Whamcloud - gitweb
LU-6245 libcfs: remove wrappers for timer functions
[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, 2016, 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 <lustre/lustre_idl.h>
34 #include <lu_object.h>
35 #include <dt_object.h>
36 #include <md_object.h>
37 #include <lustre_fid.h>
38 #include <lustre_lib.h>
39 #include <lustre_net.h>
40 #include <lustre/lustre_user.h>
41
42 #include "lfsck_internal.h"
43
44 #define LFSCK_NAMESPACE_MAGIC_V1        0xA0629D03
45 #define LFSCK_NAMESPACE_MAGIC_V2        0xA0621A0B
46
47 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
48  * as the trace file magic. When downgrade to such old release, the old LFSCK
49  * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
50  * then it will reset the whole LFSCK, and will not cause start failure. The
51  * similar case will happen when upgrade from such old release. */
52 #define LFSCK_NAMESPACE_MAGIC           LFSCK_NAMESPACE_MAGIC_V2
53
54 enum lfsck_nameentry_check {
55         LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
56         LFSCK_NAMEENTRY_REMOVED         = 2, /* The entry has been removed. */
57         LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
58 };
59
60 static struct lfsck_namespace_req *
61 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
62                                    struct lfsck_assistant_object *lso,
63                                    struct lu_dirent *ent, __u16 type)
64 {
65         struct lfsck_namespace_req *lnr;
66         int                         size;
67
68         size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
69         OBD_ALLOC(lnr, size);
70         if (lnr == NULL)
71                 return ERR_PTR(-ENOMEM);
72
73         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
74         lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
75         lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
76         lnr->lnr_fid = ent->lde_fid;
77         lnr->lnr_dir_cookie = ent->lde_hash;
78         lnr->lnr_attr = ent->lde_attrs;
79         lnr->lnr_size = size;
80         lnr->lnr_type = type;
81         lnr->lnr_namelen = ent->lde_namelen;
82         memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
83
84         return lnr;
85 }
86
87 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
88                                                struct lfsck_assistant_req *lar)
89 {
90         struct lfsck_namespace_req *lnr =
91                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
92
93         if (lnr->lnr_lmv != NULL)
94                 lfsck_lmv_put(env, lnr->lnr_lmv);
95
96         lfsck_assistant_object_put(env, lar->lar_parent);
97         OBD_FREE(lnr, lnr->lnr_size);
98 }
99
100 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
101                                       struct lfsck_namespace *src)
102 {
103         dst->ln_magic = le32_to_cpu(src->ln_magic);
104         dst->ln_status = le32_to_cpu(src->ln_status);
105         dst->ln_flags = le32_to_cpu(src->ln_flags);
106         dst->ln_success_count = le32_to_cpu(src->ln_success_count);
107         dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
108         dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
109         dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
110         dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
111         dst->ln_time_last_checkpoint =
112                                 le64_to_cpu(src->ln_time_last_checkpoint);
113         lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
114                                  &src->ln_pos_latest_start);
115         lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
116                                  &src->ln_pos_last_checkpoint);
117         lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
118                                  &src->ln_pos_first_inconsistent);
119         dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
120         dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
121         dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
122         dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
123         dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
124         dst->ln_objs_repaired_phase2 =
125                                 le64_to_cpu(src->ln_objs_repaired_phase2);
126         dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
127         dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
128         fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
129                       &src->ln_fid_latest_scanned_phase2);
130         dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
131         dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
132         dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
133         dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
134         dst->ln_unknown_inconsistency =
135                                 le64_to_cpu(src->ln_unknown_inconsistency);
136         dst->ln_unmatched_pairs_repaired =
137                                 le64_to_cpu(src->ln_unmatched_pairs_repaired);
138         dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
139         dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
140         dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
141         dst->ln_lost_dirent_repaired =
142                                 le64_to_cpu(src->ln_lost_dirent_repaired);
143         dst->ln_striped_dirs_scanned =
144                                 le64_to_cpu(src->ln_striped_dirs_scanned);
145         dst->ln_striped_dirs_repaired =
146                                 le64_to_cpu(src->ln_striped_dirs_repaired);
147         dst->ln_striped_dirs_failed =
148                                 le64_to_cpu(src->ln_striped_dirs_failed);
149         dst->ln_striped_dirs_disabled =
150                                 le64_to_cpu(src->ln_striped_dirs_disabled);
151         dst->ln_striped_dirs_skipped =
152                                 le64_to_cpu(src->ln_striped_dirs_skipped);
153         dst->ln_striped_shards_scanned =
154                                 le64_to_cpu(src->ln_striped_shards_scanned);
155         dst->ln_striped_shards_repaired =
156                                 le64_to_cpu(src->ln_striped_shards_repaired);
157         dst->ln_striped_shards_failed =
158                                 le64_to_cpu(src->ln_striped_shards_failed);
159         dst->ln_striped_shards_skipped =
160                                 le64_to_cpu(src->ln_striped_shards_skipped);
161         dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
162         dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
163         dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
164         dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
165         dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
166         dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
167         dst->ln_time_latest_reset = le32_to_cpu(src->ln_time_latest_reset);
168         dst->ln_linkea_overflow_cleared =
169                                 le64_to_cpu(src->ln_linkea_overflow_cleared);
170 }
171
172 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
173                                       struct lfsck_namespace *src)
174 {
175         dst->ln_magic = cpu_to_le32(src->ln_magic);
176         dst->ln_status = cpu_to_le32(src->ln_status);
177         dst->ln_flags = cpu_to_le32(src->ln_flags);
178         dst->ln_success_count = cpu_to_le32(src->ln_success_count);
179         dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
180         dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
181         dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
182         dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
183         dst->ln_time_last_checkpoint =
184                                 cpu_to_le64(src->ln_time_last_checkpoint);
185         lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
186                                  &src->ln_pos_latest_start);
187         lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
188                                  &src->ln_pos_last_checkpoint);
189         lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
190                                  &src->ln_pos_first_inconsistent);
191         dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
192         dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
193         dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
194         dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
195         dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
196         dst->ln_objs_repaired_phase2 =
197                                 cpu_to_le64(src->ln_objs_repaired_phase2);
198         dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
199         dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
200         fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
201                       &src->ln_fid_latest_scanned_phase2);
202         dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
203         dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
204         dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
205         dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
206         dst->ln_unknown_inconsistency =
207                                 cpu_to_le64(src->ln_unknown_inconsistency);
208         dst->ln_unmatched_pairs_repaired =
209                                 cpu_to_le64(src->ln_unmatched_pairs_repaired);
210         dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
211         dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
212         dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
213         dst->ln_lost_dirent_repaired =
214                                 cpu_to_le64(src->ln_lost_dirent_repaired);
215         dst->ln_striped_dirs_scanned =
216                                 cpu_to_le64(src->ln_striped_dirs_scanned);
217         dst->ln_striped_dirs_repaired =
218                                 cpu_to_le64(src->ln_striped_dirs_repaired);
219         dst->ln_striped_dirs_failed =
220                                 cpu_to_le64(src->ln_striped_dirs_failed);
221         dst->ln_striped_dirs_disabled =
222                                 cpu_to_le64(src->ln_striped_dirs_disabled);
223         dst->ln_striped_dirs_skipped =
224                                 cpu_to_le64(src->ln_striped_dirs_skipped);
225         dst->ln_striped_shards_scanned =
226                                 cpu_to_le64(src->ln_striped_shards_scanned);
227         dst->ln_striped_shards_repaired =
228                                 cpu_to_le64(src->ln_striped_shards_repaired);
229         dst->ln_striped_shards_failed =
230                                 cpu_to_le64(src->ln_striped_shards_failed);
231         dst->ln_striped_shards_skipped =
232                                 cpu_to_le64(src->ln_striped_shards_skipped);
233         dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
234         dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
235         dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
236         dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
237         dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
238         dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
239         dst->ln_time_latest_reset = cpu_to_le32(src->ln_time_latest_reset);
240         dst->ln_linkea_overflow_cleared =
241                                 cpu_to_le64(src->ln_linkea_overflow_cleared);
242 }
243
244 static void lfsck_namespace_record_failure(const struct lu_env *env,
245                                            struct lfsck_instance *lfsck,
246                                            struct lfsck_namespace *ns)
247 {
248         struct lfsck_position pos;
249
250         ns->ln_items_failed++;
251         lfsck_pos_fill(env, lfsck, &pos, false);
252         if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
253             lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
254                 ns->ln_pos_first_inconsistent = pos;
255
256                 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
257                        "inconsistency at the pos [%llu, "DFID", %#llx]\n",
258                        lfsck_lfsck2name(lfsck),
259                        ns->ln_pos_first_inconsistent.lp_oit_cookie,
260                        PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
261                        ns->ln_pos_first_inconsistent.lp_dir_cookie);
262         }
263 }
264
265 /**
266  * Load the MDT bitmap from the lfsck_namespace trace file.
267  *
268  * \param[in] env       pointer to the thread context
269  * \param[in] com       pointer to the lfsck component
270  *
271  * \retval              0 for success
272  * \retval              negative error number on failure or data corruption
273  */
274 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
275                                        struct lfsck_component *com)
276 {
277         struct dt_object                *obj    = com->lc_obj;
278         struct lfsck_assistant_data     *lad    = com->lc_data;
279         struct lfsck_namespace          *ns     = com->lc_file_ram;
280         struct cfs_bitmap                       *bitmap = lad->lad_bitmap;
281         ssize_t                          size;
282         __u32                            nbits;
283         int                              rc;
284         ENTRY;
285
286         if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
287             ns->ln_bitmap_size)
288                 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
289         else
290                 nbits = ns->ln_bitmap_size;
291
292         if (unlikely(nbits < BITS_PER_LONG))
293                 nbits = BITS_PER_LONG;
294
295         if (nbits > bitmap->size) {
296                 __u32 new_bits = bitmap->size;
297                 struct cfs_bitmap *new_bitmap;
298
299                 while (new_bits < nbits)
300                         new_bits <<= 1;
301
302                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
303                 if (new_bitmap == NULL)
304                         RETURN(-ENOMEM);
305
306                 lad->lad_bitmap = new_bitmap;
307                 CFS_FREE_BITMAP(bitmap);
308                 bitmap = new_bitmap;
309         }
310
311         if (ns->ln_bitmap_size == 0) {
312                 lad->lad_incomplete = 0;
313                 CFS_RESET_BITMAP(bitmap);
314
315                 RETURN(0);
316         }
317
318         size = (ns->ln_bitmap_size + 7) >> 3;
319         rc = dt_xattr_get(env, obj,
320                           lfsck_buf_get(env, bitmap->data, size),
321                           XATTR_NAME_LFSCK_BITMAP);
322         if (rc != size)
323                 RETURN(rc >= 0 ? -EINVAL : rc);
324
325         if (cfs_bitmap_check_empty(bitmap))
326                 lad->lad_incomplete = 0;
327         else
328                 lad->lad_incomplete = 1;
329
330         RETURN(0);
331 }
332
333 /**
334  * Load namespace LFSCK statistics information from the trace file.
