8 OPENFILE=${OPENFILE:-openfile}
9 OPENUNLINK=${OPENUNLINK:-openunlink}
10 export TMP=${TMP:-/tmp}
11 MOUNT_2=${MOUNT_2:-"yes"}
12 CHECK_GRANT=${CHECK_GRANT:-"yes"}
13 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
15 LUSTRE=${LUSTRE:-$(dirname $0)/..}
16 . $LUSTRE/tests/test-framework.sh
20 ALWAYS_EXCEPT="$SANITYN_EXCEPT "
21 # bug number for skipped test: LU-7105
23 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
25 if [ $mds1_FSTYPE = "zfs" ]; then
26 # bug number: LU-15757 (test_102() causes crash in umount later)
28 # LU-2829 / LU-2887 - make allowances for ZFS slowness
29 TEST33_NFILES=${TEST33_NFILES:-1000}
33 [ "$SLOW" = "no" ] && EXCEPT_SLOW="33a"
42 check_and_setup_lustre
47 rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}*
49 SAMPLE_FILE=$TMP/$(basename $0 .sh).junk
50 dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=1
52 # $RUNAS_ID may get set incorrectly somewhere else
53 [ $UID -eq 0 -a $RUNAS_ID -eq 0 ] && error "\$RUNAS_ID set to 0, but \$UID is also 0!"
55 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
58 local client2=${CLIENT2:-$HOSTNAME}
63 do_node_vp "$HOSTNAME" printf 'a b\n' |
64 diff $tmp - || error "do_node_vp mismatch"
66 do_node_vp "$client2" printf 'a b\n' |
67 diff $tmp - || error "do_node_vp mismatch"
69 do_facet_vp mds1 printf 'a b\n' |
70 diff $tmp - || error "do_facet_vp mismatch"
72 printf '%s' 1 2 3 4 5 6 7 8 \ 9 10 ' ' '"' "'" \! \' \( \) $'\n' > $tmp
74 do_node_vp "$HOSTNAME" printf '%s' 1 2 3 4 5 6 7 8 \ 9 10 ' ' '"' "'" \! \' \( \) $'\n' |
75 diff $tmp - || error "do_node_vp mismatch"
77 do_node_vp "$client2" printf '%s' 1 2 3 4 5 6 7 8 \ 9 10 ' ' '"' "'" \! \' \( \) $'\n' |
78 diff $tmp - || error "do_node_vp mismatch"
80 do_facet_vp mds1 printf '%s' 1 2 3 4 5 6 7 8 \ 9 10 ' ' '"' "'" \! \' \( \) $'\n' |
81 diff $tmp - || error "do_facet_vp mismatch"
83 run_test 0 "do_node_vp() and do_facet_vp() do the right thing"
87 [ -f $DIR2/$tfile ] || error "Check create"
88 chmod 777 $DIR2/$tfile
89 $CHECKSTAT -t file -p 0777 $DIR1/$tfile ||
90 error "Check attribute update for 0777"
92 chmod a-x $DIR2/$tfile
93 $CHECKSTAT -t file -p 0666 $DIR1/$tfile ||
94 error "Check attribute update for 0666"
97 $CHECKSTAT -a $DIR1/$tfile ||
98 error "Check unlink - removes file on other mountpoint"
100 run_test 1 "Check attribute updates on 2 mount points"
106 $CHECKSTAT -t file -p 0777 $DIR1/f2a ||
107 error "Either not file type or perms not 0777"
109 run_test 2a "check cached attribute updates on 2 mtpt's ========"
115 $CHECKSTAT -t file -p 0777 $DIR2/f2b ||
116 error "Either not file type or perms not 0777"
118 run_test 2b "check cached attribute updates on 2 mtpt's ========"
120 # NEED TO SAVE ROOT DIR MODE
123 $CHECKSTAT -t dir -p 0777 $DIR2 ||
124 error "Either not dir type or perms not 0777"
126 run_test 2c "check cached attribute updates on 2 mtpt's root ==="
130 $CHECKSTAT -t dir -p 0755 $DIR2 ||
131 error "Either not file type or perms not 0775"
133 run_test 2d "check cached attribute updates on 2 mtpt's root ==="
140 $RUNAS dd if=/dev/zero of=$DIR2/$tfile count=1 ||
143 run_test 2e "check chmod on root is propagated to others"
146 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
148 local remote_dir=$tdir/remote_dir
151 $LFS mkdir -i $MDTIDX $DIR1/$remote_dir ||
152 error "Create remote directory failed"
154 touch $DIR1/$remote_dir/$tfile ||
155 error "Create file under remote directory failed"
156 chmod 777 $DIR1/$remote_dir/$tfile ||
157 error "Chmod file under remote directory failed"
159 $CHECKSTAT -t file -p 0777 $DIR2/$remote_dir/$tfile ||
160 error "Check attr of file under remote directory failed"
162 chown $RUNAS_ID:$RUNAS_GID $DIR1/$remote_dir/$tfile ||
163 error "Chown file under remote directory failed"
165 $CHECKSTAT -u \#$RUNAS_ID -g \#$RUNAS_GID $DIR2/$remote_dir/$tfile ||
166 error "Check owner of file under remote directory failed"
168 cd $DIR2/$remote_dir || error "enter remote dir"
169 rm -rf $DIR1/$remote_dir/$tfile ||
170 error "Unlink remote directory failed"
172 $CHECKSTAT -t file $DIR2/$remote_dir/$tfile &&
173 error "unlink file still exists!"
175 cd $DIR2/$tdir || error "exit remote dir"
176 rm -rf $DIR1/$tdir || error "unlink directory failed"
178 run_test 2f "check attr/owner updates on DNE with 2 mtpt's"
181 dd if=/dev/zero of=$DIR1/$tfile oflag=sync bs=1M count=2
183 local block1=$(stat $DIR1/$tfile | awk '/Blocks/ {print $4} ')
185 local block2=$(stat $DIR2/$tfile | awk '/Blocks/ {print $4} ')
186 echo "$DIR1/$tfile has $block1 blocks"
187 echo "$DIR2/$tfile has $block2 blocks"
188 [ $block1 -eq $block2 ] || error "$block1 not equal to $block2"
190 run_test 2g "check blocks update on sync write"
193 local target="this/is/good"
194 ln -s $target $DIR1/$tfile || error "ln -s $target $DIR1/$tfile failed"
195 [ "$(ls -l $DIR2/$tfile | sed -e 's/.* -> //')" = "$target" ] ||
196 error "link $DIR2/$tfile not as expected"
198 run_test 3 "symlink on one mtpt, readlink on another ==========="
201 multifstat $DIR1/f4 $DIR2/f4
203 run_test 4 "fstat validation on multiple mount points =========="
207 $TRUNCATE $DIR2/f5 100
208 $CHECKSTAT -t file -s 100 $DIR1/f5 ||
209 error "Either not file type or size not equal to 100 bytes"
212 run_test 5 "create a file on one mount, truncate it on the other"
215 openunlink $DIR1/$tfile $DIR2/$tfile || \
216 error "openunlink $DIR1/$tfile $DIR2/$tfile"
218 run_test 6 "remove of open file on other node =================="
222 opendirunlink $DIR1/$dir $DIR2/$dir || \
223 error "opendirunlink $DIR1/$dir $DIR2/$dir"
225 run_test 7 "remove of open directory on other node ============="
228 opendevunlink $DIR1/$tfile $DIR2/$tfile || \
229 error "opendevunlink $DIR1/$tfile $DIR2/$tfile"
231 run_test 8 "remove of open special file on other node =========="
237 for C in a b c d e f g h i j k l; do
238 dir=`eval echo \\$DIR$MTPT`
239 echo -n $C >> $dir/f9
240 [ "$MTPT" -eq 1 ] && MTPT=2 || MTPT=1
242 [ "`cat $DIR1/f9`" = "abcdefghijkl" ] || \
243 error "`od -a $DIR1/f9` != abcdefghijkl"
245 run_test 9a "append of file with sub-page size on multiple mounts"
247 #LU-10681 - tiny writes & appending to sparse striped file
249 [[ $OSTCOUNT -ge 2 ]] || { skip "needs >= 2 OSTs"; return; }
251 $LFS setstripe -c 2 -S 1M $DIR/$tfile
252 echo "foo" >> $DIR/$tfile
253 dd if=/dev/zero of=$DIR2/$tfile bs=1M count=1 seek=1 conv=notrunc ||
254 error "sparse dd $DIR2/$tfile failed"
255 echo "foo" >> $DIR/$tfile
257 data=$(dd if=$DIR2/$tfile bs=1 count=3 skip=$((2 * 1048576)) conv=notrunc)
258 echo "Data read (expecting 'foo')": $data
259 [ "$data" = "foo" ] || error "append to sparse striped file failed"
261 run_test 9b "append to striped sparse file"
268 for C in a b c d e f g h i j k l; do
269 dir=`eval echo \\$DIR$MTPT`
270 echo -n $C | dd of=$dir/f10 bs=1 seek=$OFFSET count=1
271 [ "$MTPT" -eq 1 ] && MTPT=2 || MTPT=1
272 OFFSET=`expr $OFFSET + 1`
274 [ "`cat $DIR1/f10`" = "abcdefghijkl" ] || \
275 error "`od -a $DIR1/f10` != abcdefghijkl"
277 run_test 10a "write of file with sub-page size on multiple mounts "
281 yes "R" | head -c 4000 >$TMP/f10b-seed
282 dd if=$TMP/f10b-seed of=$DIR1/f10b bs=3k count=1 || error "dd $DIR1"
284 $TRUNCATE $DIR1/f10b 4096 || error "truncate 4096"
286 dd if=$DIR2/f10b of=$TMP/f10b-lustre bs=4k count=1 || error "dd $DIR2"
288 # create a test file locally to compare
289 dd if=$TMP/f10b-seed of=$TMP/f10b bs=3k count=1 || error "dd random"
290 $TRUNCATE $TMP/f10b 4096 || error "truncate 4096"
291 cmp $TMP/f10b $TMP/f10b-lustre || error "file miscompare"
292 rm $TMP/f10b $TMP/f10b-lustre $TMP/f10b-seed
294 run_test 10b "write of file with sub-page size on multiple mounts "
298 multiop_bg_pause $DIR1/d11/f O_c || return 1
300 cp -p /bin/ls $DIR1/d11/f
304 wait $MULTIPID || error "wait for PID $MULTIPID failed"
305 [ $RC -eq 0 ] && error || true
307 run_test 11 "execution of file opened for write should return error ===="
310 DIR=$DIR DIR2=$DIR2 bash lockorder.sh
312 run_test 12 "test lock ordering (link, stat, unlink)"
314 test_13() { # bug 2451 - directory coherency
316 cd $DIR1/d13 || error "cd to $DIR1/d13 failed"
318 ( touch $DIR1/d13/f13 ) # needs to be a separate shell
320 rm -f $DIR2/d13/f13 || error "Cannot remove $DIR2/d13/f13"
321 ls 2>&1 | grep f13 && error "f13 shouldn't return an error (1)" || true
322 # need to run it twice
323 ( touch $DIR1/d13/f13 ) # needs to be a separate shell
325 rm -f $DIR2/d13/f13 || error "Cannot remove $DIR2/d13/f13"
326 ls 2>&1 | grep f13 && error "f13 shouldn't return an error (2)" || true
328 run_test 13 "test directory page revocation"
331 test_mkdir $DIR1/$tdir
332 cp -p /bin/ls $DIR1/$tdir/$tfile
333 multiop_bg_pause $DIR1/$tdir/$tfile Ow_c || return 1
336 $DIR2/$tdir/$tfile && error || true
339 run_test 14aa "execution of file open for write returns -ETXTBSY"
342 test_mkdir $DIR1/$tdir
343 cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed"
344 $DIR1/$tdir/sleep 60 &
346 $MULTIOP $DIR2/$tdir/sleep Oc && error "expected error, got success"
349 run_test 14ab "open(RDWR) of executing file returns -ETXTBSY"
351 test_14b() { # bug 3192, 7040
352 test_mkdir $DIR1/$tdir
353 cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed"
354 $DIR1/$tdir/sleep 60 &
356 $TRUNCATE $DIR2/$tdir/sleep 60 && kill -9 $SLEEP_PID && \
357 error "expected truncate error, got success"
359 cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed"
361 run_test 14b "truncate of executing file returns -ETXTBSY ======"
363 test_14c() { # bug 3430, 7040
364 test_mkdir $DIR1/$tdir
365 cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed"
366 $DIR1/$tdir/sleep 60 &
368 cp /etc/hosts $DIR2/$tdir/sleep && error "expected error, got success"
370 cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed"
372 run_test 14c "open(O_TRUNC) of executing file return -ETXTBSY =="
374 test_14d() { # bug 10921
375 test_mkdir $DIR1/$tdir
376 cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed"
377 $DIR1/$tdir/sleep 60 &
380 chmod 600 $DIR1/$tdir/sleep || error "chmod failed"
382 cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed"
384 run_test 14d "chmod of executing file is still possible ========"
386 test_15() { # bug 974 - ENOSPC
388 bash oos2.sh $MOUNT1 $MOUNT2
389 wait_delete_completed
390 grant_error=$(dmesg | grep "< tot_grant")
391 [ -z "$grant_error" ] || error "$grant_error"
393 run_test 15 "test out-of-space with multiple writers ==========="
396 # The FSXNUM reduction for ZFS is needed until ORI-487 is fixed.
397 # We don't want to skip it entirely, but ZFS is VERY slow and cannot
398 # pass a 2500 operation dual-mount run within the time limit.
399 if [ "$ost1_FSTYPE" = "zfs" ]; then
400 FSXNUM=$((COUNT / 5))
402 elif [ "$SLOW" = "yes" ]; then
403 FSXNUM=$((COUNT * 5))
411 local file1=$DIR1/$tfile
412 local file2=$DIR2/$tfile
413 local stripe_size=$(do_facet $SINGLEMDS \
414 "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
418 # to allocate grant because it may run out due to test_15.
419 $LFS setstripe -c -1 $file1
420 dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync
421 dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync
424 $LFS setstripe -c -1 $file1 # b=10919
425 $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 ||
429 # O_DIRECT reads and writes must be aligned to the PAGE_SIZE.
430 $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 -Z -r $PAGE_SIZE \
431 -w $PAGE_SIZE $file1 $file2 || error "fsx with O_DIRECT failed."
433 run_test 16a "$FSXNUM iterations of dual-mount fsx"
435 # Consistency check for tiny writes, LU-9409
437 local file1=$DIR1/$tfile
438 local file2=$DIR2/$tfile
439 local stripe_size=($($LFS getstripe -S $DIR))
443 # to allocate grant because it may run out due to test_15.
444 lfs setstripe -c -1 $file1
445 dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync ||
446 error "dd failed writing to file=$file1"
447 dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync ||
448 error "dd failed writing to file=$file2"
451 lfs setstripe -c -1 $file1 # b=10919
452 # -o is set to 8192 because writes < 1 page and between 1 and 2 pages
453 # create a mix of tiny writes & normal writes
454 $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -o 8192 -S 0 \
455 $file1 $file2 || error "fsx with tiny write failed."
457 run_test 16b "$FSXNUM iterations of dual-mount fsx at small size"
460 local file1=$DIR1/$tfile
461 local file2=$DIR2/$tfile
462 local stripe_size=$(do_facet $SINGLEMDS \
463 "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
465 [ "$ost1_FSTYPE" != ldiskfs ] && skip "dio on ldiskfs only"
469 # to allocate grant because it may run out due to test_15.
470 $LFS setstripe -c -1 $file1
471 dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync
472 dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync
474 wait_delete_completed
476 local list=$(comma_list $(osts_nodes))
477 if ! get_osd_param $list '' read_cache_enable >/dev/null; then
478 skip "not cache-capable obdfilter"
481 set_osd_param $list '' read_cache_enable 0
482 set_osd_param $list '' writethrough_cache_enable 0
484 $LFS setstripe -c -1 $file1 # b=10919
485 $FSX -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 ||
489 set_osd_param $list '' read_cache_enable 1
490 set_osd_param $list '' writethrough_cache_enable 1
494 run_test 16c "verify data consistency on ldiskfs with cache disabled (b=17397)"
497 local file1=$DIR1/$tfile
498 local file2=$DIR2/$tfile
499 local file3=$DIR1/file
500 local tmpfile=$(mktemp)
501 local stripe_size=$(do_facet $SINGLEMDS \
502 "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
504 # to allocate grant because it may run out due to test_15.
505 $LFS setstripe -c -1 $file1
506 stack_trap "rm -f $file1 $file2 $file3 $tmpfile"
507 dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync
508 dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync
511 $LFS setstripe -c -1 $file1 # b=10919
512 $LCTL set_param ldlm.namespaces.*.lru_size=clear
514 # direct write on one client and direct read from another
515 dd if=/dev/urandom of=$file1 bs=1M count=100 oflag=direct
516 dd if=$file2 of=$tmpfile iflag=direct bs=1M
517 diff $file1 $tmpfile || error "file different(1)"
520 # buffer write on one client, but direct read from another
521 dd if=$tmpfile of=$file1 bs=1M count=100
522 dd if=$file2 of=$file3 bs=1M iflag=direct count=100
523 diff $file3 $tmpfile || error "file different(2)"
525 rm -f $file3 $file2 $file1
526 # direct write on one client
527 dd if=$tmpfile of=$file1 bs=1M count=100 oflag=direct
528 # buffer read from another client
529 dd if=$file2 of=$file3 bs=1M count=100
530 diff $file3 $tmpfile || error "file different(3)"
532 run_test 16d "Verify DIO and buffer IO with two clients"
534 test_16e() { # LU-13227
537 (( "$MDS1_VERSION" >= $(version_code 2.13.53) )) ||
538 skip "Need MDS version at least 2.13.53"
540 local file1=$DIR1/$tfile
541 local file2=$DIR2/$tfile
543 # client1 write 10M data
544 dd if=/dev/zero of=$file1 bs=1M count=10
546 cancel_lru_locks osc > /dev/null
547 # use lockahead to generate one PW lock to keep LVB loaded.
548 $LFS ladvise -a lockahead --start 0 --length 1M \
550 # direct write to extend file size on client2
551 dd if=/dev/zero of=$file2 bs=1M seek=20 count=1 \
552 oflag=direct conv=notrunc
553 local filesize=$(stat -c %s $file2)
554 [ "$filesize" -eq 22020096 ] ||
555 error "expected filesize 22020096 got $filesize"
558 run_test 16e "Verify size consistency for O_DIRECT write"
560 test_16f() { # LU-14541
561 local file1=$DIR1/$tfile
562 local file2=$DIR2/$tfile
566 (( $MDS1_VERSION > $(version_code 2.15.51) )) ||
567 skip "Need MDS version at least 2.15.51"
569 timeout --preserve-status --signal=USR1 $duration \
570 rw_seq_cst_vs_drop_caches $file1 $file2
573 case $((status & 0x7f)) in
575 echo OK # Computers must be fast now.
578 error "sequential consistency violation detected"
581 echo TIMEOUT # This is fine.
584 error "strange status '$status'"
590 run_test 16f "rw sequential consistency vs drop_caches"
593 local file1=$DIR1/$tfile
594 local file2=$DIR2/$tfile
598 timeout --preserve-status --signal=USR1 $duration \
599 rw_seq_cst_vs_drop_caches -m $file1 $file2
602 case $((status & 0x7f)) in
604 echo OK # Computers must be fast now.
607 error "sequential consistency violation detected"
610 echo TIMEOUT # This is fine.
613 error "strange status '$status'"
619 run_test 16g "mmap rw sequential consistency vs drop_caches"
622 local tf=$DIR/$tdir/$tfile
623 local tf2=$DIR2/$tdir/$tfile
624 local cmd="$MMAP_CAT $tf | od -x | tail -q -n4"
625 local cmd2="$MMAP_CAT $tf2 | od -x | tail -q -n4"
627 test_mkdir $DIR/$tdir
629 # create file and populate data
630 cp /etc/passwd $tf || error "cp failed"
632 local size=$(stat -c %s $tf)
635 if [[ "$c1" != "$c2" ]]; then
636 echo " ------- mount 1 read --------"
638 echo " ------- mount 2 read --------"
640 error "content mismatch"
643 echo " ------- before truncate --------"
647 $TRUNCATE $tf $((size / 2)) || error "truncate file"
650 echo " ------- after truncate --------"
652 # repeat the comparison
655 if [[ "$c1" != "$c2" ]]; then
656 echo " ------- mount 1 read --------"
658 echo " ------- mount 2 read --------"
660 error "content mismatch after truncate"
664 run_test 16h "mmap read after truncate file"
667 local tf=$DIR/$tdir/$tfile
668 local tf2=$DIR2/$tdir/$tfile
670 test_mkdir $DIR/$tdir
672 # create file and populate data
673 cp /etc/passwd $tf || error "cp failed"
675 local size=$(stat -c %s $tf)
677 c1=$(dd if=$tf bs=1 2>/dev/null | od -x | tail -q -n4)
678 c2=$(dd if=$tf2 bs=1 2>/dev/null | od -x | tail -q -n4)
680 if [[ "$c1" != "$c2" ]]; then
681 echo " ------- mount 1 read --------"
683 echo " ------- mount 2 read --------"
685 error "content mismatch"
688 echo " ------- before truncate --------"
692 $TRUNCATE $tf $((size / 2)) || error "truncate file"
694 echo " ------- after truncate --------"
696 # repeat the comparison
697 c1=$(dd if=$tf bs=1 2>/dev/null | od -x | tail -q -n4)
698 c2=$(dd if=$tf2 bs=1 2>/dev/null | od -x | tail -q -n4)
700 if [[ "$c1" != "$c2" ]]; then
701 echo " ------- mount 1 read --------"
703 echo " ------- mount 2 read --------"
705 error "content mismatch after truncate"
709 run_test 16i "read after truncate file"
713 (( $OSTCOUNT >= 2 )) || skip "needs >= 2 OSTs"
715 local stripe_size=$((1024 * 1024)) #1 MiB
716 # Max i/o below is ~ 4 * stripe_size, so this gives ~5 i/os
717 local file_size=$((25 * stripe_size))
720 $LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.1
721 stack_trap "rm -f $DIR/$tfile.1"
723 # Just a bit bigger than the largest size in the test set below
724 dd if=/dev/urandom bs=$file_size count=1 of=$DIR/$tfile.1 ||
725 error "buffered i/o to create file failed"
727 if zfs_or_rotational; then
728 # DIO on ZFS can take up to 2 seconds per IO
729 # rotational is better, but still slow.
730 # Limit testing on those media to larger sizes
731 bsizes="$((stripe_size - PAGE_SIZE)) $stripe_size \
732 $((stripe_size + 1024))"
734 bsizes="$((PAGE_SIZE / 4)) $((PAGE_SIZE - 1024)) \
735 $((PAGE_SIZE - 1)) $PAGE_SIZE $((PAGE_SIZE + 1024)) \
736 $((PAGE_SIZE * 3/2)) $((PAGE_SIZE * 4)) \
737 $((stripe_size - 1)) $stripe_size \
738 $((stripe_size + 1)) $((stripe_size * 3/2)) \
739 $((stripe_size * 4)) $((stripe_size * 4 + 1))"
742 # 1 process (BIO or DIO) on each client
743 for bs in $bsizes; do
744 $LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
745 echo "bs: $bs, file_size $file_size"
746 dd if=$DIR/$tfile.1 bs=$((bs *2 )) of=$DIR/tfile.2 \
747 conv=notrunc oflag=direct iflag=direct &
749 # Buffered I/O with similar but not the same block size
750 dd if=$DIR2/$tfile.1 bs=$((bs * 2)) of=$DIR2/$tfile.2 \
757 if (( rc1 != 0 )); then
758 error "dio copy 1 w/bsize $bs failed: $rc1"
760 if (( rc2 != 0 )); then
761 error "buffered copy 2 w/bsize $bs failed: $rc2"
764 $CHECKSTAT -t file -s $file_size $DIR/$tfile.2 ||
765 error "size incorrect"
766 $CHECKSTAT -t file -s $file_size $DIR2/$tfile.2 ||
767 error "size incorrect - mount 2"
768 cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 ||
769 error "files differ, bsize $bs"
770 cmp --verbose $DIR2/$tfile.1 $DIR2/$tfile.2 ||
771 error "files differ, bsize $bs - mount 2"
775 # 2 processes - both DIO and BIO - on each client
776 for bs in $bsizes; do
777 $LFS setstripe -c 2 -S $stripe_size $DIR/$tfile.2
778 echo "bs: $bs, file_size $file_size"
780 dd if=$DIR/$tfile.1 bs=$((bs *2 )) of=$DIR/tfile.2 \
781 conv=notrunc oflag=direct iflag=direct &
783 # Buffered I/O with similar but not the same block size
784 dd if=$DIR/$tfile.1 bs=$((bs * 2)) of=$DIR/$tfile.2 \
788 dd if=$DIR2/$tfile.1 bs=$((bs *2 )) of=$DIR2/tfile.2 \
789 conv=notrunc oflag=direct iflag=direct &
791 # Buffered I/O with similar but not the same block size
792 dd if=$DIR2/$tfile.1 bs=$((bs * 2)) of=$DIR2/$tfile.2 \
804 if (( rc1 != 0 )); then
805 error "dio copy 1 w/bsize $bs failed: $rc1"
807 if (( rc2 != 0 )); then
808 error "buffered copy 2 w/bsize $bs failed: $rc2"
810 if (( rc3 != 0 )); then
811 error "dio copy 3 w/bsize $bs failed: $rc1"
813 if (( rc4 != 0 )); then
814 error "buffered copy 4 w/bsize $bs failed: $rc2"
817 $CHECKSTAT -t file -s $file_size $DIR/$tfile.2 ||
818 error "size incorrect"
819 $CHECKSTAT -t file -s $file_size $DIR2/$tfile.2 ||
820 error "size incorrect - mount 2"
821 cmp --verbose $DIR/$tfile.1 $DIR/$tfile.2 ||
822 error "files differ, bsize $bs"
823 cmp --verbose $DIR2/$tfile.1 $DIR2/$tfile.2 ||
824 error "files differ, bsize $bs - mount 2"
828 run_test 16j "race dio with buffered i/o"
831 local fsxN=${FSX_NP:-5}
832 local fsxNops=${FSX_NOPS:-10000}
833 local fsxNparams=${FSXPARAMS_16k:-""}
834 local dropsleep=${DROP_SLEEP:-3}
839 [ "$SLOW" = "no" ] && fsxNops=1000
842 echo 3 > /proc/sys/vm/drop_caches
846 stack_trap "kill -9 $dpipd"
848 for ((i = 1; i <= fsxN; i++)); do
849 cmd="$FSX $fsxNparams -N $fsxNops $DIR/fsxfile.${i} -l $((64 * 1024 * 1024))"
854 for ((i = 1; i <= fsxN; i++)); do
855 wait ${fsxpids[$i]} && echo "+ fsx $i done: rc=$?" ||
856 error "- fsx $i FAILURE! rc=$?"
