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,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2012, 2013, Intel Corporation.
26 * lustre/lfsck/lfsck_namespace.c
28 * Author: Fan, Yong <fan.yong@intel.com>
31 #define DEBUG_SUBSYSTEM S_LFSCK
33 #include <lustre/lustre_idl.h>
34 #include <lu_object.h>
35 #include <dt_object.h>
36 #include <md_object.h>
37 #include <lustre_fid.h>
38 #include <lustre_lib.h>
39 #include <lustre_net.h>
40 #include <lustre/lustre_user.h>
42 #include "lfsck_internal.h"
44 #define LFSCK_NAMESPACE_MAGIC 0xA0629D03
46 enum lfsck_nameentry_check {
47 LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */
48 LFSCK_NAMEENTRY_REMOVED = 2, /* The entry has been removed. */
49 LFSCK_NAMEENTRY_RECREATED = 3, /* The entry has been recreated. */
52 static const char lfsck_namespace_name[] = "lfsck_namespace";
54 static struct lfsck_namespace_req *
55 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
56 struct lu_dirent *ent, __u16 type)
58 struct lfsck_namespace_req *lnr;
61 size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
64 return ERR_PTR(-ENOMEM);
66 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
67 lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
68 lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
69 lnr->lnr_fid = ent->lde_fid;
70 lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
71 lnr->lnr_dir_cookie = ent->lde_hash;
72 lnr->lnr_attr = ent->lde_attrs;
75 lnr->lnr_namelen = ent->lde_namelen;
76 memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
81 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
82 struct lfsck_assistant_req *lar)
84 struct lfsck_namespace_req *lnr =
85 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
87 if (lnr->lnr_lmv != NULL)
88 lfsck_lmv_put(env, lnr->lnr_lmv);
90 lu_object_put(env, &lnr->lnr_obj->do_lu);
91 OBD_FREE(lnr, lnr->lnr_size);
94 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
95 struct lfsck_namespace *src)
97 dst->ln_magic = le32_to_cpu(src->ln_magic);
98 dst->ln_status = le32_to_cpu(src->ln_status);
99 dst->ln_flags = le32_to_cpu(src->ln_flags);
100 dst->ln_success_count = le32_to_cpu(src->ln_success_count);
101 dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
102 dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
103 dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
104 dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
105 dst->ln_time_last_checkpoint =
106 le64_to_cpu(src->ln_time_last_checkpoint);
107 lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
108 &src->ln_pos_latest_start);
109 lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
110 &src->ln_pos_last_checkpoint);
111 lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
112 &src->ln_pos_first_inconsistent);
113 dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
114 dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
115 dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
116 dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
117 dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
118 dst->ln_objs_repaired_phase2 =
119 le64_to_cpu(src->ln_objs_repaired_phase2);
120 dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
121 dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
122 fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
123 &src->ln_fid_latest_scanned_phase2);
124 dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
125 dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
126 dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
127 dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
128 dst->ln_unknown_inconsistency =
129 le64_to_cpu(src->ln_unknown_inconsistency);
130 dst->ln_unmatched_pairs_repaired =
131 le64_to_cpu(src->ln_unmatched_pairs_repaired);
132 dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
133 dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
134 dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
135 dst->ln_lost_dirent_repaired =
136 le64_to_cpu(src->ln_lost_dirent_repaired);
137 dst->ln_striped_dirs_scanned =
138 le64_to_cpu(src->ln_striped_dirs_scanned);
139 dst->ln_striped_dirs_repaired =
140 le64_to_cpu(src->ln_striped_dirs_repaired);
141 dst->ln_striped_dirs_failed =
142 le64_to_cpu(src->ln_striped_dirs_failed);
143 dst->ln_striped_dirs_disabled =
144 le64_to_cpu(src->ln_striped_dirs_disabled);
145 dst->ln_striped_dirs_skipped =
146 le64_to_cpu(src->ln_striped_dirs_skipped);
147 dst->ln_striped_shards_scanned =
148 le64_to_cpu(src->ln_striped_shards_scanned);
149 dst->ln_striped_shards_repaired =
150 le64_to_cpu(src->ln_striped_shards_repaired);
151 dst->ln_striped_shards_failed =
152 le64_to_cpu(src->ln_striped_shards_failed);
153 dst->ln_striped_shards_skipped =
154 le64_to_cpu(src->ln_striped_shards_skipped);
155 dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
156 dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
157 dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
158 dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
159 dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
160 dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
163 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
164 struct lfsck_namespace *src)
166 dst->ln_magic = cpu_to_le32(src->ln_magic);
167 dst->ln_status = cpu_to_le32(src->ln_status);
168 dst->ln_flags = cpu_to_le32(src->ln_flags);
169 dst->ln_success_count = cpu_to_le32(src->ln_success_count);
170 dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
171 dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
172 dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
173 dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
174 dst->ln_time_last_checkpoint =
175 cpu_to_le64(src->ln_time_last_checkpoint);
176 lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
177 &src->ln_pos_latest_start);
178 lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
179 &src->ln_pos_last_checkpoint);
180 lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
181 &src->ln_pos_first_inconsistent);
182 dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
183 dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
184 dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
185 dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
186 dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
187 dst->ln_objs_repaired_phase2 =
188 cpu_to_le64(src->ln_objs_repaired_phase2);
189 dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
190 dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
191 fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
192 &src->ln_fid_latest_scanned_phase2);
193 dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
194 dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
195 dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
196 dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
197 dst->ln_unknown_inconsistency =
198 cpu_to_le64(src->ln_unknown_inconsistency);
199 dst->ln_unmatched_pairs_repaired =
200 cpu_to_le64(src->ln_unmatched_pairs_repaired);
201 dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
202 dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
203 dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
204 dst->ln_lost_dirent_repaired =
205 cpu_to_le64(src->ln_lost_dirent_repaired);
206 dst->ln_striped_dirs_scanned =
207 cpu_to_le64(src->ln_striped_dirs_scanned);
208 dst->ln_striped_dirs_repaired =
209 cpu_to_le64(src->ln_striped_dirs_repaired);
210 dst->ln_striped_dirs_failed =
211 cpu_to_le64(src->ln_striped_dirs_failed);
212 dst->ln_striped_dirs_disabled =
213 cpu_to_le64(src->ln_striped_dirs_disabled);
214 dst->ln_striped_dirs_skipped =
215 cpu_to_le64(src->ln_striped_dirs_skipped);
216 dst->ln_striped_shards_scanned =
217 cpu_to_le64(src->ln_striped_shards_scanned);
218 dst->ln_striped_shards_repaired =
219 cpu_to_le64(src->ln_striped_shards_repaired);
220 dst->ln_striped_shards_failed =
221 cpu_to_le64(src->ln_striped_shards_failed);
222 dst->ln_striped_shards_skipped =
223 cpu_to_le64(src->ln_striped_shards_skipped);
224 dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
225 dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
226 dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
227 dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
228 dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
229 dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
232 static void lfsck_namespace_record_failure(const struct lu_env *env,
233 struct lfsck_instance *lfsck,
234 struct lfsck_namespace *ns)
236 struct lfsck_position pos;
238 ns->ln_items_failed++;
239 lfsck_pos_fill(env, lfsck, &pos, false);
240 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
241 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
242 ns->ln_pos_first_inconsistent = pos;
244 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
245 "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n",
246 lfsck_lfsck2name(lfsck),
247 ns->ln_pos_first_inconsistent.lp_oit_cookie,
248 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
249 ns->ln_pos_first_inconsistent.lp_dir_cookie);
254 * Load the MDT bitmap from the lfsck_namespace trace file.
256 * \param[in] env pointer to the thread context
257 * \param[in] com pointer to the lfsck component
259 * \retval 0 for success
260 * \retval negative error number on failure or data corruption
262 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
263 struct lfsck_component *com)
265 struct dt_object *obj = com->lc_obj;
266 struct lfsck_assistant_data *lad = com->lc_data;
267 struct lfsck_namespace *ns = com->lc_file_ram;
268 cfs_bitmap_t *bitmap = lad->lad_bitmap;
274 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
276 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
278 nbits = ns->ln_bitmap_size;
280 if (unlikely(nbits < BITS_PER_LONG))
281 nbits = BITS_PER_LONG;
283 if (nbits > bitmap->size) {
284 __u32 new_bits = bitmap->size;
285 cfs_bitmap_t *new_bitmap;
287 while (new_bits < nbits)
290 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
291 if (new_bitmap == NULL)
294 lad->lad_bitmap = new_bitmap;
295 CFS_FREE_BITMAP(bitmap);
299 if (ns->ln_bitmap_size == 0) {
300 lad->lad_incomplete = 0;
301 CFS_RESET_BITMAP(bitmap);
306 size = (ns->ln_bitmap_size + 7) >> 3;
307 rc = dt_xattr_get(env, obj,
308 lfsck_buf_get(env, bitmap->data, size),
309 XATTR_NAME_LFSCK_BITMAP, BYPASS_CAPA);
311 RETURN(rc >= 0 ? -EINVAL : rc);
313 if (cfs_bitmap_check_empty(bitmap))
314 lad->lad_incomplete = 0;
316 lad->lad_incomplete = 1;
322 * \retval +ve: the lfsck_namespace is broken, the caller should reset it.
323 * \retval 0: succeed.
324 * \retval -ve: failed cases.
