set -e
ONLY=${ONLY:-"$*"}
-# bug number for skipped test: 3192 LU-1205 15528/3811 16929 9977 15528/11549 18080
-ALWAYS_EXCEPT=" 14b 18c 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 $SANITYN_EXCEPT"
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-# bug number for skipped test: 12652 12652
-grep -q 'Enterprise Server 10' /etc/SuSE-release 2> /dev/null &&
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 14" || true
-
-# 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"}
-MCREATE=${MCREATE:-mcreate}
OPENFILE=${OPENFILE:-openfile}
OPENUNLINK=${OPENUNLINK:-openunlink}
export MULTIOP=${MULTIOP:-multiop}
init_logging
if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then
-# bug number for skipped test: LU-2840 LU-2189 LU-2776
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21 36 51a"
+# 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
-
+# 23 (min)"
[ "$SLOW" = "no" ] && EXCEPT_SLOW="33a"
FAIL_ON_ERROR=false
check_and_setup_lustre
-LOVNAME=$($LCTL get_param -n llite.*.lov.common_name | tail -n 1)
-OSTCOUNT=$($LCTL get_param -n lov.$LOVNAME.numobd)
-
assert_DIR
rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}*
}
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() {
local target="this/is/good"
ln -s $target $DIR1/$tfile || error "ln -s $target $DIR1/$tfile failed"
}
run_test 13 "test directory page revocation ===================="
-test_14() {
+test_14aa() {
test_mkdir -p $DIR1/$tdir
cp -p /bin/ls $DIR1/$tdir/$tfile
multiop_bg_pause $DIR1/$tdir/$tfile Ow_c || return 1
kill -USR1 $MULTIPID
wait $MULTIPID || return 2
}
-run_test 14 "execution of file open for write returns -ETXTBSY ="
+run_test 14aa "execution of file open for write returns -ETXTBSY"
-test_14a() {
+test_14ab() {
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
wait $MULTIOP_PID || return 3
rm $TMP/test14.junk $DIR1/d14/multiop || error "removing multiop"
}
-run_test 14a "open(RDWR) of executing file returns -ETXTBSY ===="
+run_test 14ab "open(RDWR) of executing file returns -ETXTBSY"
test_14b() { # bug 3192, 7040
test_mkdir -p $DIR1/d14
test_23() { # Bug 5972
local at_diff=$(do_facet $SINGLEMDS \
- $LCTL get_param -n mdd.*MDT0000*.atime_diff | head -1)
+ $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
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/$tfile bs=8k conv=notrunc count=1 seek=0
log "dd 3 finished"
test_28() { # bug 9977
ECHO_UUID="ECHO_osc1_UUID"
- tOST=`$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 | awk '$1 == 1 {print $2}'`
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"
+ [ $(calc_osc_stats lockless_truncate) -ne 0 ] ||
+ error "not cached truncate isn't lockless"
log "disabled lockless truncate"
enable_lockless_truncate 0
varsvc=${mds}_svc
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=$(do_facet $mds "cat /proc/fs/jbd*/${dev}{,:*,-*}/info 2>/dev/null |
+ head -n1")
val=${val%% *};
stat=$(( stat + val))
done
avgjbd=0
avgtime=0
for i in 1 2 3; do
- do_node $CLIENT1 "$LFS mkdir -i $MDTIDX -p \
+ do_node $CLIENT1 "$LFS mkdir -i $MDTIDX \
$DIR1/$tdir-\\\$(hostname)-$i"
jbdold=$(print_jbd_stat)
}
run_test 33b "COS: cross create/delete, 2 clients, benchmark under remote dir"
+test_33c() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
+ skip "DNE CoS not supported" && return
+
+ sync
+
+ mkdir $DIR/$tdir
+ # remote mkdir is done on MDT2, which enqueued lock of $tdir on MDT1
+ $LFS mkdir -i 1 $DIR/$tdir/d1
+ do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
+ mkdir $DIR/$tdir/d2
+ local sync_count=$(do_facet mds1 \
+ "lctl get_param -n mdt.*MDT0000.sync_count")
+ [ $sync_count -eq 1 ] || error "Sync-Lock-Cancel not triggered"
+
+ $LFS mkdir -i 1 $DIR/$tdir/d3
+ do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
+ # during sleep remote mkdir should have been committed and canceled
+ # remote lock spontaneously, which shouldn't trigger sync
+ sleep 6
+ mkdir $DIR/$tdir/d4
+ local sync_count=$(do_facet mds1 \
+ "lctl get_param -n mdt.*MDT0000.sync_count")
+ [ $sync_count -eq 0 ] || error "Sync-Lock-Cancel triggered"
+}
+run_test 33c "Cancel cross-MDT lock should trigger Sync-Lock-Cancel"
