require_dsh_mds || exit 0
# also long tests: 19, 21a, 21e, 21f, 23, 27
-# 1 2.5 2.5 4 4 (min)"
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="17 26a 26b 50 51 57"
-[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test: LU-2194 LU-2547
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 19b 24a 24b"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
build_test_filter
}
run_test 10b "re-send BL AST"
+test_10c() {
+ local before=$(date +%s)
+ local evict
+ local mdccli
+ local mdcpath
+ local conn_uuid
+ local workdir
+ local pid
+ local rc
+
+ workdir="${DIR}/${tdir}"
+ mkdir -p ${workdir} || error "can't create workdir $?"
+ stat ${workdir} > /dev/null ||
+ error "failed to stat ${workdir}: $?"
+ mdtidx=$($LFS getdirstripe -i ${workdir})
+ mdtname=$($LFS mdts ${workdir} | grep -e "^$mdtidx:" |
+ awk '{sub("_UUID", "", $2); print $2;}')
+ #assume one client
+ mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
+ conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
+ mdcpath="mdc.${mdccli}.import=connection=${conn_uuid}"
+
+ drop_bl_callback_once "chmod 0777 ${workdir}" &
+ pid=$!
+
+ # let chmod blocked
+ sleep 1
+ # force client reconnect
+ $LCTL set_param "${mdcpath}"
+
+ # wait client reconnect
+ client_reconnect
+ wait $pid
+ rc=$?
+ evict=$($LCTL get_param mdc.${mdccli}.state |
+ awk -F"[ [,]" '/EVICTED]$/ { if (t<$4) {t=$4;} } END { print t }')
+
+ [[ $evict -le $before ]] ||
+ ( $LCTL get_param mdc.$FSNAME-MDT*.state;
+ error "eviction happened: $EVICT before:$BEFORE" )
+
+ [ $rc -eq 0 ] || error "chmod must finished OK"
+ checkstat -v -p 0777 "${workdir}" ||
+ error "client checkstat failed: $?"
+}
+run_test 10c "re-send BL AST vs reconnect race (LU-5569)"
+
test_10d() {
local before=$(date +%s)
local evict
+
+ [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ]] &&
+ skip "Need MDS version at least 2.6.90" && return
+
# sleep 1 is to make sure that BEFORE is not equal to EVICTED below
sleep 1
rm -f $TMP/$tfile
echo -n ", world" | dd of=$TMP/$tfile bs=1c seek=5
+ remount_client $MOUNT
mount_client $MOUNT2
+ cancel_lru_locks osc
$LFS setstripe -i 0 -c 1 $DIR1/$tfile
- echo -n hello > $DIR1/$tfile
+ echo -n hello | dd of=$DIR1/$tfile bs=5
stat $DIR2/$tfile >& /dev/null
$LCTL set_param fail_err=71
- drop_bl_callback_once "echo -n \\\", world\\\" >> $DIR2/$tfile"
+ drop_bl_callback "echo -n \\\", world\\\" >> $DIR2/$tfile"
client_reconnect
- cmp $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
- cmp $DIR1/$tfile $TMP/$tfile || error "wrong content found"
+ cancel_lru_locks osc
+ cmp -l $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
+ cmp -l $DIR1/$tfile $TMP/$tfile || error "wrong content found"
evict=$(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state | \
tr -d '\-\[\] ' | \
(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state;
error "no eviction: $evict before:$before")
+ $LCTL set_param fail_err=0
rm $TMP/$tfile
umount_client $MOUNT2
}
do_facet ost1 lctl set_param fail_loc=0x80000225
# force reconnect
sleep 1
- df $MOUNT > /dev/null 2>&1
+ $LFS df $MOUNT > /dev/null 2>&1
sleep 2
# my understanding is that there should be nothing in the page
- # cache after the client reconnects?
