Whamcloud - gitweb
LU-15886 lfsck: remove unreasonable assertions
[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_of(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 = lfsck_trans_create(env, dev, lfsck);
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_dir(env, dir, name, fid);
600         if (rc == -ENOENT)
601                 RETURN(LFSCK_NAMEENTRY_REMOVED);
602
603         if (rc < 0)
604                 RETURN(rc);
605
606         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
607                 RETURN(LFSCK_NAMEENTRY_RECREATED);
608
609         RETURN(0);
610 }
611
612 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
613                                             struct dt_object *obj,
614                                             struct thandle *handle)
615 {
616         int rc;
617
618         /* For remote updating LINKEA, there may be further LFSCK action
619          * on remote MDT after the updating, so update the LINKEA ASAP. */
620         if (dt_object_remote(obj))
621                 handle->th_sync = 1;
622
623         /* For destroying all invalid linkEA entries. */
624         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
625         if (rc == 0)
626                 /* For insert new linkEA entry. */
627                 rc = dt_declare_xattr_set(env, obj,
628                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
629                         XATTR_NAME_LINK, 0, handle);
630         return rc;
631 }
632
633 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
634                        struct linkea_data *ldata, bool with_rec)
635 {
636         int rc;
637
638         if (ldata->ld_buf->lb_buf == NULL)
639                 return -ENOMEM;
640
641         if (!dt_object_exists(obj))
642                 return -ENOENT;
643
644         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
645         if (rc == -ERANGE) {
646                 /* Buf was too small, figure out what we need. */
647                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
648                 if (unlikely(rc == 0))
649                         return -ENODATA;
650
651                 if (rc < 0)
652                         return rc;
653
654                 lu_buf_realloc(ldata->ld_buf, rc);
655                 if (ldata->ld_buf->lb_buf == NULL)
656                         return -ENOMEM;
657
658                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
659         }
660
661         if (unlikely(rc == 0))
662                 return -ENODATA;
663
664         if (rc > 0) {
665                 if (with_rec)
666                         rc = linkea_init_with_rec(ldata);
667                 else
668                         rc = linkea_init(ldata);
669         }
670
671         return rc;
672 }
673
674 /**
675  * Remove linkEA for the given object.
676  *
677  * The caller should take the ldlm lock before the calling.
678  *
679  * \param[in] env       pointer to the thread context
680  * \param[in] com       pointer to the lfsck component
681  * \param[in] obj       pointer to the dt_object to be handled
682  *
683  * \retval              0 for repaired cases
684  * \retval              negative error number on failure
685  */
686 static int lfsck_namespace_links_remove(const struct lu_env *env,
687                                         struct lfsck_component *com,
688                                         struct dt_object *obj)
689 {
690         struct lfsck_instance           *lfsck  = com->lc_lfsck;
691         struct dt_device                *dev    = lfsck_obj2dev(obj);
692         struct thandle                  *th     = NULL;
693         int                              rc     = 0;
694         ENTRY;
695
696         LASSERT(dt_object_remote(obj) == 0);
697
698         th = lfsck_trans_create(env, dev, lfsck);
699         if (IS_ERR(th))
700                 GOTO(log, rc = PTR_ERR(th));
701
702         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
703         if (rc != 0)
704                 GOTO(stop, rc);
705
706         rc = dt_trans_start_local(env, dev, th);
707         if (rc != 0)
708                 GOTO(stop, rc);
709
710         dt_write_lock(env, obj, 0);
711         if (unlikely(lfsck_is_dead_obj(obj)))
712                 GOTO(unlock, rc = -ENOENT);
713
714         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
715                 GOTO(unlock, rc = 0);
716
717         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
718
719         GOTO(unlock, rc);
720
721 unlock:
722         dt_write_unlock(env, obj);
723
724 stop:
725         dt_trans_stop(env, dev, th);
726
727 log:
728         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
729                "for the object "DFID": rc = %d\n",
730                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
731
732         if (rc == 0) {
733                 struct lfsck_namespace *ns = com->lc_file_ram;
734
735                 ns->ln_flags |= LF_INCONSISTENT;
736         }
737
738         return rc;
739 }
740
741 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
742                              struct linkea_data *ldata, struct thandle *handle)
743 {
744         struct lu_buf buf;
745         int rc;
746
747         lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
748
749 again:
750         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
751         if (unlikely(rc == -ENOSPC)) {
752                 rc = linkea_overflow_shrink(ldata);
753                 if (likely(rc > 0)) {
754                         buf.lb_len = rc;
755                         goto again;
756                 }
757         }
758
759         return rc;
760 }
761
762 static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
763 {
764         if (ldata->ld_reclen <= 0)
765                 return false;
766
767         if ((char *)ldata->ld_lee + ldata->ld_reclen >
768             (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
769                 return false;
770
771         return true;
772 }
773
774 static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
775                                          const struct lu_name *cname,
776                                          const struct lu_fid *pfid)
777 {
778         if (!linkea_reclen_is_valid(ldata))
779                 return false;
780
781         if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
782                 return false;
783
784         if (!fid_is_sane(pfid))
785                 return false;
786
787         return true;
788 }
789
790 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
791                                                struct lu_name *cname,
792                                                struct lu_fid *pfid,
793                                                char *buf, const int buflen)
794 {
795         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
796         if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
797                 return -EINVAL;
798
799         /* To guarantee the 'name' is terminated with '0'. */
800         memcpy(buf, cname->ln_name, cname->ln_namelen);
801         buf[cname->ln_namelen] = 0;
802         cname->ln_name = buf;
803
804         return 0;
805 }
806
807 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
808                                  const struct lu_name *lname)
809 {
810         LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
811
812         /* If current record is corrupted, all the subsequent
813          * records will be dropped. */
814         if (unlikely(!linkea_reclen_is_valid(ldata))) {
815                 void *ptr = ldata->ld_lee;
816
817                 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
818                 ldata->ld_leh->leh_reccount = 0;
819                 linkea_first_entry(ldata);
820                 while (ldata->ld_lee != NULL &&
821                        (char *)ldata->ld_lee < (char *)ptr) {
822                         int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
823                                      ldata->ld_lee->lee_reclen[1];
824
825                         ldata->ld_leh->leh_len += reclen;
826                         ldata->ld_leh->leh_reccount++;
827                         ldata->ld_lee = (struct link_ea_entry *)
828                                         ((char *)ldata->ld_lee + reclen);
829                 }
830
831                 ldata->ld_lee = NULL;
832         } else {
833                 linkea_del_buf(ldata, lname, false);
834         }
835 }
836
837 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
838                                                struct lu_name *cname,
839                                                struct lu_fid *pfid,
840                                                bool remove)
841 {
842         struct link_ea_entry    *oldlee;
843         int                      oldlen;
844         int                      repeated = 0;
845
846         oldlee = ldata->ld_lee;
847         oldlen = ldata->ld_reclen;
848         linkea_next_entry(ldata);
849         while (ldata->ld_lee != NULL) {
850                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
851                                    ldata->ld_lee->lee_reclen[1];
852                 if (unlikely(!linkea_reclen_is_valid(ldata))) {
853                         lfsck_linkea_del_buf(ldata, NULL);
854                         LASSERT(ldata->ld_lee == NULL);
855                 } else if (unlikely(ldata->ld_reclen == oldlen &&
856                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
857                         repeated++;
858                         if (!remove)
859                                 break;
860
861                         lfsck_linkea_del_buf(ldata, cname);
862                 } else {
863                         linkea_next_entry(ldata);
864                 }
865         }
866         ldata->ld_lee = oldlee;
867         ldata->ld_reclen = oldlen;
868
869         return repeated;
870 }
871
872 /**
873  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
874  *
875  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
876  * with the given type to generate the name, the detailed rules for name
877  * have been described as following.
878  *
879  * The function also generates the linkEA corresponding to the name entry
880  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
881  *
882  * \param[in] env       pointer to the thread context
883  * \param[in] com       pointer to the lfsck component
884  * \param[in] orphan    pointer to the orphan MDT-object
885  * \param[in] infix     additional information for the orphan name, such as
886  *                      the FID for original
887  * \param[in] type      the type for describing why the orphan MDT-object is
888  *                      created. The rules are as following:
889  *
890  *  type "D":           The MDT-object is a directory, it may knows its parent
891  *                      but because there is no valid linkEA, the LFSCK cannot
892  *                      know where to put it back to the namespace.
893  *  type "O":           The MDT-object has no linkEA, and there is no name
894  *                      entry that references the MDT-object.
895  *
896  *  type "S":           The orphan MDT-object is a shard of a striped directory
897  *
898  * \see lfsck_layout_recreate_parent() for more types.
899  *
900  * The orphan name will be like:
901  * ${FID}-${infix}-${type}-${conflict_version}
902  *
903  * \param[out] count    if some others inserted some linkEA entries by race,
904  *                      then return the linkEA entries count.