335  *
336  * \param[in] env       pointer to the thread context
337  * \param[in] com       pointer to the lfsck component
338  *
339  * \retval              0 for success
340  * \retval              negative error number on failure
341  */
342 static int lfsck_namespace_load(const struct lu_env *env,
343                                 struct lfsck_component *com)
344 {
345         int len = com->lc_file_size;
346         int rc;
347
348         rc = dt_xattr_get(env, com->lc_obj,
349                           lfsck_buf_get(env, com->lc_file_disk, len),
350                           XATTR_NAME_LFSCK_NAMESPACE);
351         if (rc == len) {
352                 struct lfsck_namespace *ns = com->lc_file_ram;
353
354                 lfsck_namespace_le_to_cpu(ns,
355                                 (struct lfsck_namespace *)com->lc_file_disk);
356                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
357                         CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
358                                "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
359                                ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
360                         rc = -ESTALE;
361                 } else {
362                         rc = 0;
363                 }
364         } else if (rc != -ENODATA) {
365                 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
366                        "expected = %d: rc = %d\n",
367                        lfsck_lfsck2name(com->lc_lfsck), len, rc);
368                 if (rc >= 0)
369                         rc = -ESTALE;
370         }
371
372         return rc;
373 }
374
375 static int lfsck_namespace_store(const struct lu_env *env,
376                                  struct lfsck_component *com)
377 {
378         struct dt_object                *obj    = com->lc_obj;
379         struct lfsck_instance           *lfsck  = com->lc_lfsck;
380         struct lfsck_namespace          *ns     = com->lc_file_ram;
381         struct lfsck_assistant_data     *lad    = com->lc_data;
382         struct dt_device                *dev    = lfsck_obj2dev(obj);
383         struct cfs_bitmap               *bitmap = NULL;
384         struct thandle                  *handle;
385         __u32                            nbits  = 0;
386         int                              len    = com->lc_file_size;
387         int                              rc;
388         ENTRY;
389
390         if (lad != NULL) {
391                 bitmap = lad->lad_bitmap;
392                 nbits = bitmap->size;
393
394                 LASSERT(nbits > 0);
395                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
396         }
397
398         ns->ln_bitmap_size = nbits;
399         lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
400                                   ns);
401         handle = dt_trans_create(env, dev);
402         if (IS_ERR(handle))
403                 GOTO(log, rc = PTR_ERR(handle));
404
405         rc = dt_declare_xattr_set(env, obj,
406                                   lfsck_buf_get(env, com->lc_file_disk, len),
407                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
408         if (rc != 0)
409                 GOTO(out, rc);
410
411         if (bitmap != NULL) {
412                 rc = dt_declare_xattr_set(env, obj,
413                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
414                                 XATTR_NAME_LFSCK_BITMAP, 0, handle);
415                 if (rc != 0)
416                         GOTO(out, rc);
417         }
418
419         rc = dt_trans_start_local(env, dev, handle);
420         if (rc != 0)
421                 GOTO(out, rc);
422
423         rc = dt_xattr_set(env, obj,
424                           lfsck_buf_get(env, com->lc_file_disk, len),
425                           XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
426         if (rc == 0 && bitmap != NULL)
427                 rc = dt_xattr_set(env, obj,
428                                   lfsck_buf_get(env, bitmap->data, nbits >> 3),
429                                   XATTR_NAME_LFSCK_BITMAP, 0, handle);
430
431         GOTO(out, rc);
432
433 out:
434         dt_trans_stop(env, dev, handle);
435
436 log:
437         if (rc != 0)
438                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
439                        lfsck_lfsck2name(lfsck), rc);
440         return rc;
441 }
442
443 static int lfsck_namespace_init(const struct lu_env *env,
444                                 struct lfsck_component *com)
445 {
446         struct lfsck_namespace *ns = com->lc_file_ram;
447         int rc;
448
449         memset(ns, 0, sizeof(*ns));
450         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
451         ns->ln_status = LS_INIT;
452         ns->ln_time_latest_reset = cfs_time_current_sec();
453         down_write(&com->lc_sem);
454         rc = lfsck_namespace_store(env, com);
455         if (rc == 0)
456                 rc = lfsck_load_sub_trace_files(env, com,
457                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
458         up_write(&com->lc_sem);
459
460         return rc;
461 }
462
463 /**
464  * Update the namespace LFSCK trace file for the given @fid
465  *
466  * \param[in] env       pointer to the thread context
467  * \param[in] com       pointer to the lfsck component
468  * \param[in] fid       the fid which flags to be updated in the lfsck
469  *                      trace file
470  * \param[in] add       true if add new flags, otherwise remove flags
471  *
472  * \retval              0 for success or nothing to be done
473  * \retval              negative error number on failure
474  */
475 int lfsck_namespace_trace_update(const struct lu_env *env,
476                                  struct lfsck_component *com,
477                                  const struct lu_fid *fid,
478                                  const __u8 flags, bool add)
479 {
480         struct lfsck_instance   *lfsck  = com->lc_lfsck;
481         struct dt_object        *obj;
482         struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
483         struct dt_device        *dev;
484         struct thandle          *th     = NULL;
485         int                      idx;
486         int                      rc     = 0;
487         __u8                     old    = 0;
488         __u8                     new    = 0;
489         ENTRY;
490
491         LASSERT(flags != 0);
492
493         if (unlikely(!fid_is_sane(fid)))
494                 RETURN(0);
495
496         idx = lfsck_sub_trace_file_fid2idx(fid);
497         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
498         obj = com->lc_sub_trace_objs[idx].lsto_obj;
499         if (unlikely(obj == NULL)) {
500                 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
501                 RETURN(0);
502         }
503
504         lfsck_object_get(obj);
505         dev = lfsck_obj2dev(obj);
506         fid_cpu_to_be(key, fid);
507         rc = dt_lookup(env, obj, (struct dt_rec *)&old,
508                        (const struct dt_key *)key);
509         if (rc == -ENOENT) {
510                 if (!add)
511                         GOTO(unlock, rc = 0);
512
513                 old = 0;
514                 new = flags;
515         } else if (rc == 0) {
516                 if (add) {
517                         if ((old & flags) == flags)
518                                 GOTO(unlock, rc = 0);
519
520                         new = old | flags;
521                 } else {
522                         if ((old & flags) == 0)
523                                 GOTO(unlock, rc = 0);
524
525                         new = old & ~flags;
526                 }
527         } else {
528                 GOTO(log, rc);
529         }
530
531         th = dt_trans_create(env, dev);
532         if (IS_ERR(th))
533                 GOTO(log, rc = PTR_ERR(th));
534
535         if (old != 0) {
536                 rc = dt_declare_delete(env, obj,
537                                        (const struct dt_key *)key, th);
538                 if (rc != 0)
539                         GOTO(log, rc);
540         }
541
542         if (new != 0) {
543                 rc = dt_declare_insert(env, obj,
544                                        (const struct dt_rec *)&new,
545                                        (const struct dt_key *)key, th);
546                 if (rc != 0)
547                         GOTO(log, rc);
548         }
549
550         rc = dt_trans_start_local(env, dev, th);
551         if (rc != 0)
552                 GOTO(log, rc);
553
554         if (old != 0) {
555                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
556                 if (rc != 0)
557                         GOTO(log, rc);
558         }
559
560         if (new != 0) {
561                 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
562                                (const struct dt_key *)key, th, 1);
563                 if (rc != 0)
564                         GOTO(log, rc);
565         }
566
567         GOTO(log, rc);
568
569 log:
570         if (th != NULL && !IS_ERR(th))
571                 dt_trans_stop(env, dev, th);
572
573         CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
574                "trace file, flags %x, old %x, new %x: rc = %d\n",
575                lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
576                (__u32)flags, (__u32)old, (__u32)new, rc);
577
578 unlock:
579         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
580         lfsck_object_put(env, obj);
581
582         return rc;
583 }
584
585 int lfsck_namespace_check_exist(const struct lu_env *env,
586                                 struct dt_object *dir,
587                                 struct dt_object *obj, const char *name)
588 {
589         struct lu_fid    *fid = &lfsck_env_info(env)->lti_fid;
590         int               rc;
591         ENTRY;
592
593         if (unlikely(lfsck_is_dead_obj(obj)))
594                 RETURN(LFSCK_NAMEENTRY_DEAD);
595
596         rc = dt_lookup(env, dir, (struct dt_rec *)fid,
597                        (const struct dt_key *)name);
598         if (rc == -ENOENT)
599                 RETURN(LFSCK_NAMEENTRY_REMOVED);
600
601         if (rc < 0)
602                 RETURN(rc);
603
604         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
605                 RETURN(LFSCK_NAMEENTRY_RECREATED);
606
607         RETURN(0);
608 }
609
610 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
611                                             struct dt_object *obj,
612                                             struct thandle *handle)
613 {
614         int rc;
615
616         /* For remote updating LINKEA, there may be further LFSCK action
617          * on remote MDT after the updating, so update the LINKEA ASAP. */
618         if (dt_object_remote(obj))
619                 handle->th_sync = 1;
620
621         /* For destroying all invalid linkEA entries. */
622         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
623         if (rc == 0)
624                 /* For insert new linkEA entry. */
625                 rc = dt_declare_xattr_set(env, obj,
626                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
627                         XATTR_NAME_LINK, 0, handle);
628         return rc;
629 }
630
631 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
632                        struct linkea_data *ldata, bool with_rec)
633 {
634         int rc;
635
636         if (ldata->ld_buf->lb_buf == NULL)
637                 return -ENOMEM;
638
639         if (!dt_object_exists(obj))
640                 return -ENOENT;
641
642         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
643         if (rc == -ERANGE) {
644                 /* Buf was too small, figure out what we need. */
645                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
646                 if (unlikely(rc == 0))
647                         return -ENODATA;
648
649                 if (rc < 0)
650                         return rc;
651
652                 lu_buf_realloc(ldata->ld_buf, rc);
653                 if (ldata->ld_buf->lb_buf == NULL)
654                         return -ENOMEM;
655
656                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
657         }
658
659         if (unlikely(rc == 0))
660                 return -ENODATA;
661
662         if (rc > 0) {
663                 if (with_rec)
664                         rc = linkea_init_with_rec(ldata);
665                 else
666                         rc = linkea_init(ldata);
667         }
668
669         return rc;
670 }
671
672 /**
673  * Remove linkEA for the given object.
674  *
675  * The caller should take the ldlm lock before the calling.
676  *
677  * \param[in] env       pointer to the thread context
678  * \param[in] com       pointer to the lfsck component
679  * \param[in] obj       pointer to the dt_object to be handled
680  *
681  * \retval              0 for repaired cases
682  * \retval              negative error number on failure
683  */
684 static int lfsck_namespace_links_remove(const struct lu_env *env,
685                                         struct lfsck_component *com,
686                                         struct dt_object *obj)
687 {
688         struct lfsck_instance           *lfsck  = com->lc_lfsck;
689         struct dt_device                *dev    = lfsck_obj2dev(obj);
690         struct thandle                  *th     = NULL;
691         int                              rc     = 0;
692         ENTRY;
693
694         LASSERT(dt_object_remote(obj) == 0);
695
696         th = dt_trans_create(env, dev);
697         if (IS_ERR(th))
698                 GOTO(log, rc = PTR_ERR(th));
699
700         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
701         if (rc != 0)
702                 GOTO(stop, rc);
703
704         rc = dt_trans_start_local(env, dev, th);
705         if (rc != 0)
706                 GOTO(stop, rc);
707
708         dt_write_lock(env, obj, 0);
709         if (unlikely(lfsck_is_dead_obj(obj)))
710                 GOTO(unlock, rc = -ENOENT);
711
712         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
713                 GOTO(unlock, rc = 0);
714
715         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
716
717         GOTO(unlock, rc);
718
719 unlock:
720         dt_write_unlock(env, obj);
721
722 stop:
723         dt_trans_stop(env, dev, th);
724
725 log:
726         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
727                "for the object "DFID": rc = %d\n",
728                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
729
730         if (rc == 0) {
731                 struct lfsck_namespace *ns = com->lc_file_ram;
732
733                 ns->ln_flags |= LF_INCONSISTENT;
734         }
735
736         return rc;
737 }
738
739 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
740                              struct linkea_data *ldata, struct thandle *handle)
741 {
742         struct lu_buf buf;
743         int rc;
744
745         lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
746
747 again:
748         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
749         if (unlikely(rc == -ENOSPC)) {
750                 rc = linkea_overflow_shrink(ldata);
751                 if (likely(rc > 0)) {
752                         buf.lb_len = rc;
753                         goto again;
754                 }
755         }
756
757         return rc;
758 }
759
760 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
761                                                struct lu_name *cname,
762                                                struct lu_fid *pfid,
763                                                char *buf, const int buflen)
764 {
765         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
766         if (unlikely(ldata->ld_reclen <= 0 ||
767                      ldata->ld_reclen + sizeof(struct link_ea_header) >
768                         ldata->ld_leh->leh_len ||
769                      cname->ln_namelen <= 0 ||
770                      cname->ln_namelen > NAME_MAX ||
771                      cname->ln_namelen >= buflen ||
772                      !fid_is_sane(pfid)))
773                 return -EINVAL;
774
775         /* To guarantee the 'name' is terminated with '0'. */
776         memcpy(buf, cname->ln_name, cname->ln_namelen);
777         buf[cname->ln_namelen] = 0;
778         cname->ln_name = buf;
779
780         return 0;
781 }
782
783 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
784                                  const struct lu_name *lname)
785 {
786         LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
787
788         /* If current record is corrupted, all the subsequent
789          * records will be dropped. */
790         if (unlikely(ldata->ld_reclen <= 0 ||
791                      ldata->ld_reclen + sizeof(struct link_ea_header) >
792                         ldata->ld_leh->leh_len)) {
793                 void *ptr = ldata->ld_lee;
794
795                 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
796                 ldata->ld_leh->leh_reccount = 0;
797                 linkea_first_entry(ldata);
798                 while (ldata->ld_lee != NULL &&
799                        (char *)ldata->ld_lee < (char *)ptr) {
800                         int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
801                                      ldata->ld_lee->lee_reclen[1];
802
803                         ldata->ld_leh->leh_len += reclen;
804                         ldata->ld_leh->leh_reccount++;
805                         ldata->ld_lee = (struct link_ea_entry *)
806                                         ((char *)ldata->ld_lee + reclen);
807                 }
808
809                 ldata->ld_lee = NULL;
810         } else {
811                 linkea_del_buf(ldata, lname);
812         }
813 }
814
815 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
816                                                struct lu_name *cname,
817                                                struct lu_fid *pfid,
818                                                bool remove)
819 {
820         struct link_ea_entry    *oldlee;
821         int                      oldlen;
822         int                      repeated = 0;
823
824         oldlee = ldata->ld_lee;
825         oldlen = ldata->ld_reclen;
826         linkea_next_entry(ldata);
827         while (ldata->ld_lee != NULL) {
828                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
829                                    ldata->ld_lee->lee_reclen[1];
830                 if (unlikely(ldata->ld_reclen == oldlen &&
831                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
832                         repeated++;
833                         if (!remove)
834                                 break;
835
836                         lfsck_linkea_del_buf(ldata, cname);
837                 } else {
838                         linkea_next_entry(ldata);
839                 }
840         }
841         ldata->ld_lee = oldlee;
842         ldata->ld_reclen = oldlen;
843
844         return repeated;
845 }
846
847 /**
848  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
849  *
850  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
851  * with the given type to generate the name, the detailed rules for name
852  * have been described as following.