326 static int lfsck_namespace_load(const struct lu_env *env,
327 struct lfsck_component *com)
329 int len = com->lc_file_size;
332 rc = dt_xattr_get(env, com->lc_obj,
333 lfsck_buf_get(env, com->lc_file_disk, len),
334 XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA);
336 struct lfsck_namespace *ns = com->lc_file_ram;
338 lfsck_namespace_le_to_cpu(ns,
339 (struct lfsck_namespace *)com->lc_file_disk);
340 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
341 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
342 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
343 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
348 } else if (rc != -ENODATA) {
349 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
350 "expected = %d: rc = %d\n",
351 lfsck_lfsck2name(com->lc_lfsck), len, rc);
358 static int lfsck_namespace_store(const struct lu_env *env,
359 struct lfsck_component *com)
361 struct dt_object *obj = com->lc_obj;
362 struct lfsck_instance *lfsck = com->lc_lfsck;
363 struct lfsck_namespace *ns = com->lc_file_ram;
364 struct lfsck_assistant_data *lad = com->lc_data;
365 cfs_bitmap_t *bitmap = NULL;
366 struct thandle *handle;
368 int len = com->lc_file_size;
373 bitmap = lad->lad_bitmap;
374 nbits = bitmap->size;
377 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
380 ns->ln_bitmap_size = nbits;
381 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
383 handle = dt_trans_create(env, lfsck->li_bottom);
385 GOTO(log, rc = PTR_ERR(handle));
387 rc = dt_declare_xattr_set(env, obj,
388 lfsck_buf_get(env, com->lc_file_disk, len),
389 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
393 if (bitmap != NULL) {
394 rc = dt_declare_xattr_set(env, obj,
395 lfsck_buf_get(env, bitmap->data, nbits >> 3),
396 XATTR_NAME_LFSCK_BITMAP, 0, handle);
401 rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
405 rc = dt_xattr_set(env, obj,
406 lfsck_buf_get(env, com->lc_file_disk, len),
407 XATTR_NAME_LFSCK_NAMESPACE, 0, handle, BYPASS_CAPA);
408 if (rc == 0 && bitmap != NULL)
409 rc = dt_xattr_set(env, obj,
410 lfsck_buf_get(env, bitmap->data, nbits >> 3),
411 XATTR_NAME_LFSCK_BITMAP, 0, handle,
417 dt_trans_stop(env, lfsck->li_bottom, handle);
421 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
422 lfsck_lfsck2name(lfsck), rc);
426 static int lfsck_namespace_init(const struct lu_env *env,
427 struct lfsck_component *com)
429 struct lfsck_namespace *ns = com->lc_file_ram;
432 memset(ns, 0, sizeof(*ns));
433 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
434 ns->ln_status = LS_INIT;
435 down_write(&com->lc_sem);
436 rc = lfsck_namespace_store(env, com);
437 up_write(&com->lc_sem);
442 * Update the namespace LFSCK trace file for the given @fid
444 * \param[in] env pointer to the thread context
445 * \param[in] com pointer to the lfsck component
446 * \param[in] fid the fid which flags to be updated in the lfsck
448 * \param[in] add true if add new flags, otherwise remove flags
450 * \retval 0 for succeed or nothing to be done
451 * \retval negative error number on failure
453 int lfsck_namespace_trace_update(const struct lu_env *env,
454 struct lfsck_component *com,
455 const struct lu_fid *fid,
456 const __u8 flags, bool add)
458 struct lfsck_instance *lfsck = com->lc_lfsck;
459 struct dt_object *obj = com->lc_obj;
460 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
461 struct dt_device *dev = lfsck->li_bottom;
462 struct thandle *th = NULL;
470 down_write(&com->lc_sem);
471 fid_cpu_to_be(key, fid);
472 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
473 (const struct dt_key *)key, BYPASS_CAPA);
476 GOTO(unlock, rc = 0);
480 } else if (rc == 0) {
482 if ((old & flags) == flags)
483 GOTO(unlock, rc = 0);
487 if ((old & flags) == 0)
488 GOTO(unlock, rc = 0);
496 th = dt_trans_create(env, dev);
498 GOTO(log, rc = PTR_ERR(th));
501 rc = dt_declare_delete(env, obj,
502 (const struct dt_key *)key, th);
508 rc = dt_declare_insert(env, obj,
509 (const struct dt_rec *)&new,
510 (const struct dt_key *)key, th);
515 rc = dt_trans_start_local(env, dev, th);
520 rc = dt_delete(env, obj, (const struct dt_key *)key,
527 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
528 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
536 if (th != NULL && !IS_ERR(th))
537 dt_trans_stop(env, dev, th);
539 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
540 "trace file, flags %x, old %x, new %x: rc = %d\n",
541 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
542 (__u32)flags, (__u32)old, (__u32)new, rc);
545 up_write(&com->lc_sem);
550 static int lfsck_namespace_check_exist(const struct lu_env *env,
551 struct dt_object *dir,
552 struct dt_object *obj, const char *name)
554 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
558 if (unlikely(lfsck_is_dead_obj(obj)))
559 RETURN(LFSCK_NAMEENTRY_DEAD);
561 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
562 (const struct dt_key *)name, BYPASS_CAPA);
564 RETURN(LFSCK_NAMEENTRY_REMOVED);
569 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
570 RETURN(LFSCK_NAMEENTRY_RECREATED);
575 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
576 struct dt_object *obj,
577 struct thandle *handle)
581 /* For destroying all invalid linkEA entries. */
582 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
586 /* For insert new linkEA entry. */
587 rc = dt_declare_xattr_set(env, obj,
588 lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
589 XATTR_NAME_LINK, 0, handle);
593 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
594 struct linkea_data *ldata)
598 if (ldata->ld_buf->lb_buf == NULL)
601 if (!dt_object_exists(obj))
604 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, BYPASS_CAPA);
606 /* Buf was too small, figure out what we need. */
607 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK,
612 lu_buf_realloc(ldata->ld_buf, rc);
613 if (ldata->ld_buf->lb_buf == NULL)
616 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK,
621 rc = linkea_init(ldata);
627 * Remove linkEA for the given object.
629 * The caller should take the ldlm lock before the calling.
631 * \param[in] env pointer to the thread context
632 * \param[in] com pointer to the lfsck component
633 * \param[in] obj pointer to the dt_object to be handled
635 * \retval 0 for repaired cases
636 * \retval negative error number on failure
638 static int lfsck_namespace_links_remove(const struct lu_env *env,
639 struct lfsck_component *com,
640 struct dt_object *obj)
642 struct lfsck_instance *lfsck = com->lc_lfsck;
643 struct dt_device *dev = lfsck->li_bottom;
644 struct thandle *th = NULL;
648 LASSERT(dt_object_remote(obj) == 0);
650 th = dt_trans_create(env, dev);
652 GOTO(log, rc = PTR_ERR(th));
654 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
658 rc = dt_trans_start_local(env, dev, th);
662 dt_write_lock(env, obj, 0);
663 if (unlikely(lfsck_is_dead_obj(obj)))
664 GOTO(unlock, rc = -ENOENT);
666 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
667 GOTO(unlock, rc = 0);
669 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th, BYPASS_CAPA);
674 dt_write_unlock(env, obj);
677 dt_trans_stop(env, dev, th);
680 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
681 "for the object "DFID": rc = %d\n",
682 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
685 struct lfsck_namespace *ns = com->lc_file_ram;
687 ns->ln_flags |= LF_INCONSISTENT;
693 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
694 struct linkea_data *ldata, struct thandle *handle)
696 const struct lu_buf *buf = lfsck_buf_get_const(env,
697 ldata->ld_buf->lb_buf,
698 ldata->ld_leh->leh_len);
700 return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle,
704 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
705 struct lu_name *cname,
709 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
710 /* To guarantee the 'name' is terminated with '0'. */
711 memcpy(buf, cname->ln_name, cname->ln_namelen);
712 buf[cname->ln_namelen] = 0;
713 cname->ln_name = buf;
716 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
717 struct lu_name *cname,
721 struct link_ea_entry *oldlee;
725 oldlee = ldata->ld_lee;
726 oldlen = ldata->ld_reclen;
727 linkea_next_entry(ldata);
728 while (ldata->ld_lee != NULL) {
729 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
730 ldata->ld_lee->lee_reclen[1];
731 if (unlikely(ldata->ld_reclen == oldlen &&
732 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
737 linkea_del_buf(ldata, cname);
739 linkea_next_entry(ldata);
742 ldata->ld_lee = oldlee;
743 ldata->ld_reclen = oldlen;
749 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
751 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
752 * with the given type to generate the name, the detailed rules for name
753 * have been described as following.
755 * The function also generates the linkEA corresponding to the name entry
756 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
758 * \param[in] env pointer to the thread context
759 * \param[in] com pointer to the lfsck component
760 * \param[in] orphan pointer to the orphan MDT-object
761 * \param[in] infix additional information for the orphan name, such as
762 * the FID for original
763 * \param[in] type the type for describing why the orphan MDT-object is
764 * created. The rules are as following:
766 * type "D": The MDT-object is a directory, it may knows its parent
767 * but because there is no valid linkEA, the LFSCK cannot
768 * know where to put it back to the namespace.
769 * type "O": The MDT-object has no linkEA, and there is no name
770 * entry that references the MDT-object.
772 * type "S": The orphan MDT-object is a shard of a striped directory
774 * \see lfsck_layout_recreate_parent() for more types.
776 * The orphan name will be like:
777 * ${FID}-${infix}-${type}-${conflict_version}
779 * \param[out] count if some others inserted some linkEA entries by race,
780 * then return the linkEA entries count.