860 run_test 16k "Parallel FSX and drop caches should not panic"
862 test_17() { # bug 3513, 3667
863 remote_ost_nodsh && skip "remote OST with nodsh" && return
865 lfs setstripe $DIR1/$tfile -i 0 -c 1
866 cp $SAMPLE_FILE $DIR1/$tfile
867 cancel_lru_locks osc > /dev/null
868 #define CFS_FAIL_ONCE|OBD_FAIL_LDLM_CREATE_RESOURCE 0x30a
869 do_facet ost1 lctl set_param fail_loc=0x8000030a
870 ls -ls $DIR1/$tfile | awk '{ print $1,$6 }' > $DIR1/$tfile-1 & \
871 ls -ls $DIR2/$tfile | awk '{ print $1,$6 }' > $DIR2/$tfile-2
873 diff -u $DIR1/$tfile-1 $DIR2/$tfile-2 || error "files are different"
875 run_test 17 "resource creation/LVB creation race ==============="
878 # turn e.g. ALWAYS_EXCEPT="18c" into "-e 3"
881 for idx in {a..z}; do
882 local ptr=EXCEPT_ALWAYS_18$idx
883 [ x${!ptr} = xtrue ] || continue
885 excepts="$excepts -e $(($(printf %d \'$idx)-96))"
888 excepts="$excepts -e 7 -e 8 -e 9"
889 $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2 $excepts ||
890 error "mmap_sanity test failed"
893 run_test 18 "mmap sanity check ================================="
895 test_19() { # bug3811
896 local node=$(facet_active_host ost1)
897 local device="$FSNAME-OST*"
899 [ "x$DOM" = "xyes" ] && node=$(facet_active_host $SINGLEMDS) &&
900 device="$FSNAME-MDT*"
902 # check whether obdfilter is cache capable at all
903 get_osd_param $node $device read_cache_enable >/dev/null ||
904 skip "not cache-capable obdfilter"
906 local max=$(get_osd_param $node $device readcache_max_filesize |\
908 set_osd_param $node $device readcache_max_filesize 4096
909 dd if=/dev/urandom of=$TMP/$tfile bs=512k count=32
910 local SUM=$(cksum $TMP/$tfile | cut -d" " -f 1,2)
911 cp $TMP/$tfile $DIR1/$tfile
912 for i in `seq 1 20`; do
913 [ $((i % 5)) -eq 0 ] && log "$testname loop $i"
914 cancel_lru_locks $OSC > /dev/null
915 cksum $DIR1/$tfile | cut -d" " -f 1,2 > $TMP/sum1 & \
916 cksum $DIR2/$tfile | cut -d" " -f 1,2 > $TMP/sum2
918 [ "$(cat $TMP/sum1)" = "$SUM" ] || \
919 error "$DIR1/$tfile $(cat $TMP/sum1) != $SUM"
920 [ "$(cat $TMP/sum2)" = "$SUM" ] || \
921 error "$DIR2/$tfile $(cat $TMP/sum2) != $SUM"
923 set_osd_param $node $device readcache_max_filesize $max
926 run_test 19 "test concurrent uncached read races ==============="
929 test_mkdir $DIR1/$tdir
931 CNT=$($LCTL get_param -n llite.*.dump_page_cache | wc -l)
932 $MULTIOP $DIR1/$tdir/$tfile Ow8190c
933 $MULTIOP $DIR2/$tdir/$tfile Oz8194w8190c
934 $MULTIOP $DIR1/$tdir/$tfile Oz0r8190c
936 CNT2=$($LCTL get_param -n llite.*.dump_page_cache | wc -l)
937 [[ $CNT2 == $CNT ]] ||
938 error $((CNT2 - CNT))" page left in cache after lock cancel"
940 run_test 20 "test extra readahead page left in cache ===="
947 test_21() { # Bug 5907
948 test_mkdir $DIR1/$tdir
949 mount /etc $DIR1/$tdir --bind || error "mount failed" # Poor man's mount.
951 rmdir -v $DIR1/$tdir && error "Removed mounted directory"
952 rmdir -v $DIR2/$tdir && echo "Removed mounted directory from another mountpoint, needs to be fixed"
953 test -d $DIR1/$tdir || error "Mounted directory disappeared"
955 test -d $DIR2/$tdir || test -d $DIR1/$tdir && error "Removed dir still visible after umount"
958 run_test 21 " Try to remove mountpoint on another dir ===="
960 test_23() { # Bug 5972
961 local at_diff=$(do_facet $SINGLEMDS \
962 $LCTL get_param -n mdd.*MDT0000*.atime_diff | head -n1)
963 echo "atime should be updated while another read" > $DIR1/$tfile
965 # clear the lock(mode: LCK_PW) gotten from creating operation
966 cancel_lru_locks $OSC
969 sleep $((at_diff + 1))
971 echo "starting reads"
972 multiop_bg_pause $DIR1/$tfile or20_c || return 1
973 # with SOM and opencache enabled, we need to close a file and cancel
974 # open lock to get atime propogated to MDS
975 kill -USR1 $! || return 2
978 time2=$(stat -c "%X" $DIR/$tfile)
979 echo "new atime is $time2"
981 [ $time2 -gt $time1 ] || error "atime was not updated"
982 rm -f $DIR1/$tfile || error "rm -f $DIR1/$tfile failed"
985 run_test 23 " others should see updated atime while another read===="
989 lfs df || error "lfs df failed"
990 lfs df -ih || error "lfs df -ih failed"
991 lfs df -h $DIR1 || error "lfs df -h $DIR1 failed"
992 lfs df -i $DIR2 || error "lfs df -i $DIR2 failed"
993 lfs df $DIR1/$tfile || error "lfs df $DIR1/$tfile failed"
994 lfs df -ih $DIR2/$tfile || error "lfs df -ih $DIR2/$tfile failed"
996 OSC=`lctl dl | awk '/-osc-|OSC.*MNT/ {print $4}' | head -n 1`
997 # OSC=`lctl dl | awk '/-osc-/ {print $4}' | head -n 1`
998 lctl --device %osc deactivate
999 lfs df -i || error "lfs df -i with deactivated OSC failed"
1000 lctl --device %osc activate
1001 lfs df || error "lfs df with reactivated OSC failed"
1003 run_test 24a "lfs df [-ih] [path] test ========================="
1007 fsnum=$(lfs_df | grep -c "summary")
1008 [ $fsnum -eq 2 ] || error "lfs df shows $fsnum != 2 filesystems."
1010 run_test 24b "lfs df should show both filesystems ==============="
1013 local acl=$(lctl get_param -n mdc.*MDT0000-mdc-*.connect_flags |
1015 [ "$acl" -lt 1 ] && skip "must have acl, skipping" && return
1017 mkdir -p $DIR1/$tdir
1018 touch $DIR1/$tdir/f1 || error "touch $DIR1/$tdir/f1"
1019 chmod 0755 $DIR1/$tdir/f1 || error "chmod 0755 $DIR1/$tdir/f1"
1021 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #1"
1022 setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir ||
1023 error "setfacl $DIR2/$tdir #1"
1024 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #2"
1025 setfacl -m u:$RUNAS_ID:r-x -m g:$RUNAS_GID:r-x $DIR1/$tdir ||
1026 error "setfacl $DIR2/$tdir #2"
1027 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #3"
1028 setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir ||
1029 error "setfacl $DIR2/$tdir #3"
1030 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #4"
1031 setfacl -x u:$RUNAS_ID: -x g:$RUNAS_GID: $DIR1/$tdir ||
1032 error "setfacl $DIR2/$tdir #4"
1033 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #5"
1037 run_test 25a "change ACL on one mountpoint be seen on another ==="
1040 local acl=$(lctl get_param -n mdc.*MDT0000-mdc-*.connect_flags |
1042 [ "$acl" -lt 1 ] && skip "must have acl, skipping" && return
1044 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
1047 $LFS mkdir -i 1 $DIR1/$tdir
1048 touch $DIR1/$tdir/f1 || error "touch $DIR1/$tdir/f1"
1049 chmod 0755 $DIR1/$tdir/f1 || error "chmod 0755 $DIR1/$tdir/f1"
1051 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #1"
1052 setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir ||
1053 error "setfacl $DIR2/$tdir #1"
1054 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #2"
1055 setfacl -m u:$RUNAS_ID:r-x -m g:$RUNAS_GID:r-x $DIR1/$tdir ||
1056 error "setfacl $DIR2/$tdir #2"
1057 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #3"
1058 setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir ||
1059 error "setfacl $DIR2/$tdir #3"
1060 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #4"
1061 setfacl -x u:$RUNAS_ID: -x g:$RUNAS_GID: $DIR1/$tdir ||
1062 error "setfacl $DIR2/$tdir #4"
1063 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #5"
1067 run_test 25b "change ACL under remote dir on one mountpoint be seen on another"
1070 utime $DIR1/f26a -s $DIR2/f26a || error "utime failed for $DIR1/f26a"
1072 run_test 26a "allow mtime to get older"
1077 echo "aaa" >> $DIR1/$tfile
1079 chmod a+x $DIR2/$tfile
1080 mt1=`stat -c %Y $DIR1/$tfile`
1081 mt2=`stat -c %Y $DIR2/$tfile`
1083 if [ x"$mt1" != x"$mt2" ]; then
1084 error "not equal mtime, client1: "$mt1", client2: "$mt2"."
1087 run_test 26b "sync mtime between ost and mds"
1090 cancel_lru_locks $OSC
1092 dd if=/dev/zero of=$DIR2/$tfile bs=$((4096+4))k conv=notrunc count=4 seek=3 &
1097 dd if=/dev/zero of=$DIR1/$tfile bs=$((16384-1024))k conv=notrunc count=1 seek=4 &
1102 dd if=/dev/zero of=$DIR1/$tfile bs=8k conv=notrunc count=1 seek=0
1104 lctl set_param -n ldlm.dump_namespaces ""
1105 wait $DD1_PID $DD2_PID
1106 [ $? -ne 0 ] && lctl dk $TMP/debug || true
1108 run_test 27 "align non-overlapping extent locks from request ==="
1110 test_28() { # bug 9977
1111 skip "echo_client on osc is no longer supported"
1113 run_test 28 "read/write/truncate file with lost stripes"
1115 test_30() { #b=11110, LU-2523
1116 test_mkdir $DIR1/$tdir
1117 cp -f /bin/bash $DIR1/$tdir/bash
1118 bash -c 'sleep 1; rm -f $DIR2/$tdir/bash; cp /bin/bash $DIR2/$tdir' &
1119 $DIR1/$tdir/bash -c 'sleep 2;
1120 openfile -f O_RDONLY /proc/$$/exe >& /dev/null; echo $?'
1124 run_test 30 "recreate file race"
1127 test_mkdir $DIR1/$tdir
1128 local writes=$(LANG=C dd if=/dev/zero of=$DIR/$tdir/$tfile \
1129 count=1 2>&1 | awk 'BEGIN { FS="+" } /out/ {print $1}')
1130 #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314
1131 lctl set_param fail_loc=0x314
1132 local reads=$(LANG=C dd if=$DIR2/$tdir/$tfile of=/dev/null 2>&1 |
1133 awk 'BEGIN { FS="+" } /in/ {print $1}')
1134 [ $reads -eq $writes ] || error "read" $reads "blocks, must be" $writes
1136 run_test 31a "voluntary cancel / blocking ast race=============="
1139 remote_ost || { skip "local OST" && return 0; }
1140 remote_ost_nodsh && skip "remote OST w/o dsh" && return 0
1142 # make sure there is no local locks due to destroy
1143 wait_mds_ost_sync || error "wait_mds_ost_sync()"
1144 wait_delete_completed || error "wait_delete_completed()"
1146 test_mkdir $DIR1/$tdir
1147 lfs setstripe $DIR/$tdir/$tfile -i 0 -c 1
1148 cp /etc/hosts $DIR/$tdir/$tfile
1149 #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314
1150 lctl set_param fail_loc=0x314
1151 #define OBD_FAIL_LDLM_OST_FAIL_RACE 0x316
1152 do_facet ost1 lctl set_param fail_loc=0x316
1153 # Don't crash kernel
1154 cat $DIR2/$tdir/$tfile > /dev/null 2>&1
1155 lctl set_param fail_loc=0
1156 do_facet ost1 lctl set_param fail_loc=0
1157 # cleanup: reconnect the client back
1160 run_test 31b "voluntary OST cancel / blocking ast race=============="
1162 #LU-14949 - multi-client version of the test 31r in sanity.
1164 touch $DIR/$tfile.target
1165 touch $DIR/$tfile.source
1167 ls -l $DIR/$tfile.target # cache it for sure
1169 #OBD_FAIL_LLITE_OPEN_DELAY 0x1419
1170 $LCTL set_param fail_loc=0x1419 fail_val=3
1171 cat $DIR/$tfile.target &
1174 # Guarantee open is waiting before we get here
1176 mv $DIR2/$tfile.source $DIR2/$tfile.target
1180 if [[ $RC -ne 0 ]]; then
1181 error "open with cat failed, rc=$RC"
1184 run_test 31r "open-rename(replace) race"
1186 test_32b() { # bug 11270
1187 remote_ost_nodsh && skip "remote OST with nodsh" && return
1190 local facets=$(get_facets OST)
1191 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
1193 save_lustre_params client "osc.*.contention_seconds" > $p
1194 save_lustre_params $facets \
1195 "ldlm.namespaces.filter-*.max_nolock_bytes" >> $p
1196 save_lustre_params $facets \
1197 "ldlm.namespaces.filter-*.contended_locks" >> $p
1198 save_lustre_params $facets \
1199 "ldlm.namespaces.filter-*.contention_seconds" >> $p
1200 clear_stats $OSC.*.${OSC}_stats
1202 # agressive lockless i/o settings
1203 do_nodes $(comma_list $(osts_nodes)) \
1204 "lctl set_param -n ldlm.namespaces.*.max_nolock_bytes=2000000 \
1205 ldlm.namespaces.filter-*.contended_locks=0 \
1206 ldlm.namespaces.filter-*.contention_seconds=60"
1207 lctl set_param -n $OSC.*.contention_seconds=60
1209 dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \
1211 dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \
1214 [ $(calc_stats $OSC.*.${OSC}_stats lockless_write_bytes) -ne 0 ] ||
1215 error "lockless i/o was not triggered"
1216 # disable lockless i/o (it is disabled by default)
1217 do_nodes $(comma_list $(osts_nodes)) \
1218 "lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes=0 \
1219 ldlm.namespaces.filter-*.contended_locks=32 \
1220 ldlm.namespaces.filter-*.contention_seconds=0"
1221 # set contention_seconds to 0 at client too, otherwise Lustre still
1222 # remembers lock contention
1223 lctl set_param -n $OSC.*.contention_seconds=0
1224 clear_stats $OSC.*.${OSC}_stats
1226 dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \
1228 dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \
1231 [ $(calc_stats $OSC.*.${OSC}_stats lockless_write_bytes) -eq 0 ] ||
1232 error "lockless i/o works when disabled"
1234 restore_lustre_params <$p
1237 # Disable test 32b prior to full removal
1238 #run_test 32b "lockless i/o"
1241 local mdts=$(get_facets MDS)
1247 for mds in ${mdts//,/ }; do
1250 dev=$(basename $(do_facet $mds "lctl get_param -n \
1251 osd*.${!varsvc}.mntdev | xargs readlink -f"))
1252 val=$(do_facet $mds "cat /proc/fs/jbd*/${dev}{,:*,-*}/info \
1253 2>/dev/null | head -n1")
1255 stat=$((stat + val))
1260 # commit on sharing tests
1262 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1264 [ $CLIENTCOUNT -lt 2 ] &&
1265 skip "Need two or more clients, have $CLIENTCOUNT"
1267 local nfiles=${TEST33_NFILES:-10000}
1268 local param_file=$TMP/$tfile-params
1274 save_lustre_params $(get_facets MDS) \
1275 "mdt.*.commit_on_sharing" > $param_file
1278 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=$COS
1282 do_nodes $CLIENT1,$CLIENT2 "mkdir -p $DIR1/$tdir-\\\$(hostname)-$i"
1284 [ "$mds1_FSTYPE" = ldiskfs ] && jbdold=$(print_jbd_stat)
1285 echo "=== START createmany old: $jbdold transaction"
1286 local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > /dev/null 2>&1")
1287 [ "$mds1_FSTYPE" = ldiskfs ] && jbdnew=$(print_jbd_stat)
1288 [ "$mds1_FSTYPE" = ldiskfs ] && jbd=$(( jbdnew - jbdold ))
1289 echo "=== END createmany new: $jbdnew transaction : $jbd transactions nfiles $nfiles time $elapsed COS=$COS"
1290 [ "$mds1_FSTYPE" = ldiskfs ] && avgjbd=$(( avgjbd + jbd ))
1291 avgtime=$(( avgtime + elapsed ))
1293 eval cos${COS}_jbd=$((avgjbd / 3))
1294 eval cos${COS}_time=$((avgtime / 3))
1297 echo "COS=0 transactions (avg): $cos0_jbd time (avg): $cos0_time"
1298 echo "COS=1 transactions (avg): $cos1_jbd time (avg): $cos1_time"
1299 [ "$cos0_jbd" != 0 ] &&
1300 echo "COS=1 vs COS=0 jbd: $((((cos1_jbd/cos0_jbd - 1)) * 100 )) %"
1301 [ "$cos0_time" != 0 ] &&
1302 echo "COS=1 vs COS=0 time: $((((cos1_time/cos0_time - 1)) * 100 )) %"
1304 restore_lustre_params < $param_file
1308 run_test 33a "commit on sharing, cross crete/delete, 2 clients, benchmark"
1310 # commit on sharing tests
1312 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1314 [ $CLIENTCOUNT -ge 2 ] ||
1315 { skip "Need two or more clients, have $CLIENTCOUNT" &&
1317 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
1319 local nfiles=${TEST33_NFILES:-10000}
1320 local param_file=$TMP/$tfile-params
1322 save_lustre_params $(get_facets MDS) \
1323 "mdt.*.commit_on_sharing" > $param_file
1332 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=$COS
1336 do_node $CLIENT1 "$LFS mkdir -i $MDTIDX \
1337 $DIR1/$tdir-\\\$(hostname)-$i"
1339 jbdold=$(print_jbd_stat)
1340 echo "=== START createmany old: $jbdold transaction"
1341 local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2\
1342 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- \
1343 -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > \
1345 jbdnew=$(print_jbd_stat)
1346 jbd=$(( jbdnew - jbdold ))
1347 echo "=== END createmany new: $jbdnew transaction : \
1348 $jbd transactions nfiles $nfiles time $elapsed COS=$COS"
1349 avgjbd=$(( avgjbd + jbd ))
1350 avgtime=$(( avgtime + elapsed ))
1352 eval cos${COS}_jbd=$((avgjbd / 3))
1353 eval cos${COS}_time=$((avgtime / 3))
1356 echo "COS=0 transactions (avg): $cos0_jbd time (avg): $cos0_time"
1357 echo "COS=1 transactions (avg): $cos1_jbd time (avg): $cos1_time"
1358 [ "$cos0_jbd" != 0 ] &&
1359 echo "COS=1 vs COS=0 jbd: $(((cos1_jbd/cos0_jbd - 1) * 100)) %"
1360 [ "$cos0_time" != 0 ] &&
1361 echo "COS=1 vs COS=0 time: $(((cos1_time/cos0_time - 1) * 100)) %"
1363 restore_lustre_params < $param_file
1367 run_test 33b "COS: cross create/delete, 2 clients, benchmark under remote dir"
1369 # arg1 is description, arg2 is operations before Sync-on-Lock-Cancel, arg3 is
1370 # the operation that triggers SoLC
1374 local nodes=$(comma_list $(mdts_nodes))
1378 # trigger CoS twice in case transaction commit before unlock
1381 do_nodes $nodes "$LCTL set_param -n mdt.*.sync_count=0"
1383 sync_count=$(do_nodes $nodes \
1384 "lctl get_param -n mdt.*MDT*.sync_count" | calc_sum)
1385 total=$((total + sync_count));
1393 echo " SoLC count $total"
1394 (( total > 0 )) || error "$3 didn't trigger SoLC"
1404 (( MDSCOUNT >= 2 )) || skip "needs >= 2 MDTs"
1405 (( MDS1_VERSION >= $(version_code 2.7.63) )) ||
1406 skip "DNE CoS not supported"
1410 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "start mds1 failed"
1412 mkdir_on_mdt0 $DIR/$tdir
1415 if (( MDS1_VERSION < $(version_code 2.15.55.204) )); then
1416 op_trigger_solc "create remote dir and local dir" \
1417 "$LFS mkdir -i 1 $DIR/$tdir/remote" \
1418 "$LFS mkdir -i 0 $DIR/$tdir/local"
1419 (( MDSCOUNT > 2 )) &&
1420 op_trigger_solc "create remote dirs on different MDTs" \
1421 "$LFS mkdir -i 1 $DIR/$tdir/remote.1" \
1422 "$LFS mkdir -i 2 $DIR/$tdir/remote.2"
1424 op_trigger_solc "create file on 2nd stripe under striped directory" \
1425 "$LFS mkdir -i 0 -c 2 $DIR/$tdir/striped" \
1426 "touch $DIR2/$tdir/striped/subfile"
1429 echo "Below operations shouldn't trigger Solc:"
1430 $LFS mkdir -i 0 -c 2 $DIR/$tdir/striped
1432 do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
1433 if (( MDS1_VERSION >= $(version_code 2.15.55.204) )); then
1434 test_33_run "create file on 2nd stripe after setattr" \
1435 "chmod 777 $DIR/$tdir/striped; \
1436 touch $DIR2/$tdir/striped/subfile"
1437 test_33_run "create remote dir and local dir" \
1438 "$LFS mkdir -i 1 $DIR/$tdir/remote" \
1439 "$LFS mkdir -i 0 $DIR/$tdir/local"
1440 (( MDSCOUNT > 2 )) &&
1441 test_33_run "create remote dirs on different MDTs" \
1442 "$LFS mkdir -i 1 $DIR/$tdir/remote.1" \
1443 "$LFS mkdir -i 2 $DIR/$tdir/remote.2"
1445 test_33_run "create local dir after remote dir creation transaction commit" \
1446 "$LFS mkdir -i 1 $DIR/$tdir/remote.3; \
1447 do_facet mds2 $LCTL set_param -n osd*.*MDT0001.force_sync 1;
1448 mkdir $DIR/$tdir/local.3"
1449 sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count")
1450 echo "Solc count $sync_count"
1451 [ $sync_count -eq 0 ] || error "Sync-Lock-Cancel triggered"
1453 run_test 33c "Cancel cross-MDT lock should trigger Sync-on-Lock-Cancel"
1455 # arg1 is description, arg2 is operations done before CoS, arg3 is the operation
1460 local nodes=$(comma_list $(mdts_nodes))
1464 # trigger CoS twice in case transaction commit before unlock
1467 do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
1469 commit_nr=$(do_nodes $nodes \
1470 "lctl get_param -n mdt.*.async_commit_count" | calc_sum)
1471 total=$((total + commit_nr));
1479 echo " CoS count $total"
1480 (( total > 0 )) || error "$3 didn't trigger CoS"
1484 (( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
1485 (( MDS1_VERSION >= $(version_code 2.7.63) )) ||
1486 skip "DNE CoS not supported"
1488 if (( $MDS1_VERSION < $(version_code 2.15.55.133) )); then
1489 op_trigger_cos "remote directory unlink" \
1490 "$LFS mkdir -i 1 $DIR/$tdir" "rmdir $DIR2/$tdir"
1491 op_trigger_cos "striped directory create" "mkdir $DIR/$tdir" \
1492 "$LFS mkdir -c 2 $DIR2/$tdir/subdir"
1493 op_trigger_cos "striped directory setattr" \
1494 "$LFS mkdir -c 2 $DIR/$tdir" "chmod 713 $DIR2/$tdir"
1495 op_trigger_cos "striped directory unlink" \
1496 "$LFS mkdir -c 2 $DIR/$tdir" "rmdir $DIR2/$tdir"
1497 op_trigger_cos "cross-MDT link" \
1498 "mkdir $DIR/$tdir; \
1499 $LFS mkdir -i 0 $DIR/$tdir/d1; \
1500 $LFS mkdir -i 1 $DIR/$tdir/d2; \
1501 touch $DIR/$tdir/d1/tgt" \
1502 "ln $DIR2/$tdir/d1/tgt $DIR2/$tdir/d2/src"
1505 op_trigger_cos "remote directory create" "$LFS mkdir -i 0 $DIR/$tdir" \
1506 "$LFS mkdir -i 1 $DIR2/$tdir/subdir"
1507 op_trigger_cos "cross-MDT rename" \
1508 "mkdir $DIR/$tdir; \
1509 $LFS mkdir -i 0 $DIR/$tdir/d1; \
1510 $LFS mkdir -i 1 $DIR/$tdir/d2; \
1511 touch $DIR/$tdir/d1/src" \
1512 "mv $DIR2/$tdir/d1/src $DIR2/$tdir/d2/tgt"
1513 op_trigger_cos "migrate" \
1514 "$LFS mkdir -i 0 $DIR/$tdir" \
1515 "$LFS migrate -m 1 $DIR2/$tdir"
1519 run_test 33d "dependent transactions should trigger COS"
1522 (( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs"
1523 (( MDS1_VERSION >= $(version_code 2.7.63) )) ||
1524 skip "DNE CoS not supported"
1526 $LFS mkdir -i 0 $DIR/$tdir
1527 $LFS mkdir -i 0 $DIR/$tdir/d1
1528 $LFS mkdir -i 1 $DIR/$tdir/d2
1530 local nodes=$(comma_list $(mdts_nodes))
1531 do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
1533 test_33_run "plain dir creation" "mkdir $DIR2/$tdir/plain"
1534 test_33_run "open file and write" "echo abc > $DIR2/$tdir/$tfile"
1535 test_33_run "append write" "echo dfg >> $DIR2/$tdir/$tfile"
1536 test_33_run "setattr" "touch $DIR2/$tdir/$tfile"
1537 test_33_run "file unlink" "rm $DIR2/$tdir/$tfile"
1538 test_33_run "plain dir unlink" "rmdir $DIR2/$tdir/plain"
1539 if (( MDS1_VERSION >= $(version_code 2.15.55.133) )); then
1540 test_33_run "striped directory creation" \
1541 "$LFS mkdir -i 0 -c 2 $DIR2/$tdir/striped"
1542 test_33_run "set default LMV to create striped subdir" \
1543 "$LFS setdirstripe -D -c 2 $DIR/$tdir"
1544 test_33_run "striped subdir creation" \
1545 "createmany -d $DIR/$tdir/subdir 100"
1546 test_33_run "sub file creation and write" \
1547 "createmany -o $DIR/$tdir/subfile 100; \
1548 echo abc > $DIR/$tdir/subfile1"
1549 test_33_run "sub file append write" \
1550 "echo dfg >> $DIR2/$tdir/subfile2"
1551 test_33_run "subdir setatttr" "touch $DIR2/$tdir/subdir1"
1552 test_33_run "subdir unlink" \
1553 "unlinkmany -d $DIR/$tdir/subdir 100"
1554 test_33_run "sub file unlink" \
1555 "unlinkmany $DIR2/$tdir/subfile 100"
1556 test_33_run "sub file creation follows striped dir chmod" \
1557 "chmod 777 $DIR/$tdir/striped; \
1558 touch $DIR/$tdir/striped/subfile"
1559 test_33_run "striped directory unlink" \
1560 "rm -rf $DIR2/$tdir/striped"
1563 test_33_run "directory unlink" "rm -rf $DIR2/$tdir"
1565 local async_commit_count=$(do_nodes $nodes \
1566 "lctl get_param -n mdt.*.async_commit_count" | calc_sum)
1567 echo "CoS count $async_commit_count"
1568 (( async_commit_count == 0 )) || error "CoS triggerred"
1570 run_test 33e "independent transactions shouldn't trigger COS"
1572 # End commit on sharing tests
1574 get_ost_lock_timeouts() {
1575 local nodes=${1:-$(comma_list $(osts_nodes))}
1577 local locks=$(do_nodes $nodes \
1578 "lctl get_param -n ldlm.namespaces.filter-*.lock_timeouts" | calc_sum)
1586 do_nodes $(comma_list $(osts_nodes)) \
1587 "lctl set_param -n fail_loc=0 2>/dev/null || true"
1588 for i in $(seq $OSTCOUNT); do
1589 wait_osc_import_ready client ost$i
1594 remote_ost_nodsh && skip "remote OST with nodsh" && return
1598 trap cleanup_34 EXIT RETURN
1599 for OPER in notimeout timeout ; do
1600 rm $DIR1/$tfile 2>/dev/null
1601 lock_in=$(get_ost_lock_timeouts)
1602 if [ $OPER == "timeout" ] ; then
1603 for j in `seq $OSTCOUNT`; do
1604 #define OBD_FAIL_PTLRPC_HPREQ_TIMEOUT 0x511
1605 do_facet ost$j lctl set_param fail_loc=0x511
1607 echo lock should expire
1609 for j in `seq $OSTCOUNT`; do
1610 #define OBD_FAIL_PTLRPC_HPREQ_NOTIMEOUT 0x512
1611 do_facet ost$j lctl set_param fail_loc=0x512
1613 echo lock should not expire
1615 echo writing on client1
1616 dd if=/dev/zero of=$DIR1/$tfile count=100 conv=notrunc > /dev/null 2>&1
1618 echo reading on client2
1619 dd of=/dev/null if=$DIR2/$tfile > /dev/null 2>&1
1620 # wait for a lock timeout
1622 lock_out=$(get_ost_lock_timeouts)
1623 if [ $OPER == "timeout" ] ; then
1624 if [ $lock_in == $lock_out ]; then
1625 error "no lock timeout happened"
1630 if [ $lock_in != $lock_out ]; then
1631 error "lock timeout happened"
1639 run_test 34 "no lock timeout under IO"
1641 test_35() { # bug 17645
1644 gen=$(lctl get_param mdc.$FSNAME-MDT*-mdc-*.import | grep generation |
1645 awk '/generation/{print $2}')
1647 generation[count]=$g
1651 test_mkdir $MOUNT1/$tdir
1652 cancel_lru_locks mdc
1654 # Let's initiate -EINTR situation by setting fail_loc and take
1655 # write lock on same file from same client. This will not cause
1656 # bl_ast yet as lock is already in local cache.