905  *
906  * \retval              positive number for repaired cases
907  * \retval              0 if needs to repair nothing
908  * \retval              negative error number on failure
909  */
910 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
911                                          struct lfsck_component *com,
912                                          struct dt_object *orphan,
913                                          const char *infix, const char *type,
914                                          int *count)
915 {
916         struct lfsck_thread_info        *info   = lfsck_env_info(env);
917         struct lu_name                  *cname  = &info->lti_name;
918         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
919         struct lu_attr                  *la     = &info->lti_la2;
920         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
921         const struct lu_fid             *pfid;
922         struct lu_fid                    tfid;
923         struct lfsck_instance           *lfsck  = com->lc_lfsck;
924         struct dt_device                *dev    = lfsck_obj2dev(orphan);
925         struct dt_object                *parent;
926         struct thandle                  *th     = NULL;
927         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
928         struct lustre_handle             clh    = { 0 };
929         struct linkea_data               ldata2 = { NULL };
930         struct lu_buf                    linkea_buf;
931         int                              namelen;
932         int                              idx    = 0;
933         int                              rc     = 0;
934         bool                             exist  = false;
935         ENTRY;
936
937         cname->ln_name = NULL;
938         if (unlikely(lfsck->li_lpf_obj == NULL))
939                 GOTO(log, rc = -ENXIO);
940
941         parent = lfsck->li_lpf_obj;
942         pfid = lfsck_dto2fid(parent);
943
944 again:
945         do {
946                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
947                                    PFID(cfid), infix, type, idx++);
948                 rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
949                 if (rc != 0 && rc != -ENOENT)
950                         GOTO(log, rc);
951
952                 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
953                         exist = true;
954         } while (rc == 0 && !exist);
955
956         rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
957                         MDS_INODELOCK_UPDATE, LCK_PW);
958         if (rc != 0)
959                 GOTO(log, rc);
960
961         /* Re-check whether the name conflict with othrs after taken
962          * the ldlm lock. */
963         rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
964         if (rc == 0) {
965                 if (!lu_fid_eq(cfid, &tfid)) {
966                         exist = false;
967                         lfsck_unlock(pllh);
968                         goto again;
969                 }
970
971                 exist = true;
972         } else if (rc != -ENOENT) {
973                 GOTO(log, rc);
974         } else {
975                 exist = false;
976         }
977
978         cname->ln_name = info->lti_key;
979         cname->ln_namelen = namelen;
980         rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
981                               cname, pfid);
982         if (rc != 0)
983                 GOTO(log, rc);
984
985         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
986                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
987                               MDS_INODELOCK_XATTR, LCK_EX);
988         if (rc != 0)
989                 GOTO(log, rc);
990
991         lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
992                        ldata2.ld_leh->leh_len);
993         th = lfsck_trans_create(env, dev, lfsck);
994         if (IS_ERR(th))
995                 GOTO(log, rc = PTR_ERR(th));
996
997         if (S_ISDIR(lfsck_object_type(orphan))) {
998                 rc = dt_declare_delete(env, orphan,
999                                        (const struct dt_key *)dotdot, th);
1000                 if (rc != 0)
1001                         GOTO(stop, rc);
1002
1003                 rec->rec_type = S_IFDIR;
1004                 rec->rec_fid = pfid;
1005                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1006                                        (const struct dt_key *)dotdot, th);
1007                 if (rc != 0)
1008                         GOTO(stop, rc);
1009         }
1010
1011         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1012                                   XATTR_NAME_LINK, 0, th);
1013         if (rc != 0)
1014                 GOTO(stop, rc);
1015
1016         if (!exist) {
1017                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1018                 rec->rec_fid = cfid;
1019                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1020                                        (const struct dt_key *)cname->ln_name,
1021                                        th);
1022                 if (rc != 0)
1023                         GOTO(stop, rc);
1024
1025                 if (S_ISDIR(rec->rec_type)) {
1026                         rc = dt_declare_ref_add(env, parent, th);
1027                         if (rc != 0)
1028                                 GOTO(stop, rc);
1029                 }
1030         }
1031
1032         memset(la, 0, sizeof(*la));
1033         la->la_ctime = ktime_get_real_seconds();
1034         la->la_valid = LA_CTIME;
1035         rc = dt_declare_attr_set(env, orphan, la, th);
1036         if (rc != 0)
1037                 GOTO(stop, rc);
1038
1039         rc = dt_trans_start_local(env, dev, th);
1040         if (rc != 0)
1041                 GOTO(stop, rc);
1042
1043         dt_write_lock(env, orphan, 0);
1044         rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1045         if (likely(rc == -ENODATA || rc == -EINVAL)) {
1046                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1047                         GOTO(unlock, rc = 1);
1048
1049                 if (S_ISDIR(lfsck_object_type(orphan))) {
1050                         rc = dt_delete(env, orphan,
1051                                        (const struct dt_key *)dotdot, th);
1052                         if (rc != 0)
1053                                 GOTO(unlock, rc);
1054
1055                         rec->rec_type = S_IFDIR;
1056                         rec->rec_fid = pfid;
1057                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1058                                        (const struct dt_key *)dotdot, th);
1059                         if (rc != 0)
1060                                 GOTO(unlock, rc);
1061                 }
1062
1063                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1064                                   th);
1065         } else {
1066                 if (rc == 0 && count != NULL)
1067                         *count = ldata2.ld_leh->leh_reccount;
1068
1069                 GOTO(unlock, rc);
1070         }
1071         dt_write_unlock(env, orphan);
1072
1073         if (rc == 0 && !exist) {
1074                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1075                 rec->rec_fid = cfid;
1076                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1077                                (const struct dt_key *)cname->ln_name, th);
1078                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1079                         dt_write_lock(env, parent, 0);
1080                         rc = dt_ref_add(env, parent, th);
1081                         dt_write_unlock(env, parent);
1082                 }
1083         }
1084
1085         if (rc == 0)
1086                 rc = dt_attr_set(env, orphan, la, th);
1087
1088         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1089
1090 unlock:
1091         dt_write_unlock(env, orphan);
1092
1093 stop:
1094         dt_trans_stop(env, dev, th);
1095
1096 log:
1097         lfsck_ibits_unlock(&clh, LCK_EX);
1098         lfsck_unlock(pllh);
1099         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1100                "object "DFID", name = %s: rc = %d\n",
1101                lfsck_lfsck2name(lfsck), PFID(cfid),
1102                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1103
1104         if (rc != 0) {
1105                 struct lfsck_namespace *ns = com->lc_file_ram;
1106
1107                 ns->ln_flags |= LF_INCONSISTENT;
1108         }
1109
1110         return rc;
1111 }
1112
1113 static int lfsck_lmv_set(const struct lu_env *env,
1114                          struct lfsck_instance *lfsck,
1115                          struct dt_object *obj,
1116                          struct lmv_mds_md_v1 *lmv)
1117 {
1118         struct dt_device *dev = lfsck->li_next;
1119         struct thandle *th = NULL;
1120         struct lu_buf buf = { lmv, sizeof(*lmv) };
1121         int rc;
1122
1123         ENTRY;
1124
1125         th = lfsck_trans_create(env, dev, lfsck);
1126         if (IS_ERR(th))
1127                 RETURN(PTR_ERR(th));
1128
1129         rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1130         if (rc)
1131                 GOTO(stop, rc);
1132
1133         rc = dt_trans_start_local(env, dev, th);
1134         if (rc != 0)
1135                 GOTO(stop, rc);
1136
1137         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1138         if (rc)
1139                 GOTO(stop, rc);
1140
1141         EXIT;
1142 stop:
1143         dt_trans_stop(env, dev, th);
1144
1145         return rc;
1146 }
1147
1148 static int lfsck_lmv_delete(const struct lu_env *env,
1149                             struct lfsck_instance *lfsck,
1150                             struct dt_object *obj)
1151 {
1152         struct dt_device *dev = lfsck->li_next;
1153         struct thandle *th = NULL;
1154         int rc;
1155
1156         ENTRY;
1157
1158         th = lfsck_trans_create(env, dev, lfsck);
1159         if (IS_ERR(th))
1160                 RETURN(PTR_ERR(th));
1161
1162         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
1163         if (rc)
1164                 GOTO(stop, rc);
1165
1166         rc = dt_trans_start_local(env, dev, th);
1167         if (rc != 0)
1168                 GOTO(stop, rc);
1169
1170         rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
1171         if (rc)
1172                 GOTO(stop, rc);
1173
1174         EXIT;
1175 stop:
1176         dt_trans_stop(env, dev, th);
1177
1178         return rc;
1179 }
1180
1181 static inline int lfsck_object_is_shard(const struct lu_env *env,
1182                                         struct lfsck_instance *lfsck,
1183                                         struct dt_object *obj,
1184                                         const struct lu_name *lname)
1185 {
1186         struct lfsck_thread_info *info = lfsck_env_info(env);
1187         struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1188         int rc;
1189
1190         rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
1191                                        lfsck_object_type(obj),
1192                                        lfsck_dto2fid(obj));
1193         if (rc < 0)
1194                 return 0;
1195
1196         rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
1197         if (rc == -ENODATA)
1198                 return 0;
1199
1200         if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
1201                 return 1;
1202
1203         return rc;
1204 }
1205
1206 /**
1207  * Add the specified name entry back to namespace.
1208  *
1209  * If there is a linkEA entry that back references a name entry under
1210  * some parent directory, but such parent directory does not have the
1211  * claimed name entry. On the other hand, the linkEA entries count is
1212  * not larger than the MDT-object's hard link count. Under such case,
1213  * it is quite possible that the name entry is lost. Then the LFSCK
1214  * should add the name entry back to the namespace.
1215  *
1216  * If \a child is shard, which means \a parent is a striped directory,
1217  * if \a parent has LMV, we need to delete it before insertion because
1218  * now parent's striping is broken and can't be parsed correctly.
1219  *
1220  * \param[in] env       pointer to the thread context
1221  * \param[in] com       pointer to the lfsck component
1222  * \param[in] parent    pointer to the directory under which the name entry
1223  *                      will be inserted into
1224  * \param[in] child     pointer to the object referenced by the name entry
1225  *                      that to be inserted into the parent
1226  * \param[in] lname     the name for the child in the parent directory
1227  *
1228  * \retval              positive number for repaired cases
1229  * \retval              0 if nothing to be repaired
1230  * \retval              negative error number on failure
1231  */
1232 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1233                                          struct lfsck_component *com,
1234                                          struct dt_object *parent,
1235                                          struct dt_object *child,
1236                                          const struct lu_name *lname)
1237 {
1238         struct lfsck_thread_info *info = lfsck_env_info(env);
1239         struct lu_attr *la = &info->lti_la;
1240         struct dt_insert_rec *rec = &info->lti_dt_rec;
1241         struct lfsck_instance *lfsck = com->lc_lfsck;
1242         /* The child and its name may be on different MDTs. */
1243         const struct lu_fid *pfid = lfsck_dto2fid(parent);
1244         const struct lu_fid *cfid = lfsck_dto2fid(child);
1245         struct dt_device *dev = lfsck->li_next;
1246         struct thandle *th = NULL;
1247         struct lfsck_lock_handle *llh = &info->lti_llh;
1248         struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1249         struct lu_buf buf = { lmv, sizeof(*lmv) };
1250         /* whether parent's LMV is deleted before insertion */
1251         bool parent_lmv_deleted = false;
1252         /* whether parent's LMV is missing */
1253         bool parent_lmv_lost = false;
1254         int rc = 0;
1255
1256         ENTRY;
1257
1258         /* @parent/@child may be based on lfsck->li_bottom,
1259          * but here we need the object based on the lfsck->li_next. */
1260
1261         parent = lfsck_object_locate(dev, parent);
1262         if (IS_ERR(parent))
1263                 GOTO(log, rc = PTR_ERR(parent));
1264
1265         child = lfsck_object_locate(dev, child);
1266         if (IS_ERR(child))
1267                 GOTO(log, rc = PTR_ERR(child));
1268
1269         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1270                 GOTO(log, rc = 1);
1271
1272         rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
1273                         MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1274                         MDS_INODELOCK_XATTR, LCK_EX);
1275         if (rc)
1276                 GOTO(log, rc);
1277
1278         rc = lfsck_object_is_shard(env, lfsck, child, lname);
1279         if (rc < 0)
1280                 GOTO(unlock, rc);
1281
1282         if (rc == 1) {
1283                 rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
1284                 if (!rc) {
1285                         /*
1286                          * To add a shard, we need to convert parent to a
1287                          * plain directory by deleting its LMV, and after
1288                          * insertion set it back.
