4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * lustre/utils/ll_recover_lost_found_objs.c
36 * Tool for recovering objects from lost+found that might result from a
37 * Lustre OST with a corrupted directory. Running e2fsck will fix the
38 * directory, but puts all of the objects into lost+found, where they are
39 * inaccessible to Lustre.
41 * Author: Kalpak Shah <kalpak.shah@sun.com>
54 #include <sys/types.h>
55 #include <sys/xattr.h>
58 #include <liblustre.h>
59 #include <libcfs/list.h>
65 struct obd_group_info {
71 cfs_list_t grp_info_list;
73 static void grp_info_list_destroy(cfs_list_t *list)
75 struct obd_group_info *grp, *tmp;
77 cfs_list_for_each_entry_safe(grp, tmp, list, grp_list) {
78 cfs_list_del_init(&grp->grp_list);
83 static void usage(char *progname)
85 fprintf(stderr, "Usage: %s [-hv] -d lost+found_directory\n", progname);
86 fprintf(stderr, "You need to mount the corrupted OST filesystem and "
87 "provide the path for the lost+found directory as the -d "
88 "option, for example:\n"
89 "ll_recover_lost_found_objs -d /mnt/ost/lost+found\n");
93 static int _ll_sprintf(char *buf, size_t size, const char *func, int line,
94 const char *format, ...)
100 rc = vsnprintf(buf, size, format, ap);
101 if (!(rc > -1 && rc < size)) {
103 "error: %s:%d: path \"", func, line);
104 vfprintf(stderr, format, ap);
106 fprintf(stderr, "\" is too long\n");
113 #define ll_sprintf(buf, size, format, ...) \
114 _ll_sprintf(buf, size, __FUNCTION__, __LINE__, format, ## __VA_ARGS__)
116 static int mkdir_p(const char *dest_path, const char *mount)
118 struct stat stat_buf;
122 if (stat(dest_path, &stat_buf) == 0)
125 retval = mkdir(dest_path, mode);
127 fprintf(stderr, "error: creating directory %s: "
128 "%s\n", dest_path, strerror(errno));
135 /* This is returning 0 for an error */
136 static __u64 read_last_id(char *file_path)
142 fd = open(file_path, O_RDONLY);
145 fprintf(stderr, "error: opening %s: %s\n",
146 file_path, strerror(errno));
150 count = read(fd, &last_id, sizeof(last_id));
152 fprintf(stderr, "error: reading file %s: %s\n", file_path,
157 if (count != sizeof(last_id)) {
158 fprintf(stderr, "error: Could not read full last_id from %s\n",
165 return le64_to_cpu(last_id);
168 struct obd_group_info *find_or_create_grp(cfs_list_t *list, __u64 seq,
171 struct obd_group_info *grp;
173 char tmp_path[PATH_MAX];
175 struct stat stat_buf;
179 cfs_list_for_each(entry, list) {
180 grp = (struct obd_group_info *)cfs_list_entry(entry,
181 struct obd_group_info,
183 if (grp->grp_seq == seq)
187 grp = malloc(sizeof(struct obd_group_info));
191 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
192 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
193 fid_seq_is_idif(seq) ? 0 : seq);
195 /* Check whether the obj dir has been created */
196 if (ll_sprintf(tmp_path, PATH_MAX, "%s/O/%s", mount, seq_name)) {
201 if (stat(tmp_path, &stat_buf) != 0) {
202 retval = mkdir(tmp_path, 0700);
205 fprintf(stderr, "error: creating directory %s: "
206 "%s\n", tmp_path, strerror(errno));
211 if (ll_sprintf(tmp_path, PATH_MAX, "%s/O/%s/LAST_ID",
218 * Object ID needs to be verified against last_id.
219 * LAST_ID file may not be present in the group directory
220 * due to corruption. In case of any error tyr to recover
221 * as many objects as possible by setting last_id to ~0ULL.
