Whamcloud - gitweb
LU-12616 obclass: fix MDS start/stop race
[fs/lustre-release.git] / lustre / lfsck / lfsck_namespace.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9
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.
15
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2013, 2017, Intel Corporation.
24  */
25 /*
26  * lustre/lfsck/lfsck_namespace.c
27  *
28  * Author: Fan, Yong <fan.yong@intel.com>
29  */
30
31 #define DEBUG_SUBSYSTEM S_LFSCK
32
33 #include <lu_object.h>
34 #include <dt_object.h>
35 #include <md_object.h>
36 #include <lustre_fid.h>
37 #include <lustre_lib.h>
38 #include <lustre_net.h>
39
40 #include "lfsck_internal.h"
41
42 #define LFSCK_NAMESPACE_MAGIC_V1        0xA0629D03
43 #define LFSCK_NAMESPACE_MAGIC_V2        0xA0621A0B
44 #define LFSCK_NAMESPACE_MAGIC_V3        0xA06249FF
45
46 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
47  * as the trace file magic. When downgrade to such old release, the old LFSCK
48  * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
49  * then it will reset the whole LFSCK, and will not cause start failure. The
50  * similar case will happen when upgrade from such old release. */
51 #define LFSCK_NAMESPACE_MAGIC           LFSCK_NAMESPACE_MAGIC_V3
52
53 enum lfsck_nameentry_check {
54         LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
55         LFSCK_NAMEENTRY_REMOVED         = 2, /* The entry has been removed. */
56         LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
57 };
58
59 static struct lfsck_namespace_req *
60 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
61                                    struct lfsck_assistant_object *lso,
62                                    struct lu_dirent *ent, __u16 type)
63 {
64         struct lfsck_namespace_req *lnr;
65         int                         size;
66
67         size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
68         OBD_ALLOC(lnr, size);
69         if (lnr == NULL)
70                 return ERR_PTR(-ENOMEM);
71
72         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
73         lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
74         lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
75         lnr->lnr_fid = ent->lde_fid;
76         lnr->lnr_dir_cookie = ent->lde_hash;
77         lnr->lnr_attr = ent->lde_attrs;
78         lnr->lnr_size = size;
79         lnr->lnr_type = type;
80         lnr->lnr_namelen = ent->lde_namelen;
81         memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
82
83         return lnr;
84 }
85
86 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
87                                                struct lfsck_assistant_req *lar)
88 {
89         struct lfsck_namespace_req *lnr =
90                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
91
92         if (lnr->lnr_lmv != NULL)
93                 lfsck_lmv_put(env, lnr->lnr_lmv);
94
95         lfsck_assistant_object_put(env, lar->lar_parent);
96         OBD_FREE(lnr, lnr->lnr_size);
97 }
98
99 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
100                                       struct lfsck_namespace *src)
101 {
102         dst->ln_magic = le32_to_cpu(src->ln_magic);
103         dst->ln_status = le32_to_cpu(src->ln_status);
104         dst->ln_flags = le32_to_cpu(src->ln_flags);
105         dst->ln_success_count = le32_to_cpu(src->ln_success_count);
106         dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
107         dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
108         dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
109         dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
110         dst->ln_time_last_checkpoint =
111                                 le64_to_cpu(src->ln_time_last_checkpoint);
112         lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
113                                  &src->ln_pos_latest_start);
114         lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
115                                  &src->ln_pos_last_checkpoint);
116         lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
117                                  &src->ln_pos_first_inconsistent);
118         dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
119         dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
120         dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
121         dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
122         dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
123         dst->ln_objs_repaired_phase2 =
124                                 le64_to_cpu(src->ln_objs_repaired_phase2);
125         dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
126         dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
127         fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
128                       &src->ln_fid_latest_scanned_phase2);
129         dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
130         dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
131         dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
132         dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
133         dst->ln_unknown_inconsistency =
134                                 le64_to_cpu(src->ln_unknown_inconsistency);
135         dst->ln_unmatched_pairs_repaired =
136                                 le64_to_cpu(src->ln_unmatched_pairs_repaired);
137         dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
138         dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
139         dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
140         dst->ln_lost_dirent_repaired =
141                                 le64_to_cpu(src->ln_lost_dirent_repaired);
142         dst->ln_striped_dirs_scanned =
143                                 le64_to_cpu(src->ln_striped_dirs_scanned);
144         dst->ln_striped_dirs_repaired =
145                                 le64_to_cpu(src->ln_striped_dirs_repaired);
146         dst->ln_striped_dirs_failed =
147                                 le64_to_cpu(src->ln_striped_dirs_failed);
148         dst->ln_striped_dirs_disabled =
149                                 le64_to_cpu(src->ln_striped_dirs_disabled);
150         dst->ln_striped_dirs_skipped =
151                                 le64_to_cpu(src->ln_striped_dirs_skipped);
152         dst->ln_striped_shards_scanned =
153                                 le64_to_cpu(src->ln_striped_shards_scanned);
154         dst->ln_striped_shards_repaired =
155                                 le64_to_cpu(src->ln_striped_shards_repaired);
156         dst->ln_striped_shards_failed =
157                                 le64_to_cpu(src->ln_striped_shards_failed);
158         dst->ln_striped_shards_skipped =
159                                 le64_to_cpu(src->ln_striped_shards_skipped);
160         dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
161         dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
162         dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
163         dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
164         dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
165         dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
166         dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
167         dst->ln_linkea_overflow_cleared =
168                                 le64_to_cpu(src->ln_linkea_overflow_cleared);
169         dst->ln_agent_entries_repaired =
170                                 le64_to_cpu(src->ln_agent_entries_repaired);
171 }
172
173 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
174                                       struct lfsck_namespace *src)
175 {
176         dst->ln_magic = cpu_to_le32(src->ln_magic);
177         dst->ln_status = cpu_to_le32(src->ln_status);
178         dst->ln_flags = cpu_to_le32(src->ln_flags);
179         dst->ln_success_count = cpu_to_le32(src->ln_success_count);
180         dst->ln_run_time_phase1 = cpu_to_le64(src->ln_run_time_phase1);
181         dst->ln_run_time_phase2 = cpu_to_le64(src->ln_run_time_phase2);
182         dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
183         dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
184         dst->ln_time_last_checkpoint =
185                                 cpu_to_le64(src->ln_time_last_checkpoint);
186         lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
187                                  &src->ln_pos_latest_start);
188         lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
189                                  &src->ln_pos_last_checkpoint);
190         lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
191                                  &src->ln_pos_first_inconsistent);
192         dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
193         dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
194         dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
195         dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
196         dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
197         dst->ln_objs_repaired_phase2 =
198                                 cpu_to_le64(src->ln_objs_repaired_phase2);
199         dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
200         dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
201         fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
202                       &src->ln_fid_latest_scanned_phase2);
203         dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
204         dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
205         dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
206         dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
207         dst->ln_unknown_inconsistency =
208                                 cpu_to_le64(src->ln_unknown_inconsistency);
209         dst->ln_unmatched_pairs_repaired =
210                                 cpu_to_le64(src->ln_unmatched_pairs_repaired);
211         dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
212         dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
213         dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
214         dst->ln_lost_dirent_repaired =
215                                 cpu_to_le64(src->ln_lost_dirent_repaired);
216         dst->ln_striped_dirs_scanned =
217                                 cpu_to_le64(src->ln_striped_dirs_scanned);
218         dst->ln_striped_dirs_repaired =
219                                 cpu_to_le64(src->ln_striped_dirs_repaired);
220         dst->ln_striped_dirs_failed =
221                                 cpu_to_le64(src->ln_striped_dirs_failed);
222         dst->ln_striped_dirs_disabled =
223                                 cpu_to_le64(src->ln_striped_dirs_disabled);
224         dst->ln_striped_dirs_skipped =
225                                 cpu_to_le64(src->ln_striped_dirs_skipped);
226         dst->ln_striped_shards_scanned =
227                                 cpu_to_le64(src->ln_striped_shards_scanned);
228         dst->ln_striped_shards_repaired =
229                                 cpu_to_le64(src->ln_striped_shards_repaired);
230         dst->ln_striped_shards_failed =
231                                 cpu_to_le64(src->ln_striped_shards_failed);
232         dst->ln_striped_shards_skipped =
233                                 cpu_to_le64(src->ln_striped_shards_skipped);
234         dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
235         dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
236         dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
237         dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
238         dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
239         dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
240         dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
241         dst->ln_linkea_overflow_cleared =
242                                 cpu_to_le64(src->ln_linkea_overflow_cleared);
243         dst->ln_agent_entries_repaired =
244                                 cpu_to_le64(src->ln_agent_entries_repaired);
245 }
246
247 static void lfsck_namespace_record_failure(const struct lu_env *env,
248                                            struct lfsck_instance *lfsck,
249                                            struct lfsck_namespace *ns)
250 {
251         struct lfsck_position pos;
252
253         ns->ln_items_failed++;
254         lfsck_pos_fill(env, lfsck, &pos, false);
255         if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
256             lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
257                 ns->ln_pos_first_inconsistent = pos;
258
259                 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
260                        "inconsistency at the pos [%llu, "DFID", %#llx]\n",
261                        lfsck_lfsck2name(lfsck),
262                        ns->ln_pos_first_inconsistent.lp_oit_cookie,
263                        PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
264                        ns->ln_pos_first_inconsistent.lp_dir_cookie);
265         }
266 }
267
268 /**
269  * Load the MDT bitmap from the lfsck_namespace trace file.
270  *
271  * \param[in] env       pointer to the thread context
272  * \param[in] com       pointer to the lfsck component
273  *
274  * \retval              0 for success
275  * \retval              negative error number on failure or data corruption
276  */
277 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
278                                        struct lfsck_component *com)
279 {
280         struct dt_object                *obj    = com->lc_obj;
281         struct lfsck_assistant_data     *lad    = com->lc_data;
282         struct lfsck_namespace          *ns     = com->lc_file_ram;
283         struct cfs_bitmap                       *bitmap = lad->lad_bitmap;
284         ssize_t                          size;
285         __u32                            nbits;
286         int                              rc;
287         ENTRY;
288
289         if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
290             ns->ln_bitmap_size)
291                 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
292         else
293                 nbits = ns->ln_bitmap_size;
294
295         if (unlikely(nbits < BITS_PER_LONG))
296                 nbits = BITS_PER_LONG;
297
298         if (nbits > bitmap->size) {
299                 __u32 new_bits = bitmap->size;
300                 struct cfs_bitmap *new_bitmap;
301
302                 while (new_bits < nbits)
303                         new_bits <<= 1;
304
305                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
306                 if (new_bitmap == NULL)
307                         RETURN(-ENOMEM);
308
309                 lad->lad_bitmap = new_bitmap;
310                 CFS_FREE_BITMAP(bitmap);
311                 bitmap = new_bitmap;
312         }
313
314         if (ns->ln_bitmap_size == 0) {
315                 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
316                 CFS_RESET_BITMAP(bitmap);
317
318                 RETURN(0);
319         }
320
321         size = (ns->ln_bitmap_size + 7) >> 3;
322         rc = dt_xattr_get(env, obj,
323                           lfsck_buf_get(env, bitmap->data, size),
324                           XATTR_NAME_LFSCK_BITMAP);
325         if (rc != size)
326                 RETURN(rc >= 0 ? -EINVAL : rc);
327
328         if (cfs_bitmap_check_empty(bitmap))
329                 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
330         else
331                 set_bit(LAD_INCOMPLETE, &lad->lad_flags);
332
333         RETURN(0);
334 }
335
336 /**
337  * Load namespace LFSCK statistics information from the trace file.
