkdir $DIR2/$tfile && error "mkdir must fail" + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #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 - $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #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 - link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" + PID1=$! ; pdo_sched + link $DIR2/$tfile-2 $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #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 - rmdir $DIR2/$tfile || error "unlink must succeed" + PID1=$! ; pdo_sched + rmdir $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #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 - 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #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 - mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-2 & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc #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" + PID1=$! ; pdo_sched + stat $DIR2/$tfile > /dev/null & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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 42g "pdirops: mkdir vs getattr ==============" test_42h() { + cancel_lru_locks mdc #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 - ls -lia $DIR2/ > /dev/null + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -1764,114 +1905,138 @@ run_test 42h "pdirops: mkdir vs readdir ==============" # test 43: unlink and blocking operations test_43a() { + cancel_lru_locks mdc 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" + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "mkdir must succeed" rm -rf $DIR/$tfile* return 0 } run_test 43a "pdirops: unlink vs mkdir ==============" test_43b() { + cancel_lru_locks mdc 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 - $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - rm $DIR2/$tfile && error "unlink must fail" + PID1=$! ; pdo_sched + rm $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - mv $DIR2/$tfile $DIR2/$tfile-2 && error "rename must fail" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-2 & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - stat $DIR2/$tfile > /dev/null && error "stat must fail" + PID1=$! ; pdo_sched + stat $DIR2/$tfile > /dev/null & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - ls -lia $DIR2/ > /dev/null + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -1879,15 +2044,18 @@ run_test 43h "pdirops: unlink vs readdir ==============" test_43i() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + cancel_lru_locks mdc 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 - $LFS mkdir -i 1 $DIR2/$tfile || error "remote mkdir must succeed" + PID1=$! ; pdo_sched + $LFS mkdir -i 1 $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -1895,116 +2063,141 @@ run_test 43i "pdirops: unlink vs remote mkdir" # test 44: rename tgt and blocking operations test_44a() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 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_facet $SINGLEMDS lctl set_param fail_loc=0 + 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() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 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_facet $SINGLEMDS lctl set_param fail_loc=0 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -2013,15 +2206,18 @@ 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 + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146 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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2029,114 +2225,137 @@ run_test 44i "pdirops: rename tgt vs remote mkdir" # test 45: rename src and blocking operations test_45a() { + cancel_lru_locks mdc 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" + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } + wait $PID2 ; [ $? -eq 0 ] || error "mkdir must succeed" rm -rf $DIR/$tfile* return 0 } run_test 45a "pdirops: rename src vs mkdir ==============" test_45b() { + cancel_lru_locks mdc 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 - $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - rm $DIR2/$tfile && error "unlink must fail" + PID1=$! ; pdo_sched + rm $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - mv $DIR2/$tfile $DIR2/$tfile-3 && error "rename must fail" + PID1=$! ; pdo_sched + mv $DIR2/$tfile $DIR2/$tfile-3 & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - stat $DIR2/$tfile > /dev/null && error "stat must fail" + PID1=$! ; pdo_sched + stat $DIR2/$tfile > /dev/null & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - ls -lia $DIR2/ > /dev/null + PID1=$! ; pdo_sched + ls -lia $DIR2/ > /dev/null & + do_facet $SINGLEMDS lctl set_param fail_loc=0 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -2144,15 +2363,18 @@ run_test 45h "pdirops: unlink vs readdir ==============" test_45i() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + cancel_lru_locks mdc 