1657 #define OBD_FAIL_LDLM_INTR_CP_AST 0x317
1658 do_facet client "lctl set_param fail_loc=0x80000317"
1659 local timeout=$(do_facet $SINGLEMDS lctl get_param -n timeout)
1660 let timeout=timeout*3
1662 while test $nr -lt 10; do
1663 log "Race attempt $nr"
1664 local blk1=$(lctl get_param -n ldlm.services.ldlm_cbd.stats |
1665 awk '/ldlm_bl_callback/ { print $2 }')
1666 test "x$blk1" = "x" && blk1=0
1667 createmany -o $MOUNT2/$tdir/a 4000 &
1671 # Let's make conflict and bl_ast
1672 ls -la $MOUNT1/$tdir > /dev/null &
1675 log "Wait for $pid1 $pid2 for $timeout sec..."
1677 kill -9 $pid1 $pid2 > /dev/null 2>&1
1679 local blk2=$(lctl get_param -n ldlm.services.ldlm_cbd.stats |
1680 awk '/ldlm_bl_callback/ { print $2 }')
1681 test "x$blk2" = "x" && blk2=0
1682 test $blk2 -gt $blk1 && break
1683 rm -fr $MOUNT1/$tdir
1684 cancel_lru_locks mdc
1687 do_facet client "lctl set_param fail_loc=0x0"
1688 df -h $MOUNT1 $MOUNT2
1690 gen=$(lctl get_param mdc.$FSNAME-MDT*-mdc-*.import | grep generation |
1691 awk '/generation/{print $2}')
1693 if ! test "$g" -eq "${generation[count]}"; then
1694 list=$(lctl list_param mdc.$FSNAME-MDT*-mdc-*.import)
1696 for imp in $list; do
1697 if [ $c = $count ]; then
1702 imp=$(echo "$imp" | awk -F"." '{print $2}')
1703 error "Eviction happened on import $imp"
1708 run_test 35 "-EINTR cp_ast vs. bl_ast race does not evict client"
1710 test_36() { #bug 16417
1715 test_mkdir $DIR1/$tdir
1716 $LFS setstripe -c -1 $DIR1/$tdir
1719 let SIZE_B=SIZE*1024*1024
1720 sync; sleep 2; sync # wait for delete thread
1721 wait_mds_ost_sync || error "wait_mds_ost_sync failed"
1722 wait_destroy_complete || error "wait_destroy_complete failed"
1724 while [ $i -le 10 ]; do
1725 lctl mark "start test - cycle ($i)"
1726 local before=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
1727 dd if=/dev/zero of=$DIR1/$tdir/$tfile bs=1M count=$SIZE ||
1728 error "dd $DIR1/$tdir/$tfile ${SIZE}MB failed"
1729 sync # sync data from client cache
1730 sync_all_data # sync data from server cache (delayed allocation)
1732 local after_dd=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
1733 multiop_bg_pause $DIR2/$tdir/$tfile O_r${SIZE_B}c || return 3
1735 rm -f $DIR1/$tdir/$tfile
1736 kill -USR1 $read_pid
1738 sync; sleep 2; sync # Ensure new statfs
1739 wait_delete_completed
1740 local after=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
1741 echo "*** cycle($i) *** before($before) after_dd($after_dd)" \
1743 # this free space! not used
1744 (( $after_dd <= $after)) ||
1745 error "space leaked after_dd:$after_dd > after:$after"
1749 run_test 36 "handle ESTALE/open-unlink correctly"
1751 test_37() { # bug 18695
1752 test_mkdir $DIR1/$tdir
1753 multiop_bg_pause $DIR1/$tdir D_c || return 1
1755 # create large directory (32kB seems enough from e2fsck, ~= 1000 files)
1756 createmany -m $DIR2/$tdir/f 10000
1757 # set mtime/atime backward
1758 touch -t 198001010000 $DIR2/$tdir
1759 kill -USR1 $MULTIPID
1760 nr_files=`lfs find $DIR1/$tdir -type f | wc -l`
1761 [ $nr_files -eq 10000 ] || error "$nr_files != 10000 truncated directory?"
1764 run_test 37 "check i_size is not updated for directory on close (bug 18695) =============="
1766 # this should be set to past
1767 TEST_39_MTIME=`date -d "1 year ago" +%s`
1771 local client1=${CLIENT1:-`hostname`}
1772 local client2=${CLIENT2:-`hostname`}
1774 do_node $client1 "touch $DIR1/$tfile"
1776 do_node $client1 "touch -m -d @$TEST_39_MTIME $DIR1/$tfile"
1777 local mtime1=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1778 [ "$mtime1" = $TEST_39_MTIME ] || \
1779 error "mtime is not set to past: $mtime1, should be $TEST_39_MTIME"
1781 local d1=`do_node $client1 date +%s`
1782 do_node $client1 'echo hello >> '$DIR1/$tfile
1783 local d2=`do_node $client1 date +%s`
1785 local mtime2=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1786 [ "$mtime2" -ge "$d1" ] && [ "$mtime2" -le "$d2" ] || \
1787 error "mtime is not updated on write: $d1 <= $mtime2 <= $d2"
1789 do_node $client1 "mv $DIR1/$tfile $DIR1/$tfile-1"
1791 for (( i=0; i < 2; i++ )) ; do
1792 local mtime3=`do_node $client2 "stat -c %Y $DIR1/$tfile-1"`
1793 [ "$mtime2" = "$mtime3" ] || \
1794 error "mtime ($mtime2) changed (to $mtime3) on rename"
1796 cancel_lru_locks osc
1797 if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi
1800 run_test 39a "file mtime does not change after rename"
1803 local client1=${CLIENT1:-`hostname`}
1804 local client2=${CLIENT2:-`hostname`}
1808 local mtime1=`stat -c %Y $DIR1/$tfile`
1809 local mtime2=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1812 touch -m -d @$TEST_39_MTIME $DIR1/$tfile
1814 for (( i=0; i < 2; i++ )) ; do
1815 local mtime3=`stat -c %Y $DIR1/$tfile`
1816 local mtime4=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1818 [ "$mtime3" = "$mtime4" ] || \
1819 error "different mtime on clients: $mtime3, $mtime4"
1820 [ "$mtime3" = $TEST_39_MTIME ] || \
1821 error "lost mtime: $mtime3, should be $TEST_39_MTIME"
1823 cancel_lru_locks osc
1824 if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi
1827 run_test 39b "file mtime the same on clients with/out lock"
1830 local client1=${CLIENT1:-`hostname`}
1831 local client2=${CLIENT2:-`hostname`}
1833 echo hello > $DIR1/$tfile
1835 local mtime1=`stat -c %Y $DIR1/$tfile`
1836 local mtime2=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1837 [ "$mtime1" = "$mtime2" ] || \
1838 error "create: different mtime on clients: $mtime1, $mtime2"
1841 $TRUNCATE $DIR1/$tfile 1
1843 for (( i=0; i < 2; i++ )) ; do
1844 local mtime3=`stat -c %Y $DIR1/$tfile`
1845 local mtime4=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1847 [ "$mtime3" = "$mtime4" ] || \
1848 error "different mtime on clients: $mtime3, $mtime4"
1849 [ "$mtime3" -gt $mtime2 ] || \
1850 error "truncate did not update mtime: $mtime2, $mtime3"
1852 cancel_lru_locks osc
1853 if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi
1856 run_test 39c "check truncate mtime update ======================"
1858 test_39d() { # LU-7310
1860 touch -m -d @$TEST_39_MTIME $DIR1/$tfile
1862 local mtime1=$(stat -c %Y $DIR2/$tfile)
1863 [ "$mtime1" = $TEST_39_MTIME ] ||
1864 error "mtime: $mtime1, should be $TEST_39_MTIME"
1867 # define OBD_FAIL_OSC_NO_GRANT 0x411
1868 $LCTL set_param fail_loc=0x411
1870 local d1=$(date +%s)
1871 echo hello >> $DIR1/$tfile
1872 local d2=$(date +%s)
1874 $LCTL set_param fail_loc=0
1876 cancel_lru_locks $OSC
1878 local mtime2=$(stat -c %Y $DIR2/$tfile)
1879 [ "$mtime2" -ge "$d1" ] && [ "$mtime2" -le "$d2" ] ||
1880 error "mtime is not updated on write: $d1 <= $mtime2 <= $d2"
1882 run_test 39d "sync write should update mtime"
1885 # how long 40-47 take with specific delay
1892 # for pdo testing, we must cancel MDT-MDT locks as well as client locks to
1893 # avoid unexpected delays due to previous tests
1895 cancel_lru_locks mdc
1896 do_nodes $(comma_list $(mdts_nodes)) \
1897 $LCTL set_param -n ldlm.namespaces.*mdt*.lru_size=clear
1898 do_nodes $(comma_list $(mdts_nodes)) \
1899 $LCTL get_param ldlm.namespaces.*mdt*.lock_unused_count \
1900 ldlm.namespaces.*mdt*.lock_count | grep -v '=0'
1903 # check that pid exists hence second operation wasn't blocked by first one
1904 # if it is so then there is no conflict, return 0
1905 # else second operation is conflicting with first one, return 1
1906 check_pdo_conflict() {
1909 pdo_sched # to ensure OP1 is finished on client if OP2 is blocked by OP1
1910 if [[ `ps --pid $pid | wc -l` == 1 ]]; then
1920 # test 40: check non-blocking operations
1922 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1924 mkdir_on_mdt0 $DIR2/$tdir
1926 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1927 do_nodes $(comma_list $(mdts_nodes)) \
1928 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
1929 mkdir $DIR1/$tdir/$tfile &
1931 touch $DIR2/$tdir/$tfile-2
1932 check_pdo_conflict $PID1 || error "create is blocked"
1933 mkdir $DIR2/$tdir/$tfile-3
1934 check_pdo_conflict $PID1 || error "mkdir is blocked"
1935 link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4
1936 check_pdo_conflict $PID1 || error "link is blocked"
1937 mv $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-5
1938 check_pdo_conflict $PID1 || error "rename is blocked"
1939 stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null
1940 check_pdo_conflict $PID1 || error "getattr is blocked"
1941 rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-5
1942 rmdir $DIR2/$tdir/$tfile-3
1943 check_pdo_conflict $PID1 || error "unlink is blocked"
1945 # all operations above shouldn't wait the first one
1946 check_pdo_conflict $PID1 || error "parallel operation is blocked"
1947 do_nodes $(comma_list $(mdts_nodes)) \
1948 "lctl set_param -n fail_loc=0 2>/dev/null || true"
1953 run_test 40a "pdirops: create vs others =============="
1956 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1958 mkdir_on_mdt0 $DIR2/$tdir
1960 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1961 do_nodes $(comma_list $(mdts_nodes)) \
1962 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
1963 touch $DIR1/$tdir/$tfile &
1966 touch $DIR2/$tdir/$tfile-2
1967 check_pdo_conflict $PID1 || error "create is blocked"
1968 mkdir $DIR2/$tdir/$tfile-3
1969 check_pdo_conflict $PID1 || error "mkdir is blocked"
1970 link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4
1971 check_pdo_conflict $PID1 || error "link is blocked"
1972 mv $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-5
1973 check_pdo_conflict $PID1 || error "rename is blocked"
1974 stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null
1975 check_pdo_conflict $PID1 || error "getattr is blocked"
1976 rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-5
1977 rmdir $DIR2/$tdir/$tfile-3
1978 check_pdo_conflict $PID1 || error "unlink is blocked"
1979 # all operations above shouldn't wait the first one
1981 check_pdo_conflict $PID1 || error "parallel operation is blocked"
1982 do_nodes $(comma_list $(mdts_nodes)) \
1983 "lctl set_param -n fail_loc=0 2>/dev/null || true"
1988 run_test 40b "pdirops: open|create and others =============="
1991 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1993 mkdir_on_mdt0 $DIR2/$tdir
1995 touch $DIR1/$tdir/$tfile
1996 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1997 do_nodes $(comma_list $(mdts_nodes)) \
1998 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
1999 link $DIR1/$tdir/$tfile $DIR1/$tdir/$tfile-0 &
2002 touch $DIR2/$tdir/$tfile-2
2003 check_pdo_conflict $PID1 || error "create is blocked"
2004 mkdir $DIR2/$tdir/$tfile-3
2005 check_pdo_conflict $PID1 || error "mkdir is blocked"
2006 link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4
2007 check_pdo_conflict $PID1 || error "link is blocked"
2008 mv $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-5
2009 check_pdo_conflict $PID1 || error "rename is blocked"
2010 stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null
2011 check_pdo_conflict $PID1 || error "getattr is blocked"
2012 rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-5
2013 rmdir $DIR2/$tdir/$tfile-3
2014 check_pdo_conflict $PID1 || error "unlink is blocked"
2016 # all operations above shouldn't wait the first one
2017 check_pdo_conflict $PID1 || error "parallel operation is blocked"
2018 do_nodes $(comma_list $(mdts_nodes)) \
2019 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2024 run_test 40c "pdirops: link and others =============="
2027 remote_mds_nodsh && skip "remote MDS with nodsh" && return
2029 mkdir_on_mdt0 $DIR2/$tdir
2031 touch $DIR1/$tdir/$tfile
2032 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2033 do_nodes $(comma_list $(mdts_nodes)) \
2034 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2035 rm $DIR1/$tdir/$tfile &
2038 touch $DIR2/$tdir/$tfile-2
2039 check_pdo_conflict $PID1 || error "create is blocked"
2040 mkdir $DIR2/$tdir/$tfile-3
2041 check_pdo_conflict $PID1 || error "mkdir is blocked"
2042 link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4
2043 check_pdo_conflict $PID1 || error "link is blocked"
2044 mv $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-5
2045 check_pdo_conflict $PID1 || error "rename is blocked"
2046 stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null
2047 check_pdo_conflict $PID1 || error "getattr is blocked"
2048 rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-5
2049 rmdir $DIR2/$tdir/$tfile-3
2050 check_pdo_conflict $PID1 || error "unlink is blocked"
2052 # all operations above shouldn't wait the first one
2053 check_pdo_conflict $PID1 || error "parallel operation is blocked"
2054 do_nodes $(comma_list $(mdts_nodes)) \
2055 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2059 run_test 40d "pdirops: unlink and others =============="
2062 remote_mds_nodsh && skip "remote MDS with nodsh" && return
2064 mkdir_on_mdt0 $DIR2/$tdir
2066 touch $DIR1/$tdir/$tfile
2067 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2068 do_nodes $(comma_list $(mdts_nodes)) \
2069 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2070 mv $DIR1/$tdir/$tfile $DIR1/$tdir/$tfile-0 &
2073 touch $DIR2/$tdir/$tfile-2
2074 check_pdo_conflict $PID1 || error "create is blocked"
2075 mkdir $DIR2/$tdir/$tfile-3
2076 check_pdo_conflict $PID1 || error "mkdir is blocked"
2077 link $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile-4
2078 check_pdo_conflict $PID1 || error "link is blocked"
2079 stat $DIR2/$tdir/$tfile-3 $DIR2/$tdir/$tfile-4 > /dev/null
2080 check_pdo_conflict $PID1 || error "getattr is blocked"
2081 rm $DIR2/$tdir/$tfile-4 $DIR2/$tdir/$tfile-2
2082 rmdir $DIR2/$tdir/$tfile-3
2083 check_pdo_conflict $PID1 || error "unlink is blocked"
2085 # all operations above shouldn't wait the first one
2086 check_pdo_conflict $PID1 || error "parallel operation is blocked"
2087 do_nodes $(comma_list $(mdts_nodes)) \
2088 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2093 run_test 40e "pdirops: rename and others =============="
2095 # test 41: create blocking operations
2098 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2099 do_nodes $(comma_list $(mdts_nodes)) \
2100 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2101 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
2103 mkdir $DIR2/$tfile &
2105 do_nodes $(comma_list $(mdts_nodes)) \
2106 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2107 check_pdo_conflict $PID1 && { wait $PID1; echo "mkdir isn't blocked"; }
2108 wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail"
2112 run_test 41a "pdirops: create vs mkdir =============="
2116 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2117 do_nodes $(comma_list $(mdts_nodes)) \
2118 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2119 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
2121 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c &
2123 do_nodes $(comma_list $(mdts_nodes)) \
2124 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2125 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
2126 wait $PID2 ; [ $? -ne 0 ] || error "create must fail"
2130 run_test 41b "pdirops: create vs create =============="
2134 touch $DIR1/$tfile-2
2135 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2136 do_nodes $(comma_list $(mdts_nodes)) \
2137 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2138 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
2140 link $DIR2/$tfile-2 $DIR2/$tfile &
2142 do_nodes $(comma_list $(mdts_nodes)) \
2143 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2144 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2145 wait $PID2 ; [ $? -ne 0 ] || error "link must fail"
2149 run_test 41c "pdirops: create vs link =============="
2153 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2154 do_nodes $(comma_list $(mdts_nodes)) \
2155 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2156 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
2160 do_nodes $(comma_list $(mdts_nodes)) \
2161 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2162 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
2163 wait $PID2 ; [ $? -eq 0 ] || error "unlink must succeed"
2167 run_test 41d "pdirops: create vs unlink =============="
2171 touch $DIR1/$tfile-2
2172 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2173 do_nodes $(comma_list $(mdts_nodes)) \
2174 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2175 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
2177 mv $DIR2/$tfile-2 $DIR2/$tfile &
2179 do_nodes $(comma_list $(mdts_nodes)) \
2180 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2181 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2182 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
2186 run_test 41e "pdirops: create and rename (tgt) =============="
2190 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2191 do_nodes $(comma_list $(mdts_nodes)) \
2192 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2193 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
2195 mv $DIR2/$tfile $DIR2/$tfile-2 &
2197 do_nodes $(comma_list $(mdts_nodes)) \
2198 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2199 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2200 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
2204 run_test 41f "pdirops: create and rename (src) =============="
2208 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2209 do_nodes $(comma_list $(mdts_nodes)) \
2210 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2211 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
2213 stat $DIR2/$tfile > /dev/null &
2215 do_nodes $(comma_list $(mdts_nodes)) \
2216 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2217 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
2218 wait $PID2 ; [ $? -eq 0 ] || error "stat must succeed"
2222 run_test 41g "pdirops: create vs getattr =============="
2226 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2227 do_nodes $(comma_list $(mdts_nodes)) \
2228 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2229 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
2231 ls -lia $DIR2/ > /dev/null &
2233 do_nodes $(comma_list $(mdts_nodes)) \
2234 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2235 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
2240 run_test 41h "pdirops: create vs readdir =============="
2247 do_nodes $(comma_list $(mdts_nodes)) \
2248 "lctl set_param -n fail_loc=${fail_loc} || true" &>/dev/null
2250 $MULTIOP $DIR1/$tfile oO_CREAT:O_EXCL:c 2>/dev/null &
2253 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c 2>/dev/null &
2256 if ! wait $PID1 && ! wait $PID2; then
2257 echo "Both creates failed (1 should fail, 1 should succeed)"
2259 elif wait $PID1 && wait $PID2; then
2260 echo "Both creates succeeded (1 should fail, 1 should succeed)"
2265 do_nodes $(comma_list $(mdts_nodes)) \
2266 "lctl set_param -n fail_loc=0x0 || true" &>/dev/null
2273 (( $MDS1_VERSION >= $(version_code 2.13.56) )) ||
2274 skip "Need MDS version newer than 2.13.56"
2277 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169
2278 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a
2279 for fail_loc in "0x80000169" "0x8000016a"; do
2280 echo "Begin 100 tests with fail_loc=$fail_loc"
2282 for i in {1..100}; do
2284 msg=$(sub_test_41i "$fail_loc") ||
2285 { echo; error "iter=$i : $msg"; }
2290 run_test 41i "reint_open: create vs create"
2293 # test 42: unlink and blocking operations
2296 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2297 do_nodes $(comma_list $(mdts_nodes)) \
2298 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2299 mkdir $DIR1/$tfile &
2301 mkdir $DIR2/$tfile &
2303 do_nodes $(comma_list $(mdts_nodes)) \
2304 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2305 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
2306 wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail"
2310 run_test 42a "pdirops: mkdir vs mkdir =============="
2314 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2315 do_nodes $(comma_list $(mdts_nodes)) \
2316 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2317 mkdir $DIR1/$tfile &
2319 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c &
2321 do_nodes $(comma_list $(mdts_nodes)) \
2322 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2323 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
2324 wait $PID2 ; [ $? -ne 0 ] || error "create must fail"
2328 run_test 42b "pdirops: mkdir vs create =============="
2332 touch $DIR1/$tfile-2
2333 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2334 do_nodes $(comma_list $(mdts_nodes)) \
2335 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2336 mkdir $DIR1/$tfile &
2338 link $DIR2/$tfile-2 $DIR2/$tfile &
2340 do_nodes $(comma_list $(mdts_nodes)) \
2341 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2342 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2343 wait $PID2 ; [ $? -ne 0 ] || error "link must fail"
2347 run_test 42c "pdirops: mkdir vs link =============="
2351 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2352 do_nodes $(comma_list $(mdts_nodes)) \
2353 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2354 mkdir $DIR1/$tfile &
2356 rmdir $DIR2/$tfile &
2358 do_nodes $(comma_list $(mdts_nodes)) \
2359 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2360 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
2361 wait $PID2 ; [ $? -eq 0 ] || error "unlink must succeed"
2365 run_test 42d "pdirops: mkdir vs unlink =============="
2369 touch $DIR1/$tfile-2
2370 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2371 do_nodes $(comma_list $(mdts_nodes)) \
2372 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2373 mkdir $DIR1/$tfile &
2375 mv -T $DIR2/$tfile-2 $DIR2/$tfile &
2377 do_nodes $(comma_list $(mdts_nodes)) \
2378 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2379 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2380 wait $PID2 ; [ $? -ne 0 ] || error "rename must fail"
2384 run_test 42e "pdirops: mkdir and rename (tgt) =============="
2388 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2389 do_nodes $(comma_list $(mdts_nodes)) \
2390 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2391 mkdir $DIR1/$tfile &
2393 mv $DIR2/$tfile $DIR2/$tfile-2 &
2395 do_nodes $(comma_list $(mdts_nodes)) \
2396 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2397 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2398 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
2402 run_test 42f "pdirops: mkdir and rename (src) =============="
2405 mkdir_on_mdt0 $DIR1/$tdir
2407 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2408 do_nodes $(comma_list $(mdts_nodes)) \
2409 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2410 mkdir $DIR1/$tdir/$tfile &
2412 stat $DIR2/$tdir/$tfile > /dev/null &
2414 do_nodes $(comma_list $(mdts_nodes)) \
2415 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2416 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
2417 wait $PID2 ; [ $? -eq 0 ] || error "stat must succeed"
2420 run_test 42g "pdirops: mkdir vs getattr =============="
2424 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2425 do_nodes $(comma_list $(mdts_nodes)) \
2426 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2427 mkdir $DIR1/$tfile &
2429 ls -lia $DIR2/ > /dev/null &
2431 do_nodes $(comma_list $(mdts_nodes)) \
2432 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2433 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
2438 run_test 42h "pdirops: mkdir vs readdir =============="
2440 # test 43: rmdir,mkdir won't return -EEXIST
2442 for i in {1..1000}; do
2443 mkdir $DIR1/$tdir || error "mkdir $tdir failed"
2444 rmdir $DIR2/$tdir || error "rmdir $tdir failed"
2448 run_test 43a "rmdir,mkdir doesn't return -EEXIST =============="
2453 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2454 do_nodes $(comma_list $(mdts_nodes)) \
2455 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2458 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c &
2460 do_nodes $(comma_list $(mdts_nodes)) \
2461 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2462 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
2463 wait $PID2 ; [ $? -eq 0 ] || error "create must succeed"
2467 run_test 43b "pdirops: unlink vs create =============="
2472 touch $DIR1/$tfile-2
2473 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2474 do_nodes $(comma_list $(mdts_nodes)) \
2475 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2478 link $DIR2/$tfile-2 $DIR2/$tfile &
2480 do_nodes $(comma_list $(mdts_nodes)) \
2481 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2482 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2483 wait $PID2 ; [ $? -eq 0 ] || error "link must succeed"
2487 run_test 43c "pdirops: unlink vs link =============="
2492 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2493 do_nodes $(comma_list $(mdts_nodes)) \
2494 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2499 do_nodes $(comma_list $(mdts_nodes)) \
2500 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2501 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
2502 wait $PID2 ; [ $? -ne 0 ] || error "unlink must fail"
2506 run_test 43d "pdirops: unlink vs unlink =============="
2511 touch $DIR1/$tfile-2
2512 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2513 do_nodes $(comma_list $(mdts_nodes)) \
2514 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2517 mv -u $DIR2/$tfile-2 $DIR2/$tfile &
2519 do_nodes $(comma_list $(mdts_nodes)) \
2520 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2521 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2522 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
2526 run_test 43e "pdirops: unlink and rename (tgt) =============="
2531 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2532 do_nodes $(comma_list $(mdts_nodes)) \
2533 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2536 mv $DIR2/$tfile $DIR2/$tfile-2 &
2538 do_nodes $(comma_list $(mdts_nodes)) \
2539 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2540 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2541 wait $PID2 ; [ $? -ne 0 ] || error "rename must fail"
2545 run_test 43f "pdirops: unlink and rename (src) =============="
2550 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2551 do_nodes $(comma_list $(mdts_nodes)) \
2552 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2555 stat $DIR2/$tfile > /dev/null &
2557 do_nodes $(comma_list $(mdts_nodes)) \
2558 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2559 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
2560 wait $PID2 ; [ $? -ne 0 ] || error "stat must fail"
2564 run_test 43g "pdirops: unlink vs getattr =============="
2569 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2570 do_nodes $(comma_list $(mdts_nodes)) \
2571 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2574 ls -lia $DIR2/ > /dev/null &
2576 do_nodes $(comma_list $(mdts_nodes)) \
2577 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2578 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
2583 run_test 43h "pdirops: unlink vs readdir =============="
2586 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2589 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2590 do_nodes $(comma_list $(mdts_nodes)) \
2591 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2594 $LFS mkdir -i 1 $DIR2/$tfile &
2596 do_nodes $(comma_list $(mdts_nodes)) \
2597 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2598 check_pdo_conflict $PID1 &&
2599 { wait $PID1; error "remote mkdir isn't blocked"; }
2600 wait $PID2 ; [ $? -eq 0 ] || error "remote mkdir must succeed"
2604 run_test 43i "pdirops: unlink vs remote mkdir"
2607 [[ $MDS1_VERSION -lt $(version_code 2.13.52) ]] &&
2608 skip "Need MDS version newer than 2.13.52"
2610 mkdir_on_mdt0 $DIR1/$tdir
2611 for i in {1..100}; do
2612 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_CREATE_RACE 0x167
2613 do_nodes $(comma_list $(mdts_nodes)) \
2614 "lctl set_param -n fail_loc=0x80000167 2>/dev/null ||
2617 mkdir $DIR1/$tdir/sub &
2619 mkdir $DIR2/$tdir/sub && ((OK++))
2620 wait $PID1 && ((OK++))
2621 (( OK == 1 )) || error "exactly one mkdir should succeed"
2623 rmdir $DIR1/$tdir/sub || error "rmdir failed"
2627 run_test 43j "racy mkdir return EEXIST =============="
2634 # We test in a separate directory to be able to unblock server thread in
2635 # cfs_race() if LCK_PW is taken on the parent by mdt_reint_unlink.