338  *
339  * \param[in] env       pointer to the thread context
340  * \param[in] com       pointer to the lfsck component
341  *
342  * \retval              0 for success
343  * \retval              negative error number on failure
344  */
345 static int lfsck_namespace_load(const struct lu_env *env,
346                                 struct lfsck_component *com)
347 {
348         int len = com->lc_file_size;
349         int rc;
350
351         rc = dt_xattr_get(env, com->lc_obj,
352                           lfsck_buf_get(env, com->lc_file_disk, len),
353                           XATTR_NAME_LFSCK_NAMESPACE);
354         if (rc == len) {
355                 struct lfsck_namespace *ns = com->lc_file_ram;
356
357                 lfsck_namespace_le_to_cpu(ns,
358                                 (struct lfsck_namespace *)com->lc_file_disk);
359                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
360                         CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
361                                "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
362                                ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
363                         rc = -ESTALE;
364                 } else {
365                         rc = 0;
366                 }
367         } else if (rc != -ENODATA) {
368                 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
369                        "expected = %d: rc = %d\n",
370                        lfsck_lfsck2name(com->lc_lfsck), len, rc);
371                 if (rc >= 0)
372                         rc = -ESTALE;
373         }
374
375         return rc;
376 }
377
378 static int lfsck_namespace_store(const struct lu_env *env,
379                                  struct lfsck_component *com)
380 {
381         struct dt_object                *obj    = com->lc_obj;
382         struct lfsck_instance           *lfsck  = com->lc_lfsck;
383         struct lfsck_namespace          *ns     = com->lc_file_ram;
384         struct lfsck_assistant_data     *lad    = com->lc_data;
385         struct dt_device                *dev    = lfsck_obj2dev(obj);
386         struct cfs_bitmap               *bitmap = NULL;
387         struct thandle                  *handle;
388         __u32                            nbits  = 0;
389         int                              len    = com->lc_file_size;
390         int                              rc;
391         ENTRY;
392
393         if (lad != NULL) {
394                 bitmap = lad->lad_bitmap;
395                 nbits = bitmap->size;
396
397                 LASSERT(nbits > 0);
398                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
399         }
400
401         ns->ln_bitmap_size = nbits;
402         lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
403                                   ns);
404         handle = dt_trans_create(env, dev);
405         if (IS_ERR(handle))
406                 GOTO(log, rc = PTR_ERR(handle));
407
408         rc = dt_declare_xattr_set(env, obj,
409                                   lfsck_buf_get(env, com->lc_file_disk, len),
410                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
411         if (rc != 0)
412                 GOTO(out, rc);
413
414         if (bitmap != NULL) {
415                 rc = dt_declare_xattr_set(env, obj,
416                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
417                                 XATTR_NAME_LFSCK_BITMAP, 0, handle);
418                 if (rc != 0)
419                         GOTO(out, rc);
420         }
421
422         rc = dt_trans_start_local(env, dev, handle);
423         if (rc != 0)
424                 GOTO(out, rc);
425
426         rc = dt_xattr_set(env, obj,
427                           lfsck_buf_get(env, com->lc_file_disk, len),
428                           XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
429         if (rc == 0 && bitmap != NULL)
430                 rc = dt_xattr_set(env, obj,
431                                   lfsck_buf_get(env, bitmap->data, nbits >> 3),
432                                   XATTR_NAME_LFSCK_BITMAP, 0, handle);
433
434         GOTO(out, rc);
435
436 out:
437         dt_trans_stop(env, dev, handle);
438
439 log:
440         if (rc != 0)
441                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
442                        lfsck_lfsck2name(lfsck), rc);
443         return rc;
444 }
445
446 static int lfsck_namespace_init(const struct lu_env *env,
447                                 struct lfsck_component *com)
448 {
449         struct lfsck_namespace *ns = com->lc_file_ram;
450         int rc;
451
452         memset(ns, 0, sizeof(*ns));
453         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
454         ns->ln_status = LS_INIT;
455         ns->ln_time_latest_reset = ktime_get_real_seconds();
456         down_write(&com->lc_sem);
457         rc = lfsck_namespace_store(env, com);
458         if (rc == 0)
459                 rc = lfsck_load_sub_trace_files(env, com,
460                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
461         up_write(&com->lc_sem);
462
463         return rc;
464 }
465
466 /**
467  * Update the namespace LFSCK trace file for the given @fid
468  *
469  * \param[in] env       pointer to the thread context
470  * \param[in] com       pointer to the lfsck component
471  * \param[in] fid       the fid which flags to be updated in the lfsck
472  *                      trace file
473  * \param[in] add       true if add new flags, otherwise remove flags
474  *
475  * \retval              0 for success or nothing to be done
476  * \retval              negative error number on failure
477  */
478 int lfsck_namespace_trace_update(const struct lu_env *env,
479                                  struct lfsck_component *com,
480                                  const struct lu_fid *fid,
481                                  const __u8 flags, bool add)
482 {
483         struct lfsck_instance   *lfsck  = com->lc_lfsck;
484         struct dt_object        *obj;
485         struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
486         struct dt_device        *dev;
487         struct thandle          *th     = NULL;
488         int                      idx;
489         int                      rc     = 0;
490         __u8                     old    = 0;
491         __u8                     new    = 0;
492         ENTRY;
493
494         LASSERT(flags != 0);
495
496         if (unlikely(!fid_is_sane(fid)))
497                 RETURN(0);
498
499         idx = lfsck_sub_trace_file_fid2idx(fid);
500         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
501         obj = com->lc_sub_trace_objs[idx].lsto_obj;
502         if (unlikely(obj == NULL)) {
503                 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
504                 RETURN(0);
505         }
506
507         lfsck_object_get(obj);
508         dev = lfsck_obj2dev(obj);
509         fid_cpu_to_be(key, fid);
510         rc = dt_lookup(env, obj, (struct dt_rec *)&old,
511                        (const struct dt_key *)key);
512         if (rc == -ENOENT) {
513                 if (!add)
514                         GOTO(unlock, rc = 0);
515
516                 old = 0;
517                 new = flags;
518         } else if (rc == 0) {
519                 if (add) {
520                         if ((old & flags) == flags)
521                                 GOTO(unlock, rc = 0);
522
523                         new = old | flags;
524                 } else {
525                         if ((old & flags) == 0)
526                                 GOTO(unlock, rc = 0);
527
528                         new = old & ~flags;
529                 }
530         } else {
531                 GOTO(log, rc);
532         }
533
534         th = dt_trans_create(env, dev);
535         if (IS_ERR(th))
536                 GOTO(log, rc = PTR_ERR(th));
537
538         if (old != 0) {
539                 rc = dt_declare_delete(env, obj,
540                                        (const struct dt_key *)key, th);
541                 if (rc != 0)
542                         GOTO(log, rc);
543         }
544
545         if (new != 0) {
546                 rc = dt_declare_insert(env, obj,
547                                        (const struct dt_rec *)&new,
548                                        (const struct dt_key *)key, th);
549                 if (rc != 0)
550                         GOTO(log, rc);
551         }
552
553         rc = dt_trans_start_local(env, dev, th);
554         if (rc != 0)
555                 GOTO(log, rc);
556
557         if (old != 0) {
558                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
559                 if (rc != 0)
560                         GOTO(log, rc);
561         }
562
563         if (new != 0) {
564                 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
565                                (const struct dt_key *)key, th);
566                 if (rc != 0)
567                         GOTO(log, rc);
568         }
569
570         GOTO(log, rc);
571
572 log:
573         if (th != NULL && !IS_ERR(th))
574                 dt_trans_stop(env, dev, th);
575
576         CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
577                "trace file, flags %x, old %x, new %x: rc = %d\n",
578                lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
579                (__u32)flags, (__u32)old, (__u32)new, rc);
580
581 unlock:
582         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
583         lfsck_object_put(env, obj);
584
585         return rc;
586 }
587
588 int lfsck_namespace_check_exist(const struct lu_env *env,
589                                 struct dt_object *dir,
590                                 struct dt_object *obj, const char *name)
591 {
592         struct lu_fid    *fid = &lfsck_env_info(env)->lti_fid;
593         int               rc;
594         ENTRY;
595
596         if (unlikely(lfsck_is_dead_obj(obj)))
597                 RETURN(LFSCK_NAMEENTRY_DEAD);
598
599         rc = dt_lookup(env, dir, (struct dt_rec *)fid,
600                        (const struct dt_key *)name);
601         if (rc == -ENOENT)
602                 RETURN(LFSCK_NAMEENTRY_REMOVED);
603
604         if (rc < 0)
605                 RETURN(rc);
606
607         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
608                 RETURN(LFSCK_NAMEENTRY_RECREATED);
609
610         RETURN(0);
611 }
612
613 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
614                                             struct dt_object *obj,
615                                             struct thandle *handle)
616 {
617         int rc;
618
619         /* For remote updating LINKEA, there may be further LFSCK action
620          * on remote MDT after the updating, so update the LINKEA ASAP. */
621         if (dt_object_remote(obj))
622                 handle->th_sync = 1;
623
624         /* For destroying all invalid linkEA entries. */
625         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
626         if (rc == 0)
627                 /* For insert new linkEA entry. */
628                 rc = dt_declare_xattr_set(env, obj,
629                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
630                         XATTR_NAME_LINK, 0, handle);
631         return rc;
632 }
633
634 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
635                        struct linkea_data *ldata, bool with_rec)
636 {
637         int rc;
638
639         if (ldata->ld_buf->lb_buf == NULL)
640                 return -ENOMEM;
641
642         if (!dt_object_exists(obj))
643                 return -ENOENT;
644
645         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
646         if (rc == -ERANGE) {
647                 /* Buf was too small, figure out what we need. */
648                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
649                 if (unlikely(rc == 0))
650                         return -ENODATA;
651
652                 if (rc < 0)
653                         return rc;
654
655                 lu_buf_realloc(ldata->ld_buf, rc);
656                 if (ldata->ld_buf->lb_buf == NULL)
657                         return -ENOMEM;
658
659                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
660         }
661
662         if (unlikely(rc == 0))
663                 return -ENODATA;
664
665         if (rc > 0) {
666                 if (with_rec)
667                         rc = linkea_init_with_rec(ldata);
668                 else
669                         rc = linkea_init(ldata);
670         }
671
672         return rc;
673 }
674
675 /**
676  * Remove linkEA for the given object.
677  *
678  * The caller should take the ldlm lock before the calling.
679  *
680  * \param[in] env       pointer to the thread context
681  * \param[in] com       pointer to the lfsck component
682  * \param[in] obj       pointer to the dt_object to be handled
683  *
684  * \retval              0 for repaired cases
685  * \retval              negative error number on failure
686  */
687 static int lfsck_namespace_links_remove(const struct lu_env *env,
688                                         struct lfsck_component *com,
689                                         struct dt_object *obj)
690 {
691         struct lfsck_instance           *lfsck  = com->lc_lfsck;
692         struct dt_device                *dev    = lfsck_obj2dev(obj);
693         struct thandle                  *th     = NULL;
694         int                              rc     = 0;
695         ENTRY;
696
697         LASSERT(dt_object_remote(obj) == 0);
698
699         th = dt_trans_create(env, dev);
700         if (IS_ERR(th))
701                 GOTO(log, rc = PTR_ERR(th));
702
703         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
704         if (rc != 0)
705                 GOTO(stop, rc);
706
707         rc = dt_trans_start_local(env, dev, th);
708         if (rc != 0)
709                 GOTO(stop, rc);
710
711         dt_write_lock(env, obj, 0);
712         if (unlikely(lfsck_is_dead_obj(obj)))
713                 GOTO(unlock, rc = -ENOENT);
714
715         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
716                 GOTO(unlock, rc = 0);
717
718         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
719
720         GOTO(unlock, rc);
721
722 unlock:
723         dt_write_unlock(env, obj);
724
725 stop:
726         dt_trans_stop(env, dev, th);
727
728 log:
729         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
730                "for the object "DFID": rc = %d\n",
731                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
732
733         if (rc == 0) {
734                 struct lfsck_namespace *ns = com->lc_file_ram;
735
736                 ns->ln_flags |= LF_INCONSISTENT;
737         }
738
739         return rc;
740 }
741
742 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
743                              struct linkea_data *ldata, struct thandle *handle)
744 {
745         struct lu_buf buf;
746         int rc;
747
748         lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
749
750 again:
751         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
752         if (unlikely(rc == -ENOSPC)) {
753                 rc = linkea_overflow_shrink(ldata);
754                 if (likely(rc > 0)) {
755                         buf.lb_len = rc;
756                         goto again;
757                 }
758         }
759
760         return rc;
761 }
762
763 static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
764 {
765         if (ldata->ld_reclen <= 0)
766                 return false;
767
768         if ((char *)ldata->ld_lee + ldata->ld_reclen >
769             (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
770                 return false;
771
772         return true;
773 }
774
775 static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
776                                          const struct lu_name *cname,
777                                          const struct lu_fid *pfid)
778 {
779         if (!linkea_reclen_is_valid(ldata))
780                 return false;
781
782         if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
783                 return false;
784
785         if (!fid_is_sane(pfid))
786                 return false;
787
788         return true;
789 }
790
791 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
792                                                struct lu_name *cname,
793                                                struct lu_fid *pfid,
794                                                char *buf, const int buflen)
795 {
796         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
797         if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
798                 return -EINVAL;
799
800         /* To guarantee the 'name' is terminated with '0'. */
801         memcpy(buf, cname->ln_name, cname->ln_namelen);
802         buf[cname->ln_namelen] = 0;
803         cname->ln_name = buf;
804
805         return 0;
806 }
807
808 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
809                                  const struct lu_name *lname)
810 {
811         LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
812
813         /* If current record is corrupted, all the subsequent
814          * records will be dropped. */
815         if (unlikely(!linkea_reclen_is_valid(ldata))) {
816                 void *ptr = ldata->ld_lee;
817
818                 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
819                 ldata->ld_leh->leh_reccount = 0;
820                 linkea_first_entry(ldata);
821                 while (ldata->ld_lee != NULL &&
822                        (char *)ldata->ld_lee < (char *)ptr) {
823                         int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
824                                      ldata->ld_lee->lee_reclen[1];
825
826                         ldata->ld_leh->leh_len += reclen;
827                         ldata->ld_leh->leh_reccount++;
828                         ldata->ld_lee = (struct link_ea_entry *)
829                                         ((char *)ldata->ld_lee + reclen);
830                 }
831
832                 ldata->ld_lee = NULL;
833         } else {
834                 linkea_del_buf(ldata, lname);
835         }
836 }
837
838 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
839                                                struct lu_name *cname,
840                                                struct lu_fid *pfid,
841                                                bool remove)
842 {
843         struct link_ea_entry    *oldlee;
844         int                      oldlen;
845         int                      repeated = 0;
846
847         oldlee = ldata->ld_lee;
848         oldlen = ldata->ld_reclen;
849         linkea_next_entry(ldata);
850         while (ldata->ld_lee != NULL) {
851                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
852                                    ldata->ld_lee->lee_reclen[1];
853                 if (unlikely(!linkea_reclen_is_valid(ldata))) {
854                         lfsck_linkea_del_buf(ldata, NULL);
855                         LASSERT(ldata->ld_lee == NULL);
856                 } else if (unlikely(ldata->ld_reclen == oldlen &&
857                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
858                         repeated++;
859                         if (!remove)
860                                 break;
861
862                         lfsck_linkea_del_buf(ldata, cname);
863                 } else {
864                         linkea_next_entry(ldata);
865                 }
866         }
867         ldata->ld_lee = oldlee;
868         ldata->ld_reclen = oldlen;
869
870         return repeated;
871 }
872
873 /**
874  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
875  *
876  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
877  * with the given type to generate the name, the detailed rules for name
878  * have been described as following.
