3 # lfs_migrate: a simple tool to copy and check files.
5 # To avoid allocating objects on one or more OSTs, they should be
6 # deactivated on the MDS via "lctl --device {device_number} deactivate",
7 # where {device_number} is from the output of "lctl dl" on the MDS.
9 # To guard against corruption, the file is compared after migration
10 # to verify the copy is correct and the file has not been modified.
11 # This is not a protection against the file being open by another
12 # process, but it would catch the worst cases of in-use files, but
13 # to be 100% safe the administrator needs to ensure this is safe.
16 OPT_RSYNC=${LFS_MIGRATE_RSYNC_MODE:-false}
19 RSYNC_WITH_HLINKS=false
20 LFS_MIGRATE_TMP=${TMPDIR:-/tmp}
21 MIGRATED_SET="$(mktemp ${LFS_MIGRATE_TMP}/lfs_migrate.links.XXXXXX)"
23 REMOVE_FID='s/^\[[0-9a-fx:]*\] //'
30 echo "$old_fid $path" >> "$MIGRATED_SET"
36 sed -e "$REMOVE_FID" $MIGRATED_SET | grep -q "^$path$"
42 grep "^\\$old_fid" "$MIGRATED_SET" | head -n 1 |
48 usage: lfs_migrate [--dry-run|-n] [--help|-h] [--no-rsync|--rsync] [--quiet|-q]
49 [--auto-stripe|-A [-C <cap>]
50 [--min-free|-M <min_free>] [--max-free|-X <max_free>]]
51 [--pool|-p <pool>] [--stripe-count|-c <stripe_count>]
52 [--stripe-size|-S <stripe_size>]
54 [--restripe|-R] [--skip|-s] [--verbose|-v] [--yes|-y] [-0]
56 -A restripe file using an automatically selected stripe count,
57 uses stripe_count = sqrt(size_in_GB) + 1
59 restripe file using the specified <stripe_count>
60 -C <cap> when -A is set, limit the migrated file to use on each OST
61 at most 1/<cap> of the available space of the smallest OST
62 -D do not use direct I/O to copy file contents
63 -h show this usage message
65 when -A is set, an OST must contain more available space than
66 <min_free> KB in order for it to be considered available for
68 --no-rsync do not fall back to rsync mode even if lfs migrate fails
69 -n only print the names of files to be migrated
70 -p <pool> use the specified OST pool for the destination file
71 -q run quietly (don't print filenames or status)
72 --rsync force rsync mode instead of using lfs migrate
73 -R restripe file using default directory striping
74 -s skip file data comparison after migrate
76 restripe file using the specified stripe size
77 -v show verbose debug messages
79 when -A is set, limit the amount of space on each OST that
80 can be considered available for the migration to
82 -y answer 'y' to usage question
83 -0 input file names on stdin are separated by a null character
85 Options '-A', '-c', and '-R' are mutually exclusive.
86 Options '-C', '-M', and '-X' are ignored if '-A' is not set.
88 The --rsync and --no-rsync options may not be specified at the same time.
90 If a directory is an argument, all files in the directory are migrated.
91 If no file/directory is given, the file list is read from standard input.
93 Any arguments that are not explicitly recognized by the script are passed
94 through to the 'lfs migrate' utility.
97 lfs_migrate /mnt/lustre/dir
98 lfs_migrate -p newpool /mnt/lustre/dir
99 lfs find /test -O test-OST0004 -size +4G | lfs_migrate -y
105 rm -f "$MIGRATED_SET"
106 [ -n "$NEWNAME" ] && rm -f "$NEWNAME"
131 # Examine any long options and arguments. getopts does not support long
132 # options, so they must be stripped out and classified as either options
133 # for the script, or passed through to "lfs migrate".
