set -e
ONLY=${ONLY:-"$*"}
-# bug number for skipped test: 3192 9977
-ALWAYS_EXCEPT="14b 28 $SANITYN_EXCEPT"
+# bug number for skipped test: 3192 15528/3811 16929 9977 15528/11549 18080
+ALWAYS_EXCEPT=" 14b 19 22 28 29 35 $SANITYN_EXCEPT"
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
# bug number for skipped test: 12652 12652
SIZE=${SIZE:-40960}
CHECKSTAT=${CHECKSTAT:-"checkstat -v"}
-CREATETEST=${CREATETEST:-createtest}
GETSTRIPE=${GETSTRIPE:-lfs getstripe}
SETSTRIPE=${SETSTRIPE:-lstripe}
MCREATE=${MCREATE:-mcreate}
OPENFILE=${OPENFILE:-openfile}
OPENUNLINK=${OPENUNLINK:-openunlink}
-TOEXCL=${TOEXCL:-toexcl}
TRUNCATE=${TRUNCATE:-truncate}
export TMP=${TMP:-/tmp}
MOUNT_2=${MOUNT_2:-"yes"}
init_test_env $@
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="12 16"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="12 16 23 33a"
SANITYLOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh).log}
FAIL_ON_ERROR=false
SETUP=${SETUP:-:}
TRACE=${TRACE:-""}
-LPROC=/proc/fs/lustre
-
[ "$SANITYLOG" ] && rm -f $SANITYLOG || true
check_and_setup_lustre
-LPROC=/proc/fs/lustre
LOVNAME=`lctl get_param -n llite.*.lov.common_name | tail -n 1`
OSTCOUNT=`lctl get_param -n lov.$LOVNAME.numobd`
+assert_DIR
rm -rf $DIR1/[df][0-9]* $DIR1/lnk
# $RUNAS_ID may get set incorrectly somewhere else
[ $UID -eq 0 -a $RUNAS_ID -eq 0 ] && error "\$RUNAS_ID set to 0, but \$UID is also 0!"
-check_runas_id $RUNAS_ID $RUNAS
+check_runas_id $RUNAS_ID $RUNAS_ID $RUNAS
build_test_filter
run_test 2e "check chmod on root is propagated to others"
test_3() {
- ( cd $DIR1 ; ln -s this/is/good lnk )
- [ "this/is/good" = "`perl -e 'print readlink("'$DIR2/lnk'");'`" ] || \
- error
+ ( cd $DIR1 ; ln -s this/is/good $tfile )
+ [ "this/is/good" = "`perl -e 'print readlink("'$DIR2/$tfile'");'`" ] ||
+ error "link $DIR2/$tfile not as expected"
}
run_test 3 "symlink on one mtpt, readlink on another ==========="
run_test 6 "remove of open file on other node =================="
test_7() {
- # run_one creates uniq $tdir (bug 13798)
- # opendirunlink failes if it exists
- rmdir $DIR1/$tdir || true
- opendirunlink $DIR1/$tdir $DIR2/$tdir || \
- error "opendirunlink $DIR1/$tdir $DIR2/$tdir"
+ local dir=d7
+ opendirunlink $DIR1/$dir $DIR2/$dir || \
+ error "opendirunlink $DIR1/$dir $DIR2/$dir"
}
run_test 7 "remove of open directory on other node ============="
test_9() {
MTPT=1
+ local dir
> $DIR2/f9
for C in a b c d e f g h i j k l; do
- DIR=`eval echo \\$DIR$MTPT`
- echo -n $C >> $DIR/f9
+ dir=`eval echo \\$DIR$MTPT`
+ echo -n $C >> $dir/f9
[ "$MTPT" -eq 1 ] && MTPT=2 || MTPT=1
done
[ "`cat $DIR1/f9`" = "abcdefghijkl" ] || \
test_10a() {
MTPT=1
+ local dir
OFFSET=0
> $DIR2/f10
for C in a b c d e f g h i j k l; do
- DIR=`eval echo \\$DIR$MTPT`
- echo -n $C | dd of=$DIR/f10 bs=1 seek=$OFFSET count=1
+ dir=`eval echo \\$DIR$MTPT`
+ echo -n $C | dd of=$dir/f10 bs=1 seek=$OFFSET count=1