1289                          */
1290                         rc = lfsck_lmv_delete(env, lfsck, parent);
1291                         if (rc)
1292                                 GOTO(unlock, rc);
1293                         parent_lmv_deleted = true;
1294                         lmv->lmv_layout_version++;
1295                         lfsck_lmv_header_cpu_to_le(lmv, lmv);
1296                 } else if (rc == -ENODATA) {
1297                         struct lu_seq_range *range = &info->lti_range;
1298                         struct seq_server_site *ss = lfsck_dev_site(lfsck);
1299
1300                         rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
1301                         if (rc)
1302                                 GOTO(unlock, rc);
1303
1304                         fld_range_set_mdt(range);
1305                         rc = fld_server_lookup(env, ss->ss_server_fld,
1306                                        fid_seq(lfsck_dto2fid(parent)), range);
1307                         if (rc)
1308                                 GOTO(unlock, rc);
1309
1310                         parent_lmv_lost = true;
1311                         lmv->lmv_magic = LMV_MAGIC;
1312                         lmv->lmv_master_mdt_index = range->lsr_index;
1313                         lmv->lmv_layout_version++;
1314                         lfsck_lmv_header_cpu_to_le(lmv, lmv);
1315                 } else {
1316                         GOTO(unlock, rc);
1317                 }
1318         }
1319
1320         if (unlikely(!dt_try_as_dir(env, parent)))
1321                 GOTO(unlock, rc = -ENOTDIR);
1322
1323         th = lfsck_trans_create(env, dev, lfsck);
1324         if (IS_ERR(th))
1325                 GOTO(unlock, rc = PTR_ERR(th));
1326
1327         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1328         rec->rec_fid = cfid;
1329         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1330                                (const struct dt_key *)lname->ln_name, th);
1331         if (rc != 0)
1332                 GOTO(stop, rc);
1333
1334         if (S_ISDIR(rec->rec_type)) {
1335                 rc = dt_declare_ref_add(env, parent, th);
1336                 if (rc != 0)
1337                         GOTO(stop, rc);
1338         }
1339
1340         if (parent_lmv_lost) {
1341                 rc = dt_declare_xattr_set(env, parent, &buf, XATTR_NAME_LMV,
1342                                           0, th);
1343                 if (rc)
1344                         GOTO(stop, rc);
1345         }
1346
1347         la->la_ctime = ktime_get_real_seconds();
1348         la->la_valid = LA_CTIME;
1349         rc = dt_declare_attr_set(env, parent, la, th);
1350         if (rc != 0)
1351                 GOTO(stop, rc);
1352
1353         rc = dt_declare_attr_set(env, child, la, th);
1354         if (rc != 0)
1355                 GOTO(stop, rc);
1356
1357         rc = dt_trans_start_local(env, dev, th);
1358         if (rc != 0)
1359                 GOTO(stop, rc);
1360
1361         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1362                        (const struct dt_key *)lname->ln_name, th);
1363         if (rc != 0)
1364                 GOTO(stop, rc);
1365
1366         if (S_ISDIR(rec->rec_type)) {
1367                 dt_write_lock(env, parent, 0);
1368                 rc = dt_ref_add(env, parent, th);
1369                 dt_write_unlock(env, parent);
1370                 if (rc != 0)
1371                         GOTO(stop, rc);
1372         }
1373
1374         if (parent_lmv_lost) {
1375                 rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV, 0, th);
1376                 if (rc)
1377                         GOTO(stop, rc);
1378         }
1379
1380         rc = dt_attr_set(env, parent, la, th);
1381         if (rc != 0)
1382                 GOTO(stop, rc);
1383
1384         rc = dt_attr_set(env, child, la, th);
1385
1386         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1387
1388 stop:
1389         dt_trans_stop(env, dev, th);
1390
1391 unlock:
1392         if (parent_lmv_deleted)
1393                 lfsck_lmv_set(env, lfsck, parent, lmv);
1394
1395         lfsck_unlock(llh);
1396
1397 log:
1398         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1399                "the name %s and type %o to the parent "DFID": rc = %d\n",
1400                lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
1401                lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1402
1403         if (rc != 0) {
1404                 struct lfsck_namespace *ns = com->lc_file_ram;
1405
1406                 ns->ln_flags |= LF_INCONSISTENT;
1407                 if (rc > 0)
1408                         ns->ln_lost_dirent_repaired++;
1409         }
1410
1411         return rc;
1412 }
1413
1414 /**
1415  * Create the specified orphan directory.
1416  *
1417  * For the case that the parent MDT-object stored in some MDT-object's
1418  * linkEA entry is lost, the LFSCK will re-create the parent object as
1419  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1420  * with the name ${FID}-P-${conflict_version}.
1421  *
1422  * \param[in] env       pointer to the thread context
1423  * \param[in] com       pointer to the lfsck component
1424  * \param[in] orphan    pointer to the orphan MDT-object to be created
1425  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1426  *
1427  * \retval              positive number for repaired cases
1428  * \retval              negative error number on failure
1429  */
1430 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1431                                              struct lfsck_component *com,
1432                                              struct dt_object *orphan,
1433                                              struct lmv_mds_md_v1 *lmv)
1434 {
1435         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1436         struct lu_attr                  *la     = &info->lti_la;
1437         struct dt_allocation_hint       *hint   = &info->lti_hint;
1438         struct dt_object_format         *dof    = &info->lti_dof;
1439         struct lu_name                  *cname  = &info->lti_name2;
1440         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1441         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1442         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1443         struct lu_fid                    tfid;
1444         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1445         struct lfsck_namespace          *ns     = com->lc_file_ram;
1446         struct dt_device                *dev    = lfsck_obj2dev(orphan);
1447         struct dt_object                *parent = NULL;
1448         struct thandle                  *th     = NULL;
1449         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1450         struct linkea_data               ldata  = { NULL };
1451         struct lu_buf                    linkea_buf;
1452         struct lu_buf                    lmv_buf;
1453         char                             name[32];
1454         int                              namelen;
1455         int                              idx    = 0;
1456         int                              rc     = 0;
1457         int                              rc1    = 0;
1458         ENTRY;
1459
1460         LASSERT(!dt_object_exists(orphan));
1461
1462         cname->ln_name = NULL;
1463         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1464                 GOTO(log, rc = 1);
1465
1466         if (dt_object_remote(orphan)) {
1467                 if (lfsck->li_lpf_root_obj == NULL)
1468                         GOTO(log, rc = -EBADF);
1469
1470                 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1471                 if (idx < 0)
1472                         GOTO(log, rc = idx);
1473
1474                 snprintf(name, 8, "MDT%04x", idx);
1475                 rc = dt_lookup_dir(env, lfsck->li_lpf_root_obj, name, &tfid);
1476                 if (rc != 0)
1477                         GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1478
1479                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1480                 if (IS_ERR(parent))
1481                         GOTO(log, rc = PTR_ERR(parent));
1482
1483                 if (unlikely(!dt_try_as_dir(env, parent)))
1484                         GOTO(log, rc = -ENOTDIR);
1485         } else {
1486                 if (unlikely(lfsck->li_lpf_obj == NULL))
1487                         GOTO(log, rc = -ENXIO);
1488
1489                 parent = lfsck->li_lpf_obj;
1490         }
1491
1492         dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1493         if (IS_ERR(dev))
1494                 GOTO(log, rc = PTR_ERR(dev));
1495
1496         idx = 0;
1497
1498 again:
1499         do {
1500                 namelen = snprintf(name, 31, DFID"-P-%d",
1501                                    PFID(cfid), idx++);
1502                 rc = dt_lookup_dir(env, parent, name, &tfid);
1503                 if (rc != 0 && rc != -ENOENT)
1504                         GOTO(log, rc);
1505         } while (rc == 0);
1506
1507         rc = lfsck_lock(env, lfsck, parent, name, llh,
1508                         MDS_INODELOCK_UPDATE, LCK_PW);
1509         if (rc != 0)
1510                 GOTO(log, rc);
1511
1512         /* Re-check whether the name conflict with othrs after taken
1513          * the ldlm lock. */
1514         rc = dt_lookup_dir(env, parent, name, &tfid);
1515         if (unlikely(rc == 0)) {
1516                 lfsck_unlock(llh);
1517                 goto again;
1518         }
1519
1520         if (rc != -ENOENT)
1521                 GOTO(unlock1, rc);
1522
1523         cname->ln_name = name;
1524         cname->ln_namelen = namelen;
1525
1526         memset(la, 0, sizeof(*la));
1527         la->la_mode = S_IFDIR | 0700;
1528         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1529                        LA_ATIME | LA_MTIME | LA_CTIME;
1530
1531         orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1532                                    la->la_mode & S_IFMT);
1533
1534         memset(dof, 0, sizeof(*dof));
1535         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1536
1537         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1538                               cname, lfsck_dto2fid(parent));
1539         if (rc != 0)
1540                 GOTO(unlock1, rc);
1541
1542         th = lfsck_trans_create(env, dev, lfsck);
1543         if (IS_ERR(th))
1544                 GOTO(unlock1, rc = PTR_ERR(th));
1545
1546         /* Sync the remote transaction to guarantee that the subsequent
1547          * lock against the @orphan can find the @orphan in time. */
1548         if (dt_object_remote(orphan))
1549                 th->th_sync = 1;
1550
1551         rc = dt_declare_create(env, orphan, la, hint, dof, th);
1552         if (rc != 0)
1553                 GOTO(stop, rc);
1554
1555         if (unlikely(!dt_try_as_dir(env, orphan)))
1556                 GOTO(stop, rc = -ENOTDIR);
1557
1558         rc = dt_declare_ref_add(env, orphan, th);
1559         if (rc != 0)
1560                 GOTO(stop, rc);
1561
1562         rec->rec_type = S_IFDIR;
1563         rec->rec_fid = cfid;
1564         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1565                                (const struct dt_key *)dot, th);
1566         if (rc != 0)
1567                 GOTO(stop, rc);
1568
1569         rec->rec_fid = lfsck_dto2fid(parent);
1570         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1571                                (const struct dt_key *)dotdot, th);
1572         if (rc != 0)
1573                 GOTO(stop, rc);
1574
1575         if (lmv != NULL) {
1576                 lmv->lmv_magic = LMV_MAGIC;
1577                 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1578                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1579                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1580                 rc = dt_declare_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV,
1581                                           0, th);
1582                 if (rc != 0)
1583                         GOTO(stop, rc);
1584         }
1585
1586         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1587                        ldata.ld_leh->leh_len);
1588         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1589                                   XATTR_NAME_LINK, 0, th);
1590         if (rc != 0)
1591                 GOTO(stop, rc);
1592
1593         rec->rec_fid = cfid;