879  *
880  * The function also generates the linkEA corresponding to the name entry
881  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
882  *
883  * \param[in] env       pointer to the thread context
884  * \param[in] com       pointer to the lfsck component
885  * \param[in] orphan    pointer to the orphan MDT-object
886  * \param[in] infix     additional information for the orphan name, such as
887  *                      the FID for original
888  * \param[in] type      the type for describing why the orphan MDT-object is
889  *                      created. The rules are as following:
890  *
891  *  type "D":           The MDT-object is a directory, it may knows its parent
892  *                      but because there is no valid linkEA, the LFSCK cannot
893  *                      know where to put it back to the namespace.
894  *  type "O":           The MDT-object has no linkEA, and there is no name
895  *                      entry that references the MDT-object.
896  *
897  *  type "S":           The orphan MDT-object is a shard of a striped directory
898  *
899  * \see lfsck_layout_recreate_parent() for more types.
900  *
901  * The orphan name will be like:
902  * ${FID}-${infix}-${type}-${conflict_version}
903  *
904  * \param[out] count    if some others inserted some linkEA entries by race,
905  *                      then return the linkEA entries count.
906  *
907  * \retval              positive number for repaired cases
908  * \retval              0 if needs to repair nothing
909  * \retval              negative error number on failure
910  */
911 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
912                                          struct lfsck_component *com,
913                                          struct dt_object *orphan,
914                                          const char *infix, const char *type,
915                                          int *count)
916 {
917         struct lfsck_thread_info        *info   = lfsck_env_info(env);
918         struct lu_name                  *cname  = &info->lti_name;
919         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
920         struct lu_attr                  *la     = &info->lti_la2;
921         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
922         const struct lu_fid             *pfid;
923         struct lu_fid                    tfid;
924         struct lfsck_instance           *lfsck  = com->lc_lfsck;
925         struct dt_device                *dev    = lfsck_obj2dev(orphan);
926         struct dt_object                *parent;
927         struct thandle                  *th     = NULL;
928         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
929         struct lustre_handle             clh    = { 0 };
930         struct linkea_data               ldata2 = { NULL };
931         struct lu_buf                    linkea_buf;
932         int                              namelen;
933         int                              idx    = 0;
934         int                              rc     = 0;
935         bool                             exist  = false;
936         ENTRY;
937
938         cname->ln_name = NULL;
939         if (unlikely(lfsck->li_lpf_obj == NULL))
940                 GOTO(log, rc = -ENXIO);
941
942         parent = lfsck->li_lpf_obj;
943         pfid = lfsck_dto2fid(parent);
944
945 again:
946         do {
947                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
948                                    PFID(cfid), infix, type, idx++);
949                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
950                                (const struct dt_key *)info->lti_key);
951                 if (rc != 0 && rc != -ENOENT)
952                         GOTO(log, rc);
953
954                 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
955                         exist = true;
956         } while (rc == 0 && !exist);
957
958         rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
959                         MDS_INODELOCK_UPDATE, LCK_PW);
960         if (rc != 0)
961                 GOTO(log, rc);
962
963         /* Re-check whether the name conflict with othrs after taken
964          * the ldlm lock. */
965         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
966                        (const struct dt_key *)info->lti_key);
967         if (rc == 0) {
968                 if (!lu_fid_eq(cfid, &tfid)) {
969                         exist = false;
970                         lfsck_unlock(pllh);
971                         goto again;
972                 }
973
974                 exist = true;
975         } else if (rc != -ENOENT) {
976                 GOTO(log, rc);
977         } else {
978                 exist = false;
979         }
980
981         cname->ln_name = info->lti_key;
982         cname->ln_namelen = namelen;
983         rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
984                               cname, pfid);
985         if (rc != 0)
986                 GOTO(log, rc);
987
988         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
989                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
990                               MDS_INODELOCK_XATTR, LCK_EX);
991         if (rc != 0)
992                 GOTO(log, rc);
993
994         lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
995                        ldata2.ld_leh->leh_len);
996         th = dt_trans_create(env, dev);
997         if (IS_ERR(th))
998                 GOTO(log, rc = PTR_ERR(th));
999
1000         if (S_ISDIR(lfsck_object_type(orphan))) {
1001                 rc = dt_declare_delete(env, orphan,
1002                                        (const struct dt_key *)dotdot, th);
1003                 if (rc != 0)
1004                         GOTO(stop, rc);
1005
1006                 rec->rec_type = S_IFDIR;
1007                 rec->rec_fid = pfid;
1008                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1009                                        (const struct dt_key *)dotdot, th);
1010                 if (rc != 0)
1011                         GOTO(stop, rc);
1012         }
1013
1014         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1015                                   XATTR_NAME_LINK, 0, th);
1016         if (rc != 0)
1017                 GOTO(stop, rc);
1018
1019         if (!exist) {
1020                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1021                 rec->rec_fid = cfid;
1022                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1023                                        (const struct dt_key *)cname->ln_name,
1024                                        th);
1025                 if (rc != 0)
1026                         GOTO(stop, rc);
1027
1028                 if (S_ISDIR(rec->rec_type)) {
1029                         rc = dt_declare_ref_add(env, parent, th);
1030                         if (rc != 0)
1031                                 GOTO(stop, rc);
1032                 }
1033         }
1034
1035         memset(la, 0, sizeof(*la));
1036         la->la_ctime = ktime_get_real_seconds();
1037         la->la_valid = LA_CTIME;
1038         rc = dt_declare_attr_set(env, orphan, la, th);
1039         if (rc != 0)
1040                 GOTO(stop, rc);
1041
1042         rc = dt_trans_start_local(env, dev, th);
1043         if (rc != 0)
1044                 GOTO(stop, rc);
1045
1046         dt_write_lock(env, orphan, 0);
1047         rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1048         if (likely(rc == -ENODATA || rc == -EINVAL)) {
1049                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1050                         GOTO(unlock, rc = 1);
1051
1052                 if (S_ISDIR(lfsck_object_type(orphan))) {
1053                         rc = dt_delete(env, orphan,
1054                                        (const struct dt_key *)dotdot, th);
1055                         if (rc != 0)
1056                                 GOTO(unlock, rc);
1057
1058                         rec->rec_type = S_IFDIR;
1059                         rec->rec_fid = pfid;
1060                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1061                                        (const struct dt_key *)dotdot, th);
1062                         if (rc != 0)
1063                                 GOTO(unlock, rc);
1064                 }
1065
1066                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1067                                   th);
1068         } else {
1069                 if (rc == 0 && count != NULL)
1070                         *count = ldata2.ld_leh->leh_reccount;
1071
1072                 GOTO(unlock, rc);
1073         }
1074         dt_write_unlock(env, orphan);
1075
1076         if (rc == 0 && !exist) {
1077                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1078                 rec->rec_fid = cfid;
1079                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1080                                (const struct dt_key *)cname->ln_name, th);
1081                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1082                         dt_write_lock(env, parent, 0);
1083                         rc = dt_ref_add(env, parent, th);
1084                         dt_write_unlock(env, parent);
1085                 }
1086         }
1087
1088         if (rc == 0)
1089                 rc = dt_attr_set(env, orphan, la, th);
1090
1091         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1092
1093 unlock:
1094         dt_write_unlock(env, orphan);
1095
1096 stop:
1097         dt_trans_stop(env, dev, th);
1098
1099 log:
1100         lfsck_ibits_unlock(&clh, LCK_EX);
1101         lfsck_unlock(pllh);
1102         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1103                "object "DFID", name = %s: rc = %d\n",
1104                lfsck_lfsck2name(lfsck), PFID(cfid),
1105                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1106
1107         if (rc != 0) {
1108                 struct lfsck_namespace *ns = com->lc_file_ram;
1109
1110                 ns->ln_flags |= LF_INCONSISTENT;
1111         }
1112
1113         return rc;
1114 }
1115
1116 static int lfsck_lmv_set(const struct lu_env *env,
1117                          struct lfsck_instance *lfsck,
1118                          struct dt_object *obj,
1119                          struct lmv_mds_md_v1 *lmv)
1120 {
1121         struct dt_device *dev = lfsck->li_next;
1122         struct thandle *th = NULL;
1123         struct lu_buf buf = { lmv, sizeof(*lmv) };
1124         int rc;
1125
1126         ENTRY;
1127
1128         th = dt_trans_create(env, dev);
1129         if (IS_ERR(th))
1130                 RETURN(PTR_ERR(th));
1131
1132         rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV".set", 0, th);
1133         if (rc)
1134                 GOTO(stop, rc);
1135
1136         rc = dt_trans_start_local(env, dev, th);
1137         if (rc != 0)
1138                 GOTO(stop, rc);
1139
1140         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV".set", 0, th);
1141         if (rc)
1142                 GOTO(stop, rc);
1143
1144         EXIT;
1145 stop:
1146         dt_trans_stop(env, dev, th);
1147
1148         return rc;
1149 }
1150
1151 static int lfsck_lmv_delete(const struct lu_env *env,
1152                             struct lfsck_instance *lfsck,
1153                             struct dt_object *obj)
1154 {
1155         struct dt_device *dev = lfsck->li_next;
1156         struct thandle *th = NULL;
1157         int rc;
1158
1159         ENTRY;
1160
1161         th = dt_trans_create(env, dev);
1162         if (IS_ERR(th))
1163                 RETURN(PTR_ERR(th));
1164
1165         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
1166         if (rc)
1167                 GOTO(stop, rc);
1168
1169         rc = dt_trans_start_local(env, dev, th);
1170         if (rc != 0)
1171                 GOTO(stop, rc);
1172
1173         rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
1174         if (rc)
1175                 GOTO(stop, rc);
1176
1177         EXIT;
1178 stop:
1179         dt_trans_stop(env, dev, th);
1180
1181         return rc;
1182 }
1183
1184 static inline int lfsck_object_is_shard(const struct lu_env *env,
1185                                         struct lfsck_instance *lfsck,
1186                                         struct dt_object *obj,
1187                                         const struct lu_name *lname)
1188 {
1189         struct lfsck_thread_info *info = lfsck_env_info(env);
1190         struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1191         int rc;
1192
1193         rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
1194                                        lfsck_object_type(obj),
1195                                        lfsck_dto2fid(obj));
1196         if (rc < 0)
1197                 return 0;
1198
1199         rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
1200         if (rc == -ENODATA)
1201                 return 0;
1202
1203         if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
1204                 return 1;
1205
1206         return rc;
1207 }
1208
1209 /**
1210  * Add the specified name entry back to namespace.
1211  *
1212  * If there is a linkEA entry that back references a name entry under
1213  * some parent directory, but such parent directory does not have the
1214  * claimed name entry. On the other hand, the linkEA entries count is
1215  * not larger than the MDT-object's hard link count. Under such case,
1216  * it is quite possible that the name entry is lost. Then the LFSCK
1217  * should add the name entry back to the namespace.