134 while [ -n "$*" ]; do
138 -l|--link) ;; # maintained backward compatibility for now
139 -n) OPT_DRYRUN=true; OPT_YES=true
140 echo "$PROG: -n deprecated, use --dry-run or --non-block" 1>&2;;
141 --dry-run) OPT_DRYRUN=true; OPT_YES=true;;
142 -p|--pool) OPT_POOL="$arg $2"; OPT_LAYOUT+="$OPT_POOL "; shift;;
144 -R|--restripe) OPT_RESTRIPE=true;;
145 -s|--skip) OPT_CHECK=false;;
146 -v|--verbose) OPT_DEBUG=true; ECHO=echo;;
147 -y|--yes) OPT_YES=true;;
149 -b|--block|--non-block|--non-direct|-D|--no-verify)
150 # Always pass non-layout options to 'lfs migrate'
151 OPT_PASSTHROUGH+=("$arg");;
152 --rsync) OPT_RSYNC=true;;
153 --no-rsync) OPT_NO_RSYNC=true;;
154 --copy|--yaml|--file) OPT_COMP=true;
155 # these options have files as arguments, pass both through
156 OPT_LAYOUT+="$arg $2 "; shift;;
157 --auto-stripe|-A) OPT_AUTOSTRIPE=true;;
158 -C) OPT_CAP="$2"; shift;;
159 -M|--min-free) OPT_MINFREE="$2"; shift;;
160 -X|--max-free) OPT_MAXFREE="$2"; shift;;
161 -c|--stripe-count) OPT_STRIPE_COUNT="$2"; shift;;
162 -S|--stripe-size) OPT_STRIPE_SIZE="$2"; shift;;
163 *) # Pass other non-file layout options to 'lfs migrate'
164 [ -e "$arg" ] && OPT_FILE+="$arg " && break || OPT_LAYOUT+="$arg "
169 if $OPT_RESTRIPE || $OPT_AUTOSTRIPE && [ -n "$OPT_LAYOUT" ]; then
170 echo "$PROG error: Options '$OPT_LAYOUT' can't be used with -R or -A" \
173 elif $OPT_RESTRIPE && [[ "$OPT_STRIPE_COUNT" || "$OPT_STRIPE_SIZE" ]]; then
174 echo "$PROG error: Option -R can't be used with -c or -S" 1>&2
176 elif $OPT_AUTOSTRIPE && [ -n "$OPT_STRIPE_COUNT" ]; then
177 echo "$PROG error: Option -A can't be used with -c" 1>&2
179 elif $OPT_AUTOSTRIPE && $OPT_RESTRIPE; then
180 echo "$PROG error: Option -A can't be used with -R" 1>&2
184 if $OPT_RSYNC && $OPT_NO_RSYNC; then
185 echo "$PROG: Options --rsync and --no-rsync may not be" \
186 "specified at the same time." 1>&2
192 echo "lfs_migrate is currently NOT SAFE for moving in-use files." 1>&2
193 echo "Use it only when you are sure migrated files are unused." 1>&2
195 echo "If emptying an OST that is active on the MDS, new files may" 1>&2
196 echo "use it. To stop allocating any new objects on OSTNNNN run:" 1>&2
197 echo " lctl set_param osp.<fsname>-OSTNNNN*.max_create_count=0'" 1>&2
198 echo "on each MDS using the OST(s) being emptied." 1>&2
199 echo -n "Continue? (y/n) "
201 [ "$CHECK" != "y" -a "$CHECK" != "yes" ] && exit 1
204 # if rsync has --xattr support, then try to copy the xattrs.
205 $RSYNC --help 2>&1 | grep -q xattr && RSYNC_OPTS="$RSYNC_OPTS -X"
206 $RSYNC --help 2>&1 | grep -q acls && RSYNC_OPTS="$RSYNC_OPTS -A"
207 # If rsync copies lustre xattrs in the future, then we can skip lfs (bug 22189)
208 strings $(which $RSYNC) 2>&1 | grep -q lustre && LFS=:
210 # rsync creates its temporary files with lenient permissions, even if
211 # permissions on the original files are more strict. Tighten umask here
212 # to avoid the brief window where unprivileged users might be able to
213 # access the temporary file.