+ # cache after the client reconnects?
rc=0
pgcache_empty || rc=2
rm -f $f $TMP/$tfile
replay_barrier $SINGLEMDS
createmany -o $DIR/$tdir/$tfile-%d 10
- local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
+ local oid=$(do_facet ost1 "lctl get_param -n \
+ obdfilter.${ost1_svc}.last_id" | sed -e 's/.*://')
fail_abort $SINGLEMDS
-
+
touch $DIR/$tdir/$tfile
- local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'`
+ local id=$($LFS getstripe $DIR/$tdir/$tfile |
+ awk '$1 == 0 { print $2 }')
[ $id -le $oid ] && error "the orphan objid was reused, failed"
# Cleanup
#}
#run_test 62 "Verify connection flags race - bug LU-1716"
+test_66()
+{
+ [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] ||
+ { skip "Need MDS version at least 2.7.51"; return 0; }
+
+ local list=$(comma_list $(osts_nodes))
+
+ # modify dir so that next revalidate would not obtain UPDATE lock
+ touch $DIR
+
+ # drop 1 reply with UPDATE lock
+ mcreate $DIR/$tfile || error "mcreate failed: $?"
+ drop_ldlm_reply_once "stat $DIR/$tfile" &
+ sleep 2
+
+ # make the re-sent lock to sleep
+#define OBD_FAIL_MDS_RESEND 0x136
+ do_nodes $list $LCTL set_param fail_loc=0x80000136
+
+ #initiate the re-connect & re-send
+ local mdccli=$($LCTL dl | awk '/-mdc-/ {print $4;}')
+ local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
+ $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
+ sleep 2
+
+ #initiate the client eviction while enqueue re-send is in progress
+ mds_evict_client
+
+ client_reconnect
+ wait
+}
+run_test 66 "lock enqueue re-send vs client eviction"
+
+test_65() {
+ mount_client $DIR2
+
+ #grant lock1, export2
+ $SETSTRIPE -i -0 $DIR2/$tfile || return 1
+ $MULTIOP $DIR2/$tfile Ow || return 2
+
+#define OBD_FAIL_LDLM_BL_EVICT 0x31e
+ do_facet ost $LCTL set_param fail_loc=0x31e
+ #get waiting lock2, export1
+ $MULTIOP $DIR/$tfile Ow &
+ PID1=$!
+ # let enqueue to get asleep
+ sleep 2
+
+ #get lock2 blocked
+ $MULTIOP $DIR2/$tfile Ow &
+ PID2=$!
+ sleep 2
+
+ #evict export1
+ ost_evict_client
+
+ sleep 2
+ do_facet ost $LCTL set_param fail_loc=0
+
+ wait $PID1
+ wait $PID2
+
+ umount_client $DIR2
+}
+run_test 65 "lock enqueue for destroyed export"
+
check_cli_ir_state()
{
local NODE=${1:-$HOSTNAME}
local recovery_proc=obdfilter.${!name}.recovery_status
local st
+ while : ; do
+ st=$(do_facet $target "$LCTL get_param -n $recovery_proc |
+ awk '/status:/{ print \\\$2}'")
+ [ x$st = xRECOVERING ] || break
+ done
st=$(do_facet $target "lctl get_param -n $recovery_proc |
awk '/IR:/{ print \\\$2}'")
[ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
}
run_test 107 "drop reint reply, then restart MDT"
+test_108() {
+ mkdir -p $DIR/$tdir
+ $SETSTRIPE -c 1 -i 0 $DIR/$tdir
+
+ dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
+ local dd_pid=$!