223 tmp_last_id = read_last_id(tmp_path);
224 if (tmp_last_id == 0)
226 grp->grp_last_id = tmp_last_id;
229 cfs_list_add(&grp->grp_list, list);
233 static unsigned filetype_dir_table[] = {
242 #if defined(DT_DOOR) && defined(S_IFDOOR)
247 static int traverse_lost_found(char *src_dir, const char *mount_path)
250 struct filter_fid parent_fid;
251 struct dirent64 *dirent;
252 __u64 ff_seq, ff_objid;
254 char dest_path[PATH_MAX];
256 int obj_exists, xattr_len;
257 int len, ret = 0, error = 0;
260 struct obd_group_info *grp_info;
262 len = strlen(src_dir);
264 dir_ptr = opendir(src_dir);
266 fprintf(stderr, "error: opening directory: %s\n",
271 while ((dirent = readdir64(dir_ptr)) != NULL) {
272 if (!strcmp(dirent->d_name, ".") ||
273 !strcmp(dirent->d_name, ".."))
277 if ((len + strlen(dirent->d_name) + 2) > PATH_MAX) {
278 fprintf(stderr, "error: %s/%s: path too long\n",
279 src_dir, dirent->d_name);
282 strcat(src_dir, "/");
283 strcat(src_dir, dirent->d_name);
285 if (dirent->d_type == DT_UNKNOWN) {
286 ret = stat(src_dir, &st);
289 "error: stating %s: %s\n",
290 src_dir, strerror(errno));
293 dirent->d_type = filetype_dir_table[st.st_mode &
295 if (dirent->d_type == DT_UNKNOWN) {
297 "error: %s of unknown type 0%o\n",
298 src_dir, st.st_mode);
303 switch(dirent->d_type) {
305 ret = traverse_lost_found(src_dir, mount_path);
314 xattr_len = getxattr(file_path, "trusted.fid",
318 if (xattr_len == -1 || xattr_len < sizeof(parent_fid))
320 * Its very much possible that we dont find fid
321 * on precreated files, LAST_ID
325 ff_seq = le64_to_cpu(parent_fid.ff_seq);
326 sprintf(seq_name, (fid_seq_is_rsvd(ff_seq) ||
327 fid_seq_is_mdt0(ff_seq)) ? LPU64 : LPX64i,
328 fid_seq_is_idif(ff_seq) ? 0 : ff_seq);
331 ff_objid = le64_to_cpu(parent_fid.ff_objid);
332 sprintf(obj_name, (fid_seq_is_rsvd(parent_fid.ff_seq) ||
333 fid_seq_is_mdt0(parent_fid.ff_seq) ||
334 fid_seq_is_idif(parent_fid.ff_seq)) ?
335 LPU64 : LPX64i, ff_objid);
337 grp_info = find_or_create_grp(&grp_info_list, ff_seq,
339 if (grp_info == NULL) {
344 /* might need to create the parent directories for
346 if (ll_sprintf(dest_path, PATH_MAX, "%s/O/%s/d"LPU64,
347 mount_path, seq_name, ff_objid % 32)) {
352 ret = mkdir_p(dest_path, mount_path);
358 if (ff_objid > grp_info->grp_last_id) {
359 fprintf(stderr, "error: file skipped because object ID "
360 "greater than LAST_ID\nFilename: %s\n"
361 "Group: "LPU64"\nObjectid: "LPU64"\n"
362 "LAST_ID: "LPU64, file_path, ff_seq, ff_objid,
363 grp_info->grp_last_id);
367 /* move file from lost+found to proper object
369 if (ll_sprintf(dest_path, PATH_MAX,
370 "%s/O/%s/d"LPU64"/%s", mount_path,
371 seq_name, ff_objid % 32, obj_name)) {
377 ret = stat(dest_path, &st);
384 "warning: stat for %s: %s\n",
385 dest_path, strerror(errno));
390 fprintf(stderr, "error: target object %s already "
391 "exists and will not be replaced.\n",dest_path);
395 if (rename(file_path, dest_path) < 0) {
396 fprintf(stderr, "error: rename failed for file %s: %s\n",
397 file_path, strerror(errno));
402 printf("Object %s restored.\n", dest_path);
413 * If LAST_ID file is not present in some group then restore it with the highest
414 * object ID found in that group. By the time we come here all possible objects
415 * have been restored.
