6 # bug number for skipped test: 9977/LU-7105
8 ALWAYS_EXCEPT=" 28 $SANITYN_EXCEPT"
9 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
12 PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH
15 CHECKSTAT=${CHECKSTAT:-"checkstat -v"}
16 OPENFILE=${OPENFILE:-openfile}
17 OPENUNLINK=${OPENUNLINK:-openunlink}
18 export MULTIOP=${MULTIOP:-multiop}
19 export TMP=${TMP:-/tmp}
20 MOUNT_2=${MOUNT_2:-"yes"}
21 CHECK_GRANT=${CHECK_GRANT:-"yes"}
22 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
26 export NAME=${NAME:-local}
28 LUSTRE=${LUSTRE:-`dirname $0`/..}
29 . $LUSTRE/tests/test-framework.sh
33 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
37 if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then
38 # bug number for skipped test:
39 ALWAYS_EXCEPT="$ALWAYS_EXCEPT "
40 # LU-2829 / LU-2887 - make allowances for ZFS slowness
41 TEST33_NFILES=${TEST33_NFILES:-1000}
44 [ "$SLOW" = "no" ] && EXCEPT_SLOW="33a"
51 check_and_setup_lustre
56 rm -rf $DIR1/[df][0-9]* $DIR1/lnk $DIR/[df].${TESTSUITE}*
58 SAMPLE_FILE=$TMP/$(basename $0 .sh).junk
59 dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=1
61 # $RUNAS_ID may get set incorrectly somewhere else
62 [ $UID -eq 0 -a $RUNAS_ID -eq 0 ] && error "\$RUNAS_ID set to 0, but \$UID is also 0!"
64 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
70 [ -f $DIR2/$tfile ] || error "Check create"
71 chmod 777 $DIR2/$tfile
72 $CHECKSTAT -t file -p 0777 $DIR1/$tfile ||
73 error "Check attribute update for 0777"
75 chmod a-x $DIR2/$tfile
76 $CHECKSTAT -t file -p 0666 $DIR1/$tfile ||
77 error "Check attribute update for 0666"
80 $CHECKSTAT -a $DIR1/$tfile ||
81 error "Check unlink - removes file on other mountpoint"
83 run_test 1 "Check attribute updates on 2 mount points"
89 $CHECKSTAT -t file -p 0777 $DIR1/f2a ||
90 error "Either not file type or perms not 0777"
92 run_test 2a "check cached attribute updates on 2 mtpt's ========"
98 $CHECKSTAT -t file -p 0777 $DIR2/f2b ||
99 error "Either not file type or perms not 0777"
101 run_test 2b "check cached attribute updates on 2 mtpt's ========"
103 # NEED TO SAVE ROOT DIR MODE
106 $CHECKSTAT -t dir -p 0777 $DIR2 ||
107 error "Either not dir type or perms not 0777"
109 run_test 2c "check cached attribute updates on 2 mtpt's root ==="
113 $CHECKSTAT -t dir -p 0755 $DIR2 ||
114 error "Either not file type or perms not 0775"
116 run_test 2d "check cached attribute updates on 2 mtpt's root ==="
123 $RUNAS dd if=/dev/zero of=$DIR2/$tfile count=1 ||
126 run_test 2e "check chmod on root is propagated to others"
129 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
131 local remote_dir=$tdir/remote_dir
134 $LFS mkdir -i $MDTIDX $DIR1/$remote_dir ||
135 error "Create remote directory failed"
137 touch $DIR1/$remote_dir/$tfile ||
138 error "Create file under remote directory failed"
139 chmod 777 $DIR1/$remote_dir/$tfile ||
140 error "Chmod file under remote directory failed"
142 $CHECKSTAT -t file -p 0777 $DIR2/$remote_dir/$tfile ||
143 error "Check attr of file under remote directory failed"
145 chown $RUNAS_ID:$RUNAS_GID $DIR1/$remote_dir/$tfile ||
146 error "Chown file under remote directory failed"
148 $CHECKSTAT -u \#$RUNAS_ID -g \#$RUNAS_GID $DIR2/$remote_dir/$tfile ||
149 error "Check owner of file under remote directory failed"
151 cd $DIR2/$remote_dir || error "enter remote dir"
152 rm -rf $DIR1/$remote_dir/$tfile ||
153 error "Unlink remote directory failed"
155 $CHECKSTAT -t file $DIR2/$remote_dir/$tfile &&
156 error "unlink file still exists!"
158 cd $DIR2/$tdir || error "exit remote dir"
159 rm -rf $DIR1/$tdir || error "unlink directory failed"
161 run_test 2f "check attr/owner updates on DNE with 2 mtpt's"
164 dd if=/dev/zero of=$DIR1/$tfile oflag=sync bs=1M count=2
166 local block1=$(stat $DIR1/$tfile | awk '/Blocks/ {print $4} ')
168 local block2=$(stat $DIR2/$tfile | awk '/Blocks/ {print $4} ')
169 echo "$DIR1/$tfile has $block1 blocks"
170 echo "$DIR2/$tfile has $block2 blocks"
171 [ $block1 -eq $block2 ] || error "$block1 not equal to $block2"
173 run_test 2g "check blocks update on sync write"
176 local target="this/is/good"
177 ln -s $target $DIR1/$tfile || error "ln -s $target $DIR1/$tfile failed"
178 [ "$(ls -l $DIR2/$tfile | sed -e 's/.* -> //')" = "$target" ] ||
179 error "link $DIR2/$tfile not as expected"
181 run_test 3 "symlink on one mtpt, readlink on another ==========="
184 multifstat $DIR1/f4 $DIR2/f4
186 run_test 4 "fstat validation on multiple mount points =========="
190 $TRUNCATE $DIR2/f5 100
191 $CHECKSTAT -t file -s 100 $DIR1/f5 ||
192 error "Either not file type or size not equal to 100 bytes"
195 run_test 5 "create a file on one mount, truncate it on the other"
198 openunlink $DIR1/$tfile $DIR2/$tfile || \
199 error "openunlink $DIR1/$tfile $DIR2/$tfile"
201 run_test 6 "remove of open file on other node =================="
205 opendirunlink $DIR1/$dir $DIR2/$dir || \
206 error "opendirunlink $DIR1/$dir $DIR2/$dir"
208 run_test 7 "remove of open directory on other node ============="
211 opendevunlink $DIR1/$tfile $DIR2/$tfile || \
212 error "opendevunlink $DIR1/$tfile $DIR2/$tfile"
214 run_test 8 "remove of open special file on other node =========="
220 for C in a b c d e f g h i j k l; do
221 dir=`eval echo \\$DIR$MTPT`
222 echo -n $C >> $dir/f9
223 [ "$MTPT" -eq 1 ] && MTPT=2 || MTPT=1
225 [ "`cat $DIR1/f9`" = "abcdefghijkl" ] || \
226 error "`od -a $DIR1/f9` != abcdefghijkl"
228 run_test 9a "append of file with sub-page size on multiple mounts"
230 #LU-10681 - tiny writes & appending to sparse striped file
232 [[ $OSTCOUNT -ge 2 ]] || { skip "needs >= 2 OSTs"; return; }
234 $LFS setstripe -c 2 -S 1M $DIR/$tfile
235 echo "foo" >> $DIR/$tfile
236 dd if=/dev/zero of=$DIR2/$tfile bs=1M count=1 seek=1 conv=notrunc ||
237 error "sparse dd $DIR2/$tfile failed"
238 echo "foo" >> $DIR/$tfile
240 data=$(dd if=$DIR2/$tfile bs=1 count=3 skip=$((2 * 1048576)) conv=notrunc)
241 echo "Data read (expecting 'foo')": $data
242 [ "$data" = "foo" ] || error "append to sparse striped file failed"
244 run_test 9b "append to striped sparse file"
251 for C in a b c d e f g h i j k l; do
252 dir=`eval echo \\$DIR$MTPT`
253 echo -n $C | dd of=$dir/f10 bs=1 seek=$OFFSET count=1
254 [ "$MTPT" -eq 1 ] && MTPT=2 || MTPT=1
255 OFFSET=`expr $OFFSET + 1`
257 [ "`cat $DIR1/f10`" = "abcdefghijkl" ] || \
258 error "`od -a $DIR1/f10` != abcdefghijkl"
260 run_test 10a "write of file with sub-page size on multiple mounts "
264 yes "R" | head -c 4000 >$TMP/f10b-seed
265 dd if=$TMP/f10b-seed of=$DIR1/f10b bs=3k count=1 || error "dd $DIR1"
267 $TRUNCATE $DIR1/f10b 4096 || error "truncate 4096"
269 dd if=$DIR2/f10b of=$TMP/f10b-lustre bs=4k count=1 || error "dd $DIR2"
271 # create a test file locally to compare
272 dd if=$TMP/f10b-seed of=$TMP/f10b bs=3k count=1 || error "dd random"
273 $TRUNCATE $TMP/f10b 4096 || error "truncate 4096"
274 cmp $TMP/f10b $TMP/f10b-lustre || error "file miscompare"
275 rm $TMP/f10b $TMP/f10b-lustre $TMP/f10b-seed
277 run_test 10b "write of file with sub-page size on multiple mounts "
281 multiop_bg_pause $DIR1/d11/f O_c || return 1
283 cp -p /bin/ls $DIR1/d11/f
287 wait $MULTIPID || error "wait for PID $MULTIPID failed"
288 [ $RC -eq 0 ] && error || true
290 run_test 11 "execution of file opened for write should return error ===="
293 DIR=$DIR DIR2=$DIR2 sh lockorder.sh
295 run_test 12 "test lock ordering (link, stat, unlink)"
297 test_13() { # bug 2451 - directory coherency
299 cd $DIR1/d13 || error "cd to $DIR1/d13 failed"
301 ( touch $DIR1/d13/f13 ) # needs to be a separate shell
303 rm -f $DIR2/d13/f13 || error "Cannot remove $DIR2/d13/f13"
304 ls 2>&1 | grep f13 && error "f13 shouldn't return an error (1)" || true
305 # need to run it twice
306 ( touch $DIR1/d13/f13 ) # needs to be a separate shell
308 rm -f $DIR2/d13/f13 || error "Cannot remove $DIR2/d13/f13"
309 ls 2>&1 | grep f13 && error "f13 shouldn't return an error (2)" || true
311 run_test 13 "test directory page revocation"
314 test_mkdir $DIR1/$tdir
315 cp -p /bin/ls $DIR1/$tdir/$tfile
316 multiop_bg_pause $DIR1/$tdir/$tfile Ow_c || return 1
319 $DIR2/$tdir/$tfile && error || true
322 run_test 14aa "execution of file open for write returns -ETXTBSY"
325 test_mkdir $DIR1/$tdir
326 cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed"
327 $DIR1/$tdir/sleep 60 &
329 $MULTIOP $DIR2/$tdir/sleep Oc && error "expected error, got success"
332 run_test 14ab "open(RDWR) of executing file returns -ETXTBSY"
334 test_14b() { # bug 3192, 7040
335 test_mkdir $DIR1/$tdir
336 cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed"
337 $DIR1/$tdir/sleep 60 &
339 $TRUNCATE $DIR2/$tdir/sleep 60 && kill -9 $SLEEP_PID && \
340 error "expected truncate error, got success"
342 cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed"
344 run_test 14b "truncate of executing file returns -ETXTBSY ======"
346 test_14c() { # bug 3430, 7040
347 test_mkdir $DIR1/$tdir
348 cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed"
349 $DIR1/$tdir/sleep 60 &
351 cp /etc/hosts $DIR2/$tdir/sleep && error "expected error, got success"
353 cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed"
355 run_test 14c "open(O_TRUNC) of executing file return -ETXTBSY =="
357 test_14d() { # bug 10921
358 test_mkdir $DIR1/$tdir
359 cp -p $(which sleep) $DIR1/$tdir/sleep || error "cp failed"
360 $DIR1/$tdir/sleep 60 &
363 chmod 600 $DIR1/$tdir/sleep || error "chmod failed"
365 cmp $(which sleep) $DIR1/$tdir/sleep || error "binary changed"
367 run_test 14d "chmod of executing file is still possible ========"
369 test_15() { # bug 974 - ENOSPC
371 sh oos2.sh $MOUNT1 $MOUNT2
372 wait_delete_completed
373 grant_error=`dmesg | grep "> available"`
374 [ -z "$grant_error" ] || error "$grant_error"
376 run_test 15 "test out-of-space with multiple writers ==========="
379 # The FSXNUM reduction for ZFS is needed until ORI-487 is fixed.
380 # We don't want to skip it entirely, but ZFS is VERY slow and cannot
381 # pass a 2500 operation dual-mount run within the time limit.
382 if [ "$(facet_fstype ost1)" = "zfs" ]; then
383 FSXNUM=$((COUNT / 5))
385 elif [ "$SLOW" = "yes" ]; then
386 FSXNUM=$((COUNT * 5))
394 local file1=$DIR1/$tfile
395 local file2=$DIR2/$tfile
396 local stripe_size=$(do_facet $SINGLEMDS \
397 "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
399 # to allocate grant because it may run out due to test_15.
400 $LFS setstripe -c -1 $file1
401 dd if=/dev/zero of=$file1 bs=$stripe_size count=$OSTCOUNT oflag=sync
402 dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync
405 $LFS setstripe -c -1 $file1 # b=10919
406 fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 \
407 || error "fsx failed"
410 # O_DIRECT reads and writes must be aligned to the device block size.
411 fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 -Z -r 4096 \
412 -w 4096 $file1 $file2 || error "fsx with O_DIRECT failed."
414 run_test 16a "$FSXNUM iterations of dual-mount fsx"
416 # Consistency check for tiny writes, LU-9409
418 local file1=$DIR1/$tfile
419 local file2=$DIR2/$tfile
421 # to allocate grant because it may run out due to test_15.
422 lfs setstripe -c -1 $file1
423 dd if=/dev/zero of=$file1 bs=$STRIPE_BYTES count=$OSTCOUNT oflag=sync
424 dd if=/dev/zero of=$file2 bs=$STRIPE_BYTES count=$OSTCOUNT oflag=sync
427 lfs setstripe -c -1 $file1 # b=10919
428 # -o is set to 8192 because writes < 1 page and between 1 and 2 pages
429 # create a mix of tiny writes & normal writes
430 fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -o 8192 -S 0 $file1 \
433 run_test 16b "$FSXNUM iterations of dual-mount fsx at small size"
436 local file1=$DIR1/$tfile
437 local file2=$DIR2/$tfile
438 local stripe_size=$(do_facet $SINGLEMDS \
439 "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
441 [ $(facet_fstype ost1) != ldiskfs ] && skip "dio on ldiskfs only"
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 dd if=/dev/zero of=$file2 bs=$stripe_size count=$OSTCOUNT oflag=sync
448 wait_delete_completed
450 local list=$(comma_list $(osts_nodes))
451 if ! get_osd_param $list '' read_cache_enable >/dev/null; then
452 skip "not cache-capable obdfilter"
455 set_osd_param $list '' read_cache_enable 0
456 set_osd_param $list '' writethrough_cache_enable 0
458 $LFS setstripe -c -1 $file1 # b=10919
459 fsx -c 50 -p $FSXP -N $FSXNUM -l $((SIZE * 256)) -S 0 $file1 $file2 \
460 || error "fsx failed"
463 set_osd_param $list '' read_cache_enable 1
464 set_osd_param $list '' writethrough_cache_enable 1
468 run_test 16c "verify data consistency on ldiskfs with cache disabled (b=17397)"
471 test_17() { # bug 3513, 3667
472 remote_ost_nodsh && skip "remote OST with nodsh" && return
474 lfs setstripe $DIR1/$tfile -i 0 -c 1
475 cp $SAMPLE_FILE $DIR1/$tfile
476 cancel_lru_locks osc > /dev/null
477 #define OBD_FAIL_ONCE|OBD_FAIL_LDLM_CREATE_RESOURCE 0x30a
478 do_facet ost1 lctl set_param fail_loc=0x8000030a
479 ls -ls $DIR1/$tfile | awk '{ print $1,$6 }' > $DIR1/$tfile-1 & \
480 ls -ls $DIR2/$tfile | awk '{ print $1,$6 }' > $DIR2/$tfile-2
482 diff -u $DIR1/$tfile-1 $DIR2/$tfile-2 || error "files are different"
484 run_test 17 "resource creation/LVB creation race ==============="
487 # turn e.g. ALWAYS_EXCEPT="18c" into "-e 3"
490 for idx in {a..z}; do
491 local ptr=EXCEPT_ALWAYS_18$idx
492 [ x${!ptr} = xtrue ] || continue
494 excepts="$excepts -e $(($(printf %d \'$idx)-96))"
497 $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2 $excepts
500 run_test 18 "mmap sanity check ================================="
502 test_19() { # bug3811
503 local node=$(facet_active_host ost1)
505 [ "x$DOM" = "xyes" ] && node=$(facet_active_host $SINGLEMDS)
507 # check whether obdfilter is cache capable at all
508 get_osd_param $node '' read_cache_enable >/dev/null ||
509 skip "not cache-capable obdfilter"
511 local MAX=$(get_osd_param $node '' readcache_max_filesize | head -n 1)
512 set_osd_param $node '' readcache_max_filesize 4096
513 dd if=/dev/urandom of=$TMP/$tfile bs=512k count=32
514 local SUM=$(cksum $TMP/$tfile | cut -d" " -f 1,2)
515 cp $TMP/$tfile $DIR1/$tfile
516 for i in `seq 1 20`; do
517 [ $((i % 5)) -eq 0 ] && log "$testname loop $i"
518 cancel_lru_locks $OSC > /dev/null
519 cksum $DIR1/$tfile | cut -d" " -f 1,2 > $TMP/sum1 & \
520 cksum $DIR2/$tfile | cut -d" " -f 1,2 > $TMP/sum2
522 [ "$(cat $TMP/sum1)" = "$SUM" ] || \
523 error "$DIR1/$tfile $(cat $TMP/sum1) != $SUM"
524 [ "$(cat $TMP/sum2)" = "$SUM" ] || \
525 error "$DIR2/$tfile $(cat $TMP/sum2) != $SUM"
527 set_osd_param $node '' readcache_max_filesize $MAX
530 run_test 19 "test concurrent uncached read races ==============="
534 cancel_lru_locks $OSC
535 CNT=$((`lctl get_param -n llite.*.dump_page_cache | wc -l`))
536 $MULTIOP $DIR1/f20 Ow8190c
537 $MULTIOP $DIR2/f20 Oz8194w8190c
538 $MULTIOP $DIR1/f20 Oz0r8190c
539 cancel_lru_locks $OSC
540 CNTD=$((`lctl get_param -n llite.*.dump_page_cache | wc -l` - $CNT))
542 error $CNTD" page left in cache after lock cancel" || true
544 run_test 20 "test extra readahead page left in cache ===="
551 test_21() { # Bug 5907
552 test_mkdir $DIR1/$tdir
553 mount /etc $DIR1/$tdir --bind || error "mount failed" # Poor man's mount.