[ "$MTPT" -eq 1 ] && MTPT=2 || MTPT=1
OFFSET=`expr $OFFSET + 1`
done
run_test 10a "write of file with sub-page size on multiple mounts "
test_10b() {
- yes "R" | dd of=$DIR1/f10b bs=3k count=1 || error "dd $DIR1"
+ # create a seed file
+ yes "R" | head -c 4000 >$TMP/f10b-seed
+ dd if=$TMP/f10b-seed of=$DIR1/f10b bs=3k count=1 || error "dd $DIR1"
truncate $DIR1/f10b 4096 || error "truncate 4096"
dd if=$DIR2/f10b of=$TMP/f10b-lustre bs=4k count=1 || error "dd $DIR2"
# create a test file locally to compare
- yes "R" | dd of=$TMP/f10b bs=3k count=1 || error "dd random"
+ dd if=$TMP/f10b-seed of=$TMP/f10b bs=3k count=1 || error "dd random"
truncate $TMP/f10b 4096 || error "truncate 4096"
cmp $TMP/f10b $TMP/f10b-lustre || error "file miscompare"
- rm $TMP/f10b $TMP/f10b-lustre
+ rm $TMP/f10b $TMP/f10b-lustre $TMP/f10b-seed
}
run_test 10b "write of file with sub-page size on multiple mounts "
test_11() {
mkdir $DIR1/d11
- multiop $DIR1/d11/f O_c &
+ multiop_bg_pause $DIR1/d11/f O_c || return 1
MULTIPID=$!
- usleep 200
cp -p /bin/ls $DIR1/d11/f
$DIR2/d11/f
RC=$?
run_test 13 "test directory page revocation ===================="
test_14() {
- mkdir $DIR1/d14
- cp -p /bin/ls $DIR1/d14/ls
- exec 100>> $DIR1/d14/ls
- $DIR2/d14/ls && error || true
- exec 100<&-
+ mkdir -p $DIR1/$tdir
+ cp -p /bin/ls $DIR1/$tdir/$tfile
+ multiop_bg_pause $DIR1/$tdir/$tfile Ow_c || return 1
+ MULTIPID=$!
+
+ $DIR2/$tdir/$tfile && error || true
+ kill -USR1 $MULTIPID
+ wait $MULTIPID || return 2
}
run_test 14 "execution of file open for write returns -ETXTBSY ="
test_14a() {
mkdir -p $DIR1/d14
cp -p `which multiop` $DIR1/d14/multiop || error "cp failed"
- $DIR1/d14/multiop $TMP/test14.junk O_c &
- MULTIPID=$!
- sleep 1
+ MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1
+ MULTIOP_PID=$!
multiop $DIR2/d14/multiop Oc && error "expected error, got success"
- kill -USR1 $MULTIPID || return 2
- wait $MULTIPID || return 3
+ kill -USR1 $MULTIOP_PID || return 2
+ 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 ===="
test_14b() { # bug 3192, 7040
mkdir -p $DIR1/d14
cp -p `which multiop` $DIR1/d14/multiop || error "cp failed"
- $DIR1/d14/multiop $TMP/test14.junk O_c &
- MULTIPID=$!
- sleep 1
- truncate $DIR2/d14/multiop 0 && kill -9 $MULTIPID && \
+ MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1
+ MULTIOP_PID=$!
+ truncate $DIR2/d14/multiop 0 && kill -9 $MULTIOP_PID && \
error "expected truncate error, got success"
- kill -USR1 $MULTIPID || return 2
- wait $MULTIPID || return 3
+ kill -USR1 $MULTIOP_PID || return 2
+ wait $MULTIOP_PID || return 3
cmp `which multiop` $DIR1/d14/multiop || error "binary changed"
rm $TMP/test14.junk $DIR1/d14/multiop || error "removing multiop"
}
test_14c() { # bug 3430, 7040
mkdir -p $DIR1/d14
cp -p `which multiop` $DIR1/d14/multiop || error "cp failed"
- $DIR1/d14/multiop $TMP/test14.junk O_c &
- MULTIPID=$!
