+test_51a() {
+ local filesize
+ local origfile=/etc/hosts
+
+ filesize=`stat -c %s $origfile`
+
+ # create an empty file
+ $MCREATE $DIR1/$tfile
+ # cache layout lock on both mount point
+ stat $DIR1/$tfile > /dev/null
+ stat $DIR2/$tfile > /dev/null
+
+ # open and sleep 2 seconds then read
+ $MULTIOP $DIR2/$tfile o_2r${filesize}c &
+ local pid=$!
+ sleep 1
+
+ # create the layout of testing file
+ dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null
+
+ # MULTIOP proc should be able to read enough bytes and exit
+ sleep 2
+ kill -0 $pid && error "multiop is still there"
+ cmp $origfile $DIR2/$tfile || error "$MCREATE and $DIR2/$tfile differs"
+
+ rm -f $DIR1/$tfile
+}
+run_test 51a "layout lock: refresh layout should work"
+
+test_51b() {
+ [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.59) ]] ||
+ { skip "Need MDS version at least 2.3.59"; return 0; }
+
+ local tmpfile=`mktemp`
+
+ # create an empty file
+ $MCREATE $DIR1/$tfile
+
+ # delay glimpse so that layout has changed when glimpse finish
+#define OBD_FAIL_GLIMPSE_DELAY 0x1404
+ $LCTL set_param fail_loc=0x1404
+ stat -c %s $DIR2/$tfile |tee $tmpfile &
+ local pid=$!
+ sleep 1
+
+ # create layout of testing file
+ dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc > /dev/null
+
+ wait $pid
+ local fsize=`cat $tmpfile`
+
+ [ x$fsize = x1024 ] || error "file size is $fsize, should be 1024"
+
+ rm -f $DIR1/$tfile $tmpfile
+}
+run_test 51b "layout lock: glimpse should be able to restart if layout changed"
+
+test_51c() {
+ [ $OSTCOUNT -ge 2 ] || { skip "need at least 2 osts"; return; }
+
+ # set default layout to have 1 stripe
+ mkdir -p $DIR1/$tdir
+ $LFS setstripe -c 1 $DIR1/$tdir
+
+ # create a file with empty layout
+ $MCREATE $DIR1/$tdir/$tfile
+
+#define OBD_FAIL_MDS_LL_BLOCK 0x172
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x172
+
+ # change the layout of testing file
+ echo "Setting layout to have $OSTCOUNT stripes ..."
+ $LFS setstripe -c $OSTCOUNT $DIR1/$tdir/$tfile &
+ pid=$!
+ sleep 1
+
+ # write something to the file, it should be blocked on fetching layout
+ dd if=/dev/zero of=$DIR2/$tdir/$tfile bs=1k count=1 conv=notrunc
+ local cnt=$($LFS getstripe -c $DIR2/$tdir/$tfile)
+ [ $cnt -eq $OSTCOUNT ] || error "have $cnt stripes, expected $OSTCOUNT"
+
+ rm -fr $DIR1/$tdir
+}
+run_test 51c "layout lock: IT_LAYOUT blocked and correct layout can be returned"
+
+test_51d() {
+ dd if=/dev/zero of=/$DIR1/$tfile bs=1M count=1
+ cancel_lru_locks mdc
+
+ # open should grant LAYOUT lock, mmap and read will install pages
+ $MULTIOP $DIR1/$tfile oO_RDWR:SMR_Uc &
+ local PID=$!
+ sleep 1
+
+ # rss before revoking
+ local br=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}')
+ echo "Before revoking layout lock: $br KB mapped"
+
+ # delete the file will revoke layout lock
+ rm -f $DIR2/$tfile
+
+ # rss after revoking
+ local ar=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}')
+
+ kill -USR1 $PID
+ wait $PID || error
+
+ [ $ar -eq 0 ] || error "rss before: $br, after $ar, some pages remained"
+}
+run_test 51d "layout lock: losing layout lock should clean up memory map region"
+
+test_54_part1()
+{
+ echo "==> rename vs getattr vs setxattr should not deadlock"
+ mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed"
+
+ do_facet mds $LCTL set_param fail_loc=$1
+
+ mv -T $DIR/d1/d2/d3 $DIR/d1/d3 &
+ PID1=$!
+ sleep 1
+
+ stat $DIR/d1/d2 &
+ PID2=$!