2636 test_mkdir $DIR2/$tdir
2637 touch $DIR2/$tdir/$tfile
2640 do_nodes $(comma_list $(mdts_nodes)) \
2641 "lctl set_param -n fail_loc=${fail_loc} || true" &>/dev/null
2642 echo content > $DIR1/$tdir/$tfile & PID1=$!
2644 multiop $DIR2/$tdir/$tfile u & PID2=$!
2648 echo -n "overwriting $tfile should succeed (err=$ret); "; }
2651 echo -n "unlinking $tfile should succeed (err=$ret);"; }
2654 do_nodes $(comma_list $(mdts_nodes)) \
2655 "lctl set_param -n fail_loc=0x0 || true" &>/dev/null
2662 (( $MDS1_VERSION >= $(version_code 2.13.56) )) ||
2663 skip "Need MDS version newer than 2.13.56"
2666 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169
2667 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a
2668 for fail_loc in "0x80000169" "0x8000016a"; do
2669 echo "Begin 100 tests with fail_loc=$fail_loc"
2671 for i in {1..100}; do
2673 msg=$(sub_test_43k "$fail_loc") ||
2674 { echo; error "iter=$i : $msg"; }
2684 run_test 43k "unlink vs create"
2686 # test 44: rename tgt and blocking operations
2689 touch $DIR1/$tfile-2
2690 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2691 do_nodes $(comma_list $(mdts_nodes)) \
2692 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2693 mv $DIR1/$tfile-2 $DIR1/$tfile &
2695 mkdir $DIR2/$tfile &
2697 do_nodes $(comma_list $(mdts_nodes)) \
2698 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2699 check_pdo_conflict $PID1 && { wait $PID1; date;error "mkdir isn't blocked"; }
2700 wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail"
2705 run_test 44a "pdirops: rename tgt vs mkdir =============="
2709 touch $DIR1/$tfile-2
2710 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2711 do_nodes $(comma_list $(mdts_nodes)) \
2712 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2713 mv $DIR1/$tfile-2 $DIR1/$tfile &
2715 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c &
2717 do_nodes $(comma_list $(mdts_nodes)) \
2718 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2719 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
2720 wait $PID2 ; [ $? -ne 0 ] || error "create must fail"
2724 run_test 44b "pdirops: rename tgt vs create =============="
2728 touch $DIR1/$tfile-2
2729 touch $DIR1/$tfile-3
2730 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2731 do_nodes $(comma_list $(mdts_nodes)) \
2732 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2733 mv $DIR1/$tfile-2 $DIR1/$tfile &
2735 link $DIR2/$tfile-3 $DIR2/$tfile &
2737 do_nodes $(comma_list $(mdts_nodes)) \
2738 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2739 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2740 wait $PID2 ; [ $? -ne 0 ] || error "link must fail"
2744 run_test 44c "pdirops: rename tgt vs link =============="
2748 touch $DIR1/$tfile-2
2749 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2750 do_nodes $(comma_list $(mdts_nodes)) \
2751 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2752 mv $DIR1/$tfile-2 $DIR1/$tfile &
2756 do_nodes $(comma_list $(mdts_nodes)) \
2757 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2758 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
2759 wait $PID2 ; [ $? -eq 0 ] || error "unlink must succeed"
2763 run_test 44d "pdirops: rename tgt vs unlink =============="
2768 touch $DIR1/$tfile-2
2769 touch $DIR1/$tfile-3
2770 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2771 do_nodes $(comma_list $(mdts_nodes)) \
2772 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2773 mv $DIR1/$tfile-2 $DIR1/$tfile &
2775 mv $DIR2/$tfile-3 $DIR2/$tfile &
2777 do_nodes $(comma_list $(mdts_nodes)) \
2778 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2779 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2780 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
2784 run_test 44e "pdirops: rename tgt and rename (tgt) =============="
2788 touch $DIR1/$tfile-2
2789 touch $DIR1/$tfile-3
2790 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2791 do_nodes $(comma_list $(mdts_nodes)) \
2792 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2793 mv $DIR1/$tfile-2 $DIR1/$tfile &
2795 mv $DIR2/$tfile $DIR2/$tfile-3 &
2797 do_nodes $(comma_list $(mdts_nodes)) \
2798 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2799 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2800 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
2804 run_test 44f "pdirops: rename tgt and rename (src) =============="
2808 touch $DIR1/$tfile-2
2809 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2810 do_nodes $(comma_list $(mdts_nodes)) \
2811 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2812 mv $DIR1/$tfile-2 $DIR1/$tfile &
2814 stat $DIR2/$tfile > /dev/null &
2816 do_nodes $(comma_list $(mdts_nodes)) \
2817 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2818 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
2819 wait $PID2 ; [ $? -eq 0 ] || error "stat must succeed"
2823 run_test 44g "pdirops: rename tgt vs getattr =============="
2827 touch $DIR1/$tfile-2
2828 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2829 do_nodes $(comma_list $(mdts_nodes)) \
2830 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2831 mv $DIR1/$tfile-2 $DIR1/$tfile &
2833 ls -lia $DIR2/ > /dev/null &
2835 do_nodes $(comma_list $(mdts_nodes)) \
2836 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2837 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
2842 run_test 44h "pdirops: rename tgt vs readdir =============="
2844 # test 44: rename tgt and blocking operations
2846 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2848 touch $DIR1/$tfile-2
2849 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2850 do_nodes $(comma_list $(mdts_nodes)) \
2851 "lctl set_param -n fail_loc=0x80000146 2>/dev/null || true"
2852 mv $DIR1/$tfile-2 $DIR1/$tfile &
2854 $LFS mkdir -i 1 $DIR2/$tfile &
2856 do_nodes $(comma_list $(mdts_nodes)) \
2857 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2858 check_pdo_conflict $PID1 && { wait $PID1;
2859 error "remote mkdir isn't blocked"; }
2860 wait $PID2 ; [ $? -ne 0 ] || error "remote mkdir must fail"
2864 run_test 44i "pdirops: rename tgt vs remote mkdir"
2866 # test 45: rename,mkdir doesn't fail with -EEXIST
2868 for i in {1..1000}; do
2869 mkdir $DIR1/$tdir || error "mkdir $tdir failed"
2870 mrename $DIR2/$tdir $DIR2/$tdir.$i > /dev/null ||
2871 error "mrename to $tdir.$i failed"
2876 run_test 45a "rename,mkdir doesn't return -EEXIST =============="
2881 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2882 do_nodes $(comma_list $(mdts_nodes)) \
2883 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2884 mv $DIR1/$tfile $DIR1/$tfile-2 &
2886 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c &
2888 do_nodes $(comma_list $(mdts_nodes)) \
2889 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2890 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
2891 wait $PID2 ; [ $? -eq 0 ] || error "create must succeed"
2895 run_test 45b "pdirops: rename src vs create =============="
2900 touch $DIR1/$tfile-3
2901 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2902 do_nodes $(comma_list $(mdts_nodes)) \
2903 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2904 mv $DIR1/$tfile $DIR1/$tfile-2 &
2906 link $DIR2/$tfile-3 $DIR2/$tfile &
2908 do_nodes $(comma_list $(mdts_nodes)) \
2909 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2910 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2911 wait $PID2 ; [ $? -eq 0 ] || error "link must succeed"
2915 run_test 45c "pdirops: rename src vs link =============="
2920 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2921 do_nodes $(comma_list $(mdts_nodes)) \
2922 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2923 mv $DIR1/$tfile $DIR1/$tfile-2 &
2927 do_nodes $(comma_list $(mdts_nodes)) \
2928 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2929 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
2930 wait $PID2 ; [ $? -ne 0 ] || error "unlink must fail"
2934 run_test 45d "pdirops: rename src vs unlink =============="
2939 touch $DIR1/$tfile-3
2940 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2941 do_nodes $(comma_list $(mdts_nodes)) \
2942 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2943 mv $DIR1/$tfile $DIR1/$tfile-2 &
2945 mv $DIR2/$tfile-3 $DIR2/$tfile &
2947 do_nodes $(comma_list $(mdts_nodes)) \
2948 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2949 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2950 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
2954 run_test 45e "pdirops: rename src and rename (tgt) =============="
2959 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2960 do_nodes $(comma_list $(mdts_nodes)) \
2961 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2962 mv $DIR1/$tfile $DIR1/$tfile-2 &
2964 mv $DIR2/$tfile $DIR2/$tfile-3 &
2966 do_nodes $(comma_list $(mdts_nodes)) \
2967 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2968 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2969 wait $PID2 ; [ $? -ne 0 ] || error "rename must fail"
2973 run_test 45f "pdirops: rename src and rename (src) =============="
2978 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2979 do_nodes $(comma_list $(mdts_nodes)) \
2980 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
2981 mv $DIR1/$tfile $DIR1/$tfile-2 &
2983 stat $DIR2/$tfile > /dev/null &
2985 do_nodes $(comma_list $(mdts_nodes)) \
2986 "lctl set_param -n fail_loc=0 2>/dev/null || true"
2987 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
2988 wait $PID2 ; [ $? -ne 0 ] || error "stat must fail"
2992 run_test 45g "pdirops: rename src vs getattr =============="
2997 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2998 do_nodes $(comma_list $(mdts_nodes)) \
2999 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3000 mv $DIR1/$tfile $DIR1/$tfile-2 &
3002 ls -lia $DIR2/ > /dev/null &
3003 do_nodes $(comma_list $(mdts_nodes)) \
3004 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3005 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
3010 run_test 45h "pdirops: unlink vs readdir =============="
3013 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3016 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3017 do_nodes $(comma_list $(mdts_nodes)) \
3018 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3019 mv $DIR1/$tfile $DIR1/$tfile-2 &
3021 $LFS mkdir -i 1 $DIR2/$tfile &
3023 do_nodes $(comma_list $(mdts_nodes)) \
3024 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3025 check_pdo_conflict $PID1 && { wait $PID1;
3026 error "create remote dir isn't blocked"; }
3027 wait $PID2 ; [ $? -eq 0 ] || error "create remote dir must succeed"
3031 run_test 45i "pdirops: rename src vs remote mkdir"
3038 # We test in a sparate directory to be able to unblock server thread in
3039 # cfs_race if LCK_PW is taken on the parent by mdt_reint_rename.
3040 test_mkdir $DIR2/$tdir
3041 echo file1 > $DIR2/$tdir/$tfile
3042 echo file2 > $DIR2/$tdir/$tfile-2
3045 do_nodes $(comma_list $(mdts_nodes)) \
3046 "lctl set_param -n fail_loc=${fail_loc} || true" &>/dev/null
3048 cat $DIR1/$tdir/$tfile >/dev/null &
3051 mrename $DIR2/$tdir/$tfile-2 $DIR2/$tdir/$tfile > /dev/null &
3055 { ret=$?; echo -n "cat $tfile should succeed (err=$ret); "; }
3058 echo -n "mrename $tfile-2 to $tfile failed (err=$ret);"; }
3061 do_nodes $(comma_list $(mdts_nodes)) \
3062 "lctl set_param -n fail_loc=0x0 || true" &>/dev/null
3069 (( $MDS1_VERSION >= $(version_code 2.13.56) )) ||
3070 skip "Need MDS version newer than 2.13.56"
3073 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN 0x169
3074 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_REINT_OPEN2 0x16a
3075 for fail_loc in "0x80000169" "0x8000016a"; do
3076 echo "Begin 100 tests with fail_loc=$fail_loc"
3078 for i in {1..100}; do
3080 msg=$(sub_test_45j "$fail_loc") ||
3081 { echo; error "iter=$i : $msg"; }
3086 run_test 45j "read vs rename =============="
3088 # test 46: link and blocking operations
3091 touch $DIR1/$tfile-2
3092 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3093 do_nodes $(comma_list $(mdts_nodes)) \
3094 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3095 link $DIR1/$tfile-2 $DIR1/$tfile &
3097 mkdir $DIR2/$tfile &
3099 do_nodes $(comma_list $(mdts_nodes)) \
3100 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3101 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
3102 wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail"
3106 run_test 46a "pdirops: link vs mkdir =============="
3110 touch $DIR1/$tfile-2
3111 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3112 do_nodes $(comma_list $(mdts_nodes)) \
3113 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3114 link $DIR1/$tfile-2 $DIR1/$tfile &
3116 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c &
3118 do_nodes $(comma_list $(mdts_nodes)) \
3119 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3120 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
3121 wait $PID2 ; [ $? -ne 0 ] || error "create must fail"
3125 run_test 46b "pdirops: link vs create =============="
3129 touch $DIR1/$tfile-2
3130 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3131 do_nodes $(comma_list $(mdts_nodes)) \
3132 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3133 link $DIR1/$tfile-2 $DIR1/$tfile &
3135 link $DIR2/$tfile $DIR2/$tfile &
3137 do_nodes $(comma_list $(mdts_nodes)) \
3138 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3139 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
3140 wait $PID2 ; [ $? -ne 0 ] || error "link must fail"
3144 run_test 46c "pdirops: link vs link =============="
3148 touch $DIR1/$tfile-2
3149 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3150 do_nodes $(comma_list $(mdts_nodes)) \
3151 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3152 link $DIR1/$tfile-2 $DIR1/$tfile &
3156 do_nodes $(comma_list $(mdts_nodes)) \
3157 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3158 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
3159 wait $PID2 ; [ $? -eq 0 ] || error "unlink must succeed"
3163 run_test 46d "pdirops: link vs unlink =============="
3167 touch $DIR1/$tfile-2
3168 touch $DIR1/$tfile-3
3169 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3170 do_nodes $(comma_list $(mdts_nodes)) \
3171 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3172 link $DIR1/$tfile-2 $DIR1/$tfile &
3174 mv $DIR2/$tfile-3 $DIR2/$tfile &
3176 do_nodes $(comma_list $(mdts_nodes)) \
3177 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3178 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
3179 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
3183 run_test 46e "pdirops: link and rename (tgt) =============="
3187 touch $DIR1/$tfile-2
3188 touch $DIR1/$tfile-3
3189 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3190 do_nodes $(comma_list $(mdts_nodes)) \
3191 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3192 link $DIR1/$tfile-2 $DIR1/$tfile &
3194 mv $DIR2/$tfile $DIR2/$tfile-3 &
3196 do_nodes $(comma_list $(mdts_nodes)) \
3197 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3198 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
3199 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
3203 run_test 46f "pdirops: link and rename (src) =============="
3207 touch $DIR1/$tfile-2
3208 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3209 do_nodes $(comma_list $(mdts_nodes)) \
3210 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3211 link $DIR1/$tfile-2 $DIR1/$tfile &
3213 stat $DIR2/$tfile > /dev/null &
3215 do_nodes $(comma_list $(mdts_nodes)) \
3216 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3217 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
3218 wait $PID2 ; [ $? -eq 0 ] || error "stat must succeed"
3222 run_test 46g "pdirops: link vs getattr =============="
3226 touch $DIR1/$tfile-2
3227 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3228 do_nodes $(comma_list $(mdts_nodes)) \
3229 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3230 link $DIR1/$tfile-2 $DIR1/$tfile &
3232 ls -lia $DIR2/ > /dev/null &
3234 do_nodes $(comma_list $(mdts_nodes)) \
3235 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3236 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
3241 run_test 46h "pdirops: link vs readdir =============="
3244 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3246 touch $DIR1/$tfile-2
3247 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3248 do_nodes $(comma_list $(mdts_nodes)) \
3249 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3250 link $DIR1/$tfile-2 $DIR1/$tfile &
3252 $LFS mkdir -i 1 $DIR2/$tfile &
3254 do_nodes $(comma_list $(mdts_nodes)) \
3255 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3256 check_pdo_conflict $PID1 && { wait $PID1;
3257 error "remote mkdir isn't blocked"; }
3258 wait $PID2 ; [ $? -ne 0 ] || error "remote mkdir must fail"
3262 run_test 46i "pdirops: link vs remote mkdir"
3264 # test 47: remote mkdir and blocking operations
3266 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3267 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3269 do_nodes $(comma_list $(mdts_nodes)) \
3270 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3271 $LFS mkdir -i 1 $DIR1/$tfile &
3273 mkdir $DIR2/$tfile &
3275 do_nodes $(comma_list $(mdts_nodes)) \
3276 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3277 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
3278 wait $PID2 ; [ $? -ne 0 ] || error "mkdir must fail"
3282 run_test 47a "pdirops: remote mkdir vs mkdir"
3285 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3286 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3288 do_nodes $(comma_list $(mdts_nodes)) \
3289 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3290 $LFS mkdir -i 1 $DIR1/$tfile &
3292 sleep 1 # please do not remove this sleep, see LU-10754
3293 multiop $DIR2/$tfile oO_CREAT:O_EXCL:c &
3295 do_nodes $(comma_list $(mdts_nodes)) \
3296 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3297 check_pdo_conflict $PID1 && { wait $PID1;
3298 error "create isn't blocked"; }
3299 wait $PID2 ; [ $? -ne 0 ] || error "create must fail"
3303 run_test 47b "pdirops: remote mkdir vs create"
3306 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3308 touch $DIR1/$tfile-2
3309 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3310 do_nodes $(comma_list $(mdts_nodes)) \
3311 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3312 $LFS mkdir -i 1 $DIR1/$tfile &
3314 link $DIR2/$tfile-2 $DIR2/$tfile &
3316 do_nodes $(comma_list $(mdts_nodes)) \
3317 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3318 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
3319 wait $PID2 ; [ $? -ne 0 ] || error "link must fail"
3323 run_test 47c "pdirops: remote mkdir vs link"
3326 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3328 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3329 do_nodes $(comma_list $(mdts_nodes)) \
3330 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3331 $LFS mkdir -i 1 $DIR1/$tfile &
3333 rmdir $DIR2/$tfile &
3335 do_nodes $(comma_list $(mdts_nodes)) \
3336 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3337 check_pdo_conflict $PID1 && { wait $PID1;
3338 error "unlink isn't blocked"; }
3339 wait $PID2 ; [ $? -eq 0 ] || error "rmdir must succeed"
3343 run_test 47d "pdirops: remote mkdir vs unlink"
3346 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3348 touch $DIR1/$tfile-2
3349 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3350 do_nodes $(comma_list $(mdts_nodes)) \
3351 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3352 $LFS mkdir -i 1 $DIR1/$tfile &
3354 mv -T $DIR2/$tfile-2 $DIR2/$tfile &
3356 do_nodes $(comma_list $(mdts_nodes)) \
3357 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3358 check_pdo_conflict $PID1 && { wait $PID1;
3359 error "rename isn't blocked"; }
3360 wait $PID2 ; [ $? -ne 0 ] || error "rename must fail"
3364 run_test 47e "pdirops: remote mkdir and rename (tgt)"
3367 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3369 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3370 do_nodes $(comma_list $(mdts_nodes)) \
3371 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3372 $LFS mkdir -i 1 $DIR1/$tfile &
3374 mv $DIR2/$tfile $DIR2/$tfile-2 &
3376 do_nodes $(comma_list $(mdts_nodes)) \
3377 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3378 check_pdo_conflict $PID1 && { wait $PID1;
3379 error "rename isn't blocked"; }
3380 wait $PID2 ; [ $? -eq 0 ] || error "rename must succeed"
3384 run_test 47f "pdirops: remote mkdir and rename (src)"
3387 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3391 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
3392 do_nodes $(comma_list $(mdts_nodes)) \
3393 "lctl set_param -n fail_loc=0x80000145 2>/dev/null || true"
3394 $LFS mkdir -i 1 $DIR1/$tfile &
3396 stat $DIR2/$tfile > /dev/null &
3398 do_nodes $(comma_list $(mdts_nodes)) \
3399 "lctl set_param -n fail_loc=0 2>/dev/null || true"
3400 check_pdo_conflict $PID1 && { wait $PID1;
3401 error "getattr isn't blocked"; }
3402 wait $PID2 ; [ $? -eq 0 ] || error "stat must succeed"
3406 run_test 47g "pdirops: remote mkdir vs getattr"
3410 dd if=/dev/zero of=$DIR1/$tfile bs=1K count=10
3411 #define OBD_FAIL_OSC_CP_ENQ_RACE 0x410
3412 do_facet client "lctl set_param fail_loc=0x410"
3413 $TRUNCATE $DIR2/$tfile $trunc_size
3414 do_facet client "lctl set_param fail_loc=0x0"
3416 size=`stat -c %s $DIR2/$tfile`
3417 [ $size -eq $trunc_size ] || error "wrong size"
3419 run_test 50 "osc lvb attrs: enqueue vs. CP AST =============="
3423 local origfile=/etc/hosts
3425 filesize=$(stat -c %s $origfile)
3427 # create an empty file
3428 $MCREATE $DIR1/$tfile || error "can't create $DIR1/$tfile"
3429 # cache layout lock on both mount point
3430 stat $DIR1/$tfile > /dev/null || error "stat $DIR1/$tfile failed"
3431 stat $DIR2/$tfile > /dev/null || error "stat $DIR2/$tfile failed"
3433 # open and sleep 2 seconds then read
3434 $MULTIOP $DIR2/$tfile o_2r${filesize}c &
3438 # create the layout of testing file
3439 dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null ||
3440 error "dd $DIR1/$tfile failed"
3442 # MULTIOP proc should be able to read enough bytes and exit
3443 for ((i = 0; i < 6; i++)); do
3445 kill -0 $pid || break
3447 kill -0 $pid 2> /dev/null && error "multiop is still there"
3448 cmp $origfile $DIR2/$tfile || error "$origfile and $DIR2/$tfile differs"
3452 run_test 51a "layout lock: refresh layout should work"
3455 (( $MDS1_VERSION >= $(version_code 2.3.59) )) ||
3456 skip "Need MDS version at least 2.3.59"
3458 local tmpfile=`mktemp`
3460 $LFS setstripe -E 1m -S 1M -c 1 -E -1 -c 1 $DIR1/$tfile ||
3461 error "Create $DIR1/$tfile failed"
3463 dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc ||
3464 error "dd $DIR1/$tfile failed"
3466 # delay glimpse so that layout has changed when glimpse finish
3467 #define OBD_FAIL_GLIMPSE_DELAY 0x1404
3468 $LCTL set_param fail_loc=0x1404 fail_val=4
3469 stat -c %s $DIR2/$tfile |tee $tmpfile &
3473 # extend layout of testing file
3474 dd if=/dev/zero of=$DIR1/$tfile bs=1M count=1 seek=2 conv=notrunc ||
3475 error "dd $DIR1/$tfile failed"
3478 local fsize=$(cat $tmpfile)
3480 [ x$fsize = x3145728 ] || error "file size is $fsize, should be 3145728"
3482 rm -f $DIR1/$tfile $tmpfile
3484 run_test 51b "layout lock: glimpse should be able to restart if layout changed"
3487 [ $OSTCOUNT -ge 2 ] || { skip "needs >= 2 osts"; return; }
3489 # set default layout to have 1 stripe
3491 $LFS setstripe -c 1 $DIR1/$tdir
3493 # create a file with empty layout
3494 $MCREATE $DIR1/$tdir/$tfile ||
3495 error "$MCREATE $DIR1/$tdir/$tfile failed"
3497 #define OBD_FAIL_MDS_LL_BLOCK 0x172
3498 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x172
3500 # change the layout of testing file
3501 echo "Setting layout to have $OSTCOUNT stripes ..."