782 * \retval positive number for repaired cases
783 * \retval 0 if needs to repair nothing
784 * \retval negative error number on failure
786 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
787 struct lfsck_component *com,
788 struct dt_object *orphan,
789 const char *infix, const char *type,
792 struct lfsck_thread_info *info = lfsck_env_info(env);
793 struct lu_name *cname = &info->lti_name;
794 struct dt_insert_rec *rec = &info->lti_dt_rec;
795 struct lu_fid *tfid = &info->lti_fid5;
796 struct lu_attr *la = &info->lti_la3;
797 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
798 const struct lu_fid *pfid;
799 struct lfsck_instance *lfsck = com->lc_lfsck;
800 struct dt_device *dev = lfsck->li_bottom;
801 struct dt_object *parent;
802 struct thandle *th = NULL;
803 struct lustre_handle plh = { 0 };
804 struct lustre_handle clh = { 0 };
805 struct linkea_data ldata = { 0 };
806 struct lu_buf linkea_buf;
813 cname->ln_name = NULL;
814 /* Create .lustre/lost+found/MDTxxxx when needed. */
815 if (unlikely(lfsck->li_lpf_obj == NULL)) {
816 rc = lfsck_create_lpf(env, lfsck);
821 parent = lfsck->li_lpf_obj;
822 pfid = lfsck_dto2fid(parent);
824 /* Hold update lock on the parent to prevent others to access. */
825 rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
826 MDS_INODELOCK_UPDATE, LCK_EX);
831 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
832 PFID(cfid), infix, type, idx++);
833 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
834 (const struct dt_key *)info->lti_key,
836 if (rc != 0 && rc != -ENOENT)
839 if (unlikely(rc == 0 && lu_fid_eq(cfid, tfid)))
841 } while (rc == 0 && !exist);
843 cname->ln_name = info->lti_key;
844 cname->ln_namelen = namelen;
845 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
849 rc = linkea_add_buf(&ldata, cname, pfid);
853 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
854 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP,
859 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
860 ldata.ld_leh->leh_len);
861 th = dt_trans_create(env, dev);
863 GOTO(log, rc = PTR_ERR(th));
865 if (S_ISDIR(lfsck_object_type(orphan))) {
866 rc = dt_declare_delete(env, orphan,
867 (const struct dt_key *)dotdot, th);
871 rec->rec_type = S_IFDIR;
873 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
874 (const struct dt_key *)dotdot, th);
879 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
880 XATTR_NAME_LINK, 0, th);
885 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
887 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
888 (const struct dt_key *)cname->ln_name,
893 if (S_ISDIR(rec->rec_type)) {
894 rc = dt_declare_ref_add(env, parent, th);
900 memset(la, 0, sizeof(*la));
901 la->la_ctime = cfs_time_current_sec();
902 la->la_valid = LA_CTIME;
903 rc = dt_declare_attr_set(env, orphan, la, th);
907 rc = dt_trans_start_local(env, dev, th);
911 dt_write_lock(env, orphan, 0);
912 rc = lfsck_links_read(env, orphan, &ldata);
913 if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
914 (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
915 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
916 GOTO(unlock, rc = 1);
918 if (S_ISDIR(lfsck_object_type(orphan))) {
919 rc = dt_delete(env, orphan,
920 (const struct dt_key *)dotdot, th,
925 rec->rec_type = S_IFDIR;
927 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
928 (const struct dt_key *)dotdot, th,
934 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
937 if (rc == 0 && count != NULL)
938 *count = ldata.ld_leh->leh_reccount;
942 dt_write_unlock(env, orphan);
944 if (rc == 0 && !exist) {
945 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
947 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
948 (const struct dt_key *)cname->ln_name,
950 if (rc == 0 && S_ISDIR(rec->rec_type)) {
951 dt_write_lock(env, parent, 0);
952 rc = dt_ref_add(env, parent, th);
953 dt_write_unlock(env, parent);
960 rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
962 GOTO(stop, rc = (rc == 0 ? 1 : rc));
965 dt_write_unlock(env, orphan);
968 dt_trans_stop(env, dev, th);
971 lfsck_ibits_unlock(&clh, LCK_EX);
972 lfsck_ibits_unlock(&plh, LCK_EX);
973 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
974 "object "DFID", name = %s: rc = %d\n",
975 lfsck_lfsck2name(lfsck), PFID(cfid),
976 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
979 struct lfsck_namespace *ns = com->lc_file_ram;
981 ns->ln_flags |= LF_INCONSISTENT;
988 * Add the specified name entry back to namespace.
990 * If there is a linkEA entry that back references a name entry under
991 * some parent directory, but such parent directory does not have the
992 * claimed name entry. On the other hand, the linkEA entries count is
993 * not larger than the MDT-object's hard link count. Under such case,
994 * it is quite possible that the name entry is lost. Then the LFSCK
995 * should add the name entry back to the namespace.
997 * \param[in] env pointer to the thread context
998 * \param[in] com pointer to the lfsck component
999 * \param[in] parent pointer to the directory under which the name entry
1000 * will be inserted into
1001 * \param[in] child pointer to the object referenced by the name entry
1002 * that to be inserted into the parent
1003 * \param[in] name the name for the child in the parent directory
1005 * \retval positive number for repaired cases
1006 * \retval 0 if nothing to be repaired
1007 * \retval negative error number on failure
1009 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1010 struct lfsck_component *com,
1011 struct dt_object *parent,
1012 struct dt_object *child,
1015 struct lfsck_thread_info *info = lfsck_env_info(env);
1016 struct lu_attr *la = &info->lti_la;
1017 struct dt_insert_rec *rec = &info->lti_dt_rec;
1018 struct lfsck_instance *lfsck = com->lc_lfsck;
1019 struct dt_device *dev = lfsck->li_next;
1020 struct thandle *th = NULL;
1021 struct lustre_handle lh = { 0 };
1025 if (unlikely(!dt_try_as_dir(env, parent)))
1026 GOTO(log, rc = -ENOTDIR);
1028 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1031 /* Hold update lock on the parent to prevent others to access. */
1032 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1033 MDS_INODELOCK_UPDATE, LCK_EX);
1037 th = dt_trans_create(env, dev);
1039 GOTO(unlock, rc = PTR_ERR(th));
1041 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1042 rec->rec_fid = lfsck_dto2fid(child);
1043 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1044 (const struct dt_key *)name, th);
1048 if (S_ISDIR(rec->rec_type)) {
1049 rc = dt_declare_ref_add(env, parent, th);
1054 memset(la, 0, sizeof(*la));
1055 la->la_ctime = cfs_time_current_sec();
1056 la->la_valid = LA_CTIME;
1057 rc = dt_declare_attr_set(env, parent, la, th);
1061 rc = dt_declare_attr_set(env, child, la, th);
1065 rc = dt_trans_start_local(env, dev, th);
1069 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1070 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1074 if (S_ISDIR(rec->rec_type)) {
1075 dt_write_lock(env, parent, 0);
1076 rc = dt_ref_add(env, parent, th);
1077 dt_write_unlock(env, parent);
1082 la->la_ctime = cfs_time_current_sec();
1083 rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA);
1087 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
1089 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1092 dt_trans_stop(env, dev, th);
1095 lfsck_ibits_unlock(&lh, LCK_EX);
1098 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1099 "the name %s and type %o to the parent "DFID": rc = %d\n",
1100 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
1101 lfsck_object_type(child) & S_IFMT,
1102 PFID(lfsck_dto2fid(parent)), rc);
1105 struct lfsck_namespace *ns = com->lc_file_ram;
1107 ns->ln_flags |= LF_INCONSISTENT;
1109 ns->ln_lost_dirent_repaired++;
1116 * Create the specified orphan MDT-object on remote MDT.
1118 * The LFSCK instance on this MDT will send LFSCK RPC to remote MDT to
1119 * ask the remote LFSCK instance to create the specified orphan object
1120 * under .lustre/lost+found/MDTxxxx/ directory with the name:
1121 * ${FID}-P-${conflict_version}.
1123 * \param[in] env pointer to the thread context
1124 * \param[in] com pointer to the lfsck component
1125 * \param[in] orphan pointer to the orphan MDT-object
1126 * \param[in] type the orphan's type to be created
1128 * type "P": The orphan object to be created was a parent directory
1129 * of some MDT-object which linkEA shows that the @orphan
1130 * object is missing.
1132 * \see lfsck_layout_recreate_parent() for more types.
1134 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1136 * \retval positive number for repaired cases
1137 * \retval 0 if needs to repair nothing
1138 * \retval negative error number on failure
1140 static int lfsck_namespace_create_orphan_remote(const struct lu_env *env,
1141 struct lfsck_component *com,
1142 struct dt_object *orphan,
1144 struct lmv_mds_md_v1 *lmv)
1146 struct lfsck_thread_info *info = lfsck_env_info(env);
1147 struct lfsck_request *lr = &info->lti_lr;
1148 struct lu_seq_range *range = &info->lti_range;
1149 const struct lu_fid *fid = lfsck_dto2fid(orphan);
1150 struct lfsck_namespace *ns = com->lc_file_ram;
1151 struct lfsck_instance *lfsck = com->lc_lfsck;
1152 struct seq_server_site *ss =
1153 lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
1154 struct lfsck_tgt_desc *ltd = NULL;
1155 struct ptlrpc_request *req = NULL;
1159 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1162 fld_range_set_mdt(range);
1163 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1167 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, range->lsr_index);
1169 ns->ln_flags |= LF_INCOMPLETE;
1171 GOTO(out, rc = -ENODEV);
1174 req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp),
1177 GOTO(out, rc = -ENOMEM);
1179 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1181 ptlrpc_request_free(req);
1186 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1187 memset(lr, 0, sizeof(*lr));
1188 lr->lr_event = LE_CREATE_ORPHAN;
1189 lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1190 lr->lr_active = LFSCK_TYPE_NAMESPACE;
1194 lr->lr_hash_type = lmv->lmv_hash_type;
1195 lr->lr_stripe_count = lmv->lmv_stripe_count;
1196 lr->lr_layout_version = lmv->lmv_layout_version;
1197 memcpy(lr->lr_pool_name, lmv->lmv_pool_name,
1198 sizeof(lr->lr_pool_name));
1201 ptlrpc_request_set_replen(req);
1202 rc = ptlrpc_queue_wait(req);
1203 ptlrpc_req_finished(req);
1207 else if (rc == -EEXIST)
1213 CDEBUG(D_LFSCK, "%s: namespace LFSCK create object "
1214 DFID" on the MDT %x remotely: rc = %d\n",
1215 lfsck_lfsck2name(lfsck), PFID(fid),
1216 ltd != NULL ? ltd->ltd_index : -1, rc);
1225 * Create the specified orphan MDT-object locally.