853  *
854  * The function also generates the linkEA corresponding to the name entry
855  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
856  *
857  * \param[in] env       pointer to the thread context
858  * \param[in] com       pointer to the lfsck component
859  * \param[in] orphan    pointer to the orphan MDT-object
860  * \param[in] infix     additional information for the orphan name, such as
861  *                      the FID for original
862  * \param[in] type      the type for describing why the orphan MDT-object is
863  *                      created. The rules are as following:
864  *
865  *  type "D":           The MDT-object is a directory, it may knows its parent
866  *                      but because there is no valid linkEA, the LFSCK cannot
867  *                      know where to put it back to the namespace.
868  *  type "O":           The MDT-object has no linkEA, and there is no name
869  *                      entry that references the MDT-object.
870  *
871  *  type "S":           The orphan MDT-object is a shard of a striped directory
872  *
873  * \see lfsck_layout_recreate_parent() for more types.
874  *
875  * The orphan name will be like:
876  * ${FID}-${infix}-${type}-${conflict_version}
877  *
878  * \param[out] count    if some others inserted some linkEA entries by race,
879  *                      then return the linkEA entries count.
880  *
881  * \retval              positive number for repaired cases
882  * \retval              0 if needs to repair nothing
883  * \retval              negative error number on failure
884  */
885 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
886                                          struct lfsck_component *com,
887                                          struct dt_object *orphan,
888                                          const char *infix, const char *type,
889                                          int *count)
890 {
891         struct lfsck_thread_info        *info   = lfsck_env_info(env);
892         struct lu_name                  *cname  = &info->lti_name;
893         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
894         struct lu_attr                  *la     = &info->lti_la2;
895         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
896         const struct lu_fid             *pfid;
897         struct lu_fid                    tfid;
898         struct lfsck_instance           *lfsck  = com->lc_lfsck;
899         struct dt_device                *dev    = lfsck_obj2dev(orphan);
900         struct dt_object                *parent;
901         struct thandle                  *th     = NULL;
902         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
903         struct lustre_handle             clh    = { 0 };
904         struct linkea_data               ldata2 = { NULL };
905         struct lu_buf                    linkea_buf;
906         int                              namelen;
907         int                              idx    = 0;
908         int                              rc     = 0;
909         bool                             exist  = false;
910         ENTRY;
911
912         cname->ln_name = NULL;
913         if (unlikely(lfsck->li_lpf_obj == NULL))
914                 GOTO(log, rc = -ENXIO);
915
916         parent = lfsck->li_lpf_obj;
917         pfid = lfsck_dto2fid(parent);
918
919 again:
920         do {
921                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
922                                    PFID(cfid), infix, type, idx++);
923                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
924                                (const struct dt_key *)info->lti_key);
925                 if (rc != 0 && rc != -ENOENT)
926                         GOTO(log, rc);
927
928                 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
929                         exist = true;
930         } while (rc == 0 && !exist);
931
932         rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
933                         MDS_INODELOCK_UPDATE, LCK_PW);
934         if (rc != 0)
935                 GOTO(log, rc);
936
937         /* Re-check whether the name conflict with othrs after taken
938          * the ldlm lock. */
939         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
940                        (const struct dt_key *)info->lti_key);
941         if (rc == 0) {
942                 if (!lu_fid_eq(cfid, &tfid)) {
943                         exist = false;
944                         lfsck_unlock(pllh);
945                         goto again;
946                 }
947
948                 exist = true;
949         } else if (rc != -ENOENT) {
950                 GOTO(log, rc);
951         } else {
952                 exist = false;
953         }
954
955         cname->ln_name = info->lti_key;
956         cname->ln_namelen = namelen;
957         rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
958                               cname, pfid);
959         if (rc != 0)
960                 GOTO(log, rc);
961
962         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
963                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
964                               MDS_INODELOCK_XATTR, LCK_EX);
965         if (rc != 0)
966                 GOTO(log, rc);
967
968         lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
969                        ldata2.ld_leh->leh_len);
970         th = dt_trans_create(env, dev);
971         if (IS_ERR(th))
972                 GOTO(log, rc = PTR_ERR(th));
973
974         if (S_ISDIR(lfsck_object_type(orphan))) {
975                 rc = dt_declare_delete(env, orphan,
976                                        (const struct dt_key *)dotdot, th);
977                 if (rc != 0)
978                         GOTO(stop, rc);
979
980                 rec->rec_type = S_IFDIR;
981                 rec->rec_fid = pfid;
982                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
983                                        (const struct dt_key *)dotdot, th);
984                 if (rc != 0)
985                         GOTO(stop, rc);
986         }
987
988         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
989                                   XATTR_NAME_LINK, 0, th);
990         if (rc != 0)
991                 GOTO(stop, rc);
992
993         if (!exist) {
994                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
995                 rec->rec_fid = cfid;
996                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
997                                        (const struct dt_key *)cname->ln_name,
998                                        th);
999                 if (rc != 0)
1000                         GOTO(stop, rc);
1001
1002                 if (S_ISDIR(rec->rec_type)) {
1003                         rc = dt_declare_ref_add(env, parent, th);
1004                         if (rc != 0)
1005                                 GOTO(stop, rc);
1006                 }
1007         }
1008
1009         memset(la, 0, sizeof(*la));
1010         la->la_ctime = cfs_time_current_sec();
1011         la->la_valid = LA_CTIME;
1012         rc = dt_declare_attr_set(env, orphan, la, th);
1013         if (rc != 0)
1014                 GOTO(stop, rc);
1015
1016         rc = dt_trans_start_local(env, dev, th);
1017         if (rc != 0)
1018                 GOTO(stop, rc);
1019
1020         dt_write_lock(env, orphan, 0);
1021         rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1022         if (likely(rc == -ENODATA || rc == -EINVAL)) {
1023                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1024                         GOTO(unlock, rc = 1);
1025
1026                 if (S_ISDIR(lfsck_object_type(orphan))) {
1027                         rc = dt_delete(env, orphan,
1028                                        (const struct dt_key *)dotdot, th);
1029                         if (rc != 0)
1030                                 GOTO(unlock, rc);
1031
1032                         rec->rec_type = S_IFDIR;
1033                         rec->rec_fid = pfid;
1034                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1035                                        (const struct dt_key *)dotdot, th, 1);
1036                         if (rc != 0)
1037                                 GOTO(unlock, rc);
1038                 }
1039
1040                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1041                                   th);
1042         } else {
1043                 if (rc == 0 && count != NULL)
1044                         *count = ldata2.ld_leh->leh_reccount;
1045
1046                 GOTO(unlock, rc);
1047         }
1048         dt_write_unlock(env, orphan);
1049
1050         if (rc == 0 && !exist) {
1051                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1052                 rec->rec_fid = cfid;
1053                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1054                                (const struct dt_key *)cname->ln_name, th, 1);
1055                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1056                         dt_write_lock(env, parent, 0);
1057                         rc = dt_ref_add(env, parent, th);
1058                         dt_write_unlock(env, parent);
1059                 }
1060         }
1061
1062         if (rc == 0)
1063                 rc = dt_attr_set(env, orphan, la, th);
1064
1065         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1066
1067 unlock:
1068         dt_write_unlock(env, orphan);
1069
1070 stop:
1071         dt_trans_stop(env, dev, th);
1072
1073 log:
1074         lfsck_ibits_unlock(&clh, LCK_EX);
1075         lfsck_unlock(pllh);
1076         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1077                "object "DFID", name = %s: rc = %d\n",
1078                lfsck_lfsck2name(lfsck), PFID(cfid),
1079                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1080
1081         if (rc != 0) {
1082                 struct lfsck_namespace *ns = com->lc_file_ram;
1083
1084                 ns->ln_flags |= LF_INCONSISTENT;
1085         }
1086
1087         return rc;
1088 }
1089
1090 /**
1091  * Add the specified name entry back to namespace.
1092  *
1093  * If there is a linkEA entry that back references a name entry under
1094  * some parent directory, but such parent directory does not have the
1095  * claimed name entry. On the other hand, the linkEA entries count is
1096  * not larger than the MDT-object's hard link count. Under such case,
1097  * it is quite possible that the name entry is lost. Then the LFSCK
1098  * should add the name entry back to the namespace.
1099  *
1100  * \param[in] env       pointer to the thread context
1101  * \param[in] com       pointer to the lfsck component
1102  * \param[in] parent    pointer to the directory under which the name entry
1103  *                      will be inserted into
1104  * \param[in] child     pointer to the object referenced by the name entry
1105  *                      that to be inserted into the parent
1106  * \param[in] name      the name for the child in the parent directory
1107  *
1108  * \retval              positive number for repaired cases
1109  * \retval              0 if nothing to be repaired
1110  * \retval              negative error number on failure
1111  */
1112 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1113                                          struct lfsck_component *com,
1114                                          struct dt_object *parent,
1115                                          struct dt_object *child,
1116                                          const char *name)
1117 {
1118         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1119         struct lu_attr                  *la     = &info->lti_la;
1120         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1121         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1122         /* The child and its name may be on different MDTs. */
1123         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1124         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
1125         struct dt_device                *dev    = lfsck->li_next;
1126         struct thandle                  *th     = NULL;
1127         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1128         int                              rc     = 0;
1129         ENTRY;
1130
1131         /* @parent/@child may be based on lfsck->li_bottom,
1132          * but here we need the object based on the lfsck->li_next. */
1133
1134         parent = lfsck_object_locate(dev, parent);
1135         if (IS_ERR(parent))
1136                 GOTO(log, rc = PTR_ERR(parent));
1137
1138         if (unlikely(!dt_try_as_dir(env, parent)))
1139                 GOTO(log, rc = -ENOTDIR);
1140
1141         child = lfsck_object_locate(dev, child);
1142         if (IS_ERR(child))
1143                 GOTO(log, rc = PTR_ERR(child));
1144
1145         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1146                 GOTO(log, rc = 1);
1147
1148         rc = lfsck_lock(env, lfsck, parent, name, llh,
1149                         MDS_INODELOCK_UPDATE, LCK_PW);
1150         if (rc != 0)
1151                 GOTO(log, rc);
1152
1153         th = dt_trans_create(env, dev);
1154         if (IS_ERR(th))
1155                 GOTO(unlock, rc = PTR_ERR(th));
1156
1157         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1158         rec->rec_fid = cfid;
1159         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1160                                (const struct dt_key *)name, th);
1161         if (rc != 0)
1162                 GOTO(stop, rc);
1163
1164         if (S_ISDIR(rec->rec_type)) {
1165                 rc = dt_declare_ref_add(env, parent, th);
1166                 if (rc != 0)
1167                         GOTO(stop, rc);
1168         }
1169
1170         memset(la, 0, sizeof(*la));
1171         la->la_ctime = cfs_time_current_sec();
1172         la->la_valid = LA_CTIME;
1173         rc = dt_declare_attr_set(env, parent, la, th);
1174         if (rc != 0)
1175                 GOTO(stop, rc);
1176
1177         rc = dt_declare_attr_set(env, child, la, th);
1178         if (rc != 0)
1179                 GOTO(stop, rc);
1180
1181         rc = dt_trans_start_local(env, dev, th);
1182         if (rc != 0)
1183                 GOTO(stop, rc);
1184
1185         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1186                        (const struct dt_key *)name, th, 1);
1187         if (rc != 0)
1188                 GOTO(stop, rc);
1189
1190         if (S_ISDIR(rec->rec_type)) {
1191                 dt_write_lock(env, parent, 0);
1192                 rc = dt_ref_add(env, parent, th);
1193                 dt_write_unlock(env, parent);
1194                 if (rc != 0)
1195                         GOTO(stop, rc);
1196         }
1197
1198         la->la_ctime = cfs_time_current_sec();
1199         rc = dt_attr_set(env, parent, la, th);
1200         if (rc != 0)
1201                 GOTO(stop, rc);
1202
1203         rc = dt_attr_set(env, child, la, th);
1204
1205         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1206
1207 stop:
1208         dt_trans_stop(env, dev, th);
1209
1210 unlock:
1211         lfsck_unlock(llh);
1212
1213 log:
1214         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1215                "the name %s and type %o to the parent "DFID": rc = %d\n",
1216                lfsck_lfsck2name(lfsck), PFID(cfid), name,
1217                lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1218
1219         if (rc != 0) {
1220                 struct lfsck_namespace *ns = com->lc_file_ram;
1221
1222                 ns->ln_flags |= LF_INCONSISTENT;
1223                 if (rc > 0)
1224                         ns->ln_lost_dirent_repaired++;
1225         }
1226
1227         return rc;
1228 }
1229
1230 /**
1231  * Create the specified orphan directory.
