#!/bin/sh
-set -ex
+set -e
+
+# 17 = bug 2732 2986
+ALWAYS_EXCEPT="17 19b"
+
LUSTRE=${LUSTRE:-`dirname $0`/..}
-LTESTDIR=${LTESTDIR:-$LUSTRE/../ltest}
-PATH=$PATH:$LUSTRE/utils:$LUSTRE/tests
+UPCALL=${UPCALL:-$PWD/recovery-small-upcall.sh}
+. $LUSTRE/tests/test-framework.sh
+
+init_test_env $@
+
+. ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
-RLUSTRE=${RLUSTRE:-$LUSTRE}
-RPWD=${RPWD:-$PWD}
+build_test_filter
-. $LTESTDIR/functional/llite/common/common.sh
# Allow us to override the setup if we already have a mounted system by
# setting SETUP=" " and CLEANUP=" "
SETUP=${SETUP:-"setup"}
CLEANUP=${CLEANUP:-"cleanup"}
-PDSH=${PDSH:-'pdsh -S -w'}
-# XXX I wish all this stuff was in some default-config.sh somewhere
-MDSNODE=${MDSNODE:-mdev6}
-OSTNODE=${OSTNODE:-mdev7}
-CLIENT=${CLIENT:-mdev8}
-NETWORKTYPE=${NETWORKTYPE:-tcp}
-MOUNTPT=${MOUNTPT:-/mnt/lustre}
-CONFIG=${CONFIG:-recovery-small.xml}
-MDSDEV=${MDSDEV:-/tmp/mds-`hostname`}
-MDSSIZE=${MDSSIZE:-100000}
-OSTDEV=${OSTDEV:-/tmp/ost-`hostname`}
-OSTSIZE=${OSTSIZE:-100000}
-UPCALL=${UPCALL:-$RPWD/recovery-small-upcall.sh}
-FSTYPE=${FSTYPE:-ext3}
-TIMEOUT=${TIMEOUT:-5}
-do_mds() {
- $PDSH $MDSNODE "PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests; cd $RPWD; $@" || exit $?
+make_config() {
+ rm -f $XMLCONFIG
+ add_mds mds --dev $MDSDEV --size $MDSSIZE
+ add_lov lov1 mds --stripe_sz $STRIPE_BYTES\
+ --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
+ add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
+ add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE
+ add_client client mds --lov lov1 --path $MOUNT
}
-do_client() {
- $PDSH $CLIENT "PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests; cd $RPWD; $@" || exit $?
+setup() {
+ make_config
+ start ost --reformat $OSTLCONFARGS
+ start ost2 --reformat $OSTLCONFARGS
+ [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
+ start mds $MDSLCONFARGS --reformat
+ zconf_mount `hostname` $MOUNT
}
-do_ost() {
- $PDSH $OSTNODE "PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests; cd $RPWD; $@" || exit $?
+cleanup() {
+ zconf_umount `hostname` $MOUNT
+ stop mds ${FORCE} $MDSLCONFARGS
+ stop ost2 ${FORCE} --dump cleanup.log
+ stop ost ${FORCE} --dump cleanup.log
}
-drop_request() {
-# OBD_FAIL_MDS_ALL_REQUEST_NET
- do_mds "echo 0x123 > /proc/sys/lustre/fail_loc"
- do_client "$1"
- do_mds "echo 0 > /proc/sys/lustre/fail_loc"
+replay() {
+ do_mds "sync"
+ do_mds 'echo -e "device \$mds1\\nprobe\\nnotransno\\nreadonly" | lctl'
+ do_client "$1" &
+ shutdown_mds -f
+ start_mds
+ wait
+ do_client "df -h $MOUNT" # trigger failover, if we haven't already
}
-drop_reply() {
-# OBD_FAIL_MDS_ALL_REPLY_NET
- do_mds "echo 0x122 > /proc/sys/lustre/fail_loc"
- do_client "$@"
- do_mds "echo 0 > /proc/sys/lustre/fail_loc"
-}
+if [ ! -z "$EVAL" ]; then
+ eval "$EVAL"
+ exit $?
