X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Freplay-single.sh;h=9b8be16682e7bd9233767b356a4500f2f64b264a;hp=fbf8d1986b7f172c10619a840fc3c4fdcd73a950;hb=ec9fc19d23d2ee251894634f95a71f32bbc43b20;hpb=9f59ef764e80542d82f3b78902615124ddac1020;ds=sidebyside diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh index fbf8d19..9b8be16 100755 --- a/lustre/tests/replay-single.sh +++ b/lustre/tests/replay-single.sh @@ -1,142 +1,176 @@ -#!/bin/sh +#!/bin/bash set -e +#set -v # # This test needs to be run on the client # - +SAVE_PWD=$PWD LUSTRE=${LUSTRE:-`dirname $0`/..} +SETUP=${SETUP:-} +CLEANUP=${CLEANUP:-} . $LUSTRE/tests/test-framework.sh - init_test_env $@ +. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} +CHECK_GRANT=${CHECK_GRANT:-"yes"} +GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""} -. ${CONFIG:=$LUSTRE/tests/cfg/local.sh} # Skip these tests -ALWAYS_EXCEPT="" - - -gen_config() { - rm -f $XMLCONFIG - add_mds mds --dev $MDSDEV --size $MDSSIZE - if [ ! -z "$mdsfailover_HOST" ]; then - add_mdsfailover mds --dev $MDSDEV --size $MDSSIZE - fi - - 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 -} +# bug number: 2766 4176 +ALWAYS_EXCEPT="0b 39 $REPLAY_SINGLE_EXCEPT" build_test_filter -cleanup() { - # make sure we are using the primary MDS, so the config log will - # be able to clean up properly. - activemds=`facet_active mds` - if [ $activemds != "mds" ]; then - fail mds - fi - zconf_umount `hostname` $MOUNT - stop mds ${FORCE} $MDSLCONFARGS - stop ost2 ${FORCE} --dump cleanup.log - stop ost ${FORCE} --dump cleanup.log -} - -if [ "$ONLY" == "cleanup" ]; then - sysctl -w portals.debug=0 || true - cleanup - exit -fi +cleanup_and_setup_lustre -SETUP=${SETUP:-"setup"} -CLEANUP=${CLEANUP:-"cleanup"} - -setup() { - gen_config +mkdir -p $DIR - start ost --reformat $OSTLCONFARGS - start ost2 --reformat $OSTLCONFARGS - [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE - start mds $MDSLCONFARGS --reformat - zconf_mount `hostname` $MOUNT - echo 0x3f0410 > /proc/sys/portals/debug +test_0() { + sleep 10 + mkdir $DIR/$tfile + replay_barrier $SINGLEMDS + fail $SINGLEMDS + rmdir $DIR/$tfile } +run_test 0 "empty replay" -$SETUP - -if [ "$ONLY" == "setup" ]; then - exit 0 -fi +test_0b() { + # this test attempts to trigger a race in the precreation code, + # and must run before any other objects are created on the filesystem + fail ost1 + createmany -o $DIR/$tfile 20 || return 1 + unlinkmany $DIR/$tfile 20 || return 2 +} +run_test 0b "ensure object created after recover exists. (3284)" + +seq_set_width() +{ + local mds=$1 + local width=$2 + local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width` + echo $width > $file +} -mkdir -p $DIR +seq_get_width() +{ + local mds=$1 + local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width` + cat $file +} -test_0() { - replay_barrier mds - fail mds +# This test should pass for single-mds and multi-mds configs. +# But for different configurations it tests different things. +# +# single-mds +# ---------- +# (1) fld_create replay should happen; +# +# (2) fld_create replay should not return -EEXISTS, if it does +# this means sequence manager recovery code is buggy and allocated +# same sequence two times after recovery. +# +# multi-mds +# --------- +# (1) fld_create replay may not happen, because its home MDS is +# MDS2 which is not involved to revovery; +# +# (2) as fld_create does not happen on MDS1, it does not make any +# problem. +test_0c() { + local label=`mdsdevlabel 1` + [ -z "$label" ] && echo "No label for mds1" && return 1 + + replay_barrier $SINGLEMDS + local sw=`seq_get_width $label` + + # make seq manager switch to next sequence each + # time as new fid is needed. + seq_set_width $label 1 + + # make sure that fld has created at least one new + # entry on server + touch $DIR/$tfile || return 2 + seq_set_width $label $sw + + # fail $SINGLEMDS and start recovery, replay RPCs, etc. + fail $SINGLEMDS + + # wait for recovery finish + sleep 10 + df $MOUNT + + # flush fld cache and dentry cache to make it lookup + # created entry instead of revalidating existent one + umount $MOUNT + zconf_mount `hostname` $MOUNT + + # issue lookup which should call fld lookup which + # should fail if client did not replay fld create + # correctly and server has no fld entry + touch $DIR/$tfile || return 3 + rm $DIR/$tfile || return 4 } -run_test 0 "empty replay" +run_test 0c "fld create" test_1() { - replay_barrier mds + replay_barrier $SINGLEMDS mcreate $DIR/$tfile - fail mds + fail $SINGLEMDS $CHECKSTAT -t file $DIR/$tfile || return 1 rm $DIR/$tfile } run_test 1 "simple create" test_2a() { - replay_barrier mds + replay_barrier $SINGLEMDS touch $DIR/$tfile - fail mds + fail $SINGLEMDS $CHECKSTAT -t file $DIR/$tfile || return 1 rm $DIR/$tfile } run_test 2a "touch" test_2b() { - ./mcreate $DIR/$tfile - replay_barrier mds + mcreate $DIR/$tfile + replay_barrier $SINGLEMDS touch $DIR/$tfile - fail mds + fail $SINGLEMDS $CHECKSTAT -t file $DIR/$tfile || return 1 rm $DIR/$tfile } run_test 2b "touch" test_3a() { - replay_barrier mds + replay_barrier $SINGLEMDS mcreate $DIR/$tfile o_directory $DIR/$tfile - fail mds + fail $SINGLEMDS $CHECKSTAT -t file $DIR/$tfile || return 2 rm $DIR/$tfile } run_test 3a "replay failed open(O_DIRECTORY)" test_3b() { - replay_barrier mds + replay_barrier $SINGLEMDS #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE - do_facet mds "sysctl -w lustre.fail_loc=0x80000114" + do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000114" touch $DIR/$tfile - do_facet mds "sysctl -w lustre.fail_loc=0" - fail mds + do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + fail $SINGLEMDS $CHECKSTAT -t file $DIR/$tfile && return 2 return 0 } run_test 3b "replay failed open -ENOMEM" test_3c() { - replay_barrier mds + replay_barrier $SINGLEMDS #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE - do_facet mds "sysctl -w lustre.fail_loc=0x80000128" + do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000128" touch $DIR/$tfile - do_facet mds "sysctl -w lustre.fail_loc=0" - fail mds + do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0" + fail $SINGLEMDS $CHECKSTAT -t file $DIR/$tfile && return 2 return 0 @@ -144,11 +178,11 @@ test_3c() { run_test 3c "replay failed open -ENOMEM" test_4() { - replay_barrier mds + replay_barrier $SINGLEMDS for i in `seq 10`; do echo "tag-$i" > $DIR/$tfile-$i done - fail mds + fail $SINGLEMDS for i in `seq 10`; do grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i" done @@ -156,9 +190,9 @@ test_4() { run_test 4 "|x| 10 open(O_CREAT)s" test_4b() { - replay_barrier mds + replay_barrier $SINGLEMDS rm -rf $DIR/$tfile-* - fail mds + fail $SINGLEMDS $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true } run_test 4b "|x| rm 10 files" @@ -166,11 +200,11 @@ run_test 4b "|x| rm 10 files" # The idea is to get past the first block of precreated files on both # osts, and then replay. test_5() { - replay_barrier mds + replay_barrier $SINGLEMDS for i in `seq 220`; do echo "tag-$i" > $DIR/$tfile-$i done - fail mds + fail $SINGLEMDS for i in `seq 220`; do grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i" done @@ -182,10 +216,10 @@ run_test 5 "|x| 220 open(O_CREAT)" test_6() { - replay_barrier mds + replay_barrier $SINGLEMDS mkdir $DIR/$tdir mcreate $DIR/$tdir/$tfile - fail mds + fail $SINGLEMDS $CHECKSTAT -t dir $DIR/$tdir || return 1 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2 sleep 2 @@ -194,18 +228,18 @@ test_6() { run_test 6 "mkdir + contained create" test_6b() { - replay_barrier mds + replay_barrier $SINGLEMDS rm -rf $DIR/$tdir - fail mds + fail $SINGLEMDS $CHECKSTAT -t dir $DIR/$tdir && return 1 || true } run_test 6b "|X| rmdir" test_7() { mkdir $DIR/$tdir - replay_barrier mds + replay_barrier $SINGLEMDS mcreate $DIR/$tdir/$tfile - fail mds + fail $SINGLEMDS $CHECKSTAT -t dir $DIR/$tdir || return 1 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2 rm -fr $DIR/$tdir @@ -213,11 +247,13 @@ test_7() { run_test 7 "mkdir |X| contained create" test_8() { - replay_barrier mds + # make sure no side-effect from previous test. + rm -f $DIR/$tfile + replay_barrier $SINGLEMDS multiop $DIR/$tfile mo_c & MULTIPID=$! sleep 1 - fail mds + fail $SINGLEMDS ls $DIR/$tfile $CHECKSTAT -t file $DIR/$tfile || return 1 kill -USR1 $MULTIPID || return 2 @@ -227,10 +263,10 @@ test_8() { run_test 8 "creat open |X| close" test_9() { - replay_barrier mds + replay_barrier $SINGLEMDS mcreate $DIR/$tfile local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'` - fail mds + fail $SINGLEMDS local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'` echo " old_inum == $old_inum, new_inum == $new_inum" @@ -247,10 +283,10 @@ run_test 9 "|X| create (same inum/gen)" test_10() { mcreate $DIR/$tfile - replay_barrier mds + replay_barrier $SINGLEMDS mv $DIR/$tfile $DIR/$tfile-2 rm -f $DIR/$tfile - fail mds + fail $SINGLEMDS $CHECKSTAT $DIR/$tfile && return 1 $CHECKSTAT $DIR/$tfile-2 ||return 2 rm $DIR/$tfile-2 @@ -262,11 +298,11 @@ test_11() { mcreate $DIR/$tfile echo "old" > $DIR/$tfile mv $DIR/$tfile $DIR/$tfile-2 - replay_barrier mds + replay_barrier $SINGLEMDS echo "new" > $DIR/$tfile grep new $DIR/$tfile grep old $DIR/$tfile-2 - fail mds + fail $SINGLEMDS grep new $DIR/$tfile || return 1 grep old $DIR/$tfile-2 || return 2 } @@ -279,11 +315,11 @@ test_12() { # give multiop a chance to open sleep 1 rm -f $DIR/$tfile - replay_barrier mds + replay_barrier $SINGLEMDS kill -USR1 $pid wait $pid || return 1 - fail mds + fail $SINGLEMDS [ -e $DIR/$tfile ] && return 2 return 0 } @@ -300,8 +336,8 @@ test_13() { sleep 1 chmod 0 $DIR/$tfile $CHECKSTAT -p 0 $DIR/$tfile - replay_barrier mds - fail mds + replay_barrier $SINGLEMDS + fail $SINGLEMDS kill -USR1 $pid wait $pid || return 1 @@ -316,11 +352,11 @@ test_14() { # give multiop a chance to open sleep 1 rm -f $DIR/$tfile - replay_barrier mds + replay_barrier $SINGLEMDS kill -USR1 $pid || return 1 wait $pid || return 2 - fail mds + fail $SINGLEMDS [ -e $DIR/$tfile ] && return 3 return 0 } @@ -332,12 +368,12 @@ test_15() { # give multiop a chance to open sleep 1 rm -f $DIR/$tfile - replay_barrier mds + replay_barrier $SINGLEMDS touch $DIR/g11 || return 1 kill -USR1 $pid wait $pid || return 2 - fail mds + fail $SINGLEMDS [ -e $DIR/$tfile ] && return 3 touch $DIR/h11 || return 4 return 0 @@ -346,11 +382,11 @@ run_test 15 "open(O_CREAT), unlink |X| touch new, close" test_16() { - replay_barrier mds + replay_barrier $SINGLEMDS mcreate $DIR/$tfile munlink $DIR/$tfile mcreate $DIR/$tfile-2 - fail mds + fail $SINGLEMDS [ -e $DIR/$tfile ] && return 1 [ -e $DIR/$tfile-2 ] || return 2 munlink $DIR/$tfile-2 || return 3 @@ -358,12 +394,12 @@ test_16() { run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new" test_17() { - replay_barrier mds + replay_barrier $SINGLEMDS multiop $DIR/$tfile O_c & pid=$! # give multiop a chance to open sleep 1 - fail mds + fail $SINGLEMDS kill -USR1 $pid || return 1 wait $pid || return 2 $CHECKSTAT -t file $DIR/$tfile || return 3 @@ -372,7 +408,7 @@ test_17() { run_test 17 "|X| open(O_CREAT), |replay| close" test_18() { - replay_barrier mds + replay_barrier $SINGLEMDS multiop $DIR/$tfile O_tSc & pid=$! # give multiop a chance to open @@ -383,7 +419,7 @@ test_18() { kill -USR1 $pid wait $pid || return 2 - fail mds + fail $SINGLEMDS [ -e $DIR/$tfile ] && return 3 [ -e $DIR/$tfile-2 ] || return 4 # this touch frequently fails @@ -396,25 +432,25 @@ run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink" # bug 1855 (a simpler form of test_11 above) test_19() { - replay_barrier mds + replay_barrier $SINGLEMDS mcreate $DIR/$tfile echo "old" > $DIR/$tfile mv $DIR/$tfile $DIR/$tfile-2 grep old $DIR/$tfile-2 - fail mds + fail $SINGLEMDS grep old $DIR/$tfile-2 || return 2 } run_test 19 "|X| mcreate, open, write, rename " test_20() { - replay_barrier mds + replay_barrier $SINGLEMDS multiop $DIR/$tfile O_tSc & pid=$! # give multiop a chance to open sleep 1 rm -f $DIR/$tfile - fail mds + fail $SINGLEMDS kill -USR1 $pid wait $pid || return 1 [ -e $DIR/$tfile ] && return 2 @@ -422,8 +458,53 @@ test_20() { } run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)" +test_20b() { # bug 10480 + BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'` + + dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 & + pid=$! + while [ ! -e $DIR/$tfile ] ; do + usleep 60 # give dd a chance to start + done + + lfs getstripe $DIR/$tfile || return 1 + rm -f $DIR/$tfile || return 2 # make it an orphan + mds_evict_client + df -P $DIR || df -P $DIR || true # reconnect + + fail $SINGLEMDS # start orphan recovery + df -P $DIR || df -P $DIR || true # reconnect + wait_mds_recovery_done || error "MDS recovery not done" + + AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'` + log "before $BEFOREUSED, after $AFTERUSED" + [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \ + error "after $AFTERUSED > before $BEFOREUSED" && return 5 + return 0 +} +run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)" + +test_20c() { # bug 