555 rmdir -v $DIR1/$tdir && error "Removed mounted directory"
556 rmdir -v $DIR2/$tdir && echo "Removed mounted directory from another mountpoint, needs to be fixed"
557 test -d $DIR1/$tdir || error "Mounted directory disappeared"
559 test -d $DIR2/$tdir || test -d $DIR1/$tdir && error "Removed dir still visible after umount"
562 run_test 21 " Try to remove mountpoint on another dir ===="
564 test_23() { # Bug 5972
565 local at_diff=$(do_facet $SINGLEMDS \
566 $LCTL get_param -n mdd.*MDT0000*.atime_diff | head -n1)
567 echo "atime should be updated while another read" > $DIR1/$tfile
569 # clear the lock(mode: LCK_PW) gotten from creating operation
570 cancel_lru_locks $OSC
573 sleep $((at_diff + 1))
575 echo "starting reads"
576 multiop_bg_pause $DIR1/$tfile or20_c || return 1
577 # with SOM and opencache enabled, we need to close a file and cancel
578 # open lock to get atime propogated to MDS
579 kill -USR1 $! || return 2
582 time2=$(stat -c "%X" $DIR/$tfile)
583 echo "new atime is $time2"
585 [ $time2 -gt $time1 ] || error "atime was not updated"
586 rm -f $DIR1/$tfile || error "rm -f $DIR1/$tfile failed"
589 run_test 23 " others should see updated atime while another read===="
593 lfs df || error "lfs df failed"
594 lfs df -ih || error "lfs df -ih failed"
595 lfs df -h $DIR1 || error "lfs df -h $DIR1 failed"
596 lfs df -i $DIR2 || error "lfs df -i $DIR2 failed"
597 lfs df $DIR1/$tfile || error "lfs df $DIR1/$tfile failed"
598 lfs df -ih $DIR2/$tfile || error "lfs df -ih $DIR2/$tfile failed"
600 OSC=`lctl dl | awk '/-osc-|OSC.*MNT/ {print $4}' | head -n 1`
601 # OSC=`lctl dl | awk '/-osc-/ {print $4}' | head -n 1`
602 lctl --device %osc deactivate
603 lfs df -i || error "lfs df -i with deactivated OSC failed"
604 lctl --device %osc activate
605 lfs df || error "lfs df with reactivated OSC failed"
607 run_test 24a "lfs df [-ih] [path] test ========================="
611 fsnum=$(lfs_df | grep -c "summary")
612 [ $fsnum -eq 2 ] || error "lfs df shows $fsnum != 2 filesystems."
614 run_test 24b "lfs df should show both filesystems ==============="
617 local acl=$(lctl get_param -n mdc.*MDT0000-mdc-*.connect_flags |
619 [ "$acl" -lt 1 ] && skip "must have acl, skipping" && return
622 touch $DIR1/$tdir/f1 || error "touch $DIR1/$tdir/f1"
623 chmod 0755 $DIR1/$tdir/f1 || error "chmod 0755 $DIR1/$tdir/f1"
625 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #1"
626 setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir ||
627 error "setfacl $DIR2/$tdir #1"
628 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #2"
629 setfacl -m u:$RUNAS_ID:r-x -m g:$RUNAS_GID:r-x $DIR1/$tdir ||
630 error "setfacl $DIR2/$tdir #2"
631 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #3"
632 setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir ||
633 error "setfacl $DIR2/$tdir #3"
634 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #4"
635 setfacl -x u:$RUNAS_ID: -x g:$RUNAS_GID: $DIR1/$tdir ||
636 error "setfacl $DIR2/$tdir #4"
637 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #5"
641 run_test 25a "change ACL on one mountpoint be seen on another ==="
644 local acl=$(lctl get_param -n mdc.*MDT0000-mdc-*.connect_flags |
646 [ "$acl" -lt 1 ] && skip "must have acl, skipping" && return
648 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
651 $LFS mkdir -i 1 $DIR1/$tdir
652 touch $DIR1/$tdir/f1 || error "touch $DIR1/$tdir/f1"
653 chmod 0755 $DIR1/$tdir/f1 || error "chmod 0755 $DIR1/$tdir/f1"
655 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #1"
656 setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir ||
657 error "setfacl $DIR2/$tdir #1"
658 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #2"
659 setfacl -m u:$RUNAS_ID:r-x -m g:$RUNAS_GID:r-x $DIR1/$tdir ||
660 error "setfacl $DIR2/$tdir #2"
661 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #3"
662 setfacl -m u:$RUNAS_ID:--- -m g:$RUNAS_GID:--- $DIR1/$tdir ||
663 error "setfacl $DIR2/$tdir #3"
664 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 && error "checkstat $DIR2/$tdir/f1 #4"
665 setfacl -x u:$RUNAS_ID: -x g:$RUNAS_GID: $DIR1/$tdir ||
666 error "setfacl $DIR2/$tdir #4"
667 $RUNAS $CHECKSTAT $DIR2/$tdir/f1 || error "checkstat $DIR2/$tdir/f1 #5"
671 run_test 25b "change ACL under remote dir on one mountpoint be seen on another"
674 utime $DIR1/f26a -s $DIR2/f26a || error "utime failed for $DIR1/f26a"
676 run_test 26a "allow mtime to get older"
681 echo "aaa" >> $DIR1/$tfile
683 chmod a+x $DIR2/$tfile
684 mt1=`stat -c %Y $DIR1/$tfile`
685 mt2=`stat -c %Y $DIR2/$tfile`
687 if [ x"$mt1" != x"$mt2" ]; then
688 error "not equal mtime, client1: "$mt1", client2: "$mt2"."
691 run_test 26b "sync mtime between ost and mds"
694 cancel_lru_locks $OSC
696 dd if=/dev/zero of=$DIR2/$tfile bs=$((4096+4))k conv=notrunc count=4 seek=3 &
701 dd if=/dev/zero of=$DIR1/$tfile bs=$((16384-1024))k conv=notrunc count=1 seek=4 &
706 dd if=/dev/zero of=$DIR1/$tfile bs=8k conv=notrunc count=1 seek=0
708 lctl set_param -n ldlm.dump_namespaces ""
709 wait $DD1_PID $DD2_PID
710 [ $? -ne 0 ] && lctl dk $TMP/debug || true
712 run_test 27 "align non-overlapping extent locks from request ==="
714 test_28() { # bug 9977
715 ECHO_UUID="ECHO_osc1_UUID"
716 tOST=$($LCTL dl | awk '/-osc-|OSC.*MNT/ { print $4 }' | head -n1)
718 $LFS setstripe $DIR1/$tfile -S 1048576 -i 0 -c 2
719 tOBJID=`$LFS getstripe $DIR1/$tfile | awk '$1 == 1 {print $2}'`
720 dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2
724 attach echo_client ECHO_osc1 $ECHO_UUID
728 tECHOID=`$LCTL dl | grep $ECHO_UUID | awk '{ print $1 }'`
729 $LCTL --device $tECHOID destroy "${tOBJID}:0"
737 # reading of 1st stripe should pass
738 dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error "dd failed"
739 # reading of 2nd stripe should fail (this stripe was destroyed)
740 dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 && error
742 # now, recreating test file
743 dd if=/dev/zero of=$DIR1/$tfile bs=1024k count=2 || error "dd failed"
744 # reading of 1st stripe should pass
745 dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 || error "dd failed"
746 # reading of 2nd stripe should pass
747 dd if=$DIR2/$tfile of=/dev/null bs=1024k count=1 skip=1 ||
750 run_test 28 "read/write/truncate file with lost stripes"
752 test_30() { #b=11110, LU-2523
753 test_mkdir $DIR1/$tdir
754 cp -f /bin/bash $DIR1/$tdir/bash
755 /bin/sh -c 'sleep 1; rm -f $DIR2/$tdir/bash; cp /bin/bash $DIR2/$tdir' &
756 $DIR1/$tdir/bash -c 'sleep 2;
757 openfile -f O_RDONLY /proc/$$/exe >& /dev/null; echo $?'
761 run_test 30 "recreate file race"
764 test_mkdir $DIR1/$tdir
765 local writes=$(LANG=C dd if=/dev/zero of=$DIR/$tdir/$tfile \
766 count=1 2>&1 | awk 'BEGIN { FS="+" } /out/ {print $1}')
767 #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314
768 lctl set_param fail_loc=0x314
769 local reads=$(LANG=C dd if=$DIR2/$tdir/$tfile of=/dev/null 2>&1 |
770 awk 'BEGIN { FS="+" } /in/ {print $1}')
771 [ $reads -eq $writes ] || error "read" $reads "blocks, must be" $writes
773 run_test 31a "voluntary cancel / blocking ast race=============="
776 remote_ost || { skip "local OST" && return 0; }
777 remote_ost_nodsh && skip "remote OST w/o dsh" && return 0
779 # make sure there is no local locks due to destroy
780 wait_mds_ost_sync || error "wait_mds_ost_sync()"
781 wait_delete_completed || error "wait_delete_completed()"
783 test_mkdir $DIR1/$tdir
784 lfs setstripe $DIR/$tdir/$tfile -i 0 -c 1
785 cp /etc/hosts $DIR/$tdir/$tfile
786 #define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314
787 lctl set_param fail_loc=0x314
788 #define OBD_FAIL_LDLM_OST_FAIL_RACE 0x316
789 do_facet ost1 lctl set_param fail_loc=0x316
791 cat $DIR2/$tdir/$tfile > /dev/null 2>&1
792 lctl set_param fail_loc=0
793 do_facet ost1 lctl set_param fail_loc=0
794 # cleanup: reconnect the client back
797 run_test 31b "voluntary OST cancel / blocking ast race=============="
799 # enable/disable lockless truncate feature, depending on the arg 0/1
800 enable_lockless_truncate() {
801 lctl set_param -n $OSC.*.lockless_truncate $1
804 test_32a() { # bug 11270
805 local save="$TMP/$TESTSUITE-$TESTNAME.parameters"
806 local stripe_size=$(do_facet $SINGLEMDS \
807 "$LCTL get_param -n lod.$(facet_svc $SINGLEMDS)*.stripesize")
809 save_lustre_params client "$OSC.*.lockless_truncate" > $save
810 # restore lockless_truncate default values on exit
811 stack_trap "restore_lustre_params < $save; rm -f $save" EXIT
812 cancel_lru_locks $OSC
813 enable_lockless_truncate 1
815 lfs setstripe -c -1 $DIR1/$tfile
816 dd if=/dev/zero of=$DIR1/$tfile count=$OSTCOUNT bs=$stripe_size > \
818 clear_stats $OSC.*.${OSC}_stats
820 log "checking cached lockless truncate"
821 $TRUNCATE $DIR1/$tfile 8000000
822 $CHECKSTAT -s 8000000 $DIR2/$tfile || error "wrong file size"
823 [ $(calc_stats $OSC.*.${OSC}_stats lockless_truncate) -ne 0 ] ||
824 error "cached truncate isn't lockless"
826 log "checking not cached lockless truncate"
827 $TRUNCATE $DIR2/$tfile 5000000
828 $CHECKSTAT -s 5000000 $DIR1/$tfile || error "wrong file size"
829 [ $(calc_stats $OSC.*.${OSC}_stats lockless_truncate) -ne 0 ] ||
830 error "not cached truncate isn't lockless"
832 log "disabled lockless truncate"
833 enable_lockless_truncate 0
834 clear_stats $OSC.*.${OSC}_stats
835 $TRUNCATE $DIR2/$tfile 3000000
836 $CHECKSTAT -s 3000000 $DIR1/$tfile || error "wrong file size"
837 [ $(calc_stats $OSC.*.${OSC}_stats lockless_truncate) -eq 0 ] ||
838 error "lockless truncate disabling failed"
841 run_test 32a "lockless truncate"
843 test_32b() { # bug 11270
844 remote_ost_nodsh && skip "remote OST with nodsh" && return
847 local facets=$(get_facets OST)
848 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
850 save_lustre_params client "osc.*.contention_seconds" > $p
851 save_lustre_params $facets \
852 "ldlm.namespaces.filter-*.max_nolock_bytes" >> $p
853 save_lustre_params $facets \
854 "ldlm.namespaces.filter-*.contended_locks" >> $p
855 save_lustre_params $facets \
856 "ldlm.namespaces.filter-*.contention_seconds" >> $p
857 clear_stats $OSC.*.${OSC}_stats
859 # agressive lockless i/o settings
860 do_nodes $(comma_list $(osts_nodes)) \
861 "lctl set_param -n ldlm.namespaces.*.max_nolock_bytes=2000000 \
862 ldlm.namespaces.filter-*.contended_locks=0 \
863 ldlm.namespaces.filter-*.contention_seconds=60"
864 lctl set_param -n $OSC.*.contention_seconds=60
866 dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \
868 dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \
871 [ $(calc_stats $OSC.*.${OSC}_stats lockless_write_bytes) -ne 0 ] ||
872 error "lockless i/o was not triggered"
873 # disable lockless i/o (it is disabled by default)
874 do_nodes $(comma_list $(osts_nodes)) \
875 "lctl set_param -n ldlm.namespaces.filter-*.max_nolock_bytes=0 \
876 ldlm.namespaces.filter-*.contended_locks=32 \
877 ldlm.namespaces.filter-*.contention_seconds=0"
878 # set contention_seconds to 0 at client too, otherwise Lustre still
879 # remembers lock contention
880 lctl set_param -n $OSC.*.contention_seconds=0
881 clear_stats $OSC.*.${OSC}_stats
883 dd if=/dev/zero of=$DIR1/$tfile bs=4k count=1 conv=notrunc > \
885 dd if=/dev/zero of=$DIR2/$tfile bs=4k count=1 conv=notrunc > \
888 [ $(calc_stats $OSC.*.${OSC}_stats lockless_write_bytes) -eq 0 ] ||
889 error "lockless i/o works when disabled"
891 restore_lustre_params <$p
894 run_test 32b "lockless i/o"
898 local mdts=$(get_facets MDS)
903 for mds in ${mdts//,/ }; do
905 dev=$(basename $(do_facet $mds "lctl get_param -n osd*.${!varsvc}.mntdev|\
906 xargs readlink -f" ))
907 val=$(do_facet $mds "cat /proc/fs/jbd*/${dev}{,:*,-*}/info 2>/dev/null |
910 stat=$(( stat + val))
915 # commit on sharing tests
917 remote_mds_nodsh && skip "remote MDS with nodsh" && return
919 [ -z "$CLIENTS" ] && skip "Need two or more clients, have $CLIENTS" && return 0
920 [ $CLIENTCOUNT -lt 2 ] &&
921 skip "Need two or more clients, have $CLIENTCOUNT" && return 0
