LU-988 clio: use OSC object's m/a/ctime when build write RPC

[fs/lustre-release.git] / lustre / tests / sanity.sh

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index c9bbf5a..36e2aa0 100644 (file)
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -16,12 +16,6 @@ CPU=`awk '/model/ {print $4}' /proc/cpuinfo`
 #                                    buffer i/o errs             sock spc runas
 [ "$CPU" = "UML" ] && EXCEPT="$EXCEPT 27m 27n 27o 27p 27q 27r 31d 54a  64b 99a 99b 99c 99d 99e 99f 101a"
 
-case `uname -r` in
-2.4*) FSTYPE=${FSTYPE:-ext3} ;;
-2.6*) FSTYPE=${FSTYPE:-ldiskfs} ;;
-*) error "unsupported kernel" ;;
-esac
-
 SRCDIR=$(cd $(dirname $0); echo $PWD)
 export PATH=$PATH:/sbin
 
@@ -36,6 +30,7 @@ LCTL=${LCTL:-lctl}
 MCREATE=${MCREATE:-mcreate}
 OPENFILE=${OPENFILE:-openfile}
 OPENUNLINK=${OPENUNLINK:-openunlink}
+export MULTIOP=${MULTIOP:-multiop}
 READS=${READS:-"reads"}
 MUNLINK=${MUNLINK:-munlink}
 SOCKETSERVER=${SOCKETSERVER:-socketserver}
@@ -691,13 +686,13 @@ run_test 24k "touch .../R11a/f; mv .../R11a/f .../R11a/d ======="
 # bug 2429 - rename foo foo foo creates invalid file
 test_24l() {
        f="$DIR/f24l"
-       multiop $f OcNs || error
+       $MULTIOP $f OcNs || error
 }
 run_test 24l "Renaming a file to itself ========================"
 
 test_24m() {
        f="$DIR/f24m"
-       multiop $f OcLN ${f}2 ${f}2 || error "link ${f}2 ${f}2 failed"
+       $MULTIOP $f OcLN ${f}2 ${f}2 || error "link ${f}2 ${f}2 failed"
        # on ext3 this does not remove either the source or target files
        # though the "expected" operation would be to remove the source
        $CHECKSTAT -t file ${f} || error "${f} missing"
@@ -774,7 +769,7 @@ test_24t() {
 run_test 24t "mkdir .../R16a/b/c; rename .../R16a/b/c .../R16a ="
 
 test_24u() { # bug12192
-        multiop $DIR/$tfile C2w$((2048 * 1024))c || error
+        $MULTIOP $DIR/$tfile C2w$((2048 * 1024))c || error
         $CHECKSTAT -s $((2048 * 1024)) $DIR/$tfile || error "wrong file size"
 }
 run_test 24u "create stripe file"
@@ -1356,9 +1351,6 @@ check_seq_oid()
         # compare lmm_object_id and lu_fid->oid
         [ $lmm_oid = ${fid[2]} ] || { error "OID mismatch"; return 2; }
 
-        [ "$FSTYPE" != "ldiskfs" ] &&
-                skip "cannot check filter fid FSTYPE=$FSTYPE" && return 0
-
         # check the trusted.fid attribute of the OST objects of the file
         local have_obdidx=false
         local stripe_nr=0
@@ -1371,6 +1363,11 @@ check_seq_oid()
                 local ost=$((obdidx + 1))
                 local dev=$(ostdevname $ost)
 
+               if [ $(facet_fstype ost$ost) != ldiskfs ]; then
+                       echo "Currently only works with ldiskfs-based OSTs"
+                       continue
+               fi
+
                 log "want: stripe:$stripe_nr ost:$obdidx oid:$oid/$hex seq:$seq"
 
                 #don't unmount/remount the OSTs if we don't need to do that
@@ -1543,7 +1540,7 @@ run_test 31a "open-unlink file =================================="
 test_31b() {
        touch $DIR/f31 || error
        ln $DIR/f31 $DIR/f31b || error
-       multiop $DIR/f31b Ouc || error
+       $MULTIOP $DIR/f31b Ouc || error
        $CHECKSTAT -t file $DIR/f31 || error
 }
 run_test 31b "unlink file with multiple links while open ======="
@@ -1553,7 +1550,7 @@ test_31c() {
        ln $DIR/f31 $DIR/f31c || error
        multiop_bg_pause $DIR/f31 O_uc || return 1
        MULTIPID=$!
-       multiop $DIR/f31c Ouc
+       $MULTIOP $DIR/f31c Ouc
        kill -USR1 $MULTIPID
        wait $MULTIPID
 }
@@ -2064,7 +2061,7 @@ test_34h() {
        local sz=1000
 
        dd if=/dev/zero of=$DIR/$tfile bs=1M count=10 || error
-       multiop $DIR/$tfile OG${gid}T${sz}g${gid}c &
+       $MULTIOP $DIR/$tfile OG${gid}T${sz}g${gid}c &
        MULTIPID=$!
        sleep 2
 