3502 $LFS setstripe -c $OSTCOUNT $DIR1/$tdir/$tfile &
3506 # write something to the file, it should be blocked on fetching layout
3507 dd if=/dev/zero of=$DIR2/$tdir/$tfile bs=1k count=1 conv=notrunc
3508 local stripe_count=$($LFS getstripe -c $DIR2/$tdir/$tfile)
3511 # lod_qos.c::min_stripe_count() allows setstripe with a default stripe
3512 # count to succeed with only 3/4 of the number of stripes (rounded up),
3513 # so creating striped files does not fail if an OST is offline or full
3514 [ $stripe_count -ge $((OSTCOUNT - $OSTCOUNT / 4)) ] ||
3515 error "bad layout: getstripe -c $stripe_count < $OSTCOUNT * 3/4"
3519 run_test 51c "layout lock: IT_LAYOUT blocked and correct layout can be returned"
3522 dd if=/dev/zero of=/$DIR1/$tfile bs=1M count=1
3523 cancel_lru_locks mdc
3525 # open should grant LAYOUT lock, mmap and read will install pages
3526 $MULTIOP $DIR1/$tfile oO_RDWR:SMR_Uc &
3530 # rss before revoking
3531 local br=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}')
3532 echo "Before revoking layout lock: $br KB mapped"
3534 # cancel layout lock manually
3535 cancel_lru_locks mdc
3537 # rss after revoking
3538 local ar=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}')
3541 wait $PID || error "wait PID $PID failed"
3543 [ $ar -eq 0 ] || error "rss before: $br, after $ar, some pages remained"
3545 run_test 51d "layout lock: losing layout lock should clean up memory map region"
3548 (( $MDS1_VERSION >= $(version_code 2.13.54.148) )) ||
3549 skip "MDS version must be at least 2.13.54.148"
3553 $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:eW_E+eUc &
3557 $LFS getstripe $DIR2/$tfile
3559 wait $pid || error "multiop failed"
3561 $MULTIOP $DIR/$tfile oO_RDONLY:eR_E+eUc &
3565 $LFS getstripe $DIR2/$tfile
3567 wait $pid || error "multiop failed"
3569 run_test 51e "lfs getstripe does not break leases, part 2"
3573 echo "==> rename vs getattr vs setxattr should not deadlock"
3574 mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed"
3576 do_facet mds1 $LCTL set_param fail_loc=$1
3578 mv -T $DIR/d1/d2/d3 $DIR/d1/d3 &
3586 setfattr -n user.attr1 -v value1 $DIR2/d1 || error "(2) setfattr failed"
3587 wait $PID1 || error "(3) mv failed"
3588 wait $PID2 || error "(4) stat failed"
3595 echo "==> rename vs getattr vs open vs getattr should not deadlock"
3596 mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed"
3598 do_facet mds1 $LCTL set_param fail_loc=$1
3600 mv -T $DIR/d1/d2/d3 $DIR/d1/d3 &
3608 $MULTIOP $DIR2/d1/d2 Oc &
3612 stat $DIR/d1 || error "(2) stat failed"
3614 wait $PID1 || error "(3) mv failed"
3615 wait $PID2 || error "(4) stat failed"
3616 wait $PID3 && error "(5) multiop failed"
3622 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
3623 save_lustre_params client "llite.*.xattr_cache" > $p
3624 lctl set_param llite.*.xattr_cache 1 ||
3625 { skip "xattr cache is not supported"; return 0; }
3627 #define OBD_FAIL_MDS_RENAME 0x153
3628 #define OBD_FAIL_MDS_RENAME2 0x154
3629 test_54_part1 0x80000153 || error 10
3630 test_54_part1 0x80000154 || error 11
3631 test_54_part2 0x80000153 || error 12
3632 test_54_part2 0x80000154 || error 13
3634 restore_lustre_params < $p
3637 run_test 54 "rename locking"
3640 mkdir_on_mdt0 $DIR/$tdir
3641 mkdir -p $DIR/$tdir/d1/d2 $DIR/$tdir/d3 || error "(1) mkdir failed"
3643 #define OBD_FAIL_MDS_RENAME4 0x156
3644 do_facet mds1 $LCTL set_param fail_loc=0x80000156
3646 mv -T $DIR/$tdir/d1/d2 $DIR/$tdir/d3/d2 &
3650 rm -r $DIR2/$tdir/d3
3651 wait $PID1 && error "(2) mv succeeded"
3655 run_test 55a "rename vs unlink target dir"
3659 mkdir_on_mdt0 $DIR/$tdir
3660 mkdir -p $DIR/$tdir/d1/d2 $DIR/$tdir/d3 || error "(1) mkdir failed"
3662 #define OBD_FAIL_MDS_RENAME4 0x156
3663 do_facet mds1 $LCTL set_param fail_loc=0x80000156
3665 mv -T $DIR/$tdir/d1/d2 $DIR/$tdir/d3/d2 &
3669 rm -r $DIR2/$tdir/d1
3670 wait $PID1 && error "(2) mv succeeded"
3674 run_test 55b "rename vs unlink source dir"
3678 mkdir_on_mdt0 $DIR/$tdir
3679 mkdir -p $DIR/$tdir/d1/d2 $DIR/$tdir/d3 || error "(1) mkdir failed"
3681 #define OBD_FAIL_MDS_RENAME4 0x156
3682 do_facet mds1 $LCTL set_param fail_loc=0x156
3684 mv -T $DIR/$tdir/d1/d2 $DIR/$tdir/d3/d2 &
3688 # while rename is sleeping, open and remove d3
3689 $MULTIOP $DIR2/$tdir/d3 D_c &
3692 rm -rf $DIR2/$tdir/d3
3695 # while rename is sleeping 2nd time, close d3
3697 wait $PID2 || error "(3) multiop failed"
3699 wait $PID1 && error "(2) mv succeeded"
3703 run_test 55c "rename vs unlink orphan target dir"
3707 mkdir_on_mdt0 $DIR/$tdir
3711 #define OBD_FAIL_MDS_RENAME3 0x155
3712 do_facet mds1 $LCTL set_param fail_loc=0x155
3713 mv $DIR/$tdir/f1 $DIR/$tdir/$tdir &
3717 # while rename is sleeping, create $tdir, but as a directory
3718 mkdir -p $DIR2/$tdir/$tdir || error "(1) mkdir failed"
3720 # link in reverse locking order
3721 ln $DIR2/$tdir/f1 $DIR2/$tdir/$tdir/f1 || error "(2) ln failed"
3723 ! wait $PID1 || error "(3) mv succeeded"
3726 run_test 55d "rename file vs link"
3729 $LFS setstripe -c 1 $MOUNT/$tfile || error "creating $MOUNT/$tfile"
3730 stack_trap "rm -f $MOUNT/$tfile"
3731 echo "run_llverdev $MOUNT/$tfile -p -s $((16000000)) -c 4k"
3732 run_llverdev $MOUNT/$tfile -p -s $((16000000)) -c 4k ||
3733 error "llverdev failed with rc=$?"
3735 run_test 56a "test llverdev with single large stripe"
3738 $LFS setstripe -C 2000 $MOUNT/$tfile || error "creating $MOUNT/$tfile"
3739 stack_trap "rm -f $MOUNT/$tfile"
3740 echo "run_llverdev $MOUNT/$tfile -p -s $((16000000 * OSTCOUNT)) -c 4k"
3741 run_llverdev $MOUNT/$tfile -p -s $((16000000 * OSTCOUNT)) -c 4k ||
3742 error "llverdev failed with rc=$?"
3744 run_test 56b "test llverdev and partial verify of wide stripe file"
3747 [ $MDS1_VERSION -lt $(version_code 2.3.0) ] &&
3748 skip "MDS version must be >= 2.3.0"
3751 test_mkdir $DIR1/$tdir
3752 file1=$DIR1/$tdir/file
3753 file2=$DIR2/$tdir/file
3755 echo orig > $file2 || error "Could not create $file2"
3756 version=$($LFS data_version $file1)
3759 echo append >> $file2 || error "Could not append to $file2"
3760 version2=$($LFS data_version $file1)
3761 [ "$version" != "$version2" ] ||
3762 error "append did not change data version: $version"
3765 echo overwrite > $file2 || error "Could not overwrite $file2"
3766 version3=$($LFS data_version $file1)
3767 [ "$version2" != "$version3" ] ||
3768 error "overwrite did not change data version: $version2"
3770 # Truncate before EOF
3771 $TRUNCATE $file2 3 || error "Could not truncate $file2"
3772 version4=$($LFS data_version $file1)
3773 [ "$version3" != "$version4" ] ||
3774 error "truncate did not change data version: $version3"
3776 # Truncate after EOF
3777 $TRUNCATE $file2 123456 || error "Could not truncate $file2"
3778 version5=$($LFS data_version $file1)
3779 [ "$version4" != "$version5" ] ||
3780 error "truncate did not change data version: $version4"
3782 # Chmod do not change version
3783 chmod 400 $file2 || error "Could not chmod 400 $file2"
3784 version6=$($LFS data_version $file1)
3785 [ "$version5" == "$version6" ] ||
3786 error "chmod should not change data version: $version5 != $version6"
3788 # Chown do not change version
3789 chown $RUNAS_ID $file2 || error "Could not chown $RUNAS_ID $file2"
3790 version7=$($LFS data_version $file1)
3791 [ "$version5" == "$version7" ] ||
3792 error "chown should not change data version: $version5 != $version7"
3794 run_test 60 "Verify data_version behaviour"
3797 local test_dir=$tdir/test_dir
3799 mkdir -p $DIR1/$tdir
3800 if [ $MDSCOUNT -ge 2 ]; then
3802 $LFS mkdir -i $MDTIDX $DIR1/$test_dir ||
3803 error "Create remote directory failed"
3805 mkdir -p $DIR1/$test_dir
3807 cd $DIR2/$test_dir || error "cd directory failed"
3808 rm -rf $DIR1/$test_dir || error "unlink directory failed"
3810 cd $DIR2/$tdir || error "exit directory"
3812 run_test 70a "cd directory && rm directory"
3814 test_70b() { # LU-2781
3816 mkdir -p $DIR1/$tdir
3818 touch $DIR1/$tdir/file
3819 for ((i = 0; i < 32; i++)); do
3820 $LFS rm_entry $DIR1/$tdir/non_existent_dir &>/dev/null
3822 rm $DIR1/$tdir/file || error "cannot remove file after rm_entry"
3824 touch $DIR1/$tdir/file
3825 $LFS mkdir -i0 $DIR1/$tdir/test_dir
3826 $LFS rm_entry $DIR1/$tdir/test_dir &>/dev/null
3827 rm -rf $DIR1/$tdir/test_dir ||
3828 error "cannot remove directory after rm_entry"
3829 rm $DIR1/$tdir/file || error "cannot remove file after rm_entry"
3831 run_test 70b "remove files after calling rm_entry"
3834 [[ "$MDS1_VERSION" -lt $(version_code 2.1.6) ]] &&
3835 skip "Need MDS version at least 2.1.6"
3837 # Patch not applied to 2.2 and 2.3 branches
3838 [[ "$MDS1_VERSION" -ge $(version_code 2.2.0) ]] &&
3839 [[ "$MDS1_VERSION" -lt $(version_code 2.4.0) ]] &&
3840 skip "Need MDS version earlier than 2.2.0 or at least 2.4.0"
3842 checkfiemap --test ||
3843 skip "checkfiemap not runnable: $?"
3844 # write data this way: hole - data - hole - data
3845 dd if=/dev/urandom of=$DIR1/$tfile bs=64K seek=1 count=1
3846 [ "$(facet_fstype ost$(($($LFS getstripe -i $DIR1/$tfile) + 1)))" = \
3848 skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
3849 dd if=/dev/urandom of=$DIR1/$tfile bs=64K seek=3 count=1
3850 GET_STAT="lctl get_param -n ldlm.services.ldlm_cbd.stats"
3852 local can1=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
3854 checkfiemap $DIR2/$tfile 131072 ||
3855 error "data is not flushed from client"
3856 local can2=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
3859 # common case of "create file, copy file" on a single node
3860 # should not flush data from ost
3861 dd if=/dev/urandom of=$DIR1/$tfile bs=64K seek=1 count=1
3862 dd if=/dev/urandom of=$DIR1/$tfile bs=64K seek=3 count=1
3864 local can3=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
3866 checkfiemap $DIR1/$tfile 131072 ||
3868 local can4=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
3870 [ $can3 -eq $can4 ] || error $((can2-can1)) "cancel RPC occured."
3872 run_test 71a "correct file map just after write operation is finished"
3875 [[ "$MDS1_VERSION" -lt $(version_code 2.1.6) ]] &&
3876 skip "Need MDS version at least 2.1.6"
3878 # Patch not applied to 2.2 and 2.3 branches
3879 [[ "$MDS1_VERSION" -ge $(version_code 2.2.0) ]] &&
3880 [[ "$MDS1_VERSION" -lt $(version_code 2.4.0) ]] &&
3881 skip "Need MDS version earlier than 2.2.0 or at least 2.4.0"
3882 [[ $OSTCOUNT -ge 2 ]] || skip "needs >= 2 OSTs"
3884 checkfiemap --test ||
3885 skip "error $?: checkfiemap failed"
3887 mkdir -p $DIR1/$tdir
3889 $LFS setstripe -c -1 $DIR1/$tdir || error "setstripe failed"
3890 dd if=/dev/urandom of=$DIR1/$tdir/$tfile bs=64K count=1
3891 [ "$(facet_fstype ost$(($($LFS getstripe -i $DIR1/$tdir/$tfile) + 1)))" = \
3893 skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
3894 checkfiemap $DIR1/$tdir/$tfile 65536 || error "checkfiemap failed"
3896 run_test 71b "check fiemap support for stripecount > 1"
3898 _check_last_flag_with_filefrag()
3905 echo "check last flag for file with $count extents"
3907 for ((i=0; i<$count; i++)); do
3909 dd if=/dev/zero of=$file bs=4K count=1 seek=$offset 2> /dev/null
3912 filefrag -s -v $file | grep "last" ||
3913 error "test file with $i extents failed"
3919 local file="$DIR1/$tdir/$tfile"
3921 (( $CLIENT_VERSION >= $(version_code 2.15.57) )) ||
3922 skip "Need client version >= 2.15.57"
3923 [ $(facet_fstype ost1) = "ldiskfs" ] ||
3924 skip "support only ldiskfs ost"
3925 filefrag -V | grep wc ||
3926 skip "need whamcloud version of e2fsprogs"
3928 mkdir -p $DIR1/$tdir
3930 # filefrag uses u64[2028] buffer to fetch fiemap. The number of extents
3931 # in the buffer is (8 * 2048 - 32) / 56 = 292. Test file with 291, 292
3933 _check_last_flag_with_filefrag $file 291
3934 _check_last_flag_with_filefrag $file 292
3935 _check_last_flag_with_filefrag $file 293
3937 run_test 71c "check FIEMAP_EXTENT_LAST flag with different extents number"
3939 test_71d() { #LU-17110
3940 checkfiemap --test ||
3941 skip "error $?: checkfiemap failed"
3945 # write data this way: hole - data - hole - data
3946 dd if=/dev/urandom of=$f bs=64K count=1
3947 [[ "$(facet_fstype ost$(($($LFS getstripe -i $f) + 1)))" != "zfs" ]] ||
3948 skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS"
3949 dd if=/dev/urandom of=$f bs=64K seek=2 count=1
3950 dd if=/dev/urandom of=$f bs=64K seek=4 count=1
3951 dd if=/dev/urandom of=$f bs=64K seek=6 count=1 conv=fsync
3952 echo "disk usage: $(du -B1 $f)"
3953 echo "file size: $(du -b $f)"
3955 checkfiemap --corruption_test $f $((4 * 64 *1024)) ||
3956 error "checkfiemap failed"
3958 run_test 71d "fiemap corruption test with fm_extent_count=0"
3961 local p="$TMP/sanityN-$TESTNAME.parameters"
3964 save_lustre_params client "llite.*.xattr_cache" > $p
3965 lctl set_param llite.*.xattr_cache 1 ||
3966 { skip "xattr cache is not supported"; return 0; }
3969 setfattr -n user.attr1 -v value1 $DIR1/$tfile ||
3970 error "setfattr1 failed"
3971 getfattr -n user.attr1 $DIR2/$tfile | grep value1 ||
3972 error "getfattr1 failed"
3973 setfattr -n user.attr1 -v value2 $DIR2/$tfile ||
3974 error "setfattr2 failed"
3975 getfattr -n user.attr1 $DIR1/$tfile | grep value2 ||
3976 error "getfattr2 failed"
3978 # check that trusted.link is consistent
3979 tlink1=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
3980 ln $DIR2/$tfile $DIR2/$tfile-2 || error "failed to link"
3981 tlink2=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
3982 echo "$tlink1 $tlink2"
3983 [ "$tlink1" = "$tlink2" ] && error "trusted.link should have changed!"
3987 restore_lustre_params < $p
3990 run_test 72 "getxattr/setxattr cache should be consistent between nodes"
3993 local p="$TMP/sanityN-$TESTNAME.parameters"
3994 save_lustre_params client "llite.*.xattr_cache" > $p
3995 lctl set_param llite.*.xattr_cache 1 ||
3996 { skip "xattr cache is not supported"; return 0; }
3999 setfattr -n user.attr1 -v value1 $DIR1/$tfile ||
4000 error "setfattr1 failed"
4001 getfattr -n user.attr1 $DIR2/$tfile || error "getfattr1 failed"
4002 getfattr -n user.attr1 $DIR1/$tfile || error "getfattr2 failed"
4003 clear_stats llite.*.stats
4004 # PR lock should be cached by now on both clients
4005 getfattr -n user.attr1 $DIR1/$tfile || error "getfattr3 failed"
4006 # 2 hits for getfattr(0)+getfattr(size)
4007 [ $(calc_stats llite.*.stats getxattr_hits) -eq 2 ] ||
4008 error "not cached in $DIR1"
4009 getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed"
4010 # 4 hits for more getfattr(0)+getfattr(size)
4011 [ $(calc_stats llite.*.stats getxattr_hits) -eq 4 ] ||
4012 error "not cached in $DIR2"
4015 restore_lustre_params < $p
4018 run_test 73 "getxattr should not cause xattr lock cancellation"
4021 [ "$MDS1_VERSION" -lt $(version_code 2.4.93) ] &&
4022 skip "Need MDS version at least 2.4.93"
4024 dd if=/dev/zero of=$DIR1/$tfile-1 bs=1K count=1
4025 dd if=/dev/zero of=$DIR1/$tfile-2 bs=1K count=1
4026 flocks_test 4 $DIR1/$tfile-1 $DIR2/$tfile-2
4028 run_test 74 "flock deadlock: different mounts =============="
4032 $LFS setstripe -c 2 -S 1m -i 0 $DIR1/$tfile
4033 dd if=/dev/zero of=$DIR1/$tfile bs=1M count=2
4034 cancel_lru_locks osc
4036 dd of=$DIR1/$tfile if=/dev/zero bs=1M count=1 seek=1 conv=notrunc
4039 # define OBD_FAIL_LDLM_ENQUEUE_HANG 0x31d
4040 $LCTL set_param fail_loc=0x31d
4041 stat -c %s $DIR1/$tfile &
4046 # For bad lock error handler we should ASSERT and got kernel panic here
4048 $LCTL set_param fail_loc=0
4050 run_test 75 "osc: upcall after unuse lock==================="
4053 [[ "$MDS1_VERSION" -lt $(version_code 2.5.53) ]] &&
4054 skip "Need MDS version at least 2.5.53"
4056 remote_mds_nodsh && skip "remote MDS with nodsh"
4057 local fcount=$((MDSCOUNT * 256))
4062 nid=$($LCTL list_nids | sed "s/\./\\\./g")
4068 test_mkdir $DIR/$tdir
4070 # drop all open locks and close any cached "open" files on the client
4071 cancel_lru_locks mdc
4073 local open_fids_cmd="$LCTL get_param -n mdt.*.exports.'$nid'.open_files"
4074 local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $open_fids_cmd))
4075 local already=${#fid_list[@]}
4076 for (( i = 0; i < $already; i++ )) ; do
4077 log "already open[$i]: $($LFS fid2path $DIR2 ${fid_list[i]})"
4080 echo -n "opening files: "
4081 ulimit -n $((fcount + 50))
4082 for ((i = 0; i < $fcount; i++)); do
4083 touch $DIR/$tdir/f_$i
4084 local fd=$(free_fd ${fd_list[i]})
4085 local open_cmd="exec $fd<$DIR/$tdir/f_$i"
4090 (( $i % 32 == 0 )) && echo -n "."
4094 fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $open_fids_cmd))
4096 # Possible errors in openfiles FID list.
4097 # 1. Missing FIDs. Check 1
4098 # 2. Extra FIDs. Check 1
4099 # 3. Duplicated FID. Check 2
4100 # 4. Invalid FIDs. Check 2
4101 # 5. Valid FID, points to some other file. Check 3
4104 [ ${#fid_list[@]} -ne $((fcount + already)) ] &&
4105 error "${#fid_list[@]} != $fcount (+$already old) open files"
4107 echo -n "closing files: "
4108 for (( fd = 0, fid = 0; fd < $fcount; fd++, fid++ )) ; do
4109 local close_cmd="exec ${fd_list[fd]}<&-"
4111 filename=$($LFS fid2path $DIR2 ${fid_list[fid]})
4113 while [[ ! "$filename" =~ "$DIR2/$tdir/f_" ]]; do
4114 echo "skip old open file $filename"
4116 filename=$($LFS fid2path $DIR2 ${fid_list[fid]})
4120 rm --interactive=no $filename
4122 error "Nonexisting fid ${fid_list[fid]} listed."
4123 (( $fd % 32 == 0 )) && echo -n "."
4128 ls_op=$(ls $DIR2/$tdir | wc -l)
4130 error "Some openfiles are missing in lproc output"
4134 run_test 76 "Verify MDT open_files listing"
4138 local dir=$DIR/$tdir
4140 local create_as="$2"
4142 mkdir $dir || error "mkdir $dir failed"
4143 $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed"
4146 if [[ -n "$create_as" ]]; then
4147 do_nodes $CLIENTS $create_as "touch $dir/nrs_r_\$HOSTNAME;" ||
4148 error "touch failed for $dir/nrs_r_*"
4149 do_nodes $CLIENTS $create_as touch "$dir/nrs_w_\$HOSTNAME" ||
4150 error "touch failed for $dir/nrs_w_*"
4151 do_nodes $CLIENTS $create_as "chmod 777 $dir/nrs_*_\$HOSTNAME;" ||
4152 error "chmod failed for $dir/nrs_*"
4155 do_nodes $CLIENTS $myRUNAS \
4156 dd if=/dev/zero of="$dir/nrs_r_\$HOSTNAME" bs=1M count=$n ||
4157 error "dd at 0 on client failed (1)"
4159 do_nodes $CLIENTS $myRUNAS \
4161 for ((i = 0; i < $n; i++)); do
4162 dd if=/dev/zero of=$dir/nrs_w_\$HOSTNAME bs=1M \
4163 seek=\\\$i count=1 conv=notrunc &
4164 pids_w[\\\$i]=\\\$!;
4167 for ((i = 0; i < $n; i++)); do
4168 wait \\\${pids_w[\\\$i]};
4170 [ \\\$newrc -gt \\\$rc_w ] && rc_w=\\\$newrc;
4174 do_nodes $CLIENTS sync;
4175 cancel_lru_locks osc
4177 do_nodes $CLIENTS $myRUNAS \
4179 for ((i = 0; i < $n; i++)); do
4180 dd if=$dir/nrs_r_\$HOSTNAME bs=1M of=/dev/null \
4181 seek=\\\$i count=1 &
4182 pids_r[\\\$i]=\\\$!;
4185 for ((i = 0; i < $n; i++)); do
4186 wait \\\${pids_r[\\\$i]};
4188 [ \\\$newrc -gt \\\$rc_r ] && rc_r=\\\$newrc;
4192 cancel_lru_locks osc
4194 wait $pid_w || error "dd (write) failed (2)"
4195 wait $pid_r || error "dd (read) failed (3)"
4196 rm -rvf $dir || error "rm -rf $dir failed"
4199 test_77a() { #LU-3266
4202 oss=$(comma_list $(osts_nodes))
4203 do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="fifo" ||
4205 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
4206 [[ $rc -ne 0 ]] && error "failed to set fifo policy"
4211 run_test 77a "check FIFO NRS policy"
4213 test_77b() { #LU-3266
4216 oss=$(comma_list $(osts_nodes))
4218 do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="crrn" \
4219 ost.OSS.*.nrs_crrn_quantum=1 || rc=$?
4220 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
4221 [[ $rc -ne 0 ]] && error "failed to set crrn_quantum to 1"
4223 echo "policy: crr-n, crrn_quantum 1"
4226 do_nodes $oss lctl set_param \
4227 ost.OSS.*.nrs_crrn_quantum=64 || rc=$?
4228 [[ $rc -ne 0 ]] && error "failed to set crrn_quantum to 64"
4230 echo "policy: crr-n, crrn_quantum 64"
4234 do_nodes $oss lctl set_param \
4235 ost.OSS.ost_io.nrs_policies="fifo" || rc=$?
4236 [[ $rc -ne 0 ]] && error "failed to set fifo policy"
4239 run_test 77b "check CRR-N NRS policy"
4244 oss=$(comma_list $(osts_nodes))
4246 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies=$policy \
4247 ost.OSS.*.nrs_"$policy"_quantum=1 \
4248 ost.OSS.*.nrs_"$policy"_offset_type="physical" \
4249 ost.OSS.*.nrs_"$policy"_supported="reads" || return $?
4251 echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type " \
4252 "physical, ${policy}_supported reads"
4255 do_nodes $oss lctl set_param \
4256 ost.OSS.*.nrs_${policy}_supported="writes" \
4257 ost.OSS.*.nrs_${policy}_quantum=64 || return $?
4259 echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type " \
4260 "physical, ${policy}_supported writes"
4263 do_nodes $oss lctl set_param \
4264 ost.OSS.*.nrs_${policy}_supported="reads_and_writes" \
4265 ost.OSS.*.nrs_${policy}_offset_type="logical" || return $?
4266 echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type " \
4267 "logical, ${policy}_supported reads_and_writes"
4271 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" ||
4276 test_77c() { #LU-3266
4278 orr_trr "orr" || rc=$?
4279 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
4280 [[ $rc -ne 0 ]] && error "orr_trr failed rc:$rc"
4283 run_test 77c "check ORR NRS policy"
4285 test_77d() { #LU-3266
4287 orr_trr "trr" || rc=$?
4288 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
4289 [[ $rc -ne 0 ]] && error "orr_trr failed rc:$rc"
4292 run_test 77d "check TRR nrs policy"
4299 do_facet $facet lctl set_param \
4300 ost.OSS.ost_io.nrs_tbf_rule="$*"
4302 error "failed to run operate '$*' on TBF rules"
4305 cleanup_tbf_verify()
4309 echo "cleanup_tbf $DIR/$tdir"
4312 wait_delete_completed
4317 local dir=$DIR/$tdir
4318 local client1=${CLIENT1:-$(hostname)}
4320 local create_as="$4"
4322 local np=$(check_cpt_number ost1)
4323 [ $np -gt 0 ] || error "CPU partitions should not be $np."
4324 echo "cpu_npartitions on ost1 is $np"
4326 mkdir $dir || error "mkdir $dir failed"
4327 $LFS setstripe -c 1 -i 0 $dir || error "setstripe to $dir failed"
4330 if [[ -n "$create_as" ]]; then
4331 $create_as touch $dir/tbf
4335 trap cleanup_tbf_verify EXIT
4336 echo "Limited write rate: $1, read rate: $2"
4337 echo "Verify the write rate is under TBF control"
4338 local start=$SECONDS
4339 do_node $client1 $myRUNAS dd if=/dev/zero of=$dir/tbf \
4340 bs=1M count=100 oflag=direct 2>&1
4341 local runtime=$((SECONDS - start + 1))
4342 local rate=$(bc <<< "scale=6; 100 / $runtime")
4343 echo "Write runtime is $runtime s, speed is $rate IOPS"
4345 # verify the write rate does not exceed TBF rate limit
4346 [ $(bc <<< "$rate < 1.1 * $np * $1") -eq 1 ] ||
4347 error "The write rate ($rate) exceeds 110% of rate limit ($1 * $np)"
4349 cancel_lru_locks osc
4351 echo "Verify the read rate is under TBF control"
4353 do_node $client1 $myRUNAS dd if=$dir/tbf of=/dev/null \
4354 bs=1M count=100 iflag=direct 2>&1
4355 runtime=$((SECONDS - start + 1))
4356 rate=$(bc <<< "scale=6; 100 / $runtime")
4357 echo "Read runtime is $runtime s, speed is $rate IOPS"
4359 # verify the read rate does not exceed TBF rate limit
4360 [ $(bc <<< "$rate < 1.1 * $np * $2") -eq 1 ] ||
4361 error "The read rate ($rate) exceeds 110% of rate limit ($2 * $np)"
4363 cancel_lru_locks osc
4364 cleanup_tbf_verify || error "rm -rf $dir failed"
4370 oss=$(comma_list $(osts_nodes))
4372 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" ||
4374 [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
4375 [[ $rc -ne 0 ]] && error "failed to set TBF NID policy"
4379 if [ "$OST1_VERSION" -ge $(version_code 2.8.54) ]; then
4384 # Only operate rules on ost1 since OSTs might run on the same OSS
4386 tbf_rule_operate ost1 "start\ localhost\ ${idis}{0@lo}\ ${rateis}1000"
4387 local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)")
4388 local client_nids=$(nids_list $address "\\")
4389 tbf_rule_operate ost1 "start\ clients\ ${idis}{$client_nids}\ ${rateis}100"
4390 tbf_rule_operate ost1 "start\ others\ ${idis}{*.*.*.*@$NETTYPE}\ ${rateis}50"
4394 tbf_rule_operate ost1 "change\ localhost\ ${rateis}1001"
4395 tbf_rule_operate ost1 "change\ clients\ ${rateis}101"
4396 tbf_rule_operate ost1 "change\ others\ ${rateis}51"
4400 tbf_rule_operate ost1 "stop\ localhost"
4401 tbf_rule_operate ost1 "stop\ clients"
4402 tbf_rule_operate ost1 "stop\ others"
4405 # Cleanup the TBF policy
4406 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
4407 [ $? -ne 0 ] && error "failed to set policy back to fifo"
4411 run_test 77e "check TBF NID nrs policy"
4416 oss=$(comma_list $(osts_nodes))
4418 do_nodes $oss $LCTL set_param \
4419 ost.OSS.ost_io.nrs_policies="tbf\ jobid" || rc=$?
