X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanityn.sh;h=1e65fda2f8b7662d8c73311cbc9bbf7b19212abe;hp=1978bebd80e4347215420bdfc316f44db75a38bf;hb=849d7d5b1b4cabb7578c3ab5aaf271e90dd33864;hpb=97b9148af8f7b8d31892f2f114e9461f85e9a4a4 diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh index 1978beb..1e65fda 100644 --- a/lustre/tests/sanityn.sh +++ b/lustre/tests/sanityn.sh @@ -3,29 +3,18 @@ set -e ONLY=${ONLY:-"$*"} -# bug number for skipped test: 3192 15528/3811 16929 9977 15528/11549 18080 -ALWAYS_EXCEPT=" 14b 19 22 28 29 35 $SANITYN_EXCEPT" +# bug number for skipped test: 3192 LU-1205 15528/3811 9977 15528/11549 18080 +ALWAYS_EXCEPT=" 14b 18c 19 28 29 35 $SANITYN_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! -# bug number for skipped test: 12652 12652 -grep -q 'Enterprise Server 10' /etc/SuSE-release && ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 14" || true - -# Tests that fail on uml -[ "$UML" = "true" ] && EXCEPT="$EXCEPT 7" - -# It will be ported soon. -EXCEPT="$EXCEPT 22" - -SRCDIR=`dirname $0` +SRCDIR=$(dirname $0) PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH SIZE=${SIZE:-40960} CHECKSTAT=${CHECKSTAT:-"checkstat -v"} -GETSTRIPE=${GETSTRIPE:-lfs getstripe} -SETSTRIPE=${SETSTRIPE:-lstripe} -MCREATE=${MCREATE:-mcreate} OPENFILE=${OPENFILE:-openfile} OPENUNLINK=${OPENUNLINK:-openunlink} +export MULTIOP=${MULTIOP:-multiop} export TMP=${TMP:-/tmp} MOUNT_2=${MOUNT_2:-"yes"} CHECK_GRANT=${CHECK_GRANT:-"yes"} @@ -43,7 +32,14 @@ init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} init_logging -[ "$SLOW" = "no" ] && EXCEPT_SLOW="12 16 23 33a" +if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then +# bug number for skipped test: LU-2189 LU-2776 + ALWAYS_EXCEPT="$ALWAYS_EXCEPT 36 51a" +# LU-2829 / LU-2887 - make allowances for ZFS slowness + TEST33_NFILES=${TEST33_NFILES:-1000} +fi + +[ "$SLOW" = "no" ] && EXCEPT_SLOW="33a" FAIL_ON_ERROR=false @@ -52,11 +48,11 @@ TRACE=${TRACE:-""} check_and_setup_lustre -LOVNAME=`lctl get_param -n llite.*.lov.common_name | tail -n 1` -OSTCOUNT=`lctl get_param -n lov.$LOVNAME.numobd` +LOVNAME=$($LCTL get_param -n llite.*.lov.common_name | tail -n 1) +OSTCOUNT=$($LCTL get_param -n lov.$LOVNAME.numobd) assert_DIR -rm -rf $DIR1/[df][0-9]* $DIR1/lnk +rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}* SAMPLE_FILE=$TMP/$(basename $0 .sh).junk dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=1 @@ -133,9 +129,57 @@ test_2e() { } run_test 2e "check chmod on root is propagated to others" +test_2f() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + local MDTIDX=1 + local remote_dir=$tdir/remote_dir + + mkdir -p $DIR1/$tdir + $LFS mkdir -i $MDTIDX $DIR1/$remote_dir || + error "Create remote directory failed" + + touch $DIR1/$remote_dir/$tfile || + error "Create file under remote directory failed" + chmod 777 $DIR1/$remote_dir/$tfile || + error "Chmod file under remote directory failed" + + $CHECKSTAT -t file -p 0777 $DIR2/$remote_dir/$tfile || + error "Check attr of file under remote directory failed" + + chown $RUNAS_ID:$RUNAS_GID $DIR1/$remote_dir/$tfile || + error "Chown file under remote directory failed" + + $CHECKSTAT -u \#$RUNAS_ID -g \#$RUNAS_GID $DIR2/$remote_dir/$tfile || + error "Check owner of file under remote directory failed" + + cd $DIR2/$remote_dir || error "enter remote dir" + rm -rf $DIR1/$remote_dir/$tfile || + error "Unlink remote directory failed" + + $CHECKSTAT -t file $DIR2/$remote_dir/$tfile && + error "unlink file still exists!" + + cd $DIR2/$tdir || error "exit remote dir" + rm -rf $DIR1/$tdir || error "unlink directory failed" +} +run_test 2f "check attr/owner updates on DNE with 2 mtpt's" + +test_2g() { + dd if=/dev/zero of=$DIR1/$tfile oflag=sync bs=1M count=2 + + local block1=$(stat $DIR1/$tfile | awk '/Blocks/ {print $4} ') + cancel_lru_locks osc + local block2=$(stat $DIR2/$tfile | awk '/Blocks/ {print $4} ') + echo "$DIR1/$tfile has $block1 blocks" + echo "$DIR2/$tfile has $block2 blocks" + [ $block1 -eq $block2 ] || error +} +run_test 2g "check blocks update on sync write" + test_3() { - ( cd $DIR1 ; ln -s this/is/good $tfile ) - [ "this/is/good" = "`perl -e 'print readlink("'$DIR2/$tfile'");'`" ] || + local target="this/is/good" + ln -s $target $DIR1/$tfile || error "ln -s $target $DIR1/$tfile failed" + [ "$(ls -l $DIR2/$tfile | sed -e 's/.* -> //')" = "$target" ] || error "link $DIR2/$tfile not as expected" } run_test 3 "symlink on one mtpt, readlink on another ===========" @@ -220,7 +264,7 @@ test_10b() { run_test 10b "write of file with sub-page size on multiple mounts " test_11() { - mkdir $DIR1/d11 + test_mkdir $DIR1/d11 multiop_bg_pause $DIR1/d11/f O_c || return 1 MULTIPID=$! cp -p /bin/ls $DIR1/d11/f @@ -238,8 +282,7 @@ test_12() { run_test 12 "test lock ordering (link, stat, unlink) ===========" test_13() { # bug 2451 - directory coherency - rm -rf $DIR1/d13 - mkdir $DIR1/d13 || error + test_mkdir $DIR1/d13 || error cd $DIR1/d13 || error ls ( touch $DIR1/d13/f13 ) # needs to be a separate shell @@ -255,7 +298,7 @@ test_13() { # bug 2451 - directory coherency run_test 13 "test directory page revocation ====================" test_14() { - mkdir -p $DIR1/$tdir + test_mkdir -p $DIR1/$tdir cp -p /bin/ls $DIR1/$tdir/$tfile multiop_bg_pause $DIR1/$tdir/$tfile Ow_c || return 1 MULTIPID=$! @@ -267,11 +310,11 @@ test_14() { run_test 14 "execution of file open for write returns -ETXTBSY =" test_14a() { - mkdir -p $DIR1/d14 + test_mkdir -p $DIR1/d14 cp -p `which multiop` $DIR1/d14/multiop || error "cp failed" MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1 MULTIOP_PID=$! - multiop $DIR2/d14/multiop Oc && error "expected error, got success" + $MULTIOP $DIR2/d14/multiop Oc && error "expected error, got success" kill -USR1 $MULTIOP_PID || return 2 wait $MULTIOP_PID || return 3 rm $TMP/test14.junk $DIR1/d14/multiop || error "removing multiop" @@ -279,7 +322,7 @@ test_14a() { run_test 14a "open(RDWR) of executing file returns -ETXTBSY ====" test_14b() { # bug 3192, 7040 - mkdir -p $DIR1/d14 + test_mkdir -p $DIR1/d14 cp -p `which multiop` $DIR1/d14/multiop || error "cp failed" MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1 MULTIOP_PID=$! @@ -293,7 +336,7 @@ test_14b() { # bug 3192, 7040 run_test 14b "truncate of executing file returns -ETXTBSY ======" test_14c() { # bug 3430, 7040 - mkdir -p $DIR1/d14 + test_mkdir -p $DIR1/d14 cp -p `which multiop` $DIR1/d14/multiop || error "cp failed" MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1 MULTIOP_PID=$! @@ -306,7 +349,7 @@ test_14c() { # bug 3430, 7040 run_test 14c "open(O_TRUNC) of executing file return -ETXTBSY ==" test_14d() { # bug 10921 - mkdir -p $DIR1/d14 + test_mkdir -p $DIR1/d14 cp -p `which multiop` $DIR1/d14/multiop || error "cp failed" MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1 MULTIOP_PID=$! @@ -322,17 +365,41 @@ run_test 14d "chmod of executing file is still possible ========" test_15() { # bug 974 - ENOSPC echo "PATH=$PATH" sh oos2.sh $MOUNT1 $MOUNT2 + wait_delete_completed grant_error=`dmesg | grep "> available"` [ -z "$grant_error" ] || error "$grant_error" } run_test 15 "test out-of-space with multiple writers ===========" +COUNT=${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 [ "$(facet_fstype ost1)" = "zfs" ]; then + FSXNUM=$((COUNT / 5)) + FSXP=1 +elif [ "$SLOW" = "yes" ]; then + FSXNUM=$((COUNT * 5)) + FSXP=500 +else + FSXNUM=$COUNT + FSXP=100 +fi + test_16() { - rm -f $MOUNT1/fsxfile - lfs setstripe $MOUNT1/fsxfile -c -1 # b=10919 - fsx -c 50 -p 100 -N 2500 -l $((SIZE * 256)) -S 0 $MOUNT1/fsxfile $MOUNT2/fsxfile + local file1=$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 + fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 } -run_test 16 "2500 iterations of dual-mount fsx =================" +run_test 16 "$FSXNUM iterations of dual-mount fsx" test_17() { # bug 3513, 3667 remote_ost_nodsh && skip "remote OST with nodsh" && return @@ -350,42 +417,59 @@ test_17() { # bug 3513, 3667 run_test 17 "resource creation/LVB creation race ===============" test_18() { - $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2 + # turn e.g. ALWAYS_EXCEPT="18c" into "-e 3" + local idx + local excepts= + for idx in {a..z}; do + local ptr=EXCEPT_ALWAYS_18$idx + [ x${!ptr} = xtrue ] || continue + + excepts="$excepts -e $(($(printf %d \'$idx)-96))" + done + + $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2 $excepts sync; sleep 1; sync } run_test 18 "mmap sanity check =================================" test_19() { # bug3811 - [ -d /proc/fs/lustre/obdfilter ] || return 0 + local node=$(facet_active_host ost1) + + # check whether obdfilter is cache capable at all + if ! get_osd_param $node '' read_cache_enable >/dev/null; then + echo "not cache-capable obdfilter" + return 0 + fi - MAX=`lctl get_param -n obdfilter.