1232  *
1233  * For the case that the parent MDT-object stored in some MDT-object's
1234  * linkEA entry is lost, the LFSCK will re-create the parent object as
1235  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1236  * with the name ${FID}-P-${conflict_version}.
1237  *
1238  * \param[in] env       pointer to the thread context
1239  * \param[in] com       pointer to the lfsck component
1240  * \param[in] orphan    pointer to the orphan MDT-object to be created
1241  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1242  *
1243  * \retval              positive number for repaired cases
1244  * \retval              negative error number on failure
1245  */
1246 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1247                                              struct lfsck_component *com,
1248                                              struct dt_object *orphan,
1249                                              struct lmv_mds_md_v1 *lmv)
1250 {
1251         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1252         struct lu_attr                  *la     = &info->lti_la;
1253         struct dt_allocation_hint       *hint   = &info->lti_hint;
1254         struct dt_object_format         *dof    = &info->lti_dof;
1255         struct lu_name                  *cname  = &info->lti_name2;
1256         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1257         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1258         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1259         struct lu_fid                    tfid;
1260         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1261         struct lfsck_namespace          *ns     = com->lc_file_ram;
1262         struct dt_device                *dev    = lfsck_obj2dev(orphan);
1263         struct dt_object                *parent = NULL;
1264         struct thandle                  *th     = NULL;
1265         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1266         struct linkea_data               ldata  = { NULL };
1267         struct lu_buf                    linkea_buf;
1268         struct lu_buf                    lmv_buf;
1269         char                             name[32];
1270         int                              namelen;
1271         int                              idx    = 0;
1272         int                              rc     = 0;
1273         int                              rc1    = 0;
1274         ENTRY;
1275
1276         LASSERT(!dt_object_exists(orphan));
1277
1278         cname->ln_name = NULL;
1279         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1280                 GOTO(log, rc = 1);
1281
1282         if (dt_object_remote(orphan)) {
1283                 LASSERT(lfsck->li_lpf_root_obj != NULL);
1284
1285                 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1286                 if (idx < 0)
1287                         GOTO(log, rc = idx);
1288
1289                 snprintf(name, 8, "MDT%04x", idx);
1290                 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1291                                (struct dt_rec *)&tfid,
1292                                (const struct dt_key *)name);
1293                 if (rc != 0)
1294                         GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1295
1296                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1297                 if (IS_ERR(parent))
1298                         GOTO(log, rc = PTR_ERR(parent));
1299
1300                 if (unlikely(!dt_try_as_dir(env, parent)))
1301                         GOTO(log, rc = -ENOTDIR);
1302         } else {
1303                 if (unlikely(lfsck->li_lpf_obj == NULL))
1304                         GOTO(log, rc = -ENXIO);
1305
1306                 parent = lfsck->li_lpf_obj;
1307         }
1308
1309         dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1310         if (IS_ERR(dev))
1311                 GOTO(log, rc = PTR_ERR(dev));
1312
1313         idx = 0;
1314
1315 again:
1316         do {
1317                 namelen = snprintf(name, 31, DFID"-P-%d",
1318                                    PFID(cfid), idx++);
1319                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1320                                (const struct dt_key *)name);
1321                 if (rc != 0 && rc != -ENOENT)
1322                         GOTO(log, rc);
1323         } while (rc == 0);
1324
1325         rc = lfsck_lock(env, lfsck, parent, name, llh,
1326                         MDS_INODELOCK_UPDATE, LCK_PW);
1327         if (rc != 0)
1328                 GOTO(log, rc);
1329
1330         /* Re-check whether the name conflict with othrs after taken
1331          * the ldlm lock. */
1332         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1333                        (const struct dt_key *)name);
1334         if (unlikely(rc == 0)) {
1335                 lfsck_unlock(llh);
1336                 goto again;
1337         }
1338
1339         if (rc != -ENOENT)
1340                 GOTO(unlock1, rc);
1341
1342         cname->ln_name = name;
1343         cname->ln_namelen = namelen;
1344
1345         memset(la, 0, sizeof(*la));
1346         la->la_mode = S_IFDIR | 0700;
1347         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1348                        LA_ATIME | LA_MTIME | LA_CTIME;
1349
1350         orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1351                                    la->la_mode & S_IFMT);
1352
1353         memset(dof, 0, sizeof(*dof));
1354         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1355
1356         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1357                               cname, lfsck_dto2fid(parent));
1358         if (rc != 0)
1359                 GOTO(unlock1, rc);
1360
1361         th = dt_trans_create(env, dev);
1362         if (IS_ERR(th))
1363                 GOTO(unlock1, rc = PTR_ERR(th));
1364
1365         /* Sync the remote transaction to guarantee that the subsequent
1366          * lock against the @orphan can find the @orphan in time. */
1367         if (dt_object_remote(orphan))
1368                 th->th_sync = 1;
1369
1370         rc = dt_declare_create(env, orphan, la, hint, dof, th);
1371         if (rc != 0)
1372                 GOTO(stop, rc);
1373
1374         if (unlikely(!dt_try_as_dir(env, orphan)))
1375                 GOTO(stop, rc = -ENOTDIR);
1376
1377         rc = dt_declare_ref_add(env, orphan, th);
1378         if (rc != 0)
1379                 GOTO(stop, rc);
1380
1381         rec->rec_type = S_IFDIR;
1382         rec->rec_fid = cfid;
1383         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1384                                (const struct dt_key *)dot, th);
1385         if (rc != 0)
1386                 GOTO(stop, rc);
1387
1388         rec->rec_fid = lfsck_dto2fid(parent);
1389         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1390                                (const struct dt_key *)dotdot, th);
1391         if (rc != 0)
1392                 GOTO(stop, rc);
1393
1394         if (lmv != NULL) {
1395                 lmv->lmv_magic = LMV_MAGIC;
1396                 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1397                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1398                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1399                 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1400                                           XATTR_NAME_LMV, 0, th);
1401                 if (rc != 0)
1402                         GOTO(stop, rc);
1403         }
1404
1405         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1406                        ldata.ld_leh->leh_len);
1407         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1408                                   XATTR_NAME_LINK, 0, th);
1409         if (rc != 0)
1410                 GOTO(stop, rc);
1411
1412         rec->rec_fid = cfid;
1413         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1414                                (const struct dt_key *)name, th);
1415         if (rc == 0)
1416                 rc = dt_declare_ref_add(env, parent, th);
1417
1418         if (rc != 0)
1419                 GOTO(stop, rc);
1420
1421         rc = dt_trans_start_local(env, dev, th);
1422         if (rc != 0)
1423                 GOTO(stop, rc);
1424
1425         dt_write_lock(env, orphan, 0);
1426         rc = dt_create(env, orphan, la, hint, dof, th);
1427         if (rc != 0)
1428                 GOTO(unlock2, rc);
1429
1430         rc = dt_ref_add(env, orphan, th);
1431         if (rc != 0)
1432                 GOTO(unlock2, rc);
1433
1434         rec->rec_fid = cfid;
1435         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1436                        (const struct dt_key *)dot, th, 1);
1437         if (rc != 0)
1438                 GOTO(unlock2, rc);
1439
1440         rec->rec_fid = lfsck_dto2fid(parent);
1441         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1442                        (const struct dt_key *)dotdot, th, 1);
1443         if (rc != 0)
1444                 GOTO(unlock2, rc);
1445
1446         if (lmv != NULL) {
1447                 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1448                 if (rc != 0)
1449                         GOTO(unlock2, rc);
1450         }
1451
1452         rc = dt_xattr_set(env, orphan, &linkea_buf,
1453                           XATTR_NAME_LINK, 0, th);
1454         dt_write_unlock(env, orphan);
1455         if (rc != 0)
1456                 GOTO(stop, rc);
1457
1458         rec->rec_fid = cfid;
1459         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1460                        (const struct dt_key *)name, th, 1);
1461         if (rc == 0) {
1462                 dt_write_lock(env, parent, 0);
1463                 rc = dt_ref_add(env, parent, th);
1464                 dt_write_unlock(env, parent);
1465         }
1466
1467         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1468
1469 unlock2:
1470         dt_write_unlock(env, orphan);
1471
1472 stop:
1473         rc1 = dt_trans_stop(env, dev, th);
1474         if (rc1 != 0 && rc > 0)
1475                 rc = rc1;
1476
1477 unlock1:
1478         lfsck_unlock(llh);
1479
1480 log:
1481         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1482                "the object "DFID", name = %s: rc = %d\n",
1483                lfsck_lfsck2name(lfsck), PFID(cfid),
1484                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1485
1486         if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1487                 lfsck_object_put(env, parent);
1488
1489         if (rc != 0)
1490                 ns->ln_flags |= LF_INCONSISTENT;
1491
1492         return rc;
1493 }
1494
1495 /**
1496  * Remove the specified entry from the linkEA.
1497  *
1498  * Locate the linkEA entry with the given @cname and @pfid, then
1499  * remove this entry or the other entries those are repeated with
1500  * this entry.