1594         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1595                                (const struct dt_key *)name, th);
1596         if (rc == 0)
1597                 rc = dt_declare_ref_add(env, parent, th);
1598
1599         if (rc != 0)
1600                 GOTO(stop, rc);
1601
1602         rc = dt_trans_start_local(env, dev, th);
1603         if (rc != 0)
1604                 GOTO(stop, rc);
1605
1606         dt_write_lock(env, orphan, 0);
1607         rc = dt_create(env, orphan, la, hint, dof, th);
1608         if (rc != 0)
1609                 GOTO(unlock2, rc);
1610
1611         rc = dt_ref_add(env, orphan, th);
1612         if (rc != 0)
1613                 GOTO(unlock2, rc);
1614
1615         rec->rec_fid = cfid;
1616         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1617                        (const struct dt_key *)dot, th);
1618         if (rc != 0)
1619                 GOTO(unlock2, rc);
1620
1621         rec->rec_fid = lfsck_dto2fid(parent);
1622         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1623                        (const struct dt_key *)dotdot, th);
1624         if (rc != 0)
1625                 GOTO(unlock2, rc);
1626
1627         if (lmv != NULL) {
1628                 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1629                 if (rc != 0)
1630                         GOTO(unlock2, rc);
1631         }
1632
1633         rc = dt_xattr_set(env, orphan, &linkea_buf,
1634                           XATTR_NAME_LINK, 0, th);
1635         dt_write_unlock(env, orphan);
1636         if (rc != 0)
1637                 GOTO(stop, rc);
1638
1639         rec->rec_fid = cfid;
1640         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1641                        (const struct dt_key *)name, th);
1642         if (rc == 0) {
1643                 dt_write_lock(env, parent, 0);
1644                 rc = dt_ref_add(env, parent, th);
1645                 dt_write_unlock(env, parent);
1646         }
1647
1648         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1649
1650 unlock2:
1651         dt_write_unlock(env, orphan);
1652
1653 stop:
1654         rc1 = dt_trans_stop(env, dev, th);
1655         if (rc1 != 0 && rc > 0)
1656                 rc = rc1;
1657
1658 unlock1:
1659         lfsck_unlock(llh);
1660
1661 log:
1662         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1663                "the object "DFID", name = %s: rc = %d\n",
1664                lfsck_lfsck2name(lfsck), PFID(cfid),
1665                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1666
1667         if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1668                 lfsck_object_put(env, parent);
1669
1670         if (rc != 0)
1671                 ns->ln_flags |= LF_INCONSISTENT;
1672
1673         return rc;
1674 }
1675
1676 /**
1677  * Remove the specified entry from the linkEA.
1678  *
1679  * Locate the linkEA entry with the given @cname and @pfid, then
1680  * remove this entry or the other entries those are repeated with
1681  * this entry.
1682  *
1683  * \param[in] env       pointer to the thread context
1684  * \param[in] com       pointer to the lfsck component
1685  * \param[in] obj       pointer to the dt_object to be handled
1686  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1687  * \param[in] cname     the name for the child in the parent directory
1688  * \param[in] pfid      the parent directory's FID for the linkEA
1689  * \param[in] next      if true, then remove the first found linkEA
1690  *                      entry, and move the ldata->ld_lee to next entry
1691  *
1692  * \retval              positive number for repaired cases
1693  * \retval              0 if nothing to be repaired
1694  * \retval              negative error number on failure
1695  */
1696 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1697                                          struct lfsck_component *com,
1698                                          struct dt_object *obj,
1699                                          struct linkea_data *ldata,
1700                                          struct lu_name *cname,
1701                                          struct lu_fid *pfid,
1702                                          bool next)
1703 {
1704         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1705         struct dt_device                *dev       = lfsck_obj2dev(obj);
1706         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1707         struct thandle                  *th        = NULL;
1708         struct lustre_handle             lh        = { 0 };
1709         struct linkea_data               ldata_new = { NULL };
1710         struct lu_buf                    linkea_buf;
1711         int                              buflen    = 0;
1712         int                              rc        = 0;
1713         ENTRY;
1714
1715         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1716                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1717                               LCK_EX);
1718         if (rc != 0)
1719                 GOTO(log, rc);
1720
1721         if (next)
1722                 lfsck_linkea_del_buf(ldata, cname);
1723         else
1724                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1725                                                     true);
1726         if (ldata->ld_leh->leh_reccount > 0 ||
1727             unlikely(ldata->ld_leh->leh_overflow_time)) {
1728                 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1729                                ldata->ld_leh->leh_len);
1730                 buflen = linkea_buf.lb_len;
1731         }
1732
1733 again:
1734         th = lfsck_trans_create(env, dev, lfsck);
1735         if (IS_ERR(th))
1736                 GOTO(unlock1, rc = PTR_ERR(th));
1737
1738         if (buflen != 0)
1739                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1740                                           XATTR_NAME_LINK, 0, th);
1741         else
1742                 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1743         if (rc != 0)
1744                 GOTO(stop, rc);
1745
1746         rc = dt_trans_start_local(env, dev, th);
1747         if (rc != 0)
1748                 GOTO(stop, rc);
1749
1750         dt_write_lock(env, obj, 0);
1751         if (unlikely(lfsck_is_dead_obj(obj)))
1752                 GOTO(unlock2, rc = -ENOENT);
1753
1754         rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1755         if (rc)
1756                 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1757
1758         /* The specified linkEA entry has been removed by race. */
1759         rc = linkea_links_find(&ldata_new, cname, pfid);
1760         if (rc != 0)
1761                 GOTO(unlock2, rc = 0);
1762
1763         if (bk->lb_param & LPF_DRYRUN)
1764                 GOTO(unlock2, rc = 1);
1765
1766         if (next)
1767                 lfsck_linkea_del_buf(&ldata_new, cname);
1768         else
1769                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1770                                                     true);
1771
1772         /*
1773          * linkea may change because it doesn't take lock in the first read, if
1774          * it becomes larger, restart from beginning.
1775          */
1776         if ((ldata_new.ld_leh->leh_reccount > 0 ||
1777              unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
1778             buflen < ldata_new.ld_leh->leh_len) {
1779                 dt_write_unlock(env, obj);
1780                 dt_trans_stop(env, dev, th);
1781                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1782                                ldata_new.ld_leh->leh_len);
1783                 buflen = linkea_buf.lb_len;
1784                 goto again;
1785         }
1786
1787         if (buflen)
1788                 rc = lfsck_links_write(env, obj, &ldata_new, th);
1789         else
1790                 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1791
1792         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1793
1794 unlock2:
1795         dt_write_unlock(env, obj);
1796
1797 stop:
1798         dt_trans_stop(env, dev, th);
1799
1800 unlock1:
1801         lfsck_ibits_unlock(&lh, LCK_EX);
1802
1803 log:
1804         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1805                "for the object: "DFID", parent "DFID", name %.*s\n",
1806                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1807                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1808                cname->ln_name);
1809
1810         if (rc != 0) {
1811                 struct lfsck_namespace *ns = com->lc_file_ram;
1812
1813                 ns->ln_flags |= LF_INCONSISTENT;
1814         }
1815
1816         return rc;
1817 }
1818
1819 /**
1820  * Conditionally remove the specified entry from the linkEA.
1821  *
1822  * Take the parent lock firstly, then check whether the specified
1823  * name entry exists or not: if yes, do nothing; otherwise, call
1824  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1825  *
1826  * \param[in] env       pointer to the thread context
1827  * \param[in] com       pointer to the lfsck component
1828  * \param[in] parent    pointer to the parent directory
1829  * \param[in] child     pointer to the child object that holds the linkEA
1830  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1831  * \param[in] cname     the name for the child in the parent directory
1832  * \param[in] pfid      the parent directory's FID for the linkEA
1833  *
1834  * \retval              positive number for repaired cases
1835  * \retval              0 if nothing to be repaired
1836  * \retval              negative error number on failure
1837  */
1838 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1839                                               struct lfsck_component *com,
1840                                               struct dt_object *parent,
1841                                               struct dt_object *child,
1842                                               struct linkea_data *ldata,
1843                                               struct lu_name *cname,
1844                                               struct lu_fid *pfid)
1845 {
1846         struct lfsck_thread_info *info  = lfsck_env_info(env);
1847         struct lu_fid            *cfid  = &info->lti_fid3;
1848         struct lfsck_lock_handle *llh   = &info->lti_llh;
1849         int                       rc;
1850         ENTRY;
1851
1852         rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1853                         MDS_INODELOCK_UPDATE, LCK_PR);
1854         if (rc != 0)
1855                 RETURN(rc);
1856
1857         dt_read_lock(env, parent, 0);
1858         if (unlikely(lfsck_is_dead_obj(parent))) {
1859                 dt_read_unlock(env, parent);
1860                 lfsck_unlock(llh);
1861                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1862                                                    cname, pfid, true);
1863
1864                 RETURN(rc);
1865         }
1866
1867         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1868                        (const struct dt_key *)cname->ln_name);
1869         dt_read_unlock(env, parent);
1870
1871         /* It is safe to release the ldlm lock, because when the logic come
1872          * here, we have got all the needed information above whether the
1873          * linkEA entry is valid or not. It is not important that others
1874          * may add new linkEA entry after the ldlm lock released. If other
1875          * has removed the specified linkEA entry by race, then it is OK,
1876          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1877          * such case. */
1878         lfsck_unlock(llh);
1879         if (rc == -ENOENT) {
1880                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1881                                                    cname, pfid, true);
1882
1883                 RETURN(rc);
1884         }
1885
1886         if (rc != 0)
1887                 RETURN(rc);
1888
1889         /* The LFSCK just found some internal status of cross-MDTs
1890          * create operation. That is normal. */
1891         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1892                 linkea_next_entry(ldata);
1893
1894                 RETURN(0);
1895         }
1896
1897         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1898                                            pfid, true);
1899
1900         RETURN(rc);
1901 }
1902
1903 /**
1904  * Conditionally replace name entry in the parent.