*.readcache_max_filesize | head -n 1` - lctl set_param -n obdfilter.*OST*.readcache_max_filesize=4096 - dd if=/dev/urandom of=$TMP/f19b bs=512k count=32 - SUM=`cksum $TMP/f19b | cut -d" " -f 1,2` - cp $TMP/f19b $DIR1/f19b + local MAX=$(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) + cp $TMP/$tfile $DIR1/$tfile for i in `seq 1 20`; do - [ $((i % 5)) -eq 0 ] && log "test_18 loop $i" + [ $((i % 5)) -eq 0 ] && log "$testname loop $i" cancel_lru_locks osc > /dev/null - cksum $DIR1/f19b | cut -d" " -f 1,2 > $TMP/sum1 & \ - cksum $DIR2/f19b | cut -d" " -f 1,2 > $TMP/sum2 + cksum $DIR1/$tfile | cut -d" " -f 1,2 > $TMP/sum1 & \ + cksum $DIR2/$tfile | cut -d" " -f 1,2 > $TMP/sum2 wait - [ "`cat $TMP/sum1`" = "$SUM" ] || \ - error "$DIR1/f19b `cat $TMP/sum1` != $SUM" - [ "`cat $TMP/sum2`" = "$SUM" ] || \ - error "$DIR2/f19b `cat $TMP/sum2` != $SUM" + [ "$(cat $TMP/sum1)" = "$SUM" ] || \ + error "$DIR1/$tfile $(cat $TMP/sum1) != $SUM" + [ "$(cat $TMP/sum2)" = "$SUM" ] || \ + error "$DIR2/$tfile $(cat $TMP/sum2) != $SUM" done - lctl set_param -n obdfilter.*OST*.readcache_max_filesize=$MAX - rm $DIR1/f19b + set_osd_param $node '' readcache_max_filesize $MAX + rm $DIR1/$tfile } run_test 19 "test concurrent uncached read races ===============" test_20() { - mkdir $DIR1/d20 + 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 + $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 ] && \ @@ -395,45 +479,45 @@ run_test 20 "test extra readahead page left in cache ====" cleanup_21() { trap 0 - umount $DIR1/d21 + umount $DIR1/$tdir } test_21() { # Bug 5907 - mkdir $DIR1/d21 - mount /etc $DIR1/d21 --bind || error "mount failed" # Poor man's mount. + test_mkdir $DIR1/$tdir + mount /etc $DIR1/$tdir --bind || error "mount failed" # Poor man's mount. trap cleanup_21 EXIT - rmdir -v $DIR1/d21 && error "Removed mounted directory" - rmdir -v $DIR2/d21 && echo "Removed mounted directory from another mountpoint, needs to be fixed" - test -d $DIR1/d21 || error "Mounted directory disappeared" + rmdir -v $DIR1/$tdir && error "Removed mounted directory" + rmdir -v $DIR2/$tdir && echo "Removed mounted directory from another mountpoint, needs to be fixed" + test -d $DIR1/$tdir || error "Mounted directory disappeared" cleanup_21 - test -d $DIR2/d21 || test -d $DIR1/d21 && error "Removed dir still visible after umount" + test -d $DIR2/$tdir || test -d $DIR1/$tdir && error "Removed dir still visible after umount" true } run_test 21 " Try to remove mountpoint on another dir ====" test_23() { # Bug 5972 - echo "others should see updated atime while another read" > $DIR1/f23 - + local at_diff=$(do_facet $SINGLEMDS \ + $LCTL get_param -n mdd.*MDT0000*.atime_diff | head -n1) + echo "atime should be updated while another read" > $DIR1/$tfile + # clear the lock(mode: LCK_PW) gotten from creating operation cancel_lru_locks osc - - time1=`date +%s` - #MAX_ATIME_DIFF 60, we update atime only if older than 60 seconds - sleep 61 - - multiop_bg_pause $DIR1/f23 or20_c || return 1 + time1=$(date +%s) + echo "now is $time1" + sleep $((at_diff + 1)) + + echo "starting reads" + multiop_bg_pause $DIR1/$tfile or20_c || return 1 # with SOM and opencache enabled, we need to close a file and cancel # open lock to get atime propogated to MDS - kill -USR1 $! + kill -USR1 $! || return 2 cancel_lru_locks mdc - time2=`stat -c "%X" $DIR2/f23` - - if (( $time2 <= $time1 )); then - error "atime doesn't update among nodes" - fi + time2=$(stat -c "%X" $DIR/$tfile) + echo "new atime is $time2" - rm -f $DIR1/f23 || error "rm -f $DIR1/f23 failed" + [ $time2 -gt $time1 ] || error "atime was not updated" + rm -f $DIR1/$tfile || error "rm -f $DIR1/$tfile failed" true } run_test 23 " others should see updated atime while another read====" @@ -446,7 +530,7 @@ test_24a() { lfs df -i $DIR2 || error "lfs df -i $DIR2 failed" lfs df $DIR1/$tfile || error "lfs df $DIR1/$tfile failed" lfs df -ih $DIR2/$tfile || error "lfs df -ih $DIR2/$tfile failed" - + OSC=`lctl dl | awk '/-osc-|OSC.*MNT/ {print $4}' | head -n 1` # OSC=`lctl dl | awk '/-osc-/ {print $4}' | head -n 1` lctl --device %$OSC deactivate @@ -458,32 +542,67 @@ run_test 24a "lfs df [-ih] [path] test =========================" test_24b() { touch $DIR1/$tfile - fsnum=`lfs df | grep -c "filesystem summary:"` + fsnum=$(lfs_df | grep -c "summary") [ $fsnum -eq 2 ] || error "lfs df shows $fsnum != 2 filesystems." } run_test 24b "lfs df should show both filesystems ===============" -test_25() { - [ `lctl get_param -n mdc.*-mdc-*.connect_flags | grep -c acl` -lt 2 ] && \ - skip "must have acl, skipping" && return +test_25a() { + local acl=$(lctl get_param -n mdc.*MDT0000-mdc-*.connect_flags | + grep -c acl) + [ "$acl" -lt 1 ] && skip "must have acl, skipping" && return mkdir -p $DIR1/$tdir touch $DIR1/$tdir/f1 || error "touch $DIR1/$tdir/f1" chmod 0755 $DIR1/$tdir/f1 || error "chmod 0755 $DIR1/$tdir/f1" $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #1" - setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir || error "setfacl $DIR2/$tdir #1" + setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir || + error "setfacl $DIR2/$tdir #1" $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #2" - setfacl -m u:$RUNAS_ID:r-x -m g:$RUNAS_GID:r-x $DIR1/$tdir || error "setfacl $DIR2/$tdir #2" + setfacl -m u:$RUNAS_ID:r-x -m g:$RUNAS_GID:r-x $DIR1/$tdir || + error "setfacl $DIR2/$tdir #2" $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #3" - setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir || error "setfacl $DIR2/$tdir #3" + setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir || + error "setfacl $DIR2/$tdir #3" $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #4" - setfacl -x u:$RUNAS_ID: -x g:$RUNAS_GID: $DIR1/$tdir || error "setfacl $DIR2/$tdir #4" + setfacl -x u:$RUNAS_ID: -x g:$RUNAS_GID: $DIR1/$tdir || + error "setfacl $DIR2/$tdir #4" $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #5" rm -rf $DIR1/$tdir } -run_test 25 "change ACL on one mountpoint be seen on another ===" +run_test 25a "change ACL on one mountpoint be seen on another ===" + +test_25b() { + local acl=$(lctl get_param -n mdc.