216 # Use stripe count = sqrt(size_in_GB) + 1, but cap object size per OST.
217 function calc_stripe()
222 local filegb=$((filekb / 1048576))
224 local ost_max_count=0
226 # Files up to 1GB will have 1 stripe if they fit within the object max
227 if [[ $filegb -lt 1 && "$obj_max_kb" && $filekb -le $obj_max_kb ]]; then
228 echo 1 "$obj_max_kb" && return
231 stripe_count=$(bc <<< "scale=0; 1 + sqrt($filegb)" 2> /dev/null) ||
232 { echo "cannot auto calculate stripe count" >&2; return; }
234 if [ -z "$obj_max_kb" ]; then
235 local ost_min_kb=$((1 << 62))
237 # Calculate cap on object size at 1% of smallest OST
238 # but only include OSTs that have 256MB+ available space
239 while IFS='' read avail; do
240 [[ "$OPT_MAXFREE" && $avail -gt $OPT_MAXFREE ]] &&
242 if [ $avail -ge $OPT_MINFREE ]; then
243 ost_max_count=$((ost_max_count + 1))
244 if [ $avail -lt $ost_min_kb ]; then
248 done < <($LFS df $OPT_POOL "$OLDNAME" | awk '/OST/ { print $4 }')
250 if [ $ost_max_count -eq 0 ]; then
251 # no OSTs with enough space, stripe over all of them
256 if (( ost_min_kb == (1 << 62) )); then
257 echo "warning: unable to determine minimum OST size, " \
258 "object size not capped" >&2
259 echo "$stripe_count" "0"
263 obj_max_kb=$((ost_min_kb / $OPT_CAP))
264 elif [ $obj_max_kb -eq 0 ]; then
265 echo "warning: unable to determine minimum OST size " \
266 "from previous migrate, object size not capped" >&2
267 echo "$stripe_count" "$obj_max_kb"
271 # If disk usage would exceed the cap, increase the number of stripes.
272 # Round up to the nearest MB to ensure file will fit.
273 (( filekb > stripe_count * obj_max_kb )) &&
274 stripe_count=$(((filekb + obj_max_kb - 1024) / obj_max_kb))
276 # Limit the count to the number of eligible OSTs
277 if [ "$stripe_count" -gt $ost_max_count ]; then
278 echo "$ost_max_count" "$obj_max_kb"
280 echo "$stripe_count" "$obj_max_kb"
288 while IFS='' read -d '' OLDNAME; do
293 $ECHO -n "$OLDNAME: "
295 # avoid duplicate stat call by fetching all attrs at once
296 local nlink_idx_link=0 # %h is the hard link count
297 local nlink_idx_type=1 # %F is "regular file", ignore others
298 local nlink_idx_file=2 # "file" is here
299 local nlink_idx_size=3 # %s is file size in bytes
300 local nlink_idx_dev=4 # %D is the underlying device number
301 # nlink_type=(1 regular file 1234 0x810)
302 local nlink_type=($(LANG=C stat -c "%h %F %s %D" "$OLDNAME" \
305 # skip non-regular files, since they don't have any objects
306 # and there is no point in trying to migrate them.