923 local nfiles=${TEST33_NFILES:-10000}
924 local param_file=$TMP/$tfile-params
925 local fstype=$(facet_fstype $SINGLEMDS)
927 save_lustre_params $(get_facets MDS) \
928 "mdt.*.commit_on_sharing" > $param_file
936 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=$COS
940 do_nodes $CLIENT1,$CLIENT2 "mkdir -p $DIR1/$tdir-\\\$(hostname)-$i"
942 [ $fstype = ldiskfs ] && jbdold=$(print_jbd_stat)
943 echo "=== START createmany old: $jbdold transaction"
944 local elapsed=$(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")
945 [ $fstype = ldiskfs ] && jbdnew=$(print_jbd_stat)
946 [ $fstype = ldiskfs ] && jbd=$(( jbdnew - jbdold ))
947 echo "=== END createmany new: $jbdnew transaction : $jbd transactions nfiles $nfiles time $elapsed COS=$COS"
948 [ $fstype = ldiskfs ] && avgjbd=$(( avgjbd + jbd ))
949 avgtime=$(( avgtime + elapsed ))
951 eval cos${COS}_jbd=$((avgjbd / 3))
952 eval cos${COS}_time=$((avgtime / 3))
955 echo "COS=0 transactions (avg): $cos0_jbd time (avg): $cos0_time"
956 echo "COS=1 transactions (avg): $cos1_jbd time (avg): $cos1_time"
957 [ "$cos0_jbd" != 0 ] && echo "COS=1 vs COS=0 jbd: $((((cos1_jbd/cos0_jbd - 1)) * 100 )) %"
958 [ "$cos0_time" != 0 ] && echo "COS=1 vs COS=0 time: $((((cos1_time/cos0_time - 1)) * 100 )) %"
960 restore_lustre_params < $param_file
964 run_test 33a "commit on sharing, cross crete/delete, 2 clients, benchmark"
966 # commit on sharing tests
968 remote_mds_nodsh && skip "remote MDS with nodsh" && return
970 [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; }
971 [ $CLIENTCOUNT -ge 2 ] ||
972 { skip "Need two or more clients, have $CLIENTCOUNT" &&
974 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
976 local nfiles=${TEST33_NFILES:-10000}
977 local param_file=$TMP/$tfile-params
979 save_lustre_params $(get_facets MDS) \
980 "mdt.*.commit_on_sharing" > $param_file
989 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=$COS
993 do_node $CLIENT1 "$LFS mkdir -i $MDTIDX \
994 $DIR1/$tdir-\\\$(hostname)-$i"
996 jbdold=$(print_jbd_stat)
997 echo "=== START createmany old: $jbdold transaction"
998 local elapsed=$(do_and_time "do_nodes $CLIENT1,$CLIENT2\
999 createmany -o $DIR1/$tdir-\\\$(hostname)-$i/f- \
1000 -r$DIR2/$tdir-\\\$(hostname)-$i/f- $nfiles > \
1002 jbdnew=$(print_jbd_stat)
1003 jbd=$(( jbdnew - jbdold ))
1004 echo "=== END createmany new: $jbdnew transaction : \
1005 $jbd transactions nfiles $nfiles time $elapsed COS=$COS"
1006 avgjbd=$(( avgjbd + jbd ))
1007 avgtime=$(( avgtime + elapsed ))
1009 eval cos${COS}_jbd=$((avgjbd / 3))
1010 eval cos${COS}_time=$((avgtime / 3))
1013 echo "COS=0 transactions (avg): $cos0_jbd time (avg): $cos0_time"
1014 echo "COS=1 transactions (avg): $cos1_jbd time (avg): $cos1_time"
1015 [ "$cos0_jbd" != 0 ] &&
1016 echo "COS=1 vs COS=0 jbd: $(((cos1_jbd/cos0_jbd - 1) * 100)) %"
1017 [ "$cos0_time" != 0 ] &&
1018 echo "COS=1 vs COS=0 time: $(((cos1_time/cos0_time - 1) * 100)) %"
1020 restore_lustre_params < $param_file
1024 run_test 33b "COS: cross create/delete, 2 clients, benchmark under remote dir"
1027 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
1028 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
1029 skip "DNE CoS not supported" && return
1035 do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
1036 # do twice in case transaction is committed before unlock, see LU-8200
1038 # remote dir is created on MDT1, which enqueued lock of $tdir on
1040 $LFS mkdir -i 1 $DIR/$tdir/remote.$i
1041 mkdir $DIR/$tdir/local.$i
1043 sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count")
1044 echo "sync_count $sync_count"
1045 [ $sync_count -eq 0 ] && error "Sync-Lock-Cancel not triggered"
1048 do_facet mds1 "lctl set_param -n mdt.*.sync_count=0"
1049 $LFS mkdir -i 1 $DIR/$tdir/remote.3
1050 # during sleep remote mkdir should have been committed and canceled
1051 # remote lock spontaneously, which shouldn't trigger sync
1053 mkdir $DIR/$tdir/local.3
1054 sync_count=$(do_facet mds1 "lctl get_param -n mdt.*MDT0000.sync_count")
1055 echo "sync_count $sync_count"
1056 [ $sync_count -eq 0 ] || error "Sync-Lock-Cancel triggered"
1058 run_test 33c "Cancel cross-MDT lock should trigger Sync-Lock-Cancel"
1060 # arg1 is operations done before CoS, arg2 is the operation that triggers CoS
1064 local nodes=$(comma_list $(mdts_nodes))
1068 # trigger CoS twice in case transaction commit before unlock
1071 do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
1073 commit_nr=$(do_nodes $nodes \
1074 "lctl get_param -n mdt.*.async_commit_count" | calc_sum)
1075 total=$((total + commit_nr));
1080 echo "CoS count $total"
1081 [ $total -gt 0 ] || error "$2 didn't trigger CoS"
1085 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
1086 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
1087 skip "DNE CoS not supported" && return
1089 # remote directory create
1090 op_trigger_cos "mkdir $DIR/$tdir" "$LFS mkdir -i 1 $DIR/$tdir/subdir"
1091 # remote directory unlink
1092 op_trigger_cos "$LFS mkdir -i 1 $DIR/$tdir" "rmdir $DIR/$tdir"
1093 # striped directory create
1094 op_trigger_cos "mkdir $DIR/$tdir" "$LFS mkdir -c 2 $DIR/$tdir/subdir"
1095 # striped directory setattr
1096 op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; touch $DIR/$tdir" \
1097 "chmod 713 $DIR/$tdir"
1098 # striped directory unlink
1099 op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; touch $DIR/$tdir" \
1102 op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; \
1103 $LFS mkdir -i 0 $DIR/$tdir/d1; \
1104 $LFS mkdir -i 1 $DIR/$tdir/d2; \
1105 touch $DIR/$tdir/d1/tgt" \
1106 "ln $DIR/$tdir/d1/tgt $DIR/$tdir/d2/src"
1108 op_trigger_cos "$LFS mkdir -c 2 $DIR/$tdir; \
1109 $LFS mkdir -i 0 $DIR/$tdir/d1; \
1110 $LFS mkdir -i 1 $DIR/$tdir/d2; \
1111 touch $DIR/$tdir/d1/src" \
1112 "mv $DIR/$tdir/d1/src $DIR/$tdir/d2/tgt"
1114 op_trigger_cos "$LFS mkdir -i 0 $DIR/$tdir" \
1115 "$LFS migrate -m 1 $DIR/$tdir"
1119 run_test 33d "DNE distributed operation should trigger COS"
1122 [ -n "$CLIENTS" ] || { skip "Need two or more clients" && return 0; }
1123 [ $CLIENTCOUNT -ge 2 ] ||
1124 { skip "Need two or more clients, have $CLIENTCOUNT" &&
1126 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
1127 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.63) ] &&
1128 skip "DNE CoS not supported" && return
1130 local client2=${CLIENT2:-$(hostname)}
1134 local nodes=$(comma_list $(mdts_nodes))
1135 do_nodes $nodes "lctl set_param -n mdt.*.async_commit_count=0"
1137 $LFS mkdir -c 2 $DIR/$tdir
1138 mkdir $DIR/$tdir/subdir
1139 echo abc > $DIR/$tdir/$tfile
1140 do_node $client2 echo dfg >> $DIR/$tdir/$tfile
1141 do_node $client2 touch $DIR/$tdir/subdir
1143 local async_commit_count=$(do_nodes $nodes \
1144 "lctl get_param -n mdt.*.async_commit_count" | calc_sum)
1145 [ $async_commit_count -gt 0 ] && error "CoS triggerred"
1149 run_test 33e "DNE local operation shouldn't trigger COS"
1151 # End commit on sharing tests
1153 get_ost_lock_timeouts() {
1154 local nodes=${1:-$(comma_list $(osts_nodes))}
1156 local locks=$(do_nodes $nodes \
1157 "lctl get_param -n ldlm.namespaces.filter-*.lock_timeouts" | calc_sum)
1165 do_nodes $(comma_list $(osts_nodes)) \
1166 "lctl set_param -n fail_loc=0 2>/dev/null || true"
1167 for i in $(seq $OSTCOUNT); do
1168 wait_osc_import_ready client ost$i
1173 remote_ost_nodsh && skip "remote OST with nodsh" && return
1177 trap cleanup_34 EXIT RETURN
1178 for OPER in notimeout timeout ; do
1179 rm $DIR1/$tfile 2>/dev/null
1180 lock_in=$(get_ost_lock_timeouts)
1181 if [ $OPER == "timeout" ] ; then
1182 for j in `seq $OSTCOUNT`; do
1183 #define OBD_FAIL_PTLRPC_HPREQ_TIMEOUT 0x511
1184 do_facet ost$j lctl set_param fail_loc=0x511
1186 echo lock should expire
1188 for j in `seq $OSTCOUNT`; do
1189 #define OBD_FAIL_PTLRPC_HPREQ_NOTIMEOUT 0x512
1190 do_facet ost$j lctl set_param fail_loc=0x512
1192 echo lock should not expire
1194 echo writing on client1
1195 dd if=/dev/zero of=$DIR1/$tfile count=100 conv=notrunc > /dev/null 2>&1
1197 echo reading on client2
1198 dd of=/dev/null if=$DIR2/$tfile > /dev/null 2>&1
1199 # wait for a lock timeout
1201 lock_out=$(get_ost_lock_timeouts)
1202 if [ $OPER == "timeout" ] ; then
1203 if [ $lock_in == $lock_out ]; then
1204 error "no lock timeout happened"
1209 if [ $lock_in != $lock_out ]; then
1210 error "lock timeout happened"
1218 run_test 34 "no lock timeout under IO"
1220 test_35() { # bug 17645
1223 gen=$(lctl get_param mdc.$FSNAME-MDT*-mdc-*.import | grep generation |
1224 awk '/generation/{print $2}')
1226 generation[count]=$g
1230 test_mkdir $MOUNT1/$tdir
1231 cancel_lru_locks mdc
1233 # Let's initiate -EINTR situation by setting fail_loc and take
1234 # write lock on same file from same client. This will not cause
1235 # bl_ast yet as lock is already in local cache.
1236 #define OBD_FAIL_LDLM_INTR_CP_AST 0x317
1237 do_facet client "lctl set_param fail_loc=0x80000317"
1238 local timeout=$(do_facet $SINGLEMDS lctl get_param -n timeout)
1239 let timeout=timeout*3
1241 while test $nr -lt 10; do
1242 log "Race attempt $nr"
1243 local blk1=$(lctl get_param -n ldlm.services.ldlm_cbd.stats |
1244 awk '/ldlm_bl_callback/ { print $2 }')
1245 test "x$blk1" = "x" && blk1=0
1246 createmany -o $MOUNT2/$tdir/a 4000 &
1250 # Let's make conflict and bl_ast
1251 ls -la $MOUNT1/$tdir > /dev/null &
1254 log "Wait for $pid1 $pid2 for $timeout sec..."
1256 kill -9 $pid1 $pid2 > /dev/null 2>&1
1258 local blk2=$(lctl get_param -n ldlm.services.ldlm_cbd.stats |
1259 awk '/ldlm_bl_callback/ { print $2 }')
1260 test "x$blk2" = "x" && blk2=0
1261 test $blk2 -gt $blk1 && break
1262 rm -fr $MOUNT1/$tdir
1263 cancel_lru_locks mdc
1266 do_facet client "lctl set_param fail_loc=0x0"
1267 df -h $MOUNT1 $MOUNT2
1269 gen=$(lctl get_param mdc.$FSNAME-MDT*-mdc-*.import | grep generation |
1270 awk '/generation/{print $2}')
1272 if ! test "$g" -eq "${generation[count]}"; then
1273 list=$(lctl list_param mdc.$FSNAME-MDT*-mdc-*.import)
1275 for imp in $list; do
1276 if [ $c = $count ]; then
1281 imp=$(echo "$imp" | awk -F"." '{print $2}')
1282 error "Eviction happened on import $imp"
1287 run_test 35 "-EINTR cp_ast vs. bl_ast race does not evict client"
1289 test_36() { #bug 16417
1294 test_mkdir $DIR1/$tdir
1295 $LFS setstripe -c -1 $DIR1/$tdir
1298 let SIZE_B=SIZE*1024*1024
1299 sync; sleep 2; sync # wait for delete thread
1300 wait_mds_ost_sync || error "wait_mds_ost_sync failed"
1301 wait_destroy_complete || error "wait_destroy_complete failed"
1303 while [ $i -le 10 ]; do
1304 lctl mark "start test - cycle ($i)"
1305 local before=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
1306 dd if=/dev/zero of=$DIR1/$tdir/$tfile bs=1M count=$SIZE ||
1307 error "dd $DIR1/$tdir/$tfile ${SIZE}MB failed"
1308 sync # sync data from client cache
1309 sync_all_data # sync data from server cache (delayed allocation)
1311 local after_dd=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
1312 multiop_bg_pause $DIR2/$tdir/$tfile O_r${SIZE_B}c || return 3
1314 rm -f $DIR1/$tdir/$tfile
1315 kill -USR1 $read_pid
1317 sync; sleep 2; sync # Ensure new statfs
1318 wait_delete_completed
1319 local after=$(lfs_df $MOUNT1 | awk '/^filesystem/{ print $4; exit }')
1320 echo "*** cycle($i) *** before($before) after_dd($after_dd)" \
1322 # this free space! not used
1323 (( $after_dd <= $after)) ||
1324 error "space leaked after_dd:$after_dd > after:$after"
1328 run_test 36 "handle ESTALE/open-unlink correctly"
1330 test_37() { # bug 18695
1331 test_mkdir $DIR1/$tdir
1332 multiop_bg_pause $DIR1/$tdir D_c || return 1
1334 # create large directory (32kB seems enough from e2fsck, ~= 1000 files)
1335 createmany -m $DIR2/$tdir/f 10000
1336 # set mtime/atime backward
1337 touch -t 198001010000 $DIR2/$tdir
1338 kill -USR1 $MULTIPID
1339 nr_files=`lfs find $DIR1/$tdir -type f | wc -l`
1340 [ $nr_files -eq 10000 ] || error "$nr_files != 10000 truncated directory?"