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 - $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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2160,115 +2382,138 @@ run_test 45i "pdirops: rename src vs remote mkdir" # test 46: link and blocking operations test_46a() { + cancel_lru_locks mdc 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 - mkdir $DIR2/$tfile && error "mkdir must fail" + PID1=$! ; pdo_sched + mkdir $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - $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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - link $DIR2/$tfile $DIR2/$tfile && error "link must fail" + PID1=$! ; pdo_sched + link $DIR2/$tfile $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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" + PID1=$! ; pdo_sched + rm $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 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_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - stat $DIR2/$tfile > /dev/null || error "stat must succeed" + PID1=$! ; pdo_sched + stat $DIR2/$tfile > /dev/null & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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() { + cancel_lru_locks mdc 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 - 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_facet $SINGLEMDS lctl set_param fail_loc=0 + check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } + wait $PID2 rm -rf $DIR/$tfile* return 0 } @@ -2276,15 +2521,18 @@ run_test 46h "pdirops: link vs readdir ==============" test_46i() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + cancel_lru_locks mdc 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 - $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail" + PID1=$! ; pdo_sched + $LFS mkdir -i 1 $DIR2/$tfile & + PID2=$! ; pdo_sched + do_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2292,14 +2540,18 @@ run_test 46i "pdirops: link vs remote mkdir" # test 47: remote mkdir and blocking operations test_47a() { + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + cancel_lru_locks mdc do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2308,13 +2560,17 @@ 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 + cancel_lru_locks mdc do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2322,14 +2578,17 @@ run_test 47b "pdirops: remote mkdir vs create" test_47c() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2337,14 +2596,17 @@ run_test 47c "pdirops: remote mkdir vs link" test_47d() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2352,15 +2614,18 @@ run_test 47d "pdirops: remote mkdir vs unlink" test_47e() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + cancel_lru_locks mdc touch $DIR1/$tfile-2 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2368,14 +2633,17 @@ run_test 47e "pdirops: remote mkdir and rename (tgt)" test_47f() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2383,14 +2651,19 @@ run_test 47f "pdirops: remote mkdir and rename (src)" test_47g() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + sync + sync_all_data + cancel_lru_locks mdc #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 $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_facet $SINGLEMDS lctl set_param fail_loc=0 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 } @@ -2519,14 +2792,14 @@ test_51d() { local br=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}') echo "Before revoking layout lock: $br KB mapped" - # delete the file will revoke layout lock - rm -f $DIR2/$tfile + # cancel layout lock manually + cancel_lru_locks mdc # rss after revoking local ar=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}') kill -USR1 $PID - wait $PID || error + wait $PID || error "wait PID $PID failed" [ $ar -eq 0 ] || error "rss before: $br, after $ar, some pages remained" } @@ -2788,7 +3061,7 @@ test_71a() { { skip "checkfiemap not runnable: $?" && return; } # write data this way: hole - data - hole - data dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1 - [ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tfile) + 1)))" = \ + [ "$(facet_fstype ost$(($($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 @@ -2836,7 +3109,7 @@ test_71b() { $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" @@ -3029,36 +3302,57 @@ 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 $dir failed" } test_77a() { #LU-3266 + local rc + oss=$(comma_list $(osts_nodes)) - do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="fifo" + do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="fifo" || + rc=$? + [[ $rc -eq 3 ]] && skip "no NRS exists" && return + [[ $rc -ne 0 ]] && error "failed to set fifo policy" nrs_write_read return 0 @@ -3066,21 +3360,29 @@ test_77a() { #LU-3266 run_test 77a "check FIFO NRS policy" test_77b() { #LU-3266 + local rc + oss=$(comma_list $(osts_nodes)) do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="crrn" \ - ost.OSS.