1218  *
1219  * If \a child is shard, which means \a parent is a striped directory,
1220  * if \a parent has LMV, we need to delete it before insertion because
1221  * now parent's striping is broken and can't be parsed correctly.
1222  *
1223  * \param[in] env       pointer to the thread context
1224  * \param[in] com       pointer to the lfsck component
1225  * \param[in] parent    pointer to the directory under which the name entry
1226  *                      will be inserted into
1227  * \param[in] child     pointer to the object referenced by the name entry
1228  *                      that to be inserted into the parent
1229  * \param[in] lname     the name for the child in the parent directory
1230  *
1231  * \retval              positive number for repaired cases
1232  * \retval              0 if nothing to be repaired
1233  * \retval              negative error number on failure
1234  */
1235 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1236                                          struct lfsck_component *com,
1237                                          struct dt_object *parent,
1238                                          struct dt_object *child,
1239                                          const struct lu_name *lname)
1240 {
1241         struct lfsck_thread_info *info = lfsck_env_info(env);
1242         struct lu_attr *la = &info->lti_la;
1243         struct dt_insert_rec *rec = &info->lti_dt_rec;
1244         struct lfsck_instance *lfsck = com->lc_lfsck;
1245         /* The child and its name may be on different MDTs. */
1246         const struct lu_fid *pfid = lfsck_dto2fid(parent);
1247         const struct lu_fid *cfid = lfsck_dto2fid(child);
1248         struct dt_device *dev = lfsck->li_next;
1249         struct thandle *th = NULL;
1250         struct lfsck_lock_handle *llh = &info->lti_llh;
1251         struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1252         struct lu_buf buf = { lmv, sizeof(*lmv) };
1253         /* whether parent's LMV is deleted before insertion */
1254         bool parent_lmv_deleted = false;
1255         /* whether parent's LMV is missing */
1256         bool parent_lmv_lost = false;
1257         int rc = 0;
1258
1259         ENTRY;
1260
1261         /* @parent/@child may be based on lfsck->li_bottom,
1262          * but here we need the object based on the lfsck->li_next. */
1263
1264         parent = lfsck_object_locate(dev, parent);
1265         if (IS_ERR(parent))
1266                 GOTO(log, rc = PTR_ERR(parent));
1267
1268         child = lfsck_object_locate(dev, child);
1269         if (IS_ERR(child))
1270                 GOTO(log, rc = PTR_ERR(child));
1271
1272         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1273                 GOTO(log, rc = 1);
1274
1275         rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
1276                         MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1277                         MDS_INODELOCK_XATTR, LCK_EX);
1278         if (rc)
1279                 GOTO(log, rc);
1280
1281         rc = lfsck_object_is_shard(env, lfsck, child, lname);
1282         if (rc < 0)
1283                 GOTO(unlock, rc);
1284
1285         if (rc == 1) {
1286                 rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
1287                 if (!rc) {
1288                         /*
1289                          * To add a shard, we need to convert parent to a
1290                          * plain directory by deleting its LMV, and after
1291                          * insertion set it back.
1292                          */
1293                         rc = lfsck_lmv_delete(env, lfsck, parent);
1294                         if (rc)
1295                                 GOTO(unlock, rc);
1296                         parent_lmv_deleted = true;
1297                         lmv->lmv_layout_version++;
1298                         lfsck_lmv_header_cpu_to_le(lmv, lmv);
1299                 } else if (rc == -ENODATA) {
1300                         struct lu_seq_range *range = &info->lti_range;
1301                         struct seq_server_site *ss = lfsck_dev_site(lfsck);
1302
1303                         rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
1304                         if (rc)
1305                                 GOTO(unlock, rc);
1306
1307                         fld_range_set_mdt(range);
1308                         rc = fld_server_lookup(env, ss->ss_server_fld,
1309                                        fid_seq(lfsck_dto2fid(parent)), range);
1310                         if (rc)
1311                                 GOTO(unlock, rc);
1312
1313                         parent_lmv_lost = true;
1314                         lmv->lmv_magic = LMV_MAGIC;
1315                         lmv->lmv_master_mdt_index = range->lsr_index;
1316                         lmv->lmv_layout_version++;
1317                         lfsck_lmv_header_cpu_to_le(lmv, lmv);
1318                 } else {
1319                         GOTO(unlock, rc);
1320                 }
1321         }
1322
1323         if (unlikely(!dt_try_as_dir(env, parent)))
1324                 GOTO(unlock, rc = -ENOTDIR);
1325
1326         th = dt_trans_create(env, dev);
1327         if (IS_ERR(th))
1328                 GOTO(unlock, rc = PTR_ERR(th));
1329
1330         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1331         rec->rec_fid = cfid;
1332         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1333                                (const struct dt_key *)lname->ln_name, th);
1334         if (rc != 0)
1335                 GOTO(stop, rc);
1336
1337         if (S_ISDIR(rec->rec_type)) {
1338                 rc = dt_declare_ref_add(env, parent, th);
1339                 if (rc != 0)
1340                         GOTO(stop, rc);
1341         }
1342
1343         if (parent_lmv_lost) {
1344                 rc = dt_declare_xattr_set(env, parent, &buf,
1345                                           XATTR_NAME_LMV".set", 0, th);
1346                 if (rc)
1347                         GOTO(stop, rc);
1348         }
1349
1350         la->la_ctime = ktime_get_real_seconds();
1351         la->la_valid = LA_CTIME;
1352         rc = dt_declare_attr_set(env, parent, la, th);
1353         if (rc != 0)
1354                 GOTO(stop, rc);
1355
1356         rc = dt_declare_attr_set(env, child, la, th);
1357         if (rc != 0)
1358                 GOTO(stop, rc);
1359
1360         rc = dt_trans_start_local(env, dev, th);
1361         if (rc != 0)
1362                 GOTO(stop, rc);
1363
1364         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1365                        (const struct dt_key *)lname->ln_name, th);
1366         if (rc != 0)
1367                 GOTO(stop, rc);
1368
1369         if (S_ISDIR(rec->rec_type)) {
1370                 dt_write_lock(env, parent, 0);
1371                 rc = dt_ref_add(env, parent, th);
1372                 dt_write_unlock(env, parent);
1373                 if (rc != 0)
1374                         GOTO(stop, rc);
1375         }
1376
1377         if (parent_lmv_lost) {
1378                 rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV".set", 0,
1379                                   th);
1380                 if (rc)
1381                         GOTO(stop, rc);
1382         }
1383
1384         rc = dt_attr_set(env, parent, la, th);
1385         if (rc != 0)
1386                 GOTO(stop, rc);
1387
1388         rc = dt_attr_set(env, child, la, th);
1389
1390         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1391
1392 stop:
1393         dt_trans_stop(env, dev, th);
1394
1395 unlock:
1396         if (parent_lmv_deleted)
1397                 lfsck_lmv_set(env, lfsck, parent, lmv);
1398
1399         lfsck_unlock(llh);
1400
1401 log:
1402         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1403                "the name %s and type %o to the parent "DFID": rc = %d\n",
1404                lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
1405                lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1406
1407         if (rc != 0) {
1408                 struct lfsck_namespace *ns = com->lc_file_ram;
1409
1410                 ns->ln_flags |= LF_INCONSISTENT;
1411                 if (rc > 0)
1412                         ns->ln_lost_dirent_repaired++;
1413         }
1414
1415         return rc;
1416 }
1417
1418 /**
1419  * Create the specified orphan directory.
1420  *
1421  * For the case that the parent MDT-object stored in some MDT-object's
1422  * linkEA entry is lost, the LFSCK will re-create the parent object as
1423  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1424  * with the name ${FID}-P-${conflict_version}.
1425  *
1426  * \param[in] env       pointer to the thread context
1427  * \param[in] com       pointer to the lfsck component
1428  * \param[in] orphan    pointer to the orphan MDT-object to be created
1429  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1430  *
1431  * \retval              positive number for repaired cases
1432  * \retval              negative error number on failure
1433  */
1434 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1435                                              struct lfsck_component *com,
1436                                              struct dt_object *orphan,
1437                                              struct lmv_mds_md_v1 *lmv)
1438 {
1439         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1440         struct lu_attr                  *la     = &info->lti_la;
1441         struct dt_allocation_hint       *hint   = &info->lti_hint;
1442         struct dt_object_format         *dof    = &info->lti_dof;
1443         struct lu_name                  *cname  = &info->lti_name2;
1444         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1445         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1446         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1447         struct lu_fid                    tfid;
1448         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1449         struct lfsck_namespace          *ns     = com->lc_file_ram;
1450         struct dt_device                *dev    = lfsck_obj2dev(orphan);
1451         struct dt_object                *parent = NULL;
1452         struct thandle                  *th     = NULL;
1453         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1454         struct linkea_data               ldata  = { NULL };
1455         struct lu_buf                    linkea_buf;
1456         struct lu_buf                    lmv_buf;
1457         char                             name[32];
1458         int                              namelen;
1459         int                              idx    = 0;
1460         int                              rc     = 0;
1461         int                              rc1    = 0;
1462         ENTRY;
1463
1464         LASSERT(!dt_object_exists(orphan));
1465
1466         cname->ln_name = NULL;
1467         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1468                 GOTO(log, rc = 1);
1469
1470         if (dt_object_remote(orphan)) {
1471                 LASSERT(lfsck->li_lpf_root_obj != NULL);
1472
1473                 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1474                 if (idx < 0)
1475                         GOTO(log, rc = idx);
1476
1477                 snprintf(name, 8, "MDT%04x", idx);
1478                 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1479                                (struct dt_rec *)&tfid,
1480                                (const struct dt_key *)name);
1481                 if (rc != 0)
1482                         GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1483
1484                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1485                 if (IS_ERR(parent))
1486                         GOTO(log, rc = PTR_ERR(parent));
1487
1488                 if (unlikely(!dt_try_as_dir(env, parent)))
1489                         GOTO(log, rc = -ENOTDIR);
1490         } else {
1491                 if (unlikely(lfsck->li_lpf_obj == NULL))
1492                         GOTO(log, rc = -ENXIO);
1493
1494                 parent = lfsck->li_lpf_obj;
1495         }
1496
1497         dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1498         if (IS_ERR(dev))
1499                 GOTO(log, rc = PTR_ERR(dev));
1500
1501         idx = 0;
1502
1503 again:
1504         do {
1505                 namelen = snprintf(name, 31, DFID"-P-%d",
1506                                    PFID(cfid), idx++);
1507                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1508                                (const struct dt_key *)name);
1509                 if (rc != 0 && rc != -ENOENT)
1510                         GOTO(log, rc);
1511         } while (rc == 0);
1512
1513         rc = lfsck_lock(env, lfsck, parent, name, llh,
1514                         MDS_INODELOCK_UPDATE, LCK_PW);
1515         if (rc != 0)
1516                 GOTO(log, rc);
1517
1518         /* Re-check whether the name conflict with othrs after taken
1519          * the ldlm lock. */
1520         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1521                        (const struct dt_key *)name);
1522         if (unlikely(rc == 0)) {
1523                 lfsck_unlock(llh);
1524                 goto again;
1525         }
1526
1527         if (rc != -ENOENT)
1528                 GOTO(unlock1, rc);
1529
1530         cname->ln_name = name;
1531         cname->ln_namelen = namelen;
1532
1533         memset(la, 0, sizeof(*la));
1534         la->la_mode = S_IFDIR | 0700;
1535         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1536                        LA_ATIME | LA_MTIME | LA_CTIME;
1537
1538         orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1539                                    la->la_mode & S_IFMT);
1540
1541         memset(dof, 0, sizeof(*dof));
1542         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1543
1544         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1545                               cname, lfsck_dto2fid(parent));
1546         if (rc != 0)
1547                 GOTO(unlock1, rc);
1548
1549         th = dt_trans_create(env, dev);
1550         if (IS_ERR(th))
1551                 GOTO(unlock1, rc = PTR_ERR(th));
1552
1553         /* Sync the remote transaction to guarantee that the subsequent
1554          * lock against the @orphan can find the @orphan in time. */
1555         if (dt_object_remote(orphan))
1556                 th->th_sync = 1;
1557
1558         rc = dt_declare_create(env, orphan, la, hint, dof, th);
1559         if (rc != 0)
1560                 GOTO(stop, rc);
1561
1562         if (unlikely(!dt_try_as_dir(env, orphan)))
1563                 GOTO(stop, rc = -ENOTDIR);
1564
1565         rc = dt_declare_ref_add(env, orphan, th);
1566         if (rc != 0)
1567                 GOTO(stop, rc);
1568
1569         rec->rec_type = S_IFDIR;
1570         rec->rec_fid = cfid;
1571         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1572                                (const struct dt_key *)dot, th);
1573         if (rc != 0)
1574                 GOTO(stop, rc);
1575
1576         rec->rec_fid = lfsck_dto2fid(parent);
1577         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1578                                (const struct dt_key *)dotdot, th);
1579         if (rc != 0)
1580                 GOTO(stop, rc);
1581
1582         if (lmv != NULL) {
1583                 lmv->lmv_magic = LMV_MAGIC;
1584                 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1585                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1586                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1587                 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1588                                           XATTR_NAME_LMV, 0, th);
1589                 if (rc != 0)
1590                         GOTO(stop, rc);
1591         }
1592
1593         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1594                        ldata.ld_leh->leh_len);
1595         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1596                                   XATTR_NAME_LINK, 0, th);
1597         if (rc != 0)
1598                 GOTO(stop, rc);
1599
1600         rec->rec_fid = cfid;
1601         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1602                                (const struct dt_key *)name, th);
1603         if (rc == 0)
1604                 rc = dt_declare_ref_add(env, parent, th);
1605
1606         if (rc != 0)
1607                 GOTO(stop, rc);
1608
1609         rc = dt_trans_start_local(env, dev, th);
1610         if (rc != 0)
1611                 GOTO(stop, rc);
1612
1613         dt_write_lock(env, orphan, 0);
1614         rc = dt_create(env, orphan, la, hint, dof, th);
1615         if (rc != 0)
1616                 GOTO(unlock2, rc);
1617
1618         rc = dt_ref_add(env, orphan, th);
1619         if (rc != 0)
1620                 GOTO(unlock2, rc);
1621
1622         rec->rec_fid = cfid;
1623         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1624                        (const struct dt_key *)dot, th);
1625         if (rc != 0)
1626                 GOTO(unlock2, rc);
1627
1628         rec->rec_fid = lfsck_dto2fid(parent);
1629         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1630                        (const struct dt_key *)dotdot, th);
1631         if (rc != 0)
1632                 GOTO(unlock2, rc);
1633
1634         if (lmv != NULL) {
1635                 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1636                 if (rc != 0)
1637                         GOTO(unlock2, rc);
1638         }
1639
1640         rc = dt_xattr_set(env, orphan, &linkea_buf,
1641                           XATTR_NAME_LINK, 0, th);
1642         dt_write_unlock(env, orphan);
1643         if (rc != 0)
1644                 GOTO(stop, rc);
1645
1646         rec->rec_fid = cfid;
1647         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1648                        (const struct dt_key *)name, th);
1649         if (rc == 0) {
1650                 dt_write_lock(env, parent, 0);
1651                 rc = dt_ref_add(env, parent, th);
1652                 dt_write_unlock(env, parent);
1653         }
1654
1655         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1656
1657 unlock2:
1658         dt_write_unlock(env, orphan);
1659
1660 stop:
1661         rc1 = dt_trans_stop(env, dev, th);
1662         if (rc1 != 0 && rc > 0)
1663                 rc = rc1;
1664
1665 unlock1:
1666         lfsck_unlock(llh);
1667
1668 log:
1669         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1670                "the object "DFID", name = %s: rc = %d\n",
1671                lfsck_lfsck2name(lfsck), PFID(cfid),
1672                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1673
1674         if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1675                 lfsck_object_put(env, parent);
1676
1677         if (rc != 0)
1678                 ns->ln_flags |= LF_INCONSISTENT;
1679
1680         return rc;
1681 }
1682
1683 /**
1684  * Remove the specified entry from the linkEA.