1343 run_test 37 "check i_size is not updated for directory on close (bug 18695) =============="
1345 # this should be set to past
1346 TEST_39_MTIME=`date -d "1 year ago" +%s`
1350 local client1=${CLIENT1:-`hostname`}
1351 local client2=${CLIENT2:-`hostname`}
1353 do_node $client1 "touch $DIR1/$tfile"
1355 do_node $client1 "touch -m -d @$TEST_39_MTIME $DIR1/$tfile"
1356 local mtime1=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1357 [ "$mtime1" = $TEST_39_MTIME ] || \
1358 error "mtime is not set to past: $mtime1, should be $TEST_39_MTIME"
1360 local d1=`do_node $client1 date +%s`
1361 do_node $client1 'echo hello >> '$DIR1/$tfile
1362 local d2=`do_node $client1 date +%s`
1364 local mtime2=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1365 [ "$mtime2" -ge "$d1" ] && [ "$mtime2" -le "$d2" ] || \
1366 error "mtime is not updated on write: $d1 <= $mtime2 <= $d2"
1368 do_node $client1 "mv $DIR1/$tfile $DIR1/$tfile-1"
1370 for (( i=0; i < 2; i++ )) ; do
1371 local mtime3=`do_node $client2 "stat -c %Y $DIR1/$tfile-1"`
1372 [ "$mtime2" = "$mtime3" ] || \
1373 error "mtime ($mtime2) changed (to $mtime3) on rename"
1375 cancel_lru_locks osc
1376 if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi
1379 run_test 39a "test from 11063 =================================="
1382 local client1=${CLIENT1:-`hostname`}
1383 local client2=${CLIENT2:-`hostname`}
1387 local mtime1=`stat -c %Y $DIR1/$tfile`
1388 local mtime2=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1391 touch -m -d @$TEST_39_MTIME $DIR1/$tfile
1393 for (( i=0; i < 2; i++ )) ; do
1394 local mtime3=`stat -c %Y $DIR1/$tfile`
1395 local mtime4=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1397 [ "$mtime3" = "$mtime4" ] || \
1398 error "different mtime on clients: $mtime3, $mtime4"
1399 [ "$mtime3" = $TEST_39_MTIME ] || \
1400 error "lost mtime: $mtime3, should be $TEST_39_MTIME"
1402 cancel_lru_locks osc
1403 if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi
1406 run_test 39b "11063 problem 1 =================================="
1409 local client1=${CLIENT1:-`hostname`}
1410 local client2=${CLIENT2:-`hostname`}
1412 echo hello > $DIR1/$tfile
1414 local mtime1=`stat -c %Y $DIR1/$tfile`
1415 local mtime2=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1416 [ "$mtime1" = "$mtime2" ] || \
1417 error "create: different mtime on clients: $mtime1, $mtime2"
1420 $TRUNCATE $DIR1/$tfile 1
1422 for (( i=0; i < 2; i++ )) ; do
1423 local mtime3=`stat -c %Y $DIR1/$tfile`
1424 local mtime4=`do_node $client2 "stat -c %Y $DIR1/$tfile"`
1426 [ "$mtime3" = "$mtime4" ] || \
1427 error "different mtime on clients: $mtime3, $mtime4"
1428 [ "$mtime3" -gt $mtime2 ] || \
1429 error "truncate did not update mtime: $mtime2, $mtime3"
1431 cancel_lru_locks osc
1432 if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi
1435 run_test 39c "check truncate mtime update ======================"
1437 test_39d() { # LU-7310
1439 touch -m -d @$TEST_39_MTIME $DIR1/$tfile
1441 local mtime1=$(stat -c %Y $DIR2/$tfile)
1442 [ "$mtime1" = $TEST_39_MTIME ] ||
1443 error "mtime: $mtime1, should be $TEST_39_MTIME"
1446 # define OBD_FAIL_OSC_NO_GRANT 0x411
1447 $LCTL set_param fail_loc=0x411
1449 local d1=$(date +%s)
1450 echo hello >> $DIR1/$tfile
1451 local d2=$(date +%s)
1453 $LCTL set_param fail_loc=0
1455 cancel_lru_locks $OSC
1457 local mtime2=$(stat -c %Y $DIR2/$tfile)
1458 [ "$mtime2" -ge "$d1" ] && [ "$mtime2" -le "$d2" ] ||
1459 error "mtime is not updated on write: $d1 <= $mtime2 <= $d2"
1461 run_test 39d "sync write should update mtime"
1463 # check that pid exists hence second operation wasn't blocked by first one
1464 # if it is so then there is no conflict, return 0
1465 # else second operation is conflicting with first one, return 1
1466 check_pdo_conflict() {
1469 sleep 1 # to ensure OP1 is finished on client if OP2 is blocked by OP1
1470 if [[ `ps --pid $pid | wc -l` == 1 ]]; then
1480 # test 40: check non-blocking operations
1482 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1483 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1484 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1486 mkdir $DIR1/$tfile &
1489 touch $DIR2/$tfile-2
1490 check_pdo_conflict $PID1 || error "create is blocked"
1491 mkdir $DIR2/$tfile-3
1492 check_pdo_conflict $PID1 || error "mkdir is blocked"
1493 link $DIR2/$tfile-2 $DIR2/$tfile-4
1494 check_pdo_conflict $PID1 || error "link is blocked"
1495 mv $DIR2/$tfile-2 $DIR2/$tfile-5
1496 check_pdo_conflict $PID1 || error "rename is blocked"
1497 stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null
1498 check_pdo_conflict $PID1 || error "getattr is blocked"
1499 rm $DIR2/$tfile-4 $DIR2/$tfile-5
1500 rmdir $DIR2/$tfile-3
1501 check_pdo_conflict $PID1 || error "unlink is blocked"
1503 # all operations above shouldn't wait the first one
1504 check_pdo_conflict $PID1 || error "parallel operation is blocked"
1509 run_test 40a "pdirops: create vs others =============="
1512 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1513 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1514 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1515 touch $DIR1/$tfile &
1519 touch $DIR2/$tfile-2
1520 check_pdo_conflict $PID1 || error "create is blocked"
1521 mkdir $DIR2/$tfile-3
1522 check_pdo_conflict $PID1 || error "mkdir is blocked"
1523 link $DIR2/$tfile-2 $DIR2/$tfile-4
1524 check_pdo_conflict $PID1 || error "link is blocked"
1525 mv $DIR2/$tfile-2 $DIR2/$tfile-5
1526 check_pdo_conflict $PID1 || error "rename is blocked"
1527 stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null
1528 check_pdo_conflict $PID1 || error "getattr is blocked"
1529 rm $DIR2/$tfile-4 $DIR2/$tfile-5
1530 rmdir $DIR2/$tfile-3
1531 check_pdo_conflict $PID1 || error "unlink is blocked"
1532 # all operations above shouldn't wait the first one
1534 check_pdo_conflict $PID1 || error "parallel operation is blocked"
1539 run_test 40b "pdirops: open|create and others =============="
1542 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1544 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1545 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1546 link $DIR1/$tfile $DIR1/$tfile-0 &
1550 touch $DIR2/$tfile-2
1551 check_pdo_conflict $PID1 || error "create is blocked"
1552 mkdir $DIR2/$tfile-3
1553 check_pdo_conflict $PID1 || error "mkdir is blocked"
1554 link $DIR2/$tfile-2 $DIR2/$tfile-4
1555 check_pdo_conflict $PID1 || error "link is blocked"
1556 mv $DIR2/$tfile-2 $DIR2/$tfile-5
1557 check_pdo_conflict $PID1 || error "rename is blocked"
1558 stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null
1559 check_pdo_conflict $PID1 || error "getattr is blocked"
1560 rm $DIR2/$tfile-4 $DIR2/$tfile-5
1561 rmdir $DIR2/$tfile-3
1562 check_pdo_conflict $PID1 || error "unlink is blocked"
1564 # all operations above shouldn't wait the first one
1565 check_pdo_conflict $PID1 || error "parallel operation is blocked"
1570 run_test 40c "pdirops: link and others =============="
1573 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1575 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1576 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1581 touch $DIR2/$tfile-2
1582 check_pdo_conflict $PID1 || error "create is blocked"
1583 mkdir $DIR2/$tfile-3
1584 check_pdo_conflict $PID1 || error "mkdir is blocked"
1585 link $DIR2/$tfile-2 $DIR2/$tfile-4
1586 check_pdo_conflict $PID1 || error "link is blocked"
1587 mv $DIR2/$tfile-2 $DIR2/$tfile-5
1588 check_pdo_conflict $PID1 || error "rename is blocked"
1589 stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null
1590 check_pdo_conflict $PID1 || error "getattr is blocked"
1591 rm $DIR2/$tfile-4 $DIR2/$tfile-5
1592 rmdir $DIR2/$tfile-3
1593 check_pdo_conflict $PID1 || error "unlink is blocked"
1595 # all operations above shouldn't wait the first one
1596 check_pdo_conflict $PID1 || error "parallel operation is blocked"
1600 run_test 40d "pdirops: unlink and others =============="
1603 remote_mds_nodsh && skip "remote MDS with nodsh" && return
1605 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1606 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1607 mv $DIR1/$tfile $DIR1/$tfile-0 &
1611 touch $DIR2/$tfile-2
1612 check_pdo_conflict $PID1 || error "create is blocked"
1613 mkdir $DIR2/$tfile-3
1614 check_pdo_conflict $PID1 || error "mkdir is blocked"
1615 link $DIR2/$tfile-2 $DIR2/$tfile-4
1616 check_pdo_conflict $PID1 || error "link is blocked"
1617 stat $DIR2/$tfile-3 $DIR2/$tfile-4 > /dev/null
1618 check_pdo_conflict $PID1 || error "getattr is blocked"
1619 rm $DIR2/$tfile-4 $DIR2/$tfile-2
1620 rmdir $DIR2/$tfile-3
1621 check_pdo_conflict $PID1 || error "unlink is blocked"
1623 # all operations above shouldn't wait the first one
1624 check_pdo_conflict $PID1 || error "parallel operation is blocked"
1629 run_test 40e "pdirops: rename and others =============="
1631 # test 41: create blocking operations
1633 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1634 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1635 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
1638 mkdir $DIR2/$tfile && error "mkdir must fail"
1639 check_pdo_conflict $PID1 && { wait $PID1; echo "mkdir isn't blocked"; }
1643 run_test 41a "pdirops: create vs mkdir =============="
1646 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1647 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1648 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
1651 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail"
1652 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
1656 run_test 41b "pdirops: create vs create =============="
1659 touch $DIR1/$tfile-2
1660 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1661 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1662 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
1665 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail"
1666 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
1670 run_test 41c "pdirops: create vs link =============="
1673 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1674 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1675 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
1678 rm $DIR2/$tfile || error "unlink must succeed"
1679 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
1683 run_test 41d "pdirops: create vs unlink =============="
1686 touch $DIR1/$tfile-2
1687 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1688 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1689 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
1692 mv $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed"
1693 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
1697 run_test 41e "pdirops: create and rename (tgt) =============="
1700 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1701 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1702 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
1705 mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed"
1706 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
1710 run_test 41f "pdirops: create and rename (src) =============="
1713 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1714 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1715 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
1718 stat $DIR2/$tfile > /dev/null || error "stat must succeed"
1719 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
1723 run_test 41g "pdirops: create vs getattr =============="
1726 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1727 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1728 $MULTIOP $DIR1/$tfile oO_CREAT:O_RDWR:c &
1731 ls -lia $DIR2/ > /dev/null
1732 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
1736 run_test 41h "pdirops: create vs readdir =============="
1738 # test 42: unlink and blocking operations
1740 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1741 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1742 mkdir $DIR1/$tfile &
1745 mkdir $DIR2/$tfile && error "mkdir must fail"
1746 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
1750 run_test 42a "pdirops: mkdir vs mkdir =============="
1753 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1754 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1755 mkdir $DIR1/$tfile &
1758 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail"
1759 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
1763 run_test 42b "pdirops: mkdir vs create =============="
1766 touch $DIR1/$tfile-2
1767 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1768 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1769 mkdir $DIR1/$tfile &
1772 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail"
1773 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
1777 run_test 42c "pdirops: mkdir vs link =============="
1780 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1781 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1782 mkdir $DIR1/$tfile &
1785 rmdir $DIR2/$tfile || error "unlink must succeed"
1786 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
1790 run_test 42d "pdirops: mkdir vs unlink =============="
1793 touch $DIR1/$tfile-2
1794 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1795 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1796 mkdir $DIR1/$tfile &
1799 mv -T $DIR2/$tfile-2 $DIR2/$tfile && error "rename must fail"
1800 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
1804 run_test 42e "pdirops: mkdir and rename (tgt) =============="
1807 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1808 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1809 mkdir $DIR1/$tfile &
1812 mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed"
1813 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
1817 run_test 42f "pdirops: mkdir and rename (src) =============="
1820 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1821 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1822 mkdir $DIR1/$tfile &
1825 stat $DIR2/$tfile > /dev/null || error "stat must succeed"
1826 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
1830 run_test 42g "pdirops: mkdir vs getattr =============="
1833 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1834 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1835 mkdir $DIR1/$tfile &
1838 ls -lia $DIR2/ > /dev/null
1839 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
1843 run_test 42h "pdirops: mkdir vs readdir =============="
1845 # test 43: unlink and blocking operations
1848 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1849 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1853 mkdir $DIR2/$tfile || error "mkdir must succeed"
1854 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
1858 run_test 43a "pdirops: unlink vs mkdir =============="
1862 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1863 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1867 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed"
1868 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
1872 run_test 43b "pdirops: unlink vs create =============="
1876 touch $DIR1/$tfile-2
1877 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1878 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1882 link $DIR2/$tfile-2 $DIR2/$tfile || error "link must succeed"
1883 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
1887 run_test 43c "pdirops: unlink vs link =============="
1891 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1892 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1896 rm $DIR2/$tfile && error "unlink must fail"
1897 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
1901 run_test 43d "pdirops: unlink vs unlink =============="
1905 touch $DIR1/$tfile-2
1906 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1907 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1911 mv -u $DIR2/$tfile-2 $DIR2/$tfile || error "rename must succeed"
1912 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
1916 run_test 43e "pdirops: unlink and rename (tgt) =============="
1920 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1921 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1925 mv $DIR2/$tfile $DIR2/$tfile-2 && error "rename must fail"
1926 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
1930 run_test 43f "pdirops: unlink and rename (src) =============="
1934 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1935 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1939 stat $DIR2/$tfile > /dev/null && error "stat must fail"
1940 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
1944 run_test 43g "pdirops: unlink vs getattr =============="
1948 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1949 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1953 ls -lia $DIR2/ > /dev/null
1954 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
1958 run_test 43h "pdirops: unlink vs readdir =============="
1961 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
1963 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
1964 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
1968 $LFS mkdir -i 1 $DIR2/$tfile || error "remote mkdir must succeed"
1969 check_pdo_conflict $PID1 &&
1970 { wait $PID1; error "remote mkdir isn't blocked"; }
1974 run_test 43i "pdirops: unlink vs remote mkdir"
1976 # test 44: rename tgt and blocking operations
1978 touch $DIR1/$tfile-2
1979 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
1980 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
1981 mv $DIR1/$tfile-2 $DIR1/$tfile &
1984 mkdir $DIR2/$tfile && error "mkdir must fail"
1985 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
1989 run_test 44a "pdirops: rename tgt vs mkdir =============="
1992 touch $DIR1/$tfile-2
1993 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
1994 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
1995 mv $DIR1/$tfile-2 $DIR1/$tfile &
1998 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail"
1999 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
2003 run_test 44b "pdirops: rename tgt vs create =============="
2006 touch $DIR1/$tfile-2
2007 touch $DIR1/$tfile-3
2008 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2009 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
2010 mv $DIR1/$tfile-2 $DIR1/$tfile &
2013 link $DIR2/$tfile-3 $DIR2/$tfile && error "link must fail"
2014 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2018 run_test 44c "pdirops: rename tgt vs link =============="
2021 touch $DIR1/$tfile-2
2022 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2023 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
2024 mv $DIR1/$tfile-2 $DIR1/$tfile &
2027 rm $DIR2/$tfile || error "unlink must succeed"
2028 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
2032 run_test 44d "pdirops: rename tgt vs unlink =============="
2036 touch $DIR1/$tfile-2
2037 touch $DIR1/$tfile-3
2038 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2039 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
2040 mv $DIR1/$tfile-2 $DIR1/$tfile &
2043 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed"
2044 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2048 run_test 44e "pdirops: rename tgt and rename (tgt) =============="
2051 touch $DIR1/$tfile-2
2052 touch $DIR1/$tfile-3
2053 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2054 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
2055 mv $DIR1/$tfile-2 $DIR1/$tfile &
2058 mv $DIR2/$tfile $DIR2/$tfile-3 || error "rename must succeed"
2059 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2063 run_test 44f "pdirops: rename tgt and rename (src) =============="
2066 touch $DIR1/$tfile-2
2067 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2068 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
2069 mv $DIR1/$tfile-2 $DIR1/$tfile &
2072 stat $DIR2/$tfile > /dev/null || error "stat must succeed"
2073 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
2077 run_test 44g "pdirops: rename tgt vs getattr =============="
2080 touch $DIR1/$tfile-2
2081 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2082 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
2083 mv $DIR1/$tfile-2 $DIR1/$tfile &
2086 ls -lia $DIR2/ > /dev/null
2087 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
2091 run_test 44h "pdirops: rename tgt vs readdir =============="
2093 # test 44: rename tgt and blocking operations
2095 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2096 touch $DIR1/$tfile-2
2097 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK2 0x146
2098 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000146
2099 mv $DIR1/$tfile-2 $DIR1/$tfile &
2102 $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail"
2103 check_pdo_conflict $PID1 && { wait $PID1;
2104 error "remote mkdir isn't blocked"; }
2108 run_test 44i "pdirops: rename tgt vs remote mkdir"
2110 # test 45: rename src and blocking operations