- sleep 1
+ MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1
+ MULTIOP_PID=$!
cp /etc/hosts $DIR2/d14/multiop && error "expected error, got success"
- kill -USR1 $MULTIPID || return 2
- wait $MULTIPID || return 3
+ kill -USR1 $MULTIOP_PID || return 2
+ wait $MULTIOP_PID || return 3
cmp `which multiop` $DIR1/d14/multiop || error "binary changed"
rm $TMP/test14.junk $DIR1/d14/multiop || error "removing multiop"
}
test_14d() { # bug 10921
mkdir -p $DIR1/d14
cp -p `which multiop` $DIR1/d14/multiop || error "cp failed"
- $DIR1/d14/multiop $TMP/test14.junk O_c &
- MULTIPID=$!
- sleep 1
+ MULTIOP_PROG=$DIR1/d14/multiop multiop_bg_pause $TMP/test14.junk O_c || return 1
+ MULTIOP_PID=$!
log chmod
chmod 600 $DIR1/d14/multiop || error "chmod failed"
- kill -USR1 $MULTIPID || return 2
- wait $MULTIPID || return 3
+ kill -USR1 $MULTIOP_PID || return 2
+ wait $MULTIOP_PID || return 3
cmp `which multiop` $DIR1/d14/multiop || error "binary changed"
rm $TMP/test14.junk $DIR1/d14/multiop || error "removing multiop"
}
run_test 16 "2500 iterations of dual-mount fsx ================="
test_17() { # bug 3513, 3667
- [ ! -d /proc/fs/lustre/ost ] && skip "remote OST, skipping OST-only test" && return
+ remote_ost_nodsh && skip "remote OST with nodsh" && return
- cp /etc/termcap $DIR1/f17
+ lfs setstripe $DIR1/$tfile -i 0 -c 1
+ cp /etc/termcap $DIR1/$tfile
cancel_lru_locks osc > /dev/null
#define OBD_FAIL_ONCE|OBD_FAIL_LDLM_CREATE_RESOURCE 0x30a
- sysctl -w lustre.fail_loc=0x8000030a
- ls -ls $DIR1/f17 | awk '{ print $1,$6 }' > $DIR1/f17-1 & \
- ls -ls $DIR2/f17 | awk '{ print $1,$6 }' > $DIR2/f17-2
+ do_facet ost1 lctl set_param fail_loc=0x8000030a
+ ls -ls $DIR1/$tfile | awk '{ print $1,$6 }' > $DIR1/$tfile-1 & \
+ ls -ls $DIR2/$tfile | awk '{ print $1,$6 }' > $DIR2/$tfile-2
wait
- diff -u $DIR1/f17-1 $DIR2/f17-2 || error "files are different"
+ diff -u $DIR1/$tfile-1 $DIR2/$tfile-2 || error "files are different"
}
run_test 17 "resource creation/LVB creation race ==============="
test_18() {
- ./mmap_sanity -d $MOUNT1 -m $MOUNT2
+ $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2
sync; sleep 1; sync
}
run_test 18 "mmap sanity check ================================="
test_19() { # bug3811
[ -d /proc/fs/lustre/obdfilter ] || return 0
- MAX=`cat /proc/fs/lustre/obdfilter/*/readcache_max_filesize | head -n 1`
- for O in /proc/fs/lustre/obdfilter/*OST*; do
- echo 4096 > $O/readcache_max_filesize
- done
+ MAX=`lctl get_param -n obdfilter.*.readcache_max_filesize | head -n 1`
+ lctl set_param -n obdfilter.*OST*.readcache_max_filesize=4096
dd if=/dev/urandom of=$TMP/f19b bs=512k count=32
SUM=`cksum $TMP/f19b | cut -d" " -f 1,2`
cp $TMP/f19b $DIR1/f19b
[ "`cat $TMP/sum2`" = "$SUM" ] || \
error "$DIR2/f19b `cat $TMP/sum2` != $SUM"
done
- for O in /proc/fs/lustre/obdfilter/*OST*; do
- echo $MAX > $O/readcache_max_filesize
- done