1685  *
1686  * Locate the linkEA entry with the given @cname and @pfid, then
1687  * remove this entry or the other entries those are repeated with
1688  * this entry.
1689  *
1690  * \param[in] env       pointer to the thread context
1691  * \param[in] com       pointer to the lfsck component
1692  * \param[in] obj       pointer to the dt_object to be handled
1693  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1694  * \param[in] cname     the name for the child in the parent directory
1695  * \param[in] pfid      the parent directory's FID for the linkEA
1696  * \param[in] next      if true, then remove the first found linkEA
1697  *                      entry, and move the ldata->ld_lee to next entry
1698  *
1699  * \retval              positive number for repaired cases
1700  * \retval              0 if nothing to be repaired
1701  * \retval              negative error number on failure
1702  */
1703 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1704                                          struct lfsck_component *com,
1705                                          struct dt_object *obj,
1706                                          struct linkea_data *ldata,
1707                                          struct lu_name *cname,
1708                                          struct lu_fid *pfid,
1709                                          bool next)
1710 {
1711         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1712         struct dt_device                *dev       = lfsck_obj2dev(obj);
1713         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1714         struct thandle                  *th        = NULL;
1715         struct lustre_handle             lh        = { 0 };
1716         struct linkea_data               ldata_new = { NULL };
1717         struct lu_buf                    linkea_buf;
1718         int                              buflen    = 0;
1719         int                              rc        = 0;
1720         ENTRY;
1721
1722         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1723                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1724                               LCK_EX);
1725         if (rc != 0)
1726                 GOTO(log, rc);
1727
1728         if (next)
1729                 lfsck_linkea_del_buf(ldata, cname);
1730         else
1731                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1732                                                     true);
1733         if (ldata->ld_leh->leh_reccount > 0 ||
1734             unlikely(ldata->ld_leh->leh_overflow_time)) {
1735                 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1736                                ldata->ld_leh->leh_len);
1737                 buflen = linkea_buf.lb_len;
1738         }
1739
1740 again:
1741         th = dt_trans_create(env, dev);
1742         if (IS_ERR(th))
1743                 GOTO(unlock1, rc = PTR_ERR(th));
1744
1745         if (buflen != 0)
1746                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1747                                           XATTR_NAME_LINK, 0, th);
1748         else
1749                 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1750         if (rc != 0)
1751                 GOTO(stop, rc);
1752
1753         rc = dt_trans_start_local(env, dev, th);
1754         if (rc != 0)
1755                 GOTO(stop, rc);
1756
1757         dt_write_lock(env, obj, 0);
1758         if (unlikely(lfsck_is_dead_obj(obj)))
1759                 GOTO(unlock2, rc = -ENOENT);
1760
1761         rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1762         if (rc)
1763                 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1764
1765         /* The specified linkEA entry has been removed by race. */
1766         rc = linkea_links_find(&ldata_new, cname, pfid);
1767         if (rc != 0)
1768                 GOTO(unlock2, rc = 0);
1769
1770         if (bk->lb_param & LPF_DRYRUN)
1771                 GOTO(unlock2, rc = 1);
1772
1773         if (next)
1774                 lfsck_linkea_del_buf(&ldata_new, cname);
1775         else
1776                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1777                                                     true);
1778
1779         /*
1780          * linkea may change because it doesn't take lock in the first read, if
1781          * it becomes larger, restart from beginning.
1782          */
1783         if ((ldata_new.ld_leh->leh_reccount > 0 ||
1784              unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
1785             buflen < ldata_new.ld_leh->leh_len) {
1786                 dt_write_unlock(env, obj);
1787                 dt_trans_stop(env, dev, th);
1788                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1789                                ldata_new.ld_leh->leh_len);
1790                 buflen = linkea_buf.lb_len;
1791                 goto again;
1792         }
1793
1794         if (buflen)
1795                 rc = lfsck_links_write(env, obj, &ldata_new, th);
1796         else
1797                 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1798
1799         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1800
1801 unlock2:
1802         dt_write_unlock(env, obj);
1803
1804 stop:
1805         dt_trans_stop(env, dev, th);
1806
1807 unlock1:
1808         lfsck_ibits_unlock(&lh, LCK_EX);
1809
1810 log:
1811         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1812                "for the object: "DFID", parent "DFID", name %.*s\n",
1813                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1814                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1815                cname->ln_name);
1816
1817         if (rc != 0) {
1818                 struct lfsck_namespace *ns = com->lc_file_ram;
1819
1820                 ns->ln_flags |= LF_INCONSISTENT;
1821         }
1822
1823         return rc;
1824 }
1825
1826 /**
1827  * Conditionally remove the specified entry from the linkEA.
1828  *
1829  * Take the parent lock firstly, then check whether the specified
1830  * name entry exists or not: if yes, do nothing; otherwise, call
1831  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1832  *
1833  * \param[in] env       pointer to the thread context
1834  * \param[in] com       pointer to the lfsck component
1835  * \param[in] parent    pointer to the parent directory
1836  * \param[in] child     pointer to the child object that holds the linkEA
1837  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1838  * \param[in] cname     the name for the child in the parent directory
1839  * \param[in] pfid      the parent directory's FID for the linkEA
1840  *
1841  * \retval              positive number for repaired cases
1842  * \retval              0 if nothing to be repaired
1843  * \retval              negative error number on failure
1844  */
1845 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1846                                               struct lfsck_component *com,
1847                                               struct dt_object *parent,
1848                                               struct dt_object *child,
1849                                               struct linkea_data *ldata,
1850                                               struct lu_name *cname,
1851                                               struct lu_fid *pfid)
1852 {
1853         struct lfsck_thread_info *info  = lfsck_env_info(env);
1854         struct lu_fid            *cfid  = &info->lti_fid3;
1855         struct lfsck_lock_handle *llh   = &info->lti_llh;
1856         int                       rc;
1857         ENTRY;
1858
1859         rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1860                         MDS_INODELOCK_UPDATE, LCK_PR);
1861         if (rc != 0)
1862                 RETURN(rc);
1863
1864         dt_read_lock(env, parent, 0);
1865         if (unlikely(lfsck_is_dead_obj(parent))) {
1866                 dt_read_unlock(env, parent);
1867                 lfsck_unlock(llh);
1868                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1869                                                    cname, pfid, true);
1870
1871                 RETURN(rc);
1872         }
1873
1874         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1875                        (const struct dt_key *)cname->ln_name);
1876         dt_read_unlock(env, parent);
1877
1878         /* It is safe to release the ldlm lock, because when the logic come
1879          * here, we have got all the needed information above whether the
1880          * linkEA entry is valid or not. It is not important that others
1881          * may add new linkEA entry after the ldlm lock released. If other
1882          * has removed the specified linkEA entry by race, then it is OK,
1883          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1884          * such case. */
1885         lfsck_unlock(llh);
1886         if (rc == -ENOENT) {
1887                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1888                                                    cname, pfid, true);
1889
1890                 RETURN(rc);
1891         }
1892
1893         if (rc != 0)
1894                 RETURN(rc);
1895
1896         /* The LFSCK just found some internal status of cross-MDTs
1897          * create operation. That is normal. */
1898         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1899                 linkea_next_entry(ldata);
1900
1901                 RETURN(0);
1902         }
1903
1904         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1905                                            pfid, true);
1906
1907         RETURN(rc);
1908 }
1909
1910 /**
1911  * Conditionally replace name entry in the parent.
1912  *
1913  * As required, the LFSCK may re-create the lost MDT-object for dangling
1914  * name entry, but such repairing may be wrong because of bad FID in the
1915  * name entry. As the LFSCK processing, the real MDT-object may be found,
1916  * then the LFSCK should check whether the former re-created MDT-object
1917  * has been modified or not, if not, then destroy it and update the name
1918  * entry in the parent to reference the real MDT-object.