10480 + multiop $DIR/$tfile Ow_c & + pid=$! + # give multiop a chance to open + sleep 1 + + ls -la $DIR/$tfile + + mds_evict_client + + df -P $DIR || df -P $DIR || true # reconnect + + kill -USR1 $pid + test -s $DIR/$tfile || error "File was truncated" + + return 0 +} +run_test 20c "check that client eviction does not affect file content" + test_21() { - replay_barrier mds + replay_barrier $SINGLEMDS multiop $DIR/$tfile O_tSc & pid=$! # give multiop a chance to open @@ -431,7 +512,7 @@ test_21() { rm -f $DIR/$tfile touch $DIR/g11 || return 1 - fail mds + fail $SINGLEMDS kill -USR1 $pid wait $pid || return 2 [ -e $DIR/$tfile ] && return 3 @@ -446,10 +527,10 @@ test_22() { # give multiop a chance to open sleep 1 - replay_barrier mds + replay_barrier $SINGLEMDS rm -f $DIR/$tfile - fail mds + fail $SINGLEMDS kill -USR1 $pid wait $pid || return 1 [ -e $DIR/$tfile ] && return 2 @@ -463,11 +544,11 @@ test_23() { # give multiop a chance to open sleep 1 - replay_barrier mds + replay_barrier $SINGLEMDS rm -f $DIR/$tfile touch $DIR/g11 || return 1 - fail mds + fail $SINGLEMDS kill -USR1 $pid wait $pid || return 2 [ -e $DIR/$tfile ] && return 3 @@ -482,8 +563,8 @@ test_24() { # give multiop a chance to open sleep 1 - replay_barrier mds - fail mds + replay_barrier $SINGLEMDS + fail $SINGLEMDS rm -f $DIR/$tfile kill -USR1 $pid wait $pid || return 1 @@ -499,8 +580,8 @@ test_25() { sleep 1 rm -f $DIR/$tfile - replay_barrier mds - fail mds + replay_barrier $SINGLEMDS + fail $SINGLEMDS kill -USR1 $pid wait $pid || return 1 [ -e $DIR/$tfile ] && return 2 @@ -509,7 +590,7 @@ test_25() { run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)" test_26() { - replay_barrier mds + replay_barrier $SINGLEMDS multiop $DIR/$tfile-1 O_tSc & pid1=$! multiop $DIR/$tfile-2 O_tSc & @@ -521,7 +602,7 @@ test_26() { kill -USR1 $pid2 wait $pid2 || return 1 - fail mds + fail $SINGLEMDS kill -USR1 $pid1 wait $pid1 || return 2 [ -e $DIR/$tfile-1 ] && return 3 @@ -531,7 +612,7 @@ test_26() { run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)" test_27() { - replay_barrier mds + replay_barrier $SINGLEMDS multiop $DIR/$tfile-1 O_tSc & pid1=$! multiop $DIR/$tfile-2 O_tSc & @@ -541,7 +622,7 @@ test_27() { rm -f $DIR/$tfile-1 rm -f $DIR/$tfile-2 - fail mds + fail $SINGLEMDS kill -USR1 $pid1 wait $pid1 || return 1 kill -USR1 $pid2 @@ -559,13 +640,13 @@ test_28() { pid2=$! # give multiop a chance to open sleep 1 - replay_barrier mds + replay_barrier $SINGLEMDS rm -f $DIR/$tfile-1 rm -f $DIR/$tfile-2 kill -USR1 $pid2 wait $pid2 || return 1 - fail mds + fail $SINGLEMDS kill -USR1 $pid1 wait $pid1 || return 2 [ -e $DIR/$tfile-1 ] && return 3 @@ -581,11 +662,11 @@ test_29() { pid2=$! # give multiop a chance to open sleep 1 - replay_barrier mds + replay_barrier $SINGLEMDS rm -f $DIR/$tfile-1 rm -f $DIR/$tfile-2 - fail mds + fail $SINGLEMDS kill -USR1 $pid1 wait $pid1 || return 1 kill -USR1 $pid2 @@ -606,8 +687,8 @@ test_30() { rm -f $DIR/$tfile-1 rm -f $DIR/$tfile-2 - replay_barrier mds - fail mds + replay_barrier $SINGLEMDS + fail $SINGLEMDS kill -USR1 $pid1 wait $pid1 || return 1 kill -USR1 $pid2 @@ -627,9 +708,9 @@ test_31() { sleep 1 rm -f $DIR/$tfile-1 - replay_barrier mds + replay_barrier $SINGLEMDS rm -f $DIR/$tfile-2 - fail