*.nrs_crrn_quantum=1 + ost.OSS.*.nrs_crrn_quantum=1 || rc=$? + [[ $rc -eq 3 ]] && skip "no NRS exists" && return + [[ $rc -ne 0 ]] && error "failed to set crrn_quantum to 1" echo "policy: crr-n, crrn_quantum 1" nrs_write_read - do_nodes $oss lctl set_param ost.OSS.*.nrs_crrn_quantum=64 + do_nodes $oss lctl set_param \ + ost.OSS.*.nrs_crrn_quantum=64 || rc=$? + [[ $rc -ne 0 ]] && error "failed to set crrn_quantum to 64" echo "policy: crr-n, crrn_quantum 64" nrs_write_read # cleanup - do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" + do_nodes $oss lctl set_param \ + ost.OSS.ost_io.nrs_policies="fifo" || rc=$? + [[ $rc -ne 0 ]] && error "failed to set fifo policy" return 0 } run_test 77b "check CRR-N NRS policy" @@ -3091,37 +3393,49 @@ orr_trr() { oss=$(comma_list $(osts_nodes)) do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies=$policy \ - ost.OSS.*.nrs_"$policy"_quantum=1 \ - ost.OSS.*.nrs_"$policy"_offset_type="physical" \ - ost.OSS.*.nrs_"$policy"_supported="reads" + ost.OSS.*.nrs_"$policy"_quantum=1 \ + ost.OSS.*.nrs_"$policy"_offset_type="physical" \ + ost.OSS.*.nrs_"$policy"_supported="reads" || return $? - echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type physical, ${policy}_supported reads" + echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type " \ + "physical, ${policy}_supported reads" nrs_write_read - do_nodes $oss lctl set_param ost.OSS.*.nrs_${policy}_supported="writes" \ - ost.OSS.*.nrs_${policy}_quantum=64 + do_nodes $oss lctl set_param \ + ost.OSS.*.nrs_${policy}_supported="writes" \ + ost.OSS.*.nrs_${policy}_quantum=64 || return $? - echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type physical, ${policy}_supported writes" + echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type " \ + "physical, ${policy}_supported writes" nrs_write_read - do_nodes $oss lctl set_param ost.OSS.*.nrs_${policy}_supported="reads_and_writes" \ - ost.OSS.*.nrs_${policy}_offset_type="logical" - echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type logical, ${policy}_supported reads_and_writes" + do_nodes $oss lctl set_param \ + ost.OSS.*.nrs_${policy}_supported="reads_and_writes" \ + ost.OSS.*.nrs_${policy}_offset_type="logical" || return $? + echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type " \ + "logical, ${policy}_supported reads_and_writes" nrs_write_read # cleanup - do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" || + return $? return 0 } test_77c() { #LU-3266 - orr_trr "orr" + local rc + orr_trr "orr" || rc=$? + [[ $rc -eq 3 ]] && skip "no NRS exists" && return + [[ $rc -ne 0 ]] && error "orr_trr failed rc:$rc" return 0 } run_test 77c "check ORR NRS policy" test_77d() { #LU-3266 - orr_trr "trr" + local rc + orr_trr "trr" || rc=$? + [[ $rc -eq 3 ]] && skip "no NRS exists" && return + [[ $rc -ne 0 ]] && error "orr_trr failed rc:$rc" return 0 } run_test 77d "check TRR nrs policy" @@ -3194,14 +3508,14 @@ tbf_verify() { } test_77e() { - local server_version=$(lustre_version_code ost1) - [[ $server_version -ge $(version_code 2.7.58) ]] || - { skip "Need server version newer than 2.7.57"; return 0; } + local rc oss=$(comma_list $(osts_nodes)) - do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" - [ $? -ne 0 ] && error "failed to set TBF policy" + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" || + rc=$? + [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return + [[ $rc -ne 0 ]] && error "failed to set TBF NID policy" local idis local rateis @@ -3240,23 +3554,25 @@ test_77e() { run_test 77e "check TBF NID nrs policy" test_77f() { - local server_version=$(lustre_version_code ost1) - [[ $server_version -ge $(version_code 2.7.58) ]] || - { skip "Need server version newer than 2.7.57"; return 0; } + local rc oss=$(comma_list $(osts_nodes)) + do_nodes $oss $LCTL set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ jobid" || rc=$? + [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return + [[ $rc -ne 0 ]] && error "failed to set TBF JOBID policy" + # Configure jobid_var local saved_jobid_var=$($LCTL get_param -n jobid_var) + rc=$? + [[ $rc -eq 3 ]] && skip "jobid_var not found" && return + [[ $rc -ne 0 ]] && error "failed to get param jobid_var" if [ $saved_jobid_var != procname_uid ]; then - set_conf_param_and_check client \ - "$LCTL get_param -n jobid_var" \ - "$FSNAME.sys.jobid_var" procname_uid + set_persistent_param_and_check client \ + "jobid_var" "$FSNAME.sys.jobid_var" procname_uid fi - do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid" - [ $? -ne 0 ] && error "failed to set TBF policy" - local idis local rateis if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then @@ -3289,27 +3605,28 @@ test_77f() { nrs_write_read "$RUNAS" local current_jobid_var=$($LCTL get_param -n jobid_var) + [[ $? -ne 0 ]] && error "failed to get param jobid_var" if [ $saved_jobid_var != $current_jobid_var ]; then - set_conf_param_and_check client \ - "$LCTL get_param -n jobid_var" \ - "$FSNAME.sys.jobid_var" $saved_jobid_var + set_persistent_param_and_check client \ + "jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid_var fi return 0 } run_test 77f "check TBF JobID nrs policy" test_77g() { - local server_version=$(lustre_version_code ost1) - [[ $server_version -ge $(version_code 2.7.58) ]] || - { skip "Need server version newer than 2.7.57"; return 0; } + local rc=0 oss=$(comma_list $(osts_nodes)) - do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" - [ $? -ne 0 ] && error "failed to set TBF policy" + do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" || + rc=$? + [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return + [[ $rc -ne 0 ]] && error "failed to set TBF NID policy" - do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid" - [ $? -ne 0 ] && error "failed to set TBF policy" + do_nodes $oss lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ jobid" || rc=$? + [[ $rc -ne 0 ]] && error "failed to set TBF JOBID policy" local idis local rateis @@ -3385,7 +3702,7 @@ tbf_rule_check() } test_77i() { - [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] || + [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] || { skip "Need OST version at least 2.8.55"; return 0; } for i in $(seq 1 $OSTCOUNT) @@ -3440,7 +3757,11 @@ run_test 77i "Change rank of TBF rule" test_77j() { local idis local rateis - if [ $(lustre_version_code ost1) -ge $(version_code 2.8.60) ]; then + local ost_version=$(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 $(version_code 2.8.60) ]; then idis="opcode=" rateis="rate=" fi @@ -3467,6 +3788,57 @@ test_77j() { } run_test 77j "check TBF-OPCode NRS policy" +test_id() { + local idstr="${1}id" + local policy="${idstr}={$2}" + local rate="rate=$3" + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param jobid_var=procname_uid \ + ost.OSS.ost_io.nrs_policies="tbf\ ${idstr}" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ost_${idstr}\ ${policy}\ ${rate}" + [ $? -ne 0 ] && error "failed to set tbf ${idstr} policy" + + nrs_write_read "runas $4" + tbf_verify $3 $3 "runas $4" + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_${idstr}" \ + ost.OSS.ost_io.nrs_policies="fifo" + + # sleep 3 seconds to wait the tbf policy stop completely, + # or the next test case is possible get -eagain when + # setting the tbf policy + sleep 3 +} + +test_77ja(){ + if [ $(lustre_version_code ost1) -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" +} +run_test 77ja "check TBF-UID/GID NRS policy" + +cleanup_77k() +{ + local rule_lists=$1 + local old_nrs=$2 + + trap 0 + for rule in $rule_lists; do + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ $rule" + done + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_policies="$old_nrs" + + sleep 3 +} + test_77k() { [[ $(lustre_version_code ost1) -ge $(version_code 2.9.53) ]] || { skip "Need OST version at least 2.9.53"; return 0; } @@ -3509,16 +3881,59 @@ test_77k() { tbf_verify 10 10 "$RUNAS" tbf_verify 20 10 + 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; } + do_nodes $(comma_list $(osts_nodes)) \ lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_a" \ ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_b" \ - ost.OSS.ost_io.nrs_policies="fifo" + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ug\ uid={500}\&gid={1000}\ rate=5" + nrs_write_read "runas -u 500 -g 1000" + tbf_verify 5 5 "runas -u 500 -g 1000" - sleep 3 + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ug" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_uw\ uid={500}\&opcode={ost_write}\ rate=20" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ur\ uid={500}\&opcode={ost_read}\ rate=10" + + nrs_write_read "runas -u 500" + tbf_verify 20 10 "runas -u 500" + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_uw" \ + ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ur" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ uid={500},opcode={ost_write}\ rate=20" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ uid={500},opcode={ost_read}\ rate=10" + nrs_write_read "runas -u 500" + tbf_verify 10 10 "runas -u 500" + tbf_verify 20 10 "runas -u 500" + cleanup_77k "ext_a ext_b" "fifo" } -run_test 77k "check the extended TBF policy with NID/JobID/OPCode expression" +run_test 77k "check TBF policy with NID/JobID/OPCode expression" test_77l() { + [[ $(lustre_version_code ost1) -ge $(version_code 2.10.56) ]] || + { skip "Need OST version at least 2.10.56"; return 0; } + + do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" + do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf" + + local output=$(do_facet ost1 lctl get_param \ + ost.OSS.ost_io.nrs_policies | \ + awk '/name: tbf/ {print;exit}' | \ + awk -F ': ' '{print $2}') + + if [ "$output" != "tbf" ]; then + error "The generic TBF output is '$output', not 'tbf'" + fi + + do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="fifo" +} +run_test 77l "check the output of NRS policies for generic TBF" + +test_77m() { if [ $(lustre_version_code ost1) -lt $(version_code 2.9.54) ]; then skip "Need OST version at least 2.9.54" return 0 @@ -3568,19 +3983,58 @@ test_77l() { return 0 } -run_test 77l "check NRS Delay slows write RPC processing" +run_test 77m "check NRS Delay slows write RPC processing" -test_78() { #LU-6673 - local server_version=$(lustre_version_code ost1) - [[ $server_version -ge $(version_code 2.7.58) ]] || - { skip "Need server version newer than 2.7.57"; return 0; } +test_77n() { #LU-10802 + if [ $(lustre_version_code ost1) -lt $(version_code 2.10.58) ]; then + skip "Need OST version at least 2.10.58" + return 0 + fi + + # Configure jobid_var + local saved_jobid_var=$($LCTL get_param -n jobid_var) + if [ $saved_jobid_var != procname_uid ]; then + set_persistent_param_and_check client \ + "jobid_var" "$FSNAME.sys.jobid_var" procname_uid + fi + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid" \ + ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ dd_runas\ jobid={*.$RUNAS_ID}\ rate=20" + nrs_write_read + tbf_verify 20 20 "$RUNAS" + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \ + ost.OSS.ost_io.nrs_tbf_rule="start\ dd_runas\ jobid={dd.*}\ rate=20" + + nrs_write_read + tbf_verify 20 20 + + do_nodes $(comma_list $(osts_nodes)) \ + lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \ + ost.OSS.ost_io.nrs_policies="fifo" + + sleep 3 + + local current_jobid_var=$($LCTL get_param -n jobid_var) + if [ $saved_jobid_var != $current_jobid_var ]; then + set_persistent_param_and_check client \ + "jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid_var + fi +} +run_test 77n "check wildcard support for TBF JobID NRS policy" + +test_78() { #LU-6673 local rc oss=$(comma_list $(osts_nodes)) do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="orr" & do_nodes $oss lctl set_param ost.OSS.*.nrs_orr_quantum=1 rc=$? + [[ $rc -eq 3 ]] && skip "no NRS exists" && return # Valid return codes are: # 0: Tuning succeeded # ENODEV: Policy is still stopped @@ -3607,7 +4061,7 @@ test_79() { error "setfattr -n trusted.name1=value1 $DIR/$tdir failed" #define OBD_FAIL_MDS_INTENT_DELAY 0x160 - local mdtidx=$($LFS getstripe -M $DIR/$tdir) + local mdtidx=$($LFS getstripe -m $DIR/$tdir) local facet=mds$((mdtidx + 1)) stat $DIR/$tdir set_nodes_failloc $(facet_active_host $facet) 0x80000160 @@ -3653,7 +4107,7 @@ test_80a() { echo "Finish migration, then checking.." for file in $(find $DIR1/$tdir); do - mdt_index=$($LFS getstripe -M $file) + mdt_index=$($LFS getstripe -m $file) [ $mdt_index == $MDTIDX ] || error "$file is not on MDT${MDTIDX}" done @@ -3788,7 +4242,7 @@ test_80b() { } run_test 80b "Accessing directory during migration" -test_81() { +test_81a() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return rm -rf $DIR1/$tdir @@ -3814,7 +4268,35 @@ test_81() { return 0 } -run_test 81 "rename and stat under striped directory" +run_test 81a "rename and stat under striped directory" + +test_81b() { + [ $MDSCOUNT -lt 2 ] && + skip "We need at least 2 MDTs for this test" + + local total + local setattr_pid + + total=1000 + + $LFS mkdir -c $MDSCOUNT $DIR1/$tdir || error "$LFS mkdir" + createmany -o $DIR1/$tdir/$tfile. $total || error "createmany" + + ( + while true; do + touch $DIR1/$tdir + done + ) & + setattr_pid=$! + + for i in $(seq $total); do + mrename $DIR2/$tdir/$tfile.$i $DIR2/$tdir/$tfile-new.$i \ + > /dev/null + done + + kill -9 $setattr_pid +} +run_test 81b "rename under striped directory doesn't deadlock" test_82() { [[ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.6.91) ]] || @@ -3984,10 +4466,10 @@ test_93() { #define OBD_FAIL_MDS_LOV_CREATE_RACE 0x163 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 @@ -3995,11 +4477,11 @@ test_93() { do_facet $SINGLEMDS "lctl set_param -n \ 'lod.lustre-MDT*/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 ] || @@ -4010,6 +4492,9 @@ run_test 93 "alloc_rr should not allocate on same ost" # 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 + mkdir -p $DIR/$tdir $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom @@ -4019,19 +4504,21 @@ test_100a() { $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #1" # first stat from server should return size data and save glimpse - local gls=$(lctl get_param -n mdc.*.stats | \ - awk '/ldlm_glimpse/ {print $2}') - [ -z $gls ] || error "Unexpected $gls glimpse RPCs" + local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse) + [ $gls -eq 0 ] || error "Unexpected $gls glimpse RPCs" # second stat to check size is NOT cached on client without IO lock $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #2" - local gls=$(lctl get_param -n mdc.