+ sleep 1
+
+ setfattr -n user.attr1 -v value1 $DIR2/d1 || error "(2) setfattr failed"
+ wait $PID1 || error "(3) mv failed"
+ wait $PID2 || error "(4) stat failed"
+ echo
+
+ rm -rf $DIR/d1
+}
+
+test_54_part2() {
+ echo "==> rename vs getattr vs open vs getattr should not deadlock"
+ mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed"
+
+ do_facet mds $LCTL set_param fail_loc=$1
+
+ mv -T $DIR/d1/d2/d3 $DIR/d1/d3 &
+ PID1=$!
+ sleep 1
+
+ stat $DIR/d1/d2 &
+ PID2=$!
+ sleep 1
+
+ $MULTIOP $DIR2/d1/d2 Oc &
+ PID3=$!
+ sleep 1
+
+ stat $DIR/d1 || error "(2) stat failed"
+
+ wait $PID1 || error "(3) mv failed"
+ wait $PID2 || error "(4) stat failed"
+ wait $PID3 && error "(5) multiop failed"
+ echo
+ rm -rf $DIR/d1
+}
+
+test_54() {
+ local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
+ save_lustre_params client "llite.*.xattr_cache" > $p
+ lctl set_param llite.*.xattr_cache 1 ||
+ { skip "xattr cache is not supported"; return 0; }
+
+#define OBD_FAIL_MDS_RENAME 0x153
+#define OBD_FAIL_MDS_RENAME2 0x154
+ test_54_part1 0x80000153 || error 10
+ test_54_part1 0x80000154 || error 11
+ test_54_part2 0x80000153 || error 12
+ test_54_part2 0x80000154 || error 13
+
+ restore_lustre_params < $p
+ rm -f $p
+}
+run_test 54 "rename locking"
+
+test_55a() {
+ mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
+
+#define OBD_FAIL_MDS_RENAME4 0x156
+ do_facet mds $LCTL set_param fail_loc=0x80000156
+
+ mv -T $DIR/d1/d2 $DIR/d3/d2 &
+ PID1=$!
+ sleep 1
+
+ rm -r $DIR2/d3
+ wait $PID1 && error "(2) mv succeeded"
+
+ rm -rf $DIR/d1
+}
+run_test 55a "rename vs unlink target dir"
+
+test_55b()
+{
+ mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
+
+#define OBD_FAIL_MDS_RENAME4 0x156
+ do_facet mds $LCTL set_param fail_loc=0x80000156
+
+ mv -T $DIR/d1/d2 $DIR/d3/d2 &
+ PID1=$!
+ sleep 1
+
+ rm -r $DIR2/d1
+ wait $PID1 && error "(2) mv succeeded"
+
+ rm -rf $DIR/d3
+}
+run_test 55b "rename vs unlink source dir"
+
+test_55c()
+{
+ mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
+
+#define OBD_FAIL_MDS_RENAME4 0x156
+ do_facet mds $LCTL set_param fail_loc=0x156
+
+ mv -T $DIR/d1/d2 $DIR/d3/d2 &
+ PID1=$!
+ sleep 1
+
+ # while rename is sleeping, open and remove d3
+ $MULTIOP $DIR2/d3 D_c &
+ PID2=$!
+ sleep 1
+ rm -rf $DIR2/d3
+ sleep 5
+
+ # while rename is sleeping 2nd time, close d3
+ kill -USR1 $PID2
+ wait $PID2 || error "(3) multiop failed"
+
+ wait $PID1 && error "(2) mv succeeded"
+
+ rm -rf $DIR/d1
+}
+run_test 55c "rename vs unlink orphan target dir"
+
+test_55d()
+{
+ touch $DIR/f1
+
+#define OBD_FAIL_MDS_RENAME3 0x155
+ do_facet mds $LCTL set_param fail_loc=0x155
+ mv $DIR/f1 $DIR/$tdir &
+ PID1=$!