@@ -2425,12 +2422,14 @@ test_39i() {
 run_test 39i "write, rename, stat =============================="
 
 test_39j() {
-       debugsave
-       lctl set_param debug=-1
+       start_full_debug_logging
        touch $DIR1/$tfile
        sleep 1
 
-       multiop_bg_pause $DIR1/$tfile oO_RDWR:w2097152_c || error "multiop failed"
+       #define OBD_FAIL_OSC_DELAY_SETTIME       0x412
+       lctl set_param fail_loc=0x80000412
+       multiop_bg_pause $DIR1/$tfile oO_RDWR:w2097152_c ||
+               error "multiop failed"
        local multipid=$!
        local mtime1=`stat -c %Y $DIR1/$tfile`
 
@@ -2441,13 +2440,15 @@ test_39j() {
 
        for (( i=0; i < 2; i++ )) ; do
                local mtime2=`stat -c %Y $DIR1/$tfile-1`
-               [ "$mtime1" = "$mtime2" ] || \
-                       error "mtime is lost on close: $mtime2, should be $mtime1"
+               [ "$mtime1" = "$mtime2" ] ||
+                       error "mtime is lost on close: $mtime2, " \
+                             "should be $mtime1"
 
                cancel_lru_locks osc
                if [ $i = 0 ] ; then echo "repeat after cancel_lru_locks"; fi
        done
-       debugrestore
+       lctl set_param fail_loc=0
+       stop_full_debug_logging
 }
 run_test 39j "write, rename, close, stat ======================="
 
@@ -2782,7 +2783,7 @@ run_test 42e "verify sub-RPC writes are not done synchronously"
 test_43() {
        mkdir -p $DIR/$tdir
        cp -p /bin/ls $DIR/$tdir/$tfile
-       multiop $DIR/$tdir/$tfile Ow_c &
+       $MULTIOP $DIR/$tdir/$tfile Ow_c &
        pid=$!
        # give multiop a chance to open
        sleep 1
@@ -2794,10 +2795,10 @@ run_test 43 "execution of file opened for write should return -ETXTBSY"
 
 test_43a() {
         mkdir -p $DIR/d43
-       cp -p `which multiop` $DIR/d43/multiop || cp -p multiop $DIR/d43/multiop
+       cp -p `which $MULTIOP` $DIR/d43/multiop || cp -p multiop $DIR/d43/multiop
         MULTIOP_PROG=$DIR/d43/multiop multiop_bg_pause $TMP/test43.junk O_c || return 1
         MULTIOP_PID=$!
-        multiop $DIR/d43/multiop Oc && error "expected error, got success"
+        $MULTIOP $DIR/d43/multiop Oc && error "expected error, got success"
         kill -USR1 $MULTIOP_PID || return 2
         wait $MULTIOP_PID || return 3
         rm $TMP/test43.junk
@@ -2806,7 +2807,7 @@ run_test 43a "open(RDWR) of file being executed should return -ETXTBSY"
 
 test_43b() {
         mkdir -p $DIR/d43
-       cp -p `which multiop` $DIR/d43/multiop || cp -p multiop $DIR/d43/multiop
+       cp -p `which $MULTIOP` $DIR/d43/multiop || cp -p multiop $DIR/d43/multiop
         MULTIOP_PROG=$DIR/d43/multiop multiop_bg_pause $TMP/test43.junk O_c || return 1
         MULTIOP_PID=$!
         $TRUNCATE $DIR/d43/multiop 0 && error "expected error, got success"
@@ -3031,6 +3032,30 @@ test_48e() { # bug 4134
 }
 run_test 48e "Access to recreated parent subdir (should return errors)"
 