1227 * For the case that the parent MDT-object stored in some MDT-object's
1228 * linkEA entry is lost, the LFSCK will re-create the parent object as
1229 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1230 * with the name ${FID}-P-${conflict_version}.
1232 * \param[in] env pointer to the thread context
1233 * \param[in] com pointer to the lfsck component
1234 * \param[in] orphan pointer to the orphan MDT-object to be created
1235 * \param[in] type the orphan's type to be created
1237 * type "P": The orphan object to be created was a parent directory
1238 * of some MDT-object which linkEA shows that the @orphan
1239 * object is missing.
1241 * \see lfsck_layout_recreate_parent() for more types.
1243 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1245 * \retval positive number for repaired cases
1246 * \retval negative error number on failure
1248 static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
1249 struct lfsck_component *com,
1250 struct dt_object *orphan,
1252 struct lmv_mds_md_v1 *lmv)
1254 struct lfsck_thread_info *info = lfsck_env_info(env);
1255 struct lu_attr *la = &info->lti_la;
1256 struct dt_allocation_hint *hint = &info->lti_hint;
1257 struct dt_object_format *dof = &info->lti_dof;
1258 struct lu_name *cname = &info->lti_name2;
1259 struct dt_insert_rec *rec = &info->lti_dt_rec;
1260 struct lu_fid *tfid = &info->lti_fid;
1261 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1262 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1263 const struct lu_fid *pfid;
1264 struct lfsck_instance *lfsck = com->lc_lfsck;
1265 struct dt_device *dev = lfsck->li_bottom;
1266 struct dt_object *parent = NULL;
1267 struct dt_object *child = NULL;
1268 struct thandle *th = NULL;
1269 struct lustre_handle lh = { 0 };
1270 struct linkea_data ldata = { 0 };
1271 struct lu_buf linkea_buf;
1272 struct lu_buf lmv_buf;
1279 LASSERT(!dt_object_exists(orphan));
1280 LASSERT(!dt_object_remote(orphan));
1282 /* @orphan maybe not attached to lfsck->li_bottom */
1283 child = lfsck_object_find_by_dev(env, dev, cfid);
1285 GOTO(log, rc = PTR_ERR(child));
1287 cname->ln_name = NULL;
1288 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1291 /* Create .lustre/lost+found/MDTxxxx when needed. */
1292 if (unlikely(lfsck->li_lpf_obj == NULL)) {
1293 rc = lfsck_create_lpf(env, lfsck);
1298 parent = lfsck->li_lpf_obj;
1299 pfid = lfsck_dto2fid(parent);
1301 /* Hold update lock on the parent to prevent others to access. */
1302 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1303 MDS_INODELOCK_UPDATE, LCK_EX);
1308 namelen = snprintf(name, 31, DFID"-P-%d",
1310 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1311 (const struct dt_key *)name, BYPASS_CAPA);
1312 if (rc != 0 && rc != -ENOENT)
1316 cname->ln_name = name;
1317 cname->ln_namelen = namelen;
1319 memset(la, 0, sizeof(*la));
1320 la->la_mode = type | (S_ISDIR(type) ? 0700 : 0600);
1321 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1322 LA_ATIME | LA_MTIME | LA_CTIME;
1324 child->do_ops->do_ah_init(env, hint, parent, child,
1325 la->la_mode & S_IFMT);
1327 memset(dof, 0, sizeof(*dof));
1328 dof->dof_type = dt_mode_to_dft(type);
1330 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1334 rc = linkea_add_buf(&ldata, cname, pfid);
1338 th = dt_trans_create(env, dev);
1340 GOTO(unlock1, rc = PTR_ERR(th));
1342 rc = dt_declare_create(env, child, la, hint, dof, th);
1343 if (rc == 0 && S_ISDIR(type))
1344 rc = dt_declare_ref_add(env, child, th);
1350 lmv->lmv_magic = LMV_MAGIC;
1351 lmv->lmv_master_mdt_index = lfsck_dev_idx(dev);
1352 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1353 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1354 rc = dt_declare_xattr_set(env, child, &lmv_buf,
1355 XATTR_NAME_LMV, 0, th);
1360 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1361 ldata.ld_leh->leh_len);
1362 rc = dt_declare_xattr_set(env, child, &linkea_buf,
1363 XATTR_NAME_LINK, 0, th);
1367 rec->rec_type = type;
1368 rec->rec_fid = cfid;
1369 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1370 (const struct dt_key *)name, th);
1371 if (rc == 0 && S_ISDIR(type))
1372 rc = dt_declare_ref_add(env, parent, th);
1377 rc = dt_trans_start_local(env, dev, th);
1381 dt_write_lock(env, child, 0);
1382 rc = dt_create(env, child, la, hint, dof, th);
1386 if (S_ISDIR(type)) {
1387 if (unlikely(!dt_try_as_dir(env, child)))
1388 GOTO(unlock2, rc = -ENOTDIR);
1390 rec->rec_type = S_IFDIR;
1391 rec->rec_fid = cfid;
1392 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1393 (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
1397 rec->rec_fid = pfid;
1398 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1399 (const struct dt_key *)dotdot, th,
1404 rc = dt_ref_add(env, child, th);
1410 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, 0,
1416 rc = dt_xattr_set(env, child, &linkea_buf,
1417 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1418 dt_write_unlock(env, child);
1422 rec->rec_type = type;
1423 rec->rec_fid = cfid;
1424 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1425 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1426 if (rc == 0 && S_ISDIR(type)) {
1427 dt_write_lock(env, parent, 0);
1428 rc = dt_ref_add(env, parent, th);
1429 dt_write_unlock(env, parent);
1432 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1435 dt_write_unlock(env, child);
1438 dt_trans_stop(env, dev, th);
1441 lfsck_ibits_unlock(&lh, LCK_EX);
1444 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan locally for "
1445 "the object "DFID", name = %s, type %o: rc = %d\n",
1446 lfsck_lfsck2name(lfsck), PFID(cfid),
1447 cname->ln_name != NULL ? cname->ln_name : "<NULL>", type, rc);
1449 if (child != NULL && !IS_ERR(child))
1450 lfsck_object_put(env, child);
1456 * Create the specified orphan MDT-object.
1458 * For the case that the parent MDT-object stored in some MDT-object's
1459 * linkEA entry is lost, the LFSCK will re-create the parent object as
1460 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1461 * with the name: ${FID}-P-${conflict_version}.
1463 * \param[in] env pointer to the thread context
1464 * \param[in] com pointer to the lfsck component
1465 * \param[in] orphan pointer to the orphan MDT-object
1467 * type "P": The orphan object to be created was a parent directory
1468 * of some MDT-object which linkEA shows that the @orphan
1469 * object is missing.
1471 * \see lfsck_layout_recreate_parent() for more types.
1473 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1475 * \retval positive number for repaired cases
1476 * \retval 0 if needs to repair nothing
1477 * \retval negative error number on failure
1479 static int lfsck_namespace_create_orphan(const struct lu_env *env,
1480 struct lfsck_component *com,
1481 struct dt_object *orphan,
1482 struct lmv_mds_md_v1 *lmv)
1484 struct lfsck_namespace *ns = com->lc_file_ram;
1487 if (dt_object_remote(orphan))
1488 rc = lfsck_namespace_create_orphan_remote(env, com, orphan,
1491 rc = lfsck_namespace_create_orphan_local(env, com, orphan,
1495 ns->ln_flags |= LF_INCONSISTENT;
1501 * Remove the specified entry from the linkEA.
1503 * Locate the linkEA entry with the given @cname and @pfid, then
1504 * remove this entry or the other entries those are repeated with
1507 * \param[in] env pointer to the thread context
1508 * \param[in] com pointer to the lfsck component
1509 * \param[in] obj pointer to the dt_object to be handled
1510 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1511 * \param[in] cname the name for the child in the parent directory
1512 * \param[in] pfid the parent directory's FID for the linkEA
1513 * \param[in] next if true, then remove the first found linkEA
1514 * entry, and move the ldata->ld_lee to next entry
1516 * \retval positive number for repaired cases
1517 * \retval 0 if nothing to be repaired
1518 * \retval negative error number on failure
1520 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1521 struct lfsck_component *com,
1522 struct dt_object *obj,
1523 struct linkea_data *ldata,
1524 struct lu_name *cname,
1525 struct lu_fid *pfid,
1528 struct lfsck_instance *lfsck = com->lc_lfsck;
1529 struct dt_device *dev = lfsck->li_bottom;
1530 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1531 struct thandle *th = NULL;
1532 struct lustre_handle lh = { 0 };
1533 struct linkea_data ldata_new = { 0 };
1534 struct lu_buf linkea_buf;
1538 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1539 MDS_INODELOCK_UPDATE |
1540 MDS_INODELOCK_XATTR, LCK_EX);
1545 linkea_del_buf(ldata, cname);
1547 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1549 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1550 ldata->ld_leh->leh_len);
1553 th = dt_trans_create(env, dev);
1555 GOTO(unlock1, rc = PTR_ERR(th));
1557 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1558 XATTR_NAME_LINK, 0, th);
1562 rc = dt_trans_start_local(env, dev, th);
1566 dt_write_lock(env, obj, 0);
1567 if (unlikely(lfsck_is_dead_obj(obj)))
1568 GOTO(unlock2, rc = -ENOENT);
1570 rc = lfsck_links_read2(env, obj, &ldata_new);
1574 /* The specified linkEA entry has been removed by race. */
1575 rc = linkea_links_find(&ldata_new, cname, pfid);
1577 GOTO(unlock2, rc = 0);
1579 if (bk->lb_param & LPF_DRYRUN)
1580 GOTO(unlock2, rc = 1);
1583 linkea_del_buf(&ldata_new, cname);
1585 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1588 if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1589 dt_write_unlock(env, obj);
1590 dt_trans_stop(env, dev, th);
1591 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1592 ldata_new.ld_leh->leh_len);
1596 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1597 ldata_new.ld_leh->leh_len);
1598 rc = dt_xattr_set(env, obj, &linkea_buf,
1599 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1601 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1604 dt_write_unlock(env, obj);
1607 dt_trans_stop(env, dev, th);
1610 lfsck_ibits_unlock(&lh, LCK_EX);
1613 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1614 "for the object: "DFID", parent "DFID", name %.*s\n",
1615 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1616 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1620 struct lfsck_namespace *ns = com->lc_file_ram;
1622 ns->ln_flags |= LF_INCONSISTENT;
1629 * Conditionally remove the specified entry from the linkEA.