1905  *
1906  * As required, the LFSCK may re-create the lost MDT-object for dangling
1907  * name entry, but such repairing may be wrong because of bad FID in the
1908  * name entry. As the LFSCK processing, the real MDT-object may be found,
1909  * then the LFSCK should check whether the former re-created MDT-object
1910  * has been modified or not, if not, then destroy it and update the name
1911  * entry in the parent to reference the real MDT-object.
1912  *
1913  * \param[in] env       pointer to the thread context
1914  * \param[in] com       pointer to the lfsck component
1915  * \param[in] parent    pointer to the parent directory
1916  * \param[in] child     pointer to the MDT-object that may be the real
1917  *                      MDT-object corresponding to the name entry in parent
1918  * \param[in] cfid      the current FID in the name entry
1919  * \param[in] cname     contains the name of the child in the parent directory
1920  *
1921  * \retval              positive number for repaired cases
1922  * \retval              0 if nothing to be repaired
1923  * \retval              negative error number on failure
1924  */
1925 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1926                                         struct lfsck_component *com,
1927                                         struct dt_object *parent,
1928                                         struct dt_object *child,
1929                                         const struct lu_fid *cfid,
1930                                         const struct lu_name *cname)
1931 {
1932         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1933         struct lu_attr                  *la     = &info->lti_la;
1934         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1935         struct lu_fid                    tfid;
1936         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1937         /* The child and its name may be on different MDTs. */
1938         struct dt_device                *dev    = lfsck->li_next;
1939         const char                      *name   = cname->ln_name;
1940         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1941         struct dt_object                *cobj   = NULL;
1942         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
1943         struct lustre_handle             clh    = { 0 };
1944         struct linkea_data               ldata  = { NULL };
1945         struct thandle                  *th     = NULL;
1946         bool                             exist  = true;
1947         int                              rc     = 0;
1948         ENTRY;
1949
1950         /* @parent/@child may be based on lfsck->li_bottom,
1951          * but here we need the object based on the lfsck->li_next. */
1952
1953         parent = lfsck_object_locate(dev, parent);
1954         if (IS_ERR(parent))
1955                 GOTO(log, rc = PTR_ERR(parent));
1956
1957         if (unlikely(!dt_try_as_dir(env, parent)))
1958                 GOTO(log, rc = -ENOTDIR);
1959
1960         rc = lfsck_lock(env, lfsck, parent, name, pllh,
1961                         MDS_INODELOCK_UPDATE, LCK_PW);
1962         if (rc != 0)
1963                 GOTO(log, rc);
1964
1965         if (!fid_is_sane(cfid)) {
1966                 exist = false;
1967                 goto replace;
1968         }
1969
1970         cobj = lfsck_object_find_by_dev(env, dev, cfid);
1971         if (IS_ERR(cobj)) {
1972                 rc = PTR_ERR(cobj);
1973                 if (rc == -ENOENT) {
1974                         exist = false;
1975                         goto replace;
1976                 }
1977
1978                 GOTO(log, rc);
1979         }
1980
1981         if (!dt_object_exists(cobj)) {
1982                 exist = false;
1983                 goto replace;
1984         }
1985
1986         rc = dt_lookup_dir(env, parent, name, &tfid);
1987         if (rc == -ENOENT) {
1988                 exist = false;
1989                 goto replace;
1990         }
1991
1992         if (rc != 0)
1993                 GOTO(log, rc);
1994
1995         /* Someone changed the name entry, cannot replace it. */
1996         if (!lu_fid_eq(cfid, &tfid))
1997                 GOTO(log, rc = 0);
1998
1999         /* lock the object to be destroyed. */
2000         rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
2001                               MDS_INODELOCK_UPDATE |
2002                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
2003                               LCK_EX);
2004         if (rc != 0)
2005                 GOTO(log, rc);
2006
2007         if (unlikely(lfsck_is_dead_obj(cobj))) {
2008                 exist = false;
2009                 goto replace;
2010         }
2011
2012         rc = dt_attr_get(env, cobj, la);
2013         if (rc != 0)
2014                 GOTO(log, rc);
2015
2016         /* The object has been modified by other(s), or it is not created by
2017          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
2018         if (la->la_ctime != 0)
2019                 GOTO(log, rc);
2020
2021         if (S_ISREG(la->la_mode)) {
2022                 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
2023                 /* If someone has created related OST-object(s),
2024                  * then keep it. */
2025                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
2026                         GOTO(log, rc = (rc > 0 ? 0 : rc));
2027         }
2028
2029 replace:
2030         dt_read_lock(env, child, 0);
2031         rc = lfsck_links_read2_with_rec(env, child, &ldata);
2032         dt_read_unlock(env, child);
2033
2034         /* Someone changed the child, no need to replace. */
2035         if (rc == -ENODATA)
2036                 GOTO(log, rc = 0);
2037
2038         if (rc != 0)
2039                 GOTO(log, rc);
2040
2041         rc = linkea_links_find(&ldata, cname, pfid);
2042         /* Someone moved the child, no need to replace. */
2043         if (rc != 0)
2044                 GOTO(log, rc = 0);
2045
2046         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2047                 GOTO(log, rc = 1);
2048
2049         th = lfsck_trans_create(env, dev, lfsck);
2050         if (IS_ERR(th))
2051                 GOTO(log, rc = PTR_ERR(th));
2052
2053         if (exist) {
2054                 rc = dt_declare_destroy(env, cobj, th);
2055                 if (rc != 0)
2056                         GOTO(stop, rc);
2057         }
2058
2059         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2060         if (rc != 0)
2061                 GOTO(stop, rc);
2062
2063         rec->rec_type = S_IFDIR;
2064         rec->rec_fid = lfsck_dto2fid(child);
2065         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
2066                                (const struct dt_key *)name, th);
2067         if (rc != 0)
2068                 GOTO(stop, rc);
2069
2070         rc = dt_trans_start_local(env, dev, th);
2071         if (rc != 0)
2072                 GOTO(stop, rc);
2073
2074         if (exist) {
2075                 rc = dt_destroy(env, cobj, th);
2076                 if (rc != 0)
2077                         GOTO(stop, rc);
2078         }
2079
2080         /* The old name entry maybe not exist. */
2081         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2082         if (rc != 0 && rc != -ENOENT)
2083                 GOTO(stop, rc);
2084
2085         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
2086                        (const struct dt_key *)name, th);
2087
2088         GOTO(stop, rc = (rc == 0 ? 1 : rc));
2089
2090 stop:
2091         dt_trans_stop(env, dev, th);
2092
2093 log:
2094         lfsck_ibits_unlock(&clh, LCK_EX);
2095         lfsck_unlock(pllh);
2096
2097         if (cobj != NULL && !IS_ERR(cobj))
2098                 lfsck_object_put(env, cobj);
2099
2100         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
2101                "object "DFID" because of conflict with the object "DFID
2102                " under the parent "DFID" with name %s: rc = %d\n",
2103                lfsck_lfsck2name(lfsck), PFID(cfid),
2104                PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
2105
2106         return rc;
2107 }
2108
2109 /**
2110  * Overwrite the linkEA for the object with the given ldata.
2111  *
2112  * The caller should take the ldlm lock before the calling.
2113  *
2114  * \param[in] env       pointer to the thread context
2115  * \param[in] com       pointer to the lfsck component
2116  * \param[in] obj       pointer to the dt_object to be handled
2117  * \param[in] ldata     pointer to the new linkEA data
2118  *
2119  * \retval              positive number for repaired cases
2120  * \retval              0 if nothing to be repaired
2121  * \retval              negative error number on failure
2122  */
2123 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
2124                                    struct lfsck_component *com,
2125                                    struct dt_object *obj,
2126                                    struct linkea_data *ldata)
2127 {
2128         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2129         struct dt_device                *dev    = lfsck_obj2dev(obj);
2130         struct thandle                  *th     = NULL;
2131         struct lu_buf                    linkea_buf;
2132         int                              rc     = 0;
2133         ENTRY;
2134
2135         th = lfsck_trans_create(env, dev, lfsck);
2136         if (IS_ERR(th))
2137                 GOTO(log, rc = PTR_ERR(th));
2138
2139         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
2140                        ldata->ld_leh->leh_len);
2141         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2142                                   XATTR_NAME_LINK, 0, th);
2143         if (rc != 0)
2144                 GOTO(stop, rc);
2145
2146         rc = dt_trans_start_local(env, dev, th);
2147         if (rc != 0)
2148                 GOTO(stop, rc);
2149
2150         dt_write_lock(env, obj, 0);
2151         if (unlikely(lfsck_is_dead_obj(obj)))
2152                 GOTO(unlock, rc = 0);
2153
2154         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2155                 GOTO(unlock, rc = 1);
2156
2157         rc = dt_xattr_set(env, obj, &linkea_buf,
2158                           XATTR_NAME_LINK, 0, th);
2159
2160         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2161
2162 unlock:
2163         dt_write_unlock(env, obj);
2164
2165 stop:
2166         dt_trans_stop(env, dev, th);
2167
2168 log:
2169         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2170                "object "DFID": rc = %d\n",
2171                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2172
2173         if (rc != 0) {
2174                 struct lfsck_namespace *ns = com->lc_file_ram;
2175
2176                 ns->ln_flags |= LF_INCONSISTENT;
2177         }
2178
2179         return rc;
2180 }
2181
2182 /**
2183  * Repair invalid name entry.
2184  *
2185  * If the name entry contains invalid information, such as bad file type
2186  * or (and) corrupted object FID, then either remove the name entry or
2187  * udpate the name entry with the given (right) information.