+
+ops_do_cos() {
+ local nodes=$(comma_list $(mdts_nodes))
+ do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
+ sh -c "$@"
+ local async_commit_count=$(do_nodes $nodes \
+ "lctl get_param -n mdt.*.async_commit_count" | calc_sum)
+ [ $async_commit_count -gt 0 ] || error "CoS not triggerred"
+
+ rm -rf $DIR/$tdir
+ sync
+}
+
+test_33d() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
+ skip "DNE CoS not supported" && return
+
+ sync
+ # remote directory create
+ mkdir $DIR/$tdir
+ ops_do_cos "$LFS mkdir -i 1 $DIR/$tdir/subdir"
+ # remote directory unlink
+ $LFS mkdir -i 1 $DIR/$tdir
+ ops_do_cos "rmdir $DIR/$tdir"
+ # striped directory create
+ mkdir $DIR/$tdir
+ ops_do_cos "$LFS mkdir -c 2 $DIR/$tdir/subdir"
+ # striped directory setattr
+ $LFS mkdir -c 2 $DIR/$tdir
+ touch $DIR/$tdir
+ ops_do_cos "chmod 713 $DIR/$tdir"
+ # striped directory unlink
+ $LFS mkdir -c 2 $DIR/$tdir
+ touch $DIR/$tdir
+ ops_do_cos "rmdir $DIR/$tdir"
+ # cross-MDT link
+ $LFS mkdir -c 2 $DIR/$tdir
+ $LFS mkdir -i 0 $DIR/$tdir/d1
+ $LFS mkdir -i 1 $DIR/$tdir/d2
+ touch $DIR/$tdir/d1/tgt
+ ops_do_cos "ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src"
+ # cross-MDT rename
+ $LFS mkdir -c 2 $DIR/$tdir
+ $LFS mkdir -i 0 $DIR/$tdir/d1
+ $LFS mkdir -i 1 $DIR/$tdir/d2
+ touch $DIR/$tdir/d1/src
+ ops_do_cos "mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt"
+ # migrate
+ $LFS mkdir -i 0 $DIR/$tdir
+ ops_do_cos "$LFS migrate -m 1 $DIR/$tdir"
+ return 0
+}
+run_test 33d "DNE distributed operation should trigger COS"
+
+test_33e() {
+ [ -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
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
+ skip "DNE CoS not supported" && return
+
+ local client2=${CLIENT2:-$(hostname)}
+
+ sync
+
+ local nodes=$(comma_list $(mdts_nodes))
+ do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
+
+ $LFS mkdir -c 2 $DIR/$tdir
+ mkdir $DIR/$tdir/subdir
+ echo abc > $DIR/$tdir/$tfile
+ do_node $client2 echo dfg >> $DIR/$tdir/$tfile
+ do_node $client2 touch $DIR/$tdir/subdir
+
+ local async_commit_count=$(do_nodes $nodes \
+ "lctl get_param -n mdt.*.async_commit_count" | calc_sum)
+ [ $async_commit_count -gt 0 ] && error "CoS triggerred"
+
+ return 0
+}
+run_test 33e "DNE local operation shouldn't trigger COS"
+
# End commit on sharing tests
get_ost_lock_timeouts() {
}
test_34() { #16129
+ remote_ost_nodsh && skip "remote OST with nodsh" && return
local OPER
local lock_in
local lock_out
test_35() { # bug 17645
local generation=[]
local count=0
- for imp in /proc/fs/lustre/mdc/$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
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
sync; sleep 5; sync # wait for delete thread
while [ $i -le 10 ]; do
- lctl mark "start test"
+ lctl mark "$HOSTNAME start test loop $i"
local before=$(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"
# pdirop tests
# test 40: check non-blocking operations
test_40a() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
#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
run_test 40a "pdirops: create vs others =============="
test_40b() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
touch $DIR1/$tfile &
run_test 40b "pdirops: open|create and others =============="
test_40c() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
touch $DIR1/$tfile
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
run_test 40c "pdirops: link and others =============="
test_40d() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
touch $DIR1/$tfile
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
run_test 40d "pdirops: unlink and others =============="
test_40e() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
touch $DIR1/$tfile
#define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
run_test 51b "layout lock: glimpse should be able to restart if layout changed"
test_51c() {
- # create an empty file
- $MCREATE $DIR1/$tfile
+ [ $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
- $LCTL set_param fail_loc=0x172
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x172
# change the layout of testing file
- echo "Setting layout ..."