1501  *
1502  * \param[in] env       pointer to the thread context
1503  * \param[in] com       pointer to the lfsck component
1504  * \param[in] obj       pointer to the dt_object to be handled
1505  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1506  * \param[in] cname     the name for the child in the parent directory
1507  * \param[in] pfid      the parent directory's FID for the linkEA
1508  * \param[in] next      if true, then remove the first found linkEA
1509  *                      entry, and move the ldata->ld_lee to next entry
1510  *
1511  * \retval              positive number for repaired cases
1512  * \retval              0 if nothing to be repaired
1513  * \retval              negative error number on failure
1514  */
1515 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1516                                          struct lfsck_component *com,
1517                                          struct dt_object *obj,
1518                                          struct linkea_data *ldata,
1519                                          struct lu_name *cname,
1520                                          struct lu_fid *pfid,
1521                                          bool next)
1522 {
1523         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1524         struct dt_device                *dev       = lfsck_obj2dev(obj);
1525         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1526         struct thandle                  *th        = NULL;
1527         struct lustre_handle             lh        = { 0 };
1528         struct linkea_data               ldata_new = { NULL };
1529         struct lu_buf                    linkea_buf;
1530         int                              buflen    = 0;
1531         int                              rc        = 0;
1532         ENTRY;
1533
1534         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1535                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1536                               LCK_EX);
1537         if (rc != 0)
1538                 GOTO(log, rc);
1539
1540         if (next)
1541                 lfsck_linkea_del_buf(ldata, cname);
1542         else
1543                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1544                                                     true);
1545         if (ldata->ld_leh->leh_reccount > 0 ||
1546             unlikely(ldata->ld_leh->leh_overflow_time)) {
1547                 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1548                                ldata->ld_leh->leh_len);
1549                 buflen = linkea_buf.lb_len;
1550         }
1551
1552 again:
1553         th = dt_trans_create(env, dev);
1554         if (IS_ERR(th))
1555                 GOTO(unlock1, rc = PTR_ERR(th));
1556
1557         if (buflen != 0)
1558                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1559                                           XATTR_NAME_LINK, 0, th);
1560         else
1561                 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1562         if (rc != 0)
1563                 GOTO(stop, rc);
1564
1565         rc = dt_trans_start_local(env, dev, th);
1566         if (rc != 0)
1567                 GOTO(stop, rc);
1568
1569         dt_write_lock(env, obj, 0);
1570         if (unlikely(lfsck_is_dead_obj(obj)))
1571                 GOTO(unlock2, rc = -ENOENT);
1572
1573         rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1574         if (rc)
1575                 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1576
1577         /* The specified linkEA entry has been removed by race. */
1578         rc = linkea_links_find(&ldata_new, cname, pfid);
1579         if (rc != 0)
1580                 GOTO(unlock2, rc = 0);
1581
1582         if (bk->lb_param & LPF_DRYRUN)
1583                 GOTO(unlock2, rc = 1);
1584
1585         if (next)
1586                 lfsck_linkea_del_buf(&ldata_new, cname);
1587         else
1588                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1589                                                     true);
1590
1591         if (buflen < ldata_new.ld_leh->leh_len) {
1592                 dt_write_unlock(env, obj);
1593                 dt_trans_stop(env, dev, th);
1594                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1595                                ldata_new.ld_leh->leh_len);
1596                 goto again;
1597         }
1598
1599         if (ldata_new.ld_leh->leh_reccount > 0 ||
1600             unlikely(ldata->ld_leh->leh_overflow_time))
1601                 rc = lfsck_links_write(env, obj, &ldata_new, th);
1602         else
1603                 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1604
1605         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1606
1607 unlock2:
1608         dt_write_unlock(env, obj);
1609
1610 stop:
1611         dt_trans_stop(env, dev, th);
1612
1613 unlock1:
1614         lfsck_ibits_unlock(&lh, LCK_EX);
1615
1616 log:
1617         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1618                "for the object: "DFID", parent "DFID", name %.*s\n",
1619                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1620                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1621                cname->ln_name);
1622
1623         if (rc != 0) {
1624                 struct lfsck_namespace *ns = com->lc_file_ram;
1625
1626                 ns->ln_flags |= LF_INCONSISTENT;
1627         }
1628
1629         return rc;
1630 }
1631
1632 /**
1633  * Conditionally remove the specified entry from the linkEA.
1634  *
1635  * Take the parent lock firstly, then check whether the specified
1636  * name entry exists or not: if yes, do nothing; otherwise, call
1637  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1638  *
1639  * \param[in] env       pointer to the thread context
1640  * \param[in] com       pointer to the lfsck component
1641  * \param[in] parent    pointer to the parent directory
1642  * \param[in] child     pointer to the child object that holds the linkEA
1643  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1644  * \param[in] cname     the name for the child in the parent directory
1645  * \param[in] pfid      the parent directory's FID for the linkEA
1646  *
1647  * \retval              positive number for repaired cases
1648  * \retval              0 if nothing to be repaired
1649  * \retval              negative error number on failure
1650  */
1651 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1652                                               struct lfsck_component *com,
1653                                               struct dt_object *parent,
1654                                               struct dt_object *child,
1655                                               struct linkea_data *ldata,
1656                                               struct lu_name *cname,
1657                                               struct lu_fid *pfid)
1658 {
1659         struct lfsck_thread_info *info  = lfsck_env_info(env);
1660         struct lu_fid            *cfid  = &info->lti_fid3;
1661         struct lfsck_lock_handle *llh   = &info->lti_llh;
1662         int                       rc;
1663         ENTRY;
1664
1665         rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1666                         MDS_INODELOCK_UPDATE, LCK_PR);
1667         if (rc != 0)
1668                 RETURN(rc);
1669
1670         dt_read_lock(env, parent, 0);
1671         if (unlikely(lfsck_is_dead_obj(parent))) {
1672                 dt_read_unlock(env, parent);
1673                 lfsck_unlock(llh);
1674                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1675                                                    cname, pfid, true);
1676
1677                 RETURN(rc);
1678         }
1679
1680         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1681                        (const struct dt_key *)cname->ln_name);
1682         dt_read_unlock(env, parent);
1683
1684         /* It is safe to release the ldlm lock, because when the logic come
1685          * here, we have got all the needed information above whether the
1686          * linkEA entry is valid or not. It is not important that others
1687          * may add new linkEA entry after the ldlm lock released. If other
1688          * has removed the specified linkEA entry by race, then it is OK,
1689          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1690          * such case. */
1691         lfsck_unlock(llh);
1692         if (rc == -ENOENT) {
1693                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1694                                                    cname, pfid, true);
1695
1696                 RETURN(rc);
1697         }
1698
1699         if (rc != 0)
1700                 RETURN(rc);
1701
1702         /* The LFSCK just found some internal status of cross-MDTs
1703          * create operation. That is normal. */
1704         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1705                 linkea_next_entry(ldata);
1706
1707                 RETURN(0);
1708         }
1709
1710         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1711                                            pfid, true);
1712
1713         RETURN(rc);
1714 }
1715
1716 /**
1717  * Conditionally replace name entry in the parent.
1718  *
1719  * As required, the LFSCK may re-create the lost MDT-object for dangling
1720  * name entry, but such repairing may be wrong because of bad FID in the
1721  * name entry. As the LFSCK processing, the real MDT-object may be found,
1722  * then the LFSCK should check whether the former re-created MDT-object
1723  * has been modified or not, if not, then destroy it and update the name
1724  * entry in the parent to reference the real MDT-object.
1725  *
1726  * \param[in] env       pointer to the thread context
1727  * \param[in] com       pointer to the lfsck component
1728  * \param[in] parent    pointer to the parent directory
1729  * \param[in] child     pointer to the MDT-object that may be the real
1730  *                      MDT-object corresponding to the name entry in parent
1731  * \param[in] cfid      the current FID in the name entry
1732  * \param[in] cname     contains the name of the child in the parent directory
1733  *
1734  * \retval              positive number for repaired cases
1735  * \retval              0 if nothing to be repaired
1736  * \retval              negative error number on failure
1737  */
1738 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1739                                         struct lfsck_component *com,
1740                                         struct dt_object *parent,
1741                                         struct dt_object *child,
1742                                         const struct lu_fid *cfid,
1743                                         const struct lu_name *cname)
1744 {
1745         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1746         struct lu_attr                  *la     = &info->lti_la;
1747         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1748         struct lu_fid                    tfid;
1749         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1750         /* The child and its name may be on different MDTs. */
1751         struct dt_device                *dev    = lfsck->li_next;
1752         const char                      *name   = cname->ln_name;
1753         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1754         struct dt_object                *cobj   = NULL;
1755         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
1756         struct lustre_handle             clh    = { 0 };
1757         struct linkea_data               ldata  = { NULL };
1758         struct thandle                  *th     = NULL;
1759         bool                             exist  = true;
1760         int                              rc     = 0;
1761         ENTRY;
1762
1763         /* @parent/@child may be based on lfsck->li_bottom,
1764          * but here we need the object based on the lfsck->li_next. */
1765
1766         parent = lfsck_object_locate(dev, parent);
1767         if (IS_ERR(parent))
1768                 GOTO(log, rc = PTR_ERR(parent));
1769
1770         if (unlikely(!dt_try_as_dir(env, parent)))
1771                 GOTO(log, rc = -ENOTDIR);
1772
1773         rc = lfsck_lock(env, lfsck, parent, name, pllh,
1774                         MDS_INODELOCK_UPDATE, LCK_PW);
1775         if (rc != 0)
1776                 GOTO(log, rc);
1777
1778         if (!fid_is_sane(cfid)) {
1779                 exist = false;
1780                 goto replace;
1781         }
1782
1783         cobj = lfsck_object_find_by_dev(env, dev, cfid);
1784         if (IS_ERR(cobj)) {
1785                 rc = PTR_ERR(cobj);
1786                 if (rc == -ENOENT) {
1787                         exist = false;
1788                         goto replace;
1789                 }
1790
1791                 GOTO(log, rc);
1792         }
1793
1794         if (!dt_object_exists(cobj)) {
1795                 exist = false;
1796                 goto replace;
1797         }
1798
1799         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1800                        (const struct dt_key *)name);
1801         if (rc == -ENOENT) {
1802                 exist = false;
1803                 goto replace;
1804         }
1805
1806         if (rc != 0)
1807                 GOTO(log, rc);
1808
1809         /* Someone changed the name entry, cannot replace it. */
1810         if (!lu_fid_eq(cfid, &tfid))
1811                 GOTO(log, rc = 0);
1812
1813         /* lock the object to be destroyed. */
1814         rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1815                               MDS_INODELOCK_UPDATE |
1816                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1817                               LCK_EX);
1818         if (rc != 0)
1819                 GOTO(log, rc);
1820
1821         if (unlikely(lfsck_is_dead_obj(cobj))) {
1822                 exist = false;
1823                 goto replace;
1824         }
1825
1826         rc = dt_attr_get(env, cobj, la);
1827         if (rc != 0)
1828                 GOTO(log, rc);
1829
1830         /* The object has been modified by other(s), or it is not created by
1831          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1832         if (la->la_ctime != 0)
1833                 GOTO(log, rc);
1834
1835         if (S_ISREG(la->la_mode)) {
1836                 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
1837                 /* If someone has created related OST-object(s),
1838                  * then keep it. */
1839                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1840                         GOTO(log, rc = (rc > 0 ? 0 : rc));
1841         }
1842
1843 replace:
1844         dt_read_lock(env, child, 0);
1845         rc = lfsck_links_read2_with_rec(env, child, &ldata);
1846         dt_read_unlock(env, child);
1847
1848         /* Someone changed the child, no need to replace. */
1849         if (rc == -ENODATA)
1850                 GOTO(log, rc = 0);
1851
1852         if (rc != 0)
1853                 GOTO(log, rc);
1854
1855         rc = linkea_links_find(&ldata, cname, pfid);
1856         /* Someone moved the child, no need to replace. */
1857         if (rc != 0)
1858                 GOTO(log, rc = 0);
1859
1860         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1861                 GOTO(log, rc = 1);
1862
1863         th = dt_trans_create(env, dev);
1864         if (IS_ERR(th))
1865                 GOTO(log, rc = PTR_ERR(th));
1866
1867         if (exist) {
1868                 rc = dt_declare_destroy(env, cobj, th);
1869                 if (rc != 0)
1870                         GOTO(stop, rc);
1871         }
1872
1873         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1874         if (rc != 0)
1875                 GOTO(stop, rc);
1876
1877         rec->rec_type = S_IFDIR;
1878         rec->rec_fid = lfsck_dto2fid(child);
1879         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1880                                (const struct dt_key *)name, th);
1881         if (rc != 0)
1882                 GOTO(stop, rc);
1883
1884         rc = dt_trans_start_local(env, dev, th);
1885         if (rc != 0)
1886                 GOTO(stop, rc);
1887
1888         if (exist) {
1889                 rc = dt_destroy(env, cobj, th);
1890                 if (rc != 0)
1891                         GOTO(stop, rc);
1892         }
1893
1894         /* The old name entry maybe not exist. */
1895         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
1896         if (rc != 0 && rc != -ENOENT)
1897                 GOTO(stop, rc);
1898
1899         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1900                        (const struct dt_key *)name, th, 1);
1901
1902         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1903
1904 stop:
1905         dt_trans_stop(env, dev, th);
1906
1907 log:
1908         lfsck_ibits_unlock(&clh, LCK_EX);
1909         lfsck_unlock(pllh);
1910
1911         if (cobj != NULL && !IS_ERR(cobj))
1912                 lfsck_object_put(env, cobj);
1913
1914         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1915                "object "DFID" because of conflict with the object "DFID
1916                " under the parent "DFID" with name %s: rc = %d\n",
1917                lfsck_lfsck2name(lfsck), PFID(cfid),
1918                PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
1919
1920         return rc;
1921 }
1922
1923 /**
1924  * Overwrite the linkEA for the object with the given ldata.
1925  *
1926  * The caller should take the ldlm lock before the calling.
