2 # -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
3 # vim:shiftwidth=4:softtabstop=4:tabstop=4:
5 # test e2fsck and lfsck to detect and fix filesystem corruption
10 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
11 . $LUSTRE/tests/test-framework.sh
13 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
15 NUMFILES=${NUMFILES:-10}
17 OSTIDX=${OSTIDX:-0} # the OST index in LOV
18 OBJGRP=${OBJGRP:-0} # the OST object group
20 [ -d "$SHARED_DIRECTORY" ] || \
21 { skip "SHARED_DIRECTORY should be specified with a shared directory \
22 which can be accessable on all of the nodes" && exit 0; }
23 [[ $(facet_fstype $SINGLEMDS) != ldiskfs ]] &&
24 skip "Only applicable to ldiskfs-based MDTs" && exit 0
25 [[ $(facet_fstype OST) != ldiskfs ]] &&
26 skip "Only applicable to ldiskfs-based OST" && exit 0
28 which getfattr &>/dev/null || { skip_env "could not find getfattr" && exit 0; }
29 which setfattr &>/dev/null || { skip_env "could not find setfattr" && exit 0; }
32 check_and_setup_lustre
36 SAMPLE_FILE=$TMP/$TESTSUITE.junk
37 dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=1
39 # Create some dirs and files on the filesystem.
48 for d in $(seq -f d%g $first_num $last_num); do
49 echo "creating files in $test_dir/$d"
50 for e in $(seq -f d%g $num_dirs); do
51 mkdir -p $test_dir/$d/$e || error "mkdir $test_dir/$d/$e failed"
52 for f in $(seq -f test%g $num_dirs); do
53 cp $file_name $test_dir/$d/$e/$f || \
54 error "cp $file_name $test_dir/$d/$e/$f failed"
66 # create some files on the filesystem
68 local last_num=$num_dirs
69 create_files_sub $test_dir $num_dirs /etc/fstab $first_num $last_num
71 # create files to be modified
72 for f in $(seq -f $test_dir/testfile.%g $((num_files * 3))); do
74 cp $SAMPLE_FILE $f || error "cp $SAMPLE_FILE $f failed"
77 # create some more files
78 first_num=$((num_dirs * 2 + 1))
79 last_num=$((num_dirs * 2 + 3))
80 create_files_sub $test_dir $num_dirs /etc/hosts $first_num $last_num
82 # these should NOT be taken as duplicates
83 for f in $(seq -f $test_dir/d$last_num/linkfile.%g $num_files); do
84 echo "linking files in $test_dir/d$last_num"
85 cp /etc/hosts $f || error "cp /etc/hosts $f failed"
86 ln $f $f.link || error "ln $f $f.link failed"
90 # Get the objids for files on the OST (given the OST index and object group).
98 ost_objids=$($LFS getstripe $ost_files | \
99 awk '{if ($1 == '$obdidx' && $4 == '$group') print $2 }')
103 # Get the OST nodet name (given the OST index).
110 ost_uuid=$(ostuuid_from_index $obdidx)
112 for node in $(osts_nodes); do
113 do_node $node "lctl get_param -n obdfilter.*.uuid" | grep -q $ost_uuid
114 [ ${PIPESTATUS[1]} -eq 0 ] && ost_node=$node && break
116 [ -z "$ost_node" ] && \
117 echo "failed to find the OST with index $obdidx" && return 1
121 # Get the OST target device (given the OST facet name and OST index).
128 ost_name=$(ostname_from_index $obdidx)
129 ost_dev=$(get_osd_param $node $ost_name mntdev)
130 if [ $? -ne 0 ]; then
131 printf "unable to find OST%04x on $facet\n" $obdidx
135 if [[ $ost_dev = *loop* ]]; then
136 ost_dev=$(do_node $node "losetup $ost_dev" | \
137 sed -e "s/.*(//" -e "s/).*//")
143 # Get the file names to be duplicated or removed on the MDS.