2113 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2114 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2115 mv $DIR1/$tfile $DIR1/$tfile-2 &
2118 mkdir $DIR2/$tfile || error "mkdir must succeed"
2119 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
2123 run_test 45a "pdirops: rename src vs mkdir =============="
2127 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2128 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2129 mv $DIR1/$tfile $DIR1/$tfile-2 &
2132 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c || error "create must succeed"
2133 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
2137 run_test 45b "pdirops: rename src vs create =============="
2141 touch $DIR1/$tfile-3
2142 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2143 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2144 mv $DIR1/$tfile $DIR1/$tfile-2 &
2147 link $DIR2/$tfile-3 $DIR2/$tfile || error "link must succeed"
2148 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2152 run_test 45c "pdirops: rename src vs link =============="
2156 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2157 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2158 mv $DIR1/$tfile $DIR1/$tfile-2 &
2161 rm $DIR2/$tfile && error "unlink must fail"
2162 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
2166 run_test 45d "pdirops: rename src vs unlink =============="
2170 touch $DIR1/$tfile-3
2171 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2172 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2173 mv $DIR1/$tfile $DIR1/$tfile-2 &
2176 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed"
2177 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2181 run_test 45e "pdirops: rename src and rename (tgt) =============="
2185 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2186 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2187 mv $DIR1/$tfile $DIR1/$tfile-2 &
2190 mv $DIR2/$tfile $DIR2/$tfile-3 && error "rename must fail"
2191 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2195 run_test 45f "pdirops: rename src and rename (src) =============="
2199 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2200 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2201 mv $DIR1/$tfile $DIR1/$tfile-2 &
2204 stat $DIR2/$tfile > /dev/null && error "stat must fail"
2205 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
2209 run_test 45g "pdirops: rename src vs getattr =============="
2213 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2214 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2215 mv $DIR1/$tfile $DIR1/$tfile-2 &
2218 ls -lia $DIR2/ > /dev/null
2219 check_pdo_conflict $PID1 && { wait $PID1; error "readdir isn't blocked"; }
2223 run_test 45h "pdirops: unlink vs readdir =============="
2226 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2228 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2229 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2230 mv $DIR1/$tfile $DIR1/$tfile-2 &
2233 $LFS mkdir -i 1 $DIR2/$tfile || error "create remote dir must succeed"
2234 check_pdo_conflict $PID1 && { wait $PID1;
2235 error "create remote dir isn't blocked"; }
2239 run_test 45i "pdirops: rename src vs remote mkdir"
2241 # test 46: link and blocking operations
2243 touch $DIR1/$tfile-2
2244 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2245 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2246 link $DIR1/$tfile-2 $DIR1/$tfile &
2249 mkdir $DIR2/$tfile && error "mkdir must fail"
2250 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
2254 run_test 46a "pdirops: link vs mkdir =============="
2257 touch $DIR1/$tfile-2
2258 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2259 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2260 link $DIR1/$tfile-2 $DIR1/$tfile &
2263 $MULTIOP $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail"
2264 check_pdo_conflict $PID1 && { wait $PID1; error "create isn't blocked"; }
2268 run_test 46b "pdirops: link vs create =============="
2271 touch $DIR1/$tfile-2
2272 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2273 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2274 link $DIR1/$tfile-2 $DIR1/$tfile &
2277 link $DIR2/$tfile $DIR2/$tfile && error "link must fail"
2278 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2282 run_test 46c "pdirops: link vs link =============="
2285 touch $DIR1/$tfile-2
2286 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2287 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2288 link $DIR1/$tfile-2 $DIR1/$tfile &
2291 rm $DIR2/$tfile || error "unlink must succeed"
2292 check_pdo_conflict $PID1 && { wait $PID1; error "unlink isn't blocked"; }
2296 run_test 46d "pdirops: link vs unlink =============="
2299 touch $DIR1/$tfile-2
2300 touch $DIR1/$tfile-3
2301 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2302 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2303 link $DIR1/$tfile-2 $DIR1/$tfile &
2306 mv $DIR2/$tfile-3 $DIR2/$tfile || error "rename must succeed"
2307 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2311 run_test 46e "pdirops: link and rename (tgt) =============="
2314 touch $DIR1/$tfile-2
2315 touch $DIR1/$tfile-3
2316 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2317 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2318 link $DIR1/$tfile-2 $DIR1/$tfile &
2321 mv $DIR2/$tfile $DIR2/$tfile-3 || error "rename must succeed"
2322 check_pdo_conflict $PID1 && { wait $PID1; error "rename isn't blocked"; }
2326 run_test 46f "pdirops: link and rename (src) =============="
2329 touch $DIR1/$tfile-2
2330 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2331 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2332 link $DIR1/$tfile-2 $DIR1/$tfile &
2335 stat $DIR2/$tfile > /dev/null || error "stat must succeed"
2336 check_pdo_conflict $PID1 && { wait $PID1; error "getattr isn't blocked"; }
2340 run_test 46g "pdirops: link vs getattr =============="
2343 touch $DIR1/$tfile-2
2344 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2345 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2346 link $DIR1/$tfile-2 $DIR1/$tfile &
2349 ls -lia $DIR2/ > /dev/null
2350 check_pdo_conflict $PID1 && { wait $PID1;
2351 error "readdir isn't blocked"; }
2355 run_test 46h "pdirops: link vs readdir =============="
2358 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2359 touch $DIR1/$tfile-2
2360 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2361 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2362 link $DIR1/$tfile-2 $DIR1/$tfile &
2365 $LFS mkdir -i 1 $DIR2/$tfile && error "remote mkdir must fail"
2366 check_pdo_conflict $PID1 && { wait $PID1;
2367 error "remote mkdir isn't blocked"; }
2371 run_test 46i "pdirops: link vs remote mkdir"
2373 # test 47: remote mkdir and blocking operations
2375 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2376 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2377 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2378 $LFS mkdir -i 1 $DIR1/$tfile &
2381 mkdir $DIR2/$tfile && error "mkdir must fail"
2382 check_pdo_conflict $PID1 && { wait $PID1; error "mkdir isn't blocked"; }
2386 run_test 47a "pdirops: remote mkdir vs mkdir"
2389 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2390 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2391 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2392 $LFS mkdir -i 1 $DIR1/$tfile &
2395 multiop $DIR2/$tfile oO_CREAT:O_EXCL:c && error "create must fail"
2396 check_pdo_conflict $PID1 && { wait $PID1;
2397 error "create isn't blocked"; }
2401 run_test 47b "pdirops: remote mkdir vs create"
2404 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2405 touch $DIR1/$tfile-2
2406 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2407 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2408 $LFS mkdir -i 1 $DIR1/$tfile &
2411 link $DIR2/$tfile-2 $DIR2/$tfile && error "link must fail"
2412 check_pdo_conflict $PID1 && { wait $PID1; error "link isn't blocked"; }
2416 run_test 47c "pdirops: remote mkdir vs link"
2419 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2420 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2421 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2422 $LFS mkdir -i 1 $DIR1/$tfile &
2425 rmdir $DIR2/$tfile || error "unlink must succeed"
2426 check_pdo_conflict $PID1 && { wait $PID1;
2427 error "unlink isn't blocked"; }
2431 run_test 47d "pdirops: remote mkdir vs unlink"
2434 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2435 touch $DIR1/$tfile-2
2436 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2437 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2438 $LFS mkdir -i 1 $DIR1/$tfile &
2441 mv -T $DIR2/$tfile-2 $DIR2/$tfile && error "rename must fail"
2442 check_pdo_conflict $PID1 && { wait $PID1;
2443 error "rename isn't blocked"; }
2447 run_test 47e "pdirops: remote mkdir and rename (tgt)"
2450 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2451 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2452 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2453 $LFS mkdir -i 1 $DIR1/$tfile &
2456 mv $DIR2/$tfile $DIR2/$tfile-2 || error "rename must succeed"
2457 check_pdo_conflict $PID1 && { wait $PID1;
2458 error "rename isn't blocked"; }
2462 run_test 47f "pdirops: remote mkdir and rename (src)"
2465 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
2468 #define OBD_FAIL_ONCE|OBD_FAIL_MDS_PDO_LOCK 0x145
2469 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000145
2470 $LFS mkdir -i 1 $DIR1/$tfile &
2473 stat $DIR2/$tfile > /dev/null || error "stat must succeed"
2474 check_pdo_conflict $PID1 && { wait $PID1;
2475 error "getattr isn't blocked"; }
2479 run_test 47g "pdirops: remote mkdir vs getattr"
2483 dd if=/dev/zero of=$DIR1/$tfile bs=1K count=10
2484 #define OBD_FAIL_OSC_CP_ENQ_RACE 0x410
2485 do_facet client "lctl set_param fail_loc=0x410"
2486 $TRUNCATE $DIR2/$tfile $trunc_size
2487 do_facet client "lctl set_param fail_loc=0x0"
2489 size=`stat -c %s $DIR2/$tfile`
2490 [ $size -eq $trunc_size ] || error "wrong size"
2492 run_test 50 "osc lvb attrs: enqueue vs. CP AST =============="
2496 local origfile=/etc/hosts
2498 filesize=$(stat -c %s $origfile)
2500 # create an empty file
2501 $MCREATE $DIR1/$tfile || error "can't create $DIR1/$tfile"
2502 # cache layout lock on both mount point
2503 stat $DIR1/$tfile > /dev/null || error "stat $DIR1/$tfile failed"
2504 stat $DIR2/$tfile > /dev/null || error "stat $DIR2/$tfile failed"
2506 # open and sleep 2 seconds then read
2507 $MULTIOP $DIR2/$tfile o_2r${filesize}c &
2511 # create the layout of testing file
2512 dd if=$origfile of=$DIR1/$tfile conv=notrunc > /dev/null ||
2513 error "dd $DIR1/$tfile failed"
2515 # MULTIOP proc should be able to read enough bytes and exit
2516 for ((i = 0; i < 6; i++)); do
2518 kill -0 $pid || break
2520 kill -0 $pid 2> /dev/null && error "multiop is still there"
2521 cmp $origfile $DIR2/$tfile || error "$origfile and $DIR2/$tfile differs"
2525 run_test 51a "layout lock: refresh layout should work"
2528 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.59) ]] ||
2529 { skip "Need MDS version at least 2.3.59"; return 0; }
2531 local tmpfile=`mktemp`
2533 # create an empty file
2534 $MCREATE $DIR1/$tfile || error "mcreate $DIR1/$tfile failed"
2536 # delay glimpse so that layout has changed when glimpse finish
2537 #define OBD_FAIL_GLIMPSE_DELAY 0x1404
2538 $LCTL set_param fail_loc=0x1404
2539 stat -c %s $DIR2/$tfile |tee $tmpfile &
2543 # create layout of testing file
2544 dd if=/dev/zero of=$DIR1/$tfile bs=1k count=1 conv=notrunc >/dev/null ||
2545 error "dd $DIR1/$tfile failed"
2548 local fsize=$(cat $tmpfile)
2550 [ x$fsize = x1024 ] || error "file size is $fsize, should be 1024"
2552 rm -f $DIR1/$tfile $tmpfile
2554 run_test 51b "layout lock: glimpse should be able to restart if layout changed"
2557 [ $OSTCOUNT -ge 2 ] || { skip "needs >= 2 osts"; return; }
2559 # set default layout to have 1 stripe
2561 $LFS setstripe -c 1 $DIR1/$tdir
2563 # create a file with empty layout
2564 $MCREATE $DIR1/$tdir/$tfile ||
2565 error "$MCREATE $DIR1/$tdir/$tfile failed"
2567 #define OBD_FAIL_MDS_LL_BLOCK 0x172
2568 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x172
2570 # change the layout of testing file
2571 echo "Setting layout to have $OSTCOUNT stripes ..."
2572 $LFS setstripe -c $OSTCOUNT $DIR1/$tdir/$tfile &
2576 # write something to the file, it should be blocked on fetching layout
2577 dd if=/dev/zero of=$DIR2/$tdir/$tfile bs=1k count=1 conv=notrunc
2578 local stripe_count=$($LFS getstripe -c $DIR2/$tdir/$tfile)
2581 # lod_qos.c::min_stripe_count() allows setstripe with a default stripe
2582 # count to succeed with only 3/4 of the number of stripes (rounded up),
2583 # so creating striped files does not fail if an OST is offline or full
2584 [ $stripe_count -ge $((OSTCOUNT - $OSTCOUNT / 4)) ] ||
2585 error "bad layout: getstripe -c $stripe_count < $OSTCOUNT * 3/4"
2589 run_test 51c "layout lock: IT_LAYOUT blocked and correct layout can be returned"
2592 dd if=/dev/zero of=/$DIR1/$tfile bs=1M count=1
2593 cancel_lru_locks mdc
2595 # open should grant LAYOUT lock, mmap and read will install pages
2596 $MULTIOP $DIR1/$tfile oO_RDWR:SMR_Uc &
2600 # rss before revoking
2601 local br=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}')
2602 echo "Before revoking layout lock: $br KB mapped"
2604 # cancel layout lock manually
2605 cancel_lru_locks mdc
2607 # rss after revoking
2608 local ar=$(grep -A 10 $tfile /proc/$PID/smaps | awk '/^Rss/{print $2}')
2611 wait $PID || error "wait PID $PID failed"
2613 [ $ar -eq 0 ] || error "rss before: $br, after $ar, some pages remained"
2615 run_test 51d "layout lock: losing layout lock should clean up memory map region"
2619 echo "==> rename vs getattr vs setxattr should not deadlock"
2620 mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed"
2622 do_facet mds1 $LCTL set_param fail_loc=$1
2624 mv -T $DIR/d1/d2/d3 $DIR/d1/d3 &
2632 setfattr -n user.attr1 -v value1 $DIR2/d1 || error "(2) setfattr failed"
2633 wait $PID1 || error "(3) mv failed"
2634 wait $PID2 || error "(4) stat failed"
2641 echo "==> rename vs getattr vs open vs getattr should not deadlock"
2642 mkdir -p $DIR/d1/d2/d3 || error "(1) mkdir failed"
2644 do_facet mds1 $LCTL set_param fail_loc=$1
2646 mv -T $DIR/d1/d2/d3 $DIR/d1/d3 &
2654 $MULTIOP $DIR2/d1/d2 Oc &
2658 stat $DIR/d1 || error "(2) stat failed"
2660 wait $PID1 || error "(3) mv failed"
2661 wait $PID2 || error "(4) stat failed"
2662 wait $PID3 && error "(5) multiop failed"
2668 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
2669 save_lustre_params client "llite.*.xattr_cache" > $p
2670 lctl set_param llite.*.xattr_cache 1 ||
2671 { skip "xattr cache is not supported"; return 0; }
2673 #define OBD_FAIL_MDS_RENAME 0x153
2674 #define OBD_FAIL_MDS_RENAME2 0x154
2675 test_54_part1 0x80000153 || error 10
2676 test_54_part1 0x80000154 || error 11
2677 test_54_part2 0x80000153 || error 12
2678 test_54_part2 0x80000154 || error 13
2680 restore_lustre_params < $p
2683 run_test 54 "rename locking"
2686 mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
2688 #define OBD_FAIL_MDS_RENAME4 0x156
2689 do_facet mds1 $LCTL set_param fail_loc=0x80000156
2691 mv -T $DIR/d1/d2 $DIR/d3/d2 &
2696 wait $PID1 && error "(2) mv succeeded"
2700 run_test 55a "rename vs unlink target dir"
2704 mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
2706 #define OBD_FAIL_MDS_RENAME4 0x156
2707 do_facet mds1 $LCTL set_param fail_loc=0x80000156
2709 mv -T $DIR/d1/d2 $DIR/d3/d2 &
2714 wait $PID1 && error "(2) mv succeeded"
2718 run_test 55b "rename vs unlink source dir"
2722 mkdir -p $DIR/d1/d2 $DIR/d3 || error "(1) mkdir failed"
2724 #define OBD_FAIL_MDS_RENAME4 0x156
2725 do_facet mds1 $LCTL set_param fail_loc=0x156
2727 mv -T $DIR/d1/d2 $DIR/d3/d2 &
2731 # while rename is sleeping, open and remove d3
2732 $MULTIOP $DIR2/d3 D_c &
2738 # while rename is sleeping 2nd time, close d3
2740 wait $PID2 || error "(3) multiop failed"
2742 wait $PID1 && error "(2) mv succeeded"
2746 run_test 55c "rename vs unlink orphan target dir"
2752 #define OBD_FAIL_MDS_RENAME3 0x155
2753 do_facet mds1 $LCTL set_param fail_loc=0x155
2754 mv $DIR/f1 $DIR/$tdir &
2758 # while rename is sleeping, create $tdir, but as a directory
2759 mkdir -p $DIR2/$tdir || error "(1) mkdir failed"
2761 # link in reverse locking order
2762 ln $DIR2/f1 $DIR2/$tdir/
2764 wait $PID1 && error "(2) mv succeeded"
2767 run_test 55d "rename file vs link"
2770 local MDSVER=$(lustre_build_version $SINGLEMDS)
2771 [ $(version_code $MDSVER) -lt $(version_code 2.3.0) ] &&
2772 skip "MDS version $MDSVER must be >= 2.3.0" && return 0
2775 test_mkdir $DIR1/$tdir
2776 file1=$DIR1/$tdir/file
2777 file2=$DIR2/$tdir/file
2779 echo orig > $file2 || error "Could not create $file2"
2780 version=$($LFS data_version $file1)
2783 echo append >> $file2 || error "Could not append to $file2"
2784 version2=$($LFS data_version $file1)
2785 [ "$version" != "$version2" ] ||
2786 error "append did not change data version: $version"
2789 echo overwrite > $file2 || error "Could not overwrite $file2"
2790 version3=$($LFS data_version $file1)
2791 [ "$version2" != "$version3" ] ||
2792 error "overwrite did not change data version: $version2"
2794 # Truncate before EOF
2795 $TRUNCATE $file2 3 || error "Could not truncate $file2"
2796 version4=$($LFS data_version $file1)
2797 [ "$version3" != "$version4" ] ||
2798 error "truncate did not change data version: $version3"
2800 # Truncate after EOF
2801 $TRUNCATE $file2 123456 || error "Could not truncate $file2"
2802 version5=$($LFS data_version $file1)
2803 [ "$version4" != "$version5" ] ||
2804 error "truncate did not change data version: $version4"
2806 # Chmod do not change version
2807 chmod 400 $file2 || error "Could not chmod 400 $file2"
2808 version6=$($LFS data_version $file1)
2809 [ "$version5" == "$version6" ] ||
2810 error "chmod should not change data version: $version5 != $version6"
2812 # Chown do not change version
2813 chown $RUNAS_ID $file2 || error "Could not chown $RUNAS_ID $file2"
2814 version7=$($LFS data_version $file1)
2815 [ "$version5" == "$version7" ] ||
2816 error "chown should not change data version: $version5 != $version7"
2818 run_test 60 "Verify data_version behaviour"
2821 local test_dir=$tdir/test_dir
2823 mkdir -p $DIR1/$tdir
2824 if [ $MDSCOUNT -ge 2 ]; then
2826 $LFS mkdir -i $MDTIDX $DIR1/$test_dir ||
2827 error "Create remote directory failed"
2829 mkdir -p $DIR1/$test_dir
2831 cd $DIR2/$test_dir || error "cd directory failed"
2832 rm -rf $DIR1/$test_dir || error "unlink directory failed"
2834 cd $DIR2/$tdir || error "exit directory"
2836 run_test 70a "cd directory && rm directory"
2838 test_70b() { # LU-2781
2840 mkdir -p $DIR1/$tdir
2842 touch $DIR1/$tdir/file
2843 for ((i = 0; i < 32; i++)); do
2844 $LFS rm_entry $DIR1/$tdir/non_existent_dir &>/dev/null
2846 rm $DIR1/$tdir/file || error "cannot remove file after rm_entry"
2848 touch $DIR1/$tdir/file
2849 $LFS mkdir -i0 $DIR1/$tdir/test_dir
2850 $LFS rm_entry $DIR1/$tdir/test_dir &>/dev/null
2851 rm -rf $DIR1/$tdir/test_dir ||
2852 error "cannot remove directory after rm_entry"
2853 rm $DIR1/$tdir/file || error "cannot remove file after rm_entry"
2855 run_test 70b "remove files after calling rm_entry"
2858 local server_version=$(lustre_version_code $SINGLEMDS)
2860 [[ $server_version -lt $(version_code 2.1.6) ]] &&
2861 skip "Need MDS version at least 2.1.6" && return
2863 # Patch not applied to 2.2 and 2.3 branches
2864 [[ $server_version -ge $(version_code 2.2.0) ]] &&
2865 [[ $server_version -lt $(version_code 2.4.0) ]] &&
2866 skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" &&
2869 checkfiemap --test ||
2870 { skip "checkfiemap not runnable: $?" && return; }
2871 # write data this way: hole - data - hole - data
2872 dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1
2873 [ "$(facet_fstype ost$(($($LFS getstripe -i $DIR1/$tfile) + 1)))" = \
2875 skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
2876 dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1
2877 GET_STAT="lctl get_param -n ldlm.services.ldlm_cbd.stats"
2879 local can1=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
2881 checkfiemap $DIR2/$tfile 81920 ||
2882 error "data is not flushed from client"
2883 local can2=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
2886 # common case of "create file, copy file" on a single node
2887 # should not flush data from ost
2888 dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=1 count=1
2889 dd if=/dev/urandom of=$DIR1/$tfile bs=40K seek=3 count=1
2891 local can3=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
2893 checkfiemap $DIR1/$tfile 81920 ||
2895 local can4=$($GET_STAT | awk '/ldlm_bl_callback/ {print $2}')
2897 [ $can3 -eq $can4 ] || error $((can2-can1)) "cancel RPC occured."