- $LFS setstripe -c $OSTCOUNT $DIR1/$tfile &
+ echo "Setting layout to have $OSTCOUNT stripes ..."
+ $LFS setstripe -c $OSTCOUNT $DIR1/$tdir/$tfile &
pid=$!
sleep 1
- # get layout of this file should wait until dd is finished
- local stripecnt=`$LFS getstripe -c $DIR2/$tfile`
- [ $stripecnt -eq $OSTCOUNT ] || error "layout wrong"
+ # write something to the file, it should be blocked on fetching layout
+ dd if=/dev/zero of=$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 -f $DIR1/$tfile
+ 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() {
- [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
- { skip "Need MDS version at least 2.3.0"; return; }
+ local MDSVER=$(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
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
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"
+ # 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
+ for ((i = 0; i < $fcount; i++)); do
touch $DIR/$tdir/f_$i
- fd=$(free_fd)
- cmd="exec $fd<$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
- fid_list=($(do_facet mds$mds_idx $LCTL get_param -n $proc_ofile))
+ local get_open_fids="$LCTL get_param -n mdt.*.exports.'$nid'.open_files"
+ local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $get_open_fids))
# Possible errors in openfiles FID list.
# 1. Missing FIDs. Check 1
}
run_test 76 "Verify open file for 2048 files"
+nrs_write_read() {
+ local n=16
+ local dir=$DIR/$tdir
+ local myRUNAS="$1"
+
+ mkdir $dir || error "mkdir $dir failed"
+ $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed"
+ chmod 777 $dir
+
+ do_nodes $CLIENTS $myRUNAS dd if=/dev/zero of="$dir/nrs_r_$HOSTNAME"\
+ bs=1M count=$n > /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"
+ [ $? -ne 0 ] &&
+ error "failed to set TBF policy"
+ done
+
+ # Only operate rules on ost1 since OSTs might run on the same OSS
+ # Add some rules
+ tbf_rule_operate ost1 "start\ localhost\ {0@lo}\ 1000"
+ local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)")
+ local client_nids=$(nids_list $address "\\")
+ tbf_rule_operate ost1 "start\ clients\ {$client_nids}\ 100"
+ tbf_rule_operate ost1 "start\ others\ {*.*.*.*@$NETTYPE}\ 50"
+ nrs_write_read
+
+ # Change the rules
+ tbf_rule_operate ost1 "change\ localhost\ 1001"
+ tbf_rule_operate ost1 "change\ clients\ 101"
+ tbf_rule_operate ost1 "change\ others\ 51"
+ nrs_write_read
+
+ # Stop the rules
+ tbf_rule_operate ost1 "stop\ localhost"
+ tbf_rule_operate ost1 "stop\ clients"
+ tbf_rule_operate ost1 "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"
+ [ $? -ne 0 ] &&
+ error "failed to set TBF policy"
+ done
+
+ # Only operate rules on ost1 since OSTs might run on the same OSS
+ # Add some rules
+ tbf_rule_operate ost1 "start\ runas\ {iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ 1000"
+ tbf_rule_operate ost1 "start\ iozone_runas\ {iozone.$RUNAS_ID}\ 100"
+ tbf_rule_operate ost1 "start\ dd_runas\ {dd.$RUNAS_ID}\ 50"
+ nrs_write_read "$RUNAS"
+
+ # Change the rules
+ tbf_rule_operate ost1 "change\ runas\ 1001"
+ tbf_rule_operate ost1 "change\ iozone_runas\ 101"
+ tbf_rule_operate ost1 "change\ dd_runas\ 51"
+ nrs_write_read "$RUNAS"
+
+ # Stop the rules
+ tbf_rule_operate ost1 "stop\ runas"
+ tbf_rule_operate ost1 "stop\ iozone_runas"
+ tbf_rule_operate ost1 "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"
+ [ $? -ne 0 ] &&
+ error "failed to set TBF policy"
+ done
+
+ # Add a rule that only valid for Jobid TBF. If direct change between
+ # TBF types is not supported, this operation will fail.