+ sleep 2
+
+ # while rename is sleeping, create $tdir, but as a directory
+ mkdir -p $DIR2/$tdir || error "(1) mkdir failed"
+
+ # link in reverse locking order
+ ln $DIR2/f1 $DIR2/$tdir/
+
+ wait $PID1 && error "(2) mv succeeded"
+ rm -rf $DIR/f1
+}
+run_test 55d "rename file vs link"
+
+test_60() {
+ local MDSVER=$(lustre_build_version $SINGLEMDS)
+ [ $(version_code $MDSVER) -lt $(version_code 2.3.0) ] &&
+ skip "MDS version $MDSVER must be >= 2.3.0" && return 0
+
+ # Create a file
+ test_mkdir -p $DIR1/$tdir
+ file1=$DIR1/$tdir/file
+ file2=$DIR2/$tdir/file
+
+ echo orig > $file2 || error "Could not create $file2"
+ version=$($LFS data_version $file1)
+
+ # Append data
+ echo append >> $file2 || error "Could not append to $file2"
+ version2=$($LFS data_version $file1)
+ [ "$version" != "$version2" ] ||
+ error "append did not change data version: $version"
+
+ # Overwrite data
+ echo overwrite > $file2 || error "Could not overwrite $file2"
+ version3=$($LFS data_version $file1)
+ [ "$version2" != "$version3" ] ||
+ error "overwrite did not change data version: $version2"
+
+ # Truncate before EOF
+ $TRUNCATE $file2 3 || error "Could not truncate $file2"
+ version4=$($LFS data_version $file1)
+ [ "$version3" != "$version4" ] ||
+ error "truncate did not change data version: $version3"
+
+ # Truncate after EOF
+ $TRUNCATE $file2 123456 || error "Could not truncate $file2"
+ version5=$($LFS data_version $file1)
+ [ "$version4" != "$version5" ] ||
+ error "truncate did not change data version: $version4"
+
+ # Chmod do not change version
+ chmod 400 $file2 || error "Could not chmod 400 $file2"
+ version6=$($LFS data_version $file1)
+ [ "$version5" == "$version6" ] ||
+ error "chmod should not change data version: $version5 != $version6"
+
+ # Chown do not change version
+ chown $RUNAS_ID $file2 || error "Could not chown $RUNAS_ID $file2"
+ version7=$($LFS data_version $file1)
+ [ "$version5" == "$version7" ] ||
+ error "chown should not change data version: $version5 != $version7"
+}
+run_test 60 "Verify data_version behaviour"
+
+test_70a() {
+ local test_dir=$tdir/test_dir
+
+ mkdir -p $DIR1/$tdir
+ if [ $MDSCOUNT -ge 2 ]; then
+ local MDTIDX=1
+ $LFS mkdir -i $MDTIDX $DIR1/$test_dir ||
+ error "Create remote directory failed"
+ else
+ mkdir -p $DIR1/$test_dir
+ fi
+ cd $DIR2/$test_dir || error "cd directory failed"
+ rm -rf $DIR1/$test_dir || error "unlink directory failed"
+
+ cd $DIR2/$tdir || error "exit directory"
+}
+run_test 70a "cd directory && rm directory"
+
+test_70b() { # LU-2781
+ local i
+ mkdir -p $DIR1/$tdir
+
+ touch $DIR1/$tdir/file
+ for ((i = 0; i < 32; i++)); do
+ $LFS rm_entry $DIR1/$tdir/non_existent_dir &>/dev/null
+ done
+ rm $DIR1/$tdir/file || error "cannot remove file after rm_entry"
+
+ touch $DIR1/$tdir/file
+ $LFS mkdir -i0 $DIR1/$tdir/test_dir
+ $LFS rm_entry $DIR1/$tdir/test_dir &>/dev/null
+ rm -rf $DIR1/$tdir/test_dir ||
+ error "cannot remove directory after rm_entry"
+ rm $DIR1/$tdir/file || error "cannot remove file after rm_entry"
+}
+run_test 70b "remove files after calling rm_entry"
+
+test_71() {
+ local server_version=$(lustre_version_code $SINGLEMDS)
+
+ [[ $server_version -lt $(version_code 2.1.6) ]] &&
+ skip "Need MDS version at least 2.1.6" && return
+
+ # Patch not applied to 2.2 and 2.3 branches
+ [[ $server_version -ge $(version_code 2.2.0) ]] &&
+ [[ $server_version -lt $(version_code 2.4.0) ]] &&
+ skip "Need MDS version at least 2.4.0" && return
+
+ checkfiemap --test ||
+ { skip "checkfiemap not runnable: $?" && return; }
+ # write data this way: hole - data - hole - data
+ dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1
+ [ "$(facet_fstype ost$(($($GETSTRIPE -i $DIR1/$tfile) + 1)))" = \
+ "zfs" ] &&
+ skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
+ dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1
+ GET_STAT="lctl get_param -n ldlm.services.ldlm_cbd.stats"
+ stat $DIR2/$tfile
+ local can1=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
+ echo $can1
+ checkfiemap $DIR2/$tfile 81920 ||
+ error "data is not flushed from client"
+ local can2=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
+ echo $can2
+
+ # common case of "create file, copy file" on a single node
+ # should not flush data from ost
+ dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1
+ dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1
+ stat $DIR1/$tfile
+ local can3=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
+ echo $can3
+ checkfiemap $DIR1/$tfile 81920 ||
+ error 4
+ local can4=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
+ echo $can2
+ [ $can3 -eq $can4 ] || error $((can2-can1)) "cancel RPC occured."
