LU-946 lprocfs: List open files in filesystem

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

diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh
index 63321ae..142c870 100644 (file)
--- a/lustre/tests/sanityn.sh
+++ b/lustre/tests/sanityn.sh
@@ -40,9 +40,12 @@ init_test_env $@
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 init_logging
 
-[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
+if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then
 # bug number for skipped test:        LU-2840 LU-2189 LU-2776
        ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21      36      51a"
+# LU-2829 / LU-2887 - make allowances for ZFS slowness
+       TEST33_NFILES=${TEST33_NFILES:-1000}
+fi
 
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="33a"
 
@@ -489,7 +492,8 @@ test_21() { # Bug 5907
 run_test 21 " Try to remove mountpoint on another dir ===="
 
 test_23() { # Bug 5972
-       local at_diff=$(do_facet $SINGLEMDS $LCTL get_param -n mdd.*.atime_diff)
+       local at_diff=$(do_facet $SINGLEMDS \
+               $LCTL get_param -n mdd.*MDT0000*.atime_diff | head -1)
        echo "atime should be updated while another read" > $DIR1/$tfile
 
        # clear the lock(mode: LCK_PW) gotten from creating operation
@@ -2479,6 +2483,8 @@ run_test 71 "correct file map just after write operation is finished"
 
 test_72() {
        local p="$TMP/sanityN-$TESTNAME.parameters"
+       local tlink1
+       local tlink2
        save_lustre_params client "llite.*.xattr_cache" > $p
        lctl set_param llite.*.xattr_cache 1 ||
                { skip "xattr cache is not supported"; return 0; }
@@ -2492,6 +2498,14 @@ test_72() {
                error "setfattr2 failed"
        getfattr -n user.attr1 $DIR1/$tfile | grep value2 ||
                error "getfattr2 failed"
+
+       # check that trusted.link is consistent
+       tlink1=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
+       ln $DIR2/$tfile $DIR2/$tfile-2 || error "failed to link"
+       tlink2=$(getfattr -n trusted.link $DIR1/$tfile | md5sum)
+       echo "$tlink1 $tlink2"
+       [ "$tlink1" = "$tlink2" ] && error "trusted.link should have changed!"
+
        rm -f $DIR2/$tfile
 
        restore_lustre_params < $p
@@ -2526,12 +2540,113 @@ test_73() {
 run_test 73 "getxattr should not cause xattr lock cancellation"
 
 test_74() {
+       [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.93) ] &&
+               skip "Need MDS version at least 2.4.93" && return
+
        dd if=/dev/zero of=$DIR1/$tfile-1 bs=1K count=1
        dd if=/dev/zero of=$DIR1/$tfile-2 bs=1K count=1
        flocks_test 4 $DIR1/$tfile-1 $DIR2/$tfile-2
 }
 run_test 74 "flock deadlock: different mounts =============="
 
+# LU-3889
+test_75() {
+       $LFS setstripe -c 2 -S 1m -i 0 $DIR1/$tfile
+       dd if=/dev/zero of=$DIR1/$tfile bs=1M count=2
+       cancel_lru_locks osc
+
+       dd of=$DIR1/$tfile if=/dev/zero bs=1M count=1 seek=1 conv=notrunc
+       sync
+
+       # define OBD_FAIL_LDLM_ENQUEUE_HANG 0x31d
+       $LCTL set_param fail_loc=0x31d
+       stat -c %s $DIR1/$tfile &
+       local pid=$!
+       sleep 1
+       kill -9 $pid
+
+       # For bad lock error handler we should ASSERT and got kernel panic here
+       sleep 4
+       $LCTL set_param fail_loc=0
+}
+run_test 75 "osc: upcall after unuse lock==================="
+
+test_76() { #LU-946
+       [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.53) ]] &&
+               skip "Need MDS version at least 2.5.53" && return
+
+       remote_mds_nodsh && skip "remote MDS with nodsh" && return
+       local fcount=2048
+       local fd
+       local cmd
+       local mdt_idx
+       local mds_idx
+       declare -a fd_list
+       declare -a fid_list
+
+       if remote_mds; then
+               nid=$($LCTL list_nids | sed  "s/\./\\\./g")
+       else
+               nid="0@lo"
+       fi
+
+       rm -rf $DIR/$tdir
+       test_mkdir -p $DIR/$tdir
+       if [ $MDSCOUNT -gt 1 ]; then
+               mdt_idx=$($LFS getdirstripe -i $DIR/$tdir)
+       else
+               mdt_idx=0
+       fi
+       mds_idx=$((mdt_idx + 1))
+       proc_ofile="mdt.*$mdt_idx.exports.'$nid'.open_files"
+
+       cancel_lru_locks mdc
+
+       echo -n "open files "
+       ulimit -n 8096
+       for (( i = 0; i < $fcount; i++ )) ; do
+               touch $DIR/$tdir/f_$i
+               fd=$(free_fd)
+               cmd="exec $fd<$DIR/$tdir/f_$i"
+               eval $cmd
+               fd_list[i]=$fd
+               echo -n "."
+       done
+       echo
+
+       fid_list=($(do_facet mds$mds_idx $LCTL get_param -n $proc_ofile))
+
+       # Possible errors in openfiles FID list.
+       # 1. Missing FIDs. Check 1
+       # 2. Extra FIDs. Check 1
+       # 3. Duplicated FID. Check 2
+       # 4. Invalid FIDs. Check 2
+       # 5. Valid FID, points to some other file. Check 3
+
+       # Check 1
+       [ ${#fid_list[@]} -ne $fcount ] &&
+               error "${#fid_list[@]} != $fcount open files"
+
+       for (( i = 0; i < $fcount; i++ )) ; do
+               cmd="exec ${fd_list[i]}</dev/null"
+               eval $cmd
+               filename=$($LFS fid2path $DIR2 ${fid_list[i]})
+
+               # Check 2
+               rm --interactive=no $filename
+               [ $? -ne 0 ] &&
+                       error "Nonexisting fid ${fid_list[i]} listed."
+       done
+
+       # Check 3
+       ls_op=$(ls $DIR2/$tdir | wc -l)
+       [ $ls_op -ne 0 ] &&
+               error "Some openfiles are missing in lproc output"
+
+       rm -rf $DIR/$tdir
+}
+run_test 76 "Verify open file for 2048 files"
+
 log "cleanup: ======================================================"
 
 [ "$(mount | grep $MOUNT2)" ] && umount $MOUNT2