4420 [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
4421 [[ $rc -ne 0 ]] && error "failed to set TBF JOBID policy"
4423 # Configure jobid_var
4424 local saved_jobid_var=$($LCTL get_param -n jobid_var)
4426 [[ $rc -eq 3 ]] && skip "jobid_var not found" && return
4427 [[ $rc -ne 0 ]] && error "failed to get param jobid_var"
4428 if [ $saved_jobid_var != procname_uid ]; then
4429 set_persistent_param_and_check client \
4430 "jobid_var" "$FSNAME.sys.jobid_var" procname_uid
4435 if [ "$OST1_VERSION" -ge $(version_code 2.8.54) ]; then
4440 # Only operate rules on ost1 since OSTs might run on the same OSS
4442 tbf_rule_operate ost1 "start\ runas\ ${idis}{iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ ${rateis}1000"
4443 tbf_rule_operate ost1 "start\ iozone_runas\ ${idis}{iozone.$RUNAS_ID}\ ${rateis}100"
4444 tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50"
4445 nrs_write_read "$RUNAS"
4448 tbf_rule_operate ost1 "change\ runas\ ${rateis}1001"
4449 tbf_rule_operate ost1 "change\ iozone_runas\ ${rateis}101"
4450 tbf_rule_operate ost1 "change\ dd_runas\ ${rateis}51"
4451 nrs_write_read "$RUNAS"
4454 tbf_rule_operate ost1 "stop\ runas"
4455 tbf_rule_operate ost1 "stop\ iozone_runas"
4456 tbf_rule_operate ost1 "stop\ dd_runas"
4457 nrs_write_read "$RUNAS"
4459 # Cleanup the TBF policy
4460 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
4461 [ $? -ne 0 ] && error "failed to set policy back to fifo"
4462 nrs_write_read "$RUNAS"
4464 local current_jobid_var=$($LCTL get_param -n jobid_var)
4465 [[ $? -ne 0 ]] && error "failed to get param jobid_var"
4466 if [ $saved_jobid_var != $current_jobid_var ]; then
4467 set_persistent_param_and_check client \
4468 "jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid_var
4472 run_test 77f "check TBF JobID nrs policy"
4477 oss=$(comma_list $(osts_nodes))
4479 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" ||
4481 [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
4482 [[ $rc -ne 0 ]] && error "failed to set TBF NID policy"
4484 do_nodes $oss lctl set_param \
4485 ost.OSS.ost_io.nrs_policies="tbf\ jobid" || rc=$?
4486 [[ $rc -ne 0 ]] && error "failed to set TBF JOBID policy"
4490 if [ "$OST1_VERSION" -ge $(version_code 2.8.54) ]; then
4495 # Add a rule that only valid for Jobid TBF. If direct change between
4496 # TBF types is not supported, this operation will fail.
4497 tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50"
4499 # Cleanup the TBF policy
4500 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
4501 [ $? -ne 0 ] && error "failed to set policy back to fifo"
4504 run_test 77g "Change TBF type directly"
4507 [ "$OST1_VERSION" -ge $(version_code 2.8.55) ] ||
4508 skip "Need OST version at least 2.8.55"
4510 local old_policy=$(do_facet ost1 \
4511 lctl get_param ost.OSS.ost_io.nrs_policies)
4514 do_facet ost1 lctl set_param \
4515 ost.OSS.ost_io.nrs_policies="abc"
4516 [ $? -eq 0 ] && error "should return error"
4518 do_facet ost1 lctl set_param \
4519 ost.OSS.ost_io.nrs_policies="tbf\ abc"
4520 [ $? -eq 0 ] && error "should return error"
4522 do_facet ost1 lctl set_param \
4523 ost.OSS.ost_io.nrs_policies="tbf\ reg\ abc"
4524 [ $? -eq 0 ] && error "should return error"
4526 do_facet ost1 lctl set_param \
4527 ost.OSS.ost_io.nrs_policies="tbf\ abc\ efg"
4528 [ $? -eq 0 ] && error "should return error"
4530 new_policy=$(do_facet ost1 lctl get_param ost.OSS.ost_io.nrs_policies)
4531 [ $? -eq 0 ] || error "shouldn't LBUG"
4533 [ "$old_policy" = "$new_policy" ] || error "NRS policy should be same"
4537 run_test 77h "Wrong policy name should report error, not LBUG"
4543 local error_message=$3
4545 for rule in $expected; do
4546 rule_number=$((rule_number + 1))
4548 local stop_line=$(($rule_number + 3))
4549 local awk_command="awk 'NR >= 4 && NR <= $stop_line {print \$1}'"
4551 local output=$(do_facet $facet lctl get_param \
4552 ost.OSS.ost_io.nrs_tbf_rule |
4556 if [ "$output" != "$expected" ]; then
4557 error "$error_message, expected '$expected', got '$output'"
4562 [ "$OST1_VERSION" -ge $(version_code 2.8.55) ] ||
4563 skip "Need OST version at least 2.8.55"
4565 for i in $(seq 1 $OSTCOUNT)
4567 do_facet ost"$i" lctl set_param \
4568 ost.OSS.ost_io.nrs_policies="tbf\ jobid"
4570 error "failed to set TBF policy"
4573 tbf_rule_check ost1 "default" "error before inserting any rule"
4575 tbf_rule_operate ost1 "start\ before\ jobid={jobid}\ rate=1000"
4576 tbf_rule_check ost1 "before default" \
4577 "error when inserting rule 'before'"
4579 tbf_rule_operate ost1 "start\ after\ jobid={jobid}\ rate=1000\ rank=default"
4580 tbf_rule_check ost1 "before after default" \
4581 "error when inserting rule 'after'"
4583 tbf_rule_operate ost1 "start\ target\ jobid={jobid}\ rate=1000\ rank=after"
4584 tbf_rule_check ost1 "before target after default" \
4585 "error when inserting rule 'target'"
4587 echo "Move before itself"
4588 tbf_rule_operate ost1 "change\ target\ rank=target"
4589 tbf_rule_check ost1 "before target after default" \
4590 "error when moving before itself"
4592 echo "Move to higher rank"
4593 tbf_rule_operate ost1 "change\ target\ rank=before"
4594 tbf_rule_check ost1 "target before after default" \
4595 "error when moving to higher rank"
4597 echo "Move to lower rank"
4598 tbf_rule_operate ost1 "change\ target\ rank=after"
4599 tbf_rule_check ost1 "before target after default" \
4600 "error when moving to lower rank"
4602 echo "Move before default"
4603 tbf_rule_operate ost1 "change\ target\ rank=default"
4604 tbf_rule_check ost1 "before after target default" \
4605 error "error when moving before default"
4607 # Cleanup the TBF policy
4608 do_nodes $(comma_list $(osts_nodes)) \
4609 $LCTL set_param ost.OSS.ost_io.nrs_policies=fifo
4612 run_test 77i "Change rank of TBF rule"
4618 [ "$OST1_VERSION" -ge $(version_code 2.9.53) ] ||
4619 skip "Need OST version at least 2.9.53"
4620 if [ "$OST1_VERSION" -ge $(version_code 2.8.60) ]; then
4625 do_nodes $(comma_list $(osts_nodes)) \
4626 lctl set_param jobid_var=procname_uid \
4627 ost.OSS.ost_io.nrs_policies="tbf\ opcode" \
4628 ost.OSS.ost_io.nrs_tbf_rule="start\ ost_r\ ${idis}{ost_read}\ ${rateis}5" \
4629 ost.OSS.ost_io.nrs_tbf_rule="start\ ost_w\ ${idis}{ost_write}\ ${rateis}20"
4630 [ $? -ne 0 ] && error "failed to set TBF OPCode policy"
4635 do_nodes $(comma_list $(osts_nodes)) \
4636 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_r" \
4637 ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_w" \
4638 ost.OSS.ost_io.nrs_policies="fifo"
4640 # sleep 3 seconds to wait the tbf policy stop completely,
4641 # or the next test case is possible get -EAGAIN when
4642 # setting the tbf policy
4645 run_test 77j "check TBF-OPCode NRS policy"
4648 local idstr="${1}id"
4649 local policy="${idstr}={$2}"
4650 local rate="rate=$3"
4651 local runas_args="$4"
4652 local createas_args="${5:-$runas_args}"
4654 do_nodes $(comma_list $(osts_nodes)) \
4655 lctl set_param jobid_var=procname_uid \
4656 ost.OSS.ost_io.nrs_policies="tbf\ ${idstr}" \
4657 ost.OSS.ost_io.nrs_tbf_rule="start\ ost_${idstr}\ ${policy}\ ${rate}"
4658 [ $? -ne 0 ] && error "failed to set tbf ${idstr} policy"
4660 nrs_write_read "runas $runas_args" "runas $createas_args"
4661 tbf_verify $3 $3 "runas $runas_args" "runas $createas_args"
4663 do_nodes $(comma_list $(osts_nodes)) \
4664 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_${idstr}" \
4665 ost.OSS.ost_io.nrs_policies="fifo"
4667 # sleep 3 seconds to wait the tbf policy stop completely,
4668 # or the next test case is possible get -eagain when
4669 # setting the tbf policy
4674 if [ "$OST1_VERSION" -lt $(version_code 2.11.50) ]; then
4675 skip "Need OST version at least 2.11.50"
4678 test_id "u" "$RUNAS_ID" "5" "-u $RUNAS_ID"
4679 test_id "g" "$RUNAS_GID" "5" "-u $RUNAS_ID -g $RUNAS_GID"
4681 run_test 77ja "check TBF-UID/GID NRS policy"
4683 test_77jb() { # LU-16077
4684 (( "$OST1_VERSION" >= $(version_code 2.15.51) )) ||
4685 skip "Need OST version at least 2.15.51"
4687 test_id "u" "$RUNAS_ID" "5" "-u $RUNAS_ID" "-u 0 -g $RUNAS_GID"
4688 test_id "g" "$RUNAS_GID" "5" "-u $RUNAS_ID -g $RUNAS_GID" "-u $RUNAS_ID -g 0"
4690 run_test 77jb "check TBF-UID/GID NRS policy on files that don't belong to us"
4698 for rule in $rule_lists; do
4699 do_nodes $(comma_list $(osts_nodes)) \
4700 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ $rule"
4703 do_nodes $(comma_list $(osts_nodes)) \
4704 lctl set_param ost.OSS.ost_io.nrs_policies="$old_nrs"
4710 [[ "$OST1_VERSION" -ge $(version_code 2.9.53) ]] ||
4711 skip "Need OST version at least 2.9.53"
4713 do_nodes $(comma_list $(osts_nodes)) \
4714 lctl set_param ost.OSS.ost_io.nrs_policies="tbf" \
4715 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_w\ jobid={dd.$RUNAS_ID}\&opcode={ost_write}\ rate=20" \
4716 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_r\ jobid={dd.$RUNAS_ID}\&opcode={ost_read}\ rate=10"
4718 nrs_write_read "$RUNAS"
4719 tbf_verify 20 10 "$RUNAS"
4721 local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)")
4722 local client_nids=$(nids_list $address "\\")
4723 do_nodes $(comma_list $(osts_nodes)) \
4724 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_w" \
4725 ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_r" \
4726 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_w\ nid={0@lo\ $client_nids}\&opcode={ost_write}\ rate=20" \
4727 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_r\ nid={0@lo\ $client_nids}\&opcode={ost_read}\ rate=10"
4732 do_nodes $(comma_list $(osts_nodes)) \
4733 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_w" \
4734 ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_r" \
4735 ost.OSS.ost_io.nrs_tbf_rule="start\ ext\ nid={0@lo\ $client_nids}\&jobid={dd.$RUNAS_ID}\ rate=20"
4737 nrs_write_read "$RUNAS"
4738 tbf_verify 20 20 "$RUNAS"
4740 do_nodes $(comma_list $(osts_nodes)) \
4741 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext" \
4742 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ jobid={dd.$RUNAS_ID},opcode={ost_write}\ rate=20" \
4743 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ jobid={dd.$RUNAS_ID},opcode={ost_read}\ rate=10"
4745 nrs_write_read "$RUNAS"
4746 # with parameter "RUNAS", it will match the latest rule
4747 # "ext_b" first, so the limited write rate is 10.
4748 tbf_verify 10 10 "$RUNAS"
4751 trap "cleanup_77k \"ext_a ext_b\" \"fifo\"" EXIT
4753 [[ "$OST1_VERSION" -ge $(version_code 2.10.58) ]] ||
4754 skip "Need OST version at least 2.10.58"
4756 do_nodes $(comma_list $(osts_nodes)) \
4757 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_a" \
4758 ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_b" \
4759 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ug\ uid={$RUNAS_ID}\&gid={$RUNAS_GID}\ rate=5"
4760 nrs_write_read "runas -u $RUNAS_ID -g $RUNAS_GID"
4761 tbf_verify 5 5 "runas -u $RUNAS_ID -g $RUNAS_GID"
4763 do_nodes $(comma_list $(osts_nodes)) \
4764 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ug" \
4765 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_uw\ uid={$RUNAS_ID}\&opcode={ost_write}\ rate=20" \
4766 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ur\ uid={$RUNAS_ID}\&opcode={ost_read}\ rate=10"
4768 nrs_write_read "runas -u $RUNAS_ID"
4769 tbf_verify 20 10 "runas -u $RUNAS_ID"
4771 do_nodes $(comma_list $(osts_nodes)) \
4772 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_uw" \
4773 ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ur" \
4774 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ uid={$RUNAS_ID},opcode={ost_write}\ rate=20" \
4775 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ uid={$RUNAS_ID},opcode={ost_read}\ rate=10"
4776 nrs_write_read "runas -u $RUNAS_ID"
4777 tbf_verify 10 10 "runas -u $RUNAS_ID"
4778 tbf_verify 20 10 "runas -u $RUNAS_ID"
4779 cleanup_77k "ext_a ext_b" "fifo"
4781 run_test 77k "check TBF policy with NID/JobID/OPCode expression"
4784 [[ "$OST1_VERSION" -ge $(version_code 2.10.56) ]] ||
4785 skip "Need OST version at least 2.10.56"
4787 do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid"
4788 do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf"
4790 local output=$(do_facet ost1 lctl get_param \
4791 ost.OSS.ost_io.nrs_policies | \
4792 awk '/name: tbf/ {print;exit}' | \
4793 awk -F ': ' '{print $2}')
4795 if [ "$output" != "tbf" ]; then
4796 error "The generic TBF output is '$output', not 'tbf'"
4799 do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
4801 run_test 77l "check the output of NRS policies for generic TBF"
4804 if [ "$OST1_VERSION" -lt $(version_code 2.9.54) ]; then
4805 skip "Need OST version at least 2.9.54"
4808 local dir=$DIR/$tdir
4810 mkdir $dir || error "mkdir $dir failed"
4811 $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed"
4814 local nodes=$(comma_list $(osts_nodes))
4815 do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies=delay \
4816 ost.OSS.ost_io.nrs_delay_min=4 \
4817 ost.OSS.ost_io.nrs_delay_max=4 \
4818 ost.OSS.ost_io.nrs_delay_pct=100
4819 [ $? -ne 0 ] && error "Failed to set delay policy"
4821 local start=$SECONDS
4822 do_nodes "${SINGLECLIENT:-$HOSTNAME}" "$RUNAS" \
4823 dd if=/dev/zero of="$dir/nrs_delay_$HOSTNAME" bs=1M count=1 \
4824 oflag=direct conv=fdatasync ||
4825 { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
4826 error "dd on client failed (1)"; }
4827 local elapsed=$((SECONDS - start))
4829 # NRS delay doesn't do sub-second timing, so a request enqueued at
4830 # 0.9 seconds can be dequeued at 4.0
4831 [ $elapsed -lt 3 ] &&
4832 { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
4833 error "Single 1M write should take at least 3 seconds"; }
4836 do_nodes "${SINGLECLIENT:-$HOSTNAME}" "$RUNAS" \
4837 dd if=/dev/zero of="$dir/nrs_delay_$HOSTNAME" bs=1M count=10 \
4838 oflag=direct conv=fdatasync ||
4839 { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
4840 error "dd on client failed (2)"; }
4841 elapsed=$((SECONDS - start))
4843 [ $elapsed -lt 30 ] &&
4844 { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
4845 error "Ten 1M writes should take at least 30 seconds"; }
4847 do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
4848 [ $? -ne 0 ] && error "failed to set policy back to fifo"
4852 run_test 77m "check NRS Delay slows write RPC processing"
4854 test_77n() { #LU-10802
4855 if [ "$OST1_VERSION" -lt $(version_code 2.10.58) ]; then
4856 skip "Need OST version at least 2.10.58"
4859 # Configure jobid_var
4860 local saved_jobid_var=$($LCTL get_param -n jobid_var)
4861 if [ $saved_jobid_var != procname_uid ]; then
4862 set_persistent_param_and_check client \
4863 "jobid_var" "$FSNAME.sys.jobid_var" procname_uid
4866 do_nodes $(comma_list $(osts_nodes)) \
4867 lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid" \
4868 ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
4869 ost.OSS.ost_io.nrs_tbf_rule="start\ dd_runas\ jobid={*.$RUNAS_ID}\ rate=20"
4872 tbf_verify 20 20 "$RUNAS"
4874 do_nodes $(comma_list $(osts_nodes)) \
4875 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
4876 ost.OSS.ost_io.nrs_tbf_rule="start\ dd_runas\ jobid={dd.*}\ rate=20"
4881 do_nodes $(comma_list $(osts_nodes)) \
4882 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
4883 ost.OSS.ost_io.nrs_policies="fifo"
4887 local current_jobid_var=$($LCTL get_param -n jobid_var)
4888 if [ $saved_jobid_var != $current_jobid_var ]; then
4889 set_persistent_param_and_check client \
4890 "jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid_var
4893 run_test 77n "check wildcard support for TBF JobID NRS policy"
4896 (( $OST1_VERSION > $(version_code 2.14.54) )) ||
4897 skip "need OST > 2.14.54"
4899 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies="tbf\ nid"
4900 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ name\ nid={192.168.*.*@tcp}\ rate=10000"
4901 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ name1\ nid={192.168.*.*@tcp}\ rate=10000"
4902 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="change\ name1\ rank=name"
4903 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="stop\ name"
4904 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies="fifo"
4906 run_test 77o "Changing rank should not panic"
4910 local gidlist="500 10 33 100 "
4911 local uidlist=" 500 11 3"
4913 (( $MDS1_VERSION > $(version_code 2.14.54) )) ||
4914 skip "need MDS >= 2.14.54"
4916 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies="tbf"
4917 stack_trap "do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies=fifo"
4919 # require 44cc782/LU-9859 to support list with unexpected spaces
4920 (( MDS1_VERSION >= $(version_code 2.15.57) )) ||
4921 gidlist=$(echo $gidlist) uidlist=$(echo $uidlist)
4923 for i in {1..50}; do
4926 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'start rule77q_1 uid={$uidlist}&gid={$gidlist} rate=100'" &
4928 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'start rule77q_2 uid={1000}&gid={1000} rate=100'" &
4930 wait $pid1 || error "$i: Fail to start TBF rule 'rule77q_1'"
4931 wait $pid2 || error "$i: Fail to start TBF rule 'rule77q_2'"
4933 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'stop rule77q_1'" &
4935 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'stop rule77q_2'" &
4937 wait $pid1 || error "$i: Fail to stop TBF rule 'rule77q_1'"
4938 wait $pid2 || error "$i: Fail to stop TBF rule 'rule77q_2'"
4941 run_test 77q "Parallel TBF rule definitions should not panic"
4945 local -a spec_chars=(
4946 '@' '.' '~' '#' '/' '^' '%' '*' ';' ',' '?' '<' '>' ':'
4947 '+' '=' ')' '(' '{' '}' '|' '[' ']' '!' '&' '\$' '\`' '\\')
4949 (( $MDS1_VERSION > $(version_code 2.14.54) )) ||
4950 skip "need MDS >= 2.14.54"
4952 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies="tbf"
4953 stack_trap "do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_policies=fifo"
4955 # TBF rule name size is 16 bytes
4956 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ test_77p_overflo\ uid={500}\ rate=500" &&
4957 error "The length of tbf rule name is not checked" || true
4958 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ \ uid={500}\ rate=500" &&
4959 error "The server should not accept empty tbf rule name" || true
4960 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="start\ test_77p_empty" &&
4961 error "The server should not accept 'start <tbf_rule_name>' without an expression" || true
4963 # Test with special chars
4964 for c in "${spec_chars[@]}"; do
4965 do_facet mds1 $LCTL set_param mds.MDS.mdt.nrs_tbf_rule="'start test77p${c}spec uid={500} rate=500'" &&
4966 error "Special char '${c}' should not be accepted in a tbf rule name" || true
4970 run_test 77p "Check validity of rule names for TBF policies"
4974 local saved_jobid=$2
4975 local current_jobid_var
4977 echo "cleanup 77r $pid"
4979 do_facet mds1 $LCTL set_param -n mds.MDS.mdt.nrs_policies=fifo
4980 kill $pid || echo "fail to kill md thread"
4982 current_jobid_var=$($LCTL get_param -n jobid_var)
4983 if [ $saved_jobid != $current_jobid_var ]; then
4984 set_persistent_param_and_check client \
4985 "jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid
4992 md_thread_run="true"
4996 while $md_thread_run; do
4997 printf '%s\n' {$DIR1,$DIR2}/$tdir/${tfile}-{01..20} |
4998 xargs -P20 -I{} $RUNAS bash -c 'touch {}; rm -f {}' \
4999 &> /dev/null & pid=$!
5000 trap "echo kill md_thread xargs; md_thread_run=false; kill $pid" INT TERM
5005 wait_policy_state() {
5008 local change_pid="$3"
5011 for time in {1..60}; do
5014 nbr_started=$(do_facet mds1 $LCTL get_param mds.MDS.mdt.nrs_policies |
5015 egrep -A2 "name: ${policy}$" | grep -c "state: $state")
5017 [[ "$nbr_started" != 2 ]] || return 0
5021 [[ -z "$change_pid" ]] || kill $change_pid || true
5025 test_77r() { #LU-14976
5027 (( MDS1_VERSION > $(version_code 2.15.56) )) ||
5028 skip "need MDS >= 2.15.56 c098c095 change nrs policies at run time"
5033 local saved_jobid_var
5035 rules["tbf uid"]="start md_rule uid={$RUNAS_ID} rate=1"
5036 rules["tbf gid"]="start md_rule gid={$RUNAS_GID} rate=1"
5037 rules["tbf jobid"]="start md_rule jobid={*.$RUNAS_ID} rate=1"
5038 rules["tbf"]="start md_rule uid={$RUNAS_ID} rate=1"
5047 test_mkdir -i 0 -c 1 $DIR1/$tdir
5048 chmod 777 $DIR1/$tdir
5050 # Configure jobid_var
5051 saved_jobid_var=$($LCTL get_param -n jobid_var)
5052 if [ $saved_jobid_var != procname_uid ]; then
5053 set_persistent_param_and_check client \
5054 "jobid_var" "$FSNAME.sys.jobid_var" procname_uid
5058 md_thread_77r & pid=$!
5059 stack_trap "cleanup_77r $pid '$saved_jobid_var'"
5062 for policy in "${policies[@]}"; do
5065 # wait to queue requests
5068 do_facet mds1 "$LCTL set_param mds.MDS.mdt.nrs_policies='$policy'" &
5071 wait_policy_state "started" "$policy" "$change_pid" ||
5072 error "timeout to start '$policy' policy"
5074 [[ -n "${rules[$policy]}" ]] || continue
5076 do_facet mds1 "$LCTL set_param mds.MDS.mdt.nrs_tbf_rule='${rules[$policy]}'" ||
5077 error "fail to set rule '${rules[$policy]}' to '$policy'"
5080 wait_policy_state "stopped" "tbf" ||
5081 error "fail to stop tbf policy"
5083 echo "check the number of requests in queue:"
5084 local awkcmd='/name: / {last = $3} '
5085 awkcmd+='/queued: / {printf " %s: %d\n", last, $2;'
5086 awkcmd+=' if (last == "tbf" && $2 > 0) exit 1;}'
5087 do_facet mds1 $LCTL get_param mds.MDS.mdt.nrs_policies | awk "$awkcmd" ||
5088 error "request leak in tbf policies"
5090 run_test 77r "Change type of tbf policy at run time"
5092 test_78() { #LU-6673
5095 oss=$(comma_list $(osts_nodes))
5096 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="orr" &
5097 do_nodes $oss lctl set_param ost.OSS.*.nrs_orr_quantum=1
5099 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
5100 # Valid return codes are:
5101 # 0: Tuning succeeded
5102 # ENODEV: Policy is still stopped
5103 # EAGAIN: Policy is being initialized
5104 [ $rc -eq 0 -o $rc -eq 19 -o $rc -eq 11 ] ||
5105 error "Expected set_param to return 0|ENODEV|EAGAIN"
5107 # Cleanup the ORR policy
5108 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
5109 [ $? -ne 0 ] && error "failed to set policy back to fifo"
5112 run_test 78 "Enable policy and specify tunings right away"
5115 remote_mds_nodsh && skip "remote MDS with nodsh" && return
5116 test_mkdir $DIR/$tdir
5118 # Prevent interference from layout intent RPCs due to
5119 # asynchronous writeback. These will be tested in 130c below.