417 static int check_last_id(const char *mount_path)
419 char lastid_path[PATH_MAX];
420 char dirname[PATH_MAX], subdirname[PATH_MAX];
421 DIR *groupdir, *subdir;
423 struct dirent *dirent;
429 for (group = 0; group < MAX_GROUPS; group++) {
432 if (ll_sprintf(dirname, PATH_MAX, "%s/O/"LPU64,
435 if (ll_sprintf(lastid_path, PATH_MAX, "%s/LAST_ID", dirname))
438 if (stat(lastid_path, &st) == 0)
441 groupdir = opendir(dirname);
442 if (groupdir == NULL) {
444 fprintf(stderr, "error: opening %s: %s\n",
445 dirname, strerror(errno));
449 while ((dirent = readdir(groupdir)) != NULL) {
450 if (!strcmp(dirent->d_name, ".") ||
451 !strcmp(dirent->d_name, ".."))
454 if (ll_sprintf(subdirname, PATH_MAX, "%s/%s",
455 dirname, dirent->d_name)) {
459 subdir = opendir(subdirname);
460 if (subdir == NULL) {
461 fprintf(stderr, "error: opening %s: %s\n",
462 subdirname, strerror(errno));
466 while ((dirent = readdir(subdir)) != NULL) {
470 if (!strcmp(dirent->d_name, ".") ||
471 !strcmp(dirent->d_name, ".."))
474 objid = strtoull(dirent->d_name, &end, 0);
475 if (end == dirent->d_name || *end != 0) {
476 fprintf(stderr, "error: unknown object"
477 "ID %s/%s\n", subdirname,
481 if (objid > max_objid)
488 fd = open(lastid_path, O_RDWR | O_CREAT, 0700);
490 fprintf(stderr, "error: open \"%s\" failed: %s\n",
491 lastid_path, strerror(errno));
495 max_objid = cpu_to_le64(max_objid);
496 ret = write(fd, &max_objid, sizeof(__u64));
497 if (ret < sizeof(__u64)) {
498 fprintf(stderr, "error: write \"%s\" failed: %s\n",
499 lastid_path, strerror(errno));
510 int main(int argc, char **argv)
513 struct stat stat_buf;
514 char src_dir[PATH_MAX] = "";
515 char mount_path[PATH_MAX];
516 char tmp_path[PATH_MAX];
522 while ((c = getopt(argc, argv, "d:hv")) != EOF) {
526 fprintf(stderr, "error: chdir to %s: %s\n",
527 optarg, strerror(errno));
530 if (getcwd(src_dir, PATH_MAX) == NULL) {
532 "error: getcwd of lost+found: %s\n",
537 fprintf(stderr, "error: chdir to \"..\": %s\n",
541 if (getcwd(mount_path, PATH_MAX) == NULL) {
543 "error: getcwd of mount point: %s\n",
547 if (!strcmp(src_dir, mount_path)) {
549 "error: root directory is detected\n");
552 fprintf(stdout, "\"lost+found\" directory path: %s\n",
561 fprintf(stderr, "%s: bad option '%c'\n",
570 /* Check if 'O' directory exists and create it if needed */
571 if (ll_sprintf(tmp_path, PATH_MAX, "%s/O", mount_path))
574 if (stat(tmp_path, &stat_buf) != 0) {
575 retval = mkdir(tmp_path, 0700);
577 fprintf(stderr, "error: creating objects directory %s:"
578 " %s\n", tmp_path, strerror(errno));
583 CFS_INIT_LIST_HEAD(&grp_info_list);
584 retval = traverse_lost_found(src_dir, mount_path);
586 fprintf(stderr, "error: traversing lost+found looking for "
587 "orphan objects.\n");
591 retval = check_last_id(mount_path);
593 fprintf(stderr, "error: while checking/restoring LAST_ID.\n");
596 grp_info_list_destroy(&grp_info_list);