+ tbf_rule_operate ost1 "start\ dd_runas\ {dd.$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_78() { #LU-6673
+ local rc
+
+ for i in $(seq 1 $OSTCOUNT)
+ do
+ do_facet ost"$i" lctl set_param \
+ ost.OSS.ost_io.nrs_policies="orr" &
+ do_facet ost"$i" lctl set_param \
+ ost.OSS.*.nrs_orr_quantum=1
+ rc=$?
+ # Valid return codes are:
+ # 0: Tuning succeeded
+ # ENODEV: Policy is still stopped
+ # EAGAIN: Policy is being initialized
+ [ $rc -eq 0 -o $rc -eq 19 -o $rc -eq 11 ] ||
+ error "Expected set_param to return 0|ENODEV|EAGAIN"
+ done
+
+ # Cleanup the ORR 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 78 "Enable policy and specify tunings right away"
+
+test_80a() {
+ [ $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 80a "migrate directory when some children is being opened"
+
+cleanup_80b() {
+ trap 0
+ kill -9 $migrate_pid
+}
+
+test_80b() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ local migrate_dir1=$DIR1/$tdir/migrate_dir
+ local migrate_dir2=$DIR2/$tdir/migrate_dir
+ local migrate_run=$LUSTRE/tests/migrate.sh
+ local start_time
+ local end_time
+ local show_time=1
+ local mdt_idx
+ local rc=0
+ local rc1=0
+
+ trap cleanup_80b EXIT
+ #prepare migrate directory
+ mkdir -p $migrate_dir1
+ for F in {1,2,3,4,5}; do
+ echo "$F$F$F$F$F" > $migrate_dir1/file$F
+ echo "$F$F$F$F$F" > $DIR/$tdir/file$F
+ done
+
+ #migrate the directories among MDTs
+ (
+ while true; do
+ mdt_idx=$((RANDOM % MDSCOUNT))
+ $LFS migrate -m $mdt_idx $migrate_dir1 2&>/dev/null ||
+ rc=$?
+ [ $rc -ne 0 -o $rc -ne 16 ] || break
+ done
+ ) &
+ migrate_pid=$!
+
+ echo "start migration thread $migrate_pid"
+ #Access the files at the same time
+ start_time=$(date +%s)
+ echo "accessing the migrating directory for 5 minutes..."
+ while true; do
+ ls $migrate_dir2 > /dev/null || {
+ echo "read dir fails"
+ break
+ }
+ diff -u $DIR2/$tdir/file1 $migrate_dir2/file1 || {
+ echo "access file1 fails"
+ break
+ }
+
+ cat $migrate_dir2/file2 > $migrate_dir2/file3 || {
+ echo "access file2/3 fails"
+ break
+ }
+
+ echo "aaaaa" > $migrate_dir2/file4 > /dev/null || {
+ echo "access file4 fails"
+ break
+ }
+
+ stat $migrate_dir2/file5 > /dev/null || {
+ echo "stat file5 fails"
+ break
+ }
+
+ touch $migrate_dir2/source_file > /dev/null || rc1=$?
+ [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
+ echo "touch file failed with $rc1"
+ break;
+ }
+
+ if [ -e $migrate_dir2/source_file ]; then
+ ln $migrate_dir2/source_file $migrate_dir2/link_file \
+ 2&>/dev/null || rc1=$?
+ if [ -e $migrate_dir2/link_file ]; then
+ rm -rf $migrate_dir2/link_file
+ fi
+
+ mrename $migrate_dir2/source_file \
+ $migrate_dir2/target_file 2&>/dev/null || rc1=$?
+ [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
+ echo "rename failed with $rc1"
+ break
+ }
+
+ if [ -e $migrate_dir2/target_file ]; then
+ rm -rf $migrate_dir2/target_file 2&>/dev/null ||
+ rc1=$?
+ else
+ rm -rf $migrate_dir2/source_file 2&>/dev/null ||
+ rc1=$?