5120 do_nodes ${CLIENTS:-$HOSTNAME} sync
5122 setfattr -n trusted.name1 -v value1 $DIR/$tdir ||
5123 error "setfattr -n trusted.name1=value1 $DIR/$tdir failed"
5125 #define OBD_FAIL_MDS_INTENT_DELAY 0x160
5126 local mdtidx=$($LFS getstripe -m $DIR/$tdir)
5127 local facet=mds$((mdtidx + 1))
5129 set_nodes_failloc $(facet_active_host $facet) 0x80000160
5130 getfattr -n trusted.name1 $DIR/$tdir 2> /dev/null &
5134 #define OBD_FAIL_MDS_GETXATTR_PACK 0x131
5135 set_nodes_failloc $(facet_active_host $facet) 0x80000131
5140 run_test 79 "xattr: intent error"
5143 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5150 mkdir_on_mdt0 $DIR1/$tdir
5151 mkdir -p $DIR1/$tdir/dir
5152 createmany -o $DIR1/$tdir/dir/f 10 ||
5153 error "create files under remote dir failed $i"
5155 cp /etc/passwd $DIR1/$tdir/$tfile
5157 #migrate open file should fails
5158 multiop_bg_pause $DIR2/$tdir/$tfile O_c || error "open $file failed"
5160 # give multiop a chance to open
5163 $LFS migrate -m $MDTIDX $DIR1/$tdir &&
5164 error "migrate open files should failed with open files"
5168 $LFS migrate -m $MDTIDX $DIR1/$tdir ||
5169 error "migrate remote dir error"
5171 echo "Finish migration, then checking.."
5172 for file in $(find $DIR1/$tdir); do
5173 mdt_index=$($LFS getstripe -m $file)
5174 [ $mdt_index == $MDTIDX ] ||
5175 error "$file is not on MDT${MDTIDX}"
5178 diff /etc/passwd $DIR1/$tdir/$tfile ||
5179 error "file different after migration"
5181 rm -rf $DIR1/$tdir || error "rm dir failed after migration"
5183 run_test 80a "migrate directory when some children is being opened"
5187 kill -9 $migrate_pid
5195 eval $@ &>/dev/null && success_count=$((success_count + 1)) ||
5196 failure_count=$((failure_count + 1))
5200 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5201 local migdir1=$DIR1/$tdir/migrate_dir
5202 local migdir2=$DIR2/$tdir/migrate_dir
5208 trap cleanup_80b EXIT
5209 # prepare migrate directory
5211 for F in {1,2,3,4,5}; do
5212 echo "$F$F$F$F$F" > $migdir1/file$F
5213 echo "$F$F$F$F$F" > $DIR/$tdir/file$F
5216 # migrate the directories among MDTs
5219 mdt_idx=$((RANDOM % MDSCOUNT))
5221 $LFS migrate -m $mdt_idx $migdir1 &>/dev/null
5226 echo "start migration thread $migrate_pid"
5227 # access the files at the same time
5229 echo "accessing the migrating directory for 1 minute..."
5230 while ((SECONDS - start_time < 60)); do
5231 run_and_count ls $migdir2
5232 run_and_count diff -u $DIR2/$tdir/file1 $migdir2/file1
5233 run_and_count "cat $migdir2/file2 > $migdir2/file3"
5234 run_and_count "echo "aaaaa" > $migdir2/file4"
5235 run_and_count stat $migdir2/file5
5236 run_and_count touch $migdir2/source_file
5237 if [ -e $migdir2/source_file ]; then
5238 run_and_count ln $migdir2/source_file \
5241 [ -e $migdir2/link_file ] &&
5242 rm -rf $migdir2/link_file
5244 run_and_count mrename $migdir2/source_file \
5245 $migdir2/target_file
5247 [ -e $migdir2/target_file ] &&
5248 run_and_count rm -rf $migdir2/target_file ||
5249 run_and_count rm -rf $migdir2/source_file
5252 kill -0 $migrate_pid || {
5253 echo "migration stopped 1"
5258 # check migration are still there
5259 kill -0 $migrate_pid || error "migration stopped 2"
5261 # access during migration may fail
5262 echo "concurrent access $failure_count failures, $success_count successes"
5264 run_test 80b "Accessing directory during migration"
5267 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5271 mkdir -p $DIR1/$tdir
5273 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR1/$tdir/d0
5274 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR1/$tdir/d1
5277 touch d0/0 || error "create 0 failed"
5278 mv d0/0 d1/0 || error "rename d0/0 d1/0 failed"
5279 stat d0/0 && error "stat mv filed succeed"
5280 mv $DIR2/$tdir/d1/0 $DIR2/$tdir/d0/0 || error "rename d1/0 d0/0 failed"
5281 stat d0/0 || error "stat failed"
5283 local t=$(ls -ai $DIR1/$tdir/d0 | sort -u | wc -l)
5285 if [ $t -ne 3 ]; then
5286 ls -ai $DIR1/$tdir/d0
5287 error "expect 3 get $t"
5292 run_test 81a "rename and stat under striped directory"
5295 [ $MDSCOUNT -lt 2 ] &&
5296 skip "We need at least 2 MDTs for this test"
5300 $LFS mkdir -c $MDSCOUNT $DIR1/$tdir || error "$LFS mkdir"
5301 createmany -o $DIR1/$tdir/$tfile. $COUNT || error "createmany"
5310 for ((i = 0; i < COUNT; i++)); do
5311 mrename $DIR2/$tdir/$tfile.$i $DIR2/$tdir/$tfile-new.$i \
5315 kill -9 $setattr_pid
5317 run_test 81b "rename under striped directory doesn't deadlock"
5320 [ $MDSCOUNT -lt 4 ] && skip_env "needs >= 4 MDTs"
5321 [ $MDS1_VERSION -lt $(version_code 2.13.52) ] &&
5322 skip "Need MDS version at least 2.13.52"
5324 # source is local, source parent is remote
5325 $LFS mkdir -i 0 $DIR1/${tdir}_src || error "mkdir ${tdir}_src"
5326 $LFS mkdir -i 1 $DIR1/${tdir}_tgt || error "mkdir ${tdir}_tgt"
5327 $LFS mkdir -i 3 $DIR1/${tdir}_src/sub || error "mkdir sub"
5328 $LFS mkdir -i 3 $DIR1/${tdir}_tgt/sub || error "mkdir sub"
5329 stat $DIR2/${tdir}_src/sub || error "stat sub failed"
5330 mv $DIR1/${tdir}_src/sub $DIR1/${tdir}_tgt/ || error "mv failed"
5331 [ -f $DIR2/${tdir}_src/sub ] && error "sub should be gone"
5332 rm -rf $DIR1/${tdir}_src $DIR1/${tdir}_tgt
5334 # source is remote, source parent is local
5335 $LFS mkdir -i 3 $DIR1/${tdir}_src || error "mkdir ${tdir}_src"
5336 $LFS mkdir -i 1 $DIR1/${tdir}_tgt || error "mkdir ${tdir}_tgt"
5337 $LFS mkdir -i 0 $DIR1/${tdir}_src/sub || error "mkdir sub"
5338 $LFS mkdir -i 3 $DIR1/${tdir}_tgt/sub || error "mkdir sub"
5339 stat $DIR2/${tdir}_src/sub || error "stat sub failed"
5340 mv $DIR1/${tdir}_src/sub $DIR1/${tdir}_tgt/ || error "mv failed"
5341 [ -f $DIR2/${tdir}_src/sub ] && error "sub should be gone"
5342 rm -rf $DIR1/${tdir}_src $DIR1/${tdir}_tgt
5344 # source and source parent are remote
5345 $LFS mkdir -i 0 $DIR1/${tdir}_src || error "mkdir ${tdir}_src"
5346 $LFS mkdir -i 1 $DIR1/${tdir}_tgt || error "mkdir ${tdir}_tgt"
5347 mkdir $DIR1/${tdir}_src/sub || error "mkdir sub"
5348 $LFS mkdir -i 3 $DIR1/${tdir}_tgt/sub || error "mkdir sub"
5349 stat $DIR2/${tdir}_src/sub || error "stat sub failed"
5350 mv $DIR1/${tdir}_src/sub $DIR1/${tdir}_tgt/ || error "mv failed"
5351 [ -f $DIR2/${tdir}_src/sub ] && error "sub should be gone"
5352 rm -rf $DIR1/${tdir}_src $DIR1/${tdir}_tgt
5354 # source and source parent are remote, and source is remote object
5355 $LFS mkdir -i 0 $DIR1/${tdir}_src || error "mkdir ${tdir}_src"
5356 $LFS mkdir -i 1 $DIR1/${tdir}_tgt || error "mkdir ${tdir}_tgt"
5357 $LFS mkdir -i 2 $DIR1/${tdir}_src/sub || error "mkdir sub"
5358 $LFS mkdir -i 3 $DIR1/${tdir}_tgt/sub || error "mkdir sub"
5359 stat $DIR2/${tdir}_src/sub || error "stat sub failed"
5360 mv $DIR1/${tdir}_src/sub $DIR1/${tdir}_tgt/ || error "mv failed"
5361 [ -f $DIR2/${tdir}_src/sub ] && error "sub should be gone" || true
5363 run_test 81c "rename revoke LOOKUP lock for remote object"
5366 for ((mds = 0; mds < $MDSCOUNT; mds++)); do
5367 local d2=$DIR2/$tdir-$mds
5371 wait || error "rm failed"
5376 local mdts=$(comma_list $(mdts_nodes))
5378 do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear > /dev/null"
5380 stack_trap cleanup_81d
5381 for ((mds = 0; mds < $MDSCOUNT; mds++)); do
5382 local d1=$DIR1/$tdir-$mds
5384 $LFS mkdir -i $mds $d1 $d1/_temporary || error "mkdir failed"
5385 createmany -o $d1/_temporary/$tfile. $COUNT ||
5386 error "createmany failed for $d1/_temporary"
5389 for ((mds = 0; mds < $MDSCOUNT; mds++)); do
5390 local d1=$DIR1/$tdir-$mds
5391 local d2=$DIR2/$tdir-$mds
5393 for ((i = 0; i < COUNT; i++)); do
5394 mrename $d1/_temporary/$tfile.$i $d1/$tfile.$i &
5396 mrename $d2/_temporary/$tfile.$i $d2/$tfile.$i &
5399 wait || error "rename failed"
5402 local stats=$DIR1/md_stats
5403 local total=$((MDSCOUNT * COUNT))
5405 do_nodes $mdts "$LCTL get_param -n mdt.*.md_stats" > $stats
5407 crossdir=$(awk '/crossdir_rename/ {sum+=$2} END {print sum}' $stats)
5408 (( crossdir == total )) ||
5409 error "not crossdir: $crossdir != $total"
5410 samedir=$(awk '/samedir_rename/ {sum+=$2} END {print sum}' $stats)
5411 (( samedir == 0 )) || error "considered samedir: $samedir"
5412 pardir=$(awk '/parallel_rename_dir/ {sum+=$2} END {print sum}' $stats)
5413 (( pardir == 0 )) || error "considered directory: $pardir"
5416 (( MDS1_VERSION >= $(version_code 2.15.60) )) ||
5417 { echo "need MDS >= 2.15.60 for parallel cross-dir"; return 0; }
5419 parfile=$(awk '/parallel_rename_file/ {sum+=$2} END {print sum}' $stats)
5420 (( parfile == total )) ||
5421 error "not considered file: $parfile != $total"
5423 run_test 81d "parallel rename file cross-dir on same MDT"
5426 [[ "$MDS1_VERSION" -gt $(version_code 2.6.91) ]] ||
5427 skip "Need MDS version at least 2.6.92"
5429 # Client 1 creates a file.
5430 multiop_bg_pause $DIR1/$tfile O_ac || error "multiop_bg_pause 1"
5432 # Client 2 opens the file.
5433 multiop_bg_pause $DIR2/$tfile o_Ac || error "multiop_bg_pause 2"
5435 # Client 1 makes the file an orphan.
5436 rm $DIR1/$tfile || error "rm"
5437 # Client 2 sets EA "user.multiop".
5439 wait $pid2 || error "multiop 2"
5440 # Client 1 gets EA "user.multiop". This used to fail because the EA
5441 # cache refill would get "trusted.link" from mdd_xattr_list() but
5442 # -ENOENT when trying to get "trusted.link"'s value. See also sanity
5445 wait $pid1 || error "multiop 1"
5447 run_test 82 "fsetxattr and fgetxattr on orphan files"
5450 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5457 $LFS mkdir -i1 -c2 $tdir
5462 echo "start pid $pid1 to create/unlink striped directory"
5464 # Access the directory at the same time
5468 stat $tdir > /dev/null 2>&1
5472 echo "start pid $pid2 to stat striped directory"
5480 run_test 83 "access striped directory while it is being created/unlinked"
5483 [ $MDS1_VERSION -lt $(version_code 2.12.55) ] &&
5484 skip "lustre < 2.12.55 does not contain LU-12485 fix"
5488 $MULTIOP $DIR/$tfile oO_RDWR:O_CREAT:O_LOV_DELAY_CREATE:c ||
5489 error "create $tfile failed"
5490 mtime=$(stat -c%Y $DIR/$tfile)
5491 mtime=$((mtime + 200))
5493 #define OBD_FAIL_OBD_0NLINK_RACE 0x60b
5494 do_facet mds1 $LCTL set_param fail_loc=0x8000060b
5496 touch -c -m $mtime $DIR/$tfile &
5498 # sleep a while to let 'touch' run first
5500 rm -f $DIR2/$tfile || error "unlink $tfile failed"
5503 wait $setattr_pid || true
5505 run_test 84 "0-nlink race in lu_object_find()"
5508 llapi_root_test $DIR/$tfile $DIR2/$tfile
5510 run_test 85 "Lustre API root cache race"
5513 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5518 [ "$SLOW" = "yes" ] && duration=600
5519 # Open/Create under striped directory
5523 $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1
5524 touch $tdir/f{0..3} > /dev/null 2>&1
5528 echo "start pid $pid1 to open/create under striped directory"
5530 # unlink the striped directory at the same time
5534 rm -rf $tdir > /dev/null 2>&1
5538 echo "start pid $pid2 to unlink striped directory"
5547 run_test 90 "open/create and unlink striped directory"
5550 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5555 [ "$SLOW" = "yes" ] && duration=600
5556 # chmod striped directory
5560 $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1
5561 chmod go+w $tdir > /dev/null 2>&1
5565 echo "start pid $pid1 to chmod striped directory"
5567 # unlink the striped directory at the same time
5571 rm -rf $tdir > /dev/null 2>&1
5575 echo "start pid $pid2 to unlink striped directory"
5584 run_test 91 "chmod and unlink striped directory"
5587 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5590 local cmd="exec $fd<$DIR1/$tdir"
5591 $LFS setdirstripe -c$MDSCOUNT $DIR1/$tdir || error "mkdir $tdir fails"
5594 trap "eval $cmd" EXIT
5595 cd $DIR1/$tdir || error "cd $DIR1/$tdir fails"
5596 rmdir ../$tdir || error "rmdir ../$tdir fails"
5598 #define OBD_FAIL_LLITE_NO_CHECK_DEAD 0x1408
5599 $LCTL set_param fail_loc=0x1408
5600 mkdir $DIR2/$tdir/dir && error "create dir succeeds"
5601 $LFS setdirstripe -i1 $DIR2/$tdir/remote_dir &&
5602 error "create remote dir succeeds"
5603 $LCTL set_param fail_loc=0
5607 run_test 92 "create remote directory under orphan directory"
5614 mkdir -p $DIR1/$tfile-1/
5615 mkdir -p $DIR2/$tfile-2/
5616 local old_rr=$(do_facet $SINGLEMDS "$LCTL get_param -n \
5617 lod.$FSNAME-MDT0000-*/qos_threshold_rr" | sed -e 's/%//')
5618 do_facet $SINGLEMDS "$LCTL set_param -n \
5619 lod.$FSNAME-MDT0000-*/qos_threshold_rr=100"
5620 #define OBD_FAIL_MDS_LOV_CREATE_RACE 0x163
5621 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x00000163"
5623 $LFS setstripe -c -1 $DIR1/$tfile-1/file1 &
5626 $LFS setstripe -c -1 $DIR2/$tfile-2/file2 &
5630 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x0"
5631 do_facet $SINGLEMDS "$LCTL set_param -n \
5632 lod.$FSNAME-MDT0000-*/qos_threshold_rr=$old_rr"
5634 $LFS getstripe $DIR1/$tfile-1/file1
5635 rc1=$($LFS getstripe -q $DIR1/$tfile-1/file1 |
5636 awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l)
5637 $LFS getstripe $DIR2/$tfile-2/file2
5638 rc2=$($LFS getstripe -q $DIR2/$tfile-2/file2 |
5639 awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l)
5640 echo "rc1=$rc1 and rc2=$rc2 "
5641 [ $rc1 -eq 0 ] && [ $rc2 -eq 0 ] ||
5642 error "object allocate on same ost detected"
5644 run_test 93 "alloc_rr should not allocate on same ost"
5647 $LCTL set_param osc.*.idle_timeout=0
5648 dd if=/dev/zero of=$DIR2/$tfile bs=4k count=2 conv=fsync
5650 local before=$(date +%s)
5654 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
5655 $LCTL set_param fail_val=5 fail_loc=0x80000312
5656 dd if=/dev/zero of=$DIR/$tfile conv=notrunc oflag=append bs=4k count=1 &
5660 #define OBD_FAIL_LDLM_PAUSE_CANCEL_LOCAL 0x329
5661 $LCTL set_param fail_val=6 fail_loc=0x80000329
5662 $LCTL mark kill $pid
5665 dd if=/dev/zero of=$DIR2/$tfile conv=notrunc oflag=append bs=4k count=1
5668 dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 conv=fsync
5670 evict=$(do_facet client $LCTL get_param \
5671 osc.$FSNAME-OST*-osc-*/state |
5672 awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }')
5674 [ -z "$evict" ] || [[ $evict -le $before ]] ||
5675 (do_facet client $LCTL get_param \
5676 osc.$FSNAME-OST*-osc-*/state;
5677 error "eviction happened: $evict before:$before")
5678 $LCTL set_param osc.*.idle_timeout=debug
5680 run_test 94 "signal vs CP callback race"
5683 local file=$DIR/$tfile
5684 local file2=$DIR2/$tfile
5685 local fast_read_save
5688 fast_read_save=$($LCTL get_param -n llite.*.fast_read | head -n 1)
5689 [ -z "$fast_read_save" ] && skip "no fast read support"
5691 stack_trap "$LCTL set_param llite.*.fast_read=$fast_read_save" EXIT
5692 $LCTL set_param llite.*.fast_read=0
5694 $LFS setstripe -c $OSTCOUNT $file || error "failed to setstripe $file"
5695 dd if=/dev/zero of=$file bs=1M count=2 || error "failed to write $file"
5696 cancel_lru_locks $OSC
5697 $MULTIOP $file Oz1048576w4096c || error "failed to write $file"
5698 $MULTIOP $file oz1044480r4096c || error "failed to read $file"
5700 # OBD_FAIL_LLITE_PAGE_INVALIDATE_PAUSE 0x1421
5701 $LCTL set_param fail_loc=0x80001421 fail_val=7
5702 $MULTIOP $file2 Oz1048576w4096_c &
5706 # OBD_FAIL_LLITE_READPAGE_PAUSE 0x1422
5707 $LCTL set_param fail_loc=0x80001422 fail_val=10
5708 $MULTIOP $file oz1044480r4096c || error "failed to read $file"
5710 kill -USR1 $pid && wait $pid || error "wait for PID $pid failed"
5712 run_test 95a "Check readpage() on a page that was removed from page cache"
5715 local file=$DIR/$tfile
5716 local file2=$DIR2/$tfile
5717 local fast_read_save
5720 fast_read_save=$($LCTL get_param -n llite.*.fast_read | head -n 1)
5721 [ -z "$fast_read_save" ] && skip "no fast read support"
5723 stack_trap "$LCTL set_param llite.*.fast_read=$fast_read_save" EXIT
5724 $LCTL set_param llite.*.fast_read=0
5726 $LFS setstripe -c $OSTCOUNT $file || error "failed to setstripe $file"
5727 dd if=/dev/zero of=$file bs=$((PAGE_SIZE * 3)) count=1 ||
5728 error "failed to write $file"
5730 # This does the read from the second mount, so this flushes the pages
5731 # the first mount and creates new ones on the second mount
5732 # OBD_FAIL_LLITE_READPAGE_PAUSE2 0x1424
5733 $LCTL set_param fail_loc=0x80001424 fail_val=5
5734 $MULTIOP $file2 or${PAGE_SIZE}c &
5738 fadvise_dontneed_helper $file2
5739 $LCTL set_param fail_loc=0
5741 wait $pid || error "failed to read file"
5743 run_test 95b "Check readpage() on a page that is no longer uptodate"
5747 skip "Reserved for glimpse-ahead" && return
5748 [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] &&
5749 skip "Need MDS version at least 2.10.55"
5753 $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
5755 lctl set_param -n mdc.*.stats=clear
5756 dd if=/dev/zero of=$DIR2/$tdir/dom bs=4096 count=1 || return 1
5758 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #1"
5759 # first stat from server should return size data and save glimpse
5760 local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
5761 [ $gls -eq 0 ] || error "Unexpected $gls glimpse RPCs"
5762 # second stat to check size is NOT cached on client without IO lock
5763 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #2"
5765 local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
5766 [ $gls -ge 1 ] || error "Expect glimpse RPCs but none"
5769 run_test 100a "DoM: glimpse RPCs for stat without IO lock (DoM only file)"
5772 [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] &&
5773 skip "Need MDS version at least 2.10.55"
5777 $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
5779 lctl set_param -n mdc.*.stats=clear
5780 dd if=/dev/zero of=$DIR2/$tdir/dom bs=4096 count=1 || return 1
5781 cancel_lru_locks mdc
5782 # first stat data from server should have size
5783 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #1"
5784 # second stat to check size is cached on client
5785 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #2"
5787 local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
5788 # both stats should cause no glimpse requests
5789 [ $gls == 0 ] || error "Unexpected $gls glimpse RPCs"
5792 run_test 100b "DoM: no glimpse RPC for stat with IO lock (DoM only file)"
5795 [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] &&
5796 skip "Need MDS version at least 2.10.55"
5800 $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
5802 lctl set_param -n mdc.*.stats=clear
5803 lctl set_param -n osc.*.stats=clear
5804 dd if=/dev/zero of=$DIR2/$tdir/dom bs=2048K count=1 || return 1
5806 # check that size is merged from MDT and OST correctly
5807 $CHECKSTAT -t file -s 2097152 $DIR/$tdir/dom ||
5808 error "Wrong size from stat #1"
5810 local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse)
5811 [ $gls -eq 0 ] && error "Expect OST glimpse RPCs but got none"
5815 run_test 100c "DoM: write vs stat without IO lock (combined file)"
5818 [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] &&
5819 skip "Need MDS version at least 2.10.55"
5823 $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
5826 dd if=/dev/zero of=$DIR2/$tdir/dom bs=2048K count=1 || return 1
5827 lctl set_param -n mdc.*.stats=clear
5828 $TRUNCATE $DIR2/$tdir/dom 4096
5830 # check that reported size is valid after file grows to OST and
5831 # is truncated back to MDT stripe size
5832 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom ||
5833 error "Wrong size from stat #1"
5835 local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse)
5836 [ $gls -eq 0 ] && error "Expect OST glimpse but got none"
5840 run_test 100d "DoM: write+truncate vs stat without IO lock (combined file)"
5843 [ "$MDS1_VERSION" -lt $(version_code 2.11.50) ] &&
5844 skip "Need MDS version at least 2.11.50"
5846 local dom=$DIR/$tdir/dom
5847 local dom2=$DIR2/$tdir/dom
5850 $LFS setstripe -E 1024K -L mdt $DIR/$tdir
5852 cancel_lru_locks mdc
5853 dd if=/dev/urandom of=$dom bs=12000 count=1
5854 $TRUNCATE $dom2 6000
5855 cancel_lru_locks mdc
5856 lctl set_param -n mdc.*.stats=clear
5857 # expect read-on-open to return all data before write
5858 cat /etc/hosts >> $dom
5859 local read=$(lctl get_param -n mdc.*.stats | grep -c ost_read)
5860 [[ $read -eq 0 ]] || error "Unexpected $read READ RPCs"
5862 run_test 100e "DoM: read on open and file size"
5865 [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] &&
5866 skip "Need MDS version at least 2.10.55"
5868 $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
5870 $CHECKSTAT -t file $DIR1/$tfile
5872 local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
5873 sysctl -wq vm.dirty_writeback_centisecs=0
5875 trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
5878 dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 ||
5879 error_noexit "Write fails"
5880 # must discard pages
5881 lctl set_param -n mdc.*.stats=clear
5882 rm $DIR2/$tfile || error "Unlink fails"
5884 local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
5885 [ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
5887 run_test 101a "Discard DoM data on unlink"
5890 [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] &&
5891 skip "Need MDS version at least 2.10.55"
5893 $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
5894 touch $DIR1/${tfile}_2
5896 $CHECKSTAT -t file $DIR1/$tfile
5898 local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
5899 sysctl -wq vm.dirty_writeback_centisecs=0
5901 trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
5904 dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails"
5905 # must discard pages
5906 lctl set_param -n mdc.*.stats=clear
5907 mv $DIR2/${tfile}_2 $DIR2/$tfile || error "Rename fails"
5909 local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
5910 [ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
5912 run_test 101b "Discard DoM data on rename"
5915 [ "$MDS1_VERSION" -lt $(version_code 2.10.55) ] &&
5916 skip "Need MDS version at least 2.10.55"
5918 $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
5920 $CHECKSTAT -t file $DIR1/$tfile
5922 local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
5923 sysctl -wq vm.dirty_writeback_centisecs=0
5925 trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
5928 dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails"
5929 $MULTIOP $DIR1/$tfile O_c &
5932 lctl set_param -n mdc.*.stats=clear
5933 rm $DIR2/$tfile > /dev/null || error "Unlink fails for opened file"
5934 kill -USR1 $MULTIOP_PID && wait $MULTIOP_PID || error "multiop failure"
5936 local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
5937 [ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
5939 run_test 101c "Discard DoM data on close-unlink"
5941 # test to verify file handle related system calls
5942 # (name_to_handle_at/open_by_handle_at)
5943 # The new system calls are supported in glibc >= 2.14.
5945 # test to verify we can open by handle an unlinked file from > 1 client
5946 # This test opens the file normally on $DIR1, which is on one mount, and then
5947 # opens it by handle on $DIR2, which is on a different mount.
5949 (( "$MDS1_VERSION" >= $(version_code 2.11.57) )) ||
5950 skip "Needs MDS version 2.11.57 or later"
5952 echo "Test file_handle syscalls" > $DIR/$tfile ||
5953 error "write failed"
5954 check_fhandle_syscalls $DIR/$tfile $DIR2 ||
5955 error "check_fhandle_syscalls $tfile failed"
5957 # test this is working on DNE directories also
5958 if (( MDSCOUNT > 1 && MDS1_VERSION >= $(version_code 2.14.52) )); then
5959 $LFS mkdir -i 1 $DIR/$tdir.remote
5960 cancel_lru_locks mdc
5961 check_fhandle_syscalls $DIR/$tdir.remote $DIR2 ||
5962 error "check_fhandle_syscalls $tdir.remote failed"
5963 $LFS mkdir -c -1 $DIR/$tdir.remote/subdir
5964 cancel_lru_locks mdc
5965 check_fhandle_syscalls $DIR/$tdir.remote/subdir $DIR2 ||
5966 error "check_fhandle_syscalls $tdir.remote/subdir fail"
5968 $LFS mkdir -c -1 $DIR/$tdir.stripe
5969 cancel_lru_locks mdc
5970 check_fhandle_syscalls $DIR/$tdir.stripe $DIR2 ||
5971 error "check_fhandle_syscalls $tdir.stripe failed"
5972 $LFS mkdir -c -1 $DIR/$tdir.stripe/subdir
5973 cancel_lru_locks mdc
5974 check_fhandle_syscalls $DIR/$tdir.stripe/subdir $DIR2 ||
5975 error "check_fhandle_syscalls $tdir.stripe/subdir fail"
5978 run_test 102 "Test open by handle of unlinked file"
5980 # Compare file size between first & second mount, ensuring the client correctly
5981 # glimpses even with unused speculative locks - LU-11670
5983 [ $OST1_VERSION -lt $(version_code 2.10.50) ] &&
5984 skip "Lockahead needs OST version at least 2.10.50"
5988 test_mkdir -p $DIR/$tdir
5990 # Force file on to OST0
5991 $LFS setstripe -i 0 $DIR/$tdir
5993 # Do not check multiple locks on glimpse
5994 # OBD_FAIL_OSC_NO_SIZE_DATA 0x415
5995 $LCTL set_param fail_loc=0x415
5997 # Delay write commit by 2 seconds to guarantee glimpse wins race
5998 # The same fail_loc is used on client & server so it can work in the
5999 # single node sanity setup
6000 do_facet ost1 $LCTL set_param fail_loc=0x415 fail_val=2
6002 echo "Incorrect size expected (no glimpse fix):"
6003 lockahead_test -d $DIR/$tdir -D $DIR2/$tdir -t $locktest -f $tfile
6005 if [ $rc -eq 0 ]; then
6006 echo "This doesn't work 100%, but this is just reproducing the bug, not testing the fix, so OK to not fail test."
