X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity.sh;h=c70826e01325b24511c4ba383b14919e97331f5d;hb=ac37e7b4d101761bbff401ed12fcf671d6b68f9c;hp=16a041078b847e997aaaf237524f419dc3d1cdd3;hpb=413b6c2e365cf3ff986611e20dd77186ed25a3ac;p=fs%2Flustre-release.git diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh index 16a0410..c70826e 100644 --- a/lustre/tests/sanity.sh +++ b/lustre/tests/sanity.sh @@ -9,7 +9,7 @@ set -e ONLY=${ONLY:-"$*"} # bug number for skipped test: 13297 2108 9789 3637 9789 3561 12622 5188 -ALWAYS_EXCEPT=" 27u 42a 42b 42c 42d 45 51d 68b $SANITY_EXCEPT" +ALWAYS_EXCEPT=" 42a 42b 42c 42d 45 51d 68b $SANITY_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! # with LOD/OSP landing @@ -58,11 +58,11 @@ init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/${NAME}.sh} init_logging -[ "$SLOW" = "no" ] && EXCEPT_SLOW="24o 24v 27m 36f 36g 36h 51b 60c 63 64b 68 71 73 77f 78 101a 103 115 120g 124b" +[ "$SLOW" = "no" ] && EXCEPT_SLOW="24o 27m 64b 68 71 77f 78 115 124b" [ $(facet_fstype $SINGLEMDS) = "zfs" ] && -# bug number for skipped test: LU-2834 LU-1593 LU-2610 LU-2833 LU-1957 LU-2805 - ALWAYS_EXCEPT="$ALWAYS_EXCEPT 18 34h 40 48a 180 184c" +# bug number for skipped test: LU-1593 LU-2610 LU-2833 LU-1957 LU-2805 + ALWAYS_EXCEPT="$ALWAYS_EXCEPT 34h 40 48a 180 184c" FAIL_ON_ERROR=false @@ -89,6 +89,13 @@ check_kernel_version() { return 1 } +check_swap_layouts_support() +{ + $LCTL get_param -n llite.*.sbi_flags | grep -q layout || + { skip "Does not support layout lock."; return 0; } + return 1 +} + if [ "$ONLY" == "cleanup" ]; then sh llmountcleanup.sh exit 0 @@ -468,9 +475,18 @@ str_repeat() { # Long symlinks and LU-2241 test_17g() { - test_mkdir -p $DIR/$tdir + test_mkdir -p $DIR/$tdir local TESTS="59 60 61 4094 4095" + # Fix for inode size boundary in 2.1.4 + [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.1.4) ] && + TESTS="4094 4095" + + # Patch not applied to 2.2 or 2.3 branches + [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.2.0) ] && + [ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.3.55) ] && + TESTS="4094 4095" + for i in $TESTS; do local SYMNAME=$(str_repeat 'x' $i) ln -s $SYMNAME $DIR/$tdir/f$i || error "failed $i-char symlink" @@ -519,12 +535,12 @@ test_17k() { #bug 22301 rsync --help | grep -q xattr || skip_env "$(rsync --version| head -1) does not support xattrs" [ $PARALLEL == "yes" ] && skip "skip parallel run" && return - test_mkdir -p $DIR/$tdir + test_mkdir -p $DIR/$tdir test_mkdir -p $DIR/$tdir.new - touch $DIR/$tdir/$tfile - ln -s $DIR/$tdir/$tfile $DIR/$tdir/$tfile.lnk - rsync -av -X $DIR/$tdir/ $DIR/$tdir.new || - error "rsync failed with xattrs enabled" + touch $DIR/$tdir/$tfile + ln -s $DIR/$tdir/$tfile $DIR/$tdir/$tfile.lnk + rsync -av -X $DIR/$tdir/ $DIR/$tdir.new || + error "rsync failed with xattrs enabled" } run_test 17k "symlinks: rsync with xattrs enabled =========================" @@ -661,9 +677,33 @@ test_17n() { } run_test 17n "run e2fsck against master/slave MDT which contains remote dir" +test_17o() { + local WDIR=$DIR/${tdir}o + local mdt_index + local mdtdevname + local rc=0 + + mkdir -p $WDIR + mdt_index=$($LFS getstripe -M $WDIR) + mdt_index=$((mdt_index+1)) + mdtdevname=$(mdsdevname $mdt_index) + + touch $WDIR/$tfile + stop mds${mdt_index} + start mds${mdt_index} $mdtdevname $MDS_MOUNT_OPTS + + #define OBD_FAIL_OSD_LMA_INCOMPAT 0x194 + do_facet mds${mdt_index} lctl set_param fail_loc=0x194 + ls -l $WDIR/$tfile && rc=1 + do_facet mds${mdt_index} lctl set_param fail_loc=0 + [[ $rc -ne 0 ]] && error "stat file should fail" + true +} +run_test 17o "stat file with incompat LMA feature" + test_18() { - touch $DIR/f - ls $DIR || error + touch $DIR/f || error "Failed to touch $DIR/f: $?" + ls $DIR || error "Failed to ls $DIR: $?" } run_test 18 "touch .../f ; ls ... ==============================" @@ -1103,6 +1143,11 @@ test_24z() { mrename $remote_src $remote_tgt && error "rename remote dirs should not work!" + # If target dir does not exists, it should succeed + rm -rf $remote_tgt + mrename $remote_src $remote_tgt || + error "rename remote dirs(tgt dir does not exists) failed!" + rm -rf $DIR/$tdir || error "Can not delete directories" } run_test 24z "rename one remote dir to another remote dir should fail" @@ -1225,7 +1270,7 @@ run_test 27e "setstripe existing file (should return error) ======" test_27f() { test_mkdir -p $DIR/d27 $SETSTRIPE -S 100 -i 0 -c 1 $DIR/d27/fbad && error "setstripe failed" - dd if=/dev/zero of=$DIR/d27/f12 bs=4k count=4 || error "dd failed" + dd if=/dev/zero of=$DIR/d27/fbad bs=4k count=4 || error "dd failed" $GETSTRIPE $DIR/d27/fbad || error "$GETSTRIPE failed" } run_test 27f "setstripe with bad stripe size (should return error)" @@ -1478,21 +1523,24 @@ test_27t() { # bug 10864 run_test 27t "check that utils parse path correctly" test_27u() { # bug 4900 - [ "$OSTCOUNT" -lt "2" ] && skip_env "too few OSTs" && return - remote_mds_nodsh && skip "remote MDS with nodsh" && return + [ "$OSTCOUNT" -lt "2" ] && skip_env "too few OSTs" && return + remote_mds_nodsh && skip "remote MDS with nodsh" && return + local index + local list=$(comma_list $(mdts_nodes)) #define OBD_FAIL_MDS_OSC_PRECREATE 0x139 - do_facet $SINGLEMDS lctl set_param fail_loc=0x139 - test_mkdir -p $DIR/$tdir - createmany -o $DIR/$tdir/t- 1000 - do_facet $SINGLEMDS lctl set_param fail_loc=0 + do_nodes $list $LCTL set_param fail_loc=0x139 + test_mkdir -p $DIR/$tdir + rm -rf $DIR/$tdir/* + createmany -o $DIR/$tdir/t- 1000 + do_nodes $list $LCTL set_param fail_loc=0 - TLOG=$DIR/$tfile.getstripe - $GETSTRIPE $DIR/$tdir > $TLOG - OBJS=`awk -vobj=0 '($1 == 0) { obj += 1 } END { print obj;}' $TLOG` - unlinkmany $DIR/$tdir/t- 1000 - [ $OBJS -gt 0 ] && \ - error "$OBJS objects created on OST-0. See $TLOG" || pass + TLOG=$DIR/$tfile.getstripe + $GETSTRIPE $DIR/$tdir > $TLOG + OBJS=`awk -vobj=0 '($1 == 0) { obj += 1 } END { print obj;}' $TLOG` + unlinkmany $DIR/$tdir/t- 1000 + [ $OBJS -gt 0 ] && \ + error "$OBJS objects created on OST-0. See $TLOG" || pass } run_test 27u "skip object creation on OSC w/o objects ==========" @@ -1638,6 +1686,19 @@ test_27y() { do_facet $SINGLEMDS lctl --device %$OSC activate fi done + + # all osp devices get activated, hence -1 stripe count restored + local stripecnt=0 + + # sleep 2*lod_qos_maxage seconds waiting for lod qos to notice osp + # devices get activated. + sleep_maxage + $SETSTRIPE -c -1 $DIR/$tfile + stripecnt=$($GETSTRIPE -c $DIR/$tfile) + rm -f $DIR/$tfile + [ $stripecnt -ne $OSTCOUNT ] && + error "Of $OSTCOUNT OSTs, only $stripecnt is available" + return 0 } run_test 27y "create files while OST0 is degraded and the rest inactive" @@ -1671,8 +1732,9 @@ check_seq_oid() [ "$obdidx" = "obdidx" ] && have_obdidx=true && continue $have_obdidx || continue - local ost=$((obdidx + 1)) - local dev=$(ostdevname $ost) + local ost=$((obdidx + 1)) + local dev=$(ostdevname $ost) + local oid_hex if [ $(facet_fstype ost$ost) != ldiskfs ]; then echo "Currently only works with ldiskfs-based OSTs" @@ -1682,29 +1744,38 @@ check_seq_oid() log "want: stripe:$stripe_nr ost:$obdidx oid:$oid/$hex seq:$seq" #don't unmount/remount the OSTs if we don't need to do that - #local dir=$(facet_mntpt ost$ost) - #stop ost$dev - #do_facet ost$dev mount -t $FSTYPE $dev $dir $OST_MOUNT_OPTS || - # { error "mounting $dev as $FSTYPE failed"; return 3; } - #local obj_file=$(do_facet ost$ost find $dir/O/$seq -name $oid) - #local ff=$(do_facet ost$ost $LL_DECODE_FILTER_FID $obj_file) + # LU-2577 changes filter_fid to be smaller, so debugfs needs + # update too, until that use mount/ll_decode_filter_fid/mount + local dir=$(facet_mntpt ost$ost) + local opts=${OST_MOUNT_OPTS} + + if ! do_facet ost$ost test -b ${dev}; then + opts=$(csa_add "$opts" -o loop) + fi + + stop ost$ost + do_facet ost$ost mount -t $(facet_fstype ost$ost) $opts $dev $dir || + { error "mounting $dev as $FSTYPE failed"; return 3; } + seq=$(echo $seq | sed -e "s/^0x//g") if [ $seq == 0 ]; then oid_hex=$(echo $oid) else oid_hex=$(echo $hex | sed -e "s/^0x//g") fi - local obj_file="O/$seq/d$((oid %32))/$oid_hex" - local ff=$(do_facet ost$ost "$DEBUGFS -c -R 'stat $obj_file' \ - $dev 2>/dev/null" | grep "parent=") + local obj_file=$(do_facet ost$ost find $dir/O/$seq -name $oid_hex) + local ff=$(do_facet ost$ost $LL_DECODE_FILTER_FID $obj_file) + do_facet ost$ost umount -d $dir + start ost$ost $dev $OST_MOUNT_OPTS + + # re-enable when debugfs will understand new filter_fid + #local ff=$(do_facet ost$ost "$DEBUGFS -c -R 'stat $obj_file' \ + # $dev 2>/dev/null" | grep "parent=") [ -z "$ff" ] && error "$obj_file: no filter_fid info" echo "$ff" | sed -e 's#.*objid=#got: objid=#' - #do_facet ost$ost umount -d $dir - #start ost$ost $dev $OST_MOUNT_OPTS - # /mnt/O/0/d23/23: objid=23 seq=0 parent=[0x200000400:0x1e:0x1] # fid: objid=23 seq=0 parent=[0x200000400:0x1e:0x0] stripe=1 local ff_parent=$(echo $ff|sed -e 's/.*parent=.//') @@ -1768,6 +1839,67 @@ test_27A() { # b=19102 } run_test 27A "check filesystem-wide default LOV EA values" +test_27B() { # LU-2523 + test_mkdir -p $DIR/$tdir + rm -f $DIR/$tdir/f0 $DIR/$tdir/f1 + touch $DIR/$tdir/f0 + # open f1 with O_LOV_DELAY_CREATE + # rename f0 onto f1 + # call setstripe ioctl on open file descriptor for f1 + # close + multiop $DIR/$tdir/f1 oO_RDWR:O_CREAT:O_LOV_DELAY_CREATE:nB1c \ + $DIR/$tdir/f0 + + rm -f $DIR/$tdir/f1 + # open f1 with O_LOV_DELAY_CREATE + # unlink f1 + # call setstripe ioctl on open file descriptor for f1 + # close + multiop $DIR/$tdir/f1 oO_RDWR:O_CREAT:O_LOV_DELAY_CREATE:uB1c + + # Allow multiop to fail in imitation of NFS's busted semantics. + true +} +run_test 27B "call setstripe on open unlinked file/rename victim" + +test_27C() { #LU-2871 + [ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return + + declare -a ost_idx + local index + local found + local i + local j + + test_mkdir -p $DIR/$tdir + cd $DIR/$tdir + for i in $(seq 0 $((OSTCOUNT - 1))); do + # set stripe across all OSTs starting from OST$i + $SETSTRIPE -i $i -c -1 $tfile$i + # get striping information + ost_idx=($($GETSTRIPE $tfile$i | + tail -n $((OSTCOUNT + 1)) | awk '{print $1}')) + echo ${ost_idx[@]} + + # check the layout + [ ${#ost_idx[@]} -eq $OSTCOUNT ] || + error "${#ost_idx[@]} != $OSTCOUNT" + + for index in $(seq 0 $((OSTCOUNT - 1))); do + found=0 + for j in $(echo ${ost_idx[@]}); do + if [ $index -eq $j ]; then + found=1 + break + fi + done + [ $found = 1 ] || + error "Can not find $index in ${ost_idx[@]}" + done + done +} +run_test 27C "check full striping across all OSTs" + # createtest also checks that device nodes are created and # then visible correctly (#2091) test_28() { # bug 2091 @@ -2300,7 +2432,7 @@ test_33c() { for ostnum in $(seq $OSTCOUNT); do # test-framework's OST numbering is one-based, while Lustre's # is zero-based - ostname=$(printf "$FSNAME-OST%.4d" $((ostnum - 1))) + ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) # Parsing llobdstat's output sucks; we could grep the /proc # path, but that's likely to not be as portable as using the # llobdstat utility. So we parse lctl output instead. @@ -2324,7 +2456,7 @@ test_33c() { # Total up write_bytes after writing. We'd better find non-zeros. for ostnum in $(seq $OSTCOUNT); do - ostname=$(printf "$FSNAME-OST%.4d" $((ostnum - 1))) + ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) write_bytes=$(do_facet ost$ostnum lctl get_param -n \ obdfilter/$ostname/stats | awk '/^write_bytes/ {print $7}' ) @@ -2339,7 +2471,7 @@ test_33c() { if $all_zeros then for ostnum in $(seq $OSTCOUNT); do - ostname=$(printf "$FSNAME-OST%.4d" $((ostnum - 1))) + ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) echo "Check that write_bytes is present in obdfilter/*/stats:" do_facet ost$ostnum lctl get_param -n \ obdfilter/$ostname/stats @@ -2461,6 +2593,8 @@ test_34h() { local sz=1000 dd if=/dev/zero of=$DIR/$tfile bs=1M count=10 || error + sync # Flush the cache so that multiop below does not block on cache + # flush when getting the group lock $MULTIOP $DIR/$tfile OG${gid}T${sz}g${gid}c & MULTIPID=$! sleep 2 @@ -3666,27 +3800,47 @@ test_53() { remote_ost_nodsh && skip "remote OST with nodsh" && return local param + local param_seq local ostname local mds_last + local mds_last_seq local ost_last + local ost_last_seq + local ost_last_id local ostnum local node + local found=0 # only test MDT0000 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS) for value in $(do_facet $SINGLEMDS lctl get_param osc.