+test_49() { # LU-1030
+       # get ost1 size - lustre-OST0000
+       ost1_size=$(do_facet ost1 lfs df |grep ${ost1_svc} |awk '{print $4}')
+       # write 800M at maximum
+       [ $ost1_size -gt 819200 ] && ost1_size=819200
+
+       lfs setstripe -c 1 -i 0 $DIR/$tfile
+       dd if=/dev/zero of=$DIR/$tfile bs=4k count=$((ost1_size >> 2)) &
+       local dd_pid=$!
+
+       # change max_pages_per_rpc while writing the file
+       local osc1_mppc=osc.$(get_osc_import_name client ost1).max_pages_per_rpc
+       local orig_mppc=`$LCTL get_param -n $osc1_mppc`
+       # loop until dd process exits
+       while ps ax -opid | grep -q $dd_pid; do
+               $LCTL set_param $osc1_mppc=$((RANDOM % 256 + 1))
+               sleep $((RANDOM % 5 + 1))
+       done
+       # restore original max_pages_per_rpc
+       $LCTL set_param $osc1_mppc=$orig_mppc
+       rm $DIR/$tfile || error "rm $DIR/$tfile failed"
+}
+run_test 49 "Change max_pages_per_rpc won't break osc extent"
+
 test_50() {
        # bug 1485
        mkdir $DIR/d50
@@ -4021,7 +4046,7 @@ test_61() {
        f="$DIR/f61"
        dd if=/dev/zero of=$f bs=`page_size` count=1
        cancel_lru_locks osc
-       multiop $f OSMWUc || error
+       $MULTIOP $f OSMWUc || error
        sync
 }
 run_test 61 "mmap() writes don't make sync hang ================"
@@ -4067,7 +4092,7 @@ test_63b() {
 
        #define OBD_FAIL_OSC_BRW_PREP_REQ        0x406
        lctl set_param fail_loc=0x80000406
-       multiop $DIR/$tfile Owy && \
+       $MULTIOP $DIR/$tfile Owy && \
                error "sync didn't return ENOMEM"
        sync; sleep 2; sync     # do a real sync this time to flush page
        lctl get_param -n llite.*.dump_page_cache | grep locked && \
@@ -4437,11 +4462,11 @@ test_73() {
        pid1=$!
 
        lctl set_param fail_loc=0x80000129
-       multiop $DIR/d73-1/f73-2 Oc &
+       $MULTIOP $DIR/d73-1/f73-2 Oc &
        sleep 1
        lctl set_param fail_loc=0
 
-       multiop $DIR/d73-2/f73-3 Oc &
+       $MULTIOP $DIR/d73-2/f73-3 Oc &
        pid3=$!
 
        kill -USR1 $pid1
@@ -4696,7 +4721,7 @@ test_77i() { # bug 13805
        for VALUE in `lctl get_param osc.*osc-[^mM]*.checksum_type`; do
                PARAM=`echo ${VALUE[0]} | cut -d "=" -f1`
                algo=`lctl get_param -n $PARAM | sed 's/.*\[\(.*\)\].*/\1/g'`
-               [ "$algo" = "crc32" ] || error "algo set to $algo instead of crc32"
+               [ "$algo" = "adler" ] || error "algo set to $algo instead of adler"
        done
        remount_client $MOUNT
 }
@@ -4817,7 +4842,7 @@ test_81a() { # LU-456
 
         # write should trigger a retry and success
         $SETSTRIPE -i 0 -c 1 $DIR/$tfile
-        multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+        $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
         RC=$?
         if [ $RC -ne 0 ] ; then
                 error "write should success, but failed for $RC"
@@ -4833,7 +4858,7 @@ test_81b() { # LU-456
 
         # write should retry several times and return -ENOSPC finally
         $SETSTRIPE -i 0 -c 1 $DIR/$tfile
-        multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+        $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
         RC=$?
         ENOSPC=28
         if [ $RC -ne $ENOSPC ] ; then
@@ -4842,6 +4867,27 @@ test_81b() { # LU-456
 }
 run_test 81b "OST should return -ENOSPC when retry still fails ======="
 
+test_82() { # LU-1031
+       dd if=/dev/zero of=$DIR/$tfile bs=1M count=10
+       local gid1=14091995
+       local gid2=16022000
+
+       multiop_bg_pause $DIR/$tfile OG${gid1}_g${gid1}c || return 1
+       local MULTIPID1=$!
+       multiop_bg_pause $DIR/$tfile O_G${gid2}r10g${gid2}c || return 2
+       local MULTIPID2=$!
+       kill -USR1 $MULTIPID2
+       sleep 2
+       if [[ `ps h -o comm -p $MULTIPID2` == "" ]]; then
+               error "First grouplock does not block second one"
+       else
+               echo "Second grouplock blocks first one"
+       fi
+       kill -USR1 $MULTIPID1
+       wait $MULTIPID1
+       wait $MULTIPID2
+}
+run_test 82 "Basic grouplock test ==============================="
 
 test_99a() {
         [ -z "$(which cvs 2>/dev/null)" ] && skip_env "could not find cvs" && \
@@ -5903,7 +5949,7 @@ test_118a() #bug 11710
 {
        reset_async
 