1927  *
1928  * \param[in] env       pointer to the thread context
1929  * \param[in] com       pointer to the lfsck component
1930  * \param[in] obj       pointer to the dt_object to be handled
1931  * \param[in] ldata     pointer to the new linkEA data
1932  *
1933  * \retval              positive number for repaired cases
1934  * \retval              0 if nothing to be repaired
1935  * \retval              negative error number on failure
1936  */
1937 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1938                                    struct lfsck_component *com,
1939                                    struct dt_object *obj,
1940                                    struct linkea_data *ldata)
1941 {
1942         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1943         struct dt_device                *dev    = lfsck_obj2dev(obj);
1944         struct thandle                  *th     = NULL;
1945         struct lu_buf                    linkea_buf;
1946         int                              rc     = 0;
1947         ENTRY;
1948
1949         th = dt_trans_create(env, dev);
1950         if (IS_ERR(th))
1951                 GOTO(log, rc = PTR_ERR(th));
1952
1953         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1954                        ldata->ld_leh->leh_len);
1955         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1956                                   XATTR_NAME_LINK, 0, th);
1957         if (rc != 0)
1958                 GOTO(stop, rc);
1959
1960         rc = dt_trans_start_local(env, dev, th);
1961         if (rc != 0)
1962                 GOTO(stop, rc);
1963
1964         dt_write_lock(env, obj, 0);
1965         if (unlikely(lfsck_is_dead_obj(obj)))
1966                 GOTO(unlock, rc = 0);
1967
1968         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1969                 GOTO(unlock, rc = 1);
1970
1971         rc = dt_xattr_set(env, obj, &linkea_buf,
1972                           XATTR_NAME_LINK, 0, th);
1973
1974         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1975
1976 unlock:
1977         dt_write_unlock(env, obj);
1978
1979 stop:
1980         dt_trans_stop(env, dev, th);
1981
1982 log:
1983         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1984                "object "DFID": rc = %d\n",
1985                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1986
1987         if (rc != 0) {
1988                 struct lfsck_namespace *ns = com->lc_file_ram;
1989
1990                 ns->ln_flags |= LF_INCONSISTENT;
1991         }
1992
1993         return rc;
1994 }
1995
1996 /**
1997  * Repair invalid name entry.
1998  *
1999  * If the name entry contains invalid information, such as bad file type
2000  * or (and) corrupted object FID, then either remove the name entry or
2001  * udpate the name entry with the given (right) information.
2002  *
2003  * \param[in] env       pointer to the thread context
2004  * \param[in] com       pointer to the lfsck component
2005  * \param[in] parent    pointer to the parent directory
2006  * \param[in] child     pointer to the object referenced by the name entry
2007  * \param[in] name      the old name of the child under the parent directory
2008  * \param[in] name2     the new name of the child under the parent directory
2009  * \param[in] type      the type claimed by the name entry
2010  * \param[in] update    update the name entry if true; otherwise, remove it
2011  * \param[in] dec       decrease the parent nlink count if true
2012  *
2013  * \retval              positive number for repaired successfully
2014  * \retval              0 if nothing to be repaired
2015  * \retval              negative error number on failure
2016  */
2017 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2018                                   struct lfsck_component *com,
2019                                   struct dt_object *parent,
2020                                   struct dt_object *child,
2021                                   const char *name, const char *name2,
2022                                   __u16 type, bool update, bool dec)
2023 {
2024         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2025         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2026         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2027         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
2028         struct lu_fid                    tfid;
2029         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2030         struct dt_device                *dev    = lfsck->li_next;
2031         struct thandle                  *th     = NULL;
2032         struct lfsck_lock_handle        *llh    = &info->lti_llh;
2033         struct lustre_handle             lh     = { 0 };
2034         int                              rc     = 0;
2035         ENTRY;
2036
2037         parent = lfsck_object_locate(dev, parent);
2038         if (IS_ERR(parent))
2039                 GOTO(log, rc = PTR_ERR(parent));
2040
2041         if (unlikely(!dt_try_as_dir(env, parent)))
2042                 GOTO(log, rc = -ENOTDIR);
2043
2044         if (!update || strcmp(name, name2) == 0)
2045                 rc = lfsck_lock(env, lfsck, parent, name, llh,
2046                                 MDS_INODELOCK_UPDATE, LCK_PW);
2047         else
2048                 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2049                                       MDS_INODELOCK_UPDATE, LCK_PW);
2050         if (rc != 0)
2051                 GOTO(log, rc);
2052
2053         th = dt_trans_create(env, dev);
2054         if (IS_ERR(th))
2055                 GOTO(unlock1, rc = PTR_ERR(th));
2056
2057         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2058         if (rc != 0)
2059                 GOTO(stop, rc);
2060
2061         if (update) {
2062                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2063                 rec->rec_fid = cfid;
2064                 rc = dt_declare_insert(env, parent,
2065                                        (const struct dt_rec *)rec,
2066                                        (const struct dt_key *)name2, th);
2067                 if (rc != 0)
2068                         GOTO(stop, rc);
2069         }
2070
2071         if (dec && S_ISDIR(type)) {
2072                 rc = dt_declare_ref_del(env, parent, th);
2073                 if (rc != 0)
2074                         GOTO(stop, rc);
2075         }
2076
2077         rc = dt_trans_start_local(env, dev, th);
2078         if (rc != 0)
2079                 GOTO(stop, rc);
2080
2081
2082         dt_write_lock(env, parent, 0);
2083         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2084                        (const struct dt_key *)name);
2085         /* Someone has removed the bad name entry by race. */
2086         if (rc == -ENOENT)
2087                 GOTO(unlock2, rc = 0);
2088
2089         if (rc != 0)
2090                 GOTO(unlock2, rc);
2091
2092         /* Someone has removed the bad name entry and reused it for other
2093          * object by race. */
2094         if (!lu_fid_eq(&tfid, cfid))
2095                 GOTO(unlock2, rc = 0);
2096
2097         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2098                 GOTO(unlock2, rc = 1);
2099
2100         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2101         if (rc != 0)
2102                 GOTO(unlock2, rc);
2103
2104         if (update) {
2105                 rc = dt_insert(env, parent,
2106                                (const struct dt_rec *)rec,
2107                                (const struct dt_key *)name2, th, 1);
2108                 if (rc != 0)
2109                         GOTO(unlock2, rc);
2110         }
2111
2112         if (dec && S_ISDIR(type)) {
2113                 rc = dt_ref_del(env, parent, th);
2114                 if (rc != 0)
2115                         GOTO(unlock2, rc);
2116         }
2117
2118         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2119
2120 unlock2:
2121         dt_write_unlock(env, parent);
2122
2123 stop:
2124         dt_trans_stop(env, dev, th);
2125
2126         /* We are not sure whether the child will become orphan or not.
2127          * Record it in the LFSCK trace file for further checking in
2128          * the second-stage scanning. */
2129         if (!update && !dec && rc == 0)
2130                 lfsck_namespace_trace_update(env, com, cfid,
2131                                              LNTF_CHECK_LINKEA, true);
2132
2133 unlock1:
2134         /* It is harmless even if unlock the unused lock_handle */
2135         lfsck_ibits_unlock(&lh, LCK_PW);
2136         lfsck_unlock(llh);
2137
2138 log:
2139         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2140                "entry for: parent "DFID", child "DFID", name %s, type "
2141                "in name entry %o, type claimed by child %o. repair it "
2142                "by %s with new name2 %s: rc = %d\n",
2143                lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2144                name, type, update ? lfsck_object_type(child) : 0,
2145                update ? "updating" : "removing", name2, rc);
2146
2147         if (rc != 0) {
2148                 struct lfsck_namespace *ns = com->lc_file_ram;
2149
2150                 ns->ln_flags |= LF_INCONSISTENT;
2151         }
2152
2153         return rc;
2154 }
2155
2156 /**
2157  * Update the ".." name entry for the given object.
2158  *
2159  * The object's ".." is corrupted, this function will update the ".." name
2160  * entry with the given pfid, and the linkEA with the given ldata.
2161  *
2162  * The caller should take the ldlm lock before the calling.
2163  *
2164  * \param[in] env       pointer to the thread context
2165  * \param[in] com       pointer to the lfsck component
2166  * \param[in] obj       pointer to the dt_object to be handled
2167  * \param[in] pfid      the new fid for the object's ".." name entry
2168  * \param[in] cname     the name for the @obj in the parent directory
2169  *
2170  * \retval              positive number for repaired cases
2171  * \retval              0 if nothing to be repaired
2172  * \retval              negative error number on failure
2173  */
2174 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2175                                                   struct lfsck_component *com,
2176                                                   struct dt_object *obj,
2177                                                   const struct lu_fid *pfid,
2178                                                   struct lu_name *cname)
2179 {
2180         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2181         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2182         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2183         struct dt_device                *dev    = lfsck_obj2dev(obj);
2184         struct thandle                  *th     = NULL;
2185         struct linkea_data               ldata  = { NULL };
2186         struct lu_buf                    linkea_buf;
2187         int                              rc     = 0;
2188         ENTRY;
2189
2190         LASSERT(!dt_object_remote(obj));
2191         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2192
2193         rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2194         if (rc != 0)
2195                 GOTO(log, rc);
2196
2197         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2198                        ldata.ld_leh->leh_len);
2199
2200         th = dt_trans_create(env, dev);
2201         if (IS_ERR(th))
2202                 GOTO(log, rc = PTR_ERR(th));
2203
2204         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2205         if (rc != 0)
2206                 GOTO(stop, rc);
2207
2208         rec->rec_type = S_IFDIR;
2209         rec->rec_fid = pfid;
2210         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2211                                (const struct dt_key *)dotdot, th);
2212         if (rc != 0)
2213                 GOTO(stop, rc);
2214
2215         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2216                                   XATTR_NAME_LINK, 0, th);
2217         if (rc != 0)
2218                 GOTO(stop, rc);
2219
2220         rc = dt_trans_start_local(env, dev, th);
2221         if (rc != 0)
2222                 GOTO(stop, rc);
2223
2224         dt_write_lock(env, obj, 0);
2225         if (unlikely(lfsck_is_dead_obj(obj)))
2226                 GOTO(unlock, rc = 0);
2227
2228         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2229                 GOTO(unlock, rc = 1);
2230
2231         /* The old ".." name entry maybe not exist. */
2232         dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2233
2234         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2235                        (const struct dt_key *)dotdot, th, 1);
2236         if (rc != 0)
2237                 GOTO(unlock, rc);
2238
2239         rc = lfsck_links_write(env, obj, &ldata, th);
2240
2241         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2242
2243 unlock:
2244         dt_write_unlock(env, obj);
2245
2246 stop:
2247         dt_trans_stop(env, dev, th);
2248
2249 log:
2250         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2251                "the object "DFID", new parent "DFID": rc = %d\n",
2252                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2253                PFID(pfid), rc);
2254
2255         if (rc != 0) {
2256                 struct lfsck_namespace *ns = com->lc_file_ram;
2257
2258                 ns->ln_flags |= LF_INCONSISTENT;
2259         }
2260
2261         return rc;
2262 }
2263
2264 /**
2265  * Handle orphan @obj during Double Scan Directory.
2266  *
2267  * Remove the @obj's current (invalid) linkEA entries, and insert
2268  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2269  * ${FID}-${PFID}-D-${conflict_version}
2270  *
2271  * The caller should take the ldlm lock before the calling.
2272  *
2273  * \param[in] env       pointer to the thread context
2274  * \param[in] com       pointer to the lfsck component
2275  * \param[in] obj       pointer to the orphan object to be handled
2276  * \param[in] pfid      the new fid for the object's ".." name entry
2277  * \param[in,out] lh    ldlm lock handler for the given @obj
2278  * \param[out] type     to tell the caller what the inconsistency is
2279  *
2280  * \retval              positive number for repaired cases
2281  * \retval              0 if nothing to be repaired
2282  * \retval              negative error number on failure
2283  */
2284 static int
2285 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2286                            struct lfsck_component *com,
2287                            struct dt_object *obj,
2288                            const struct lu_fid *pfid,
2289                            struct lustre_handle *lh,
2290                            enum lfsck_namespace_inconsistency_type *type)
2291 {
2292         struct lfsck_thread_info *info = lfsck_env_info(env);
2293         struct lfsck_namespace   *ns   = com->lc_file_ram;
2294         int                       rc;
2295         ENTRY;
2296
2297         /* Remove the unrecognized linkEA. */
2298         rc = lfsck_namespace_links_remove(env, com, obj);
2299         lfsck_ibits_unlock(lh, LCK_EX);
2300         if (rc < 0 && rc != -ENODATA)
2301                 RETURN(rc);
2302
2303         *type = LNIT_MUL_REF;
2304
2305         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2306          * ever tried to verify some remote MDT-object that resides on this
2307          * MDT, but this MDT failed to respond such request. So means there
2308          * may be some remote name entry on other MDT that references this
2309          * object with another name, so we cannot know whether this linkEA
2310          * is valid or not. So keep it there and maybe resolved when next
2311          * LFSCK run. */
2312         if (ns->ln_flags & LF_INCOMPLETE)
2313                 RETURN(0);
2314
2315         /* The unique linkEA is invalid, even if the ".." name entry may be
2316          * valid, we still cannot know via which name entry this directory
2317          * will be referenced. Then handle it as pure orphan. */
2318         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2319                  "-"DFID, PFID(pfid));
2320         rc = lfsck_namespace_insert_orphan(env, com, obj,
2321                                            info->lti_tmpbuf, "D", NULL);
2322
2323         RETURN(rc);
2324 }
2325
2326 /**
2327  * Double Scan Directory object for single linkEA entry case.