1631 * Take the parent lock firstly, then check whether the specified
1632 * name entry exists or not: if yes, do nothing; otherwise, call
1633 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1635 * \param[in] env pointer to the thread context
1636 * \param[in] com pointer to the lfsck component
1637 * \param[in] parent pointer to the parent directory
1638 * \param[in] child pointer to the child object that holds the linkEA
1639 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1640 * \param[in] cname the name for the child in the parent directory
1641 * \param[in] pfid the parent directory's FID for the linkEA
1643 * \retval positive number for repaired cases
1644 * \retval 0 if nothing to be repaired
1645 * \retval negative error number on failure
1647 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1648 struct lfsck_component *com,
1649 struct dt_object *parent,
1650 struct dt_object *child,
1651 struct linkea_data *ldata,
1652 struct lu_name *cname,
1653 struct lu_fid *pfid)
1655 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
1656 struct lustre_handle lh = { 0 };
1660 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
1661 MDS_INODELOCK_UPDATE, LCK_EX);
1665 dt_read_lock(env, parent, 0);
1666 if (unlikely(lfsck_is_dead_obj(parent))) {
1667 dt_read_unlock(env, parent);
1668 lfsck_ibits_unlock(&lh, LCK_EX);
1669 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1675 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1676 (const struct dt_key *)cname->ln_name,
1678 dt_read_unlock(env, parent);
1680 /* It is safe to release the ldlm lock, because when the logic come
1681 * here, we have got all the needed information above whether the
1682 * linkEA entry is valid or not. It is not important that others
1683 * may add new linkEA entry after the ldlm lock released. If other
1684 * has removed the specified linkEA entry by race, then it is OK,
1685 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1687 lfsck_ibits_unlock(&lh, LCK_EX);
1688 if (rc == -ENOENT) {
1689 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1698 /* The LFSCK just found some internal status of cross-MDTs
1699 * create operation. That is normal. */
1700 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1701 linkea_next_entry(ldata);
1706 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1713 * Conditionally replace name entry in the parent.
1715 * As required, the LFSCK may re-create the lost MDT-object for dangling
1716 * name entry, but such repairing may be wrong because of bad FID in the
1717 * name entry. As the LFSCK processing, the real MDT-object may be found,
1718 * then the LFSCK should check whether the former re-created MDT-object
1719 * has been modified or not, if not, then destroy it and update the name
1720 * entry in the parent to reference the real MDT-object.
1722 * \param[in] env pointer to the thread context
1723 * \param[in] com pointer to the lfsck component
1724 * \param[in] parent pointer to the parent directory
1725 * \param[in] child pointer to the MDT-object that may be the real
1726 * MDT-object corresponding to the name entry in parent
1727 * \param[in] cfid the current FID in the name entry
1728 * \param[in] cname contains the name of the child in the parent directory
1730 * \retval positive number for repaired cases
1731 * \retval 0 if nothing to be repaired
1732 * \retval negative error number on failure
1734 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1735 struct lfsck_component *com,
1736 struct dt_object *parent,
1737 struct dt_object *child,
1738 const struct lu_fid *cfid,
1739 const struct lu_name *cname)
1741 struct lfsck_thread_info *info = lfsck_env_info(env);
1742 struct lu_fid *tfid = &info->lti_fid5;
1743 struct lu_attr *la = &info->lti_la;
1744 struct dt_insert_rec *rec = &info->lti_dt_rec;
1745 struct lfsck_instance *lfsck = com->lc_lfsck;
1746 struct dt_device *dev = lfsck->li_next;
1747 const char *name = cname->ln_name;
1748 struct dt_object *obj = NULL;
1749 struct lustre_handle plh = { 0 };
1750 struct lustre_handle clh = { 0 };
1751 struct linkea_data ldata = { 0 };
1752 struct thandle *th = NULL;
1757 rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
1758 MDS_INODELOCK_UPDATE, LCK_EX);
1762 if (!fid_is_sane(cfid)) {
1767 obj = lfsck_object_find(env, lfsck, cfid);
1770 if (rc == -ENOENT) {
1778 if (!dt_object_exists(obj)) {
1783 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1784 (const struct dt_key *)name, BYPASS_CAPA);
1785 if (rc == -ENOENT) {
1793 /* Someone changed the name entry, cannot replace it. */
1794 if (!lu_fid_eq(cfid, tfid))
1797 /* lock the object to be destroyed. */
1798 rc = lfsck_ibits_lock(env, lfsck, obj, &clh,
1799 MDS_INODELOCK_UPDATE |
1800 MDS_INODELOCK_XATTR, LCK_EX);
1804 if (unlikely(lfsck_is_dead_obj(obj))) {
1809 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
1813 /* The object has been modified by other(s), or it is not created by
1814 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1815 if (la->la_ctime != 0)
1818 if (S_ISREG(la->la_mode)) {
1819 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
1821 /* If someone has created related OST-object(s),
1823 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1824 GOTO(log, rc = (rc > 0 ? 0 : rc));
1828 dt_read_lock(env, child, 0);
1829 rc = lfsck_links_read2(env, child, &ldata);
1830 dt_read_unlock(env, child);
1832 /* Someone changed the child, no need to replace. */
1839 rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(parent));
1840 /* Someone moved the child, no need to replace. */
1844 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1847 th = dt_trans_create(env, dev);
1849 GOTO(log, rc = PTR_ERR(th));
1852 rc = dt_declare_destroy(env, obj, th);
1857 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1861 rec->rec_type = S_IFDIR;
1862 rec->rec_fid = lfsck_dto2fid(child);
1863 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1864 (const struct dt_key *)name, th);
1868 rc = dt_trans_start(env, dev, th);
1873 rc = dt_destroy(env, obj, th);
1878 /* The old name entry maybe not exist. */
1879 dt_delete(env, parent, (const struct dt_key *)name, th,
1882 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1883 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1885 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1888 dt_trans_stop(env, dev, th);
1891 lfsck_ibits_unlock(&clh, LCK_EX);
1892 lfsck_ibits_unlock(&plh, LCK_EX);
1893 if (obj != NULL && !IS_ERR(obj))
1894 lfsck_object_put(env, obj);
1896 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1897 "object "DFID" because of conflict with the object "DFID
1898 " under the parent "DFID" with name %s: rc = %d\n",
1899 lfsck_lfsck2name(lfsck), PFID(cfid),
1900 PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)),
1907 * Overwrite the linkEA for the object with the given ldata.
1909 * The caller should take the ldlm lock before the calling.
1911 * \param[in] env pointer to the thread context
1912 * \param[in] com pointer to the lfsck component
1913 * \param[in] obj pointer to the dt_object to be handled
1914 * \param[in] ldata pointer to the new linkEA data
1916 * \retval positive number for repaired cases
1917 * \retval 0 if nothing to be repaired
1918 * \retval negative error number on failure
1920 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1921 struct lfsck_component *com,
1922 struct dt_object *obj,
1923 struct linkea_data *ldata)
1925 struct lfsck_instance *lfsck = com->lc_lfsck;
1926 struct dt_device *dev = lfsck->li_bottom;
1927 struct thandle *th = NULL;
1928 struct lu_buf linkea_buf;
1932 LASSERT(!dt_object_remote(obj));
1934 th = dt_trans_create(env, dev);
1936 GOTO(log, rc = PTR_ERR(th));
1938 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1939 ldata->ld_leh->leh_len);
1940 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1941 XATTR_NAME_LINK, 0, th);
1945 rc = dt_trans_start_local(env, dev, th);
1949 dt_write_lock(env, obj, 0);
1950 if (unlikely(lfsck_is_dead_obj(obj)))
1951 GOTO(unlock, rc = 0);
1953 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1954 GOTO(unlock, rc = 1);
1956 rc = dt_xattr_set(env, obj, &linkea_buf,
1957 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1959 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1962 dt_write_unlock(env, obj);
1965 dt_trans_stop(env, dev, th);
1968 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1969 "object "DFID": rc = %d\n",
1970 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1973 struct lfsck_namespace *ns = com->lc_file_ram;
1975 ns->ln_flags |= LF_INCONSISTENT;
1982 * Repair invalid name entry.
1984 * If the name entry contains invalid information, such as bad file type
1985 * or (and) corrupted object FID, then either remove the name entry or
1986 * udpate the name entry with the given (right) information.