mds + fail $SINGLEMDS kill -USR1 $pid1 wait $pid1 || return 1 kill -USR1 $pid2 @@ -650,7 +731,7 @@ test_32() { # give multiop a chance to open sleep 1 mds_evict_client - df $MOUNT || df $MOUNT || return 1 + df $MOUNT || sleep 1 && df $MOUNT || return 1 kill -USR1 $pid1 kill -USR1 $pid2 sleep 1 @@ -660,15 +741,29 @@ run_test 32 "close() notices client eviction; close() after client eviction" # Abort recovery before client complete test_33() { - replay_barrier mds - touch $DIR/$tfile - fail_abort mds + replay_barrier $SINGLEMDS + createmany -o $DIR/$tfile-%d 100 + fail_abort $SINGLEMDS # this file should be gone, because the replay was aborted - $CHECKSTAT -t file $DIR/$tfile && return 1 + $CHECKSTAT -t file $DIR/$tfile-* && return 3 + unlinkmany $DIR/$tfile-%d 0 100 return 0 } run_test 33 "abort recovery before client does replay" +# Stale FID sequence +test_33a() { + replay_barrier $SINGLEMDS + createmany -o $DIR/$tfile-%d 10 + fail_abort $SINGLEMDS + unlinkmany $DIR/$tfile-%d 0 10 + # recreate shouldn't fail + createmany -o $DIR/$tfile-%d 10 || return 3 + unlinkmany $DIR/$tfile-%d 0 10 + return 0 +} +run_test 33a "fid shouldn't be reused after abort recovery" + test_34() { multiop $DIR/$tfile O_c & pid=$! @@ -676,8 +771,8 @@ test_34() { sleep 1 rm -f $DIR/$tfile - replay_barrier mds - fail_abort mds + replay_barrier $SINGLEMDS + fail_abort $SINGLEMDS kill -USR1 $pid [ -e $DIR/$tfile ] && return 1 sync @@ -690,13 +785,13 @@ test_35() { touch $DIR/$tfile #define OBD_FAIL_MDS_REINT_NET_REP 0x119 - do_facet mds "sysctl -w lustre.fail_loc=0x80000119" + do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119" rm -f $DIR/$tfile & sleep 1 sync sleep 1 # give a chance to remove from MDS - fail_abort mds + fail_abort $SINGLEMDS $CHECKSTAT -t file $DIR/$tfile && return 1 || true } run_test 35 "test recovery from llog for unlink op" @@ -704,11 +799,11 @@ run_test 35 "test recovery from llog for unlink op" # b=2432 resent cancel after replay uses wrong cookie, # so don't resend cancels test_36() { - replay_barrier mds + replay_barrier $SINGLEMDS touch $DIR/$tfile checkstat $DIR/$tfile - facet_failover mds - cancel_lru_locks MDC + facet_failover $SINGLEMDS + cancel_lru_locks mdc if dmesg | grep "unknown lock cookie"; then echo "cancel after replay failed" return 1 @@ -726,10 +821,10 @@ test_37() { sleep 1 rmdir $DIR/$tfile - replay_barrier mds + replay_barrier $SINGLEMDS # clear the dmesg buffer so we only see errors from this recovery dmesg -c >/dev/null - fail_abort mds + fail_abort $SINGLEMDS kill -USR1 $pid dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1 sync @@ -740,19 +835,19 @@ run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans test_38() { createmany -o $DIR/$tfile-%d 800 unlinkmany $DIR/$tfile-%d 0 400 - replay_barrier mds - fail mds + 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() { +test_39() { # bug 4176 createmany -o $DIR/$tfile-%d 800 - replay_barrier mds + replay_barrier $SINGLEMDS unlinkmany $DIR/$tfile-%d 0 400 - fail mds + fail $SINGLEMDS unlinkmany $DIR/$tfile-%d 400 400 sleep 2 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true @@ -760,8 +855,7 @@ test_39() { run_test 39 "test recovery from unlink llog (test llog_gen_rec) " count_ost_writes() { - cat /proc/fs/lustre/osc/*/stats | - awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' + awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats } #b=2477,2532 @@ -772,9 +866,9 @@ test_40(){ writeme -s $MOUNT/${tfile}-2 & WRITE_PID=$! sleep 1 - facet_failover mds + facet_failover $SINGLEMDS #define OBD_FAIL_MDS_CONNECT_NET 0x117 - do_facet mds "sysctl -w lustre.fail_loc=0x80000117" + do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117" kill -USR1 $PID stat1=`count_ost_writes` sleep $TIMEOUT @@ -808,68 +902,92 @@ run_test 40 "cause recovery in ptlrpc, ensure IO continues" # 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 + cancel_lru_locks osc # fail ost2 and read from ost1 - local osc2_dev=`$LCTL device_list | \ - awk '(/ost2.*client_facet/){print $4}' ` - $LCTL --device %$osc2_dev deactivate + 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 - $LCTL --device %$osc2_dev activate + 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 $MOUNT | tail -1 | awk '{ print $1 }'` + blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'` createmany -o $DIR/$tfile-%d 800 - replay_barrier ost + replay_barrier ost1 unlinkmany $DIR/$tfile-%d 0 400 - facet_failover ost + debugsave + sysctl -w lnet.debug=-1 + facet_failover ost1 - # osc is evicted, fs is smaller - blocks_after=`df $MOUNT | tail -1 | awk '{ print $1 }'` - [ $blocks_after -lt $blocks ] || return 1 + # 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" -# b=2530 -# directory orphans can't be unlinked from PENDING directory -test_43() { - replay_barrier mds +# timeout in MDS/OST recovery RPC will LBUG MDS +test_43() { # bug 2530 + replay_barrier $SINGLEMDS # OBD_FAIL_OST_CREATE_NET 0x204 - do_facet ost "sysctl -w lustre.fail_loc=0x80000204" - facet_failover mds - df $MOUNT || return 1 + do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204" + fail $SINGLEMDS sleep 10 - do_facet ost "sysctl -w lustre.fail_loc=0" + 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 '/mds_svc_MNT/ {print $1}' < /proc/fs/lustre/devices` - do_facet mds "sysctl -w lustre.fail_loc=0x80000701" - $LCTL --device $mdcdev recover - df $MOUNT - do_facet mds "sysctl -w lustre.fail_loc=0" + 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 '/mds_svc_MNT/ {print $1}' < /proc/fs/lustre/devices` + mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices` + [ "$mdcdev" ] || exit 2 $LCTL --device $mdcdev recover multiop $DIR/$tfile O_c & @@ -878,19 +996,183 @@ test_45() { # This will cause the CLOSE to fail before even # allocating a reply buffer - $LCTL --device $mdcdev deactivate + $LCTL --device $mdcdev deactivate || return 4 # try the close kill -USR1 $pid wait $pid || return 1 - $LCTL --device $mdcdev activate + $LCTL --device $mdcdev activate || return 5 + sleep 1 $CHECKSTAT -t file $DIR/$tfile || return 2 return 0 } run_test 45 "Handle failed close" -equals_msg test complete, cleaning up -$CLEANUP +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=`do_facet $SINGLEMDS grep ${ost1_svc}-osc-MDT0000 $LPROC/devices | awk '{print $1}'` + [ "$oscdev" ] || return 1 + do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2 + do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3 + # 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