+ lctl set_param -n obdfilter.*OST*.readcache_max_filesize=$MAX
rm $DIR1/f19b
}
-#run_test 19 "test concurrent uncached read races ==============="
+run_test 19 "test concurrent uncached read races ==============="
test_20() {
mkdir $DIR1/d20
cancel_lru_locks osc
- CNT=$((`cat /proc/fs/lustre/llite/*/dump_page_cache | wc -l`))
+ CNT=$((`lctl get_param -n llite.*.dump_page_cache | wc -l`))
multiop $DIR1/f20 Ow8190c
multiop $DIR2/f20 Oz8194w8190c
multiop $DIR1/f20 Oz0r8190c
cancel_lru_locks osc
- CNTD=$((`cat /proc/fs/lustre/llite/*/dump_page_cache | wc -l` - $CNT))
+ CNTD=$((`lctl get_param -n llite.*.dump_page_cache | wc -l` - $CNT))
[ $CNTD -gt 0 ] && \
error $CNTD" page left in cache after lock cancel" || true
}
cat $DIR2/d21/no_joined || error "cat error"
rm -rf $DIR2/d21/no_joined || error "unlink normal file error"
}
-run_test 22 " After joining in one dir, open/close unlink file in anther dir"
+run_test 22 " After joining in one dir, open/close unlink file in anther dir"
test_23() { # Bug 5972
echo "others should see updated atime while another read" > $DIR1/f23
cancel_lru_locks osc
time1=`date +%s`
- sleep 2
+ #MAX_ATIME_DIFF 60, we update atime only if older than 60 seconds
+ sleep 61
- multiop $DIR1/f23 or20_c &
- MULTIPID=$!
+ multiop_bg_pause $DIR1/f23 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 $!
+ cancel_lru_locks mdc
- sleep 2
time2=`stat -c "%X" $DIR2/f23`
if (( $time2 <= $time1 )); then
- kill -USR1 $MULTIPID
error "atime doesn't update among nodes"
fi
- kill -USR1 $MULTIPID || return 1
rm -f $DIR1/f23 || error "rm -f $DIR1/f23 failed"
true
}
[ `lctl get_param -n mdc.*-mdc-*.connect_flags | grep -c acl` -lt 2 ] && \
skip "must have acl, skipping" && return
+ mkdir -p $DIR1/$tdir
touch $DIR1/$tdir/f1 || error "touch $DIR1/$tdir/f1"
chmod 0755 $DIR1/$tdir/f1 || error "chmod 0755 $DIR1/$tdir/f1"
chmod a+x $DIR2/$tfile
mt1=`stat -c %Y $DIR1/$tfile`
mt2=`stat -c %Y $DIR2/$tfile`
-
- if [ x"$mt1" != x"$mt2" ]; then
+
+ if [ x"$mt1" != x"$mt2" ]; then
error "not equal mtime, client1: "$mt1", client2: "$mt2"."
fi
}
tECHOID=`$LCTL dl | grep $ECHO_UUID | awk '{print $1}'`
$LCTL --device $tECHOID destroy "${tOBJID}:0"
-
+
$LCTL <<-EOF
cfg_device ECHO_osc1
cleanup
test_29() { # bug 10999
touch $DIR1/$tfile
#define OBD_FAIL_LDLM_GLIMPSE 0x30f
- sysctl -w lustre.fail_loc=0x8000030f
+ lctl set_param fail_loc=0x8000030f
ls -l $DIR2/$tfile &
usleep 500
dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1
wait
}
#bug 11549 - permanently turn test off in b1_5
-#run_test 29 "lock put race between glimpse and enqueue ========="
+run_test 29 "lock put race between glimpse and enqueue ========="
test_30() { #bug #11110
+ mkdir -p $DIR1/$tdir
cp -f /bin/bash $DIR1/$tdir/bash
/bin/sh -c 'sleep 1; rm -f $DIR2/$tdir/bash; cp /bin/bash $DIR2/$tdir' &
err=$($DIR1/$tdir/bash -c 'sleep 2; openfile -f O_RDONLY /proc/$$/exe >& /dev/null; echo $?')