*MDT0000-mdc-*.connect_flags | + grep -c acl) + [ "$acl" -lt 1 ] && skip "must have acl, skipping" && return + + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + + rm -rf $DIR1/$tdir + $LFS mkdir -i 1 $DIR1/$tdir + touch $DIR1/$tdir/f1 || error "touch $DIR1/$tdir/f1" + chmod 0755 $DIR1/$tdir/f1 || error "chmod 0755 $DIR1/$tdir/f1" + + $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #1" + setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir || + error "setfacl $DIR2/$tdir #1" + $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #2" + setfacl -m u:$RUNAS_ID:r-x -m g:$RUNAS_GID:r-x $DIR1/$tdir || + error "setfacl $DIR2/$tdir #2" + $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #3" + setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir || + error "setfacl $DIR2/$tdir #3" + $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #4" + setfacl -x u:$RUNAS_ID: -x g:$RUNAS_GID: $DIR1/$tdir || + error "setfacl $DIR2/$tdir #4" + $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #5" + + rm -rf $DIR1/$tdir +} +run_test 25b "change ACL under remote dir on one mountpoint be seen on another" test_26a() { utime $DIR1/f26a -s $DIR2/f26a || error @@ -512,11 +631,11 @@ test_27() { DD2_PID=$! usleep 50 log "dd 1 started" - + dd if=/dev/zero of=$DIR1/$tfile bs=$((16384-1024))k conv=notrunc count=1 seek=4 & DD1_PID=$! log "dd 2 started" - + sleep 1 dd if=/dev/zero of=$DIR1/$tfile bs=8k conv=notrunc count=1 seek=0 log "dd 3 finished" @@ -528,10 +647,10 @@ run_test 27 "align non-overlapping extent locks from request ===" test_28() { # bug 9977 ECHO_UUID="ECHO_osc1_UUID" - tOST=`$LCTL dl | | awk '/-osc-|OSC.*MNT/ { print $4 }' | head -1` + tOST=$($LCTL dl | awk '/-osc-|OSC.*MNT/ { print $4 }' | head -n1) - lfs setstripe $DIR1/$tfile -s 1048576 -i 0 -c 2 - tOBJID=`lfs getstripe $DIR1/$tfile |grep "^[[:space:]]\+1" |awk '{print $2}'` + $LFS setstripe $DIR1/$tfile -S 1048576 -i 0 -c 2 + tOBJID=`$LFS getstripe $DIR1/$tfile | awk '$1 == 1 {print $2}'` dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 $LCTL <<-EOF @@ -540,7 +659,7 @@ test_28() { # bug 9977 setup $tOST EOF - tECHOID=`$LCTL dl | grep $ECHO_UUID | awk '{print $1}'` + tECHOID=`$LCTL dl | grep $ECHO_UUID | awk '{ print $1 }'` $LCTL --device $tECHOID destroy "${tOBJID}:0" $LCTL <<-EOF @@ -575,34 +694,40 @@ test_29() { # bug 10999 #bug 11549 - permanently turn test off in b1_5 run_test 29 "lock put race between glimpse and enqueue =========" -test_30() { #bug #11110 - mkdir -p $DIR1/$tdir - cp -f /bin/bash $DIR1/$tdir/bash - /bin/sh -c 'sleep 1; rm -f $DIR2/$tdir/bash; cp /bin/bash $DIR2/$tdir' & - err=$($DIR1/$tdir/bash -c 'sleep 2; openfile -f O_RDONLY /proc/$$/exe >& /dev/null; echo $?') - wait - [ $err -ne 116 ] && error_ignore 12900 "return code ($err) != -ESTALE" && return - true +test_30() { #bug #11110, LU-2523 + test_mkdir -p $DIR1/$tdir + cp -f /bin/bash $DIR1/$tdir/bash + /bin/sh -c 'sleep 1; rm -f $DIR2/$tdir/bash; + cp /bin/bash $DIR2/$tdir' & + $DIR1/$tdir/bash -c 'sleep 2; + openfile -f O_RDONLY /proc/$$/exe >& /dev/null; echo $?' + wait + true } -run_test 30 "recreate file race =========" +run_test 30 "recreate file race" test_31a() { - mkdir -p $DIR1/$tdir || error "Creating dir $DIR1/$tdir" - writes=`LANG=C dd if=/dev/zero of=$DIR/$tdir/$tfile count=1 2>&1 | - awk 'BEGIN { FS="+" } /out/ {print $1}'` - #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314 - lctl set_param fail_loc=0x314 - reads=`LANG=C dd if=$DIR2/$tdir/$tfile of=/dev/null 2>&1 | - awk 'BEGIN { FS="+" } /in/ {print $1}'` - [ $reads -eq $writes ] || error "read" $reads "blocks, must be" $writes + test_mkdir -p $DIR1/$tdir || error "Creating dir $DIR1/$tdir" + local writes=$(LANG=C dd if=/dev/zero of=$DIR/$tdir/$tfile \ + count=1 2>&1 | awk 'BEGIN { FS="+" } /out/ {print $1}') + #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314 + lctl set_param fail_loc=0x314 + local reads=$(LANG=C dd if=$DIR2/$tdir/$tfile of=/dev/null 2>&1 | + awk 'BEGIN { FS="+" } /in/ {print $1}') + [ $reads -eq $writes ] || error "read" $reads "blocks, must be" $writes } run_test 31a "voluntary cancel / blocking ast race==============" test_31b() { - remote_ost || { skip "local OST" && return 0; } - remote_ost_nodsh && skip "remote OST w/o dsh" && return 0 - mkdir -p $DIR1/$tdir || error "Creating dir $DIR1/$tdir" + remote_ost || { skip "local OST" && return 0; } + remote_ost_nodsh && skip "remote OST w/o dsh" && return 0 + + # make sure there is no local locks due to destroy + wait_mds_ost_sync || error "wait_mds_ost_sync()" + wait_delete_completed || error "wait_delete_completed()" + + test_mkdir -p $DIR1/$tdir || error "Creating dir $DIR1/$tdir" lfs setstripe $DIR/$tdir/$tfile -i 0 -c 1 cp /etc/hosts $DIR/$tdir/$tfile #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314 @@ -624,26 +749,27 @@ enable_lockless_truncate() { } test_32a() { # bug 11270 - local p="$TMP/sanityN-$TESTNAME.parameters" - save_lustre_params $HOSTNAME osc.*.lockless_truncate > $p + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + save_lustre_params client "osc.*.lockless_truncate" > $p cancel_lru_locks osc enable_lockless_truncate 1 rm -f $DIR1/$tfile lfs setstripe -c -1 $DIR1/$tfile - dd if=/dev/zero of=$DIR1/$tfile count=10 bs=1M > /dev/null 2>&1 + dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$STRIPE_BYTES > \ + /dev/null 2>&1 clear_osc_stats log "checking cached lockless truncate" $TRUNCATE $DIR1/$tfile 8000000 $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size" - [ $(calc_osc_stats lockless_truncate) -eq 0 ] || - error "lockless truncate doesn't use cached locks" + [ $(calc_osc_stats lockless_truncate) -ne 0 ] || + error "cached truncate isn't lockless" log "checking not cached lockless truncate" $TRUNCATE $DIR2/$tfile 5000000 $CHECKSTAT -s 5000000 $DIR1/$tfile || error "wrong file size" - [ $(calc_osc_stats lockless_truncate) -ne 0 ] || - error "not cached trancate isn't lockless" + [ $(calc_osc_stats lockless_truncate) -ne 0 ] || + error "not cached truncate isn't lockless" log "disabled lockless truncate" enable_lockless_truncate 0 @@ -662,15 +788,19 @@ run_test 32a "lockless truncate" test_32b() { # bug 11270 remote_ost_nodsh && skip "remote OST with nodsh" && return - local node - local p="$TMP/sanityN-$TESTNAME.parameters" - save_lustre_params $HOSTNAME "osc.*.contention_seconds" > $p - for node in $(osts_nodes); do - save_lustre_params $node "ldlm.namespaces.filter-*.max_nolock_bytes" >> $p - save_lustre_params $node "ldlm.namespaces.filter-*.contended_locks" >> $p - save_lustre_params $node "ldlm.namespaces.filter-*.contention_seconds" >> $p - done - clear_osc_stats + local node + local facets=$(get_facets OST) + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + + save_lustre_params client "osc.*.contention_seconds" > $p + save_lustre_params $facets \ + "ldlm.namespaces.filter-*.max_nolock_bytes" >> $p + save_lustre_params $facets \ + "ldlm.namespaces.filter-*.contended_locks" >> $p + save_lustre_params $facets \ + "ldlm.namespaces.filter-*.contention_seconds" >> $p + clear_osc_stats + # agressive lockless i/o settings for node in $(osts_nodes); do do_node $node 'lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes 2000000; lctl set_param -n ldlm.namespaces.filter-*.contended_locks 0; lctl set_param -n ldlm.namespaces.filter-*.contention_seconds 60' @@ -710,11 +840,10 @@ print_jbd_stat () { local stat=0 for mds in ${mdts//,/ }; do varsvc=${mds}_svc - dev=$(basename $(do_facet $mds lctl get_param -n osd*.${!varsvc}.mntdev)) - val=$(do_facet $mds "procfile=/proc/fs/jbd/$dev/info; -[ -f \\\$procfile ] || procfile=/proc/fs/jbd2/$dev/info; -[ -f \\\$procfile ] || procfile=/proc/fs/jbd2/${dev}\:\\\*/info; -cat \\\$procfile | head -1;") + dev=$(basename $(do_facet $mds "lctl get_param -n osd*.${!varsvc}.mntdev|\ + xargs readlink -f" )) + val=$(do_facet $mds "cat /proc/fs/jbd*/${dev}{,:*,-*}/info 2>/dev/null | + head -n1") val=${val%% *}; stat=$(( stat + val)) done @@ -725,18 +854,20 @@ cat \\\$procfile | head -1;") test_33a() { remote_mds_nodsh && skip "remote MDS with nodsh" && return - [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; } - [ $CLIENTCOUNT -ge 2 ] || \ - { skip "Need two or more clients, have $CLIENTCOUNT" && return 0; } + [ -z "$CLIENTS" ] && skip "Need two or more clients, have $CLIENTS" && return 0 + [ $CLIENTCOUNT -lt 2 ] && + skip "Need two or more clients, have $CLIENTCOUNT" && return 0 local nfiles=${TEST33_NFILES:-10000} local param_file=$TMP/$tfile-params + local fstype=$(facet_fstype $SINGLEMDS) - save_lustre_params $(comma_list $(mdts_nodes)) "mdt.