+test_38() {
+ createmany -o $DIR/$tfile-%d 800
+ unlinkmany $DIR/$tfile-%d 0 400
+ replay_barrier $SINGLEMDS
+ fail $SINGLEMDS
+ unlinkmany $DIR/$tfile-%d 400 400
+ sleep 2
+ $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
+}
+run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
+
+test_39() { # bug 4176
+ createmany -o $DIR/$tfile-%d 800
+ replay_barrier $SINGLEMDS
+ unlinkmany $DIR/$tfile-%d 0 400
+ fail $SINGLEMDS
+ unlinkmany $DIR/$tfile-%d 400 400
+ sleep 2
+ $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
+}
+run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
+
+count_ost_writes() {
+ awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
+}
+
+#b=2477,2532
+test_40(){
+ $LCTL mark multiop $MOUNT/$tfile OS_c
+ multiop $MOUNT/$tfile OS_c &
+ PID=$!
+ writeme -s $MOUNT/${tfile}-2 &
+ WRITE_PID=$!
+ sleep 1
+ facet_failover $SINGLEMDS
+#define OBD_FAIL_MDS_CONNECT_NET 0x117
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117"
+ kill -USR1 $PID
+ stat1=`count_ost_writes`
+ sleep $TIMEOUT
+ stat2=`count_ost_writes`
+ echo "$stat1, $stat2"
+ if [ $stat1 -lt $stat2 ]; then
+ echo "writes continuing during recovery"
+ RC=0
+ else
+ echo "writes not continuing during recovery, bug 2477"
+ RC=4
+ fi
+ echo "waiting for writeme $WRITE_PID"
+ kill $WRITE_PID
+ wait $WRITE_PID
+
+ echo "waiting for multiop $PID"
+ wait $PID || return 2
+ do_facet client munlink $MOUNT/$tfile || return 3
+ do_facet client munlink $MOUNT/${tfile}-2 || return 3
+ return $RC
+}
+run_test 40 "cause recovery in ptlrpc, ensure IO continues"
+
+
+#b=2814
+# make sure that a read to one osc doesn't try to double-unlock its page just
+# because another osc is invalid. trigger_group_io used to mistakenly return
+# an error if any oscs were invalid even after having successfully put rpcs
+# on valid oscs. This was fatal if the caller was ll_readpage who unlocked
+# the page, guarnateeing that the unlock from the RPC completion would
+# assert on trying to unlock the unlocked page.
+test_41() {
+ [ $OSTCOUNT -lt 2 ] && \
+ skip "skipping test 41: we don't have a second OST to test with" && \
+ return
+
+ local f=$MOUNT/$tfile
+ # make sure the start of the file is ost1
+ lfs setstripe $f $((128 * 1024)) 0 0
+ do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
+ cancel_lru_locks osc
+ # fail ost2 and read from ost1
+ local osc2dev=`do_facet mds "grep ${ost2_svc}-osc-MDT0000 $LPROC/devices" | awk '{print $1}'`
+ [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && cat $LPROC/devices && return 4
+ do_facet mds $LCTL --device $osc2dev deactivate || return 1
+ do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
+ do_facet mds $LCTL --device $osc2dev activate || return 2
+ return 0
+}
+run_test 41 "read from a valid osc while other oscs are invalid"
+
+# test MDS recovery after ost failure
+test_42() {
+ blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
+ createmany -o $DIR/$tfile-%d 800
+ replay_barrier ost1
+ unlinkmany $DIR/$tfile-%d 0 400
+ debugsave
+ sysctl -w lnet.debug=-1
+ facet_failover ost1
+
+ # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
+ #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
+ #[ $blocks_after -lt $blocks ] || return 1
+ echo wait for MDS to timeout and recover
+ sleep $((TIMEOUT * 2))
+ debugrestore
+ unlinkmany $DIR/$tfile-%d 400 400
+ $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
+}
+run_test 42 "recovery after ost failure"
+
+# timeout in MDS/OST recovery RPC will LBUG MDS
+test_43() { # bug 2530
+ replay_barrier $SINGLEMDS
+
+ # OBD_FAIL_OST_CREATE_NET 0x204
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
+ fail $SINGLEMDS
+ sleep 10
+ do_facet ost1 "sysctl -w lustre.fail_loc=0"
+
+ return 0
+}
+run_test 43 "mds osc import failure during recovery; don't LBUG"
+
+test_44() {
+ mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
+ [ "$mdcdev" ] || exit 2
+ for i in `seq 1 10`; do
+ #define OBD_FAIL_TGT_CONN_RACE 0x701
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000701"
+ $LCTL --device $mdcdev recover
+ df $MOUNT
+ done
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+ return 0
+}
+run_test 44 "race in target handle connect"
+
+test_44b() {
+ mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
+ [ "$mdcdev" ] || exit 2
+ for i in `seq 1 10`; do
+ #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000704"
+ $LCTL --device $mdcdev recover
+ df $MOUNT
+ done
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
+ return 0
+}
+run_test 44b "race in target handle connect"
+
+# Handle failed close
+test_45() {
+ mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
+ [ "$mdcdev" ] || exit 2
+ $LCTL --device $mdcdev recover
+
+ multiop $DIR/$tfile O_c &
+ pid=$!