+}
+run_test 71 "correct file map just after write operation is finished"
+
+test_72() {
+ local p="$TMP/sanityN-$TESTNAME.parameters"
+ local tlink1
+ local tlink2
+ save_lustre_params client "llite.*.xattr_cache" > $p
+ lctl set_param llite.*.xattr_cache 1 ||
+ { skip "xattr cache is not supported"; return 0; }
+
+ touch $DIR1/$tfile
+ setfattr -n user.attr1 -v value1 $DIR1/$tfile ||
+ error "setfattr1 failed"
+ getfattr -n user.attr1 $DIR2/$tfile | grep value1 ||
+ error "getfattr1 failed"
+ setfattr -n user.attr1 -v value2 $DIR2/$tfile ||
+ error "setfattr2 failed"
+ getfattr -n user.attr1 $DIR1/$tfile | grep value2 ||
+ error "getfattr2 failed"
+
+ # check that trusted.link is consistent
+ tlink1=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
+ ln $DIR2/$tfile $DIR2/$tfile-2 || error "failed to link"
+ tlink2=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
+ echo "$tlink1 $tlink2"
+ [ "$tlink1" = "$tlink2" ] && error "trusted.link should have changed!"
+
+ rm -f $DIR2/$tfile
+
+ restore_lustre_params < $p
+ rm -f $p
+}
+run_test 72 "getxattr/setxattr cache should be consistent between nodes"
+
+test_73() {
+ local p="$TMP/sanityN-$TESTNAME.parameters"
+ save_lustre_params client "llite.*.xattr_cache" > $p
+ lctl set_param llite.*.xattr_cache 1 ||
+ { skip "xattr cache is not supported"; return 0; }
+
+ touch $DIR1/$tfile
+ setfattr -n user.attr1 -v value1 $DIR1/$tfile ||
+ error "setfattr1 failed"
+ getfattr -n user.attr1 $DIR2/$tfile || error "getfattr1 failed"
+ getfattr -n user.attr1 $DIR1/$tfile || error "getfattr2 failed"
+ clear_llite_stats
+ # PR lock should be cached by now on both clients
+ getfattr -n user.attr1 $DIR1/$tfile || error "getfattr3 failed"
+ # 2 hits for getfattr(0)+getfattr(size)
+ [ $(calc_llite_stats getxattr_hits) -eq 2 ] || error "not cached in $DIR1"
+ getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed"
+ # 4 hits for more getfattr(0)+getfattr(size)
+ [ $(calc_llite_stats getxattr_hits) -eq 4 ] || error "not cached in $DIR2"
+ rm -f $DIR2/$tfile
+
+ restore_lustre_params < $p
+ rm -f $p
+}
+run_test 73 "getxattr should not cause xattr lock cancellation"
+
+test_74() {
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.93) ] &&
+ skip "Need MDS version at least 2.4.93" && return
+
+ dd if=/dev/zero of=$DIR1/$tfile-1 bs=1K count=1
+ dd if=/dev/zero of=$DIR1/$tfile-2 bs=1K count=1
+ flocks_test 4 $DIR1/$tfile-1 $DIR2/$tfile-2
+}
+run_test 74 "flock deadlock: different mounts =============="
+
+# LU-3889
+test_75() {
+ $LFS setstripe -c 2 -S 1m -i 0 $DIR1/$tfile
+ dd if=/dev/zero of=$DIR1/$tfile bs=1M count=2
+ cancel_lru_locks osc
+
+ dd of=$DIR1/$tfile if=/dev/zero bs=1M count=1 seek=1 conv=notrunc
+ sync
+
+ # define OBD_FAIL_LDLM_ENQUEUE_HANG 0x31d
+ $LCTL set_param fail_loc=0x31d
+ stat -c %s $DIR1/$tfile &
+ local pid=$!
+ sleep 1
+ kill -9 $pid
+
+ # For bad lock error handler we should ASSERT and got kernel panic here
+ sleep 4
+ $LCTL set_param fail_loc=0
+}
+run_test 75 "osc: upcall after unuse lock==================="
+
+test_76() { #LU-946
+ [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.53) ]] &&
+ skip "Need MDS version at least 2.5.53" && return
+
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
+ local fcount=2048
+ declare -a fd_list
+ declare -a fid_list
+
+ if remote_mds; then
+ nid=$($LCTL list_nids | sed "s/\./\\\./g")
+ else
+ nid="0@lo"
+ fi
+
+ rm -rf $DIR/$tdir
+ test_mkdir -p $DIR/$tdir
+
+ # drop all open locks and close any cached "open" files on the client
+ cancel_lru_locks mdc
+
+ echo -n "open files "
+ ulimit -n 8096
+ for ((i = 0; i < $fcount; i++)); do
+ touch $DIR/$tdir/f_$i
+ local fd=$(free_fd)
+ local cmd="exec $fd<$DIR/$tdir/f_$i"
+ eval $cmd
+ fd_list[i]=$fd
+ echo -n "."