*.stats | grep ldlm_glimpse | wc -l) - [ "1" == "$gls" ] || error "Expect 1 glimpse RPCs but got $gls" + local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse) + [ $gls -ge 1 ] || error "Expect glimpse RPCs but none" rm -f $dom } 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 + mkdir -p $DIR/$tdir $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom @@ -4044,15 +4531,17 @@ test_100b() { # second stat to check size is cached on client $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #2" - local gls=$(lctl get_param -n mdc.*.stats | - awk '/ldlm_glimpse/ {print $2}') + local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse) # both stats should cause no glimpse requests - [ -z $gls ] || error "Unexpected $gls glimpse RPCs" + [ $gls == 0 ] || error "Unexpected $gls glimpse RPCs" rm -f $dom } 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 + mkdir -p $DIR/$tdir $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom @@ -4065,7 +4554,7 @@ test_100c() { $CHECKSTAT -t file -s 2097152 $DIR/$tdir/dom || error "Wrong size from stat #1" - local gls=$(lctl get_param -n osc.*.stats | grep ldlm_glimpse | wc -l) + local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse) [ $gls -eq 0 ] && error "Expect OST glimpse RPCs but got none" rm -f $dom @@ -4073,6 +4562,9 @@ 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 + mkdir -p $DIR/$tdir $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom @@ -4087,62 +4579,109 @@ test_100d() { $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "Wrong size from stat #1" - local gls=$(lctl get_param -n osc.*.stats | grep ldlm_glimpse | wc -l) + local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse) [ $gls -eq 0 ] && error "Expect OST glimpse but got none" rm -f $dom } run_test 100d "DoM: write+truncate vs stat without IO lock (combined file)" - test_101a() { + [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] && + skip "Need MDS version at least 2.10.55" && return + $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile - lctl set_param -n mdc.*.stats=clear # to get layout $CHECKSTAT -t file $DIR1/$tfile + + local old_wb=$(sysctl -n vm.dirty_writeback_centisecs) + sysctl -wq vm.dirty_writeback_centisecs=0 + + trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT + # open + IO lock - dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails" + dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || + error_noexit "Write fails" # must discard pages + lctl set_param -n mdc.*.stats=clear rm $DIR2/$tfile || error "Unlink fails" - local writes=$(lctl get_param -n mdc.*.stats | grep ost_write | wc -l) + + local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write) [ $writes -eq 0 ] || error "Found WRITE RPC but expect none" } 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 + $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile touch $DIR1/${tfile}_2 - lctl set_param -n mdc.*.stats=clear # to get layout $CHECKSTAT -t file $DIR1/$tfile + + local old_wb=$(sysctl -n vm.dirty_writeback_centisecs) + sysctl -wq vm.dirty_writeback_centisecs=0 + + trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT + # open + IO lock dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails" # must discard pages + lctl set_param -n mdc.*.stats=clear mv $DIR2/${tfile}_2 $DIR2/$tfile || error "Rename fails" - local writes=$(lctl get_param -n mdc.*.stats | grep ost_write | wc -l) + + local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write) [ $writes -eq 0 ] || error "Found WRITE RPC but expect none" } 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 + $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile - lctl set_param -n mdc.*.stats=clear # to get layout $CHECKSTAT -t file $DIR1/$tfile + + local old_wb=$(sysctl -n vm.dirty_writeback_centisecs) + sysctl -wq vm.dirty_writeback_centisecs=0 + + trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT + # open + IO lock dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails" - $MULTIOP $DIR1/$tfile O_c & MULTIOP_PID=$! - sleep 2 - rm $DIR2/$tfile > /dev/null || error "Unlink fails" - kill -USR1 $MULTIOP_PID || return 2 - wait $MULTIOP_PID || return 3 - local writes=$(lctl get_param -n mdc.*.stats | grep ost_write | wc -l) + sleep 1 + lctl set_param -n mdc.*.stats=clear + rm $DIR2/$tfile > /dev/null || error "Unlink fails for opened file" + kill -USR1 $MULTIOP_PID && wait $MULTIOP_PID || error "multiop failure" + + local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write) [ $writes -eq 0 ] || error "Found WRITE RPC but expect none" } run_test 101c "Discard DoM data on close-unlink" +# test to verify file handle related system calls +# (name_to_handle_at/open_by_handle_at) +# The new system calls are supported in glibc >= 2.14. + +# test to verify we can open by handle an unlinked file from > 1 client +# 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 +} +run_test 102 "Test open by handle of unlinked file" + log "cleanup: ======================================================" # kill and wait in each test only guarentee script finish, but command in script