2188  *
2189  * \param[in] env       pointer to the thread context
2190  * \param[in] com       pointer to the lfsck component
2191  * \param[in] parent    pointer to the parent directory
2192  * \param[in] child     pointer to the object referenced by the name entry
2193  * \param[in] name      the old name of the child under the parent directory
2194  * \param[in] name2     the new name of the child under the parent directory
2195  * \param[in] type      the type claimed by the name entry
2196  * \param[in] update    update the name entry if true; otherwise, remove it
2197  * \param[in] dec       decrease the parent nlink count if true
2198  *
2199  * \retval              positive number for repaired successfully
2200  * \retval              0 if nothing to be repaired
2201  * \retval              negative error number on failure
2202  */
2203 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2204                                   struct lfsck_component *com,
2205                                   struct dt_object *parent,
2206                                   struct dt_object *child,
2207                                   const char *name, const char *name2,
2208                                   __u16 type, bool update, bool dec)
2209 {
2210         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2211         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2212         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2213         struct lu_fid                   cfid    = {0};
2214         struct lu_fid                    tfid;
2215         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2216         struct dt_device                *dev    = lfsck->li_next;
2217         struct thandle                  *th     = NULL;
2218         struct lfsck_lock_handle        *llh    = &info->lti_llh;
2219         struct lustre_handle             lh     = { 0 };
2220         int                              rc     = 0;
2221         ENTRY;
2222
2223         if (child)
2224                 cfid = *lfsck_dto2fid(child);
2225         parent = lfsck_object_locate(dev, parent);
2226         if (IS_ERR(parent))
2227                 GOTO(log, rc = PTR_ERR(parent));
2228
2229         if (unlikely(!dt_try_as_dir(env, parent)))
2230                 GOTO(log, rc = -ENOTDIR);
2231
2232         if (!update || strcmp(name, name2) == 0)
2233                 rc = lfsck_lock(env, lfsck, parent, name, llh,
2234                                 MDS_INODELOCK_UPDATE, LCK_PW);
2235         else
2236                 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2237                                       MDS_INODELOCK_UPDATE, LCK_PW);
2238         if (rc != 0)
2239                 GOTO(log, rc);
2240
2241         th = lfsck_trans_create(env, dev, lfsck);
2242         if (IS_ERR(th))
2243                 GOTO(unlock1, rc = PTR_ERR(th));
2244
2245         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2246         if (rc != 0)
2247                 GOTO(stop, rc);
2248
2249         if (update) {
2250                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2251                 LASSERT(!fid_is_zero(&cfid));
2252                 rec->rec_fid = &cfid;
2253                 rc = dt_declare_insert(env, parent,
2254                                        (const struct dt_rec *)rec,
2255                                        (const struct dt_key *)name2, th);
2256                 if (rc != 0)
2257                         GOTO(stop, rc);
2258         }
2259
2260         if (dec && S_ISDIR(type)) {
2261                 rc = dt_declare_ref_del(env, parent, th);
2262                 if (rc != 0)
2263                         GOTO(stop, rc);
2264         }
2265
2266         rc = dt_trans_start_local(env, dev, th);
2267         if (rc != 0)
2268                 GOTO(stop, rc);
2269
2270
2271         dt_write_lock(env, parent, 0);
2272         rc = dt_lookup_dir(env, dt_object_child(parent), name, &tfid);
2273         /* Someone has removed the bad name entry by race. */
2274         if (rc == -ENOENT)
2275                 GOTO(unlock2, rc = 0);
2276
2277         if (rc != 0)
2278                 GOTO(unlock2, rc);
2279
2280         /* Someone has removed the bad name entry and reused it for other
2281          * object by race. */
2282         if (!lu_fid_eq(&tfid, &cfid))
2283                 GOTO(unlock2, rc = 0);
2284
2285         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2286                 GOTO(unlock2, rc = 1);
2287
2288         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2289         if (rc != 0)
2290                 GOTO(unlock2, rc);
2291
2292         if (update) {
2293                 rc = dt_insert(env, parent,
2294                                (const struct dt_rec *)rec,
2295                                (const struct dt_key *)name2, th);
2296                 if (rc != 0)
2297                         GOTO(unlock2, rc);
2298         }
2299
2300         if (dec && S_ISDIR(type)) {
2301                 rc = dt_ref_del(env, parent, th);
2302                 if (rc != 0)
2303                         GOTO(unlock2, rc);
2304         }
2305
2306         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2307
2308 unlock2:
2309         dt_write_unlock(env, parent);
2310
2311 stop:
2312         dt_trans_stop(env, dev, th);
2313
2314         /* We are not sure whether the child will become orphan or not.
2315          * Record it in the LFSCK trace file for further checking in
2316          * the second-stage scanning. */
2317         if (!update && !dec && child && rc == 0)
2318                 lfsck_namespace_trace_update(env, com, &cfid,
2319                                              LNTF_CHECK_LINKEA, true);
2320
2321 unlock1:
2322         /* It is harmless even if unlock the unused lock_handle */
2323         lfsck_ibits_unlock(&lh, LCK_PW);
2324         lfsck_unlock(llh);
2325
2326 log:
2327         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2328                "entry for: parent "DFID", child "DFID", name %s, type "
2329                "in name entry %o, type claimed by child %o. repair it "
2330                "by %s with new name2 %s: rc = %d\n",
2331                lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2332                name, type, update ? lfsck_object_type(child) : 0,
2333                update ? "updating" : "removing", name2, rc);
2334
2335         if (rc != 0) {
2336                 struct lfsck_namespace *ns = com->lc_file_ram;
2337
2338                 ns->ln_flags |= LF_INCONSISTENT;
2339         }
2340
2341         return rc;
2342 }
2343
2344 /**
2345  * Update the ".." name entry for the given object.
2346  *
2347  * The object's ".." is corrupted, this function will update the ".." name
2348  * entry with the given pfid, and the linkEA with the given ldata.
2349  *
2350  * The caller should take the ldlm lock before the calling.
2351  *
2352  * \param[in] env       pointer to the thread context
2353  * \param[in] com       pointer to the lfsck component
2354  * \param[in] obj       pointer to the dt_object to be handled
2355  * \param[in] pfid      the new fid for the object's ".." name entry
2356  * \param[in] cname     the name for the @obj in the parent directory
2357  *
2358  * \retval              positive number for repaired cases
2359  * \retval              0 if nothing to be repaired
2360  * \retval              negative error number on failure
2361  */
2362 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2363                                                   struct lfsck_component *com,
2364                                                   struct dt_object *obj,
2365                                                   const struct lu_fid *pfid,
2366                                                   struct lu_name *cname)
2367 {
2368         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2369         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2370         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2371         struct dt_device                *dev    = lfsck_obj2dev(obj);
2372         struct thandle                  *th     = NULL;
2373         struct linkea_data               ldata  = { NULL };
2374         struct lu_buf                    linkea_buf;
2375         int                              rc     = 0;
2376         ENTRY;
2377
2378         LASSERT(!dt_object_remote(obj));
2379         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2380
2381         rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2382         if (rc != 0)
2383                 GOTO(log, rc);
2384
2385         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2386                        ldata.ld_leh->leh_len);
2387
2388         th = lfsck_trans_create(env, dev, lfsck);
2389         if (IS_ERR(th))
2390                 GOTO(log, rc = PTR_ERR(th));
2391
2392         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2393         if (rc != 0)
2394                 GOTO(stop, rc);
2395
2396         rec->rec_type = S_IFDIR;
2397         rec->rec_fid = pfid;
2398         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2399                                (const struct dt_key *)dotdot, th);
2400         if (rc != 0)
2401                 GOTO(stop, rc);
2402
2403         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2404                                   XATTR_NAME_LINK, 0, th);
2405         if (rc != 0)
2406                 GOTO(stop, rc);
2407
2408         rc = dt_trans_start_local(env, dev, th);
2409         if (rc != 0)
2410                 GOTO(stop, rc);
2411
2412         dt_write_lock(env, obj, 0);
2413         if (unlikely(lfsck_is_dead_obj(obj)))
2414                 GOTO(unlock, rc = 0);
2415
2416         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2417                 GOTO(unlock, rc = 1);
2418
2419         /* The old ".." name entry maybe not exist. */
2420         dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2421
2422         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2423                        (const struct dt_key *)dotdot, th);
2424         if (rc != 0)
2425                 GOTO(unlock, rc);
2426
2427         rc = lfsck_links_write(env, obj, &ldata, th);
2428
2429         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2430
2431 unlock:
2432         dt_write_unlock(env, obj);
2433
2434 stop:
2435         dt_trans_stop(env, dev, th);
2436
2437 log:
2438         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2439                "the object "DFID", new parent "DFID": rc = %d\n",
2440                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2441                PFID(pfid), rc);
2442
2443         if (rc != 0) {
2444                 struct lfsck_namespace *ns = com->lc_file_ram;
2445
2446                 ns->ln_flags |= LF_INCONSISTENT;
2447         }
2448
2449         return rc;
2450 }
2451
2452 /**
2453  * Handle orphan @obj during Double Scan Directory.
2454  *
2455  * Remove the @obj's current (invalid) linkEA entries, and insert
2456  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2457  * ${FID}-${PFID}-D-${conflict_version}
2458  *
2459  * The caller should take the ldlm lock before the calling.