+ done
+ echo
+
+ local get_open_fids="$LCTL get_param -n mdt.*.exports.'$nid'.open_files"
+ local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $get_open_fids))
+
+ # Possible errors in openfiles FID list.
+ # 1. Missing FIDs. Check 1
+ # 2. Extra FIDs. Check 1
+ # 3. Duplicated FID. Check 2
+ # 4. Invalid FIDs. Check 2
+ # 5. Valid FID, points to some other file. Check 3
+
+ # Check 1
+ [ ${#fid_list[@]} -ne $fcount ] &&
+ error "${#fid_list[@]} != $fcount open files"
+
+ for (( i = 0; i < $fcount; i++ )) ; do
+ cmd="exec ${fd_list[i]}</dev/null"
+ 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"
+
+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
+
+ local idis
+ local rateis
+ if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then
+ idis="nid="
+ rateis="rate="
+ fi
+
+ # Only operate rules on ost1 since OSTs might run on the same OSS
+ # Add some rules
+ tbf_rule_operate ost1 "start\ localhost\ ${idis}{0@lo}\ ${rateis}1000"
+ local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)")
+ local client_nids=$(nids_list $address "\\")
+ tbf_rule_operate ost1 "start\ clients\ ${idis}{$client_nids}\ ${rateis}100"
+ tbf_rule_operate ost1 "start\ others\ ${idis}{*.*.*.*@$NETTYPE}\ ${rateis}50"
+ nrs_write_read
+
+ # Change the rules
+ tbf_rule_operate ost1 "change\ localhost\ ${rateis}1001"
+ tbf_rule_operate ost1 "change\ clients\ ${rateis}101"
+ tbf_rule_operate ost1 "change\ others\ ${rateis}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
+
+ local idis
+ local rateis
+ if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then
+ idis="jobid="
+ rateis="rate="
+ fi
+
+ # Only operate rules on ost1 since OSTs might run on the same OSS
+ # Add some rules
+ tbf_rule_operate ost1 "start\ runas\ ${idis}{iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ ${rateis}1000"
+ tbf_rule_operate ost1 "start\ iozone_runas\ ${idis}{iozone.$RUNAS_ID}\ ${rateis}100"
+ tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50"
+ nrs_write_read "$RUNAS"
+
+ # Change the rules
+ tbf_rule_operate ost1 "change\ runas\ ${rateis}1001"
+ tbf_rule_operate ost1 "change\ iozone_runas\ ${rateis}101"
+ tbf_rule_operate ost1 "change\ dd_runas\ ${rateis}51"
+ nrs_write_read "$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
+
+ local idis
+ local rateis
+ if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then
+ idis="jobid="
+ rateis="rate="
+ fi
+
+ # 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\ ${idis}{dd.$RUNAS_ID}\ ${rateis}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_77h() {
+ [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] ||
+ { skip "Need OST version at least 2.8.55"; return 0; }
+
+ local old_policy=$(do_facet ost1 \
+ lctl get_param ost.OSS.ost_io.nrs_policies)
+ local new_policy
+
+ do_facet ost1 lctl set_param \
+ ost.OSS.ost_io.nrs_policies="abc"
+ [ $? -eq 0 ] && error "should return error"
+
+ do_facet ost1 lctl set_param \
+ ost.OSS.ost_io.nrs_policies="tbf\ abc"
+ [ $? -eq 0 ] && error "should return error"
+
+ do_facet ost1 lctl set_param \
+ ost.OSS.ost_io.nrs_policies="tbf\ reg"
+ [ $? -eq 0 ] && error "should return error"
+
+ do_facet ost1 lctl set_param \
+ ost.OSS.ost_io.nrs_policies="tbf\ reg\ abc"
+ [ $? -eq 0 ] && error "should return error"
+
+ do_facet ost1 lctl set_param \
+ ost.OSS.ost_io.nrs_policies="tbf\ abc\ efg"
+ [ $? -eq 0 ] && error "should return error"
+
+ new_policy=$(do_facet ost1 lctl get_param ost.OSS.ost_io.nrs_policies)
+ [ $? -eq 0 ] || error "shouldn't LBUG"
+
+ [ "$old_policy" = "$new_policy" ] || error "NRS policy should be same"
+
+ return 0
+}
+run_test 77h "Wrong policy name should report error, not LBUG"
+
+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 || error "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"
+