1988 * \param[in] env pointer to the thread context
1989 * \param[in] com pointer to the lfsck component
1990 * \param[in] parent pointer to the parent directory
1991 * \param[in] child pointer to the object referenced by the name entry
1992 * \param[in] name the old name of the child under the parent directory
1993 * \param[in] name2 the new name of the child under the parent directory
1994 * \param[in] type the type claimed by the name entry
1995 * \param[in] update update the name entry if true; otherwise, remove it
1996 * \param[in] dec decrease the parent nlink count if true
1998 * \retval positive number for repaired successfully
1999 * \retval 0 if nothing to be repaired
2000 * \retval negative error number on failure
2002 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2003 struct lfsck_component *com,
2004 struct dt_object *parent,
2005 struct dt_object *child,
2006 const char *name, const char *name2,
2007 __u16 type, bool update, bool dec)
2009 struct lfsck_thread_info *info = lfsck_env_info(env);
2010 struct dt_insert_rec *rec = &info->lti_dt_rec;
2011 const struct lu_fid *cfid = lfsck_dto2fid(child);
2012 struct lu_fid *tfid = &info->lti_fid5;
2013 struct lfsck_instance *lfsck = com->lc_lfsck;
2014 struct dt_device *dev = lfsck->li_next;
2015 struct thandle *th = NULL;
2016 struct lustre_handle lh = { 0 };
2020 if (unlikely(!dt_try_as_dir(env, parent)))
2021 GOTO(log, rc = -ENOTDIR);
2023 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2024 MDS_INODELOCK_UPDATE, LCK_EX);
2028 th = dt_trans_create(env, dev);
2030 GOTO(unlock1, rc = PTR_ERR(th));
2032 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2037 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2038 rec->rec_fid = cfid;
2039 rc = dt_declare_insert(env, parent,
2040 (const struct dt_rec *)rec,
2041 (const struct dt_key *)name2, th);
2047 rc = dt_declare_ref_del(env, parent, th);
2052 rc = dt_trans_start(env, dev, th);
2056 dt_write_lock(env, parent, 0);
2057 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2058 (const struct dt_key *)name, BYPASS_CAPA);
2059 /* Someone has removed the bad name entry by race. */
2061 GOTO(unlock2, rc = 0);
2066 /* Someone has removed the bad name entry and reused it for other
2067 * object by race. */
2068 if (!lu_fid_eq(tfid, cfid))
2069 GOTO(unlock2, rc = 0);
2071 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2072 GOTO(unlock2, rc = 1);
2074 rc = dt_delete(env, parent, (const struct dt_key *)name, th,
2080 rc = dt_insert(env, parent,
2081 (const struct dt_rec *)rec,
2082 (const struct dt_key *)name2, th,
2089 rc = dt_ref_del(env, parent, th);
2094 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2097 dt_write_unlock(env, parent);
2100 dt_trans_stop(env, dev, th);
2102 /* We are not sure whether the child will become orphan or not.
2103 * Record it in the LFSCK trace file for further checking in
2104 * the second-stage scanning. */
2105 if (!update && !dec && rc == 0)
2106 lfsck_namespace_trace_update(env, com, cfid,
2107 LNTF_CHECK_LINKEA, true);
2110 lfsck_ibits_unlock(&lh, LCK_EX);
2113 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2114 "entry for: parent "DFID", child "DFID", name %s, type "
2115 "in name entry %o, type claimed by child %o. repair it "
2116 "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
2117 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2118 name, type, update ? lfsck_object_type(child) : 0,
2119 update ? "updating" : "removing", name2, rc);
2122 struct lfsck_namespace *ns = com->lc_file_ram;
2124 ns->ln_flags |= LF_INCONSISTENT;
2131 * Update the ".." name entry for the given object.
2133 * The object's ".." is corrupted, this function will update the ".." name
2134 * entry with the given pfid, and the linkEA with the given ldata.
2136 * The caller should take the ldlm lock before the calling.
2138 * \param[in] env pointer to the thread context
2139 * \param[in] com pointer to the lfsck component
2140 * \param[in] obj pointer to the dt_object to be handled
2141 * \param[in] pfid the new fid for the object's ".." name entry
2142 * \param[in] cname the name for the @obj in the parent directory
2144 * \retval positive number for repaired cases
2145 * \retval 0 if nothing to be repaired
2146 * \retval negative error number on failure
2148 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2149 struct lfsck_component *com,
2150 struct dt_object *obj,
2151 const struct lu_fid *pfid,
2152 struct lu_name *cname)
2154 struct lfsck_thread_info *info = lfsck_env_info(env);
2155 struct dt_insert_rec *rec = &info->lti_dt_rec;
2156 struct lfsck_instance *lfsck = com->lc_lfsck;
2157 struct dt_device *dev = lfsck->li_bottom;
2158 struct thandle *th = NULL;
2159 struct linkea_data ldata = { 0 };
2160 struct lu_buf linkea_buf;
2164 LASSERT(!dt_object_remote(obj));
2165 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2167 rc = linkea_data_new(&ldata, &info->lti_big_buf);
2171 rc = linkea_add_buf(&ldata, cname, pfid);
2175 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2176 ldata.ld_leh->leh_len);
2178 th = dt_trans_create(env, dev);
2180 GOTO(log, rc = PTR_ERR(th));
2182 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2186 rec->rec_type = S_IFDIR;
2187 rec->rec_fid = pfid;
2188 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2189 (const struct dt_key *)dotdot, th);
2193 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2194 XATTR_NAME_LINK, 0, th);
2198 rc = dt_trans_start_local(env, dev, th);
2202 dt_write_lock(env, obj, 0);
2203 if (unlikely(lfsck_is_dead_obj(obj)))
2204 GOTO(unlock, rc = 0);
2206 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2207 GOTO(unlock, rc = 1);
2209 /* The old ".." name entry maybe not exist. */
2210 dt_delete(env, obj, (const struct dt_key *)dotdot, th,
2213 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2214 (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
2218 rc = dt_xattr_set(env, obj, &linkea_buf,
2219 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2221 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2224 dt_write_unlock(env, obj);
2227 dt_trans_stop(env, dev, th);
2230 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2231 "the object "DFID", new parent "DFID": rc = %d\n",
2232 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2236 struct lfsck_namespace *ns = com->lc_file_ram;
2238 ns->ln_flags |= LF_INCONSISTENT;
2245 * Handle orphan @obj during Double Scan Directory.
2247 * Remove the @obj's current (invalid) linkEA entries, and insert
2248 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2249 * ${FID}-${PFID}-D-${conflict_version}
2251 * The caller should take the ldlm lock before the calling.
2253 * \param[in] env pointer to the thread context
2254 * \param[in] com pointer to the lfsck component
2255 * \param[in] obj pointer to the orphan object to be handled
2256 * \param[in] pfid the new fid for the object's ".." name entry
2257 * \param[in,out] lh ldlm lock handler for the given @obj
2258 * \param[out] type to tell the caller what the inconsistency is
2260 * \retval positive number for repaired cases
2261 * \retval 0 if nothing to be repaired
2262 * \retval negative error number on failure
2265 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2266 struct lfsck_component *com,
2267 struct dt_object *obj,
2268 const struct lu_fid *pfid,
2269 struct lustre_handle *lh,
2270 enum lfsck_namespace_inconsistency_type *type)
2272 struct lfsck_thread_info *info = lfsck_env_info(env);
2273 struct lfsck_namespace *ns = com->lc_file_ram;
2277 /* Remove the unrecognized linkEA. */
2278 rc = lfsck_namespace_links_remove(env, com, obj);
2279 lfsck_ibits_unlock(lh, LCK_EX);
2280 if (rc < 0 && rc != -ENODATA)
2283 *type = LNIT_MUL_REF;
2285 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2286 * ever tried to verify some remote MDT-object that resides on this
2287 * MDT, but this MDT failed to respond such request. So means there
2288 * may be some remote name entry on other MDT that references this
2289 * object with another name, so we cannot know whether this linkEA
2290 * is valid or not. So keep it there and maybe resolved when next
2292 if (ns->ln_flags & LF_INCOMPLETE)
2295 /* The unique linkEA is invalid, even if the ".." name entry may be
2296 * valid, we still cannot know via which name entry this directory
2297 * will be referenced. Then handle it as pure orphan. */
2298 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2299 "-"DFID, PFID(pfid));
2300 rc = lfsck_namespace_insert_orphan(env, com, obj,
2301 info->lti_tmpbuf, "D", NULL);
2307 * Double Scan Directory object for single linkEA entry case.