+ sleep 0.1
+
+ ost_evict_client
+
+ wait $dd_pid
+
+ client_up || error "reconnect failed"
+ rm -f $DIR/$tdir/$tfile
+}
+run_test 108 "client eviction don't crash"
+
test_110a () {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
local remote_dir=$DIR/$tdir/remote_dir
createmany -o $remote_dir/f 100
- #define OBD_FAIL_MIGRATE_NET_REP 0x1702
- do_facet mds$MDTIDX lctl set_param fail_loc=0x1702
- $LFS mv -M $MDTIDX $remote_dir || error "migrate failed"
+ #define OBD_FAIL_MIGRATE_NET_REP 0x1800
+ do_facet mds$MDTIDX lctl set_param fail_loc=0x1800
+ $LFS migrate -m $MDTIDX $remote_dir || error "migrate failed"
do_facet mds$MDTIDX lctl set_param fail_loc=0x0
for file in $(find $remote_dir); do
}
run_test 110g "drop reply during migration"
+test_110h () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ local src_dir=$DIR/$tdir/source_dir
+ local tgt_dir=$DIR/$tdir/target_dir
+ local MDTIDX=1
+
+ mkdir -p $src_dir
+ $LFS mkdir -i $MDTIDX $tgt_dir
+
+ dd if=/etc/hosts of=$src_dir/src_file
+ touch $tgt_dir/tgt_file
+ drop_update_reply $MDTIDX \
+ "mrename $src_dir/src_file $tgt_dir/tgt_file" ||
+ error "mrename failed"
+
+ $CHECKSTAT -t file $src_dir/src_file &&
+ error "src_file present after rename"
+
+ diff /etc/hosts $tgt_dir/tgt_file ||
+ error "file changed after rename"
+
+}
+run_test 110h "drop update reply during cross-MDT file rename"
+
+test_110i () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ local src_dir=$DIR/$tdir/source_dir
+ local tgt_dir=$DIR/$tdir/target_dir
+ local MDTIDX=1
+
+ mkdir -p $src_dir
+ $LFS mkdir -i $MDTIDX $tgt_dir
+
+ mkdir $src_dir/src_dir
+ touch $src_dir/src_dir/a
+ mkdir $tgt_dir/tgt_dir
+ drop_update_reply $MDTIDX \
+ "mrename $src_dir/src_dir $tgt_dir/tgt_dir" ||
+ error "mrename failed"
+
+ $CHECKSTAT -t dir $src_dir/src_dir &&
+ error "src_dir present after rename"
+
+ $CHECKSTAT -t dir $tgt_dir/tgt_dir ||
+ error "tgt_dir not present after rename"
+
+ $CHECKSTAT -t file $tgt_dir/tgt_dir/a ||
+ error "a not present after rename"
+}
+run_test 110i "drop update reply during cross-MDT dir rename"
+
+test_110j () {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ local remote_dir=$DIR/$tdir/remote_dir
+ local local_dir=$DIR/$tdir/local_dir
+ local MDTIDX=1
+
+ mkdir -p $DIR/$tdir
+ mkdir $DIR/$tdir/local_dir
+ $LFS mkdir -i $MDTIDX $remote_dir
+
+ touch $local_dir/local_file
+ drop_update_reply $MDTIDX \
+ "ln $local_dir/local_file $remote_dir/remote_file" ||
+ error "ln failed"
+
+ $CHECKSTAT -t file $remote_dir/remote_file ||
+ error "remote not present after ln"
+}
+run_test 110j "drop update reply during cross-MDT ln"
+
# LU-2844 mdt prepare fail should not cause umount oops
test_111 ()
{
}
run_test 112a "bulk resend while orignal request is in progress"
+test_115_read() {
+ local fail1=$1
+ local fail2=$2
+
+ df $DIR
+ dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1
+ cancel_lru_locks osc
+
+ # OST_READ = 3,
+ $LCTL set_param fail_loc=$fail1 fail_val=3
+ dd of=/dev/null if=$DIR/$tfile bs=4096 count=1 &
+ pid=$!
+ sleep 1
+
+ set_nodes_failloc "$(osts_nodes)" $fail2
+
+ wait $pid || error "dd failed"
+ return 0
+}
+
+test_115_write() {
+ local fail1=$1
+ local fail2=$2
+ local error=$3
+ local fail_val2=${4:-0}
+
+ df $DIR
+ touch $DIR/$tfile
+
+ # OST_WRITE = 4,
+ $LCTL set_param fail_loc=$fail1 fail_val=4
+ dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 oflag=dsync &
+ pid=$!
+ sleep 1
+
+ df $MOUNT
+ set_nodes_failloc "$(osts_nodes)" $fail2 $fail_val2
+
+ wait $pid
+ rc=$?