2328  *
2329  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2330  * then check whether the name is in the namespace or not, if not, add the
2331  * missing name entry back to namespace. If the linkEA entry is invalid,
2332  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2333  * as an orphan.
2334  *
2335  * \param[in] env       pointer to the thread context
2336  * \param[in] com       pointer to the lfsck component
2337  * \param[in] child     pointer to the directory to be double scanned
2338  * \param[in] pfid      the FID corresponding to the ".." entry
2339  * \param[in] ldata     pointer to the linkEA data for the given @child
2340  * \param[in,out] lh    ldlm lock handler for the given @child
2341  * \param[out] type     to tell the caller what the inconsistency is
2342  * \param[in] retry     if found inconsistency, but the caller does not hold
2343  *                      ldlm lock on the @child, then set @retry as true
2344  * \param[in] unknown   set if does not know how to repair the inconsistency
2345  *
2346  * \retval              positive number for repaired cases
2347  * \retval              0 if nothing to be repaired
2348  * \retval              negative error number on failure
2349  */
2350 static int
2351 lfsck_namespace_dsd_single(const struct lu_env *env,
2352                            struct lfsck_component *com,
2353                            struct dt_object *child,
2354                            const struct lu_fid *pfid,
2355                            struct linkea_data *ldata,
2356                            struct lustre_handle *lh,
2357                            enum lfsck_namespace_inconsistency_type *type,
2358                            bool *retry, bool *unknown)
2359 {
2360         struct lfsck_thread_info *info          = lfsck_env_info(env);
2361         struct lu_name           *cname         = &info->lti_name;
2362         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2363         struct lu_fid             tfid;
2364         struct lfsck_namespace   *ns            = com->lc_file_ram;
2365         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2366         struct dt_object         *parent        = NULL;
2367         struct lmv_mds_md_v1     *lmv;
2368         int                       rc            = 0;
2369         ENTRY;
2370
2371         rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2372                                                  info->lti_key,
2373                                                  sizeof(info->lti_key));
2374         /* The unique linkEA entry with bad parent will be handled as orphan. */
2375         if (rc != 0) {
2376                 if (!lustre_handle_is_used(lh) && retry != NULL)
2377                         *retry = true;
2378                 else
2379                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2380                                                         pfid, lh, type);
2381
2382                 GOTO(out, rc);
2383         }
2384
2385         parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2386         if (IS_ERR(parent))
2387                 GOTO(out, rc = PTR_ERR(parent));
2388
2389         /* We trust the unique linkEA entry in spite of whether it matches the
2390          * ".." name entry or not. Because even if the linkEA entry is wrong
2391          * and the ".." name entry is right, we still cannot know via which
2392          * name entry the child will be referenced, since all known entries
2393          * have been verified during the first-stage scanning. */
2394         if (!dt_object_exists(parent)) {
2395                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2396                  * has ever tried to verify some remote MDT-object that resides
2397                  * on this MDT, but this MDT failed to respond such request. So
2398                  * means there may be some remote name entry on other MDT that
2399                  * references this object with another name, so we cannot know
2400                  * whether this linkEA is valid or not. So keep it there and
2401                  * maybe resolved when next LFSCK run. */
2402                 if (ns->ln_flags & LF_INCOMPLETE)
2403                         GOTO(out, rc = 0);
2404
2405                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2406                         *retry = true;
2407
2408                         GOTO(out, rc = 0);
2409                 }
2410
2411                 lfsck_ibits_unlock(lh, LCK_EX);
2412
2413 lost_parent:
2414                 lmv = &info->lti_lmv;
2415                 rc = lfsck_read_stripe_lmv(env, child, lmv);
2416                 if (rc != 0 && rc != -ENODATA)
2417                         GOTO(out, rc);
2418
2419                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2420                         lmv = NULL;
2421                 } else if (lfsck_shard_name_to_index(env,
2422                                         cname->ln_name, cname->ln_namelen,
2423                                         S_IFDIR, cfid) < 0) {
2424                         /* It is an invalid name entry, we
2425                          * cannot trust the parent also. */
2426                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2427                                                 ldata, cname, &tfid, true);
2428                         if (rc < 0)
2429                                 GOTO(out, rc);
2430
2431                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2432                                  "-"DFID, PFID(pfid));
2433                         rc = lfsck_namespace_insert_orphan(env, com, child,
2434                                                 info->lti_tmpbuf, "S", NULL);
2435
2436                         GOTO(out, rc);
2437                 }
2438
2439                 /* Create the lost parent as an orphan. */
2440                 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2441                 if (rc >= 0) {
2442                         /* Add the missing name entry to the parent. */
2443                         rc = lfsck_namespace_insert_normal(env, com, parent,
2444                                                         child, cname->ln_name);
2445                         if (unlikely(rc == -EEXIST)) {
2446                                 /* Unfortunately, someone reused the name
2447                                  * under the parent by race. So we have
2448                                  * to remove the linkEA entry from
2449                                  * current child object. It means that the
2450                                  * LFSCK cannot recover the system
2451                                  * totally back to its original status,
2452                                  * but it is necessary to make the
2453                                  * current system to be consistent. */
2454                                 rc = lfsck_namespace_shrink_linkea(env,
2455                                                 com, child, ldata,
2456                                                 cname, &tfid, true);
2457                                 if (rc >= 0) {
2458                                         snprintf(info->lti_tmpbuf,
2459                                                  sizeof(info->lti_tmpbuf),
2460                                                  "-"DFID, PFID(pfid));
2461                                         rc = lfsck_namespace_insert_orphan(env,
2462                                                 com, child, info->lti_tmpbuf,
2463                                                 "D", NULL);
2464                                 }
2465                         }
2466                 }
2467
2468                 GOTO(out, rc);
2469         } /* !dt_object_exists(parent) */
2470
2471         /* The unique linkEA entry with bad parent will be handled as orphan. */
2472         if (unlikely(!dt_try_as_dir(env, parent))) {
2473                 if (!lustre_handle_is_used(lh) && retry != NULL)
2474                         *retry = true;
2475                 else
2476                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2477                                                         pfid, lh, type);
2478
2479                 GOTO(out, rc);
2480         }
2481
2482         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2483                        (const struct dt_key *)cname->ln_name);
2484         if (rc == -ENOENT) {
2485                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2486                  * has ever tried to verify some remote MDT-object that resides
2487                  * on this MDT, but this MDT failed to respond such request. So
2488                  * means there may be some remote name entry on other MDT that
2489                  * references this object with another name, so we cannot know
2490                  * whether this linkEA is valid or not. So keep it there and
2491                  * maybe resolved when next LFSCK run. */
2492                 if (ns->ln_flags & LF_INCOMPLETE)
2493                         GOTO(out, rc = 0);
2494
2495                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2496                         *retry = true;
2497
2498                         GOTO(out, rc = 0);
2499                 }
2500
2501                 lfsck_ibits_unlock(lh, LCK_EX);
2502                 rc = lfsck_namespace_check_name(env, parent, child, cname);
2503                 if (rc == -ENOENT)
2504                         goto lost_parent;
2505
2506                 if (rc < 0)
2507                         GOTO(out, rc);
2508
2509                 /* It is an invalid name entry, drop it. */
2510                 if (unlikely(rc > 0)) {
2511                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2512                                                 ldata, cname, &tfid, true);
2513                         if (rc >= 0) {
2514                                 snprintf(info->lti_tmpbuf,
2515                                          sizeof(info->lti_tmpbuf),
2516                                          "-"DFID, PFID(pfid));
2517                                 rc = lfsck_namespace_insert_orphan(env, com,
2518                                         child, info->lti_tmpbuf, "D", NULL);
2519                         }
2520
2521                         GOTO(out, rc);
2522                 }
2523
2524                 /* Add the missing name entry back to the namespace. */
2525                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2526                                                    cname->ln_name);
2527                 if (unlikely(rc == -ESTALE))
2528                         /* It may happen when the remote object has been
2529                          * removed, but the local MDT is not aware of that. */
2530                         goto lost_parent;
2531
2532                 if (unlikely(rc == -EEXIST)) {
2533                         /* Unfortunately, someone reused the name under the
2534                          * parent by race. So we have to remove the linkEA
2535                          * entry from current child object. It means that the
2536                          * LFSCK cannot recover the system totally back to
2537                          * its original status, but it is necessary to make
2538                          * the current system to be consistent.
2539                          *
2540                          * It also may be because of the LFSCK found some
2541                          * internal status of create operation. Under such
2542                          * case, nothing to be done. */
2543                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2544                                         parent, child, ldata, cname, &tfid);
2545                         if (rc >= 0) {
2546                                 snprintf(info->lti_tmpbuf,
2547                                          sizeof(info->lti_tmpbuf),
2548                                          "-"DFID, PFID(pfid));
2549                                 rc = lfsck_namespace_insert_orphan(env, com,
2550                                         child, info->lti_tmpbuf, "D", NULL);
2551                         }
2552                 }
2553
2554                 GOTO(out, rc);
2555         } /* rc == -ENOENT */
2556
2557         if (rc != 0)
2558                 GOTO(out, rc);
2559
2560         if (!lu_fid_eq(&tfid, cfid)) {
2561                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2562                         *retry = true;
2563
2564                         GOTO(out, rc = 0);
2565                 }
2566
2567                 lfsck_ibits_unlock(lh, LCK_EX);
2568                 /* The name entry references another MDT-object that
2569                  * may be created by the LFSCK for repairing dangling
2570                  * name entry. Try to replace it. */
2571                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2572                                                   &tfid, cname);
2573                 if (rc == 0)
2574                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2575                                                         pfid, lh, type);
2576
2577                 GOTO(out, rc);
2578         }
2579
2580         /* Zero FID may because the remote directroy object has invalid linkEA,
2581          * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2582          * how to repair the inconsistency, but the namespace LFSCK on the MDT
2583          * where its name entry resides may has more information (name, FID) to
2584          * repair such inconsistency. So here, keep the inconsistency to avoid
2585          * some imporper repairing. */
2586         if (fid_is_zero(pfid)) {
2587                 if (unknown)
2588                         *unknown = true;
2589
2590                 GOTO(out, rc = 0);
2591         }
2592
2593         /* The ".." name entry is wrong, update it. */
2594         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2595                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2596                         *retry = true;
2597
2598                         GOTO(out, rc = 0);
2599                 }
2600
2601                 *type = LNIT_UNMATCHED_PAIRS;
2602                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2603                                                 lfsck_dto2fid(parent), cname);
2604         }
2605
2606         GOTO(out, rc);
2607
2608 out:
2609         if (parent != NULL && !IS_ERR(parent))
2610                 lfsck_object_put(env, parent);
2611
2612         return rc;
2613 }
2614
2615 /**
2616  * Double Scan Directory object for multiple linkEA entries case.
2617  *
2618  * The given @child has multiple linkEA entries. There is at most one linkEA
2619  * entry will be valid, all the others will be removed. Firstly, the function
2620  * will try to find out the linkEA entry for which the name entry exists under
2621  * the given parent (@pfid). If there is no linkEA entry that matches the given
2622  * ".." name entry, then tries to find out the first linkEA entry that both the
2623  * parent and the name entry exist to rebuild a new ".." name entry.
