sanityn.sh

lustre 1.6.5 source code

test_14d() { # bug 10921
	mkdir -p $DIR1/d14
	cp -p `which multiop` $DIR1/d14/multiop || error "cp failed"
	MULTIOP_PROG=$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 $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 14d "chmod of executing file is still possible ========"

test_15() {	# bug 974 - ENOSPC
	echo "PATH=$PATH"
	sh oos2.sh $MOUNT1 $MOUNT2
	grant_error=`dmesg | grep "> available"`
	[ -z "$grant_error" ] || error "$grant_error"
}
run_test 15 "test out-of-space with multiple writers ==========="

test_16() {
	rm -f $MOUNT1/fsxfile
	lfs setstripe $MOUNT1/fsxfile -c -1 # b=10919
	fsx -c 50 -p 100 -N 2500 -l $((SIZE * 256)) -S 0 $MOUNT1/fsxfile $MOUNT2/fsxfile
}
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

	cp /etc/termcap $DIR1/f17
	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
	wait
	diff -u $DIR1/f17-1 $DIR2/f17-2 || error "files are different"
}
run_test 17 "resource creation/LVB creation race ==============="

test_18() {
	./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 > $O/readcache_max_filesize
	done
	dd if=/dev/urandom of=$TMP/f19b bs=512k count=32
	SUM=`cksum $TMP/f19b | cut -d" " -f 1,2`
	cp $TMP/f19b $DIR1/f19b
	for i in `seq 1 20`; do
		[ $((i % 5)) -eq 0 ] && log "test_18 loop $i"
		cancel_lru_locks osc > /dev/null
		cksum $DIR1/f19b | cut -d" " -f 1,2 > $TMP/sum1 & \
		cksum $DIR2/f19b | cut -d" " -f 1,2 > $TMP/sum2
		wait
		[ "`cat $TMP/sum1`" = "$SUM" ] || \
			error "$DIR1/f19b `cat $TMP/sum1` != $SUM"
		[ "`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
	rm $DIR1/f19b
}
#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`))
	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 -gt 0 ] && \
	    error $CNTD" page left in cache after lock cancel" || true
}
run_test 20 "test extra readahead page left in cache ===="

cleanup_21() {
	trap 0
	umount $DIR1/d21
}

test_21() { # Bug 5907
	mkdir $DIR1/d21
	mount /etc $DIR1/d21 --bind || error "mount failed" # Poor man's mount.
	trap cleanup_21 EXIT
	rmdir -v $DIR1/d21 && error "Removed mounted directory"
	rmdir -v $DIR2/d21 && echo "Removed mounted directory from another mountpoint, needs to be fixed"
	test -d $DIR1/d21 || error "Mounted directory disappeared"
	cleanup_21
	test -d $DIR2/d21 || test -d $DIR1/d21 && error "Removed dir still visible after umount"
	true
}
run_test 21 " Try to remove mountpoint on another dir ===="

JOIN=${JOIN:-"lfs join"}

test_22() { # Bug 9926
	mkdir $DIR1/d21
	dd if=/dev/urandom of=$DIR1/d21/128k bs=1024 count=128
	cp -p $DIR1/d21/128k $DIR1/d21/f_head
	for ((i=0;i<10;i++)); do
		cp -p $DIR1/d21/128k $DIR1/d21/f_tail
		$JOIN $DIR1/d21/f_head $DIR1/d21/f_tail || error "join error"
		$CHECKSTAT -a $DIR1/d21/f_tail || error "tail file exist after join"
	done
	echo aaaaaaaaaaa >> $DIR1/d21/no_joined

	mv $DIR2/d21/f_head $DIR2/
	munlink $DIR2/f_head || error "unlink joined file error"
	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" 

test_23() { # Bug 5972
	echo "others should see updated atime while another read" > $DIR1/f23
	
	# clear the lock(mode: LCK_PW) gotten from creating operation
	cancel_lru_locks osc
	
	time1=`date +%s`	
	sleep 2
	
	multiop_bg_pause $DIR1/f23 or20_c || return 1
	MULTIPID=$!