307 if [ "${nlink_type[$nlink_idx_type]}" != "regular" ]; then
308 echo -e "\r$OLDNAME: not a regular file, skipped" 1>&2
312 # working out write perms is hard, let the shell do it
313 if [ ! -w "$OLDNAME" ]; then
314 echo -e "\r$OLDNAME: no write permission, skipped" 1>&2
318 if $OPT_DRYRUN && ! $OPT_DEBUG; then
319 $ECHO "dry run, skipped"
323 # xattrs use absolute file paths, so ensure provided path is
324 # also absolute so that the names can be compared
325 local oldname_absolute=$(readlink -f "$OLDNAME")
326 if [ -z "$oldname_absolute" ]; then
327 echo -e "\r$OLDNAME: cannot resolve full path, skipped" 1>&2
330 OLDNAME=$oldname_absolute
332 if [[ ${nlink_type[$nlink_idx_link]} -gt 1 ]] ||
333 $RSYNC_WITH_HLINKS; then
334 fid=$($LFS path2fid "$OLDNAME" 2> /dev/null)
335 if [ $? -ne 0 ]; then
336 echo -e "\r$OLDNAME: cannot get FID, skipping; is this a Lustre file system?" 1>&2
340 # don't migrate a hard link if it was already migrated
341 if path_in_set "$OLDNAME"; then
342 $ECHO "already migrated via another hard link"
346 # There is limited space available in the xattrs
347 # to store all of the hard links for a file, so it's
348 # possible that $OLDNAME is part of a link set but is
349 # not listed in xattrs and therefore not listed as
351 local migrated=$(old_fid_in_set "$fid")
352 if [ -n "$migrated" ]; then
353 $ECHO "already migrated via another hard link"
354 # Only the rsync case has to relink. The
355 # "lfs migrate" case keeps the same inode so
356 # all of the links are already correct.
357 $OPT_RSYNC && [ "$migrated" != "$OLDNAME" ] &&
358 ln -f "$migrated" "$OLDNAME"
360 add_to_set "$fid" "$OLDNAME"
365 local olddir=$(dirname "$OLDNAME")
366 local stripe_size="$OPT_STRIPE_SIZE"
367 local stripe_count="$OPT_STRIPE_COUNT"
368 local stripe_opts="-N --comp-count -c -S -p -y"
369 local parent_count=""
371 local stripe_pool="${OPT_POOL#-p }"
374 # avoid multiple getstripe calls
375 # lcm_mirror_count: 1
377 # lmm_stripe_count: 1
378 # lmm_stripe_size: 1048576
380 local l_mirror_count=0
382 local l_stripe_count=2
383 local l_stripe_size=3
384 local l_stripe_pool=4
385 local layout_info=($($LFS getstripe $stripe_opts "$OLDNAME" \
386 2>/dev/null | awk '{ print $2 }'))
388 layout="${OPT_PASSTHROUGH[@]} "
390 if $OPT_RESTRIPE; then
392 layout+="--copy $olddir"
395 # If rsync copies Lustre xattrs properly in the future
396 # (i.e. before the file data, so that it preserves
397 # striping) then we don't need this getstripe stuff.
400 [ -n "$OPT_POOL" ] ||
401 stripe_pool=${layout_info[$l_stripe_pool]}
402 mirror_count=${layout_info[$l_mirror_count]}
404 if $OPT_AUTOSTRIPE; then
405 local filekb=$((${nlink_type[$nlink_idx_size]} /
408 read stripe_count OBJ_MAX_KB < <(calc_stripe \
409 "$OLDNAME" "$filekb" "$OBJ_MAX_KB")
410 [ -z "$stripe_count" ] && exit 1
411 [ $stripe_count -lt 1 ] && stripe_count=1
413 [ -n "$stripe_count" ] ||
414 stripe_count=${layout_info[$l_stripe_count]}
416 [ -n "$stripe_size" ] ||
417 stripe_size=${layout_info[$l_stripe_size]}
419 [ -z "$stripe_count" -o -z "$stripe_size" ] && UNLINK=""
427 if $OPT_RESTRIPE; then
428 parent_layout=($($LFS getstripe $stripe_opts \
429 -d "$olddir" 2>/dev/null |