run_test 30 "recreate file race ========="
-test_31() {
+test_31a() {
mkdir -p $DIR1/$tdir || error "Creating dir $DIR1/$tdir"
writes=`LANG=C dd if=/dev/zero of=$DIR/$tdir/$tfile count=1 2>&1 |
awk 'BEGIN { FS="+" } /out/ {print $1}'`
#define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314
- sysctl -w lustre.fail_loc=0x314
+ lctl set_param fail_loc=0x314
reads=`LANG=C dd if=$DIR2/$tdir/$tfile of=/dev/null 2>&1 |
awk 'BEGIN { FS="+" } /in/ {print $1}'`
[ $reads -eq $writes ] || error "read" $reads "blocks, must be" $writes
}
-run_test 31 "voluntary cancel / blocking ast race=============="
+run_test 31a "voluntary cancel / blocking ast race=============="
+
+test_31b() {
+ remote_ost || { skip "local OST" && return 0; }
+ remote_ost_nodsh && skip "remote OST w/o dsh" && return 0
+ mkdir -p $DIR1/$tdir || error "Creating dir $DIR1/$tdir"
+ lfs setstripe $DIR/$tdir/$tfile -i 0 -c 1
+ cp /etc/hosts $DIR/$tdir/$tfile
+ #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314
+ lctl set_param fail_loc=0x314
+ #define OBD_FAIL_LDLM_OST_FAIL_RACE 0x316
+ do_facet ost1 lctl set_param fail_loc=0x316
+ # Don't crash kernel
+ cat $DIR2/$tdir/$tfile > /dev/null 2>&1
+ lctl set_param fail_loc=0
+ do_facet ost1 lctl set_param fail_loc=0
+}
+run_test 31b "voluntary OST cancel / blocking ast race=============="
+
+# enable/disable lockless truncate feature, depending on the arg 0/1
+enable_lockless_truncate() {
+ lctl set_param -n osc.*.lockless_truncate $1
+}
+
+test_32a() { # bug 11270
+ local p="$TMP/sanityN-$TESTNAME.parameters"
+ save_lustre_params $HOSTNAME osc.*.lockless_truncate > $p
+ cancel_lru_locks osc
+ enable_lockless_truncate 1
+ rm -f $DIR1/$tfile
+ lfs setstripe -c -1 $DIR1/$tfile
+ dd if=/dev/zero of=$DIR1/$tfile count=10 bs=1M > /dev/null 2>&1
+ clear_osc_stats
+
+ log "checking cached lockless truncate"
+ $TRUNCATE $DIR1/$tfile 8000000
+ $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size"
+ [ $(calc_osc_stats lockless_truncate) -eq 0 ] ||
+ error "lockless truncate doesn't use cached locks"
+
+ 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"
+
+ log "disabled lockless truncate"
+ enable_lockless_truncate 0
+ clear_osc_stats
+ $TRUNCATE $DIR2/$tfile 3000000
+ $CHECKSTAT -s 3000000 $DIR1/$tfile || error "wrong file size"
+ [ $(calc_osc_stats lockless_truncate) -eq 0 ] ||
+ error "lockless truncate disabling failed"
+ rm $DIR1/$tfile
+ # restore lockless_truncate default values
+ restore_lustre_params < $p
+ rm -f $p
+}
+run_test 32a "lockless truncate"
+
+test_32b() { # bug 11270
+ remote_ost_nodsh && skip "remote OST with nodsh" && return
+
+ local node
+ local p="$TMP/sanityN-$TESTNAME.parameters"
+ save_lustre_params $HOSTNAME "osc.*.contention_seconds" > $p
+ for node in $(osts_nodes); do
+ save_lustre_params $node "ldlm.namespaces.filter-*.max_nolock_bytes" >> $p
+ save_lustre_params $node "ldlm.namespaces.filter-*.contended_locks" >> $p
+ save_lustre_params $node "ldlm.namespaces.filter-*.contention_seconds" >> $p
+ done
+ clear_osc_stats
+ # 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'
+ done
+ lctl set_param -n osc.*.contention_seconds 60
+ for i in $(seq 5); do
+ dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1
+ dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1
+ done
+ [ $(calc_osc_stats lockless_write_bytes) -ne 0 ] || error "lockless i/o was not triggered"