*.commit_on_sharing" > $param_file + save_lustre_params $(get_facets MDS) \ + "mdt.*.commit_on_sharing" > $param_file local COS - local jbdold - local jbdnew + local jbdold="N/A" + local jbdnew="N/A" local jbd for COS in 0 1; do @@ -746,13 +877,13 @@ test_33a() { for i in 1 2 3; do do_nodes $CLIENT1,$CLIENT2 "mkdir -p $DIR1/$tdir-\\\$(hostname)-$i" - jbdold=$(print_jbd_stat) + [ $fstype = ldiskfs ] && jbdold=$(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- $nfiles > /dev/null 2>&1") - jbdnew=$(print_jbd_stat) - jbd=$(( jbdnew - jbdold )) + [ $fstype = ldiskfs ] && jbdnew=$(print_jbd_stat) + [ $fstype = ldiskfs ] && jbd=$(( jbdnew - jbdold )) echo "=== END createmany new: $jbdnew transaction : $jbd transactions nfiles $nfiles time $elapsed COS=$COS" - avgjbd=$(( avgjbd + jbd )) + [ $fstype = ldiskfs ] && avgjbd=$(( avgjbd + jbd )) avgtime=$(( avgtime + elapsed )) done eval cos${COS}_jbd=$((avgjbd / 3)) @@ -770,6 +901,66 @@ test_33a() { } run_test 33a "commit on sharing, cross crete/delete, 2 clients, benchmark" +# commit on sharing tests +test_33b() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return + + [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; } + [ $CLIENTCOUNT -ge 2 ] || + { skip "Need two or more clients, have $CLIENTCOUNT" && + return 0; } + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + + local nfiles=${TEST33_NFILES:-10000} + local param_file=$TMP/$tfile-params + + save_lustre_params $(get_facets MDS) \ + "mdt.*.commit_on_sharing" > $param_file + + local COS + local jbdold + local jbdnew + local jbd + local MDTIDX=1 + + 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_node $CLIENT1 "$LFS mkdir -i $MDTIDX -p \ + $DIR1/$tdir-\\\$(hostname)-$i" + + jbdold=$(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- $nfiles > \ + /dev/null 2>&1") + jbdnew=$(print_jbd_stat) + jbd=$(( jbdnew - jbdold )) + echo "=== END createmany new: $jbdnew transaction : \ + $jbd transactions nfiles $nfiles time $elapsed COS=$COS" + 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)) %" + + restore_lustre_params < $param_file + rm -f $param_file + return 0 +} +run_test 33b "COS: cross create/delete, 2 clients, benchmark under remote dir" + # End commit on sharing tests get_ost_lock_timeouts() { @@ -781,10 +972,21 @@ get_ost_lock_timeouts() { echo $locks } +cleanup_34() { + local i + trap 0 + 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 + done +} + test_34() { #16129 local OPER local lock_in local lock_out + trap cleanup_34 EXIT RETURN for OPER in notimeout timeout ; do rm $DIR1/$tfile 2>/dev/null lock_in=$(get_ost_lock_timeouts) @@ -823,19 +1025,21 @@ test_34() { #16129 fi fi done + cleanup_34 } run_test 34 "no lock timeout under IO" test_35() { # bug 17645 local generation=[] local count=0 - for imp in /proc/fs/lustre/mdc/$FSNAME-MDT*-mdc-*; do - g=$(awk '/generation/{print $2}' $imp/import) + gen=$(lctl get_param mdc.$FSNAME-MDT*-mdc-*.import | grep generation | + awk '/generation/{print $2}') + for g in $gen; do generation[count]=$g let count=count+1 done - mkdir -p $MOUNT1/$tfile + test_mkdir -p $MOUNT1/$tfile cancel_lru_locks mdc # Let's initiate -EINTR situation by setting fail_loc and take @@ -872,10 +1076,20 @@ test_35() { # bug 17645 do_facet client "lctl set_param fail_loc=0x0" df -h $MOUNT1 $MOUNT2 count=0 - for imp in /proc/fs/lustre/mdc/$FSNAME-MDT*-mdc-*; do - g=$(awk '/generation/{print $2}' $imp/import) - if ! test "$g" -eq "${generation[count]}"; then - error "Eviction happened on import $(basename $imp)" + gen=$(lctl get_param mdc.$FSNAME-MDT*-mdc-*.import | grep generation | + awk '/generation/{print $2}') + for g in $gen; do + if ! test "$g" -eq "${generation[count]}"; then + list=$(lctl list_param mdc.$FSNAME-MDT*-mdc-*.import) + local c = 0 + for imp in $list; do + if [ $c = $count ]; then + break + fi + c=c+1 + done + imp=$(echo "$imp" | awk -F"." '{print $2}') + error "Eviction happened on import $imp" fi let count=count+1 done @@ -883,43 +1097,45 @@ test_35() { # bug 17645 run_test 35 "-EINTR cp_ast vs. bl_ast race does not evict client" test_36() { #bug 16417 - local SIZE - local SIZE_B - local i - - mkdir -p $DIR1/$tdir - $LFS setstripe -c -1 $DIR1/$tdir - i=0 - SIZE=50 - let SIZE_B=SIZE*1024*1024 - - while [ $i -le 10 ]; do - lctl mark "start test" - local before=$($LFS df | awk '{if ($1 ~/^filesystem/) {print $5; exit} }') - dd if=/dev/zero of=$DIR1/$tdir/file000 bs=1M count=$SIZE - sync - sleep 1 - local after_dd=$($LFS df | awk '{if ($1 ~/^filesystem/) {print $5; exit} }') - multiop_bg_pause $DIR2/$tdir/file000 O_r${SIZE_B}c || return 3 - read_pid=$! - rm -f $DIR1/$tdir/file000 - kill -USR1 $read_pid - wait $read_pid - sleep 1 - local after=$($LFS df | awk '{if ($1 ~/^filesystem/) {print $5; exit} }') - echo "*** cycle($i) *** before($before):after_dd($after_dd):after($after)" - # this free space! not used - if [ $after_dd -ge $after ]; then - error "space leaked" - return 1; - fi - let i=i+1 - done + local SIZE + local SIZE_B + local i + + test_mkdir -p $DIR1/$tdir + $LFS setstripe -c -1 $DIR1/$tdir + i=0 + SIZE=50 + let SIZE_B=SIZE*1024*1024 + sync; sleep 5; sync # wait for delete thread + + while [ $i -le 10 ]; do + lctl mark "start test" + local before=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }') + dd if=/dev/zero of=$DIR1/$tdir/$tfile bs=1M count=$SIZE || + error "dd $DIR1/$tdir/$tfile ${SIZE}MB failed" + sync # sync data from client cache + sync_all_data # sync data from server cache (delayed allocation) + sleep 2 + local after_dd=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }') + multiop_bg_pause $DIR2/$tdir/$tfile O_r${SIZE_B}c || return 3 + read_pid=$! + rm -f $DIR1/$tdir/$tfile + kill -USR1 $read_pid + wait $read_pid + wait_delete_completed + local after=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }') + echo "*** cycle($i) *** before($before) after_dd($after_dd)" \ + "after($after)" + # this free space! not used + (( $after_dd <= $after)) || + error "space leaked after_dd:$after_dd > after:$after" + let i=i+1 + done } -run_test 36 "handle ESTALE/open-unlink corectly" +run_test 36 "handle ESTALE/open-unlink correctly" test_37() { # bug 18695 - mkdir -p $DIR1/$tdir + test_mkdir -p $DIR1/$tdir multiop_bg_pause $DIR1/$tdir D_c || return 1 MULTIPID=$! # create large directory (32kB seems enough from e2fsck, ~= 1000 files) @@ -1046,6 +1262,7 @@ check_pdo_conflict() { test_40a() { #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145 + touch $DIR2 mkdir $DIR1/$tfile & PID1=$! sleep 1 @@ -1062,6 +1279,7 @@ test_40a() { rm $DIR2/$tfile-4 $DIR2/$tfile-5 rmdir $DIR2/$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" wait $PID1 @@ -1091,7 +1309,8 @@ test_40b() { rmdir $DIR2/$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" + + check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 rm -r $DIR1/* return 0 @@ -1119,7 +1338,8 @@ test_40c() { rm $DIR2/$tfile-4 $DIR2/$tfile-5 rmdir $DIR2/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" - # all operations above shouldn't wait the first one + + # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 rm -r $DIR1/* @@ -1148,6 +1368,7 @@ test_40d() { rm $DIR2/$tfile-4 $DIR2/$tfile-5 rmdir $DIR2/$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" wait $PID1 @@ -1174,7 +1395,8 @@ test_40e() { rm $DIR2/$tfile-4 $DIR2/$tfile-2 rmdir $DIR2/$tfile-3 check_pdo_conflict $PID1 || error "unlink is blocked" - # all operations above shouldn't wait the first one + + # all operations above shouldn't wait the first one check_pdo_conflict $PID1 || error "parallel operation is blocked" wait $PID1 rm -r $DIR1/* @@ -1186,7 +1408,7 @@ run_test 40e "pdirops: rename and others ==============" test_41a() { #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 & + $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & PID1=$! sleep 1 mkdir $DIR2/$tfile && error "mkdir must fail" @@ -1199,10 +1421,10 @@ run_test 41a "pdirops: create vs mkdir ==============" test_41b() { #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 & + $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & PID1=$! sleep 1 - multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } rm -r $DIR1/* return 0 @@ -1213,7 +1435,7 @@ test_41c() { 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 & + $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & PID1=$! sleep 1 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail" @@ -1226,7 +1448,7 @@ run_test 41c "pdirops: create vs link ==============" test_41d() { #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 & + $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & PID1=$! sleep 1 rm $DIR2/$tfile || error "unlink must succeed" @@ -1240,7 +1462,7 @@ test_41e() { 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 & + $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & PID1=$! sleep 1 mv $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed" @@ -1253,7 +1475,7 @@ run_test 41e "pdirops: create and rename (tgt) ==============" test_41f() { #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 & + $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & PID1=$! sleep 1 mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed" @@ -1266,7 +1488,7 @@ run_test 41f "pdirops: create and rename (src) ==============" test_41g() { #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 & + $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & PID1=$! sleep 1 stat $DIR2/$tfile > /dev/null || error "stat must succeed" @@ -1279,7 +1501,7 @@ run_test 41g "pdirops: create vs getattr ==============" test_41h() { #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 & + $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c & PID1=$! sleep 1 ls -lia $DIR2/ > /dev/null @@ -1309,7 +1531,7 @@ test_42b() { mkdir $DIR1/$tfile & PID1=$! sleep 1 - multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } rm -r $DIR1/* return 0 @@ -1418,7 +1640,7 @@ test_43b() { rm $DIR1/$tfile & PID1=$! sleep 1 - multiop $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } rm -r $DIR1/* return 0 @@ -1511,6 +1733,22 @@ test_43h() { } run_test 43h "pdirops: unlink vs readdir ==============" +test_43i() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + 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" + check_pdo_conflict $PID1 && + { wait $PID1; error "remote mkdir isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 43i "pdirops: unlink vs remote mkdir" + # test 44: rename tgt and blocking operations test_44a() { touch $DIR1/$tfile-2 @@ -1533,7 +1771,7 @@ test_44b() { mv $DIR1/$tfile-2 $DIR1/$tfile & PID1=$! sleep 1 - multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } rm -r $DIR1/* return 0 @@ -1628,6 +1866,23 @@ test_44h() { } 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 + 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" + check_pdo_conflict $PID1 && { wait $PID1; + error "remote mkdir isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 44i "pdirops: rename tgt vs remote mkdir" + # test 45: rename src and blocking operations test_45a() { touch $DIR1/$tfile @@ -1650,7 +1905,7 @@ test_45b() { mv $DIR1/$tfile $DIR1/$tfile-2 & PID1=$! sleep 1 - multiop $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } rm -r $DIR1/* return 0 @@ -1722,7 +1977,7 @@ test_45g() { mv $DIR1/$tfile $DIR1/$tfile-2 & PID1=$! sleep 1 - stat $DIR2/$tfile > /dev/null && "stat must fail" + stat $DIR2/$tfile > /dev/null && error "stat must fail" check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; } rm -r $DIR1/* return 0 @@ -1743,6 +1998,22 @@ test_45h() { } run_test 45h "pdirops: unlink vs readdir ==============" +test_45i() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + 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" + check_pdo_conflict $PID1 && { wait $PID1; + error "create remote dir isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 45i "pdirops: rename src vs remote mkdir" + # test 46: link and blocking operations test_46a() { touch $DIR1/$tfile-2 @@ -1765,7 +2036,7 @@ test_46b() { link $DIR1/$tfile-2 $DIR1/$tfile & PID1=$! sleep 1 - multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; } rm -r $DIR1/* return 0 @@ -1852,12 +2123,135 @@ test_46h() { PID1=$! sleep 1 ls -lia $DIR2/ > /dev/null - check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; } + check_pdo_conflict $PID1 && { wait $PID1; + error "readdir isn't blocked"; } rm -r $DIR1/* return 0 } run_test 46h "pdirops: link vs readdir ==============" +test_46i() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + 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" + check_pdo_conflict $PID1 && { wait $PID1; + error "remote mkdir isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 46i "pdirops: link vs remote mkdir" + +# test 47: remote mkdir and blocking operations +test_47a() { +#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145 + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + 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" + check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; } + rm -r $DIR1/* + return 0 +} +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 + $LFS mkdir -i 1 $DIR1/$tfile & + PID1=$! + sleep 1 + multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail" + check_pdo_conflict $PID1 && { wait $PID1; + error "create isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 47b "pdirops: remote mkdir vs create" + +test_47c() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + 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" + check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 47c "pdirops: remote mkdir vs link" + +test_47d() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return +#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" + check_pdo_conflict $PID1 && { wait $PID1; + error "unlink isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 47d "pdirops: remote mkdir vs unlink" + +test_47e() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + 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" + check_pdo_conflict $PID1 && { wait $PID1; + error "rename isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 47e "pdirops: remote mkdir and rename (tgt)" + +test_47f() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return +#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" + check_pdo_conflict $PID1 && { wait $PID1; + error "rename isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 47f "pdirops: remote mkdir and rename (src)" + +test_47g() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return +#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" + check_pdo_conflict $PID1 && { wait $PID1; + error "getattr isn't blocked"; } + rm -r $DIR1/* + return 0 +} +run_test 47g "pdirops: remote mkdir vs getattr" + test_50() { trunc_size=4096 dd if=/dev/zero of=$DIR1/$tfile bs=1K count=10 @@ -1871,10 +2265,915 @@ test_50() { } run_test 50 "osc lvb attrs: enqueue vs. CP AST ==============" +test_51a() { + local filesize + local origfile=/etc/hosts + + filesize=`stat -c %s $origfile` + + # create an empty file + $MCREATE $DIR1/$tfile + # cache layout lock on both mount point + stat $DIR1/$tfile > /dev/null + stat $DIR2/$tfile > /dev/null + + # open and sleep 2 seconds then read + $MULTIOP $DIR2/$tfile o_2r${filesize}c & + local pid=$! + sleep 1 + + # create the layout of testing file + dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null + + # MULTIOP proc