1919  *
1920  * \param[in] env       pointer to the thread context
1921  * \param[in] com       pointer to the lfsck component
1922  * \param[in] parent    pointer to the parent directory
1923  * \param[in] child     pointer to the MDT-object that may be the real
1924  *                      MDT-object corresponding to the name entry in parent
1925  * \param[in] cfid      the current FID in the name entry
1926  * \param[in] cname     contains the name of the child in the parent directory
1927  *
1928  * \retval              positive number for repaired cases
1929  * \retval              0 if nothing to be repaired
1930  * \retval              negative error number on failure
1931  */
1932 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1933                                         struct lfsck_component *com,
1934                                         struct dt_object *parent,
1935                                         struct dt_object *child,
1936                                         const struct lu_fid *cfid,
1937                                         const struct lu_name *cname)
1938 {
1939         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1940         struct lu_attr                  *la     = &info->lti_la;
1941         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1942         struct lu_fid                    tfid;
1943         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1944         /* The child and its name may be on different MDTs. */
1945         struct dt_device                *dev    = lfsck->li_next;
1946         const char                      *name   = cname->ln_name;
1947         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1948         struct dt_object                *cobj   = NULL;
1949         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
1950         struct lustre_handle             clh    = { 0 };
1951         struct linkea_data               ldata  = { NULL };
1952         struct thandle                  *th     = NULL;
1953         bool                             exist  = true;
1954         int                              rc     = 0;
1955         ENTRY;
1956
1957         /* @parent/@child may be based on lfsck->li_bottom,
1958          * but here we need the object based on the lfsck->li_next. */
1959
1960         parent = lfsck_object_locate(dev, parent);
1961         if (IS_ERR(parent))
1962                 GOTO(log, rc = PTR_ERR(parent));
1963
1964         if (unlikely(!dt_try_as_dir(env, parent)))
1965                 GOTO(log, rc = -ENOTDIR);
1966
1967         rc = lfsck_lock(env, lfsck, parent, name, pllh,
1968                         MDS_INODELOCK_UPDATE, LCK_PW);
1969         if (rc != 0)
1970                 GOTO(log, rc);
1971
1972         if (!fid_is_sane(cfid)) {
1973                 exist = false;
1974                 goto replace;
1975         }
1976
1977         cobj = lfsck_object_find_by_dev(env, dev, cfid);
1978         if (IS_ERR(cobj)) {
1979                 rc = PTR_ERR(cobj);
1980                 if (rc == -ENOENT) {
1981                         exist = false;
1982                         goto replace;
1983                 }
1984
1985                 GOTO(log, rc);
1986         }
1987
1988         if (!dt_object_exists(cobj)) {
1989                 exist = false;
1990                 goto replace;
1991         }
1992
1993         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1994                        (const struct dt_key *)name);
1995         if (rc == -ENOENT) {
1996                 exist = false;
1997                 goto replace;
1998         }
1999
2000         if (rc != 0)
2001                 GOTO(log, rc);
2002
2003         /* Someone changed the name entry, cannot replace it. */
2004         if (!lu_fid_eq(cfid, &tfid))
2005                 GOTO(log, rc = 0);
2006
2007         /* lock the object to be destroyed. */
2008         rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
2009                               MDS_INODELOCK_UPDATE |
2010                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
2011                               LCK_EX);
2012         if (rc != 0)
2013                 GOTO(log, rc);
2014
2015         if (unlikely(lfsck_is_dead_obj(cobj))) {
2016                 exist = false;
2017                 goto replace;
2018         }
2019
2020         rc = dt_attr_get(env, cobj, la);
2021         if (rc != 0)
2022                 GOTO(log, rc);
2023
2024         /* The object has been modified by other(s), or it is not created by
2025          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
2026         if (la->la_ctime != 0)
2027                 GOTO(log, rc);
2028
2029         if (S_ISREG(la->la_mode)) {
2030                 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
2031                 /* If someone has created related OST-object(s),
2032                  * then keep it. */
2033                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
2034                         GOTO(log, rc = (rc > 0 ? 0 : rc));
2035         }
2036
2037 replace:
2038         dt_read_lock(env, child, 0);
2039         rc = lfsck_links_read2_with_rec(env, child, &ldata);
2040         dt_read_unlock(env, child);
2041
2042         /* Someone changed the child, no need to replace. */
2043         if (rc == -ENODATA)
2044                 GOTO(log, rc = 0);
2045
2046         if (rc != 0)
2047                 GOTO(log, rc);
2048
2049         rc = linkea_links_find(&ldata, cname, pfid);
2050         /* Someone moved the child, no need to replace. */
2051         if (rc != 0)
2052                 GOTO(log, rc = 0);
2053
2054         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2055                 GOTO(log, rc = 1);
2056
2057         th = dt_trans_create(env, dev);
2058         if (IS_ERR(th))
2059                 GOTO(log, rc = PTR_ERR(th));
2060
2061         if (exist) {
2062                 rc = dt_declare_destroy(env, cobj, th);
2063                 if (rc != 0)
2064                         GOTO(stop, rc);
2065         }
2066
2067         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2068         if (rc != 0)
2069                 GOTO(stop, rc);
2070
2071         rec->rec_type = S_IFDIR;
2072         rec->rec_fid = lfsck_dto2fid(child);
2073         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
2074                                (const struct dt_key *)name, th);
2075         if (rc != 0)
2076                 GOTO(stop, rc);
2077
2078         rc = dt_trans_start_local(env, dev, th);
2079         if (rc != 0)
2080                 GOTO(stop, rc);
2081
2082         if (exist) {
2083                 rc = dt_destroy(env, cobj, th);
2084                 if (rc != 0)
2085                         GOTO(stop, rc);
2086         }
2087
2088         /* The old name entry maybe not exist. */
2089         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2090         if (rc != 0 && rc != -ENOENT)
2091                 GOTO(stop, rc);
2092
2093         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
2094                        (const struct dt_key *)name, th);
2095
2096         GOTO(stop, rc = (rc == 0 ? 1 : rc));
2097
2098 stop:
2099         dt_trans_stop(env, dev, th);
2100
2101 log:
2102         lfsck_ibits_unlock(&clh, LCK_EX);
2103         lfsck_unlock(pllh);
2104
2105         if (cobj != NULL && !IS_ERR(cobj))
2106                 lfsck_object_put(env, cobj);
2107
2108         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
2109                "object "DFID" because of conflict with the object "DFID
2110                " under the parent "DFID" with name %s: rc = %d\n",
2111                lfsck_lfsck2name(lfsck), PFID(cfid),
2112                PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
2113
2114         return rc;
2115 }
2116
2117 /**
2118  * Overwrite the linkEA for the object with the given ldata.
2119  *
2120  * The caller should take the ldlm lock before the calling.
2121  *
2122  * \param[in] env       pointer to the thread context
2123  * \param[in] com       pointer to the lfsck component
2124  * \param[in] obj       pointer to the dt_object to be handled
2125  * \param[in] ldata     pointer to the new linkEA data
2126  *
2127  * \retval              positive number for repaired cases
2128  * \retval              0 if nothing to be repaired
2129  * \retval              negative error number on failure
2130  */
2131 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
2132                                    struct lfsck_component *com,
2133                                    struct dt_object *obj,
2134                                    struct linkea_data *ldata)
2135 {
2136         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2137         struct dt_device                *dev    = lfsck_obj2dev(obj);
2138         struct thandle                  *th     = NULL;
2139         struct lu_buf                    linkea_buf;
2140         int                              rc     = 0;
2141         ENTRY;
2142
2143         th = dt_trans_create(env, dev);
2144         if (IS_ERR(th))
2145                 GOTO(log, rc = PTR_ERR(th));
2146
2147         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
2148                        ldata->ld_leh->leh_len);
2149         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2150                                   XATTR_NAME_LINK, 0, th);
2151         if (rc != 0)
2152                 GOTO(stop, rc);
2153
2154         rc = dt_trans_start_local(env, dev, th);
2155         if (rc != 0)
2156                 GOTO(stop, rc);
2157
2158         dt_write_lock(env, obj, 0);
2159         if (unlikely(lfsck_is_dead_obj(obj)))
2160                 GOTO(unlock, rc = 0);
2161
2162         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2163                 GOTO(unlock, rc = 1);
2164
2165         rc = dt_xattr_set(env, obj, &linkea_buf,
2166                           XATTR_NAME_LINK, 0, th);
2167
2168         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2169
2170 unlock:
2171         dt_write_unlock(env, obj);
2172
2173 stop:
2174         dt_trans_stop(env, dev, th);
2175
2176 log:
2177         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2178                "object "DFID": rc = %d\n",
2179                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2180
2181         if (rc != 0) {
2182                 struct lfsck_namespace *ns = com->lc_file_ram;
2183
2184                 ns->ln_flags |= LF_INCONSISTENT;
2185         }
2186
2187         return rc;
2188 }
2189
2190 /**
2191  * Repair invalid name entry.
2192  *
2193  * If the name entry contains invalid information, such as bad file type
2194  * or (and) corrupted object FID, then either remove the name entry or
2195  * udpate the name entry with the given (right) information.
2196  *
2197  * \param[in] env       pointer to the thread context
2198  * \param[in] com       pointer to the lfsck component
2199  * \param[in] parent    pointer to the parent directory
2200  * \param[in] child     pointer to the object referenced by the name entry
2201  * \param[in] name      the old name of the child under the parent directory
2202  * \param[in] name2     the new name of the child under the parent directory
2203  * \param[in] type      the type claimed by the name entry
2204  * \param[in] update    update the name entry if true; otherwise, remove it
2205  * \param[in] dec       decrease the parent nlink count if true
2206  *
2207  * \retval              positive number for repaired successfully
2208  * \retval              0 if nothing to be repaired
2209  * \retval              negative error number on failure
2210  */
2211 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2212                                   struct lfsck_component *com,
2213                                   struct dt_object *parent,
2214                                   struct dt_object *child,
2215                                   const char *name, const char *name2,
2216                                   __u16 type, bool update, bool dec)
2217 {
2218         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2219         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2220         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2221         struct lu_fid                   cfid    = {0};
2222         struct lu_fid                    tfid;
2223         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2224         struct dt_device                *dev    = lfsck->li_next;
2225         struct thandle                  *th     = NULL;
2226         struct lfsck_lock_handle        *llh    = &info->lti_llh;
2227         struct lustre_handle             lh     = { 0 };
2228         int                              rc     = 0;
2229         ENTRY;
2230
2231         if (child)
2232                 cfid = *lfsck_dto2fid(child);
2233         parent = lfsck_object_locate(dev, parent);
2234         if (IS_ERR(parent))
2235                 GOTO(log, rc = PTR_ERR(parent));
2236
2237         if (unlikely(!dt_try_as_dir(env, parent)))
2238                 GOTO(log, rc = -ENOTDIR);
2239
2240         if (!update || strcmp(name, name2) == 0)
2241                 rc = lfsck_lock(env, lfsck, parent, name, llh,
2242                                 MDS_INODELOCK_UPDATE, LCK_PW);
2243         else
2244                 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2245                                       MDS_INODELOCK_UPDATE, LCK_PW);
2246         if (rc != 0)
2247                 GOTO(log, rc);
2248
2249         th = dt_trans_create(env, dev);
2250         if (IS_ERR(th))
2251                 GOTO(unlock1, rc = PTR_ERR(th));
2252
2253         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2254         if (rc != 0)
2255                 GOTO(stop, rc);
2256
2257         if (update) {
2258                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2259                 LASSERT(!fid_is_zero(&cfid));
2260                 rec->rec_fid = &cfid;
2261                 rc = dt_declare_insert(env, parent,
2262                                        (const struct dt_rec *)rec,
2263                                        (const struct dt_key *)name2, th);
2264                 if (rc != 0)
2265                         GOTO(stop, rc);
2266         }
2267
2268         if (dec && S_ISDIR(type)) {
2269                 rc = dt_declare_ref_del(env, parent, th);
2270                 if (rc != 0)
2271                         GOTO(stop, rc);
2272         }
2273
2274         rc = dt_trans_start_local(env, dev, th);
2275         if (rc != 0)
2276                 GOTO(stop, rc);
2277
2278
2279         dt_write_lock(env, parent, 0);
2280         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2281                        (const struct dt_key *)name);
2282         /* Someone has removed the bad name entry by race. */
2283         if (rc == -ENOENT)
2284                 GOTO(unlock2, rc = 0);
2285
2286         if (rc != 0)
2287                 GOTO(unlock2, rc);
2288
2289         /* Someone has removed the bad name entry and reused it for other
2290          * object by race. */
2291         if (!lu_fid_eq(&tfid, &cfid))
2292                 GOTO(unlock2, rc = 0);
2293
2294         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2295                 GOTO(unlock2, rc = 1);
2296
2297         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2298         if (rc != 0)
2299                 GOTO(unlock2, rc);
2300
2301         if (update) {
2302                 rc = dt_insert(env, parent,
2303                                (const struct dt_rec *)rec,
2304                                (const struct dt_key *)name2, th);
2305                 if (rc != 0)
2306                         GOTO(unlock2, rc);
2307         }
2308
2309         if (dec && S_ISDIR(type)) {
2310                 rc = dt_ref_del(env, parent, th);
2311                 if (rc != 0)
2312                         GOTO(unlock2, rc);
2313         }
2314
2315         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2316
2317 unlock2:
2318         dt_write_unlock(env, parent);
2319
2320 stop:
2321         dt_trans_stop(env, dev, th);
2322
2323         /* We are not sure whether the child will become orphan or not.