+ # disable lockless i/o (it is disabled by default)
+ for node in $(osts_nodes); do
+ do_node $node 'lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes 0; lctl set_param -n ldlm.namespaces.filter-*.contended_locks 32; lctl set_param -n ldlm.namespaces.filter-*.contention_seconds 0'
+ done
+ # set contention_seconds to 0 at client too, otherwise Lustre still
+ # remembers lock contention
+ lctl set_param -n osc.*.contention_seconds 0
+ clear_osc_stats
+ for i in $(seq 1); do
+ dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1
+ dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > /dev/null 2>&1
+ done
+ [ $(calc_osc_stats lockless_write_bytes) -eq 0 ] ||
+ error "lockless i/o works when disabled"
+ rm -f $DIR1/$tfile
+ restore_lustre_params <$p
+ rm -f $p
+}
+run_test 32b "lockless i/o"
+
+print_jbd_stat () {
+ local dev
+ local mdts=$(get_facets MDS)
+ local varcvs
+ local mds
+
+ 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 $mds cat /proc/fs/jbd/$dev/info | head -1 | cut -d" " -f1)
+ stat=$(( stat + val))
+ done
+ echo $stat
+}
+
+# commit on sharing tests
+test_33a() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
+
+ [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; }
+ [ $CLIENTCOUNT -ge 2 ] || \
+ { skip "Need two or more clients, have $CLIENTCOUNT" && return 0; }
+
+ zconf_mount_clients $CLIENT1,$CLIENT2 $DIR1
+ zconf_mount_clients $CLIENT1,$CLIENT2 $DIR2
+
+ local nfiles=${TEST33_NFILES:-10000}
+ local param_file=$TMP/$tfile-params
+
+ save_lustre_params $(comma_list $(mdts_nodes)) "mdt.*.commit_on_sharing" > $param_file
+
+ local COS
+ local jbdold
+ local jbdnew
+ local jbd
+
+ for COS in 0 1; do
+ do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=$COS
+ avgjbd=0
+ avgtime=0
+ for i in 1 2 3; do
+ do_nodes $CLIENT1,$CLIENT2 "mkdir -p $DIR1/$tdir-\\\$(hostname)-$i"
+
+ jbdold=$(print_jbd_stat)
+ echo "=== START createmany old: $jbdold transaction"
+ local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- -r $DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > /dev/null 2>&1")
+ jbdnew=$(print_jbd_stat)
+ jbd=$(( jbdnew - jbdold ))
+ echo "=== END createmany new: $jbdnew transaction : $jbd transactions nfiles $nfiles time $ELAPSED COS=$COS"
+ avgjbd=$(( avgjbd + jbd ))
+ avgtime=$(( avgtime + elapsed ))
+ done
+ eval cos${COS}_jbd=$((avgjbd / 3))
+ eval cos${COS}_time=$((avgtime / 3))
+ done
+
+ echo "COS=0 transactions (avg): $cos0_jbd time (avg): $cos0_time"
+ echo "COS=1 transactions (avg): $cos1_jbd time (avg): $cos1_time"
+ [ "$cos0_jbd" != 0 ] && echo "COS=1 vs COS=0 jbd: $((((cos1_jbd/cos0_jbd - 1)) * 100 )) %"
+ [ "$cos0_time" != 0 ] && echo "COS=1 vs COS=0 time: $((((cos1_time/cos0_time - 1)) * 100 )) %"
+
+ restore_lustre_params < $param_file
+ rm -f $param_file
+ return 0
+}
+run_test 33a "commit on sharing, cross crete/delete, 2 clients, benchmark"
+
+# End commit on sharing tests
+
+test_34() { #16129
+ local OPER
+ local lock_in
+ local lock_out
+ for OPER in notimeout timeout ; do
+ rm $DIR1/$tfile 2>/dev/null
+ lock_in=$(do_nodes $(osts_nodes) "lctl get_param -n ldlm.namespaces.filter-*.lock_timeouts" | calc_sum)
+ if [ $OPER == "timeout" ] ; then
+ for j in `seq $OSTCOUNT`; do
+ #define OBD_FAIL_PTLRPC_HPREQ_TIMEOUT 0x511
+ do_facet ost$j lctl set_param fail_loc=0x511