should be able to read enough bytes and exit + sleep 2 + kill -0 $pid && error "multiop is still there" + cmp $origfile $DIR2/$tfile || error "$MCREATE and $DIR2/$tfile differs" + + rm -f $DIR1/$tfile +} +run_test 51a "layout lock: refresh layout should work" + +test_51b() { + [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.59) ]] || + { skip "Need MDS version at least 2.3.59"; return 0; } + + local tmpfile=`mktemp` + + # create an empty file + $MCREATE $DIR1/$tfile + + # delay glimpse so that layout has changed when glimpse finish +#define OBD_FAIL_GLIMPSE_DELAY 0x1404 + $LCTL set_param fail_loc=0x1404 + stat -c %s $DIR2/$tfile |tee $tmpfile & + local pid=$! + sleep 1 + + # create layout of testing file + dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc > /dev/null + + wait $pid + local fsize=`cat $tmpfile` + + [ x$fsize = x1024 ] || error "file size is $fsize, should be 1024" + + rm -f $DIR1/$tfile $tmpfile +} +run_test 51b "layout lock: glimpse should be able to restart if layout changed" + +test_51c() { + [ $OSTCOUNT -ge 2 ] || { skip "need at least 2 osts"; return; } + + # set default layout to have 1 stripe + mkdir -p $DIR1/$tdir + $LFS setstripe -c 1 $DIR1/$tdir + + # create a file with empty layout + $MCREATE $DIR1/$tdir/$tfile + +#define OBD_FAIL_MDS_LL_BLOCK 0x172 + do_facet $SINGLEMDS $LCTL set_param fail_loc=0x172 + + # change the layout of testing file + echo "Setting layout to have $OSTCOUNT stripes ..." + $LFS setstripe -c $OSTCOUNT $DIR1/$tdir/$tfile & + pid=$! + sleep 1 + + # write something to the file, it should be blocked on fetching layout + dd if=/dev/zero of=$DIR2/$tdir/$tfile bs=1k count=1 conv=notrunc + local cnt=$($LFS getstripe -c $DIR2/$tdir/$tfile) + [ $cnt -eq $OSTCOUNT ] || error "have $cnt stripes, expected $OSTCOUNT" + + rm -fr $DIR1/$tdir +} +run_test 51c "layout lock: IT_LAYOUT blocked and correct layout can be returned" + +test_51d() { + dd if=/dev/zero of=/$DIR1/$tfile bs=1M count=1 + cancel_lru_locks mdc + + # open should grant LAYOUT lock, mmap and read will install pages + $MULTIOP $DIR1/$tfile oO_RDWR:SMR_Uc & + local PID=$! + sleep 1 + + # rss before revoking + local br=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}') + echo "Before revoking layout lock: $br KB mapped" + + # delete the file will revoke layout lock + rm -f $DIR2/$tfile + + # rss after revoking + local ar=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}') + + kill -USR1 $PID + wait $PID || error + + [ $ar -eq 0 ] || error "rss before: $br, after $ar, some pages remained" +} +run_test 51d "layout lock: losing layout lock should clean up memory map region" + +test_54_part1() +{ + echo "==> rename vs getattr vs setxattr should not deadlock" + mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed" + + do_facet mds $LCTL set_param fail_loc=$1 + + mv -T $DIR/d1/d2/d3 $DIR/d1/d3 & + PID1=$! + sleep 1 + + stat $DIR/d1/d2 & + PID2=$! + sleep 1 + + setfattr -n user.attr1 -v value1 $DIR2/d1 || error "(2) setfattr failed" + wait $PID1 || error "(3) mv failed" + wait $PID2 || error "(4) stat failed" + echo + + rm -rf $DIR/d1 +} + +test_54_part2() { + echo "==> rename vs getattr vs open vs getattr should not deadlock" + mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed" + + do_facet mds $LCTL set_param fail_loc=$1 + + mv -T $DIR/d1/d2/d3 $DIR/d1/d3 & + PID1=$! + sleep 1 + + stat $DIR/d1/d2 & + PID2=$! + sleep 1 + + $MULTIOP $DIR2/d1/d2 Oc & + PID3=$! + sleep 1 + + stat $DIR/d1 || error "(2) stat failed" + + wait $PID1 || error "(3) mv failed" + wait $PID2 || error "(4) stat failed" + wait $PID3 && error "(5) multiop failed" + echo + rm -rf $DIR/d1 +} + +test_54() { + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + save_lustre_params client "llite.*.xattr_cache" > $p + lctl set_param llite.*.xattr_cache 1 || + { skip "xattr cache is not supported"; return 0; } + +#define OBD_FAIL_MDS_RENAME 0x153 +#define OBD_FAIL_MDS_RENAME2 0x154 + test_54_part1 0x80000153 || error 10 + test_54_part1 0x80000154 || error 11 + test_54_part2 0x80000153 || error 12 + test_54_part2 0x80000154 || error 13 + + restore_lustre_params < $p + rm -f $p +} +run_test 54 "rename locking" + +test_55a() { + mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" + +#define OBD_FAIL_MDS_RENAME4 0x156 + do_facet mds $LCTL set_param fail_loc=0x80000156 + + mv -T $DIR/d1/d2 $DIR/d3/d2 & + PID1=$! + sleep 1 + + rm -r $DIR2/d3 + wait $PID1 && error "(2) mv succeeded" + + rm -rf $DIR/d1 +} +run_test 55a "rename vs unlink target dir" + +test_55b() +{ + mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" + +#define OBD_FAIL_MDS_RENAME4 0x156 + do_facet mds $LCTL set_param fail_loc=0x80000156 + + mv -T $DIR/d1/d2 $DIR/d3/d2 & + PID1=$! + sleep 1 + + rm -r $DIR2/d1 + wait $PID1 && error "(2) mv succeeded" + + rm -rf $DIR/d3 +} +run_test 55b "rename vs unlink source dir" + +test_55c() +{ + mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed" + +#define OBD_FAIL_MDS_RENAME4 0x156 + do_facet mds $LCTL set_param fail_loc=0x156 + + mv -T $DIR/d1/d2 $DIR/d3/d2 & + PID1=$! + sleep 1 + + # while rename is sleeping, open and remove d3 + $MULTIOP $DIR2/d3 D_c & + PID2=$! + sleep 1 + rm -rf $DIR2/d3 + sleep 5 + + # while rename is sleeping 2nd time, close d3 + kill -USR1 $PID2 + wait $PID2 || error "(3) multiop failed" + + wait $PID1 && error "(2) mv succeeded" + + rm -rf $DIR/d1 +} +run_test 55c "rename vs unlink orphan target dir" + +test_55d() +{ + touch $DIR/f1 + +#define OBD_FAIL_MDS_RENAME3 0x155 + do_facet mds $LCTL set_param fail_loc=0x155 + mv $DIR/f1 $DIR/$tdir & + PID1=$! + sleep 2 + + # while rename is sleeping, create $tdir, but as a directory + mkdir -p $DIR2/$tdir || error "(1) mkdir failed" + + # link in reverse locking order + ln $DIR2/f1 $DIR2/$tdir/ + + wait $PID1 && error "(2) mv succeeded" + rm -rf $DIR/f1 +} +run_test 55d "rename file vs link" + +test_60() { + local MDSVER=$(lustre_build_version $SINGLEMDS) + [ $(version_code $MDSVER) -lt $(version_code 2.3.0) ] && + skip "MDS version $MDSVER must be >= 2.3.0" && return 0 + + # Create a file + test_mkdir -p $DIR1/$tdir + file1=$DIR1/$tdir/file + file2=$DIR2/$tdir/file + + echo orig > $file2 || error "Could not create $file2" + version=$($LFS data_version $file1) + + # Append data + echo append >> $file2 || error "Could not append to $file2" + version2=$($LFS data_version $file1) + [ "$version" != "$version2" ] || + error "append did not change data version: $version" + + # Overwrite data + echo overwrite > $file2 || error "Could not overwrite $file2" + version3=$($LFS data_version $file1) + [ "$version2" != "$version3" ] || + error "overwrite did not change data version: $version2" + + # Truncate before EOF + $TRUNCATE $file2 3 || error "Could not truncate $file2" + version4=$($LFS data_version $file1) + [ "$version3" != "$version4" ] || + error "truncate did not change data version: $version3" + + # Truncate after EOF + $TRUNCATE $file2 123456 || error "Could not truncate $file2" + version5=$($LFS data_version $file1) + [ "$version4" != "$version5" ] || + error "truncate did not change data version: $version4" + + # Chmod do not change version + chmod 400 $file2 || error "Could not chmod 400 $file2" + version6=$($LFS data_version $file1) + [ "$version5" == "$version6" ] || + error "chmod should not change data version: $version5 != $version6" + + # Chown do not change version + chown $RUNAS_ID $file2 || error "Could not chown $RUNAS_ID $file2" + version7=$($LFS data_version $file1) + [ "$version5" == "$version7" ] || + error "chown should not change data version: $version5 != $version7" +} +run_test 60 "Verify data_version behaviour" + +test_70a() { + local test_dir=$tdir/test_dir + + mkdir -p $DIR1/$tdir + if [ $MDSCOUNT -ge 2 ]; then + local MDTIDX=1 + $LFS mkdir -i $MDTIDX $DIR1/$test_dir || + error "Create remote directory failed" + else + mkdir -p $DIR1/$test_dir + fi + cd $DIR2/$test_dir || error "cd directory failed" + rm -rf $DIR1/$test_dir || error "unlink directory failed" + + cd $DIR2/$tdir || error "exit directory" +} +run_test 70a "cd directory && rm directory" + +test_70b() { # LU-2781 + local i + mkdir -p $DIR1/$tdir + + touch $DIR1/$tdir/file + for ((i = 0; i < 32; i++)); do + $LFS rm_entry $DIR1/$tdir/non_existent_dir &>/dev/null + done + rm $DIR1/$tdir/file || error "cannot remove file after rm_entry" + + touch $DIR1/$tdir/file + $LFS mkdir -i0 $DIR1/$tdir/test_dir + $LFS rm_entry $DIR1/$tdir/test_dir &>/dev/null + rm -rf $DIR1/$tdir/test_dir || + error "cannot remove directory after rm_entry" + rm $DIR1/$tdir/file || error "cannot remove file after rm_entry" +} +run_test 70b "remove files after calling rm_entry" + +test_71() { + local server_version=$(lustre_version_code $SINGLEMDS) + + [[ $server_version -lt $(version_code 2.1.6) ]] && + skip "Need MDS version at least 2.1.6" && return + + # Patch not applied to 2.2 and 2.3 branches + [[ $server_version -ge $(version_code 2.2.0) ]] && + [[ $server_version -lt $(version_code 2.4.0) ]] && + skip "Need MDS version at least 2.4.0" && return + + checkfiemap --test || + { skip "checkfiemap not runnable: $?" && return; } + # write data this way: hole - data - hole - data + dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1 + [ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tfile) + 1)))" = \ + "zfs" ] && + skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0 + dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1 + GET_STAT="lctl get_param -n ldlm.services.ldlm_cbd.stats" + stat $DIR2/$tfile + local can1=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') + echo $can1 + checkfiemap $DIR2/$tfile 81920 || + error "data is not flushed from client" + local can2=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') + echo $can2 + + # common case of "create file, copy file" on a single node + # should not flush data from ost + dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1 + dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1 + stat $DIR1/$tfile + local can3=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') + echo $can3 + checkfiemap $DIR1/$tfile 81920 || + error 4 + local can4=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}') + echo $can2 + [ $can3 -eq $can4 ] || error $((can2-can1)) "cancel RPC occured." +} +run_test 71 "correct file map just after write operation is finished" + +test_72() { + local p="$TMP/sanityN-$TESTNAME.parameters" + local tlink1 + local tlink2 + save_lustre_params client "llite.*.xattr_cache" > $p + lctl set_param llite.*.xattr_cache 1 || + { skip "xattr cache is not supported"; return 0; } + + touch $DIR1/$tfile + setfattr -n user.attr1 -v value1 $DIR1/$tfile || + error "setfattr1 failed" + getfattr -n user.attr1 $DIR2/$tfile | grep value1 || + error "getfattr1 failed" + setfattr -n user.attr1 -v value2 $DIR2/$tfile || + error "setfattr2 failed" + getfattr -n user.attr1 $DIR1/$tfile | grep value2 || + error "getfattr2 failed" + + # check that trusted.link is consistent + tlink1=$(getfattr -n trusted.link $DIR1/$tfile | md5sum) + ln $DIR2/$tfile $DIR2/$tfile-2 || error "failed to link" + tlink2=$(getfattr -n trusted.link $DIR1/$tfile | md5sum) + echo "$tlink1 $tlink2" + [ "$tlink1" = "$tlink2" ] && error "trusted.link should have changed!" + + rm -f $DIR2/$tfile + + restore_lustre_params < $p + rm -f $p +} +run_test 72 "getxattr/setxattr cache should be consistent between nodes" + +test_73() { + local p="$TMP/sanityN-$TESTNAME.parameters" + save_lustre_params client "llite.*.xattr_cache" > $p + lctl set_param llite.*.xattr_cache 1 || + { skip "xattr cache is not supported"; return 0; } + + touch $DIR1/$tfile + setfattr -n user.attr1 -v value1 $DIR1/$tfile || + error "setfattr1 failed" + getfattr -n user.attr1 $DIR2/$tfile || error "getfattr1 failed" + getfattr -n user.attr1 $DIR1/$tfile || error "getfattr2 failed" + clear_llite_stats + # PR lock should be cached by now on both clients + getfattr -n user.attr1 $DIR1/$tfile || error "getfattr3 failed" + # 2 hits for getfattr(0)+getfattr(size) + [ $(calc_llite_stats getxattr_hits) -eq 2 ] || error "not cached in $DIR1" + getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed" + # 4 hits for more getfattr(0)+getfattr(size) + [ $(calc_llite_stats getxattr_hits) -eq 4 ] || error "not cached in $DIR2" + rm -f $DIR2/$tfile + + restore_lustre_params < $p + rm -f $p +} +run_test 73 "getxattr should not cause xattr lock cancellation" + +test_74() { + [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.93) ] && + skip "Need MDS version at least 2.4.93" && return + + dd if=/dev/zero of=$DIR1/$tfile-1 bs=1K count=1 + dd if=/dev/zero of=$DIR1/$tfile-2 bs=1K count=1 + flocks_test 4 $DIR1/$tfile-1 $DIR2/$tfile-2 +} +run_test 74 "flock deadlock: different mounts ==============" + +# LU-3889 +test_75() { + $LFS setstripe -c 2 -S 1m -i 0 $DIR1/$tfile + dd if=/dev/zero of=$DIR1/$tfile bs=1M count=2 + cancel_lru_locks osc + + dd of=$DIR1/$tfile if=/dev/zero bs=1M count=1 seek=1 conv=notrunc + sync + + # define OBD_FAIL_LDLM_ENQUEUE_HANG 0x31d + $LCTL set_param fail_loc=0x31d + stat -c %s $DIR1/$tfile & + local pid=$! + sleep 1 + kill -9 $pid + + # For bad lock error handler we should ASSERT and got kernel panic here + sleep 4 + $LCTL set_param fail_loc=0 +} +run_test 75 "osc: upcall after unuse lock===================" + +test_76() { #LU-946 + [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.53) ]] && + skip "Need MDS version at least 2.5.53" && return + + remote_mds_nodsh && skip "remote MDS with nodsh" && return + local fcount=2048 + declare -a fd_list + declare -a fid_list + + if remote_mds; then + nid=$($LCTL list_nids | sed "s/\./\\\./g") + else + nid="0@lo" + fi + + rm -rf $DIR/$tdir + test_mkdir -p $DIR/$tdir + + # drop all open locks and close any cached "open" files on the client + cancel_lru_locks mdc + + echo -n "open files " + ulimit -n 8096 + for ((i = 0; i < $fcount; i++)); do + touch $DIR/$tdir/f_$i + local fd=$(free_fd) + local cmd="exec $fd<$DIR/$tdir/f_$i" + eval $cmd + fd_list[i]=$fd + echo -n "." + done + echo + + local get_open_fids="$LCTL get_param -n mdt.*.exports.'$nid'.open_files" + local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $get_open_fids)) + + # Possible errors in openfiles FID list. + # 1. Missing FIDs. Check 1 + # 2. Extra FIDs. Check 1 + # 3. Duplicated FID. Check 2 + # 4. Invalid FIDs. Check 2 + # 5. Valid FID, points to some other file. Check 3 + + # Check 1 + [ ${#fid_list[@]} -ne $fcount ] && + error "${#fid_list[@]} != $fcount open files" + + for (( i = 0; i < $fcount; i++ )) ; do + cmd="exec ${fd_list[i]} /dev/null 2>&1 + + for ((i = 0; i < $n; i++)); do + do_nodes $CLIENTS $myRUNAS dd if=/dev/zero\ + of="$dir/nrs_w_$HOSTNAME" bs=1M seek=$i count=1\ + > /dev/null 2>&1 & + local pids_w[$i]=$! + done + do_nodes $CLIENTS sync; + cancel_lru_locks osc + + for ((i = 0; i < $n; i++)); do + do_nodes $CLIENTS $myRUNAS dd if="$dir/nrs_w_$HOSTNAME"\ + of=/dev/zero bs=1M seek=$i count=1 > /dev/null 2>&1 & + local pids_r[$i]=$! + done + cancel_lru_locks osc + + for ((i = 0; i < $n; i++)); do + wait ${pids_w[$i]} + wait ${pids_r[$i]} + done + rm -rf $dir || error "rm -rf $dir failed" +} + +test_77a() { #LU-3266 + do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_policies="fifo" + nrs_write_read + + return 0 +} +run_test 77a "check FIFO NRS policy" + + +test_77b() { #LU-3266 + do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_policies="crrn" + do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_crrn_quantum=1 + + echo "policy: crr-n, crrn_quantum 1" + nrs_write_read + + do_facet $SINGLEMDS lctl set_param ost.OSS.