+fi
-pause_bulk() {
-#define OBD_FAIL_OST_BRW_PAUSE_BULK 0x214
- do_ost "echo 0x214 > /proc/sys/lustre/fail_loc"
- do_client "$1"
- do_client "sync"
- do_ost "echo 0 > /proc/sys/lustre/fail_loc"
-}
-make_config() {
- rm -f $CONFIG
- for NODE in $CLIENT $MDSNODE $OSTNODE; do
- lmc -m $CONFIG --add net --node $NODE --nid `h2$NETWORKTYPE $NODE` \
- --nettype $NETWORKTYPE || exit 4
- done
- lmc -m $CONFIG --add mds --node $MDSNODE --mds mds1 --dev $MDSDEV \
- --size $MDSSIZE --fstype $FSTYPE || exit 5
- lmc -m $CONFIG --add ost --node $OSTNODE --ost ost1 --dev $OSTDEV \
- --size $OSTSIZE --fstype $FSTYPE || exit 6
- lmc -m $CONFIG --add mtpt --node $CLIENT --path $MOUNTPT --mds mds1 \
- --ost ost1 || exit 7
-}
+if [ "$ONLY" == "cleanup" ]; then
+ sysctl -w portals.debug=0 || true
+ cleanup
+ exit
+fi
-start_mds() {
- do_mds "lconf $@ $CONFIG"
-}
+REFORMAT=--reformat $SETUP
+unset REFORMAT
-shutdown_mds() {
- do_mds "lconf $@ --cleanup $CONFIG"
+test_1() {
+ drop_request "mcreate $MOUNT/1" || return 1
+ drop_reply "mcreate $MOUNT/2" || return 2
}
+run_test 1 "mcreate: drop req, drop rep"
-start_ost() {
- do_ost "lconf $@ $CONFIG"
+test_2() {
+ drop_request "tchmod 111 $MOUNT/2" || return 1
+ drop_reply "tchmod 666 $MOUNT/2" || return 2
}
+run_test 2 "chmod: drop req, drop rep"
-shutdown_ost() {
- do_ost "lconf $@ --cleanup $CONFIG"
+test_3() {
+ drop_request "statone $MOUNT/2" || return 1
+ drop_reply "statone $MOUNT/2" || return 2
}
+run_test 3 "stat: drop req, drop rep"
-mount_client() {
- do_client "lconf $@ $CONFIG"
+test_4() {
+ do_facet client "cp /etc/resolv.conf $MOUNT/resolv.conf" || return 1
+ drop_request "cat $MOUNT/resolv.conf > /dev/null" || return 2
+ drop_reply "cat $MOUNT/resolv.conf > /dev/null" || return 3
}
+run_test 4 "open: drop req, drop rep"
-unmount_client() {
- do_client "lconf $@ --cleanup $CONFIG"
+test_5() {
+ drop_request "mv $MOUNT/resolv.conf $MOUNT/renamed" || return 1
+ drop_reply "mv $MOUNT/renamed $MOUNT/renamed-again" || return 2
+ do_facet client "checkstat -v $MOUNT/renamed-again" || return 3
}
+run_test 5 "rename: drop req, drop rep"
-setup() {
- start_mds --timeout=$TIMEOUT ${REFORMAT}
- start_ost --timeout=$TIMEOUT ${REFORMAT}
- # XXX we should write our own upcall, when we move this somewhere better.
- mount_client --timeout=${TIMEOUT} \
- --lustre_upcall=$UPCALL
+test_6() {
+ drop_request "mlink $MOUNT/renamed-again $MOUNT/link1" || return 1
+ drop_reply "mlink $MOUNT/renamed-again $MOUNT/link2" || return 2
}
+run_test 6 "link: drop req, drop rep"
-cleanup() {
- do_mds "echo 0 > /proc/sys/lustre/fail_loc"
- unmount_client $@ || true
- shutdown_mds $@ || true
- shutdown_ost $@ || true
+test_7() {
+ drop_request "munlink $MOUNT/link1" || return 1
+ drop_reply "munlink $MOUNT/link2" || return 2
}
+run_test 7 "unlink: drop req, drop rep"
-replay() {
- do_mds "sync"
- do_mds 'echo -e "device \$mds1\\nprobe\\nnotransno\\nreadonly" | lctl'
- do_client "$1" &
- shutdown_mds -f
- start_mds
- wait
- do_client "df -h $MOUNTPT" # trigger failover, if we haven't already
+
+#bug 1423
+test_8() {
+ drop_reply "touch $MOUNT/renamed" || return 1
}
+run_test 8 "touch: drop rep (bug 1423)"
-if [ ! -z "$ONLY" ]; then
- eval "$ONLY"
- exit $?