$mdtosc.prealloc_last_id) ; do - param=`echo ${value[0]} | cut -d "=" -f1` - ostname=`echo $param | cut -d "." -f2 | cut -d - -f 1-2` - mds_last=$(do_facet $SINGLEMDS lctl get_param -n $param) + param=$(echo ${value[0]} | cut -d "=" -f1) + ostname=$(echo $param | cut -d "." -f2 | cut -d - -f 1-2) + param_seq=$(echo ${param} | + sed -e s/prealloc_last_id/prealloc_last_seq/g) + mds_last_seq=$(do_facet $SINGLEMDS lctl get_param -n $param_seq) + mds_last=$(do_facet $SINGLEMDS lctl get_param -n $param) + ostnum=$(index_from_ostuuid ${ostname}_UUID) node=$(facet_active_host ost$((ostnum+1))) param="obdfilter.$ostname.last_id" - ost_last=$(do_node $node lctl get_param -n $param | head -n 1) - echo "$ostname.last_id=$ost_last ; MDS.last_id=$mds_last" - if [ $ost_last != $mds_last ]; then - error "$ostname.last_id=$ost_last ; MDS.last_id=$mds_last" - fi + for ost_last in $(do_node $node lctl get_param -n $param) ; do + echo "$ostname.last_id=$ost_last ;MDS.last_id=$mds_last" + ost_last_id=$(echo $ost_last | awk -F':' '{print $2}' | + sed -e "s/^0x//g") + ost_last_seq=$(echo $ost_last | awk -F':' '{print $1}') + if [ $ost_last_seq = $mds_last_seq ]; then + if [ $ost_last_id != $mds_last ]; then + error "$ost_last != $mds_last_id" + else + found=1 + break + fi + fi + done done + [ $found = 0 ] && error "can not match last_seq/last_id for $mdtosc" + return 0 } run_test 53 "verify that MDS and OSTs agree on pre-creation ====" @@ -4204,11 +4358,13 @@ test_56u() { # LU-611 [ $NUMS -eq $EXPECTED ] || error "\"$CMD\" wrong: found $NUMS, expected $EXPECTED" - EXPECTED=$(((NUMDIRS + 1) * NUMFILES + ONESTRIPE)) - CMD="$LFIND -stripe-index 0,1 -type f $TDIR" - NUMS=$($CMD | wc -l) - [ $NUMS -eq $EXPECTED ] || - error "\"$CMD\" wrong: found $NUMS, expected $EXPECTED" + if [ $OSTCOUNT -gt 1 ]; then + EXPECTED=$(((NUMDIRS + 1) * NUMFILES + ONESTRIPE)) + CMD="$LFIND -stripe-index 0,1 -type f $TDIR" + NUMS=$($CMD | wc -l) + [ $NUMS -eq $EXPECTED ] || + error "\"$CMD\" wrong: found $NUMS, expected $EXPECTED" + fi } run_test 56u "check lfs find -stripe-index works" @@ -4329,6 +4485,29 @@ test_56w() { } run_test 56w "check lfs_migrate -c stripe_count works" +test_56x() { + check_swap_layouts_support && return 0 + [ "$OSTCOUNT" -lt "2" ] && + skip_env "need 2 OST, skipping test" && return + + local dir0=$DIR/$tdir/$testnum + mkdir -p $dir0 || error "creating dir $dir0" + + local ref1=/etc/passwd + local file1=$dir0/file1 + + $SETSTRIPE -c 2 $file1 + cp $ref1 $file1 + $LFS migrate -c 1 $file1 || error "migrate failed rc = $?" + stripe=$($GETSTRIPE -c $file1) + [[ $stripe == 1 ]] || error "stripe of $file1 is $stripe != 1" + cmp $file1 $ref1 || error "content mismatch $file1 differs from $ref1" + + # clean up + rm -f $file1 +} +run_test 56x "lfs migration support" + test_57a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" && return # note test will not do anything if MDS is not local @@ -6088,7 +6267,7 @@ test_102k() { local default_size=`$GETSTRIPE -S $test_kdir` local default_count=`$GETSTRIPE -c $test_kdir` local default_offset=`$GETSTRIPE -i $test_kdir` - $SETSTRIPE -S 65536 -i 1 -c $OSTCOUNT $test_kdir || + $SETSTRIPE -S 65536 -i 0 -c $OSTCOUNT $test_kdir || error 'dir setstripe failed' setfattr -n trusted.lov $test_kdir local stripe_size=`$GETSTRIPE -S $test_kdir` @@ -6162,8 +6341,12 @@ test_103 () { run_acl_subtest inheritance || error "inheritance test failed" rm -f make-tree - echo "LU-974 ignore umask when acl is enabled..." - run_acl_subtest 974 || error "LU-974 test failed" + echo "LU-974 ignore umask when acl is enabled..." + run_acl_subtest 974 || error "LU-974 test failed" + if [ $MDSCOUNT -ge 2 ]; then + run_acl_subtest 974_remote || + error "LU-974 test failed under remote dir" + fi echo "LU-2561 newly created file is same size as directory..." run_acl_subtest 2561 || error "LU-2561 test failed" @@ -6874,6 +7057,15 @@ test_118l() } run_test 118l "fsync dir =========" +test_118m() # LU-3066 +{ + [ $PARALLEL == "yes" ] && skip "skip parallel run" && return + test_mkdir -p $DIR/$tdir + $MULTIOP $DIR/$tdir DY || error "fdatasync dir failed" + rm -rf $DIR/$tdir +} +run_test 118m "fdatasync dir =========" + [ "$SLOW" = "no" ] && [ -n "$OLD_RESENDCOUNT" ] && set_resend_count $OLD_RESENDCOUNT test_119a() # bug 11737 @@ -6945,9 +7137,13 @@ test_120a() { test_mkdir -p $DIR/$tdir [ -z "`lctl get_param -n mdc.