-       multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+       $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
        DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
         WRITEBACK=$(lctl get_param -n llite.*.dump_page_cache | grep -c writeback)
 
@@ -5923,7 +5969,7 @@ test_118b()
 
        #define OBD_FAIL_OST_ENOENT 0x217
        set_nodes_failloc "$(osts_nodes)" 0x217
-       multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+       $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
        RC=$?
        set_nodes_failloc "$(osts_nodes)" 0
         DIRTY=$(lctl get_param -n llite.*.dump_page_cache | grep -c dirty)
@@ -5944,7 +5990,7 @@ test_118b()
 
        # Due to the above error the OSC will issue all RPCs syncronously
        # until a subsequent RPC completes successfully without error.
-       multiop $DIR/$tfile Ow4096yc
+       $MULTIOP $DIR/$tfile Ow4096yc
        rm -f $DIR/$tfile
 
        return 0
@@ -5961,7 +6007,7 @@ test_118c()
        set_nodes_failloc "$(osts_nodes)" 0x216
 
        # multiop should block due to fsync until pages are written
-       multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
+       $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
        MULTIPID=$!
        sleep 1
 
@@ -6004,7 +6050,7 @@ test_118d()
        #define OBD_FAIL_OST_BRW_PAUSE_BULK
        set_nodes_failloc "$(osts_nodes)" 0x214
        # multiop should block due to fsync until pages are written
-       multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
+       $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
        MULTIPID=$!
        sleep 1
 
@@ -6041,7 +6087,7 @@ test_118f() {
         lctl set_param fail_loc=0x8000040a
 
        # Should simulate EINVAL error which is fatal
-        multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+        $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
         RC=$?
        if [[ $RC -eq 0 ]]; then
                error "Must return error due to dropped pages, rc=$RC"
@@ -6076,7 +6122,7 @@ test_118g() {
        lctl set_param fail_loc=0x406
 
        # simulate local -ENOMEM
-       multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+       $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
        RC=$?
 
        lctl set_param fail_loc=0
@@ -6112,7 +6158,7 @@ test_118h() {
        #define OBD_FAIL_OST_BRW_WRITE_BULK      0x20e
         set_nodes_failloc "$(osts_nodes)" 0x20e
        # Should simulate ENOMEM error which is recoverable and should be handled by timeout
-        multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+        $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
         RC=$?
 
         set_nodes_failloc "$(osts_nodes)" 0
@@ -6148,7 +6194,7 @@ test_118i() {
         set_nodes_failloc "$(osts_nodes)" 0x20e
 
        # Should simulate ENOMEM error which is recoverable and should be handled by timeout
-        multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
+        $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c &
        PID=$!
        sleep 5
        set_nodes_failloc "$(osts_nodes)" 0
@@ -6186,7 +6232,7 @@ test_118j() {
         set_nodes_failloc "$(osts_nodes)" 0x220
 
        # return -EIO from OST
-        multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
+        $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
         RC=$?
         set_nodes_failloc "$(osts_nodes)" 0x0
        if [[ $RC -eq 0 ]]; then
@@ -6238,7 +6284,7 @@ test_118l()
 {
        # LU-646
        mkdir -p $DIR/$tdir
-       multiop $DIR/$tdir Dy || error "fsync dir failed"
+       $MULTIOP $DIR/$tdir Dy || error "fsync dir failed"
        rm -rf $DIR/$tdir
 }
 run_test 118l "fsync dir ========="
@@ -6268,7 +6314,7 @@ test_119b() # bug 11737
         $SETSTRIPE -c 2 $DIR/$tfile || error "setstripe failed"
         dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 seek=1 || error "dd failed"
         sync
-        multiop $DIR/$tfile oO_RDONLY:O_DIRECT:r$((2048 * 1024)) || \
+        $MULTIOP $DIR/$tfile oO_RDONLY:O_DIRECT:r$((2048 * 1024)) || \
                 error "direct read failed"
         rm -f $DIR/$tfile
 }
@@ -6893,7 +6939,10 @@ set_dir_limits () {
        done
 }
 test_129() {
-       [ "$FSTYPE" != "ldiskfs" ] && skip "not needed for FSTYPE=$FSTYPE" && return 0
+       if [ "$(facet_type_fstype MDS)" != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
        remote_mds_nodsh && skip "remote MDS with nodsh" && return
 