2324          * Record it in the LFSCK trace file for further checking in
2325          * the second-stage scanning. */
2326         if (!update && !dec && child && rc == 0)
2327                 lfsck_namespace_trace_update(env, com, &cfid,
2328                                              LNTF_CHECK_LINKEA, true);
2329
2330 unlock1:
2331         /* It is harmless even if unlock the unused lock_handle */
2332         lfsck_ibits_unlock(&lh, LCK_PW);
2333         lfsck_unlock(llh);
2334
2335 log:
2336         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2337                "entry for: parent "DFID", child "DFID", name %s, type "
2338                "in name entry %o, type claimed by child %o. repair it "
2339                "by %s with new name2 %s: rc = %d\n",
2340                lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2341                name, type, update ? lfsck_object_type(child) : 0,
2342                update ? "updating" : "removing", name2, rc);
2343
2344         if (rc != 0) {
2345                 struct lfsck_namespace *ns = com->lc_file_ram;
2346
2347                 ns->ln_flags |= LF_INCONSISTENT;
2348         }
2349
2350         return rc;
2351 }
2352
2353 /**
2354  * Update the ".." name entry for the given object.
2355  *
2356  * The object's ".." is corrupted, this function will update the ".." name
2357  * entry with the given pfid, and the linkEA with the given ldata.
2358  *
2359  * The caller should take the ldlm lock before the calling.
2360  *
2361  * \param[in] env       pointer to the thread context
2362  * \param[in] com       pointer to the lfsck component
2363  * \param[in] obj       pointer to the dt_object to be handled
2364  * \param[in] pfid      the new fid for the object's ".." name entry
2365  * \param[in] cname     the name for the @obj in the parent directory
2366  *
2367  * \retval              positive number for repaired cases
2368  * \retval              0 if nothing to be repaired
2369  * \retval              negative error number on failure
2370  */
2371 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2372                                                   struct lfsck_component *com,
2373                                                   struct dt_object *obj,
2374                                                   const struct lu_fid *pfid,
2375                                                   struct lu_name *cname)
2376 {
2377         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2378         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2379         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2380         struct dt_device                *dev    = lfsck_obj2dev(obj);
2381         struct thandle                  *th     = NULL;
2382         struct linkea_data               ldata  = { NULL };
2383         struct lu_buf                    linkea_buf;
2384         int                              rc     = 0;
2385         ENTRY;
2386
2387         LASSERT(!dt_object_remote(obj));
2388         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2389
2390         rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2391         if (rc != 0)
2392                 GOTO(log, rc);
2393
2394         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2395                        ldata.ld_leh->leh_len);
2396
2397         th = dt_trans_create(env, dev);
2398         if (IS_ERR(th))
2399                 GOTO(log, rc = PTR_ERR(th));
2400
2401         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2402         if (rc != 0)
2403                 GOTO(stop, rc);
2404
2405         rec->rec_type = S_IFDIR;
2406         rec->rec_fid = pfid;
2407         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2408                                (const struct dt_key *)dotdot, th);
2409         if (rc != 0)
2410                 GOTO(stop, rc);
2411
2412         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2413                                   XATTR_NAME_LINK, 0, th);
2414         if (rc != 0)
2415                 GOTO(stop, rc);
2416
2417         rc = dt_trans_start_local(env, dev, th);
2418         if (rc != 0)
2419                 GOTO(stop, rc);
2420
2421         dt_write_lock(env, obj, 0);
2422         if (unlikely(lfsck_is_dead_obj(obj)))
2423                 GOTO(unlock, rc = 0);
2424
2425         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2426                 GOTO(unlock, rc = 1);
2427
2428         /* The old ".." name entry maybe not exist. */
2429         dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2430
2431         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2432                        (const struct dt_key *)dotdot, th);
2433         if (rc != 0)
2434                 GOTO(unlock, rc);
2435
2436         rc = lfsck_links_write(env, obj, &ldata, th);
2437
2438         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2439
2440 unlock:
2441         dt_write_unlock(env, obj);
2442
2443 stop:
2444         dt_trans_stop(env, dev, th);
2445
2446 log:
2447         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2448                "the object "DFID", new parent "DFID": rc = %d\n",
2449                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2450                PFID(pfid), rc);
2451
2452         if (rc != 0) {
2453                 struct lfsck_namespace *ns = com->lc_file_ram;
2454
2455                 ns->ln_flags |= LF_INCONSISTENT;
2456         }
2457
2458         return rc;
2459 }
2460
2461 /**
2462  * Handle orphan @obj during Double Scan Directory.
2463  *
2464  * Remove the @obj's current (invalid) linkEA entries, and insert
2465  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2466  * ${FID}-${PFID}-D-${conflict_version}
2467  *
2468  * The caller should take the ldlm lock before the calling.
2469  *
2470  * \param[in] env       pointer to the thread context
2471  * \param[in] com       pointer to the lfsck component
2472  * \param[in] obj       pointer to the orphan object to be handled
2473  * \param[in] pfid      the new fid for the object's ".." name entry
2474  * \param[in,out] lh    ldlm lock handler for the given @obj
2475  * \param[out] type     to tell the caller what the inconsistency is
2476  *
2477  * \retval              positive number for repaired cases
2478  * \retval              0 if nothing to be repaired
2479  * \retval              negative error number on failure
2480  */
2481 static int
2482 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2483                            struct lfsck_component *com,
2484                            struct dt_object *obj,
2485                            const struct lu_fid *pfid,
2486                            struct lustre_handle *lh,
2487                            enum lfsck_namespace_inconsistency_type *type)
2488 {
2489         struct lfsck_thread_info *info = lfsck_env_info(env);
2490         struct lfsck_namespace   *ns   = com->lc_file_ram;
2491         int                       rc;
2492         ENTRY;
2493
2494         /* Remove the unrecognized linkEA. */
2495         rc = lfsck_namespace_links_remove(env, com, obj);
2496         lfsck_ibits_unlock(lh, LCK_EX);
2497         if (rc < 0 && rc != -ENODATA)
2498                 RETURN(rc);
2499
2500         *type = LNIT_MUL_REF;
2501
2502         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2503          * ever tried to verify some remote MDT-object that resides on this
2504          * MDT, but this MDT failed to respond such request. So means there
2505          * may be some remote name entry on other MDT that references this
2506          * object with another name, so we cannot know whether this linkEA
2507          * is valid or not. So keep it there and maybe resolved when next
2508          * LFSCK run. */
2509         if (ns->ln_flags & LF_INCOMPLETE)
2510                 RETURN(0);
2511
2512         /* The unique linkEA is invalid, even if the ".." name entry may be
2513          * valid, we still cannot know via which name entry this directory
2514          * will be referenced. Then handle it as pure orphan. */
2515         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2516                  "-"DFID, PFID(pfid));
2517         rc = lfsck_namespace_insert_orphan(env, com, obj,
2518                                            info->lti_tmpbuf, "D", NULL);
2519
2520         RETURN(rc);
2521 }
2522
2523 /**
2524  * Double Scan Directory object for single linkEA entry case.
2525  *
2526  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2527  * then check whether the name is in the namespace or not, if not, add the
2528  * missing name entry back to namespace. If the linkEA entry is invalid,
2529  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2530  * as an orphan.
2531  *
2532  * \param[in] env       pointer to the thread context
2533  * \param[in] com       pointer to the lfsck component
2534  * \param[in] child     pointer to the directory to be double scanned
2535  * \param[in] pfid      the FID corresponding to the ".." entry
2536  * \param[in] ldata     pointer to the linkEA data for the given @child
2537  * \param[in,out] lh    ldlm lock handler for the given @child
2538  * \param[out] type     to tell the caller what the inconsistency is
2539  * \param[in] retry     if found inconsistency, but the caller does not hold
2540  *                      ldlm lock on the @child, then set @retry as true
2541  * \param[in] unknown   set if does not know how to repair the inconsistency
2542  *
2543  * \retval              positive number for repaired cases
2544  * \retval              0 if nothing to be repaired
2545  * \retval              negative error number on failure
2546  */
2547 static int
2548 lfsck_namespace_dsd_single(const struct lu_env *env,
2549                            struct lfsck_component *com,
2550                            struct dt_object *child,
2551                            const struct lu_fid *pfid,
2552                            struct linkea_data *ldata,
2553                            struct lustre_handle *lh,
2554                            enum lfsck_namespace_inconsistency_type *type,
2555                            bool *retry, bool *unknown)
2556 {
2557         struct lfsck_thread_info *info          = lfsck_env_info(env);
2558         struct lu_name           *cname         = &info->lti_name;
2559         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2560         struct lu_fid             tfid;
2561         struct lfsck_namespace   *ns            = com->lc_file_ram;
2562         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2563         struct dt_object         *parent        = NULL;
2564         struct lmv_mds_md_v1     *lmv;
2565         int                       rc            = 0;
2566         ENTRY;
2567
2568         rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2569                                                  info->lti_key,
2570                                                  sizeof(info->lti_key));
2571         /* The unique linkEA entry with bad parent will be handled as orphan. */
2572         if (rc != 0) {
2573                 if (!lustre_handle_is_used(lh) && retry != NULL)
2574                         *retry = true;
2575                 else
2576                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2577                                                         pfid, lh, type);
2578
2579                 GOTO(out, rc);
2580         }
2581
2582         parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2583         if (IS_ERR(parent))
2584                 GOTO(out, rc = PTR_ERR(parent));
2585
2586         /* We trust the unique linkEA entry in spite of whether it matches the
2587          * ".." name entry or not. Because even if the linkEA entry is wrong
2588          * and the ".." name entry is right, we still cannot know via which
2589          * name entry the child will be referenced, since all known entries
2590          * have been verified during the first-stage scanning. */
2591         if (!dt_object_exists(parent)) {
2592                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2593                  * has ever tried to verify some remote MDT-object that resides
2594                  * on this MDT, but this MDT failed to respond such request. So
2595                  * means there may be some remote name entry on other MDT that
2596                  * references this object with another name, so we cannot know
2597                  * whether this linkEA is valid or not. So keep it there and
2598                  * maybe resolved when next LFSCK run. */
2599                 if (ns->ln_flags & LF_INCOMPLETE)
2600                         GOTO(out, rc = 0);
2601
2602                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2603                         *retry = true;
2604
2605                         GOTO(out, rc = 0);
2606                 }
2607
2608                 lfsck_ibits_unlock(lh, LCK_EX);
2609
2610 lost_parent:
2611                 lmv = &info->lti_lmv;
2612                 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
2613                 if (rc != 0 && rc != -ENODATA)
2614                         GOTO(out, rc);
2615
2616                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2617                         lmv = NULL;
2618                 } else if (lfsck_shard_name_to_index(env,
2619                                         cname->ln_name, cname->ln_namelen,
2620                                         S_IFDIR, cfid) < 0) {
2621                         /* It is an invalid name entry, we
2622                          * cannot trust the parent also. */
2623                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2624                                                 ldata, cname, &tfid, true);
2625                         if (rc < 0)
2626                                 GOTO(out, rc);
2627
2628                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2629                                  "-"DFID, PFID(pfid));
2630                         rc = lfsck_namespace_insert_orphan(env, com, child,
2631                                                 info->lti_tmpbuf, "S", NULL);
2632
2633                         GOTO(out, rc);
2634                 }
2635
2636                 /* Create the lost parent as an orphan. */
2637                 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2638                 if (rc >= 0) {
2639                         /* Add the missing name entry to the parent. */
2640                         rc = lfsck_namespace_insert_normal(env, com, parent,
2641                                                            child, cname);
2642                         if (unlikely(rc == -EEXIST)) {
2643                                 /* Unfortunately, someone reused the name
2644                                  * under the parent by race. So we have
2645                                  * to remove the linkEA entry from
2646                                  * current child object. It means that the
2647                                  * LFSCK cannot recover the system
2648                                  * totally back to its original status,
2649                                  * but it is necessary to make the
2650                                  * current system to be consistent. */
2651                                 rc = lfsck_namespace_shrink_linkea(env,
2652                                                 com, child, ldata,
2653                                                 cname, &tfid, true);
2654                                 if (rc >= 0) {
2655                                         snprintf(info->lti_tmpbuf,
2656                                                  sizeof(info->lti_tmpbuf),
2657                                                  "-"DFID, PFID(pfid));
2658                                         rc = lfsck_namespace_insert_orphan(env,
2659                                                 com, child, info->lti_tmpbuf,
2660                                                 "D", NULL);
2661                                 }
2662                         }
2663                 }
2664
2665                 GOTO(out, rc);
2666         } /* !dt_object_exists(parent) */
2667
2668         /* The unique linkEA entry with bad parent will be handled as orphan. */
2669         if (unlikely(!dt_try_as_dir(env, parent))) {
2670                 if (!lustre_handle_is_used(lh) && retry != NULL)
2671                         *retry = true;
2672                 else
2673                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2674                                                         pfid, lh, type);
2675
2676                 GOTO(out, rc);
2677         }
2678
2679         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2680                        (const struct dt_key *)cname->ln_name);
2681         if (rc == -ENOENT) {
2682                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2683                  * has ever tried to verify some remote MDT-object that resides
2684                  * on this MDT, but this MDT failed to respond such request. So
2685                  * means there may be some remote name entry on other MDT that
2686                  * references this object with another name, so we cannot know
2687                  * whether this linkEA is valid or not. So keep it there and
2688                  * maybe resolved when next LFSCK run. */
2689                 if (ns->ln_flags & LF_INCOMPLETE)
2690                         GOTO(out, rc = 0);
2691
2692                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2693                         *retry = true;
2694
2695                         GOTO(out, rc = 0);
2696                 }
2697
2698                 lfsck_ibits_unlock(lh, LCK_EX);
2699                 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
2700                                                 cname);
2701                 if (rc == -ENOENT)
2702                         goto lost_parent;
2703
2704                 if (rc < 0)
2705                         GOTO(out, rc);
2706
2707                 /* It is an invalid name entry, drop it. */
2708                 if (unlikely(rc > 0)) {
2709                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2710                                                 ldata, cname, &tfid, true);
2711                         if (rc >= 0) {
2712                                 snprintf(info->lti_tmpbuf,
2713                                          sizeof(info->lti_tmpbuf),
2714                                          "-"DFID, PFID(pfid));
2715                                 rc = lfsck_namespace_insert_orphan(env, com,
2716                                         child, info->lti_tmpbuf, "D", NULL);
2717                         }
2718
2719                         GOTO(out, rc);
2720                 }
2721
2722                 /* Add the missing name entry back to the namespace. */
2723                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2724                                                    cname);
2725                 if (unlikely(rc == -ESTALE))
2726                         /* It may happen when the remote object has been
2727                          * removed, but the local MDT is not aware of that. */
2728                         goto lost_parent;
2729
2730                 if (unlikely(rc == -EEXIST)) {
2731                         /* Unfortunately, someone reused the name under the
2732                          * parent by race. So we have to remove the linkEA
2733                          * entry from current child object. It means that the
2734                          * LFSCK cannot recover the system totally back to
2735                          * its original status, but it is necessary to make
2736                          * the current system to be consistent.