*.connect_flags | grep early_lock_cancel`" ] && \ skip "no early lock cancel on server" && return 0 - lru_resize_disable mdc - lru_resize_disable osc - cancel_lru_locks mdc + + lru_resize_disable mdc + lru_resize_disable osc + cancel_lru_locks mdc + # asynchronous object destroy at MDT could cause bl ast to client + cancel_lru_locks osc + stat $DIR/$tdir > /dev/null can1=`lctl get_param -n ldlm.services.ldlm_canceld.stats | awk '/ldlm_cancel/ {print $2}'` blk1=`lctl get_param -n ldlm.services.ldlm_cbd.stats | awk '/ldlm_bl_callback/ {print $2}'` @@ -7638,6 +7834,9 @@ test_130b() { [ "$OSTCOUNT" -lt "2" ] && skip_env "skipping FIEMAP on 2-stripe file test" && return + [ "$OSTCOUNT" -ge "10" ] && + skip_env "skipping FIEMAP with >= 10 OSTs" && return + local filefrag_op=$(filefrag -e 2>&1 | grep "invalid option") [ -n "$filefrag_op" ] && skip_env "filefrag does not support FIEMAP" && return @@ -7694,6 +7893,9 @@ test_130c() { [ "$OSTCOUNT" -lt "2" ] && skip_env "skipping FIEMAP on 2-stripe file" && return + [ "$OSTCOUNT" -ge "10" ] && + skip_env "skipping FIEMAP with >= 10 OSTs" && return + filefrag_op=$(filefrag -e 2>&1 | grep "invalid option") [ -n "$filefrag_op" ] && skip "filefrag does not support FIEMAP" && return @@ -7753,6 +7955,9 @@ run_test 130c "FIEMAP (2-stripe file with hole)" test_130d() { [ "$OSTCOUNT" -lt "3" ] && skip_env "skipping FIEMAP on N-stripe file test" && return + [ "$OSTCOUNT" -ge "10" ] && + skip_env "skipping FIEMAP with >= 10 OSTs" && return + filefrag_op=$(filefrag -e 2>&1 | grep "invalid option") [ -n "$filefrag_op" ] && skip "filefrag does not support FIEMAP" && return @@ -7804,6 +8009,9 @@ run_test 130d "FIEMAP (N-stripe file)" test_130e() { [ "$OSTCOUNT" -lt "2" ] && skip_env "skipping continuation FIEMAP test" && return + [ "$OSTCOUNT" -ge "10" ] && + skip_env "skipping FIEMAP with >= 10 OSTs" && return + filefrag_op=$(filefrag -e 2>&1 | grep "invalid option") [ -n "$filefrag_op" ] && skip "filefrag does not support FIEMAP" && return @@ -8155,20 +8363,20 @@ get_rename_size() { test_133d() { [ $PARALLEL == "yes" ] && skip "skip parallel run" && return - remote_ost_nodsh && skip "remote OST with nodsh" && return - remote_mds_nodsh && skip "remote MDS with nodsh" && return - do_facet $SINGLEMDS $LCTL list_param mdt.*.rename_stats || - { skip "MDS doesn't support rename stats"; return; } + remote_ost_nodsh && skip "remote OST with nodsh" && return + remote_mds_nodsh && skip "remote MDS with nodsh" && return + do_facet $SINGLEMDS $LCTL list_param mdt.*.rename_stats || + { skip "MDS doesn't support rename stats"; return; } - local testdir1=$DIR/${tdir}/stats_testdir1 - local testdir2=$DIR/${tdir}/stats_testdir2 + local testdir1=$DIR/${tdir}/stats_testdir1 + local testdir2=$DIR/${tdir}/stats_testdir2 - do_facet $SINGLEMDS $LCTL set_param mdt.*.rename_stats=clear + do_facet $SINGLEMDS $LCTL set_param mdt.*.rename_stats=clear - test_mkdir -p ${testdir1} || error "mkdir failed" - test_mkdir -p ${testdir2} || error "mkdir failed" + mkdir -p ${testdir1} || error "mkdir failed" + mkdir -p ${testdir2} || error "mkdir failed" - createmany -o $testdir1/test 512 || error "createmany failed" + createmany -o $testdir1/test 512 || error "createmany failed" # check samedir rename size mv ${testdir1}/test0 ${testdir1}/test_0 @@ -8187,18 +8395,18 @@ test_133d() { echo "source rename dir size: ${testdir1_size}" echo "target rename dir size: ${testdir2_size}" - local cmd="do_facet $SINGLEMDS $LCTL get_param mdt.*.rename_stats" - eval $cmd || error "$cmd failed" - local samedir=$($cmd | grep 'same_dir') - local same_sample=$(get_rename_size $testdir1_size) - [ -z "$samedir" ] && error "samedir_rename_size count error" - [ "$same_sample" -eq 1 ] || error "samedir_rename_size error $same_sample" - echo "Check same dir rename stats success" + local cmd="do_facet $SINGLEMDS $LCTL get_param mdt.*.rename_stats" + eval $cmd || error "$cmd failed" + local samedir=$($cmd | grep 'same_dir') + local same_sample=$(get_rename_size $testdir1_size) + [ -z "$samedir" ] && error "samedir_rename_size count error" + [ "$same_sample" -eq 1 ] || error "samedir_rename_size error $same_sample" + echo "Check same dir rename stats success" - do_facet $SINGLEMDS $LCTL set_param mdt.*.rename_stats=clear + do_facet $SINGLEMDS $LCTL set_param mdt.*.rename_stats=clear - # check crossdir rename size - mv ${testdir1}/test_0 ${testdir2}/test_0 + # check crossdir rename size + mv ${testdir1}/test_0 ${testdir2}/test_0 testdir1_size=$(ls -l $DIR/${tdir} | awk '/stats_testdir1/ {print $5}') @@ -8214,15 +8422,15 @@ test_133d() { echo "source rename dir size: ${testdir1_size}" echo "target rename dir size: ${testdir2_size}" - eval $cmd || error "$cmd failed" - local crossdir=$($cmd | grep 'crossdir') - local src_sample=$(get_rename_size $testdir1_size crossdir_src) - local tgt_sample=$(get_rename_size $testdir2_size crossdir_tgt) - [ -z "$crossdir" ] && error "crossdir_rename_size count error" - [ "$src_sample" -eq 1 ] || error "crossdir_rename_size error $src_sample" - [ "$tgt_sample" -eq 1 ] || error "crossdir_rename_size error $tgt_sample" - echo "Check cross dir rename stats success" - rm -rf $DIR/${tdir} + eval $cmd || error "$cmd failed" + local crossdir=$($cmd | grep 'crossdir') + local src_sample=$(get_rename_size $testdir1_size crossdir_src) + local tgt_sample=$(get_rename_size $testdir2_size crossdir_tgt) + [ -z "$crossdir" ] && error "crossdir_rename_size count error" + [ "$src_sample" -eq 1 ] || error "crossdir_rename_size error $src_sample" + [ "$tgt_sample" -eq 1 ] || error "crossdir_rename_size error $tgt_sample" + echo "Check cross dir rename stats success" + rm -rf $DIR/${tdir} } run_test 133d "Verifying rename_stats ========================================" @@ -8272,6 +8480,36 @@ test_133e() { } run_test 133e "Verifying OST {read,write}_bytes nid stats =================" +test_133f() { + local proc_dirs="/proc/fs/lustre/ /proc/sys/lnet/ /proc/sys/lustre/" + local facet + + # First without trusting modes. + find $proc_dirs \ + -exec cat '{}' \; &> /dev/null + + # Second verifying readability. + find $proc_dirs \ + -type f \ + -readable \ + -exec cat '{}' \; > /dev/null || + error "proc file read failed" + + for facet in $SINGLEMDS ost1; do + do_facet $facet find $proc_dirs \ + -not -name req_history \ + -exec cat '{}' \\\; &> /dev/null + + do_facet $facet find $proc_dirs \ + -not -name req_history \ + -type f \ + -readable \ + -exec cat '{}' \\\; > /dev/null || + error "proc file read failed" + done +} +run_test 133f "Check for LBUGs/Oopses/unreadable files in /proc" + test_140() { #bug-17379 [ $PARALLEL == "yes" ] && skip "skip parallel run" && return test_mkdir -p $DIR/$tdir || error "Creating dir $DIR/$tdir" @@ -8345,7 +8583,8 @@ run_test 150 "truncate/append tests" function roc_hit() { local list=$(comma_list $(osts_nodes)) - + #debug temp debug for LU-2902: lets see what values we get back + echo $(get_osd_param $list '' stats) 1>&2 echo $(get_osd_param $list '' stats | awk '/'cache_hit'/ {sum+=$2} END {print sum}') } @@ -8383,15 +8622,29 @@ test_151() { set_osd_param $list '' writethrough_cache_enable 1 - # pages should be in the case right after write - dd if=/dev/urandom of=$DIR/$tfile bs=4k count=$CPAGES || error "dd failed" - local BEFORE=`roc_hit` - cancel_lru_locks osc - cat $DIR/$tfile >/dev/null - local AFTER=`roc_hit` - if ! let "AFTER - BEFORE == CPAGES"; then - error "NOT IN CACHE: before: $BEFORE, after: $AFTER" - fi + # check write cache is enabled on all obdfilters + if get_osd_param $list '' writethrough_cache_enable | grep 0; then + echo "oss write cache is NOT enabled" + return 0 + fi + +#define OBD_FAIL_OBD_NO_LRU 0x609 + do_nodes $list $LCTL set_param fail_loc=0x609 + + # pages should be in the case right after write + dd if=/dev/urandom of=$DIR/$tfile bs=4k count=$CPAGES || + error "dd failed" + + local BEFORE=`roc_hit` + cancel_lru_locks osc + cat $DIR/$tfile >/dev/null + local AFTER=`roc_hit` + + do_nodes $list $LCTL set_param fail_loc=0 + + if ! let "AFTER - BEFORE == CPAGES"; then + error "NOT IN CACHE: before: $BEFORE, after: $AFTER" + fi # the following read invalidates the cache cancel_lru_locks osc @@ -8441,18 +8694,10 @@ test_153() { } run_test 153 "test if fdatasync does not crash =======================" -test_154() { - [ $PARALLEL == "yes" ] && skip "skip parallel run" && return - [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.2.51) ]] || - { skip "Need MDS version at least 2.2.51"; return 0; } - - cp /etc/hosts $DIR/$tfile - - fid=$($LFS path2fid $DIR/$tfile) - rc=$? - [ $rc -ne 0 ] && error "error: could not get fid for $DIR/$tfile." - - ffid=$DIR/.lustre/fid/$fid +dot_lustre_fid_permission_check() { + local fid=$1 + local ffid=$MOUNT/.lustre/fid/$fid + local test_dir=$2 echo "stat fid $fid" stat $ffid > /dev/null || error "stat $ffid failed." @@ -8465,14 +8710,16 @@ test_154() { echo "append write to fid $fid" cat /etc/hosts >> $ffid || error "append write $ffid failed." echo "rename fid $fid" - mv $ffid $DIR/$tfile.1 && error "rename $ffid to $tfile.1 should fail." - touch $DIR/$tfile.1 - mv $DIR/$tfile.1 $ffid && error "rename $tfile.1 to $ffid should fail." - rm -f $DIR/$tfile.1 + mv $ffid $test_dir/$tfile.1 && + error "rename $ffid to $tfile.1 should fail." + touch $test_dir/$tfile.1 + mv $test_dir/$tfile.1 $ffid && + error "rename $tfile.1 to $ffid should fail." + rm -f $test_dir/$tfile.1 echo "truncate fid $fid" $TRUNCATE $ffid 777 || error "truncate $ffid failed." echo "link fid $fid" - ln -f $ffid $DIR/tfile.lnk || error "link $ffid failed." + ln -f $ffid $test_dir/tfile.lnk || error "link $ffid failed." if [ -n $(lctl get_param -n mdc.*-mdc-*.connect_flags | grep acl) ]; then echo "setfacl fid $fid" setfacl -R -m u:bin:rwx $ffid || error "setfacl $ffid failed." @@ -8480,80 +8727,147 @@ test_154() { getfacl $ffid >/dev/null || error "getfacl $ffid failed." fi echo "unlink fid $fid" - unlink $DIR/.lustre/fid/$fid && error "unlink $ffid should fail." + unlink $MOUNT/.lustre/fid/$fid && error "unlink $ffid should fail." echo "mknod fid $fid" mknod $ffid c 1 3 && error "mknod $ffid should fail." fid=[0xf00000400:0x1:0x0] - ffid=$DIR/.lustre/fid/$fid + ffid=$MOUNT/.lustre/fid/$fid echo "stat non-exist fid $fid" stat $ffid > /dev/null && error "stat non-exist $ffid should fail." echo "write to non-exist fid $fid" cat /etc/hosts > $ffid && error "write non-exist $ffid should fail." echo "link new fid $fid" - ln $DIR/$tfile $ffid && error "link $ffid should fail." + ln $test_dir/$tfile $ffid && error "link $ffid should fail." - test_mkdir -p $DIR/$tdir - touch $DIR/$tdir/$tfile - fid=$($LFS path2fid $DIR/$tdir) + mkdir -p $test_dir/$tdir + touch $test_dir/$tdir/$tfile + fid=$($LFS path2fid $test_dir/$tdir) rc=$? - [ $rc -ne 0 ] && error "error: could not get fid for $DIR/$tfile." + [ $rc -ne 0 ] && + error "error: could not get fid for $test_dir/$dir/$tfile." - ffid=$DIR/.lustre/fid/$fid + ffid=$MOUNT/.lustre/fid/$fid echo "ls $fid" ls $ffid > /dev/null || error "ls $ffid failed." echo "touch $fid/$tfile.1" touch $ffid/$tfile.1 || error "touch $ffid/$tfile.1 failed." - echo "touch $DIR/.lustre/fid/$tfile" - touch $DIR/.lustre/fid/$tfile && \ - error "touch $DIR/.lustre/fid/$tfile should fail." + echo "touch $MOUNT/.lustre/fid/$tfile" + touch $MOUNT/.lustre/fid/$tfile && \ + error "touch $MOUNT/.lustre/fid/$tfile should fail." - echo "setxattr to $DIR/.lustre/fid" - setfattr -n trusted.name1 -v value1 $DIR/.lustre/fid && - error "setxattr should fail." + echo "setxattr to $MOUNT/.lustre/fid" + setfattr -n trusted.name1 -v value1 $MOUNT/.lustre/fid - echo "listxattr for $DIR/.lustre/fid" - getfattr -d -m "^trusted" $DIR/.lustre/fid && - error "listxattr should fail." + echo "listxattr for $MOUNT/.lustre/fid" + getfattr -d -m "^trusted" $MOUNT/.lustre/fid - echo "delxattr from $DIR/.lustre/fid" - setfattr -x trusted.name1 $DIR/.lustre/fid && - error "delxattr should fail." + echo "delxattr from $MOUNT/.lustre/fid" + setfattr -x trusted.name1 $MOUNT/.lustre/fid - echo "touch invalid fid: $DIR/.lustre/fid/[0x200000400:0x2:0x3]" - touch $DIR/.lustre/fid/[0x200000400:0x2:0x3] && + echo "touch invalid fid: $MOUNT/.lustre/fid/[0x200000400:0x2:0x3]" + touch $MOUNT/.lustre/fid/[0x200000400:0x2:0x3] && error "touch invalid fid should fail." - echo "touch non-normal fid: $DIR/.lustre/fid/[0x1:0x2:0x0]" - touch $DIR/.lustre/fid/[0x1:0x2:0x0] && + echo "touch non-normal fid: $MOUNT/.lustre/fid/[0x1:0x2:0x0]" + touch $MOUNT/.lustre/fid/[0x1:0x2:0x0] && error "touch non-normal fid should fail." - echo "rename $tdir to $DIR/.lustre/fid" - mrename $DIR/$tdir $DIR/.lustre/fid && - error "rename to $DIR/.lustre/fid should fail." + echo "rename $tdir to $MOUNT/.lustre/fid" + mrename $test_dir/$tdir $MOUNT/.lustre/fid && + error "rename to $MOUNT/.lustre/fid should fail." echo "rename .lustre to itself" - fid=$($LFS path2fid $DIR) - mrename $DIR/.lustre $DIR/.lustre/fid/$fid/.lustre && + fid=$($LFS path2fid $MOUNT) + mrename $MOUNT/.lustre $MOUNT/.lustre/fid/$fid/.lustre && error "rename .lustre to itself should fail." - $OPENFILE -f O_LOV_DELAY_CREATE:O_CREAT $DIR/$tfile-2 - fid=$($LFS path2fid $DIR/$tfile-2) - echo "cp /etc/passwd $DIR/.lustre/fid/$fid" - cp /etc/passwd $DIR/.lustre/fid/$fid && + local old_obf_mode=$(stat --format="%a" $DIR/.lustre/fid) + local new_obf_mode=777 + + echo "change mode of $DIR/.lustre/fid to $new_obf_mode" + chmod $new_obf_mode $DIR/.lustre/fid || + error "chmod $new_obf_mode $DIR/.lustre/fid failed" + + local obf_mode=$(stat --format=%a $DIR/.lustre/fid) + [ $obf_mode -eq $new_obf_mode ] || + error "stat $DIR/.lustre/fid returned wrong mode $obf_mode" + + echo "restore mode of $DIR/.lustre/fid to $old_obf_mode" + chmod $old_obf_mode $DIR/.lustre/fid || + error "chmod $old_obf_mode $DIR/.lustre/fid failed" + + $OPENFILE -f O_LOV_DELAY_CREATE:O_CREAT $test_dir/$tfile-2 + fid=$($LFS path2fid $test_dir/$tfile-2) + echo "cp /etc/passwd $MOUNT/.lustre/fid/$fid" + cp /etc/passwd $MOUNT/.lustre/fid/$fid && error "create lov data thru .lustre should fail." - echo "cp /etc/passwd $DIR/$tfile-2" - cp /etc/passwd $DIR/$tfile-2 || error "copy to $DIR/$tfile-2 failed." - echo "diff /etc/passwd $DIR/.lustre/fid/$fid" - diff /etc/passwd $DIR/.lustre/fid/$fid || - error "diff /etc/passwd $DIR/.lustre/fid/$fid failed." + echo "cp /etc/passwd $test_dir/$tfile-2" + cp /etc/passwd $test_dir/$tfile-2 || + error "copy to $test_dir/$tfile-2 failed." + echo "diff /etc/passwd $MOUNT/.lustre/fid/$fid" + diff /etc/passwd $MOUNT/.lustre/fid/$fid || + error "diff /etc/passwd $MOUNT/.lustre/fid/$fid failed." + + rm -rf $test_dir/tfile.lnk + rm -rf $test_dir/$tfile-2 +} + +test_154a() { + [ $PARALLEL == "yes" ] && skip "skip parallel run" && return + [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.2.51) ]] || + { skip "Need MDS version at least 2.2.51"; return 0; } + + cp /etc/hosts $DIR/$tfile + + fid=$($LFS path2fid $DIR/$tfile) + rc=$? + [ $rc -ne 0 ] && error "error: could not get fid for $DIR/$tfile." + + dot_lustre_fid_permission_check "$fid" $DIR || + error "dot lustre permission check $fid failed" + + rm -rf $MOUNT/.lustre && error ".lustre is not allowed to be unlinked" + + touch $MOUNT/.lustre/file && + error "creation is not allowed under .lustre" + + mkdir $MOUNT/.lustre/dir && + error "mkdir is not allowed under .lustre" - echo "Open-by-FID succeeded" + rm -rf $DIR/$tfile +} +run_test 154a "Open-by-FID" + +test_154b() { + [ $PARALLEL == "yes" ] && skip "skip parallel run" && return + [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.2.51) ]] || + { skip "Need MDS version at least 2.2.51"; return 0; } + + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + + local remote_dir=$DIR/$tdir/remote_dir + local MDTIDX=1 + local rc=0 + + mkdir -p $DIR/$tdir + $LFS mkdir -i $MDTIDX $remote_dir || + error "create remote directory failed" + + cp /etc/hosts $remote_dir/$tfile + + fid=$($LFS path2fid $remote_dir/$tfile) + rc=$? + [ $rc -ne 0 ] && error "error: could not get fid for $remote_dir/$tfile" + + dot_lustre_fid_permission_check "$fid" $remote_dir || + error "dot lustre permission check $fid failed" + rm -rf $DIR/$tdir } -run_test 154 "Open-by-FID" +run_test 154b "Open-by-FID for remote directory" test_155_small_load() { local temp=$TMP/$tfile @@ -8930,7 +9244,7 @@ test_160() { } run_test 160 "changelog sanity" -test_161() { +test_161a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" && return test_mkdir -p $DIR/$tdir cp /etc/hosts $DIR/$tdir/$tfile @@ -8940,11 +9254,11 @@ test_161() { ln $DIR/$tdir/$tfile $DIR/$tdir/foo2/zachary ln $DIR/$tdir/$tfile $DIR/$tdir/foo1/luna ln $DIR/$tdir/$tfile $DIR/$tdir/foo2/thor - local FID=$($LFS path2fid $DIR/$tdir/$tfile | tr -d '[') - if [ "$($LFS fid2path $DIR $FID | wc -l)" != "5" ]; then - $LFS fid2path $DIR $FID - err17935 "bad link ea" - fi + local FID=$($LFS path2fid $DIR/$tdir/$tfile | tr -d '[]') + if [ "$($LFS fid2path $DIR $FID | wc -l)" != "5" ]; then + $LFS fid2path $DIR $FID + err17935 "bad link ea" + fi # middle rm $DIR/$tdir/foo2/zachary # last @@ -8970,7 +9284,59 @@ test_161() { unlinkmany $DIR/$tdir/foo2/$longname 1000 || \ error "failed to unlink many hardlinks" } -run_test 161 "link ea sanity" +run_test 161a "link ea sanity" + +test_161b() { + [ $PARALLEL == "yes" ] && skip "skip parallel run" && return + [ $MDSCOUNT -lt 2 ] && + skip "skipping remote directory test" && return + local MDTIDX=1 + local remote_dir=$DIR/$tdir/remote_dir + + mkdir -p $DIR/$tdir + $LFS mkdir -i $MDTIDX $remote_dir || + error "create remote directory failed" + + cp /etc/hosts $remote_dir/$tfile + mkdir -p $remote_dir/foo1 + mkdir -p $remote_dir/foo2 + ln $remote_dir/$tfile $remote_dir/foo1/sofia + ln $remote_dir/$tfile $remote_dir/foo2/zachary + ln $remote_dir/$tfile $remote_dir/foo1/luna + ln $remote_dir/$tfile $remote_dir/foo2/thor + + local FID=$($LFS path2fid $remote_dir/$tfile | tr -d '[' | + tr -d ']') + if [ "$($LFS fid2path $DIR $FID | wc -l)" != "5" ]; then + $LFS fid2path $DIR $FID + err17935 "bad link ea" + fi + # middle + rm $remote_dir/foo2/zachary + # last + rm $remote_dir/foo2/thor + # first + rm $remote_dir/$tfile + # rename + mv $remote_dir/foo1/sofia $remote_dir/foo2/maggie + local link_path=$($LFS fid2path $FSNAME --link 1 $FID) + if [ "$DIR/$link_path" != "$remote_dir/foo2/maggie" ]; then + $LFS fid2path $DIR $FID + err17935 "bad link rename" + fi + rm $remote_dir/foo2/maggie + + # overflow the EA + local longname=filename_avg_len_is_thirty_two_ + createmany -l$remote_dir/foo1/luna $remote_dir/foo2/$longname 1000 || + error "failed to hardlink many files" + links=$($LFS fid2path $DIR $FID | wc -l) + echo -n "${links}/1000 links in link EA" + [ ${links} -gt 60 ] || err17935 "expected at least 60 links in link EA" + unlinkmany $remote_dir/foo2/$longname 1000 || + error "failed to unlink many hardlinks" +} +run_test 161b "link ea sanity under remote directory" check_path() { local expected=$1 @@ -9294,43 +9660,6 @@ test_183() { # LU-2275 } run_test 183 "No crash or request leak in case of strange dispositions ========" -test_185() { # LU-2441 - mkdir -p $DIR/$tdir || error "creating dir $DIR/$tdir" - touch $DIR/$tdir/spoo - local mtime1=$(stat -c "%Y" $DIR/$tdir) - local fid=$($MULTIOP $DIR/$tdir VFw4096c) || - error "cannot create/write a volatile file" - $CHECKSTAT -t file $MOUNT/.lustre/fid/$fid 2>/dev/null && - error "FID is still valid after close" - - multiop_bg_pause $DIR/$tdir vVw4096_c - local multi_pid=$! - - local OLD_IFS=$IFS - IFS=":" - local fidv=($fid) - IFS=$OLD_IFS - # assume that the next FID for this client is sequential, since stdout - # is unfortunately eaten by multiop_bg_pause - local n=$((${fidv[1]} + 1)) - local next_fid="${fidv[0]}:$(printf "0x%x" $n):${fidv[2]}" - $CHECKSTAT -t file $MOUNT/.lustre/fid/$next_fid || - error "FID is missing before close" - kill -USR1 $multi_pid - # 1 second delay, so if mtime change we will see it - sleep 1 - local mtime2=$(stat -c "%Y" $DIR/$tdir) - [[ $mtime1 == $mtime2 ]] || error "mtime has changed" -} -run_test 185 "Volatile file support" - -check_swap_layouts_support() -{ - $LCTL get_param -n llite.*.sbi_flags | grep -q layout || - { skip "Does not support layout lock."; return 0; } - return 1 -} - # test suite 184 is for LU-2016, LU-2017 test_184a() { check_swap_layouts_support && return 0 @@ -9394,23 +9723,27 @@ run_test 184b "Forbidden layout swap (will generate errors)" test_184c() { check_swap_layouts_support && return 0 - dir0=$DIR/$tdir/$testnum + local dir0=$DIR/$tdir/$testnum mkdir -p $dir0 || error "creating dir $dir0" - ref1=$dir0/ref1 - ref2=$dir0/ref2 - file1=$dir0/file1 - file2=$dir0/file2 + + local ref1=$dir0/ref1 + local ref2=$dir0/ref2 + local file1=$dir0/file1 + local file2=$dir0/file2 # create a file large enough for the concurent test dd if=/dev/urandom of=$ref1 bs=1M count=$((RANDOM % 50 + 20)) dd if=/dev/urandom of=$ref2 bs=1M count=$((RANDOM % 50 + 20)) echo "ref file size: ref1(`stat -c %s $ref1`), ref2(`stat -c %s $ref2`)" cp $ref2 $file2 - dd if=$ref1 of=$file1 bs=64k & - sleep 0.$((RANDOM % 5 + 1)) + dd if=$ref1 of=$file1 bs=16k & + local DD_PID=$! + + # Make sure dd starts to copy file + while [ ! -f $file1 ]; do sleep 0.1; done $LFS swap_layouts $file1 $file2 - rc=$? + local rc=$? wait $DD_PID [[ $? == 0 ]] || error "concurrent write on $file1 failed" [[ $rc == 0 ]] || error "swap of $file1 and $file2 failed" @@ -9421,7 +9754,7 @@ test_184c() { remaining=$((remaining - copied)) echo "Copied $copied bytes before swapping layout..." - cmp -n $copied $file1 $ref2 || + cmp -n $copied $file1 $ref2 | grep differ && error "Content mismatch [0, $copied) of ref2 and file1" cmp -n $copied $file2 $ref1 || error "Content mismatch [0, $copied) of ref1 and file2" @@ -9433,6 +9766,67 @@ test_184c() { } run_test 184c "Concurrent write and layout swap" +test_184d() { + check_swap_layouts_support && return 0 + + local file1=$DIR/$tdir/$tfile-1 + local file2=$DIR/$tdir/$tfile-2 + local file3=$DIR/$tdir/$tfile-3 + local lovea1 + local lovea2 + + mkdir -p $DIR/$tdir + touch $file1 || error "create $file1 failed" + $OPENFILE -f O_CREAT:O_LOV_DELAY_CREATE $file2 || + error "create $file2 failed" + $OPENFILE -f O_CREAT:O_LOV_DELAY_CREATE $file3 || + error "create $file3 failed" + lovea1=$($LFS getstripe $file1 | sed 1d) + + $LFS swap_layouts $file2 $file3 || + error "swap $file2 $file3 layouts failed" + $LFS swap_layouts $file1 $file2 || + error "swap $file1 $file2 layouts failed" + + lovea2=$($LFS getstripe $file2 | sed 1d) + [ "$lovea1" == "$lovea2" ] || error "lovea $lovea1 != $lovea2" + + lovea1=$(getfattr -n trusted.lov $file1 | grep ^trusted) + [ -z $lovea1 ] || error "$file1 shouldn't have lovea" +} +run_test 184d "allow stripeless layouts swap" + + +test_185() { # LU-2441 + mkdir -p $DIR/$tdir || error "creating dir $DIR/$tdir" + touch $DIR/$tdir/spoo + local mtime1=$(stat -c "%Y" $DIR/$tdir) + local fid=$($MULTIOP $DIR/$tdir VFw4096c) || + error "cannot create/write a volatile file" + $CHECKSTAT -t file $MOUNT/.lustre/fid/$fid 2>/dev/null && + error "FID is still valid after close" + + multiop_bg_pause $DIR/$tdir vVw4096_c + local multi_pid=$! + + local OLD_IFS=$IFS + IFS=":" + local fidv=($fid) + IFS=$OLD_IFS + # assume that the next FID for this client is sequential, since stdout + # is unfortunately eaten by multiop_bg_pause + local n=$((${fidv[1]} + 1)) + local next_fid="${fidv[0]}:$(printf "0x%x" $n):${fidv[2]}" + $CHECKSTAT -t file $MOUNT/.lustre/fid/$next_fid || + error "FID is missing before close" + kill -USR1 $multi_pid + # 1 second delay, so if mtime change we will see it + sleep 1 + local mtime2=$(stat -c "%Y" $DIR/$tdir) + [[ $mtime1 == $mtime2 ]] || error "mtime has changed" +} +run_test 185 "Volatile file support" + # OST pools tests check_file_in_pool() { @@ -9891,6 +10285,13 @@ verify_jobstats() { fi } +jobstats_set() { + trap 0 + NEW_JOBENV=${1:-$OLD_JOBENV} + do_facet mgs $LCTL conf_param $FSNAME.sys.jobid_var=$NEW_JOBENV + wait_update $HOSTNAME "$LCTL get_param -n jobid_var" $NEW_JOBENV +} + test_205() { # Job stats [ $PARALLEL == "yes" ] && skip "skip parallel run" && return [ -z "$(lctl get_param -n mdc.