6009 # guarantee write commit timeout has expired
6012 # Clear fail_loc on client
6013 $LCTL set_param fail_loc=0
6015 # Delay write commit by 2 seconds to guarantee glimpse wins race
6016 # OBD_FAIL_OST_BRW_PAUSE_BULK 0x214
6017 do_facet ost1 $LCTL set_param fail_loc=0x214 fail_val=2
6019 # Write commit is still delayed by 2 seconds
6020 lockahead_test -d $DIR/$tdir -D $DIR2/$tdir -t $locktest -f $tfile
6022 [ $rc -eq 0 ] || error "Lockahead test$locktest failed, $rc"
6024 # guarantee write commit timeout has expired
6027 rm -f $DIR/$tfile || error "unable to delete $DIR/$tfile"
6029 run_test 103 "Test size correctness with lockahead"
6035 xtimes=$(stat -c "%X %Y %Z" $DIR/$tfile)
6046 output=$(do_facet mds1 "$DEBUGFS -c -R 'stat ROOT/$tfile' $mdtdev")
6047 ((xtimes[0]=$(awk -F ':' /atime/'{ print $2 }' <<< "$output")))
6048 ((xtimes[1]=$(awk -F ':' /mtime/'{ print $2 }' <<< "$output")))
6049 ((xtimes[2]=$(awk -F ':' /ctime/'{ print $2 }' <<< "$output")))
6057 local xtimes=($(get_stat_xtimes))
6058 local mdt_xtimes=($(get_mdt_xtimes $mdtdev))
6060 echo "STAT a|m|ctime ${xtimes[*]}"
6061 echo "MDT a|m|ctime ${mdt_xtimes[*]}"
6062 [[ ${xtimes[0]} == ${mdt_xtimes[0]} ]] ||
6063 error "$DIR/$tfile atime (${xtimes[0]}:${mdt_xtimes[0]}) diff"
6064 [[ ${xtimes[1]} == ${mdt_xtimes[1]} ]] ||
6065 error "$DIR/$tfile mtime (${xtimes[1]}:${mdt_xtimes[1]}) diff"
6066 [[ ${xtimes[2]} == ${mdt_xtimes[2]} ]] ||
6067 error "$DIR/$tfile ctime (${xtimes[2]}:${mdt_xtimes[2]}) diff"
6071 [ "$mds1_FSTYPE" == "ldiskfs" ] || skip_env "ldiskfs only test"
6072 [ $MDS1_VERSION -lt $(version_code 2.12.4) ] &&
6073 skip "Need MDS version at least 2.12.4"
6076 local mdtdev=$(mdsdevname ${SINGLEMDS//mds/})
6077 local atime_diff=$(do_facet $SINGLEMDS \
6078 lctl get_param -n mdd.*MDT0000*.atime_diff)
6080 do_facet $SINGLEMDS \
6081 lctl set_param -n mdd.*MDT0000*.atime_diff=0
6083 stack_trap "do_facet $SINGLEMDS \
6084 lctl set_param -n mdd.*MDT0000*.atime_diff=$atime_diff" EXIT
6086 dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc
6087 check_mdt_xtimes $mdtdev
6090 dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc
6091 check_mdt_xtimes $mdtdev
6093 $MULTIOP $DIR2/$tfile Oz8192w8192_c &
6096 dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc
6098 kill -USR1 $pid && wait $pid || error "multiop failure"
6099 check_mdt_xtimes $mdtdev
6104 # Verify mtime/ctime is NOT upated on MDS when there is no modification
6105 # on the client side
6106 xtimes=($(get_stat_xtimes))
6107 $MULTIOP $DIR/$tfile O_c &
6110 kill -USR1 $pid && wait $pid || error "multiop failure"
6111 mdt_xtimes=($(get_mdt_xtimes $mdtdev))
6112 [[ ${xtimes[1]} == ${mdt_xtimes[1]} ]] ||
6113 error "$DIR/$tfile mtime (${xtimes[1]}:${mdt_xtimes[1]}) diff"
6114 [[ ${xtimes[2]} == ${mdt_xtimes[2]} ]] ||
6115 error "$DIR/$tfile ctime (${xtimes[2]}:${mdt_xtimes[2]}) diff"
6116 check_mdt_xtimes $mdtdev
6119 # Change ctime via chmod
6120 $MULTIOP $DIR/$tfile o_tc &
6123 kill -USR1 $pid && wait $pid || error "multiop failure"
6124 check_mdt_xtimes $mdtdev
6126 run_test 104 "Verify that MDS stores atime/mtime/ctime during close"
6129 test_mkdir -p $DIR/$tdir
6130 echo test > $DIR/$tdir/$tfile
6131 $LCTL set_param fail_loc=0x416
6132 cancel_lru_locks osc & sleep 1
6133 fsize1=$(stat -c %s $DIR2/$tdir/$tfile)
6135 [[ $fsize1 = 5 ]] || error "Glimpse returned wrong file size $fsize1"
6137 run_test 105 "Glimpse and lock cancel race"
6140 [ "$mds1_FSTYPE" == "ldiskfs" ] && statx_supported ||
6141 skip_env "Test only for ldiskfs and statx() supported"
6146 local mdtdev=$(mdsdevname ${SINGLEMDS//mds/})
6148 dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc
6149 btime=$($STATX -c %W $DIR/$tfile)
6150 output=$(do_facet mds1 "$DEBUGFS -c -R 'stat ROOT/$tfile' $mdtdev")
6152 ((mdt_btime=$(awk -F ':' /crtime/'{ print $2 }' <<< "$output")))
6153 [[ $btime == $mdt_btime ]] ||
6154 error "$DIR/$tfile btime ($btime:$mdt_btime) diff"
6157 run_test 106a "Verify the btime via statx()"
6160 statx_supported || skip_env "statx() only test"
6165 $LFS setstripe -c 1 $DIR/$tfile || error "$DIR/$tfile setstripe failed"
6166 dd if=/dev/zero of=$DIR/$tfile bs=1k count=1 conv=notrunc
6167 cancel_lru_locks $OSC
6168 rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6170 rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6171 [ $rpcs_after -eq $((rpcs_before + 1)) ] ||
6172 error "$STATX should send 1 glimpse RPC to $OSC"
6174 cancel_lru_locks $OSC
6175 rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6176 # %n: FILENAME; %i: STATX_INO; %A STATX_MODE; %h STATX_NLINK;
6177 # %u: STATX_UID; %g: STATX_GID; %W STATX_BTIME; %X STATX_ATIME;
6179 $STATX -c "%n %i %A %h %u %g %W %X %Z" $DIR/$tfile
6180 rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6181 [ $rpcs_after -eq $rpcs_before ] ||
6182 error "$STATX should not send glimpse RPCs to $OSC"
6184 cancel_lru_locks $OSC
6185 rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6186 $STATX --cached=always $DIR/$tfile
6187 rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6188 [ $rpcs_after -eq $rpcs_before ] ||
6189 error "$STATX should not send glimpse RPCs to $OSC"
6191 cancel_lru_locks $OSC
6192 rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6193 $STATX -c %Y $DIR/$tfile
6194 rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6195 [ $rpcs_after -eq $((rpcs_before + 1)) ] ||
6196 error "$STATX -c %Y should send 1 glimpse RPC to $OSC"
6198 cancel_lru_locks $OSC
6199 rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6200 $STATX -c %s $DIR/$tfile
6201 rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6202 [ $rpcs_after -eq $((rpcs_before + 1)) ] ||
6203 error "$STATX -c %s should send 1 glimpse RPC to $OSC"
6205 cancel_lru_locks $OSC
6206 rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6207 $STATX -c %b $DIR/$tfile
6208 rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue)
6209 [ $rpcs_after -eq $((rpcs_before + 1)) ] ||
6210 error "$STATX -c %b should send 1 glimpse RPC to $OSC"
6212 run_test 106b "Glimpse RPCs test for statx"
6215 statx_supported || skip_env "statx() only test"
6220 # Mask supported in stx_attributes by Lustre is
6221 # STATX_ATTR_IMMUTABLE(0x10) | STATX_ATTR_APPEND(0x20) : (0x30).
6222 mask=$($STATX -c %p $DIR/$tfile)
6223 (( (0x$mask & 0x30) == 0x30 )) ||
6224 error "supported stx_attributes: got '0x$mask', expected '0x30' at least"
6225 chattr +i $DIR/$tfile || error "chattr +i $DIR/$tfile failed"
6226 mask=$($STATX -c %r $DIR/$tfile)
6227 [[ $mask == "10" ]] ||
6228 error "got immutable flags '$mask', expected '10'"
6229 chattr -i $DIR/$tfile || error "chattr -i $DIR/$tfile failed"
6230 mask=$($STATX -c %r $DIR/$tfile)
6231 [[ $mask == "0" ]] || error "got flags '$mask', expected '0'"
6232 chattr +a $DIR/$tfile || error "chattr +a $DIR/$tfile failed"
6233 mask=$($STATX -c %r $DIR/$tfile)
6234 [[ $mask == "20" ]] || error "got flags '$mask', expected '20'"
6235 chattr -a $DIR/$tfile || error "chattr -a $DIR/$tfile failed"
6236 mask=$($STATX -c %r $DIR/$tfile)
6237 [[ $mask == "0" ]] || error "got flags '$mask', expected '0'"
6238 chattr +ia $DIR/$tfile || error "chattr +ia $DIR/$tfile failed"
6239 mask=$($STATX -c %r $DIR/$tfile)
6240 [[ $mask == "30" ]] || error "got flags '$mask', expected '30'"
6241 chattr -ia $DIR/$tfile || error "chattr -ia $DIR/$tfile failed"
6242 mask=$($STATX -c %r $DIR/$tfile)
6243 [[ $mask == "0" ]] || error "got flags '$mask', expected '0'"
6245 run_test 106c "Verify statx attributes mask"
6247 test_107a() { # LU-1031
6248 dd if=/dev/zero of=$DIR1/$tfile bs=1M count=10
6252 $LFS getstripe $DIR1/$tfile
6254 multiop_bg_pause $DIR1/$tfile OG${gid1}_g${gid1}c || return 1
6256 multiop_bg_pause $DIR2/$tfile O_G${gid2}r10g${gid2}c || return 2
6258 kill -USR1 $MULTIPID2
6260 if [[ $(ps h -o comm -p $MULTIPID2) == "" ]]; then
6261 error "First grouplock does not block second one"
6263 echo "First grouplock blocks second one"
6265 kill -USR1 $MULTIPID1
6269 run_test 107a "Basic grouplock conflict"
6272 dd if=/dev/zero of=$DIR1/$tfile bs=1M count=10
6276 $LFS getstripe $DIR1/$tfile
6278 multiop_bg_pause $DIR1/$tfile OG${gid1}_g${gid1}c || return 1
6280 multiop $DIR2/$tfile Or10c &
6284 if [[ $(ps h -o comm -p $MULTIPID2) == "" ]]; then
6285 error "Grouplock does not block IO"
6287 echo "Grouplock blocks IO"
6290 multiop $DIR2/$tfile OG${gid2}_g${gid2}c &
6293 if [[ $(ps h -o comm -p $MULTIPID3) == "" ]]; then
6294 error "First grouplock does not block second one"
6296 echo "First grouplock blocks second one"
6299 kill -USR1 $MULTIPID1
6302 if [[ $(ps h -o comm -p $MULTIPID3) == "" ]]; then
6303 error "Second grouplock thread disappeared"
6306 if [[ $(ps h -o comm -p $MULTIPID2) == "" ]]; then
6307 error "Second grouplock does not block IO"
6309 echo "Second grouplock blocks IO"
6312 kill -USR1 $MULTIPID3
6317 run_test 107b "Grouplock is added to the head of waiting list"
6322 $LFS setstripe -E 1M -c 1 -E -1 $DIR1/$tfile ||
6323 error "Create $DIR1/$tfile failed"
6325 dd if=/dev/zero of=$DIR1/$tfile bs=10000 count=1 ||
6326 error "dd $DIR1/$tfile failed"
6327 offset=$(lseek_test -d 5000 $DIR2/$tfile)
6328 [[ $offset == 5000 ]] || error "offset $offset != 5000"
6330 $TRUNCATE $DIR1/$tfile 2000
6331 offset=$(lseek_test -l 1000 $DIR2/$tfile)
6332 [[ $offset == 2000 ]] || error "offset $offset != 2000"
6334 #define OBD_FAIL_OSC_DELAY_IO 0x414
6335 $LCTL set_param fail_val=4 fail_loc=0x80000414
6336 dd if=/dev/zero of=$DIR1/$tfile count=1 bs=8M conv=notrunc oflag=dsync &
6340 offset=$(lseek_test -l 8000 $DIR2/$tfile)
6342 [[ $offset == 8388608 ]] || error "offset $offset != 8388608"
6344 run_test 108a "lseek: parallel updates"
6352 skip "Clients need to be on different nodes than the servers"
6354 umount_client $MOUNT
6355 umount_client $MOUNT2
6357 echo "Starting race between client mount instances (50 iterations):"
6358 for i in {1..50}; do
6361 #define CFS_FAIL_ONCE|OBD_FAIL_LLITE_RACE_MOUNT 0x80001417
6362 $LCTL set_param -n fail_loc=0x80001417
6364 mount_client $MOUNT & pid1=$!
6365 mount_client $MOUNT2 & pid2=$!
6366 wait $pid1 || error "Mount $MOUNT fails with $?"
6367 wait $pid2 || error "Mount $MOUNT2 fails with $?"
6369 umount_client $MOUNT & pid1=$!
6370 umount_client $MOUNT2 & pid2=$!
6371 wait $pid1 || error "Umount $MOUNT fails with $?"
6372 wait $pid2 || error "Umount $MOUNT2 fails with $?"
6374 $LUSTRE_RMMOD || error "Fail to remove lustre modules"
6380 mount_client $MOUNT2
6383 run_test 109 "Race with several mount instances on 1 node"
6386 local before=$(date +%s)
6393 #define OBD_FAIL_PTLRPC_RESEND_RACE 0x525
6394 do_facet mds1 lctl set_param fail_loc=0x525 fail_val=3
6396 # disable last_xid logic by dropping link reply
6397 ln $DIR/$tdir/f1 $DIR/$tdir/f2 &
6400 #define OBD_FAIL_PTLRPC_ENQ_RESEND 0x534
6401 do_facet mds1 lctl set_param fail_loc=0x534
6403 # RPC will race with its Resend and the Resend will sleep to let
6404 # the original lock to get granted & cancelled.
6406 # AST_SENT is set artificially, so an explicit conflict is not needed
6408 # The woken up Resend gets a new lock, but client does not wait for it
6411 do_facet mds1 lctl set_param fail_loc=0 fail_val=0
6413 # Take a conflict to wait long enough to see the eviction
6416 # let the client reconnect
6418 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
6419 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
6421 [ -z "$evict" ] || [[ $evict -le $before ]] ||
6422 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
6423 error "eviction happened: $evict before:$before")
6425 run_test 110 "do not grant another lock on resend"
6428 [ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs"
6429 [[ $(facet_active_host mds1) = $(facet_active_host mds2) ]] ||
6430 skip "MDT0 and MDT1 should be on the same node"
6433 $LFS mkdir -i 0 $DIR1/$tdir/mdt0dir
6434 $LFS mkdir -i 1 $DIR1/$tdir/mdt1dir
6436 mkdir $DIR1/$tdir/mdt0dir/foodir
6437 touch $DIR1/$tdir/mdt0dir/foodir/{file1,file2}
6439 $MULTIOP $DIR2/$tdir/mdt0dir/foodir/file2 Ow4096_c &
6441 ln $DIR1/$tdir/mdt0dir/foodir/file2 $DIR1/$tdir/mdt1dir/file2
6443 #define OBD_FAIL_MDS_LINK_RENAME_RACE 0x18a
6444 do_facet mds1 $LCTL set_param fail_loc=0x8000018a
6446 ln $DIR1/$tdir/mdt0dir/foodir/file2 $DIR1/$tdir/mdt1dir/file2x &
6449 rm $DIR2/$tdir/mdt1dir/file2
6452 mv $DIR2/$tdir/mdt0dir/foodir/file1 $DIR2/$tdir/mdt0dir/foodir/file2
6457 rm -r $DIR1/$tdir || error "Removing test dir failed"
6459 run_test 111 "A racy rename/link an open file should not cause fs corruption"
6462 (( MDSCOUNT >= 2 )) ||
6463 skip "We need at least 2 MDTs for this test"
6465 (( MDS1_VERSION >= $(version_code 2.14.54) )) ||
6466 skip "Need server version at least 2.14.54"
6471 rr=$($LCTL get_param -n lmv.*.qos_threshold_rr | head -n1)
6473 stack_trap "$LCTL set_param lmv.*.qos_threshold_rr=$rr > /dev/null"
6475 mkdir -p $DIR1/$tdir/s1/s2 || error "mkdir s2 failed"
6476 $LFS mkdir -i 0 $DIR1/$tdir/s1/s2/s3 || error "mkdir s3 failed"
6477 $LFS setdirstripe -D -i -1 --max-inherit-rr=0 $DIR1/$tdir/s1/s2/s3 ||
6478 error "setdirstripe s3 failed"
6479 $LCTL set_param lmv.*.qos_threshold_rr=90
6480 mkdir $DIR2/$tdir/s1/s2/s3/d{1..64}
6481 count=$($LFS getstripe -m $DIR2/$tdir/s1/s2/s3/d* | grep ^0 | wc -l)
6482 (( count == 64 )) || error "only $count subdirs created on MDT0"
6484 $LFS setdirstripe -D -i -1 --max-inherit-rr=3 $DIR1/$tdir/s1/s2/s3 ||
6485 error "setdirstripe s3 failed"
6486 mkdir $DIR2/$tdir/s1/s2/s3/s{1..64}
6487 count=$($LFS getstripe -m $DIR2/$tdir/s1/s2/s3/s* | grep ^0 | wc -l)
6488 (( count == 64 / MDSCOUNT )) || error "$count subdirs created on MDT0"
6490 run_test 112 "update max-inherit in default LMV"
6493 (( MDS1_VERSION >= $(version_code 2.15.50) )) ||
6494 skip "Need server version at least 2.15.50"
6499 instance=$($LFS getname -i $DIR1) ||
6500 error "cannot get instance of $DIR1"
6502 $LFS check osts $DIR1 | grep $instance ||
6503 error "cannot find OSTs of instance $instance"
6505 $LFS check osts $DIR1 | grep -v $instance
6506 if (( $? == 0 )); then
6507 error "find OSTs other than instance $instance"
6510 $LFS check osts | grep $instance ||
6511 error "cannot find other OSTs"
6513 nid=$(df $DIR2 | tail -1 | sed 's%:/.*%%') ||
6514 error "cannot parse nid for $DIR2"
6516 $LFS check mgts $DIR2 | grep MGC$nid ||
6517 error "cannot find mgc of $nid"
6519 $LFS check mgts $DIR2 | grep -v MGC$nid
6520 if (( $? == 0 )); then
6521 error "find MGTs other than nid $nid"
6524 run_test 113 "check servers of specified fs"
6526 check_default_lmv() {
6537 enabled=$(do_facet mds1 \
6538 $LCTL get_param -n mdt.*-MDT0000*.enable_dmv_implicit_inherit)
6540 dmv=$($LFS getdirstripe -D $dir)
6542 index=$(echo $dmv | awk '{ print $4 }')
6543 (( index == $2 )) || error "$dir default stripe index $index != $2"
6545 count=$(echo $dmv | awk '{ print $2 }')
6546 (( count == $3 )) || error "$dir default stripe count $count != $3"
6548 inherit=$(echo $dmv | awk '{ print $8 }')
6549 (( inherit == $4 )) || error "$dir default max-inherit $inherit != $4"
6551 if [ $index -eq -1 ]; then
6552 inherit_rr=$(echo $dmv | awk '{ print $10 }')
6553 (( inherit_rr == $5 )) ||
6554 error "$dir default max-inherit-rr $inherit_rr != $5"
6557 # with --raw, print default LMV stored in inode, otherwise print nothing
6558 raw=$($LFS getdirstripe -D --raw $dir)
6559 if (( enabled == 1 )); then
6561 error "implicit inherited DMV is printed with --raw"
6563 # if disabled, dmv is stored in inode, which will always
6564 # print max-inherit-rr
6566 [[ $raw =~ $dmv.* ]] || error "$dir raw $raw != dmv $dmv"
6570 test_dmv_imp_inherit() {
6578 rm -rf $DIR/$tdir || error "rm $tdir failed"
6579 mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
6581 # set dir default LMV
6582 $LFS setdirstripe -D -c1 -X4 --max-inherit-rr 2 $DIR/$tdir ||
6583 error "setdirstripe -D $tdir failed"
6584 dmv=$($LFS getdirstripe -D $DIR/$tdir)
6585 raw=$($LFS getdirstripe -D --raw $DIR/$tdir)
6586 [ "$dmv" == "$raw" ] || error "$dmv != $raw"
6588 mkdir -p $DIR/$tdir/l1/l2/l3 || error "mkdir $DIR/$tdir/l1/l2/l3 failed"
6589 check_default_lmv $DIR/$tdir/l1/l2/l3 -1 1 1 0
6590 check_default_lmv $DIR2/$tdir/l1/l2/l3 -1 1 1 0
6592 # below tests are valid only when this feature is enabled
6593 local enabled=$(do_facet mds1 \
6594 $LCTL get_param -n mdt.*-MDT0000*.enable_dmv_implicit_inherit)
6596 (( enabled == 1 )) || return 0
6598 # set l2 default LMV, dmv of l3 should change immediately
6599 $LFS setdirstripe -D -i1 -c2 -X4 $DIR/$tdir/l1/l2 ||
6600 error "setdirstripe -D $tdir/l1/l2 failed"
6602 check_default_lmv $DIR/$tdir/l1/l2/l3 1 2 3
6603 check_default_lmv $DIR2/$tdir/l1/l2/l3 1 2 3
6605 # change tdir default LMV, dmv of l3 should be unchanged because dmv
6606 # of l2 is explicitly set
6607 $LFS setdirstripe -D -i2 -c2 -X3 $DIR/$tdir ||
6608 error "setdirstripe -D $tdir failed"
6610 check_default_lmv $DIR/$tdir/l1 2 2 2
6611 check_default_lmv $DIR2/$tdir/l1 2 2 2
6612 check_default_lmv $DIR/$tdir/l1/l2/l3 1 2 3
6613 check_default_lmv $DIR2/$tdir/l1/l2/l3 1 2 3
6617 (( MDSCOUNT >= 2 )) ||
6618 skip "We need at least 2 MDTs for this test"
6620 (( MDS1_VERSION >= $(version_code 2.15.55.45) )) ||
6621 skip "Need server version at least 2.15.54.45"
6623 test_dmv_imp_inherit
6625 # disable dmv_imp_inherit to simulate old client
6626 local mdts=$(comma_list $(mdts_nodes))
6628 do_nodes $mdts $LCTL set_param -n \
6629 mdt.*MDT*.enable_dmv_implicit_inherit=0
6630 test_dmv_imp_inherit
6631 do_nodes $mdts $LCTL set_param -n \
6632 mdt.*MDT*.enable_dmv_implicit_inherit=1
6634 run_test 114 "implicit default LMV inherit"
6639 [ "$mds1_FSTYPE" == "ldiskfs" ] || skip_env "ldiskfs only test"
6641 mkdir_on_mdt0 $td || error "can't mkdir"
6642 # turn it htree (don't really needed)
6643 createmany -m $td/f 3000 || error "can't createmany"
6645 # here is an example of debugfs output for htree command:
6646 # Entry #0: Hash 0x00000000, block 27
6647 # Reading directory block 27, phys 16760
6648 # 938 0x0016fb58-7f3d21f5 (32) f775 834 0x001db8c8-d31a4e0e (32) f671
6649 # 1085 0x0040cb70-4498abd4 (32) f922 1850 0x0066a1e6-f6f0dc69 (32) f1687
6650 # 2005 0x006c1a46-ef466058 (32) f1842 2025 0x007e64d4-8b28b734 (32) f1862
6651 # 642 0x008b53a0-77adc601 (32) f479 447 0x009ec152-af54eea3 (32) f284
6652 # 1740 0x00c38f56-ed310e61 (32) f1577 2165 0x00cdfd66-f429a93f (32) f2002
6653 # 930 0x00d7ada4-b80421c9 (32) f767 1946 0x00da6a7a-e8080600 (32) f1783
6654 # 273 0x00f8ea00-760bf97c (32) f110 1589 0x0103c4ee-94fad5dd (32) f1426
6655 # 1383 0x01193516-83120b48 (32) f1220 2379 0x01431e3c-e85b5bd9 (32) f2216
6657 # find couple names in a same htree block of the same size
6658 mdt_dev=$(facet_device $SINGLEMDS)
6659 de=( $(do_facet $SINGLEMDS "debugfs -c -R 'htree /ROOT/$tdir' $mdt_dev" |
6660 awk '/Reading directory block/ { getline; print $4,$8; exit; }' ))
6663 [[ $de1 == "" || $de2 == "" ]] && error "de1=$de1 de2=$de2"
6664 echo "USE: $de1 $de2"
6665 # release one mkdir will lookup
6667 #define OBD_FAIL_MDS_PAUSE_CREATE_AFTER_LOOKUP 0x2401
6668 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80002401 fail_val=5
6669 mkdir $DIR/$tdir/$de2 &
6673 mkdir $DIR2/$tdir/$de2
6674 # release space $de1 (should be enough to save $de2)
6676 # ready to create a dup of $de2
6678 local found=$(ls $DIR/$tdir/|grep "^$de2\$"|wc -l)
6679 (( $found == 1 )) || error "found $found"
6681 run_test 115 "ldiskfs doesn't check direntry for uniqueness"
6683 log "cleanup: ======================================================"
6685 # kill and wait in each test only guarentee script finish, but command in script
6686 # like 'rm' 'chmod' may still be running, wait for all commands to finish
6687 # otherwise umount below will fail
6688 [ "$(mount | grep $MOUNT2)" ] && wait_update $HOSTNAME "fuser -m $MOUNT2" "" ||
6691 complete_test $SECONDS
6693 check_and_cleanup_lustre