+ [ $error -eq 0 ] && [ $rc -ne 0 ] && error "dd error ($rc)"
+ [ $error -ne 0 ] && [ $rc -eq 0 ] && error "dd success"
+ return 0
+}
+
+test_115a() {
+ [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST" && return 0
+
+ #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
+ #define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
+ test_115_read 0x8000051b 0x8000051a
+}
+run_test 115a "read: late REQ MDunlink and no bulk"
+
+test_115b() {
+ [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST" && return 0
+
+ #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
+ #define OBD_FAIL_OST_ENOSPC 0x215
+
+ # pass $OSTCOUNT for the fail_loc to be caught
+ # appropriately by the IO thread
+ test_115_write 0x8000051b 0x80000215 1 $OSTCOUNT
+}
+run_test 115b "write: late REQ MDunlink and no bulk"
+
+test_115c() {
+ [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST" && return 0
+
+ #define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
+ #define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
+ test_115_read 0x8000050f 0x8000051a
+}
+run_test 115c "read: late Reply MDunlink and no bulk"
+
+test_115d() {
+ [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST" && return 0
+
+ #define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
+ #define OBD_FAIL_OST_ENOSPC 0x215
+ test_115_write 0x8000050f 0x80000215 0
+}
+run_test 115d "write: late Reply MDunlink and no bulk"
+
+test_115e() {
+ [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST" && return 0
+
+ #define OBD_FAIL_PTLRPC_LONG_BULK_UNLINK 0x510
+ #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
+ test_115_read 0x80000510 0x80000211
+}
+run_test 115e "read: late Bulk MDunlink and no reply"
+
+test_115f() {
+ [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST" && return 0
+
+ #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
+ #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
+ test_115_read 0x8000051b 0x80000211
+}
+run_test 115f "read: late REQ MDunlink and no reply"
+
+test_115g() {
+ [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST" && return 0
+
+ #define OBD_FAIL_PTLRPC_LONG_BOTH_UNLINK 0x51c
+ test_115_read 0x8000051c 0
+}
+run_test 115g "read: late REQ MDunlink and Reply MDunlink"
+
# parameters: fail_loc CMD RC
test_120_reply() {
local PID
}
run_test 113 "ldlm enqueue dropped reply should not cause deadlocks"
+T130_PID=0
+test_130_base() {
+ test_mkdir -p $DIR/$tdir
+
+ # Prevent interference from layout intent RPCs due to
+ # asynchronous writeback. These will be tested in 130c below.
+ do_nodes ${CLIENTS:-$HOSTNAME} sync
+
+ # get only LOOKUP lock on $tdir
+ cancel_lru_locks mdc
+ ls $DIR/$tdir/$tfile 2>/dev/null
+
+ # get getattr by fid on $tdir
+ #
+ # we need to race with unlink, unlink must complete before we will
+ # take a DLM lock, otherwise unlink will wait until getattr will
+ # complete; but later than getattr starts so that getattr found
+ # the object
+#define OBD_FAIL_MDS_INTENT_DELAY 0x160
+ set_nodes_failloc "$(mdts_nodes)" 0x80000160
+ stat $DIR/$tdir &
+ T130_PID=$!
+ sleep 2
+
+ rm -rf $DIR/$tdir
+
+ # drop the reply so that resend happens on an unlinked file.
+#define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
+ set_nodes_failloc "$(mdts_nodes)" 0x80000157
+}
+
+test_130a() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
+ test_130_base
+
+ wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
+ return 0
+}
+run_test 130a "enqueue resend on not existing file"
+
+test_130b() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
+ test_130_base
+ # let the reply to be dropped
+ sleep 10
+
+#define OBD_FAIL_SRV_ENOENT 0x217
+ set_nodes_failloc "$(mdts_nodes)" 0x80000217
+
+ wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
+ return 0
+}
+run_test 130b "enqueue resend on a stale inode"
+
+test_130c() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return
+
+ do_nodes ${CLIENTS:-$HOSTNAME} sync
+ echo XXX > $DIR/$tfile
+
+ cancel_lru_locks mdc
+
+ # Trigger writeback on $tfile.