+ sleep 1
+
+ # This will cause the CLOSE to fail before even
+ # allocating a reply buffer
+ $LCTL --device $mdcdev deactivate || return 4
+
+ # try the close
+ kill -USR1 $pid
+ wait $pid || return 1
+
+ $LCTL --device $mdcdev activate || return 5
+ sleep 1
+
+ $CHECKSTAT -t file $DIR/$tfile || return 2
+ return 0
+}
+run_test 45 "Handle failed close"
+
+test_46() {
+ dmesg -c >/dev/null
+ drop_reply "touch $DIR/$tfile"
+ fail $SINGLEMDS
+ # ironically, the previous test, 45, will cause a real forced close,
+ # so just look for one for this test
+ dmesg | grep -i "force closing client file handle for $tfile" && return 1
+ return 0
+}
+run_test 46 "Don't leak file handle after open resend (3325)"
+
+test_47() { # bug 2824
+ # create some files to make sure precreate has been done on all
+ # OSTs. (just in case this test is run independently)
+ createmany -o $DIR/$tfile 20 || return 1
+
+ # OBD_FAIL_OST_CREATE_NET 0x204
+ fail ost1
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
+ df $MOUNT || return 2
+
+ # let the MDS discover the OST failure, attempt to recover, fail
+ # and recover again.
+ sleep $((3 * TIMEOUT))
+
+ # Without 2824, this createmany would hang
+ createmany -o $DIR/$tfile 20 || return 3
+ unlinkmany $DIR/$tfile 20 || return 4
+
+ do_facet ost1 "sysctl -w lustre.fail_loc=0"
+ return 0
+}
+run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
+
+test_48() {
+ replay_barrier $SINGLEMDS
+ createmany -o $DIR/$tfile 20 || return 1
+ # OBD_FAIL_OST_EROFS 0x216
+ fail $SINGLEMDS
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
+ df $MOUNT || return 2
+
+ createmany -o $DIR/$tfile 20 20 || return 2
+ unlinkmany $DIR/$tfile 40 || return 3
+
+ do_facet ost1 "sysctl -w lustre.fail_loc=0"
+ return 0
+}
+run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
+
+test_50() {
+ local oscdev=`grep ${ost1_svc}-osc-MDT0000 $LPROC/devices | awk '{print $1}'`
+ [ "$oscdev" ] || return 1
+ $LCTL --device $oscdev recover && $LCTL --device $oscdev recover
+ # give the mds_lov_sync threads a chance to run
+ sleep 5
+}
+run_test 50 "Double OSC recovery, don't LASSERT (3812)"
+
+# b3764 timed out lock replay
+test_52() {
+ touch $DIR/$tfile
+ cancel_lru_locks mdc
+
+ multiop $DIR/$tfile s || return 1
+ replay_barrier $SINGLEMDS
+#define OBD_FAIL_LDLM_REPLY 0x30c
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
+ fail $SINGLEMDS || return 2
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
+
+ $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
+}
+run_test 52 "time out lock replay (3764)"
+
+#b_cray 53 "|X| open request and close reply while two MDC requests in flight"
+#b_cray 54 "|X| open request and close reply while two MDC requests in flight"
+
+#b3761 ASSERTION(hash != 0) failed
+test_55() {
+# OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
+ touch $DIR/$tfile &
+ # give touch a chance to run
+ sleep 5
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
+ rm $DIR/$tfile
+ return 0
+}
+run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
+
+#b3440 ASSERTION(rec->ur_fid2->id) failed
+test_56() {
+ ln -s foo $DIR/$tfile
+ replay_barrier $SINGLEMDS
+ #drop_reply "cat $DIR/$tfile"
+ fail $SINGLEMDS
+ sleep 10
+}
+run_test 56 "don't replay a symlink open request (3440)"
+
+#recovery one mds-ost setattr from llog
+test_57() {
+#define OBD_FAIL_MDS_OST_SETATTR 0x12c
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
+ touch $DIR/$tfile
+ replay_barrier $SINGLEMDS
+ fail $SINGLEMDS
+ sleep 1
+ $CHECKSTAT -t file $DIR/$tfile || return 1
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
+ rm $DIR/$tfile
+}
+run_test 57 "test recovery from llog for setattr op"
+
+#recovery many mds-ost setattr from llog
+test_58() {
+#define OBD_FAIL_MDS_OST_SETATTR 0x12c
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
+ mkdir $DIR/$tdir
+ createmany -o $DIR/$tdir/$tfile-%d 2500
+ replay_barrier $SINGLEMDS
+ fail $SINGLEMDS
+ sleep 2
+ $CHECKSTAT -t file $DIR/$tdir/$tfile-* || return 1
+ do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
+ unlinkmany $DIR/$tdir/$tfile-%d 2500
+ rmdir $DIR/$tdir
+}
+run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
+
+# log_commit_thread vs filter_destroy race used to lead to import use after free
+# bug 11658
+test_59() {
+ mkdir $DIR/$tdir
+ createmany -o $DIR/$tdir/$tfile-%d 200
+ sync
+ unlinkmany $DIR/$tdir/$tfile-%d 200
+#define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x507"
+ fail ost1
+ fail $SINGLEMDS
+ do_facet ost1 "sysctl -w lustre.fail_loc=0x0"
+ sleep 20
+ rmdir $DIR/$tdir
+}
+run_test 59 "test log_commit_thread vs filter_destroy race"
+
+# race between add unlink llog vs cat log init in post_recovery (only for b1_6)
+# bug 12086: should no oops and No ctxt error for this test
+test_60() {
+ mkdir $DIR/$tdir
+ createmany -o $DIR/$tdir/$tfile-%d 200
+ replay_barrier $SINGLEMDS
+ unlinkmany $DIR/$tdir/$tfile-%d 0 100
+ fail $SINGLEMDS
+ unlinkmany $DIR/$tdir/$tfile-%d 100 100
+ local no_ctxt=`dmesg | grep "No ctxt"`
+ [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
+}
+run_test 60 "test llog post recovery init vs llog unlink"
+
+equals_msg `basename $0`: test complete, cleaning up
+check_and_cleanup_lustre
+[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true