2737                          *
2738                          * It also may be because of the LFSCK found some
2739                          * internal status of create operation. Under such
2740                          * case, nothing to be done. */
2741                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2742                                         parent, child, ldata, cname, &tfid);
2743                         if (rc >= 0) {
2744                                 snprintf(info->lti_tmpbuf,
2745                                          sizeof(info->lti_tmpbuf),
2746                                          "-"DFID, PFID(pfid));
2747                                 rc = lfsck_namespace_insert_orphan(env, com,
2748                                         child, info->lti_tmpbuf, "D", NULL);
2749                         }
2750                 }
2751
2752                 GOTO(out, rc);
2753         } /* rc == -ENOENT */
2754
2755         if (rc != 0)
2756                 GOTO(out, rc);
2757
2758         if (!lu_fid_eq(&tfid, cfid)) {
2759                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2760                         *retry = true;
2761
2762                         GOTO(out, rc = 0);
2763                 }
2764
2765                 lfsck_ibits_unlock(lh, LCK_EX);
2766                 /* The name entry references another MDT-object that
2767                  * may be created by the LFSCK for repairing dangling
2768                  * name entry. Try to replace it. */
2769                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2770                                                   &tfid, cname);
2771                 if (rc == 0)
2772                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2773                                                         pfid, lh, type);
2774
2775                 GOTO(out, rc);
2776         }
2777
2778         /* Zero FID may because the remote directroy object has invalid linkEA,
2779          * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2780          * how to repair the inconsistency, but the namespace LFSCK on the MDT
2781          * where its name entry resides may has more information (name, FID) to
2782          * repair such inconsistency. So here, keep the inconsistency to avoid
2783          * some imporper repairing. */
2784         if (fid_is_zero(pfid)) {
2785                 if (unknown)
2786                         *unknown = true;
2787
2788                 GOTO(out, rc = 0);
2789         }
2790
2791         /* The ".." name entry is wrong, update it. */
2792         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2793                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2794                         *retry = true;
2795
2796                         GOTO(out, rc = 0);
2797                 }
2798
2799                 *type = LNIT_UNMATCHED_PAIRS;
2800                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2801                                                 lfsck_dto2fid(parent), cname);
2802         }
2803
2804         GOTO(out, rc);
2805
2806 out:
2807         if (parent != NULL && !IS_ERR(parent))
2808                 lfsck_object_put(env, parent);
2809
2810         return rc;
2811 }
2812
2813 /**
2814  * Double Scan Directory object for multiple linkEA entries case.
2815  *
2816  * The given @child has multiple linkEA entries. There is at most one linkEA
2817  * entry will be valid, all the others will be removed. Firstly, the function
2818  * will try to find out the linkEA entry for which the name entry exists under
2819  * the given parent (@pfid). If there is no linkEA entry that matches the given
2820  * ".." name entry, then tries to find out the first linkEA entry that both the
2821  * parent and the name entry exist to rebuild a new ".." name entry.
2822  *
2823  * \param[in] env       pointer to the thread context
2824  * \param[in] com       pointer to the lfsck component
2825  * \param[in] child     pointer to the directory to be double scanned
2826  * \param[in] pfid      the FID corresponding to the ".." entry
2827  * \param[in] ldata     pointer to the linkEA data for the given @child
2828  * \param[in,out] lh    ldlm lock handler for the given @child
2829  * \param[out] type     to tell the caller what the inconsistency is
2830  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2831  * \param[in] unknown   set if does not know how to repair the inconsistency
2832  *
2833  * \retval              positive number for repaired cases
2834  * \retval              0 if nothing to be repaired
2835  * \retval              negative error number on failure
2836  */
2837 static int
2838 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2839                              struct lfsck_component *com,
2840                              struct dt_object *child,
2841                              const struct lu_fid *pfid,
2842                              struct linkea_data *ldata,
2843                              struct lustre_handle *lh,
2844                              enum lfsck_namespace_inconsistency_type *type,
2845                              bool lpf, bool *unknown)
2846 {
2847         struct lfsck_thread_info *info          = lfsck_env_info(env);
2848         struct lu_name           *cname         = &info->lti_name;
2849         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2850         struct lu_fid            *pfid2         = &info->lti_fid3;
2851         struct lu_fid             tfid;
2852         struct lfsck_namespace   *ns            = com->lc_file_ram;
2853         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2854         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2855         struct dt_object         *parent        = NULL;
2856         struct linkea_data        ldata_new     = { NULL };
2857         int                       dirent_count  = 0;
2858         int                       rc            = 0;
2859         bool                      once          = true;
2860         ENTRY;
2861
2862 again:
2863         while (ldata->ld_lee != NULL) {
2864                 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2865                                                          info->lti_key,
2866                                                          sizeof(info->lti_key));
2867                 /* Drop invalid linkEA entry. */
2868                 if (rc != 0) {
2869                         lfsck_linkea_del_buf(ldata, cname);
2870                         continue;
2871                 }
2872
2873                 /* Drop repeated linkEA entries. */
2874                 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2875
2876                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2877                  * then it is possible that: the directry object has ever
2878                  * been lost, but its name entry was there. In the former
2879                  * LFSCK run, during the first-stage scanning, the LFSCK
2880                  * found the dangling name entry, but it did not recreate
2881                  * the lost object, and when moved to the second-stage
2882                  * scanning, some children objects of the lost directory
2883                  * object were found, then the LFSCK recreated such lost
2884                  * directory object as an orphan.
2885                  *
2886                  * When the LFSCK runs again, if the dangling name is still
2887                  * there, the LFSCK should move the orphan directory object
2888                  * back to the normal namespace. */
2889                 if (!lpf && !fid_is_zero(pfid) &&
2890                     !lu_fid_eq(pfid, &tfid) && once) {
2891                         linkea_next_entry(ldata);
2892                         continue;
2893                 }
2894
2895                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2896                 if (IS_ERR(parent))
2897                         RETURN(PTR_ERR(parent));
2898
2899                 if (!dt_object_exists(parent)) {
2900                         lfsck_object_put(env, parent);
2901                         if (ldata->ld_leh->leh_reccount > 1) {
2902                                 /* If it is NOT the last linkEA entry, then
2903                                  * there is still other chance to make the
2904                                  * child to be visible via other parent, then
2905                                  * remove this linkEA entry. */
2906                                 lfsck_linkea_del_buf(ldata, cname);
2907                                 continue;
2908                         }
2909
2910                         break;
2911                 }
2912
2913                 /* The linkEA entry with bad parent will be removed. */
2914                 if (unlikely(!dt_try_as_dir(env, parent))) {
2915                         lfsck_object_put(env, parent);
2916                         lfsck_linkea_del_buf(ldata, cname);
2917                         continue;
2918                 }
2919
2920                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2921                                (const struct dt_key *)cname->ln_name);
2922                 *pfid2 = *lfsck_dto2fid(parent);
2923                 if (rc == -ENOENT) {
2924                         lfsck_object_put(env, parent);
2925                         linkea_next_entry(ldata);
2926                         continue;
2927                 }
2928
2929                 if (rc != 0) {
2930                         lfsck_object_put(env, parent);
2931
2932                         RETURN(rc);
2933                 }
2934
2935                 if (lu_fid_eq(&tfid, cfid)) {
2936                         lfsck_object_put(env, parent);
2937                         /* If the parent (that is declared via linkEA entry)
2938                          * directory contains the specified child, but such
2939                          * parent does not match the dotdot name entry, then
2940                          * trust the linkEA. */
2941                         if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2942                                 *type = LNIT_UNMATCHED_PAIRS;
2943                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2944                                                 com, child, pfid2, cname);
2945
2946                                 RETURN(rc);
2947                         }
2948
2949 rebuild:
2950                         /* It is the most common case that we find the
2951                          * name entry corresponding to the linkEA entry
2952                          * that matches the ".." name entry. */
2953                         rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2954                                               cname, pfid2);
2955                         if (rc != 0)
2956                                 RETURN(rc);
2957
2958                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2959                                                             &ldata_new);
2960                         if (rc < 0)
2961                                 RETURN(rc);
2962
2963                         lfsck_linkea_del_buf(ldata, cname);
2964                         linkea_first_entry(ldata);
2965                         /* There may be some invalid dangling name entries under
2966                          * other parent directories, remove all of them. */
2967                         while (ldata->ld_lee != NULL) {
2968                                 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2969                                                 cname, &tfid, info->lti_key,
2970                                                 sizeof(info->lti_key));
2971                                 if (rc != 0)
2972                                         goto next;
2973
2974                                 parent = lfsck_object_find_bottom(env, lfsck,
2975                                                                   &tfid);
2976                                 if (IS_ERR(parent)) {
2977                                         rc = PTR_ERR(parent);
2978                                         if (rc != -ENOENT &&
2979                                             bk->lb_param & LPF_FAILOUT)
2980                                                 RETURN(rc);
2981
2982                                         goto next;
2983                                 }
2984
2985                                 if (!dt_object_exists(parent)) {
2986                                         lfsck_object_put(env, parent);
2987                                         goto next;
2988                                 }
2989
2990                                 rc = lfsck_namespace_repair_dirent(env, com,
2991                                         parent, child, cname->ln_name,
2992                                         cname->ln_name, S_IFDIR, false, true);
2993                                 lfsck_object_put(env, parent);
2994                                 if (rc < 0) {
2995                                         if (bk->lb_param & LPF_FAILOUT)
2996                                                 RETURN(rc);
2997
2998                                         goto next;
2999                                 }
3000
3001                                 dirent_count += rc;
3002
3003 next:
3004                                 lfsck_linkea_del_buf(ldata, cname);
3005                         }
3006
3007                         ns->ln_dirent_repaired += dirent_count;
3008
3009                         RETURN(rc);
3010                 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
3011
3012                 lfsck_ibits_unlock(lh, LCK_EX);
3013                 /* The name entry references another MDT-object that may be
3014                  * created by the LFSCK for repairing dangling name entry.
3015                  * Try to replace it. */
3016                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
3017                                                   &tfid, cname);
3018                 lfsck_object_put(env, parent);
3019                 if (rc < 0)
3020                         RETURN(rc);
3021
3022                 if (rc > 0)
3023                         goto rebuild;
3024
3025                 lfsck_linkea_del_buf(ldata, cname);
3026         } /* while (ldata->ld_lee != NULL) */
3027
3028         /* If there is still linkEA overflow, return. */
3029         if (unlikely(ldata->ld_leh->leh_overflow_time))
3030                 RETURN(0);
3031
3032         linkea_first_entry(ldata);
3033         if (ldata->ld_leh->leh_reccount == 1) {
3034                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
3035                                                 lh, type, NULL, unknown);
3036
3037                 RETURN(rc);
3038         }
3039
3040         /* All linkEA entries are invalid and removed, then handle the @child
3041          * as an orphan.*/
3042         if (ldata->ld_leh->leh_reccount == 0) {