153 first=$((num_files + 1))
154 last=$((num_files * 2))
157 first=$((num_files * 2 + 1))
158 last=$((num_files * 3))
160 *) echo "get_files(): invalid flavor" && return 1 ;;
165 for f in $(seq -f testfile.%g $first $last); do
166 test_file=$test_dir/$f
167 files="$files $test_file"
169 files=$(echo $files | sed "s#$DIR/##g")
173 # Remove objects associated with files.
180 local facet=ost$((OSTIDX + 1))
181 local mntpt=$(facet_mntpt $facet)
182 local opts=$OST_MOUNT_OPTS
186 echo "removing objects from $ostdev on $facet: $objids"
187 if ! do_facet $facet test -b $ostdev; then
188 opts=$(csa_add "$opts" -o loop)
190 mount -t $(facet_fstype $facet) $opts $ostdev $mntpt ||
194 rm $mntpt/O/$group/d$((i % 32))/$i || { rc=$?; break; }
196 umount -f $mntpt || return $?
200 # Remove files from MDS.
202 do_rpc_nodes $(facet_host $1) remove_mdt_files $@
205 # Create EAs on files so objects are referenced from different files.
207 do_rpc_nodes $(facet_host $1) duplicate_mdt_files $@
210 #********************************* Main Flow **********************************#
214 # get the server target devices
217 TESTDIR=$DIR/d0.$TESTSUITE
218 if is_empty_fs $MOUNT; then
219 # create test directory
220 mkdir -p $TESTDIR || error "mkdir $TESTDIR failed"
222 # create some dirs and files on the filesystem
223 create_files $TESTDIR $NUMDIRS $NUMFILES
225 # get the objids for files in group $OBJGRP on the OST with index $OSTIDX
226 OST_REMOVE=$(get_objects $OSTIDX $OBJGRP \
227 $(seq -f $TESTDIR/testfile.%g $NUMFILES))
229 # get the node name and target device for the OST with index $OSTIDX
230 OSTNODE=$(get_ost_node $OSTIDX) || error "get_ost_node by index $OSTIDX failed"
231 OSTDEV=$(get_ost_dev $OSTNODE $OSTIDX) ||
232 error "get_ost_dev $OSTNODE $OSTIDX failed"
234 # get the file names to be duplicated on the MDS
235 MDS_DUPE=$(get_files dup $TESTDIR $NUMFILES) || error "$MDS_DUPE"
236 # get the file names to be removed from the MDS
237 MDS_REMOVE=$(get_files remove $TESTDIR $NUMFILES) || error "$MDS_REMOVE"
239 stopall -f || error "cleanupall failed"
241 # remove objects associated with files in group $OBJGRP
242 # on the OST with index $OSTIDX
243 remove_objects $OSTDEV $OBJGRP $OST_REMOVE ||
244 error "removing objects failed"
246 # remove files from MDS
247 remove_files $SINGLEMDS $MDTDEV $MDS_REMOVE || error "removing files failed"
249 # create EAs on files so objects are referenced from different files
250 duplicate_files $SINGLEMDS $MDTDEV $MDS_DUPE ||
251 error "duplicating files failed"
252 FSCK_MAX_ERR=1 # file system errors corrected
253 else # is_empty_fs $MOUNT
254 FSCK_MAX_ERR=4 # file system errors left uncorrected
257 # Test 1a - check and repair the filesystem
258 # lfsck will return 1 if the filesystem had errors fixed
259 # run e2fsck to generate databases used for lfsck
263 ORIG_REFORMAT=$REFORMAT
265 check_and_setup_lustre
266 REFORMAT=$ORIG_REFORMAT
271 if [ $rc -eq 0 ]; then
272 echo "clean after the first check"
274 # run e2fsck again to generate databases used for lfsck
280 if [ $rc -eq 0 ]; then
281 echo "clean after the second check"
283 error "lfsck test 2 - finished with rc=$rc"
288 # The test directory contains some files referencing to some object
289 # which could cause error when removing the directory.
291 while [ -d $TESTDIR ]; do
293 rm -fr $TESTDIR || echo "$RMCNT round: rm $TESTDIR failed"
294 [ $RMCNT -ge 10 ] && error "cleanup $TESTDIR failed $RMCNT times"
295 remount_client $MOUNT
297 check_and_cleanup_lustre