2899 run_test 71a "correct file map just after write operation is finished"
2902 local server_version=$(lustre_version_code $SINGLEMDS)
2904 [[ $server_version -lt $(version_code 2.1.6) ]] &&
2905 skip "Need MDS version at least 2.1.6" && return
2907 # Patch not applied to 2.2 and 2.3 branches
2908 [[ $server_version -ge $(version_code 2.2.0) ]] &&
2909 [[ $server_version -lt $(version_code 2.4.0) ]] &&
2910 skip "Need MDS version earlier than 2.2.0 or at least 2.4.0" &&
2912 [[ $OSTCOUNT -ge 2 ]] || { skip "needs >= 2 OSTs"; return; }
2914 checkfiemap --test ||
2915 { skip "error $?: checkfiemap failed" && return; }
2917 mkdir -p $DIR1/$tdir
2919 $LFS setstripe -c -1 $DIR1/$tdir || error "setstripe failed"
2920 dd if=/dev/urandom of=$DIR1/$tdir/$tfile bs=40K count=1
2921 [ "$(facet_fstype ost$(($($LFS getstripe -i $DIR1/$tdir/$tfile) + 1)))" = \
2923 skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS" && return 0
2924 checkfiemap $DIR1/$tdir/$tfile 40960 || error "checkfiemap failed"
2926 run_test 71b "check fiemap support for stripecount > 1"
2929 local p="$TMP/sanityN-$TESTNAME.parameters"
2932 save_lustre_params client "llite.*.xattr_cache" > $p
2933 lctl set_param llite.*.xattr_cache 1 ||
2934 { skip "xattr cache is not supported"; return 0; }
2937 setfattr -n user.attr1 -v value1 $DIR1/$tfile ||
2938 error "setfattr1 failed"
2939 getfattr -n user.attr1 $DIR2/$tfile | grep value1 ||
2940 error "getfattr1 failed"
2941 setfattr -n user.attr1 -v value2 $DIR2/$tfile ||
2942 error "setfattr2 failed"
2943 getfattr -n user.attr1 $DIR1/$tfile | grep value2 ||
2944 error "getfattr2 failed"
2946 # check that trusted.link is consistent
2947 tlink1=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
2948 ln $DIR2/$tfile $DIR2/$tfile-2 || error "failed to link"
2949 tlink2=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
2950 echo "$tlink1 $tlink2"
2951 [ "$tlink1" = "$tlink2" ] && error "trusted.link should have changed!"
2955 restore_lustre_params < $p
2958 run_test 72 "getxattr/setxattr cache should be consistent between nodes"
2961 local p="$TMP/sanityN-$TESTNAME.parameters"
2962 save_lustre_params client "llite.*.xattr_cache" > $p
2963 lctl set_param llite.*.xattr_cache 1 ||
2964 { skip "xattr cache is not supported"; return 0; }
2967 setfattr -n user.attr1 -v value1 $DIR1/$tfile ||
2968 error "setfattr1 failed"
2969 getfattr -n user.attr1 $DIR2/$tfile || error "getfattr1 failed"
2970 getfattr -n user.attr1 $DIR1/$tfile || error "getfattr2 failed"
2971 clear_stats llite.*.stats
2972 # PR lock should be cached by now on both clients
2973 getfattr -n user.attr1 $DIR1/$tfile || error "getfattr3 failed"
2974 # 2 hits for getfattr(0)+getfattr(size)
2975 [ $(calc_stats llite.*.stats getxattr_hits) -eq 2 ] ||
2976 error "not cached in $DIR1"
2977 getfattr -n user.attr1 $DIR2/$tfile || error "getfattr4 failed"
2978 # 4 hits for more getfattr(0)+getfattr(size)
2979 [ $(calc_stats llite.*.stats getxattr_hits) -eq 4 ] ||
2980 error "not cached in $DIR2"
2983 restore_lustre_params < $p
2986 run_test 73 "getxattr should not cause xattr lock cancellation"
2989 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.93) ] &&
2990 skip "Need MDS version at least 2.4.93" && return
2992 dd if=/dev/zero of=$DIR1/$tfile-1 bs=1K count=1
2993 dd if=/dev/zero of=$DIR1/$tfile-2 bs=1K count=1
2994 flocks_test 4 $DIR1/$tfile-1 $DIR2/$tfile-2
2996 run_test 74 "flock deadlock: different mounts =============="
3000 $LFS setstripe -c 2 -S 1m -i 0 $DIR1/$tfile
3001 dd if=/dev/zero of=$DIR1/$tfile bs=1M count=2
3002 cancel_lru_locks osc
3004 dd of=$DIR1/$tfile if=/dev/zero bs=1M count=1 seek=1 conv=notrunc
3007 # define OBD_FAIL_LDLM_ENQUEUE_HANG 0x31d
3008 $LCTL set_param fail_loc=0x31d
3009 stat -c %s $DIR1/$tfile &
3014 # For bad lock error handler we should ASSERT and got kernel panic here
3016 $LCTL set_param fail_loc=0
3018 run_test 75 "osc: upcall after unuse lock==================="
3021 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.53) ]] &&
3022 skip "Need MDS version at least 2.5.53" && return
3024 remote_mds_nodsh && skip "remote MDS with nodsh" && return
3025 local fcount=$((MDSCOUNT * 256))
3030 nid=$($LCTL list_nids | sed "s/\./\\\./g")
3036 test_mkdir $DIR/$tdir
3038 # drop all open locks and close any cached "open" files on the client
3039 cancel_lru_locks mdc
3041 local open_fids_cmd="$LCTL get_param -n mdt.*.exports.'$nid'.open_files"
3042 local fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $open_fids_cmd))
3043 local already=${#fid_list[@]}
3044 for (( i = 0; i < $already; i++ )) ; do
3045 log "already open[$i]: $($LFS fid2path $DIR2 ${fid_list[i]})"
3048 echo -n "opening files: "
3049 ulimit -n $((fcount + 50))
3050 for ((i = 0; i < $fcount; i++)); do
3051 touch $DIR/$tdir/f_$i
3052 local fd=$(free_fd ${fd_list[i]})
3053 local open_cmd="exec $fd<$DIR/$tdir/f_$i"
3058 (( $i % 32 == 0 )) && echo -n "."
3062 fid_list=($(do_nodes $(comma_list $(mdts_nodes)) $open_fids_cmd))
3064 # Possible errors in openfiles FID list.
3065 # 1. Missing FIDs. Check 1
3066 # 2. Extra FIDs. Check 1
3067 # 3. Duplicated FID. Check 2
3068 # 4. Invalid FIDs. Check 2
3069 # 5. Valid FID, points to some other file. Check 3
3072 [ ${#fid_list[@]} -ne $((fcount + already)) ] &&
3073 error "${#fid_list[@]} != $fcount (+$already old) open files"
3075 echo -n "closing files: "
3076 for (( fd = 0, fid = 0; fd < $fcount; fd++, fid++ )) ; do
3077 local close_cmd="exec ${fd_list[fd]}<&-"
3079 filename=$($LFS fid2path $DIR2 ${fid_list[fid]})
3081 while [[ ! "$filename" =~ "$DIR2/$tdir/f_" ]]; do
3082 echo "skip old open file $filename"
3084 filename=$($LFS fid2path $DIR2 ${fid_list[fid]})
3088 rm --interactive=no $filename
3090 error "Nonexisting fid ${fid_list[fid]} listed."
3091 (( $fd % 32 == 0 )) && echo -n "."
3096 ls_op=$(ls $DIR2/$tdir | wc -l)
3098 error "Some openfiles are missing in lproc output"
3102 run_test 76 "Verify MDT open_files listing"
3106 local dir=$DIR/$tdir
3109 mkdir $dir || error "mkdir $dir failed"
3110 $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed"
3113 do_nodes $CLIENTS $myRUNAS \
3114 dd if=/dev/zero of="$dir/nrs_r_\$HOSTNAME" bs=1M count=$n ||
3115 error "dd at 0 on client failed (1)"
3117 do_nodes $CLIENTS $myRUNAS \
3119 for ((i = 0; i < $n; i++)); do
3120 dd if=/dev/zero of=$dir/nrs_w_\$HOSTNAME bs=1M \
3121 seek=\\\$i count=1 conv=notrunc &
3122 pids_w[\\\$i]=\\\$!;
3125 for ((i = 0; i < $n; i++)); do
3126 wait \\\${pids_w[\\\$i]};
3128 [ \\\$newrc -gt \\\$rc_w ] && rc_w=\\\$newrc;
3132 do_nodes $CLIENTS sync;
3133 cancel_lru_locks osc
3135 do_nodes $CLIENTS $myRUNAS \
3137 for ((i = 0; i < $n; i++)); do
3138 dd if=$dir/nrs_r_\$HOSTNAME bs=1M of=/dev/null \
3139 seek=\\\$i count=1 &
3140 pids_r[\\\$i]=\\\$!;
3143 for ((i = 0; i < $n; i++)); do
3144 wait \\\${pids_r[\\\$i]};
3146 [ \\\$newrc -gt \\\$rc_r ] && rc_r=\\\$newrc;
3150 cancel_lru_locks osc
3152 wait $pid_w || error "dd (write) failed (2)"
3153 wait $pid_r || error "dd (read) failed (3)"
3154 rm -rvf $dir || error "rm -rf $dir failed"
3157 test_77a() { #LU-3266
3160 oss=$(comma_list $(osts_nodes))
3161 do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="fifo" ||
3163 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
3164 [[ $rc -ne 0 ]] && error "failed to set fifo policy"
3169 run_test 77a "check FIFO NRS policy"
3171 test_77b() { #LU-3266
3174 oss=$(comma_list $(osts_nodes))
3176 do_nodes $oss lctl set_param ost.OSS.*.nrs_policies="crrn" \
3177 ost.OSS.*.nrs_crrn_quantum=1 || rc=$?
3178 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
3179 [[ $rc -ne 0 ]] && error "failed to set crrn_quantum to 1"
3181 echo "policy: crr-n, crrn_quantum 1"
3184 do_nodes $oss lctl set_param \
3185 ost.OSS.*.nrs_crrn_quantum=64 || rc=$?
3186 [[ $rc -ne 0 ]] && error "failed to set crrn_quantum to 64"
3188 echo "policy: crr-n, crrn_quantum 64"
3192 do_nodes $oss lctl set_param \
3193 ost.OSS.ost_io.nrs_policies="fifo" || rc=$?
3194 [[ $rc -ne 0 ]] && error "failed to set fifo policy"
3197 run_test 77b "check CRR-N NRS policy"
3202 oss=$(comma_list $(osts_nodes))
3204 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies=$policy \
3205 ost.OSS.*.nrs_"$policy"_quantum=1 \
3206 ost.OSS.*.nrs_"$policy"_offset_type="physical" \
3207 ost.OSS.*.nrs_"$policy"_supported="reads" || return $?
3209 echo "policy: $policy, ${policy}_quantum 1, ${policy}_offset_type " \
3210 "physical, ${policy}_supported reads"
3213 do_nodes $oss lctl set_param \
3214 ost.OSS.*.nrs_${policy}_supported="writes" \
3215 ost.OSS.*.nrs_${policy}_quantum=64 || return $?
3217 echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type " \
3218 "physical, ${policy}_supported writes"
3221 do_nodes $oss lctl set_param \
3222 ost.OSS.*.nrs_${policy}_supported="reads_and_writes" \
3223 ost.OSS.*.nrs_${policy}_offset_type="logical" || return $?
3224 echo "policy: $policy, ${policy}_quantum 64, ${policy}_offset_type " \
3225 "logical, ${policy}_supported reads_and_writes"
3229 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo" ||
3234 test_77c() { #LU-3266
3236 orr_trr "orr" || rc=$?
3237 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
3238 [[ $rc -ne 0 ]] && error "orr_trr failed rc:$rc"
3241 run_test 77c "check ORR NRS policy"
3243 test_77d() { #LU-3266
3245 orr_trr "trr" || rc=$?
3246 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
3247 [[ $rc -ne 0 ]] && error "orr_trr failed rc:$rc"
3250 run_test 77d "check TRR nrs policy"
3257 do_facet $facet lctl set_param \
3258 ost.OSS.ost_io.nrs_tbf_rule="$*"
3260 error "failed to run operate '$*' on TBF rules"
3263 cleanup_tbf_verify()
3267 echo "cleanup_tbf $DIR/$tdir"
3270 wait_delete_completed
3275 local dir=$DIR/$tdir
3276 local client1=${CLIENT1:-$(hostname)}
3279 local np=$(check_cpt_number ost1)
3280 [ $np -gt 0 ] || error "CPU partitions should not be $np."
3281 echo "cpu_npartitions on ost1 is $np"
3283 mkdir $dir || error "mkdir $dir failed"
3284 $LFS setstripe -c 1 -i 0 $dir || error "setstripe to $dir failed"
3287 trap cleanup_tbf_verify EXIT
3288 echo "Limited write rate: $1, read rate: $2"
3289 echo "Verify the write rate is under TBF control"
3290 local start=$SECONDS
3291 do_node $client1 $myRUNAS dd if=/dev/zero of=$dir/tbf \
3292 bs=1M count=100 oflag=direct 2>&1
3293 local runtime=$((SECONDS - start + 1))
3294 local rate=$(bc <<< "scale=6; 100 / $runtime")
3295 echo "Write runtime is $runtime s, speed is $rate IOPS"
3297 # verify the write rate does not exceed TBF rate limit
3298 [ $(bc <<< "$rate < 1.1 * $np * $1") -eq 1 ] ||
3299 error "The write rate ($rate) exceeds 110% of rate limit ($1 * $np)"
3301 cancel_lru_locks osc
3303 echo "Verify the read rate is under TBF control"
3305 do_node $client1 $myRUNAS dd if=$dir/tbf of=/dev/null \
3306 bs=1M count=100 iflag=direct 2>&1
3307 runtime=$((SECONDS - start + 1))
3308 rate=$(bc <<< "scale=6; 100 / $runtime")
3309 echo "Read runtime is $runtime s, speed is $rate IOPS"
3311 # verify the read rate does not exceed TBF rate limit
3312 [ $(bc <<< "$rate < 1.1 * $np * $2") -eq 1 ] ||
3313 error "The read rate ($rate) exceeds 110% of rate limit ($2 * $np)"
3315 cancel_lru_locks osc
3316 cleanup_tbf_verify || error "rm -rf $dir failed"
3322 oss=$(comma_list $(osts_nodes))
3324 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" ||
3326 [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
3327 [[ $rc -ne 0 ]] && error "failed to set TBF NID policy"
3331 if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then
3336 # Only operate rules on ost1 since OSTs might run on the same OSS
3338 tbf_rule_operate ost1 "start\ localhost\ ${idis}{0@lo}\ ${rateis}1000"
3339 local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)")
3340 local client_nids=$(nids_list $address "\\")
3341 tbf_rule_operate ost1 "start\ clients\ ${idis}{$client_nids}\ ${rateis}100"
3342 tbf_rule_operate ost1 "start\ others\ ${idis}{*.*.*.*@$NETTYPE}\ ${rateis}50"
3346 tbf_rule_operate ost1 "change\ localhost\ ${rateis}1001"
3347 tbf_rule_operate ost1 "change\ clients\ ${rateis}101"
3348 tbf_rule_operate ost1 "change\ others\ ${rateis}51"
3352 tbf_rule_operate ost1 "stop\ localhost"
3353 tbf_rule_operate ost1 "stop\ clients"
3354 tbf_rule_operate ost1 "stop\ others"
3357 # Cleanup the TBF policy
3358 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
3359 [ $? -ne 0 ] && error "failed to set policy back to fifo"
3363 run_test 77e "check TBF NID nrs policy"
3368 oss=$(comma_list $(osts_nodes))
3370 do_nodes $oss $LCTL set_param \
3371 ost.OSS.ost_io.nrs_policies="tbf\ jobid" || rc=$?