-fi
-make_config
+#bug 1420
+test_9() {
+ pause_bulk "cp /etc/profile $MOUNT" || return 1
+ do_facet client "cp /etc/termcap $MOUNT" || return 2
+ do_facet client "sync"
+ do_facet client "rm $MOUNT/termcap $MOUNT/profile" || return 3
+}
+run_test 9 "pause bulk on OST (bug 1420)"
+
+#bug 1521
+test_10() {
+ do_facet client mcreate $MOUNT/f10 || return 1
+ drop_bl_callback "chmod 0777 $MOUNT/f10" || return 2
+ # wait for the mds to evict the client
+ #echo "sleep $(($TIMEOUT*2))"
+ #sleep $(($TIMEOUT*2))
+ do_facet client touch $MOUNT/f10 || echo "touch failed, evicted"
+ do_facet client checkstat -v -p 0777 $MOUNT/f10 || return 3
+ do_facet client "munlink $MOUNT/f10"
+}
+run_test 10 "finish request on server after client eviction (bug 1521)"
-REFORMAT=--reformat $SETUP
-unset REFORMAT
+#bug 2460
+# wake up a thead waiting for completion after eviction
+test_11(){
+ do_facet client multiop $MOUNT/$tfile Ow || return 1
+ do_facet client multiop $MOUNT/$tfile or || return 2
-drop_request "mcreate /mnt/lustre/1"
-drop_reply "mcreate /mnt/lustre/2"
-# replay "mcreate /mnt/lustre/3"
+ cancel_lru_locks OSC
-drop_request "tchmod 111 /mnt/lustre/2"
-drop_reply "tchmod 666 /mnt/lustre/2"
-# replay "tchmod 444 /mnt/lustre/2"
+ do_facet client multiop $MOUNT/$tfile or || return 3
+ drop_bl_callback multiop $MOUNT/$tfile Ow ||
+ echo "client evicted, as expected"
-drop_request "statone /mnt/lustre/2"
-drop_reply "statone /mnt/lustre/2"
-# replay "statone /mnt/lustre/2"
+ do_facet client munlink $MOUNT/$tfile || return 4
+}
+run_test 11 "wake up a thead waiting for completion after eviction (b=2460)"
+
+#b=2494
+test_12(){
+ $LCTL mark multiop $MOUNT/$tfile OS_c
+ do_facet mds "sysctl -w lustre.fail_loc=0x115"
+ clear_failloc mds $((TIMEOUT * 2)) &
+ multiop $MOUNT/$tfile OS_c &
+ PID=$!
+#define OBD_FAIL_MDS_CLOSE_NET 0x115
+ sleep 2
+ kill -USR1 $PID
+ echo "waiting for multiop $PID"
+ wait $PID || return 2
+ do_facet client munlink $MOUNT/$tfile || return 3
+}
+run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
+
+# Bug 113, check that readdir lost recv timeout works.
+test_13() {
+ mkdir /mnt/lustre/readdir
+ touch /mnt/lustre/readdir/newentry
+# OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000104"
+ ls /mnt/lustre/readdir || return 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+ rm -rf /mnt/lustre/readdir
+}
+run_test 13 "mdc_readpage restart test (bug 1138)"
+
+# Bug 113, check that readdir lost send timeout works.