	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
}
run_test 23 " others should see updated atime while another read===="

test_24() {
	touch $DIR1/$tfile
	lfs df || error "lfs df failed"
	lfs df -ih || error "lfs df -ih failed"
	lfs df -h $DIR1 || error "lfs df -h $DIR1 failed"
	lfs df -i $DIR2 || error "lfs df -i $DIR2 failed"
	lfs df $DIR1/$tfile || error "lfs df $DIR1/$tfile failed"
	lfs df -ih $DIR2/$tfile || error "lfs df -ih $DIR2/$tfile failed"
	
	OSC=`lctl dl | awk '/-osc-|OSC.*MNT/ {print $4}' | head -n 1`
	lctl --device %$OSC deactivate
	lfs df -i || error "lfs df -i with deactivated OSC failed"
	lctl --device %$OSC recover
	lfs df || error "lfs df with reactivated OSC failed"
}
run_test 24 "lfs df [-ih] [path] test ========================="

test_25() {
	[ `lctl get_param -n mdc.*-mdc-*.connect_flags | grep -c acl` -lt 2 ] && \
	    skip "must have acl, skipping" && return

	touch $DIR1/$tdir/f1 || error "touch $DIR1/$tdir/f1"
	chmod 0755 $DIR1/$tdir/f1 || error "chmod 0755 $DIR1/$tdir/f1"

	$RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #1"
	setfacl -m u:$RUNAS_ID:--- $DIR1/$tdir || error "setfacl $DIR2/$tdir #1"
	$RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #2"
	setfacl -m u:$RUNAS_ID:r-x $DIR1/$tdir || error "setfacl $DIR2/$tdir #2"
	$RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #3"
	setfacl -m u:$RUNAS_ID:--- $DIR1/$tdir || error "setfacl $DIR2/$tdir #3"
	$RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #4"
	setfacl -x u:$RUNAS_ID: $DIR1/$tdir || error "setfacl $DIR2/$tdir #4"
	$RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #5"

	rm -rf $DIR1/$tdir
}
run_test 25 "change ACL on one mountpoint be seen on another ==="

test_26a() {
        utime $DIR1/f26a -s $DIR2/f26a || error
}
run_test 26a "allow mtime to get older"

test_26b() {
        touch $DIR1/$tfile
        sleep 1
        echo "aaa" >> $DIR1/$tfile
        sleep 1
        chmod a+x $DIR2/$tfile
        mt1=`stat -c %Y $DIR1/$tfile`
        mt2=`stat -c %Y $DIR2/$tfile`
        
        if [ x"$mt1" != x"$mt2" ]; then 
                error "not equal mtime, client1: "$mt1", client2: "$mt2"."
        fi
}
run_test 26b "sync mtime between ost and mds"

test_27() {
	cancel_lru_locks OSC
	lctl clear
	dd if=/dev/zero of=$DIR2/$tfile bs=$((4096+4))k conv=notrunc count=4 seek=3 &
	DD2_PID=$!
	sleep 0.050s
	log "dd 1 started"
	
	dd if=/dev/zero of=$DIR1/$tfile bs=$((16384-1024))k conv=notrunc count=1 seek=4 &
	DD1_PID=$!
	log "dd 2 started"
	
	sleep 1
	dd if=/dev/zero of=$DIR1/$tfile bs=8k conv=notrunc count=1 seek=0
	log "dd 3 finished"
	lctl set_param -n ldlm.dump_namespaces ""
	wait $DD1_PID $DD2_PID
	[ $? -ne 0 ] && lctl dk $TMP/debug || true
}
run_test 27 "align non-overlapping extent locks from request ==="

test_28() { # bug 9977
	ostID=`$LCTL dl | awk '/-osc-|OSC.*MNT/ { ost++; if (ost == 2) { print $1 } }'`

	lfs setstripe $DIR1/$tfile -s 1048576 -i 0 -c 2
	tOBJID=`lfs getstripe $DIR1/$tfile | awk '/^[[:space:]]+1/ {print $2}'`
	dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2

	$LCTL --device $ostID destroy "${tOBJID}"
    
	# reading of 1st stripe should pass
	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error
	# reading of 2nd stripe should fail (this stripe was destroyed)
	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 && error

	# now, recreating test file
	dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 || error
	# reading of 1st stripe should pass
	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error
	# reading of 2nd stripe should pass
	dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 || error
}
run_test 28 "read/write/truncate file with lost stripes"

test_29() { # bug 10999
	touch $DIR1/$tfile
	#define OBD_FAIL_LDLM_GLIMPSE  0x30f
	sysctl -w lustre.fail_loc=0x8000030f
	ls -l $DIR2/$tfile &
	sleep 0.500s
	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 ========="

test_30() { #bug #11110
    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 $?')
    wait
    [ $err -ne 116 ] && error_ignore 12900 "return code ($err) != -ESTALE" && return
    true
}

run_test 30 "recreate file race ========="

test_31() {
        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 $1}'`
        #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE   0x314
        sysctl -w lustre.fail_loc=0x314
        reads=`LANG=C dd if=$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=============="

log "cleanup: ======================================================"

check_and_cleanup_lustre

echo '=========================== finished ==============================='
[ -f "$SANITYLOG" ] && cat $SANITYLOG && grep -q FAIL $SANITYLOG && exit 1 || true
echo "$0: completed"