. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
init_logging
-[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test: LU-2840 LU-2776
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21 51a"
+if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then
+# bug number for skipped test: LU-2840 LU-2189 LU-2776
+ ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21 36 51a"
+# LU-2829 / LU-2887 - make allowances for ZFS slowness
+ TEST33_NFILES=${TEST33_NFILES:-1000}
+fi
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="12 23 33a"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="33a"
FAIL_ON_ERROR=false
OSTCOUNT=$($LCTL get_param -n lov.$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
cleanup_21() {
trap 0
- umount $DIR1/d21
+ umount $DIR1/$tdir
}
test_21() { # Bug 5907
- test_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 -1)
+ 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" $DIR2/f23`
+ time2=$(stat -c "%X" $DIR/$tfile)
+ echo "new atime is $time2"
- if (( $time2 <= $time1 )); then
- error "atime doesn't update among nodes"
- fi
-
- 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===="
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
#bug 11549 - permanently turn test off in b1_5
run_test 29 "lock put race between glimpse and enqueue ========="
-test_30() { #bug #11110
+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' &
- local err=$($DIR1/$tdir/bash -c 'sleep 2;
- openfile -f O_RDONLY /proc/$$/exe >& /dev/null; echo $?')
+ $DIR1/$tdir/bash -c 'sleep 2;
+ openfile -f O_RDONLY /proc/$$/exe >& /dev/null; echo $?'
wait
- [ $err -ne 116 ] &&
- error_ignore 12900 "return code ($err) != -ESTALE" && return
true
}
}
test_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
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'
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 -1")
val=${val%% *};
stat=$(( stat + val))
done
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
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=$((avgjbd / 3))
local nfiles=${TEST33_NFILES:-10000}
local param_file=$TMP/$tfile-params
- 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
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 | awk '{ if ($1 ~/^filesystem/) \
- { print $5; exit} }')
+ local before=$(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 ${SIZE}MB failed"
sync # sync data from client cache
sync_all_data # sync data from server cache (delayed allocation)
- sleep 1
- local after_dd=$($LFS df | awk '{ if ($1 ~/^filesystem/) \
- { print $5; exit} }')
+ sleep 2
+ local after_dd=$(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 | awk '{ if ($1 ~/^filesystem/) \
- { print $5; exit} }')
+ local after=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
echo "*** cycle($i) *** before($before) after_dd($after_dd)" \
"after($after)"
# this free space! not used
- if [ $after_dd -ge $after ]; then
- error "space leaked"
- return 1;
- fi
+ (( $after_dd <= $after)) ||
+ error "space leaked after_dd:$after_dd > after:$after"
let i=i+1
done
}
# open and sleep 2 seconds then read
$MULTIOP $DIR2/$tfile o_2r${filesize}c &
local pid=$!
- sleep 0.1
+ sleep 1
# create the layout of testing file
dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null
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
$LCTL set_param fail_loc=0x1404
stat -c %s $DIR2/$tfile |tee $tmpfile &
local pid=$!
- sleep 0.1
+ sleep 1
# create layout of testing file
dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc > /dev/null
echo "Setting layout ..."
$LFS setstripe -c $OSTCOUNT $DIR1/$tfile &
pid=$!
- sleep 0.1
+ sleep 1
# get layout of this file should wait until dd is finished
local stripecnt=`$LFS getstripe -c $DIR2/$tfile`
}
run_test 60 "Verify data_version behaviour"
-test_70() {
+test_70a() {
local test_dir=$tdir/test_dir
mkdir -p $DIR1/$tdir
cd $DIR2/$tdir || error "exit directory"
}
-run_test 70 "cd directory && rm 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 $(version_code 2.5.53) ]] &&
+ skip "Need MDS version at least 2.5.53" && return
+
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
+ local fcount=2048
+ local fd
+ local cmd
+ local mdt_idx
+ local mds_idx
+ declare -a fd_list
+ declare -a fid_list
+
+ if remote_mds; then
+ nid=$($LCTL list_nids | sed "s/\./\\\./g")
+ else
+ nid="0@lo"
+ fi
+
+ rm -rf $DIR/$tdir
+ test_mkdir -p $DIR/$tdir
+ if [ $MDSCOUNT -gt 1 ]; then
+ mdt_idx=$($LFS getdirstripe -i $DIR/$tdir)
+ else
+ mdt_idx=0
+ fi
+ mds_idx=$((mdt_idx + 1))
+ proc_ofile="mdt.*$mdt_idx.exports.'$nid'.open_files"
+
+ cancel_lru_locks mdc
+
+ echo -n "open files "
+ ulimit -n 8096
+ for (( i = 0; i < $fcount; i++ )) ; do
+ touch $DIR/$tdir/f_$i
+ fd=$(free_fd)
+ cmd="exec $fd<$DIR/$tdir/f_$i"
+ eval $cmd
+ fd_list[i]=$fd
+ echo -n "."
+ done
+ echo
+
+ fid_list=($(do_facet mds$mds_idx $LCTL get_param -n $proc_ofile))
+
+ # 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"
+ eval $cmd
+ filename=$($LFS fid2path $DIR2 ${fid_list[i]})
+
+ # Check 2
+ rm --interactive=no $filename
+ [ $? -ne 0 ] &&
+ error "Nonexisting fid ${fid_list[i]} listed."
+ done
+
+ # Check 3
+ ls_op=$(ls $DIR2/$tdir | wc -l)
+ [ $ls_op -ne 0 ] &&
+ error "Some openfiles are missing in lproc output"
+
+ rm -rf $DIR/$tdir
+}
+run_test 76 "Verify open file for 2048 files"
log "cleanup: ======================================================"
git://git.whamcloud.com / fs / lustre-release.git / blobdiff