+test_14() {
+ mkdir /mnt/lustre/readdir
+ touch /mnt/lustre/readdir/newentry
+# OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000106"
+ ls /mnt/lustre/readdir || return 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+}
+run_test 14 "mdc_readpage resend test (bug 1138)"
-do_client "cp /etc/resolv.conf /mnt/lustre/resolv.conf"
-drop_request "cat /mnt/lustre/resolv.conf > /dev/null"
-drop_reply "cat /mnt/lustre/resolv.conf > /dev/null"
+test_15() {
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
+ touch $DIR/$tfile && return 1
+ return 0
+}
+run_test 15 "failed open (-ENOMEM)"
+
+test_16() {
+ do_facet client cp /etc/termcap $MOUNT
+ sync
+
+#define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
+ sysctl -w lustre.fail_loc=0x80000504
+ cancel_lru_locks OSC
+ # will get evicted here
+ do_facet client "diff /etc/termcap $MOUNT/termcap" && return 1
+ sysctl -w lustre.fail_loc=0
+ do_facet client "diff /etc/termcap $MOUNT/termcap" || return 2
+}
+run_test 16 "timeout bulk put, evict client (2732)"
-drop_request "mv /mnt/lustre/resolv.conf /mnt/lustre/renamed"
-drop_reply "mv /mnt/lustre/renamed /mnt/lustre/renamed-again"
+test_17() {
+# OBD_FAIL_PTLRPC_BULK_GET_NET | OBD_FAIL_ONCE
+ # wil get evicted here
+ sysctl -w lustre.fail_loc=0x80000503
+ do_facet client cp /etc/termcap $MOUNT && return 1
-drop_request "mlink /mnt/lustre/renamed-again /mnt/lustre/link1"
-drop_reply "mlink /mnt/lustre/renamed-again /mnt/lustre/link2"
+ do_facet client "diff /etc/termcap $MOUNT/termcap" && return 1
+ sysctl -w lustre.fail_loc=0
+ do_facet client "diff /etc/termcap $MOUNT/termcap" || return 2
-drop_request "munlink /mnt/lustre/link1"
-drop_reply "munlink /mnt/lustre/link2"
+}
+run_test 17 "timeout bulk get, evict client (2732)"
+
+test_18a() {
+ do_facet client mkdir -p $MOUNT/$tdir
+ f=$MOUNT/$tdir/$tfile
+
+ cancel_lru_locks OSC
+ pgcache_empty || return 1
+
+ # 1 stripe on ost2
+ lfs setstripe $f $((128 * 1024)) 1 1
+
+ do_facet client cp /etc/termcap $f
+ sync
+ local osc2_dev=`$LCTL device_list | \
+ awk '(/ost2.*client_facet/){print $4}' `
+ $LCTL --device %$osc2_dev deactivate
+ # my understanding is that there should be nothing in the page
+ # cache after the client reconnects?
+ rc=0
+ pgcache_empty || rc=2
+ $LCTL --device %$osc2_dev activate
+ rm -f $f
+ return $rc
+}
+run_test 18a "manual ost invalidate clears page cache immediately"
+
+test_18b() {
+# OBD_FAIL_PTLRPC_BULK_PUT_NET|OBD_FAIL_ONCE
+ do_facet client mkdir -p $MOUNT/$tdir
+ f=$MOUNT/$tdir/$tfile
+ f2=$MOUNT/$tdir/${tfile}-2
+
+ cancel_lru_locks OSC
+ pgcache_empty || return 1
+
+ # shouldn't have to set stripe size of count==1
+ lfs setstripe $f $((128 * 1024)) 0 1
+ lfs setstripe $f2 $((128 * 1024)) 0 1
+
+ do_facet client cp /etc/termcap $f
+ sync
+ # just use this write to trigger the client's eviction from the ost
+ sysctl -w lustre.fail_loc=0x80000503
+ do_facet client dd if=/dev/zero of=$f2 bs=4k count=1
+ sync
+ sysctl -w lustre.fail_loc=0
+ # allow recovery to complete
+ sleep $((TIMEOUT + 2))
+ # my understanding is that there should be nothing in the page
+ # cache after the client reconnects?
+ rc=0
+ pgcache_empty || rc=2
+ rm -f $f $f2
+ return $rc
+}
+run_test 18b "eviction and reconnect clears page cache (2766)"
+
+test_19a() { # bug 2983 - ldlm_handle_enqueue cleanup
+ mkdir -p $DIR/d19
+ multiop $DIR/d19/f19a O_wc &
+ MULTI_PID=$!
+ usleep 500
+ cancel_lru_locks OSC
+#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
+ do_facet ost sysctl -w lustre.fail_loc=0x80000308
+ kill -USR1 $MULTI_PID
+ wait $MULTI_PID && error "multiop didn't fail enqueue" || true
+}
+run_test 19a "ldlm_handle_enqueue error (should return error) ==="
+
+test_19b() { # bug 2986 - ldlm_handle_enqueue error during open
+ mkdir $DIR/d19
+ touch $DIR/d19/f19b
+ cancel_lru_locks OSC
+#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
+ do_facet ost sysctl -w lustre.fail_loc=0x80000308
+ dd if=/etc/hosts of=$DIR/d19/f19b && error "didn't fail enqueue" || true
+}
+run_test 19b "ldlm_handle_enqueue error (should return error) ==="
-#bug 1423
-drop_reply "touch /mnt/lustre/renamed"
$CLEANUP
git://git.whamcloud.com / fs / lustre-release.git / blobdiff