+ done
+ echo lock should expire
+ else
+ for j in `seq $OSTCOUNT`; do
+ #define OBD_FAIL_PTLRPC_HPREQ_NOTIMEOUT 0x512
+ do_facet ost$j lctl set_param fail_loc=0x512
+ done
+ echo lock should not expire
+ fi
+ echo writing on client1
+ dd if=/dev/zero of=$DIR1/$tfile count=100 conv=notrunc > /dev/null 2>&1
+ sync &
+ echo reading on client2
+ dd of=/dev/null if=$DIR2/$tfile > /dev/null 2>&1
+ # wait for a lock timeout
+ sleep 4
+ lock_out=$(do_nodes $(osts_nodes) "lctl get_param -n ldlm.namespaces.filter-*.lock_timeouts" | calc_sum)
+ if [ $OPER == "timeout" ] ; then
+ if [ $lock_in == $lock_out ]; then
+ error "no lock timeout happened"
+ else
+ echo "success"
+ fi
+ else
+ if [ $lock_in != $lock_out ]; then
+ error "lock timeout happened"
+ else
+ echo "success"
+ fi
+ fi
+ done
+}
+run_test 34 "no lock timeout under IO"
+
+test_35() { # bug 17645
+ local generation=[]
+ local count=0
+ for imp in /proc/fs/lustre/mdc/$FSNAME-MDT*-mdc-*; do
+ g=$(awk '/generation/{print $2}' $imp/import)
+ generation[count]=$g
+ let count=count+1
+ done
+
+ mkdir -p $MOUNT1/$tfile
+ cancel_lru_locks mdc
+
+ # Let's initiate -EINTR situation by setting fail_loc and take
+ # write lock on same file from same client. This will not cause
+ # bl_ast yet as lock is already in local cache.
+#define OBD_FAIL_LDLM_INTR_CP_AST 0x317
+ do_facet client "lctl set_param fail_loc=0x80000317"
+ local timeout=`do_facet $SINGLEMDS lctl get_param -n timeout`
+ let timeout=timeout*3
+ local nr=0
+ while test $nr -lt 10; do
+ log "Race attempt $nr"
+ local blk1=`lctl get_param -n ldlm.services.ldlm_cbd.stats | awk '/ldlm_bl_callback/ {print $2}'`
+ test "x$blk1" = "x" && blk1=0
+ createmany -o $MOUNT2/$tfile/a 4000 &
+ pid1=$!
+ sleep 1
+
+ # Let's make conflict and bl_ast
+ ls -la $MOUNT1/$tfile > /dev/null &
+ pid2=$!
+
+ log "Wait for $pid1 $pid2 for $timeout sec..."
+ sleep $timeout
+ kill -9 $pid1 $pid2 > /dev/null 2>&1
+ wait
+ local blk2=`lctl get_param -n ldlm.services.ldlm_cbd.stats | awk '/ldlm_bl_callback/ {print $2}'`
+ test "x$blk2" = "x" && blk2=0
+ test $blk2 -gt $blk1 && break
+ rm -fr $MOUNT1/$tfile/*
+ cancel_lru_locks mdc
+ let nr=nr+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)"
+ fi
+ let count=count+1
+ done
+}
+run_test 35 "-EINTR cp_ast vs. bl_ast race does not evict client"
+
+test_36() { #bug 16417
+ local SIZE
+ local SIZE_B
+ local i
+
+ mkdir -p $DIR1/$tdir
+ $LFS setstripe -c -1 $DIR1/$tdir
+ i=0
+ SIZE=50
+ let SIZE_B=SIZE*1024*1024
+
+ while [ $i -le 10 ]; do
+ lctl mark "start test"
+ local before=$($LFS df | awk '{if ($1 ~/^filesystem/) {print $5; exit} }')
+ dd if=/dev/zero of=$DIR1/$tdir/file000 bs=1M count=$SIZE
+ sync
+ sleep 1
+ local after_dd=$($LFS df | awk '{if ($1 ~/^filesystem/) {print $5; exit} }')
+ multiop_bg_pause $DIR2/$tdir/file000 O_r${SIZE_B}c || return 3
+ read_pid=$!
+ rm -f $DIR1/$tdir/file000
+ kill -USR1 $read_pid
+ wait $read_pid
+ sleep 1
+ local after=$($LFS df | awk '{if ($1 ~/^filesystem/) {print $5; exit} }')
+ echo "*** cycle($i) *** before($before):after_dd($after_dd):after($after)"
+ # this free space! not used
+ if [ $after_dd -ge $after ]; then
+ error "space leaked"
+ return 1;
+ fi
+ let i=i+1
+ done
+}
+run_test 36 "handle ESTALE/open-unlink corectly"
log "cleanup: ======================================================"