3372 [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
3373 [[ $rc -ne 0 ]] && error "failed to set TBF JOBID policy"
3375 # Configure jobid_var
3376 local saved_jobid_var=$($LCTL get_param -n jobid_var)
3378 [[ $rc -eq 3 ]] && skip "jobid_var not found" && return
3379 [[ $rc -ne 0 ]] && error "failed to get param jobid_var"
3380 if [ $saved_jobid_var != procname_uid ]; then
3381 set_persistent_param_and_check client \
3382 "jobid_var" "$FSNAME.sys.jobid_var" procname_uid
3387 if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then
3392 # Only operate rules on ost1 since OSTs might run on the same OSS
3394 tbf_rule_operate ost1 "start\ runas\ ${idis}{iozone.$RUNAS_ID\ dd.$RUNAS_ID\ tiotest.$RUNAS_ID}\ ${rateis}1000"
3395 tbf_rule_operate ost1 "start\ iozone_runas\ ${idis}{iozone.$RUNAS_ID}\ ${rateis}100"
3396 tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50"
3397 nrs_write_read "$RUNAS"
3400 tbf_rule_operate ost1 "change\ runas\ ${rateis}1001"
3401 tbf_rule_operate ost1 "change\ iozone_runas\ ${rateis}101"
3402 tbf_rule_operate ost1 "change\ dd_runas\ ${rateis}51"
3403 nrs_write_read "$RUNAS"
3406 tbf_rule_operate ost1 "stop\ runas"
3407 tbf_rule_operate ost1 "stop\ iozone_runas"
3408 tbf_rule_operate ost1 "stop\ dd_runas"
3409 nrs_write_read "$RUNAS"
3411 # Cleanup the TBF policy
3412 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
3413 [ $? -ne 0 ] && error "failed to set policy back to fifo"
3414 nrs_write_read "$RUNAS"
3416 local current_jobid_var=$($LCTL get_param -n jobid_var)
3417 [[ $? -ne 0 ]] && error "failed to get param jobid_var"
3418 if [ $saved_jobid_var != $current_jobid_var ]; then
3419 set_persistent_param_and_check client \
3420 "jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid_var
3424 run_test 77f "check TBF JobID nrs policy"
3429 oss=$(comma_list $(osts_nodes))
3431 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid" ||
3433 [[ $rc -eq 3 ]] && skip "no NRS TBF exists" && return
3434 [[ $rc -ne 0 ]] && error "failed to set TBF NID policy"
3436 do_nodes $oss lctl set_param \
3437 ost.OSS.ost_io.nrs_policies="tbf\ jobid" || rc=$?
3438 [[ $rc -ne 0 ]] && error "failed to set TBF JOBID policy"
3442 if [ $(lustre_version_code ost1) -ge $(version_code 2.8.54) ]; then
3447 # Add a rule that only valid for Jobid TBF. If direct change between
3448 # TBF types is not supported, this operation will fail.
3449 tbf_rule_operate ost1 "start\ dd_runas\ ${idis}{dd.$RUNAS_ID}\ ${rateis}50"
3451 # Cleanup the TBF policy
3452 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
3453 [ $? -ne 0 ] && error "failed to set policy back to fifo"
3456 run_test 77g "Change TBF type directly"
3459 [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] ||
3460 { skip "Need OST version at least 2.8.55"; return 0; }
3462 local old_policy=$(do_facet ost1 \
3463 lctl get_param ost.OSS.ost_io.nrs_policies)
3466 do_facet ost1 lctl set_param \
3467 ost.OSS.ost_io.nrs_policies="abc"
3468 [ $? -eq 0 ] && error "should return error"
3470 do_facet ost1 lctl set_param \
3471 ost.OSS.ost_io.nrs_policies="tbf\ abc"
3472 [ $? -eq 0 ] && error "should return error"
3474 do_facet ost1 lctl set_param \
3475 ost.OSS.ost_io.nrs_policies="tbf\ reg\ abc"
3476 [ $? -eq 0 ] && error "should return error"
3478 do_facet ost1 lctl set_param \
3479 ost.OSS.ost_io.nrs_policies="tbf\ abc\ efg"
3480 [ $? -eq 0 ] && error "should return error"
3482 new_policy=$(do_facet ost1 lctl get_param ost.OSS.ost_io.nrs_policies)
3483 [ $? -eq 0 ] || error "shouldn't LBUG"
3485 [ "$old_policy" = "$new_policy" ] || error "NRS policy should be same"
3489 run_test 77h "Wrong policy name should report error, not LBUG"
3495 local error_message=$3
3497 for rule in $expected; do
3498 rule_number=$((rule_number + 1))
3500 local stop_line=$(($rule_number + 3))
3501 local awk_command="awk 'NR >= 4 && NR <= $stop_line {print \$1}'"
3503 local output=$(do_facet $facet lctl get_param \
3504 ost.OSS.ost_io.nrs_tbf_rule |
3508 if [ "$output" != "$expected" ]; then
3509 error "$error_message, expected '$expected', got '$output'"
3514 [ $(lustre_version_code ost1) -ge $(version_code 2.8.55) ] ||
3515 { skip "Need OST version at least 2.8.55"; return 0; }
3517 for i in $(seq 1 $OSTCOUNT)
3519 do_facet ost"$i" lctl set_param \
3520 ost.OSS.ost_io.nrs_policies="tbf\ jobid"
3522 error "failed to set TBF policy"
3525 tbf_rule_check ost1 "default" "error before inserting any rule"
3527 tbf_rule_operate ost1 "start\ before\ jobid={jobid}\ rate=1000"
3528 tbf_rule_check ost1 "before default" \
3529 "error when inserting rule 'before'"
3531 tbf_rule_operate ost1 "start\ after\ jobid={jobid}\ rate=1000\ rank=default"
3532 tbf_rule_check ost1 "before after default" \
3533 "error when inserting rule 'after'"
3535 tbf_rule_operate ost1 "start\ target\ jobid={jobid}\ rate=1000\ rank=after"
3536 tbf_rule_check ost1 "before target after default" \
3537 "error when inserting rule 'target'"
3539 echo "Move before itself"
3540 tbf_rule_operate ost1 "change\ target\ rank=target"
3541 tbf_rule_check ost1 "before target after default" \
3542 "error when moving before itself"
3544 echo "Move to higher rank"
3545 tbf_rule_operate ost1 "change\ target\ rank=before"
3546 tbf_rule_check ost1 "target before after default" \
3547 "error when moving to higher rank"
3549 echo "Move to lower rank"
3550 tbf_rule_operate ost1 "change\ target\ rank=after"
3551 tbf_rule_check ost1 "before target after default" \
3552 "error when moving to lower rank"
3554 echo "Move before default"
3555 tbf_rule_operate ost1 "change\ target\ rank=default"
3556 tbf_rule_check ost1 "before after target default" \
3557 error "error when moving before default"
3559 # Cleanup the TBF policy
3560 do_nodes $(comma_list $(osts_nodes)) \
3561 $LCTL set_param ost.OSS.ost_io.nrs_policies=fifo
3564 run_test 77i "Change rank of TBF rule"
3569 local ost_version=$(lustre_version_code ost1)
3571 [ $ost_version -ge $(version_code 2.9.53) ] ||
3572 { skip "Need OST version at least 2.9.53"; return 0; }
3573 if [ $ost_version -ge $(version_code 2.8.60) ]; then
3578 do_nodes $(comma_list $(osts_nodes)) \
3579 lctl set_param jobid_var=procname_uid \
3580 ost.OSS.ost_io.nrs_policies="tbf\ opcode" \
3581 ost.OSS.ost_io.nrs_tbf_rule="start\ ost_r\ ${idis}{ost_read}\ ${rateis}5" \
3582 ost.OSS.ost_io.nrs_tbf_rule="start\ ost_w\ ${idis}{ost_write}\ ${rateis}20"
3583 [ $? -ne 0 ] && error "failed to set TBF OPCode policy"
3588 do_nodes $(comma_list $(osts_nodes)) \
3589 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_r" \
3590 ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_w" \
3591 ost.OSS.ost_io.nrs_policies="fifo"
3593 # sleep 3 seconds to wait the tbf policy stop completely,
3594 # or the next test case is possible get -EAGAIN when
3595 # setting the tbf policy
3598 run_test 77j "check TBF-OPCode NRS policy"
3601 local idstr="${1}id"
3602 local policy="${idstr}={$2}"
3603 local rate="rate=$3"
3605 do_nodes $(comma_list $(osts_nodes)) \
3606 lctl set_param jobid_var=procname_uid \
3607 ost.OSS.ost_io.nrs_policies="tbf\ ${idstr}" \
3608 ost.OSS.ost_io.nrs_tbf_rule="start\ ost_${idstr}\ ${policy}\ ${rate}"
3609 [ $? -ne 0 ] && error "failed to set tbf ${idstr} policy"
3611 nrs_write_read "runas $4"
3612 tbf_verify $3 $3 "runas $4"
3614 do_nodes $(comma_list $(osts_nodes)) \
3615 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ost_${idstr}" \
3616 ost.OSS.ost_io.nrs_policies="fifo"
3618 # sleep 3 seconds to wait the tbf policy stop completely,
3619 # or the next test case is possible get -eagain when
3620 # setting the tbf policy
3625 if [ $(lustre_version_code ost1) -lt $(version_code 2.11.50) ]; then
3626 skip "Need OST version at least 2.11.50"
3629 test_id "u" "500" "5" "-u 500"
3630 test_id "g" "500" "5" "-u 500 -g 500"
3632 run_test 77ja "check TBF-UID/GID NRS policy"
3640 for rule in $rule_lists; do
3641 do_nodes $(comma_list $(osts_nodes)) \
3642 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ $rule"
3645 do_nodes $(comma_list $(osts_nodes)) \
3646 lctl set_param ost.OSS.ost_io.nrs_policies="$old_nrs"
3652 [[ $(lustre_version_code ost1) -ge $(version_code 2.9.53) ]] ||
3653 { skip "Need OST version at least 2.9.53"; return 0; }
3655 do_nodes $(comma_list $(osts_nodes)) \
3656 lctl set_param ost.OSS.ost_io.nrs_policies="tbf" \
3657 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_w\ jobid={dd.$RUNAS_ID}\&opcode={ost_write}\ rate=20" \
3658 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_r\ jobid={dd.$RUNAS_ID}\&opcode={ost_read}\ rate=10"
3660 nrs_write_read "$RUNAS"
3661 tbf_verify 20 10 "$RUNAS"
3663 local address=$(comma_list "$(host_nids_address $CLIENTS $NETTYPE)")
3664 local client_nids=$(nids_list $address "\\")
3665 do_nodes $(comma_list $(osts_nodes)) \
3666 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_w" \
3667 ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_r" \
3668 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_w\ nid={0@lo\ $client_nids}\&opcode={ost_write}\ rate=20" \
3669 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_r\ nid={0@lo\ $client_nids}\&opcode={ost_read}\ rate=10"
3674 do_nodes $(comma_list $(osts_nodes)) \
3675 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_w" \
3676 ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_r" \
3677 ost.OSS.ost_io.nrs_tbf_rule="start\ ext\ nid={0@lo\ $client_nids}\&jobid={dd.$RUNAS_ID}\ rate=20"
3679 nrs_write_read "$RUNAS"
3680 tbf_verify 20 20 "$RUNAS"
3682 do_nodes $(comma_list $(osts_nodes)) \
3683 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext" \
3684 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ jobid={dd.$RUNAS_ID},opcode={ost_write}\ rate=20" \
3685 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ jobid={dd.$RUNAS_ID},opcode={ost_read}\ rate=10"
3687 nrs_write_read "$RUNAS"
3688 # with parameter "RUNAS", it will match the latest rule
3689 # "ext_b" first, so the limited write rate is 10.
3690 tbf_verify 10 10 "$RUNAS"
3693 trap "cleanup_77k \"ext_a ext_b\" \"fifo\"" EXIT
3695 [[ $(lustre_version_code ost1) -ge $(version_code 2.10.58) ]] ||
3696 { skip "Need OST version at least 2.10.58"; return 0; }
3698 do_nodes $(comma_list $(osts_nodes)) \
3699 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_a" \
3700 ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_b" \
3701 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ug\ uid={500}\&gid={1000}\ rate=5"
3702 nrs_write_read "runas -u 500 -g 1000"
3703 tbf_verify 5 5 "runas -u 500 -g 1000"
3705 do_nodes $(comma_list $(osts_nodes)) \
3706 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ug" \
3707 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_uw\ uid={500}\&opcode={ost_write}\ rate=20" \
3708 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_ur\ uid={500}\&opcode={ost_read}\ rate=10"
3710 nrs_write_read "runas -u 500"
3711 tbf_verify 20 10 "runas -u 500"
3713 do_nodes $(comma_list $(osts_nodes)) \
3714 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_uw" \
3715 ost.OSS.ost_io.nrs_tbf_rule="stop\ ext_ur" \
3716 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_a\ uid={500},opcode={ost_write}\ rate=20" \
3717 ost.OSS.ost_io.nrs_tbf_rule="start\ ext_b\ uid={500},opcode={ost_read}\ rate=10"
3718 nrs_write_read "runas -u 500"
3719 tbf_verify 10 10 "runas -u 500"
3720 tbf_verify 20 10 "runas -u 500"
3721 cleanup_77k "ext_a ext_b" "fifo"
3723 run_test 77k "check TBF policy with NID/JobID/OPCode expression"
3726 [[ $(lustre_version_code ost1) -ge $(version_code 2.10.56) ]] ||
3727 { skip "Need OST version at least 2.10.56"; return 0; }
3729 do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ nid"
3730 do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="tbf"
3732 local output=$(do_facet ost1 lctl get_param \
3733 ost.OSS.ost_io.nrs_policies | \
3734 awk '/name: tbf/ {print;exit}' | \
3735 awk -F ': ' '{print $2}')
3737 if [ "$output" != "tbf" ]; then
3738 error "The generic TBF output is '$output', not 'tbf'"
3741 do_facet ost1 lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
3743 run_test 77l "check the output of NRS policies for generic TBF"
3746 if [ $(lustre_version_code ost1) -lt $(version_code 2.9.54) ]; then
3747 skip "Need OST version at least 2.9.54"
3751 local dir=$DIR/$tdir
3753 mkdir $dir || error "mkdir $dir failed"
3754 $LFS setstripe -c $OSTCOUNT $dir || error "setstripe to $dir failed"
3757 local nodes=$(comma_list $(osts_nodes))
3758 do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies=delay \
3759 ost.OSS.ost_io.nrs_delay_min=4 \
3760 ost.OSS.ost_io.nrs_delay_max=4 \
3761 ost.OSS.ost_io.nrs_delay_pct=100
3762 [ $? -ne 0 ] && error "Failed to set delay policy"
3764 local start=$SECONDS
3765 do_nodes "${SINGLECLIENT:-$HOSTNAME}" "$RUNAS" \
3766 dd if=/dev/zero of="$dir/nrs_delay_$HOSTNAME" bs=1M count=1 \
3767 oflag=direct conv=fdatasync ||
3768 { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
3769 error "dd on client failed (1)"; }
3770 local elapsed=$((SECONDS - start))
3772 # NRS delay doesn't do sub-second timing, so a request enqueued at
3773 # 0.9 seconds can be dequeued at 4.0
3774 [ $elapsed -lt 3 ] &&
3775 { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
3776 error "Single 1M write should take at least 3 seconds"; }
3779 do_nodes "${SINGLECLIENT:-$HOSTNAME}" "$RUNAS" \
3780 dd if=/dev/zero of="$dir/nrs_delay_$HOSTNAME" bs=1M count=10 \
3781 oflag=direct conv=fdatasync ||
3782 { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
3783 error "dd on client failed (2)"; }
3784 elapsed=$((SECONDS - start))
3786 [ $elapsed -lt 30 ] &&
3787 { do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo";
3788 error "Ten 1M writes should take at least 30 seconds"; }
3790 do_nodes $nodes lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
3791 [ $? -ne 0 ] && error "failed to set policy back to fifo"
3795 run_test 77m "check NRS Delay slows write RPC processing"
3797 test_77n() { #LU-10802
3798 if [ $(lustre_version_code ost1) -lt $(version_code 2.10.58) ]; then
3799 skip "Need OST version at least 2.10.58"
3803 # Configure jobid_var
3804 local saved_jobid_var=$($LCTL get_param -n jobid_var)
3805 if [ $saved_jobid_var != procname_uid ]; then
3806 set_persistent_param_and_check client \
3807 "jobid_var" "$FSNAME.sys.jobid_var" procname_uid
3810 do_nodes $(comma_list $(osts_nodes)) \
3811 lctl set_param ost.OSS.ost_io.nrs_policies="tbf\ jobid" \
3812 ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
3813 ost.OSS.ost_io.nrs_tbf_rule="start\ dd_runas\ jobid={*.$RUNAS_ID}\ rate=20"
3816 tbf_verify 20 20 "$RUNAS"
3818 do_nodes $(comma_list $(osts_nodes)) \
3819 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
3820 ost.OSS.ost_io.nrs_tbf_rule="start\ dd_runas\ jobid={dd.*}\ rate=20"
3825 do_nodes $(comma_list $(osts_nodes)) \
3826 lctl set_param ost.OSS.ost_io.nrs_tbf_rule="stop\ dd_runas" \
3827 ost.OSS.ost_io.nrs_policies="fifo"
3831 local current_jobid_var=$($LCTL get_param -n jobid_var)
3832 if [ $saved_jobid_var != $current_jobid_var ]; then
3833 set_persistent_param_and_check client \
3834 "jobid_var" "$FSNAME.sys.jobid_var" $saved_jobid_var
3837 run_test 77n "check wildcard support for TBF JobID NRS policy"
3839 test_78() { #LU-6673
3842 oss=$(comma_list $(osts_nodes))
3843 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="orr" &
3844 do_nodes $oss lctl set_param ost.OSS.*.nrs_orr_quantum=1
3846 [[ $rc -eq 3 ]] && skip "no NRS exists" && return
3847 # Valid return codes are:
3848 # 0: Tuning succeeded
3849 # ENODEV: Policy is still stopped
3850 # EAGAIN: Policy is being initialized
3851 [ $rc -eq 0 -o $rc -eq 19 -o $rc -eq 11 ] ||
3852 error "Expected set_param to return 0|ENODEV|EAGAIN"
3854 # Cleanup the ORR policy
3855 do_nodes $oss lctl set_param ost.OSS.ost_io.nrs_policies="fifo"
3856 [ $? -ne 0 ] && error "failed to set policy back to fifo"
3859 run_test 78 "Enable policy and specify tunings right away"
3862 remote_mds_nodsh && skip "remote MDS with nodsh" && return
3863 test_mkdir $DIR/$tdir
3865 # Prevent interference from layout intent RPCs due to
3866 # asynchronous writeback. These will be tested in 130c below.