*.connect_flags | grep jobstats)" ] && @@ -9899,9 +10300,8 @@ test_205() { # Job stats local cmd OLD_JOBENV=`$LCTL get_param -n jobid_var` if [ $OLD_JOBENV != $JOBENV ]; then - do_facet mgs $LCTL conf_param $FSNAME.sys.jobid_var=$JOBENV - wait_update $HOSTNAME "$LCTL get_param -n jobid_var" \ - $JOBENV || return 1 + jobstats_set $JOBENV + trap jobstats_set EXIT fi # mkdir @@ -9938,11 +10338,7 @@ test_205() { # Job stats # cleanup rm -f $DIR/jobstats_test_rename - if [ $OLD_JOBENV != $JOBENV ]; then - do_facet mgs $LCTL conf_param $FSNAME.sys.jobid_var=$OLD_JOBENV - wait_update $HOSTNAME "$LCTL get_param -n jobid_var" \ - $OLD_JOBENV || return 1 - fi + [ $OLD_JOBENV != $JOBENV ] && jobstats_set $OLD_JOBENV } run_test 205 "Verify job stats" @@ -10157,16 +10553,20 @@ run_test 215 "/proc/sys/lnet exists and has proper content - bugs 18102, 21079, test_216() { # bug 20317 [ $PARALLEL == "yes" ] && skip "skip parallel run" && return - remote_ost_nodsh && skip "remote OST with nodsh" && return - local node - local p="$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" >> $p - done - clear_osc_stats + remote_ost_nodsh && skip "remote OST with nodsh" && return + + local node + local facets=$(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 $(osts_nodes); do @@ -10383,9 +10783,13 @@ run_test 224b "Don't panic on bulk IO failure" MDSSURVEY=${MDSSURVEY:-$(which mds-survey 2>/dev/null || true)} test_225a () { [ $PARALLEL == "yes" ] && skip "skip parallel run" && return - if [ -z ${MDSSURVEY} ]; then - skip_env "mds-survey not found" && return - fi + if [ -z ${MDSSURVEY} ]; then + skip_env "mds-survey not found" && return + fi + + [ $MDSCOUNT -ge 2 ] && + skip "skipping now for more than one MDT" && return + [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.2.51) ] || { skip "Need MDS version at least 2.2.51"; return; } @@ -10408,16 +10812,19 @@ run_test 225a "Metadata survey sanity with zero-stripe" test_225b () { [ $PARALLEL == "yes" ] && skip "skip parallel run" && return - if [ -z ${MDSSURVEY} ]; then - skip_env "mds-survey not found" && return - fi - [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.2.51) ] || - { skip "Need MDS version at least 2.2.51"; return; } - if [ $($LCTL dl | grep -c osc) -eq 0 ]; then - skip_env "Need to mount OST to test" && return - fi + if [ -z ${MDSSURVEY} ]; then + skip_env "mds-survey not found" && return + fi + [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.2.51) ] || + { skip "Need MDS version at least 2.2.51"; return; } + + if [ $($LCTL dl | grep -c osc) -eq 0 ]; then + skip_env "Need to mount OST to test" && return + fi + [ $MDSCOUNT -ge 2 ] && + skip "skipping now for more than one MDT" && return local mds=$(facet_host $SINGLEMDS) local target=$(do_nodes $mds 'lctl dl' | \ awk "{if (\$2 == \"UP\" && \$3 == \"mdt\") {print \$4}}") @@ -10463,7 +10870,7 @@ mcreate_path2fid () { echo "pass with $path and $fid" } -test_226 () { +test_226a () { rm -rf $DIR/$tdir mkdir -p $DIR/$tdir @@ -10476,7 +10883,29 @@ test_226 () { mcreate_path2fid 0120666 0 0 link "symbolic link" mcreate_path2fid 0140666 0 0 sock "socket" } -run_test 226 "call path2fid and fid2path on files of all type" +run_test 226a "call path2fid and fid2path on files of all type" + +test_226b () { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return + rm -rf $DIR/$tdir + local MDTIDX=1 + + mkdir -p $DIR/$tdir + $LFS setdirstripe -i $MDTIDX $DIR/$tdir/remote_dir || + error "create remote directory failed" + mcreate_path2fid 0010666 0 0 "remote_dir/fifo" "FIFO" + mcreate_path2fid 0020666 1 3 "remote_dir/null" \ + "character special file (null)" + mcreate_path2fid 0020666 1 255 "remote_dir/none" \ + "character special file (no device)" + mcreate_path2fid 0040666 0 0 "remote_dir/dir" "directory" + mcreate_path2fid 0060666 7 0 "remote_dir/loop0" \ + "block special file (loop)" + mcreate_path2fid 0100666 0 0 "remote_dir/file" "regular file" + mcreate_path2fid 0120666 0 0 "remote_dir/link" "symbolic link" + mcreate_path2fid 0140666 0 0 "remote_dir/sock" "socket" +} +run_test 226b "call path2fid and fid2path on files of all type under remote dir" # LU-1299 Executing or running ldd on a truncated executable does not # cause an out-of-memory condition. @@ -10672,7 +11101,20 @@ test_230b() { [ $rc -ne 0 ] && error "create remote directory failed after set enable_remote_dir" - rm -r $DIR/$tdir || error "unlink remote directory failed" + rm -rf $remote_dir || error "first unlink remote directory failed" + + $RUNAS -G$RUNAS_GID $LFS mkdir -i $MDTIDX $DIR/$tfile && + error "chown worked" + + do_facet mds$MDTIDX lctl set_param \ + mdt.*.enable_remote_dir_gid=$RUNAS_GID + $LFS mkdir -i $MDTIDX $remote_dir || rc=$? + do_facet mds$MDTIDX lctl set_param mdt.*.enable_remote_dir_gid=0 + + [ $rc -ne 0 ] && + error "create remote dir failed after set enable_remote_dir_gid" + + rm -r $DIR/$tdir || error "second unlink remote directory failed" } run_test 230b "nested remote directory should be failed" @@ -10726,6 +11168,20 @@ test_231b() { } run_test 231b "must not assert on fully utilized OST request buffer" +test_232() { + mkdir -p $DIR/$tdir + #define OBD_FAIL_LDLM_OST_LVB 0x31c + $LCTL set_param fail_loc=0x31c + + # ignore dd failure + dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=1 || true + + $LCTL set_param fail_loc=0 + umount_client $MOUNT || error "umount failed" + mount_client $MOUNT || error "mount failed" +} +run_test 232 "failed lock should not block umount" + # # tests that do cleanup/setup should be run at the end #