2460  *
2461  * \param[in] env       pointer to the thread context
2462  * \param[in] com       pointer to the lfsck component
2463  * \param[in] obj       pointer to the orphan object to be handled
2464  * \param[in] pfid      the new fid for the object's ".." name entry
2465  * \param[in,out] lh    ldlm lock handler for the given @obj
2466  * \param[out] type     to tell the caller what the inconsistency is
2467  *
2468  * \retval              positive number for repaired cases
2469  * \retval              0 if nothing to be repaired
2470  * \retval              negative error number on failure
2471  */
2472 static int
2473 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2474                            struct lfsck_component *com,
2475                            struct dt_object *obj,
2476                            const struct lu_fid *pfid,
2477                            struct lustre_handle *lh,
2478                            enum lfsck_namespace_inconsistency_type *type)
2479 {
2480         struct lfsck_thread_info *info = lfsck_env_info(env);
2481         struct lfsck_namespace   *ns   = com->lc_file_ram;
2482         int                       rc;
2483         ENTRY;
2484
2485         /* Remove the unrecognized linkEA. */
2486         rc = lfsck_namespace_links_remove(env, com, obj);
2487         lfsck_ibits_unlock(lh, LCK_EX);
2488         if (rc < 0 && rc != -ENODATA)
2489                 RETURN(rc);
2490
2491         *type = LNIT_MUL_REF;
2492
2493         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2494          * ever tried to verify some remote MDT-object that resides on this
2495          * MDT, but this MDT failed to respond such request. So means there
2496          * may be some remote name entry on other MDT that references this
2497          * object with another name, so we cannot know whether this linkEA
2498          * is valid or not. So keep it there and maybe resolved when next
2499          * LFSCK run. */
2500         if (ns->ln_flags & LF_INCOMPLETE)
2501                 RETURN(0);
2502
2503         /* The unique linkEA is invalid, even if the ".." name entry may be
2504          * valid, we still cannot know via which name entry this directory
2505          * will be referenced. Then handle it as pure orphan. */
2506         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2507                  "-"DFID, PFID(pfid));
2508         rc = lfsck_namespace_insert_orphan(env, com, obj,
2509                                            info->lti_tmpbuf, "D", NULL);
2510
2511         RETURN(rc);
2512 }
2513
2514 /**
2515  * Double Scan Directory object for single linkEA entry case.
2516  *
2517  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2518  * then check whether the name is in the namespace or not, if not, add the
2519  * missing name entry back to namespace. If the linkEA entry is invalid,
2520  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2521  * as an orphan.
2522  *
2523  * \param[in] env       pointer to the thread context
2524  * \param[in] com       pointer to the lfsck component
2525  * \param[in] child     pointer to the directory to be double scanned
2526  * \param[in] pfid      the FID corresponding to the ".." entry
2527  * \param[in] ldata     pointer to the linkEA data for the given @child
2528  * \param[in,out] lh    ldlm lock handler for the given @child
2529  * \param[out] type     to tell the caller what the inconsistency is
2530  * \param[in] retry     if found inconsistency, but the caller does not hold
2531  *                      ldlm lock on the @child, then set @retry as true
2532  * \param[in] unknown   set if does not know how to repair the inconsistency
2533  *
2534  * \retval              positive number for repaired cases
2535  * \retval              0 if nothing to be repaired
2536  * \retval              negative error number on failure
2537  */
2538 static int
2539 lfsck_namespace_dsd_single(const struct lu_env *env,
2540                            struct lfsck_component *com,
2541                            struct dt_object *child,
2542                            const struct lu_fid *pfid,
2543                            struct linkea_data *ldata,
2544                            struct lustre_handle *lh,
2545                            enum lfsck_namespace_inconsistency_type *type,
2546                            bool *retry, bool *unknown)
2547 {
2548         struct lfsck_thread_info *info          = lfsck_env_info(env);
2549         struct lu_name           *cname         = &info->lti_name;
2550         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2551         struct lu_fid             tfid;
2552         struct lfsck_namespace   *ns            = com->lc_file_ram;
2553         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2554         struct dt_object         *parent        = NULL;
2555         struct lmv_mds_md_v1     *lmv;
2556         int                       rc            = 0;
2557         ENTRY;
2558
2559         rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2560                                                  info->lti_key,
2561                                                  sizeof(info->lti_key));
2562         /* The unique linkEA entry with bad parent will be handled as orphan. */
2563         if (rc != 0) {
2564                 if (!lustre_handle_is_used(lh) && retry != NULL)
2565                         *retry = true;
2566                 else
2567                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2568                                                         pfid, lh, type);
2569
2570                 GOTO(out, rc);
2571         }
2572
2573         parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2574         if (IS_ERR(parent))
2575                 GOTO(out, rc = PTR_ERR(parent));
2576
2577         /* We trust the unique linkEA entry in spite of whether it matches the
2578          * ".." name entry or not. Because even if the linkEA entry is wrong
2579          * and the ".." name entry is right, we still cannot know via which
2580          * name entry the child will be referenced, since all known entries
2581          * have been verified during the first-stage scanning. */
2582         if (!dt_object_exists(parent)) {
2583                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2584                  * has ever tried to verify some remote MDT-object that resides
2585                  * on this MDT, but this MDT failed to respond such request. So
2586                  * means there may be some remote name entry on other MDT that
2587                  * references this object with another name, so we cannot know
2588                  * whether this linkEA is valid or not. So keep it there and
2589                  * maybe resolved when next LFSCK run. */
2590                 if (ns->ln_flags & LF_INCOMPLETE)
2591                         GOTO(out, rc = 0);
2592
2593                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2594                         *retry = true;
2595
2596                         GOTO(out, rc = 0);
2597                 }
2598
2599                 lfsck_ibits_unlock(lh, LCK_EX);
2600
2601 lost_parent:
2602                 lmv = &info->lti_lmv;
2603                 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
2604                 if (rc != 0 && rc != -ENODATA)
2605                         GOTO(out, rc);
2606
2607                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2608                         lmv = NULL;
2609                 } else if (lfsck_shard_name_to_index(env,
2610                                         cname->ln_name, cname->ln_namelen,
2611                                         S_IFDIR, cfid) < 0) {
2612                         /* It is an invalid name entry, we
2613                          * cannot trust the parent also. */
2614                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2615                                                 ldata, cname, &tfid, true);
2616                         if (rc < 0)
2617                                 GOTO(out, rc);
2618
2619                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2620                                  "-"DFID, PFID(pfid));
2621                         rc = lfsck_namespace_insert_orphan(env, com, child,
2622                                                 info->lti_tmpbuf, "S", NULL);
2623
2624                         GOTO(out, rc);
2625                 }
2626
2627                 /* Create the lost parent as an orphan. */
2628                 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2629                 if (rc >= 0) {
2630                         /* Add the missing name entry to the parent. */
2631                         rc = lfsck_namespace_insert_normal(env, com, parent,
2632                                                            child, cname);
2633                         if (unlikely(rc == -EEXIST)) {
2634                                 /* Unfortunately, someone reused the name
2635                                  * under the parent by race. So we have
2636                                  * to remove the linkEA entry from
2637                                  * current child object. It means that the
2638                                  * LFSCK cannot recover the system
2639                                  * totally back to its original status,
2640                                  * but it is necessary to make the
2641                                  * current system to be consistent. */
2642                                 rc = lfsck_namespace_shrink_linkea(env,
2643                                                 com, child, ldata,
2644                                                 cname, &tfid, true);
2645                                 if (rc >= 0) {
2646                                         snprintf(info->lti_tmpbuf,
2647                                                  sizeof(info->lti_tmpbuf),
2648                                                  "-"DFID, PFID(pfid));
2649                                         rc = lfsck_namespace_insert_orphan(env,
2650                                                 com, child, info->lti_tmpbuf,
2651                                                 "D", NULL);
2652                                 }
2653                         }
2654                 }
2655
2656                 GOTO(out, rc);
2657         } /* !dt_object_exists(parent) */
2658
2659         /* The unique linkEA entry with bad parent will be handled as orphan. */
2660         if (unlikely(!dt_try_as_dir(env, parent))) {
2661                 if (!lustre_handle_is_used(lh) && retry != NULL)
2662                         *retry = true;
2663                 else
2664                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2665                                                         pfid, lh, type);
2666
2667                 GOTO(out, rc);
2668         }
2669
2670         rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
2671         if (rc == -ENOENT) {
2672                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2673                  * has ever tried to verify some remote MDT-object that resides
2674                  * on this MDT, but this MDT failed to respond such request. So
2675                  * means there may be some remote name entry on other MDT that
2676                  * references this object with another name, so we cannot know
2677                  * whether this linkEA is valid or not. So keep it there and
2678                  * maybe resolved when next LFSCK run. */
2679                 if (ns->ln_flags & LF_INCOMPLETE)
2680                         GOTO(out, rc = 0);
2681
2682                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2683                         *retry = true;
2684
2685                         GOTO(out, rc = 0);
2686                 }
2687
2688                 lfsck_ibits_unlock(lh, LCK_EX);
2689                 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
2690                                                 cname);
2691                 if (rc == -ENOENT)
2692                         goto lost_parent;
2693
2694                 if (rc < 0)
2695                         GOTO(out, rc);
2696
2697                 /* It is an invalid name entry, drop it. */
2698                 if (unlikely(rc > 0)) {
2699                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2700                                                 ldata, cname, &tfid, true);
2701                         if (rc >= 0) {
2702                                 snprintf(info->lti_tmpbuf,
2703                                          sizeof(info->lti_tmpbuf),
2704                                          "-"DFID, PFID(pfid));
2705                                 rc = lfsck_namespace_insert_orphan(env, com,
2706                                         child, info->lti_tmpbuf, "D", NULL);
2707                         }
2708
2709                         GOTO(out, rc);
2710                 }
2711
2712                 /* Add the missing name entry back to the namespace. */
2713                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2714                                                    cname);
2715                 if (unlikely(rc == -ESTALE))
2716                         /* It may happen when the remote object has been
2717                          * removed, but the local MDT is not aware of that. */
2718                         goto lost_parent;
2719
2720                 if (unlikely(rc == -EEXIST)) {
2721                         /* Unfortunately, someone reused the name under the
2722                          * parent by race. So we have to remove the linkEA
2723                          * entry from current child object. It means that the
2724                          * LFSCK cannot recover the system totally back to
2725                          * its original status, but it is necessary to make
2726                          * the current system to be consistent.