        EFBIG=27
@@ -6906,7 +6955,7 @@ test_129() {
        I=0
        J=0
        while [ ! $I -gt $((MAX * MDSCOUNT)) ]; do
-               multiop $DIR/$tdir/$J Oc
+               $MULTIOP $DIR/$tdir/$J Oc
                rc=$?
                if [ $rc -eq $EFBIG ]; then
                        set_dir_limits 0
@@ -7673,7 +7722,7 @@ test_152() {
 run_test 152 "test read/write with enomem ============================"
 
 test_153() {
-        multiop $DIR/$tfile Ow4096Ycu || error "multiop failed"
+        $MULTIOP $DIR/$tfile Ow4096Ycu || error "multiop failed"
 }
 run_test 153 "test if fdatasync does not crash ======================="
 
@@ -8226,11 +8275,11 @@ run_test 163 "kernel <-> userspace comms"
 test_169() {
        # do directio so as not to populate the page cache
        log "creating a 10 Mb file"
-       multiop $DIR/$tfile oO_CREAT:O_DIRECT:O_RDWR:w$((10*1048576))c || error "multiop failed while creating a file"
+       $MULTIOP $DIR/$tfile oO_CREAT:O_DIRECT:O_RDWR:w$((10*1048576))c || error "multiop failed while creating a file"
        log "starting reads"
        dd if=$DIR/$tfile of=/dev/null bs=4096 &
        log "truncating the file"
-       multiop $DIR/$tfile oO_TRUNC:c || error "multiop failed while truncating the file"
+       $MULTIOP $DIR/$tfile oO_TRUNC:c || error "multiop failed while truncating the file"
        log "killing dd"
        kill %+ || true # reads might have finished
        echo "wait until dd is finished"
@@ -8416,6 +8465,32 @@ test_181() { # bug 22177
 }
 run_test 181 "Test open-unlinked dir ========================"
 
+test_182() {
+       # disable MDC RPC lock wouldn't crash client
+       local fcount=1000
+       local tcount=4
+
+       mkdir -p $DIR/$tdir || error "creating dir $DIR/$tdir"
+#define OBD_FAIL_MDC_RPCS_SEM          0x804
+       $LCTL set_param fail_loc=0x804
+
+       for (( i=0; i < $tcount; i++ )) ; do
+               mkdir $DIR/$tdir/$i
+               createmany -o $DIR/$tdir/$i/f- $fcount &
+       done
+       wait
+
+       for (( i=0; i < $tcount; i++ )) ; do
+               unlinkmany $DIR/$tdir/$i/f- $fcount &
+       done
+       wait
+
+       rm -rf $DIR/$tdir
+
+       $LCTL set_param fail_loc=0
+}
+run_test 182 "Disable MDC RPCs semaphore wouldn't crash client ================"
+
 # OST pools tests
 POOL=${POOL:-cea1}
 TGT_COUNT=$OSTCOUNT
@@ -9032,11 +9107,11 @@ run_test 217 "check lctl ping for hostnames with hiphen ('-')"
 test_218() {
        # do directio so as not to populate the page cache
        log "creating a 10 Mb file"
-       multiop $DIR/$tfile oO_CREAT:O_DIRECT:O_RDWR:w$((10*1048576))c || error "multiop failed while creating a file"
+       $MULTIOP $DIR/$tfile oO_CREAT:O_DIRECT:O_RDWR:w$((10*1048576))c || error "multiop failed while creating a file"
        log "starting reads"
        dd if=$DIR/$tfile of=/dev/null bs=4096 &
        log "truncating the file"
-       multiop $DIR/$tfile oO_TRUNC:c || error "multiop failed while truncating the file"
+       $MULTIOP $DIR/$tfile oO_TRUNC:c || error "multiop failed while truncating the file"
        log "killing dd"
        kill %+ || true # reads might have finished
        echo "wait until dd is finished"
@@ -9285,6 +9360,18 @@ test_226 () {
 }
 run_test 226 "call path2fid and fid2path on files of all type"
 
+# LU-1299 Executing or running ldd on a truncated executable does not
+# cause an out-of-memory condition.
+test_227() {
+       dd if=`which date` of=$MOUNT/date bs=1k count=1
+       chmod +x $MOUNT/date
+
+       $MOUNT/date > /dev/null
+       ldd $MOUNT/date > /dev/null
+       rm -f $MOUNT/date
+}
+run_test 227 "running truncated executable does not cause OOM"
+
 #
 # tests that do cleanup/setup should be run at the end
 #