2309 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2310 * then check whether the name is in the namespace or not, if not, add the
2311 * missing name entry back to namespace. If the linkEA entry is invalid,
2312 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2315 * \param[in] env pointer to the thread context
2316 * \param[in] com pointer to the lfsck component
2317 * \param[in] child pointer to the directory to be double scanned
2318 * \param[in] pfid the FID corresponding to the ".." entry
2319 * \param[in] ldata pointer to the linkEA data for the given @child
2320 * \param[in,out] lh ldlm lock handler for the given @child
2321 * \param[out] type to tell the caller what the inconsistency is
2322 * \param[in] retry if found inconsistency, but the caller does not hold
2323 * ldlm lock on the @child, then set @retry as true
2325 * \retval positive number for repaired cases
2326 * \retval 0 if nothing to be repaired
2327 * \retval negative error number on failure
2330 lfsck_namespace_dsd_single(const struct lu_env *env,
2331 struct lfsck_component *com,
2332 struct dt_object *child,
2333 const struct lu_fid *pfid,
2334 struct linkea_data *ldata,
2335 struct lustre_handle *lh,
2336 enum lfsck_namespace_inconsistency_type *type,
2339 struct lfsck_thread_info *info = lfsck_env_info(env);
2340 struct lu_name *cname = &info->lti_name;
2341 const struct lu_fid *cfid = lfsck_dto2fid(child);
2342 struct lu_fid *tfid = &info->lti_fid3;
2343 struct lfsck_namespace *ns = com->lc_file_ram;
2344 struct lfsck_instance *lfsck = com->lc_lfsck;
2345 struct dt_object *parent = NULL;
2346 struct lmv_mds_md_v1 *lmv;
2350 lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid, info->lti_key);
2351 /* The unique linkEA entry with bad parent will be handled as orphan. */
2352 if (!fid_is_sane(tfid)) {
2353 if (!lustre_handle_is_used(lh) && retry != NULL)
2356 rc = lfsck_namespace_dsd_orphan(env, com, child,
2362 parent = lfsck_object_find_bottom(env, lfsck, tfid);
2364 GOTO(out, rc = PTR_ERR(parent));
2366 /* We trust the unique linkEA entry in spite of whether it matches the
2367 * ".." name entry or not. Because even if the linkEA entry is wrong
2368 * and the ".." name entry is right, we still cannot know via which
2369 * name entry the child will be referenced, since all known entries
2370 * have been verified during the first-stage scanning. */
2371 if (!dt_object_exists(parent)) {
2372 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2373 * has ever tried to verify some remote MDT-object that resides
2374 * on this MDT, but this MDT failed to respond such request. So
2375 * means there may be some remote name entry on other MDT that
2376 * references this object with another name, so we cannot know
2377 * whether this linkEA is valid or not. So keep it there and
2378 * maybe resolved when next LFSCK run. */
2379 if (ns->ln_flags & LF_INCOMPLETE)
2382 if (!lustre_handle_is_used(lh) && retry != NULL) {
2388 lfsck_ibits_unlock(lh, LCK_EX);
2391 lmv = &info->lti_lmv;
2392 rc = lfsck_read_stripe_lmv(env, child, lmv);
2393 if (rc != 0 && rc != -ENODATA)
2396 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2398 } else if (lfsck_shard_name_to_index(env,
2399 cname->ln_name, cname->ln_namelen,
2400 S_IFDIR, cfid) < 0) {
2401 /* It is an invalid name entry, we
2402 * cannot trust the parent also. */
2403 rc = lfsck_namespace_shrink_linkea(env, com, child,
2404 ldata, cname, tfid, true);
2408 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2409 "-"DFID, PFID(pfid));
2410 rc = lfsck_namespace_insert_orphan(env, com, child,
2411 info->lti_tmpbuf, "S", NULL);
2416 /* Create the lost parent as an orphan. */
2417 rc = lfsck_namespace_create_orphan(env, com, parent, lmv);
2419 /* Add the missing name entry to the parent. */
2420 rc = lfsck_namespace_insert_normal(env, com, parent,
2421 child, cname->ln_name);
2422 if (unlikely(rc == -EEXIST)) {
2423 /* Unfortunately, someone reused the name
2424 * under the parent by race. So we have
2425 * to remove the linkEA entry from
2426 * current child object. It means that the
2427 * LFSCK cannot recover the system
2428 * totally back to its original status,
2429 * but it is necessary to make the
2430 * current system to be consistent. */
2431 rc = lfsck_namespace_shrink_linkea(env,
2435 snprintf(info->lti_tmpbuf,
2436 sizeof(info->lti_tmpbuf),
2437 "-"DFID, PFID(pfid));
2438 rc = lfsck_namespace_insert_orphan(env,
2439 com, child, info->lti_tmpbuf,
2448 /* The unique linkEA entry with bad parent will be handled as orphan. */
2449 if (unlikely(!dt_try_as_dir(env, parent))) {
2450 if (!lustre_handle_is_used(lh) && retry != NULL)
2453 rc = lfsck_namespace_dsd_orphan(env, com, child,
2459 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2460 (const struct dt_key *)cname->ln_name, BYPASS_CAPA);
2461 if (rc == -ENOENT) {
2462 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2463 * has ever tried to verify some remote MDT-object that resides
2464 * on this MDT, but this MDT failed to respond such request. So
2465 * means there may be some remote name entry on other MDT that
2466 * references this object with another name, so we cannot know
2467 * whether this linkEA is valid or not. So keep it there and
2468 * maybe resolved when next LFSCK run. */
2469 if (ns->ln_flags & LF_INCOMPLETE)
2472 if (!lustre_handle_is_used(lh) && retry != NULL) {
2478 lfsck_ibits_unlock(lh, LCK_EX);
2479 rc = lfsck_namespace_check_name(env, parent, child, cname);
2486 /* It is an invalid name entry, drop it. */
2487 if (unlikely(rc > 0)) {
2488 rc = lfsck_namespace_shrink_linkea(env, com, child,
2489 ldata, cname, tfid, true);
2491 snprintf(info->lti_tmpbuf,
2492 sizeof(info->lti_tmpbuf),
2493 "-"DFID, PFID(pfid));
2494 rc = lfsck_namespace_insert_orphan(env, com,
2495 child, info->lti_tmpbuf, "D", NULL);
2501 /* Add the missing name entry back to the namespace. */
2502 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2504 if (unlikely(rc == -ESTALE))
2505 /* It may happen when the remote object has been
2506 * removed, but the local MDT is not aware of that. */
2509 if (unlikely(rc == -EEXIST)) {
2510 /* Unfortunately, someone reused the name under the
2511 * parent by race. So we have to remove the linkEA
2512 * entry from current child object. It means that the
2513 * LFSCK cannot recover the system totally back to
2514 * its original status, but it is necessary to make
2515 * the current system to be consistent.
2517 * It also may be because of the LFSCK found some
2518 * internal status of create operation. Under such
2519 * case, nothing to be done. */
2520 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2521 parent, child, ldata, cname, tfid);
2523 snprintf(info->lti_tmpbuf,
2524 sizeof(info->lti_tmpbuf),
2525 "-"DFID, PFID(pfid));
2526 rc = lfsck_namespace_insert_orphan(env, com,
2527 child, info->lti_tmpbuf, "D", NULL);
2537 if (!lu_fid_eq(tfid, cfid)) {
2538 if (!lustre_handle_is_used(lh) && retry != NULL) {
2544 lfsck_ibits_unlock(lh, LCK_EX);
2545 /* The name entry references another MDT-object that
2546 * may be created by the LFSCK for repairing dangling
2547 * name entry. Try to replace it. */
2548 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2551 rc = lfsck_namespace_dsd_orphan(env, com, child,
2557 /* The ".." name entry is wrong, update it. */
2558 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2559 if (!lustre_handle_is_used(lh) && retry != NULL) {
2565 *type = LNIT_UNMATCHED_PAIRS;
2566 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2567 lfsck_dto2fid(parent), cname);
2573 if (parent != NULL && !IS_ERR(parent))
2574 lfsck_object_put(env, parent);
2580 * Double Scan Directory object for multiple linkEA entries case.
2582 * The given @child has multiple linkEA entries. There is at most one linkEA
2583 * entry will be valid, all the others will be removed. Firstly, the function
2584 * will try to find out the linkEA entry for which the name entry exists under
2585 * the given parent (@pfid). If there is no linkEA entry that matches the given
2586 * ".." name entry, then tries to find out the first linkEA entry that both the
2587 * parent and the name entry exist to rebuild a new ".." name entry.
2589 * \param[in] env pointer to the thread context
2590 * \param[in] com pointer to the lfsck component
2591 * \param[in] child pointer to the directory to be double scanned
2592 * \param[in] pfid the FID corresponding to the ".." entry
2593 * \param[in] ldata pointer to the linkEA data for the given @child
2594 * \param[in,out] lh ldlm lock handler for the given @child
2595 * \param[out] type to tell the caller what the inconsistency is
2596 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2598 * \retval positive number for repaired cases
2599 * \retval 0 if nothing to be repaired
2600 * \retval negative error number on failure
2603 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2604 struct lfsck_component *com,
2605 struct dt_object *child,
2606 const struct lu_fid *pfid,
2607 struct linkea_data *ldata,
2608 struct lustre_handle *lh,
2609 enum lfsck_namespace_inconsistency_type *type,
2612 struct lfsck_thread_info *info = lfsck_env_info(env);
2613 struct lu_name *cname = &info->lti_name;
2614 const struct lu_fid *cfid = lfsck_dto2fid(child);
2615 struct lu_fid *tfid = &info->lti_fid3;
2616 struct lu_fid *pfid2 = &info->lti_fid4;
2617 struct lfsck_namespace *ns = com->lc_file_ram;
2618 struct lfsck_instance *lfsck = com->lc_lfsck;
2619 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2620 struct dt_object *parent = NULL;
2621 struct linkea_data ldata_new = { 0 };
2628 while (ldata->ld_lee != NULL) {
2629 lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid,
2631 /* Drop repeated linkEA entries. */
2632 lfsck_namespace_filter_linkea_entry(ldata, cname, tfid, true);
2633 /* Drop invalid linkEA entry. */
2634 if (!fid_is_sane(tfid)) {
2635 linkea_del_buf(ldata, cname);
2639 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2640 * then it is possible that: the directry object has ever
2641 * been lost, but its name entry was there. In the former
2642 * LFSCK run, during the first-stage scanning, the LFSCK
2643 * found the dangling name entry, but it did not recreate
2644 * the lost object, and when moved to the second-stage
2645 * scanning, some children objects of the lost directory
2646 * object were found, then the LFSCK recreated such lost
2647 * directory object as an orphan.