431 parent_count=${parent_layout[$l_stripe_count]}
432 parent_size=${parent_layout[$l_stripe_size]}
433 stripe_pool=${parent_layout[$l_stripe_pool]}
434 mirror_count=${parent_layout[$l_mirror_count]}
437 $ECHO -n "stripe_count=${stripe_count:-$parent_count},stripe_size=${stripe_size:-$parent_size}"
438 [ -n "$stripe_pool" ] &&
439 $ECHO -n ",pool=${stripe_pool}"
440 [[ $mirror_count -gt 1 ]] &&
441 $ECHO -n ",mirror_count=${mirror_count}"
446 $ECHO " dry run, skipped"
450 if ! $OPT_COMP && [ ${layout_info[$l_comp_count]} -gt 0 ]; then
451 layout+="--copy $OLDNAME"
455 [ -n "$stripe_count" ] && layout+="-c $stripe_count "
456 [ -n "$stripe_size" ] && layout+="-S $stripe_size "
457 [ -n "$OPT_POOL" -a -n "$stripe_pool" ] &&
458 layout+="-p $stripe_pool "
459 [[ $mirror_count -gt 1 ]] && layout+="-N $mirror_count "
461 layout+="$OPT_LAYOUT"
463 # detect other hard links and store them on a global
464 # list so we don't re-migrate them
465 if [[ ${nlink_type[$nlink_idx_link]} -gt 1 ]]; then
466 [ "${nlink_type[$nlink_idx_dev]}" == "$last_dev" ] ||
467 mntpoint=$(df -P "$OLDNAME" |
468 awk 'NR==2 { print $NF }')
469 if [ -z "$mntpoint" ]; then
470 echo -e "\r$OLDNAME: cannot determine mount point; skipped" 1>&2
473 hlinks=$($LFS fid2path "$mntpoint" "$fid" 2> /dev/null)
474 if $OPT_RSYNC && [ $? -ne 0 ]; then
475 echo -e "\r$OLDNAME: cannot determine hard link paths, skipped" 1>&2
483 # first try to migrate via Lustre tools, then fall back to rsync
484 if ! $OPT_RSYNC; then
486 echo -e "\n$LFS migrate $layout \"$OLDNAME\""
488 if $LFS migrate $layout "$OLDNAME"; then
490 # no-op if hlinks empty for 1-link files
491 for link in ${hlinks[*]}; do
492 add_to_set "$fid" "$link"
495 elif $OPT_NO_RSYNC; then
496 echo -e "\r$OLDNAME: refusing to fall back to rsync, skipped" 1>&2
499 $ECHO -n "falling back to rsync: "
504 local oldfile=$(basename "$OLDNAME")
505 NEWNAME=$(mktemp $UNLINK "$olddir/.$oldfile.XXXXXX")
506 if [ $? -ne 0 -o -z "$NEWNAME" ]; then
507 echo -e "\r$OLDNAME: cannot make temp file, skipped" 1>&2
511 if [ "$UNLINK" ]; then
512 if ! $LFS setstripe $layout "$NEWNAME"; then
513 echo -e "\r$NEWNAME: setstripe failed, exiting" 1>&2
518 # we use --inplace, since we created our own temp file already
519 if ! $RSYNC -a --inplace $RSYNC_OPTS "$OLDNAME" "$NEWNAME";then
520 echo -e "\r$OLDNAME: copy error, exiting" 1>&2
524 if $OPT_CHECK && ! cmp -s "$OLDNAME" "$NEWNAME"; then
525 echo -e "\r$NEWNAME: compare failed, exiting" 1>&2
529 if ! mv "$NEWNAME" "$OLDNAME"; then
530 echo -e "\r$OLDNAME: rename error, exiting" 1>&2
535 # no-op if hlinks empty for 1-link files
536 for link in ${hlinks[*]}; do
537 if [ "$link" != "$OLDNAME" ]; then
538 ln -f "$OLDNAME" "$link"
540 add_to_set "$fid" "$link"
543 # If the number of hlinks exceeds the space in the xattrs,
544 # when the final path is statted it will have a link count
545 # of 1 (all other links will point to the new inode).
546 # This flag indicates that even paths with a link count of
547 # 1 are potentially part of a link set.
548 (( ${#hlinks[*]} == 1 )) || RSYNC_WITH_HLINKS=true
552 if [ "$#" -eq 0 ]; then
556 tr '\n' '\0' | lfs_migrate
561 $LFS find "$1" -type f -print0