+ fi
+ [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
+ echo "unlink failed with $rc1"
+ break
+ }
+ fi
+
+ end_time=$(date +%s)
+ duration=$((end_time - start_time))
+ if [ $((duration % 10)) -eq 0 ]; then
+ if [ $show_time -eq 1 ]; then
+ echo "...$duration seconds"
+ show_time=0
+ fi
+ else
+ show_time=1
+ fi
+
+ kill -0 $migrate_pid || {
+ echo "migration stopped 1"
+ break
+ }
+
+ [ $duration -ge 300 ] && break
+ done
+
+ #check migration are still there
+ kill -0 $migrate_pid || error "migration stopped 2"
+ cleanup_80b
+}
+run_test 80b "Accessing directory during migration"
+
+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"
+
+test_83() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ local pid1
+ local pid2
+
+ (
+ cd $DIR1
+ while true; do
+ $LFS mkdir -i1 -c2 $tdir
+ rmdir $tdir
+ done
+ ) &
+ pid1=$!
+ echo "start pid $pid1 to create/unlink striped directory"
+
+ # Access the directory at the same time
+ (
+ cd $DIR2
+ while true; do
+ stat $tdir > /dev/null 2>&1
+ done
+ ) &
+ pid2=$!
+ echo "start pid $pid2 to stat striped directory"
+
+ sleep 120
+ kill $pid1 $pid2
+ wait $pid1 $pid2
+
+ return 0
+}
+run_test 83 "access striped directory while it is being created/unlinked"
+
+test_90() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ local pid1
+ local pid2
+ local duration=180
+
+ [ "$SLOW" = "yes" ] && duration=600
+ # Open/Create under striped directory
+ (
+ cd $DIR1
+ while true; do
+ $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1
+ touch $tdir/f{0..3} > /dev/null 2>&1
+ done
+ ) &
+ pid1=$!
+ echo "start pid $pid1 to open/create under striped directory"
+
+ # unlink the striped directory at the same time
+ (
+ cd $DIR2
+ while true; do
+ rm -rf $tdir > /dev/null 2>&1
+ done
+ ) &
+ pid2=$!
+ echo "start pid $pid2 to unlink striped directory"
+
+ sleep $duration
+
+ kill $pid1 $pid2
+ wait $pid1 $pid2
+
+ return 0
+}
+run_test 90 "open/create and unlink striped directory"
+
+test_91() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ local pid1
+ local pid2
+ local duration=180
+
+ [ "$SLOW" = "yes" ] && duration=600
+ # chmod striped directory
+ (
+ cd $DIR1
+ while true; do
+ $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1
+ chmod go+w $tdir > /dev/null 2>&1
+ done
+ ) &
+ pid1=$!
+ echo "start pid $pid1 to chmod striped directory"
+
+ # unlink the striped directory at the same time
+ (
+ cd $DIR2
+ while true; do
+ rm -rf $tdir > /dev/null 2>&1
+ done
+ ) &
+ pid2=$!
+ echo "start pid $pid2 to unlink striped directory"
+
+ sleep $duration
+
+ kill $pid1 $pid2
+ wait $pid1 $pid2
+
+ return 0
+}
+run_test 91 "chmod and unlink striped directory"
+
+test_92() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+ local fd=$(free_fd)
+ local cmd="exec $fd<$DIR1/$tdir"
+ $LFS setdirstripe -c$MDSCOUNT $DIR1/$tdir || error "mkdir $tdir fails"
+ eval $cmd
+ cmd="exec $fd<&-"
+ trap "eval $cmd" EXIT
+ cd $DIR1/$tdir || error "cd $DIR1/$tdir fails"
+ rmdir ../$tdir || error "rmdir ../$tdir fails"
+
+ #define OBD_FAIL_LLITE_NO_CHECK_DEAD 0x1408
+ $LCTL set_param fail_loc=0x1408
+ mkdir $DIR2/$tdir/dir && error "create dir succeeds"
+ $LFS setdirstripe -i1 $DIR2/$tdir/remote_dir &&
+ error "create remote dir succeeds"
+ $LCTL set_param fail_loc=0
+ eval $cmd
+ return 0
+}
+run_test 92 "create remote directory under orphan directory"
+
log "cleanup: ======================================================"
+# kill and wait in each test only guarentee script finish, but command in script
+# like 'rm' 'chmod' may still be running, wait for all commands to finish
+# otherwise umount below will fail
+wait_update $HOSTNAME "fuser -m $MOUNT2" "" || true
+
[ "$(mount | grep $MOUNT2)" ] && umount $MOUNT2
complete $SECONDS