*.nrs_crrn_quantum=64 + + echo "policy: crr-n, crrn_quantum 64" + nrs_write_read + + return 0 +} +run_test 77b "check CRR-N NRS policy" + +orr_trr() { + local policy=$1 + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies=$policy + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_"$policy"_quantum=1 + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_"$policy"_offset_type="physical" + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_"$policy"_supported="reads" + done + + echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type \ + physical, ${policy}_supported reads" + nrs_write_read + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_${policy}_supported="writes" + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_${policy}_quantum=64 + done + echo "policy: $policy, ${policy}_quantum 64, \ + ${policy}_offset_type physical, ${policy}_supported writes" + nrs_write_read + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_${policy}_supported="reads_and_writes" + do_facet ost"$i" lctl set_param \ + ost.OSS.*.nrs_${policy}_offset_type="logical" + done + echo "policy: $policy, ${policy}_quantum 64, \ + ${policy}_offset_type logical, ${policy}_supported reads_and_writes" + nrs_write_read + + return 0 +} + +test_77c() { #LU-3266 + orr_trr "orr" + return 0 +} +run_test 77c "check ORR NRS policy" + +test_77d() { #LU-3266 + orr_trr "trr" + return 0 +} +run_test 77d "check TRR nrs policy" + +tbf_rule_operate() +{ + local facet=$1 + shift 1 + + do_facet $facet lctl set_param \ + ost.OSS.ost_io.nrs_tbf_rule="$@" + [ $? -ne 0 ] && + error "failed to operate on TBF rules" +} + +test_77e() { + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ nid" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + # Only operate rules on ost0 since OSTs might run on the same OSS + # Add some rules + tbf_rule_operate ost0 "start\ localhost\ {0@lo}\ 1000" + local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)") + local client_nids=$(nids_list $address "\\") + tbf_rule_operate ost0 "start\ clients\ {$client_nids}\ 100" + tbf_rule_operate ost0 "start\ others\ {*.*.*.*@$NETTYPE}\ 50" + nrs_write_read + + # Change the rules + tbf_rule_operate ost0 "change\ localhost\ 1001" + tbf_rule_operate ost0 "change\ clients\ 101" + tbf_rule_operate ost0 "change\ others\ 51" + nrs_write_read + + # Stop the rules + tbf_rule_operate ost0 "stop\ localhost" + tbf_rule_operate ost0 "stop\ clients" + tbf_rule_operate ost0 "stop\ others" + nrs_write_read + + # Cleanup the TBF policy + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && + error "failed to set policy back to fifo" + done + nrs_write_read + return 0 +} +run_test 77e "check TBF NID nrs policy" + +test_77f() { + # Configure jobid_var + local saved_jobid_var=$($LCTL get_param -n jobid_var) + if [ $saved_jobid_var != procname_uid ]; then + set_conf_param_and_check client \ + "$LCTL get_param -n jobid_var" \ + "$FSNAME.sys.jobid_var" procname_uid + fi + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ jobid" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + # Only operate rules on ost0 since OSTs might run on the same OSS + # Add some rules + tbf_rule_operate ost0 "start\ runas\ {iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ 1000" + tbf_rule_operate ost0 "start\ iozone_runas\ {iozone.$RUNAS_ID}\ 100" + tbf_rule_operate ost0 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50" + nrs_write_read "$RUNAS" + + # Change the rules + tbf_rule_operate ost0 "change\ runas\ 1001" + tbf_rule_operate ost0 "change\ iozone_runas\ 101" + tbf_rule_operate ost0 "change\ dd_runas\ 51" + nrs_write_read "$RUNAS" + + # Stop the rules + tbf_rule_operate ost0 "stop\ runas" + tbf_rule_operate ost0 "stop\ iozone_runas" + tbf_rule_operate ost0 "stop\ dd_runas" + nrs_write_read "$RUNAS" + + # Cleanup the TBF policy + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && + error "failed to set policy back to fifo" + done + nrs_write_read "$RUNAS" + + local current_jobid_var=$($LCTL get_param -n jobid_var) + if [ $saved_jobid_var != $current_jobid_var ]; then + set_conf_param_and_check client \ + "$LCTL get_param -n jobid_var" \ + "$FSNAME.sys.jobid_var" $saved_jobid_var + fi + return 0 +} +run_test 77f "check TBF JobID nrs policy" + +test_77g() { + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ nid" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="tbf\ jobid" + [ $? -ne 0 ] && + error "failed to set TBF policy" + done + + # Add a rule that only valid for Jobid TBF. If direct change between + # TBF types is not supported, this operation will fail. + tbf_rule_operate ost0 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50" + + # Cleanup the TBF policy + for i in $(seq 1 $OSTCOUNT) + do + do_facet ost"$i" lctl set_param \ + ost.OSS.ost_io.nrs_policies="fifo" + [ $? -ne 0 ] && + error "failed to set policy back to fifo" + done + return 0 +} +run_test 77g "Change TBF type directly" + +test_80() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + local MDTIDX=1 + local mdt_index + local i + local file + local pid + + mkdir -p $DIR1/$tdir/dir + createmany -o $DIR1/$tdir/dir/f 10 || + error "create files under remote dir failed $i" + + cp /etc/passwd $DIR1/$tdir/$tfile + + #migrate open file should fails + multiop_bg_pause $DIR2/$tdir/$tfile O_c || error "open $file failed" + pid=$! + # give multiop a chance to open + sleep 1 + + $LFS migrate -m $MDTIDX $DIR1/$tdir && + error "migrate open files should failed with open files" + + kill -USR1 $pid + + $LFS migrate -m $MDTIDX $DIR1/$tdir || + error "migrate remote dir error" + + echo "Finish migration, then checking.." + for file in $(find $DIR1/$tdir); do + mdt_index=$($LFS getstripe -M $file) + [ $mdt_index == $MDTIDX ] || + error "$file is not on MDT${MDTIDX}" + done + + diff /etc/passwd $DIR1/$tdir/$tfile || + error "file different after migration" + + rm -rf $DIR1/$tdir || error "rm dir failed after migration" +} +run_test 80 "migrate directory when some children is being opened" + +test_81() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + + rm -rf $DIR1/$tdir + + mkdir -p $DIR1/$tdir + + $LFS setdirstripe -i0 -c$MDSCOUNT $DIR1/$tdir/d0 + $LFS setdirstripe -i0 -c$MDSCOUNT $DIR1/$tdir/d1 + + cd $DIR1/$tdir + touch d0/0 || error "create 0 failed" + mv d0/0 d1/0 || error "rename d0/0 d1/0 failed" + stat d0/0 && error "stat mv filed succeed" + mv $DIR2/$tdir/d1/0 $DIR2/$tdir/d0/0 || "rename d1/0 d0/0 failed" + stat d0/0 || error "stat failed" + + local t=$(ls -ai $DIR1/$tdir/d0 | sort -u | wc -l) + + if [ $t -ne 3 ]; then + ls -ai $DIR1/$tdir/d0 + error "expect 3 get $t" + fi + + return 0 +} +run_test 81 "rename and stat under striped directory" + +test_82() { + [[ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.6.91) ]] || + { skip "Need MDS version at least 2.6.92"; return 0; } + + # Client 1 creates a file. + multiop_bg_pause $DIR1/$tfile O_ac || error "multiop_bg_pause 1" + pid1=$! + # Client 2 opens the file. + multiop_bg_pause $DIR2/$tfile o_Ac || error "multiop_bg_pause 2" + pid2=$! + # Client 1 makes the file an orphan. + rm $DIR1/$tfile || error "rm" + # Client 2 sets EA "user.multiop". + kill -s USR1 $pid2 + wait $pid2 || error "multiop 2" + # Client 1 gets EA "user.multiop". This used to fail because the EA + # cache refill would get "trusted.link" from mdd_xattr_list() but + # -ENOENT when trying to get "trusted.link"'s value. See also sanity + # 102q. + kill -s USR1 $pid1 + wait $pid1 || error "multiop 1" +} +run_test 82 "fsetxattr and fgetxattr on orphan files" + log "cleanup: ======================================================" [ "$(mount | grep $MOUNT2)" ] && umount $MOUNT2 -complete $(basename $0) $SECONDS +complete $SECONDS +rm -f $SAMPLE_FILE check_and_cleanup_lustre exit_status