3867 do_nodes ${CLIENTS:-$HOSTNAME} sync
3869 setfattr -n trusted.name1 -v value1 $DIR/$tdir ||
3870 error "setfattr -n trusted.name1=value1 $DIR/$tdir failed"
3872 #define OBD_FAIL_MDS_INTENT_DELAY 0x160
3873 local mdtidx=$($LFS getstripe -m $DIR/$tdir)
3874 local facet=mds$((mdtidx + 1))
3876 set_nodes_failloc $(facet_active_host $facet) 0x80000160
3877 getfattr -n trusted.name1 $DIR/$tdir 2> /dev/null &
3881 #define OBD_FAIL_MDS_GETXATTR_PACK 0x131
3882 set_nodes_failloc $(facet_active_host $facet) 0x80000131
3887 run_test 79 "xattr: intent error"
3890 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3897 mkdir -p $DIR1/$tdir/dir
3898 createmany -o $DIR1/$tdir/dir/f 10 ||
3899 error "create files under remote dir failed $i"
3901 cp /etc/passwd $DIR1/$tdir/$tfile
3903 #migrate open file should fails
3904 multiop_bg_pause $DIR2/$tdir/$tfile O_c || error "open $file failed"
3906 # give multiop a chance to open
3909 $LFS migrate -m $MDTIDX $DIR1/$tdir &&
3910 error "migrate open files should failed with open files"
3914 $LFS migrate -m $MDTIDX $DIR1/$tdir ||
3915 error "migrate remote dir error"
3917 echo "Finish migration, then checking.."
3918 for file in $(find $DIR1/$tdir); do
3919 mdt_index=$($LFS getstripe -m $file)
3920 [ $mdt_index == $MDTIDX ] ||
3921 error "$file is not on MDT${MDTIDX}"
3924 diff /etc/passwd $DIR1/$tdir/$tfile ||
3925 error "file different after migration"
3927 rm -rf $DIR1/$tdir || error "rm dir failed after migration"
3929 run_test 80a "migrate directory when some children is being opened"
3933 kill -9 $migrate_pid
3937 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
3938 local migrate_dir1=$DIR1/$tdir/migrate_dir
3939 local migrate_dir2=$DIR2/$tdir/migrate_dir
3940 local migrate_run=$LUSTRE/tests/migrate.sh
3948 trap cleanup_80b EXIT
3949 #prepare migrate directory
3950 mkdir -p $migrate_dir1
3951 for F in {1,2,3,4,5}; do
3952 echo "$F$F$F$F$F" > $migrate_dir1/file$F
3953 echo "$F$F$F$F$F" > $DIR/$tdir/file$F
3956 #migrate the directories among MDTs
3959 mdt_idx=$((RANDOM % MDSCOUNT))
3960 $LFS migrate -m $mdt_idx $migrate_dir1 &>/dev/null ||
3962 [ $rc -ne 0 -o $rc -ne 16 ] || break
3967 echo "start migration thread $migrate_pid"
3968 #Access the files at the same time
3969 start_time=$(date +%s)
3970 echo "accessing the migrating directory for 5 minutes..."
3972 ls $migrate_dir2 > /dev/null || {
3973 echo "read dir fails"
3976 diff -u $DIR2/$tdir/file1 $migrate_dir2/file1 || {
3977 echo "access file1 fails"
3981 cat $migrate_dir2/file2 > $migrate_dir2/file3 || {
3982 echo "access file2/3 fails"
3986 echo "aaaaa" > $migrate_dir2/file4 > /dev/null || {
3987 echo "access file4 fails"
3991 stat $migrate_dir2/file5 > /dev/null || {
3992 echo "stat file5 fails"
3996 touch $migrate_dir2/source_file > /dev/null || rc1=$?
3997 [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
3998 echo "touch file failed with $rc1"
4002 if [ -e $migrate_dir2/source_file ]; then
4003 ln $migrate_dir2/source_file $migrate_dir2/link_file \
4004 &>/dev/null || rc1=$?
4005 if [ -e $migrate_dir2/link_file ]; then
4006 rm -rf $migrate_dir2/link_file
4009 mrename $migrate_dir2/source_file \
4010 $migrate_dir2/target_file &>/dev/null || rc1=$?
4011 [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
4012 echo "rename failed with $rc1"
4016 if [ -e $migrate_dir2/target_file ]; then
4017 rm -rf $migrate_dir2/target_file &>/dev/null ||
4020 rm -rf $migrate_dir2/source_file &>/dev/null ||
4023 [ $rc1 -ne 0 -o $rc1 -ne 1 ] || {
4024 echo "unlink failed with $rc1"
4029 end_time=$(date +%s)
4030 duration=$((end_time - start_time))
4031 if [ $((duration % 10)) -eq 0 ]; then
4032 if [ $show_time -eq 1 ]; then
4033 echo "...$duration seconds"
4040 kill -0 $migrate_pid || {
4041 echo "migration stopped 1"
4045 [ $duration -ge 300 ] && break
4048 #check migration are still there
4049 kill -0 $migrate_pid || error "migration stopped 2"
4052 run_test 80b "Accessing directory during migration"
4055 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4059 mkdir -p $DIR1/$tdir
4061 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR1/$tdir/d0
4062 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR1/$tdir/d1
4065 touch d0/0 || error "create 0 failed"
4066 mv d0/0 d1/0 || error "rename d0/0 d1/0 failed"
4067 stat d0/0 && error "stat mv filed succeed"
4068 mv $DIR2/$tdir/d1/0 $DIR2/$tdir/d0/0 || error "rename d1/0 d0/0 failed"
4069 stat d0/0 || error "stat failed"
4071 local t=$(ls -ai $DIR1/$tdir/d0 | sort -u | wc -l)
4073 if [ $t -ne 3 ]; then
4074 ls -ai $DIR1/$tdir/d0
4075 error "expect 3 get $t"
4080 run_test 81a "rename and stat under striped directory"
4083 [ $MDSCOUNT -lt 2 ] &&
4084 skip "We need at least 2 MDTs for this test"
4091 $LFS mkdir -c $MDSCOUNT $DIR1/$tdir || error "$LFS mkdir"
4092 createmany -o $DIR1/$tdir/$tfile. $total || error "createmany"
4101 for i in $(seq $total); do
4102 mrename $DIR2/$tdir/$tfile.$i $DIR2/$tdir/$tfile-new.$i \
4106 kill -9 $setattr_pid
4108 run_test 81b "rename under striped directory doesn't deadlock"
4111 [[ $(lustre_version_code $SINGLEMDS) -gt $(version_code 2.6.91) ]] ||
4112 { skip "Need MDS version at least 2.6.92"; return 0; }
4114 # Client 1 creates a file.
4115 multiop_bg_pause $DIR1/$tfile O_ac || error "multiop_bg_pause 1"
4117 # Client 2 opens the file.
4118 multiop_bg_pause $DIR2/$tfile o_Ac || error "multiop_bg_pause 2"
4120 # Client 1 makes the file an orphan.
4121 rm $DIR1/$tfile || error "rm"
4122 # Client 2 sets EA "user.multiop".
4124 wait $pid2 || error "multiop 2"
4125 # Client 1 gets EA "user.multiop". This used to fail because the EA
4126 # cache refill would get "trusted.link" from mdd_xattr_list() but
4127 # -ENOENT when trying to get "trusted.link"'s value. See also sanity
4130 wait $pid1 || error "multiop 1"
4132 run_test 82 "fsetxattr and fgetxattr on orphan files"
4135 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4142 $LFS mkdir -i1 -c2 $tdir
4147 echo "start pid $pid1 to create/unlink striped directory"
4149 # Access the directory at the same time
4153 stat $tdir > /dev/null 2>&1
4157 echo "start pid $pid2 to stat striped directory"
4165 run_test 83 "access striped directory while it is being created/unlinked"
4168 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4173 [ "$SLOW" = "yes" ] && duration=600
4174 # Open/Create under striped directory
4178 $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1
4179 touch $tdir/f{0..3} > /dev/null 2>&1
4183 echo "start pid $pid1 to open/create under striped directory"
4185 # unlink the striped directory at the same time
4189 rm -rf $tdir > /dev/null 2>&1
4193 echo "start pid $pid2 to unlink striped directory"
4202 run_test 90 "open/create and unlink striped directory"
4205 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4210 [ "$SLOW" = "yes" ] && duration=600
4211 # chmod striped directory
4215 $LFS mkdir -c$MDSCOUNT $tdir > /dev/null 2>&1
4216 chmod go+w $tdir > /dev/null 2>&1
4220 echo "start pid $pid1 to chmod striped directory"
4222 # unlink the striped directory at the same time
4226 rm -rf $tdir > /dev/null 2>&1
4230 echo "start pid $pid2 to unlink striped directory"
4239 run_test 91 "chmod and unlink striped directory"
4242 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4245 local cmd="exec $fd<$DIR1/$tdir"
4246 $LFS setdirstripe -c$MDSCOUNT $DIR1/$tdir || error "mkdir $tdir fails"
4249 trap "eval $cmd" EXIT
4250 cd $DIR1/$tdir || error "cd $DIR1/$tdir fails"
4251 rmdir ../$tdir || error "rmdir ../$tdir fails"
4253 #define OBD_FAIL_LLITE_NO_CHECK_DEAD 0x1408
4254 $LCTL set_param fail_loc=0x1408
4255 mkdir $DIR2/$tdir/dir && error "create dir succeeds"
4256 $LFS setdirstripe -i1 $DIR2/$tdir/remote_dir &&
4257 error "create remote dir succeeds"
4258 $LCTL set_param fail_loc=0
4262 run_test 92 "create remote directory under orphan directory"
4269 mkdir -p $DIR1/$tfile-1/
4270 mkdir -p $DIR2/$tfile-2/
4271 local old_rr=$(do_facet $SINGLEMDS lctl get_param -n \
4272 'lod.lustre-MDT*/qos_threshold_rr' | sed -e 's/%//')
4273 do_facet $SINGLEMDS lctl set_param -n \
4274 'lod.lustre-MDT*/qos_threshold_rr' 100
4275 #define OBD_FAIL_MDS_LOV_CREATE_RACE 0x163
4276 do_facet $SINGLEMDS "lctl set_param fail_loc=0x00000163"
4278 $LFS setstripe -c -1 $DIR1/$tfile-1/file1 &
4281 $LFS setstripe -c -1 $DIR2/$tfile-2/file2 &
4285 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
4286 do_facet $SINGLEMDS "lctl set_param -n \
4287 'lod.lustre-MDT*/qos_threshold_rr' $old_rr"
4289 $LFS getstripe $DIR1/$tfile-1/file1
4290 rc1=$($LFS getstripe -q $DIR1/$tfile-1/file1 |
4291 awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l)
4292 $LFS getstripe $DIR2/$tfile-2/file2
4293 rc2=$($LFS getstripe -q $DIR2/$tfile-2/file2 |
4294 awk '{if (/[0-9]/) print $1 }' | sort | uniq -d | wc -l)
4295 echo "rc1=$rc1 and rc2=$rc2 "
4296 [ $rc1 -eq 0 ] && [ $rc2 -eq 0 ] ||
4297 error "object allocate on same ost detected"
4299 run_test 93 "alloc_rr should not allocate on same ost"
4303 skip "Reserved for glimpse-ahead" && return
4304 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
4305 skip "Need MDS version at least 2.10.55" && return
4309 $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
4311 lctl set_param -n mdc.*.stats=clear
4312 dd if=/dev/zero of=$DIR2/$tdir/dom bs=4096 count=1 || return 1
4314 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #1"
4315 # first stat from server should return size data and save glimpse
4316 local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
4317 [ $gls -eq 0 ] || error "Unexpected $gls glimpse RPCs"
4318 # second stat to check size is NOT cached on client without IO lock
4319 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #2"
4321 local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
4322 [ $gls -ge 1 ] || error "Expect glimpse RPCs but none"
4325 run_test 100a "DoM: glimpse RPCs for stat without IO lock (DoM only file)"
4328 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
4329 skip "Need MDS version at least 2.10.55" && return
4333 $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
4335 lctl set_param -n mdc.*.stats=clear
4336 dd if=/dev/zero of=$DIR2/$tdir/dom bs=4096 count=1 || return 1
4337 cancel_lru_locks mdc
4338 # first stat data from server should have size
4339 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #1"
4340 # second stat to check size is cached on client
4341 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom || error "stat #2"
4343 local gls=$(lctl get_param -n mdc.*.stats | grep -c ldlm_glimpse)
4344 # both stats should cause no glimpse requests
4345 [ $gls == 0 ] || error "Unexpected $gls glimpse RPCs"
4348 run_test 100b "DoM: no glimpse RPC for stat with IO lock (DoM only file)"
4351 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
4352 skip "Need MDS version at least 2.10.55" && return
4356 $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
4358 lctl set_param -n mdc.*.stats=clear
4359 lctl set_param -n osc.*.stats=clear
4360 dd if=/dev/zero of=$DIR2/$tdir/dom bs=2048K count=1 || return 1
4362 # check that size is merged from MDT and OST correctly
4363 $CHECKSTAT -t file -s 2097152 $DIR/$tdir/dom ||
4364 error "Wrong size from stat #1"
4366 local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse)
4367 [ $gls -eq 0 ] && error "Expect OST glimpse RPCs but got none"
4371 run_test 100c "DoM: write vs stat without IO lock (combined file)"
4374 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
4375 skip "Need MDS version at least 2.10.55" && return
4379 $LFS setstripe -E 1024K -L mdt -E EOF $DIR/$tdir/dom
4382 dd if=/dev/zero of=$DIR2/$tdir/dom bs=2048K count=1 || return 1
4383 lctl set_param -n mdc.*.stats=clear
4384 $TRUNCATE $DIR2/$tdir/dom 4096
4386 # check that reported size is valid after file grows to OST and
4387 # is truncated back to MDT stripe size
4388 $CHECKSTAT -t file -s 4096 $DIR/$tdir/dom ||
4389 error "Wrong size from stat #1"
4391 local gls=$(lctl get_param -n osc.*.stats | grep -c ldlm_glimpse)
4392 [ $gls -eq 0 ] && error "Expect OST glimpse but got none"
4396 run_test 100d "DoM: write+truncate vs stat without IO lock (combined file)"
4399 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
4400 skip "Need MDS version at least 2.10.55" && return
4402 $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
4404 $CHECKSTAT -t file $DIR1/$tfile
4406 local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
4407 sysctl -wq vm.dirty_writeback_centisecs=0
4409 trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
4412 dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 ||
4413 error_noexit "Write fails"
4414 # must discard pages
4415 lctl set_param -n mdc.*.stats=clear
4416 rm $DIR2/$tfile || error "Unlink fails"
4418 local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
4419 [ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
4421 run_test 101a "Discard DoM data on unlink"
4424 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
4425 skip "Need MDS version at least 2.10.55" && return
4427 $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
4428 touch $DIR1/${tfile}_2
4430 $CHECKSTAT -t file $DIR1/$tfile
4432 local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
4433 sysctl -wq vm.dirty_writeback_centisecs=0
4435 trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
4438 dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails"
4439 # must discard pages
4440 lctl set_param -n mdc.*.stats=clear
4441 mv $DIR2/${tfile}_2 $DIR2/$tfile || error "Rename fails"
4443 local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
4444 [ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
4446 run_test 101b "Discard DoM data on rename"
4449 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.55) ] &&
4450 skip "Need MDS version at least 2.10.55" && return
4452 $LFS setstripe -E 1024K -L mdt -E EOF $DIR1/$tfile
4454 $CHECKSTAT -t file $DIR1/$tfile
4456 local old_wb=$(sysctl -n vm.dirty_writeback_centisecs)
4457 sysctl -wq vm.dirty_writeback_centisecs=0
4459 trap "sysctl -wq vm.dirty_writeback_centisecs=$old_wb" EXIT
4462 dd if=/dev/zero of=$DIR1/$tfile bs=4096 count=1 || error "Write fails"
4463 $MULTIOP $DIR1/$tfile O_c &
4466 lctl set_param -n mdc.*.stats=clear
4467 rm $DIR2/$tfile > /dev/null || error "Unlink fails for opened file"
4468 kill -USR1 $MULTIOP_PID && wait $MULTIOP_PID || error "multiop failure"
4470 local writes=$(lctl get_param -n mdc.*.stats | grep -c ost_write)
4471 [ $writes -eq 0 ] || error "Found WRITE RPC but expect none"
4473 run_test 101c "Discard DoM data on close-unlink"
4475 # test to verify file handle related system calls
4476 # (name_to_handle_at/open_by_handle_at)
4477 # The new system calls are supported in glibc >= 2.14.
4479 # test to verify we can open by handle an unlinked file from > 1 client
4480 # This test opens the file normally on $DIR1, which is on one mount, and then
4481 # opens it by handle on $DIR2, which is on a different mount.
4483 [ $MDS1_VERSION -lt $(version_code 2.11.57) ] &&
4484 skip "Needs MDS version 2.11.57 or later"
4486 echo "Test file_handle syscalls" > $DIR/$tfile ||
4487 error "write failed"
4488 check_fhandle_syscalls $DIR/$tfile $DIR2 ||
4489 error "check_fhandle_syscalls failed"
4492 run_test 102 "Test open by handle of unlinked file"
4494 log "cleanup: ======================================================"
4496 # kill and wait in each test only guarentee script finish, but command in script
4497 # like 'rm' 'chmod' may still be running, wait for all commands to finish
4498 # otherwise umount below will fail
4499 [ "$(mount | grep $MOUNT2)" ] && wait_update $HOSTNAME "fuser -m $MOUNT2" "" ||
4504 check_and_cleanup_lustre