2727                          *
2728                          * It also may be because of the LFSCK found some
2729                          * internal status of create operation. Under such
2730                          * case, nothing to be done. */
2731                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2732                                         parent, child, ldata, cname, &tfid);
2733                         if (rc >= 0) {
2734                                 snprintf(info->lti_tmpbuf,
2735                                          sizeof(info->lti_tmpbuf),
2736                                          "-"DFID, PFID(pfid));
2737                                 rc = lfsck_namespace_insert_orphan(env, com,
2738                                         child, info->lti_tmpbuf, "D", NULL);
2739                         }
2740                 }
2741
2742                 GOTO(out, rc);
2743         } /* rc == -ENOENT */
2744
2745         if (rc != 0)
2746                 GOTO(out, rc);
2747
2748         if (!lu_fid_eq(&tfid, cfid)) {
2749                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2750                         *retry = true;
2751
2752                         GOTO(out, rc = 0);
2753                 }
2754
2755                 lfsck_ibits_unlock(lh, LCK_EX);
2756                 /* The name entry references another MDT-object that
2757                  * may be created by the LFSCK for repairing dangling
2758                  * name entry. Try to replace it. */
2759                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2760                                                   &tfid, cname);
2761                 if (rc == 0)
2762                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2763                                                         pfid, lh, type);
2764
2765                 GOTO(out, rc);
2766         }
2767
2768         /* Zero FID may because the remote directroy object has invalid linkEA,
2769          * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2770          * how to repair the inconsistency, but the namespace LFSCK on the MDT
2771          * where its name entry resides may has more information (name, FID) to
2772          * repair such inconsistency. So here, keep the inconsistency to avoid
2773          * some imporper repairing. */
2774         if (fid_is_zero(pfid)) {
2775                 if (unknown)
2776                         *unknown = true;
2777
2778                 GOTO(out, rc = 0);
2779         }
2780
2781         /* The ".." name entry is wrong, update it. */
2782         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2783                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2784                         *retry = true;
2785
2786                         GOTO(out, rc = 0);
2787                 }
2788
2789                 *type = LNIT_UNMATCHED_PAIRS;
2790                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2791                                                 lfsck_dto2fid(parent), cname);
2792         }
2793
2794         GOTO(out, rc);
2795
2796 out:
2797         if (parent != NULL && !IS_ERR(parent))
2798                 lfsck_object_put(env, parent);
2799
2800         return rc;
2801 }
2802
2803 /**
2804  * Double Scan Directory object for multiple linkEA entries case.
2805  *
2806  * The given @child has multiple linkEA entries. There is at most one linkEA
2807  * entry will be valid, all the others will be removed. Firstly, the function
2808  * will try to find out the linkEA entry for which the name entry exists under
2809  * the given parent (@pfid). If there is no linkEA entry that matches the given
2810  * ".." name entry, then tries to find out the first linkEA entry that both the
2811  * parent and the name entry exist to rebuild a new ".." name entry.
2812  *
2813  * \param[in] env       pointer to the thread context
2814  * \param[in] com       pointer to the lfsck component
2815  * \param[in] child     pointer to the directory to be double scanned
2816  * \param[in] pfid      the FID corresponding to the ".." entry
2817  * \param[in] ldata     pointer to the linkEA data for the given @child
2818  * \param[in,out] lh    ldlm lock handler for the given @child
2819  * \param[out] type     to tell the caller what the inconsistency is
2820  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2821  * \param[in] unknown   set if does not know how to repair the inconsistency
2822  *
2823  * \retval              positive number for repaired cases
2824  * \retval              0 if nothing to be repaired
2825  * \retval              negative error number on failure
2826  */
2827 static int
2828 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2829                              struct lfsck_component *com,
2830                              struct dt_object *child,
2831                              const struct lu_fid *pfid,
2832                              struct linkea_data *ldata,
2833                              struct lustre_handle *lh,
2834                              enum lfsck_namespace_inconsistency_type *type,
2835                              bool lpf, bool *unknown)
2836 {
2837         struct lfsck_thread_info *info          = lfsck_env_info(env);
2838         struct lu_name           *cname         = &info->lti_name;
2839         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2840         struct lu_fid            *pfid2         = &info->lti_fid3;
2841         struct lu_fid             tfid;
2842         struct lfsck_namespace   *ns            = com->lc_file_ram;
2843         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2844         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2845         struct dt_object         *parent        = NULL;
2846         struct linkea_data        ldata_new     = { NULL };
2847         int                       dirent_count  = 0;
2848         int                       rc            = 0;
2849         bool                      once          = true;
2850         ENTRY;
2851
2852 again:
2853         while (ldata->ld_lee != NULL) {
2854                 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2855                                                          info->lti_key,
2856                                                          sizeof(info->lti_key));
2857                 /* Drop invalid linkEA entry. */
2858                 if (rc != 0) {
2859                         lfsck_linkea_del_buf(ldata, cname);
2860                         continue;
2861                 }
2862
2863                 /* Drop repeated linkEA entries. */
2864                 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2865
2866                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2867                  * then it is possible that: the directry object has ever
2868                  * been lost, but its name entry was there. In the former
2869                  * LFSCK run, during the first-stage scanning, the LFSCK
2870                  * found the dangling name entry, but it did not recreate
2871                  * the lost object, and when moved to the second-stage
2872                  * scanning, some children objects of the lost directory
2873                  * object were found, then the LFSCK recreated such lost
2874                  * directory object as an orphan.
2875                  *
2876                  * When the LFSCK runs again, if the dangling name is still
2877                  * there, the LFSCK should move the orphan directory object
2878                  * back to the normal namespace. */
2879                 if (!lpf && !fid_is_zero(pfid) &&
2880                     !lu_fid_eq(pfid, &tfid) && once) {
2881                         linkea_next_entry(ldata);
2882                         continue;
2883                 }
2884
2885                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2886                 if (IS_ERR(parent)) {
2887                         rc = PTR_ERR(parent);
2888                         /* if @pfid doesn't have a valid OI mapping, it will
2889                          * trigger OI scrub, and -ENONET is is returned if it's
2890                          * remote, -EINPROGRESS if local.
2891                          */
2892                         if ((rc == -ENOENT || rc == -EINPROGRESS) &&
2893                             ldata->ld_leh->leh_reccount > 1) {
2894                                 lfsck_linkea_del_buf(ldata, cname);
2895                                 continue;
2896                         }
2897
2898                         RETURN(rc);
2899                 }
2900
2901                 if (!dt_object_exists(parent)) {
2902                         lfsck_object_put(env, parent);
2903                         if (ldata->ld_leh->leh_reccount > 1) {
2904                                 /* If it is NOT the last linkEA entry, then
2905                                  * there is still other chance to make the
2906                                  * child to be visible via other parent, then
2907                                  * remove this linkEA entry. */
2908                                 lfsck_linkea_del_buf(ldata, cname);
2909                                 continue;
2910                         }
2911
2912                         break;
2913                 }
2914
2915                 /* The linkEA entry with bad parent will be removed. */
2916                 if (unlikely(!dt_try_as_dir(env, parent))) {
2917                         lfsck_object_put(env, parent);
2918                         lfsck_linkea_del_buf(ldata, cname);
2919                         continue;
2920                 }
2921
2922                 rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
2923                 *pfid2 = *lfsck_dto2fid(parent);
2924                 if (rc == -ENOENT) {
2925                         lfsck_object_put(env, parent);
2926                         linkea_next_entry(ldata);
2927                         continue;
2928                 }
2929
2930                 if (rc != 0) {
2931                         lfsck_object_put(env, parent);
2932
2933                         RETURN(rc);
2934                 }
2935
2936                 if (lu_fid_eq(&tfid, cfid)) {
2937                         lfsck_object_put(env, parent);
2938                         /* If the parent (that is declared via linkEA entry)
2939                          * directory contains the specified child, but such
2940                          * parent does not match the dotdot name entry, then
2941                          * trust the linkEA. */
2942                         if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2943                                 *type = LNIT_UNMATCHED_PAIRS;
2944                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2945                                                 com, child, pfid2, cname);
2946
2947                                 RETURN(rc);
2948                         }
2949
2950 rebuild:
2951                         /* It is the most common case that we find the
2952                          * name entry corresponding to the linkEA entry
2953                          * that matches the ".." name entry. */
2954                         rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2955                                               cname, pfid2);
2956                         if (rc != 0)
2957                                 RETURN(rc);
2958
2959                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2960                                                             &ldata_new);
2961                         if (rc < 0)
2962                                 RETURN(rc);
2963
2964                         lfsck_linkea_del_buf(ldata, cname);
2965                         linkea_first_entry(ldata);
2966                         /* There may be some invalid dangling name entries under
2967                          * other parent directories, remove all of them. */
2968                         while (ldata->ld_lee != NULL) {
2969                                 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2970                                                 cname, &tfid, info->lti_key,
2971                                                 sizeof(info->lti_key));
2972                                 if (rc != 0)
2973                                         goto next;
2974
2975                                 parent = lfsck_object_find_bottom(env, lfsck,
2976                                                                   &tfid);
2977                                 if (IS_ERR(parent)) {
2978                                         rc = PTR_ERR(parent);
2979                                         if (rc != -ENOENT &&
2980                                             bk->lb_param & LPF_FAILOUT)
2981                                                 RETURN(rc);
2982
2983                                         goto next;
2984                                 }
2985
2986                                 if (!dt_object_exists(parent)) {
2987                                         lfsck_object_put(env, parent);
2988                                         goto next;
2989                                 }
2990
2991                                 rc = lfsck_namespace_repair_dirent(env, com,
2992                                         parent, child, cname->ln_name,
2993                                         cname->ln_name, S_IFDIR, false, true);
2994                                 lfsck_object_put(env, parent);
2995                                 if (rc < 0) {
2996                                         if (bk->lb_param & LPF_FAILOUT)
2997                                                 RETURN(rc);
2998
2999                                         goto next;
3000                                 }
3001
3002                                 dirent_count += rc;
3003
3004 next:
3005                                 lfsck_linkea_del_buf(ldata, cname);
3006                         }
3007
3008                         ns->ln_dirent_repaired += dirent_count;
3009
3010                         RETURN(rc);
3011                 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
3012
3013                 lfsck_ibits_unlock(lh, LCK_EX);
3014                 /* The name entry references another MDT-object that may be
3015                  * created by the LFSCK for repairing dangling name entry.
3016                  * Try to replace it. */
3017                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
3018                                                   &tfid, cname);
3019                 lfsck_object_put(env, parent);
3020                 if (rc < 0)
3021                         RETURN(rc);
3022
3023                 if (rc > 0)
3024                         goto rebuild;
3025
3026                 lfsck_linkea_del_buf(ldata, cname);
3027         } /* while (ldata->ld_lee != NULL) */
3028
3029         /* If there is still linkEA overflow, return. */
3030         if (unlikely(ldata->ld_leh->leh_overflow_time))
3031                 RETURN(0);
3032
3033         linkea_first_entry(ldata);
3034         if (ldata->ld_leh->leh_reccount == 1) {
3035                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
3036                                                 lh, type, NULL, unknown);
3037
3038                 RETURN(rc);
3039         }
3040
3041         /* All linkEA entries are invalid and removed, then handle the @child
3042          * as an orphan.*/