2649 * When the LFSCK runs again, if the dangling name is still
2650 * there, the LFSCK should move the orphan directory object
2651 * back to the normal namespace. */
2652 if (!lpf && !lu_fid_eq(pfid, tfid) && once) {
2653 linkea_next_entry(ldata);
2657 parent = lfsck_object_find_bottom(env, lfsck, tfid);
2659 RETURN(PTR_ERR(parent));
2661 if (!dt_object_exists(parent)) {
2662 lfsck_object_put(env, parent);
2663 if (ldata->ld_leh->leh_reccount > 1) {
2664 /* If it is NOT the last linkEA entry, then
2665 * there is still other chance to make the
2666 * child to be visible via other parent, then
2667 * remove this linkEA entry. */
2668 linkea_del_buf(ldata, cname);
2675 /* The linkEA entry with bad parent will be removed. */
2676 if (unlikely(!dt_try_as_dir(env, parent))) {
2677 lfsck_object_put(env, parent);
2678 linkea_del_buf(ldata, cname);
2682 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2683 (const struct dt_key *)cname->ln_name,
2685 *pfid2 = *lfsck_dto2fid(parent);
2686 if (rc == -ENOENT) {
2687 lfsck_object_put(env, parent);
2688 linkea_next_entry(ldata);
2693 lfsck_object_put(env, parent);
2698 if (lu_fid_eq(tfid, cfid)) {
2699 lfsck_object_put(env, parent);
2700 if (!lu_fid_eq(pfid, pfid2)) {
2701 *type = LNIT_UNMATCHED_PAIRS;
2702 rc = lfsck_namespace_repair_unmatched_pairs(env,
2703 com, child, pfid2, cname);
2709 /* It is the most common case that we find the
2710 * name entry corresponding to the linkEA entry
2711 * that matches the ".." name entry. */
2712 rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2716 rc = linkea_add_buf(&ldata_new, cname, pfid2);
2720 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2725 linkea_del_buf(ldata, cname);
2726 linkea_first_entry(ldata);
2727 /* There may be some invalid dangling name entries under
2728 * other parent directories, remove all of them. */
2729 while (ldata->ld_lee != NULL) {
2730 lfsck_namespace_unpack_linkea_entry(ldata,
2731 cname, tfid, info->lti_key);
2732 if (!fid_is_sane(tfid))
2735 parent = lfsck_object_find_bottom(env, lfsck,
2737 if (IS_ERR(parent)) {
2738 rc = PTR_ERR(parent);
2739 if (rc != -ENOENT &&
2740 bk->lb_param & LPF_FAILOUT)
2746 if (!dt_object_exists(parent)) {
2747 lfsck_object_put(env, parent);
2751 rc = lfsck_namespace_repair_dirent(env, com,
2752 parent, child, cname->ln_name,
2753 cname->ln_name, S_IFDIR, false, true);
2754 lfsck_object_put(env, parent);
2756 if (bk->lb_param & LPF_FAILOUT)
2765 linkea_del_buf(ldata, cname);
2768 ns->ln_dirent_repaired += count;
2773 lfsck_ibits_unlock(lh, LCK_EX);
2774 /* The name entry references another MDT-object that may be
2775 * created by the LFSCK for repairing dangling name entry.
2776 * Try to replace it. */
2777 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2779 lfsck_object_put(env, parent);
2786 linkea_del_buf(ldata, cname);
2789 if (ldata->ld_leh->leh_reccount == 1) {
2790 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2796 /* All linkEA entries are invalid and removed, then handle the @child
2798 if (ldata->ld_leh->leh_reccount == 0) {
2799 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2805 linkea_first_entry(ldata);
2806 /* If the dangling name entry for the orphan directory object has
2807 * been remvoed, then just check whether the directory object is
2808 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2814 /* There is no linkEA entry that matches the ".." name entry. Find
2815 * the first linkEA entry that both parent and name entry exist to
2816 * rebuild a new ".." name entry. */
2826 * Repair the object's nlink attribute.
2828 * If all the known name entries have been verified, then the object's hard
2829 * link attribute should match the object's linkEA entries count unless the
2830 * object's has too much hard link to be recorded in the linkEA. Such cases
2831 * should have been marked in the LFSCK trace file. Otherwise, trust the
2832 * linkEA to update the object's nlink attribute.
2834 * \param[in] env pointer to the thread context
2835 * \param[in] com pointer to the lfsck component
2836 * \param[in] obj pointer to the dt_object to be handled
2837 * \param[in,out] nlink pointer to buffer to object's hard lock count before
2838 * and after the repairing
2840 * \retval positive number for repaired cases
2841 * \retval 0 if nothing to be repaired
2842 * \retval negative error number on failure
2844 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2845 struct lfsck_component *com,
2846 struct dt_object *obj, __u32 *nlink)
2848 struct lfsck_thread_info *info = lfsck_env_info(env);
2849 struct lu_attr *la = &info->lti_la3;
2850 struct lu_fid *tfid = &info->lti_fid3;
2851 struct lfsck_namespace *ns = com->lc_file_ram;
2852 struct lfsck_instance *lfsck = com->lc_lfsck;
2853 struct dt_device *dev = lfsck->li_bottom;
2854 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2855 struct dt_object *child = NULL;
2856 struct thandle *th = NULL;
2857 struct linkea_data ldata = { 0 };
2858 struct lustre_handle lh = { 0 };
2864 LASSERT(!dt_object_remote(obj));
2865 LASSERT(S_ISREG(lfsck_object_type(obj)));
2867 child = lfsck_object_find_by_dev(env, dev, cfid);
2869 GOTO(log, rc = PTR_ERR(child));
2871 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
2872 MDS_INODELOCK_UPDATE |
2873 MDS_INODELOCK_XATTR, LCK_EX);
2877 th = dt_trans_create(env, dev);
2879 GOTO(log, rc = PTR_ERR(th));
2881 la->la_valid = LA_NLINK;
2882 rc = dt_declare_attr_set(env, child, la, th);
2886 rc = dt_trans_start_local(env, dev, th);
2890 dt_write_lock(env, child, 0);
2891 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2892 * ever tried to verify some remote MDT-object that resides on this
2893 * MDT, but this MDT failed to respond such request. So means there
2894 * may be some remote name entry on other MDT that references this
2895 * object with another name, so we cannot know whether this linkEA
2896 * is valid or not. So keep it there and maybe resolved when next
2898 if (ns->ln_flags & LF_INCOMPLETE)
2899 GOTO(unlock, rc = 0);
2901 fid_cpu_to_be(tfid, cfid);
2902 rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)&flags,
2903 (const struct dt_key *)tfid, BYPASS_CAPA);
2907 if (flags & LNTF_SKIP_NLINK)
2908 GOTO(unlock, rc = 0);
2910 rc = lfsck_links_read2(env, child, &ldata);
2912 GOTO(unlock, rc = 0);
2917 if (*nlink == ldata.ld_leh->leh_reccount)
2918 GOTO(unlock, rc = 0);
2920 la->la_nlink = *nlink = ldata.ld_leh->leh_reccount;
2921 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2922 GOTO(unlock, rc = 1);
2924 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
2926 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2929 dt_write_unlock(env, child);
2932 dt_trans_stop(env, dev, th);
2935 lfsck_ibits_unlock(&lh, LCK_EX);
2936 if (child != NULL && !IS_ERR(child))
2937 lfsck_object_put(env, child);
2939 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2940 "nlink count from %u to %u: rc = %d\n",
2941 lfsck_lfsck2name(lfsck), PFID(cfid), old, *nlink, rc);
2944 ns->ln_flags |= LF_INCONSISTENT;
2950 * Double scan the directory object for namespace LFSCK.
2952 * This function will verify the <parent, child> pairs in the namespace tree:
2953 * the parent references the child via some name entry that should be in the
2954 * child's linkEA entry, the child should back references the parent via its
2957 * The LFSCK will scan every linkEA entry in turn until find out the first
2958 * matched pairs. If found, then all other linkEA entries will be dropped.
2959 * If all the linkEA entries cannot match the ".." name entry, then there
2960 * are serveral possible cases:
2962 * 1) If there is only one linkEA entry, then trust it as long as the PFID
2963 * in the linkEA entry is valid.
2965 * 2) If there are multiple linkEA entries, then try to find the linkEA
2966 * that matches the ".." name entry. If found, then all other entries
2967 * are invalid; otherwise, it is quite possible that the ".." name entry
2968 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
2969 * entry according to the first valid linkEA entry (both the parent and
2970 * the name entry should exist).
2972 * 3) If the directory object has no (valid) linkEA entry, then the
2973 * directory object will be handled as pure orphan and inserted
2974 * in the .lustre/lost+found/MDTxxxx/ with the name:
2975 * ${self_FID}-${PFID}-D-${conflict_version}
2977 * \param[in] env pointer to the thread context
2978 * \param[in] com pointer to the lfsck component
2979 * \param[in] child pointer to the directory object to be handled
2980 * \param[in] flags to indicate the specical checking on the @child
2982 * \retval positive number for repaired cases
2983 * \retval 0 if nothing to be repaired
2984 * \retval negative error number on failure
2986 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
2987 struct lfsck_component *com,
2988 struct dt_object *child, __u8 flags)
2990 struct lfsck_thread_info *info = lfsck_env_info(env);
2991 const struct lu_fid *cfid = lfsck_dto2fid(child);
2992 struct lu_fid *pfid = &info->lti_fid2;
2993 struct lfsck_namespace *ns = com->lc_file_ram;
2994 struct lfsck_instance *lfsck = com->lc_lfsck;
2995 struct lustre_handle lh = { 0 };
2996 struct linkea_data ldata = { 0 };
2997 bool unknown = false;
3000 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3004 LASSERT(!dt_object_remote(child));
3006 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3007 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3008 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3009 "the namespace LFSCK, then the LFSCK cannot guarantee"
3010 "all the name entries have been verified in first-stage"
3011 "scanning. So have to skip orphan related handling for"
3012 "the directory object "DFID" with remote name entry\n",
3013 lfsck_lfsck2name(lfsck), PFID(cfid));
3018 if (unlikely(!dt_try_as_dir(env, child)))
3019 GOTO(out, rc = -ENOTDIR);
3021 /* We only take ldlm lock on the @child when required. When the
3022 * logic comes here for the first time, it is always false. */
3026 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3027 MDS_INODELOCK_UPDATE |
3028 MDS_INODELOCK_XATTR, LCK_EX);