+ #
+ # we need to race with unlink, unlink must complete before we will
+ # take a DLM lock, otherwise unlink will wait until intent will
+ # complete; but later than intent starts so that intent found
+ # the object
+#define OBD_FAIL_MDS_INTENT_DELAY 0x160
+ set_nodes_failloc "$(mdts_nodes)" 0x80000160
+ sync &
+ T130_PID=$!
+ sleep 2
+
+ rm $DIR/$tfile
+
+ # drop the reply so that resend happens on an unlinked file.
+#define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
+ set_nodes_failloc "$(mdts_nodes)" 0x80000157
+
+ # let the reply to be dropped
+ sleep 10
+
+#define OBD_FAIL_SRV_ENOENT 0x217
+ set_nodes_failloc "$(mdts_nodes)" 0x80000217
+
+ wait $T130_PID
+
+ return 0
+}
+run_test 130c "layout intent resend on a stale inode"
+
+test_132() {
+ local before=$(date +%s)
+ local evict
+
+ mount_client $MOUNT2 || error "mount filed"
+
+ rm -f $DIR/$tfile
+ # get a lock on client so that export would reach the stale list
+ $SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
+ dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync ||
+ error "dd failed"
+
+ #define OBD_FAIL_OST_PAUSE_PUNCH 0x236
+ do_facet ost1 $LCTL set_param fail_val=120 fail_loc=0x80000236
+
+ $TRUNCATE $DIR/$tfile 100 &
+
+ sleep 1
+ dd if=/dev/zero of=$DIR2/$tfile bs=4096 count=1 conv=notrunc ||
+ error "dd failed"
+
+ wait
+ umount_client $MOUNT2
+
+ evict=$(do_facet client $LCTL get_param \
+ osc.$FSNAME-OST0000-osc-*/state |
+ awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }')
+
+ [ -z "$evict" ] || [[ $evict -le $before ]] ||
+ (do_facet client $LCTL get_param \
+ osc.$FSNAME-OST0000-osc-*/state;
+ error "eviction happened: $evict before:$before")
+}
+run_test 132 "long punch"
+
+test_131() {
+ remote_ost_nodsh && skip "remote OST with nodsh" && return 0
+
+ rm -f $DIR/$tfile
+ # get a lock on client so that export would reach the stale list
+ $SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
+ dd if=/dev/zero of=$DIR/$tfile count=1 || error "dd failed"
+
+ # another IO under the same lock
+ #define OBD_FAIL_OSC_DELAY_IO 0x414
+ $LCTL set_param fail_loc=0x80000414
+ dd if=/dev/zero of=$DIR/$tfile count=1 conv=notrunc oflag=dsync &
+ local pid=$!
+ sleep 1
+
+ #define OBD_FAIL_LDLM_BL_EVICT 0x31e
+ set_nodes_failloc "$(osts_nodes)" 0x8000031e
+ ost_evict_client
+ client_reconnect
+
+ wait $pid && error "dd succeeded"
+ return 0
+}
+run_test 131 "IO vs evict results to IO under staled lock"
+
+test_133() {
+ local list=$(comma_list $(mdts_nodes))
+
+ local t=$((TIMEOUT * 2))
+ touch $DIR/$tfile
+
+ flock $DIR/$tfile -c "echo bl lock;sleep $t;echo bl flock unlocked" &
+ sleep 1
+ multiop_bg_pause $DIR/$tfile O_jc || return 1
+ PID=$!
+
+ #define OBD_FAIL_LDLM_REPLY 0x30c
+ do_nodes $list $LCTL set_param fail_loc=0x8000030c
+ kill -USR1 $PID
+ echo "waiting for multiop $PID"
+ wait $PID || return 2
+
+ rm -f $DIR/$tfile
+
+ return 0
+}
+run_test 133 "don't fail on flock resend"
+
complete $SECONDS
check_and_cleanup_lustre
exit_status
git://git.whamcloud.com / fs / lustre-release.git / blobdiff