2624  *
2625  * \param[in] env       pointer to the thread context
2626  * \param[in] com       pointer to the lfsck component
2627  * \param[in] child     pointer to the directory to be double scanned
2628  * \param[in] pfid      the FID corresponding to the ".." entry
2629  * \param[in] ldata     pointer to the linkEA data for the given @child
2630  * \param[in,out] lh    ldlm lock handler for the given @child
2631  * \param[out] type     to tell the caller what the inconsistency is
2632  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2633  * \param[in] unknown   set if does not know how to repair the inconsistency
2634  *
2635  * \retval              positive number for repaired cases
2636  * \retval              0 if nothing to be repaired
2637  * \retval              negative error number on failure
2638  */
2639 static int
2640 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2641                              struct lfsck_component *com,
2642                              struct dt_object *child,
2643                              const struct lu_fid *pfid,
2644                              struct linkea_data *ldata,
2645                              struct lustre_handle *lh,
2646                              enum lfsck_namespace_inconsistency_type *type,
2647                              bool lpf, bool *unknown)
2648 {
2649         struct lfsck_thread_info *info          = lfsck_env_info(env);
2650         struct lu_name           *cname         = &info->lti_name;
2651         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2652         struct lu_fid            *pfid2         = &info->lti_fid3;
2653         struct lu_fid             tfid;
2654         struct lfsck_namespace   *ns            = com->lc_file_ram;
2655         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2656         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2657         struct dt_object         *parent        = NULL;
2658         struct linkea_data        ldata_new     = { NULL };
2659         int                       dirent_count  = 0;
2660         int                       rc            = 0;
2661         bool                      once          = true;
2662         ENTRY;
2663
2664 again:
2665         while (ldata->ld_lee != NULL) {
2666                 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2667                                                          info->lti_key,
2668                                                          sizeof(info->lti_key));
2669                 /* Drop invalid linkEA entry. */
2670                 if (rc != 0) {
2671                         lfsck_linkea_del_buf(ldata, cname);
2672                         continue;
2673                 }
2674
2675                 /* Drop repeated linkEA entries. */
2676                 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2677
2678                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2679                  * then it is possible that: the directry object has ever
2680                  * been lost, but its name entry was there. In the former
2681                  * LFSCK run, during the first-stage scanning, the LFSCK
2682                  * found the dangling name entry, but it did not recreate
2683                  * the lost object, and when moved to the second-stage
2684                  * scanning, some children objects of the lost directory
2685                  * object were found, then the LFSCK recreated such lost
2686                  * directory object as an orphan.
2687                  *
2688                  * When the LFSCK runs again, if the dangling name is still
2689                  * there, the LFSCK should move the orphan directory object
2690                  * back to the normal namespace. */
2691                 if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
2692                         linkea_next_entry(ldata);
2693                         continue;
2694                 }
2695
2696                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2697                 if (IS_ERR(parent))
2698                         RETURN(PTR_ERR(parent));
2699
2700                 if (!dt_object_exists(parent)) {
2701                         lfsck_object_put(env, parent);
2702                         if (ldata->ld_leh->leh_reccount > 1) {
2703                                 /* If it is NOT the last linkEA entry, then
2704                                  * there is still other chance to make the
2705                                  * child to be visible via other parent, then
2706                                  * remove this linkEA entry. */
2707                                 lfsck_linkea_del_buf(ldata, cname);
2708                                 continue;
2709                         }
2710
2711                         break;
2712                 }
2713
2714                 /* The linkEA entry with bad parent will be removed. */
2715                 if (unlikely(!dt_try_as_dir(env, parent))) {
2716                         lfsck_object_put(env, parent);
2717                         lfsck_linkea_del_buf(ldata, cname);
2718                         continue;
2719                 }
2720
2721                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2722                                (const struct dt_key *)cname->ln_name);
2723                 *pfid2 = *lfsck_dto2fid(parent);
2724                 if (rc == -ENOENT) {
2725                         lfsck_object_put(env, parent);
2726                         linkea_next_entry(ldata);
2727                         continue;
2728                 }
2729
2730                 if (rc != 0) {
2731                         lfsck_object_put(env, parent);
2732
2733                         RETURN(rc);
2734                 }
2735
2736                 if (lu_fid_eq(&tfid, cfid)) {
2737                         lfsck_object_put(env, parent);
2738                         /* If the parent (that is declared via linkEA entry)
2739                          * directory contains the specified child, but such
2740                          * parent does not match the dotdot name entry, then
2741                          * trust the linkEA. */
2742                         if (!lu_fid_eq(pfid, pfid2)) {
2743                                 *type = LNIT_UNMATCHED_PAIRS;
2744                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2745                                                 com, child, pfid2, cname);
2746
2747                                 RETURN(rc);
2748                         }
2749
2750 rebuild:
2751                         /* It is the most common case that we find the
2752                          * name entry corresponding to the linkEA entry
2753                          * that matches the ".." name entry. */
2754                         rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2755                                               cname, pfid2);
2756                         if (rc != 0)
2757                                 RETURN(rc);
2758
2759                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2760                                                             &ldata_new);
2761                         if (rc < 0)
2762                                 RETURN(rc);
2763
2764                         lfsck_linkea_del_buf(ldata, cname);
2765                         linkea_first_entry(ldata);
2766                         /* There may be some invalid dangling name entries under
2767                          * other parent directories, remove all of them. */
2768                         while (ldata->ld_lee != NULL) {
2769                                 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2770                                                 cname, &tfid, info->lti_key,
2771                                                 sizeof(info->lti_key));
2772                                 if (rc != 0)
2773                                         goto next;
2774
2775                                 parent = lfsck_object_find_bottom(env, lfsck,
2776                                                                   &tfid);
2777                                 if (IS_ERR(parent)) {
2778                                         rc = PTR_ERR(parent);
2779                                         if (rc != -ENOENT &&
2780                                             bk->lb_param & LPF_FAILOUT)
2781                                                 RETURN(rc);
2782
2783                                         goto next;
2784                                 }
2785
2786                                 if (!dt_object_exists(parent)) {
2787                                         lfsck_object_put(env, parent);
2788                                         goto next;
2789                                 }
2790
2791                                 rc = lfsck_namespace_repair_dirent(env, com,
2792                                         parent, child, cname->ln_name,
2793                                         cname->ln_name, S_IFDIR, false, true);
2794                                 lfsck_object_put(env, parent);
2795                                 if (rc < 0) {
2796                                         if (bk->lb_param & LPF_FAILOUT)
2797                                                 RETURN(rc);
2798
2799                                         goto next;
2800                                 }
2801
2802                                 dirent_count += rc;
2803
2804 next:
2805                                 lfsck_linkea_del_buf(ldata, cname);
2806                         }
2807
2808                         ns->ln_dirent_repaired += dirent_count;
2809
2810                         RETURN(rc);
2811                 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
2812
2813                 lfsck_ibits_unlock(lh, LCK_EX);
2814                 /* The name entry references another MDT-object that may be
2815                  * created by the LFSCK for repairing dangling name entry.
2816                  * Try to replace it. */
2817                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2818                                                   &tfid, cname);
2819                 lfsck_object_put(env, parent);
2820                 if (rc < 0)
2821                         RETURN(rc);
2822
2823                 if (rc > 0)
2824                         goto rebuild;
2825
2826                 lfsck_linkea_del_buf(ldata, cname);
2827         } /* while (ldata->ld_lee != NULL) */
2828
2829         /* If there is still linkEA overflow, return. */
2830         if (unlikely(ldata->ld_leh->leh_overflow_time))
2831                 RETURN(0);
2832
2833         linkea_first_entry(ldata);
2834         if (ldata->ld_leh->leh_reccount == 1) {
2835                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2836                                                 lh, type, NULL, unknown);
2837
2838                 RETURN(rc);
2839         }
2840
2841         /* All linkEA entries are invalid and removed, then handle the @child
2842          * as an orphan.*/
2843         if (ldata->ld_leh->leh_reccount == 0) {
2844                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2845                                                 type);
2846
2847                 RETURN(rc);
2848         }
2849
2850         /* If the dangling name entry for the orphan directory object has
2851          * been remvoed, then just check whether the directory object is
2852          * still under the .lustre/lost+found/MDTxxxx/ or not. */
2853         if (lpf) {
2854                 lpf = false;
2855                 goto again;
2856         }
2857
2858         /* There is no linkEA entry that matches the ".." name entry. Find
2859          * the first linkEA entry that both parent and name entry exist to
2860          * rebuild a new ".." name entry. */
2861         if (once) {
2862                 once = false;
2863                 goto again;
2864         }
2865
2866         RETURN(rc);
2867 }
2868
2869 /**
2870  * Repair the object's nlink attribute.
2871  *
2872  * If all the known name entries have been verified, then the object's hard
2873  * link attribute should match the object's linkEA entries count unless the
2874  * object's has too many hard link to be recorded in the linkEA. Such cases
2875  * should have been marked in the LFSCK trace file. Otherwise, trust the
2876  * linkEA to update the object's nlink attribute.
2877  *
2878  * \param[in] env       pointer to the thread context
2879  * \param[in] com       pointer to the lfsck component
2880  * \param[in] obj       pointer to the dt_object to be handled
2881  * \param[in,out] la    pointer to buffer to object's attribute before
2882  *                      and after the repairing
2883  *
2884  * \retval              positive number for repaired cases
2885  * \retval              0 if nothing to be repaired
2886  * \retval              negative error number on failure
2887  */
2888 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2889                                         struct lfsck_component *com,
2890                                         struct dt_object *obj,
2891                                         struct lu_attr *la)
2892 {
2893         struct lfsck_namespace          *ns     = com->lc_file_ram;
2894         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2895         struct dt_device                *dev    = lfsck_obj2dev(obj);
2896         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
2897         struct thandle                  *th     = NULL;
2898         struct linkea_data               ldata  = { NULL };
2899         struct lustre_handle             lh     = { 0 };
2900         __u32                            old    = la->la_nlink;
2901         int                              rc     = 0;
2902         ENTRY;
2903
2904         LASSERT(!dt_object_remote(obj));
2905
2906         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2907                               MDS_INODELOCK_UPDATE, LCK_PW);
2908         if (rc != 0)
2909                 GOTO(log, rc);
2910
2911         th = dt_trans_create(env, dev);
2912         if (IS_ERR(th))
2913                 GOTO(log, rc = PTR_ERR(th));
2914
2915         la->la_valid = LA_NLINK;
2916         rc = dt_declare_attr_set(env, obj, la, th);
2917         if (rc != 0)
2918                 GOTO(stop, rc);
2919
2920         rc = dt_trans_start_local(env, dev, th);
2921         if (rc != 0)
2922                 GOTO(stop, rc);
2923
2924         dt_write_lock(env, obj, 0);
2925         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2926          * ever tried to verify some remote MDT-object that resides on this
2927          * MDT, but this MDT failed to respond such request. So means there
2928          * may be some remote name entry on other MDT that references this
2929          * object with another name, so we cannot know whether this linkEA
2930          * is valid or not. So keep it there and maybe resolved when next
2931          * LFSCK run. */
2932         if (ns->ln_flags & LF_INCOMPLETE)
2933                 GOTO(unlock, rc = 0);
2934
2935         rc = dt_attr_get(env, obj, la);
2936         if (rc != 0)
2937                 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2938
2939         rc = lfsck_links_read2_with_rec(env, obj, &ldata);
2940         if (rc)
2941                 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2942
2943         /* XXX: Currently, we only update the nlink attribute if the known
2944          *      linkEA entries is larger than the nlink attribute. That is
2945          *      safe action. */
2946         if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
2947             unlikely(la->la_nlink == 0 ||
2948                      ldata.ld_leh->leh_overflow_time))
2949                 GOTO(unlock, rc = 0);
2950
2951         la->la_nlink = ldata.ld_leh->leh_reccount;
2952         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2953                 GOTO(unlock, rc = 1);
2954
2955         rc = dt_attr_set(env, obj, la, th);
2956
2957         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2958
2959 unlock:
2960         dt_write_unlock(env, obj);
2961
2962 stop:
2963         dt_trans_stop(env, dev, th);
2964
2965 log:
2966         lfsck_ibits_unlock(&lh, LCK_PW);
2967         CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2968                "nlink count from %u to %u: rc = %d\n",
2969                lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2970
2971         if (rc != 0)
2972                 ns->ln_flags |= LF_INCONSISTENT;
2973
2974         return rc;
2975 }
2976
2977 /**
2978  * Double scan the directory object for namespace LFSCK.
2979  *
2980  * This function will verify the <parent, child> pairs in the namespace tree:
2981  * the parent references the child via some name entry that should be in the
2982  * child's linkEA entry, the child should back references the parent via its
2983  * ".." name entry.
2984  *
2985  * The LFSCK will scan every linkEA entry in turn until find out the first
2986  * matched pairs. If found, then all other linkEA entries will be dropped.
2987  * If all the linkEA entries cannot match the ".." name entry, then there
2988  * are serveral possible cases:
2989  *
2990  * 1) If there is only one linkEA entry, then trust it as long as the PFID
2991  *    in the linkEA entry is valid.
2992  *
2993  * 2) If there are multiple linkEA entries, then try to find the linkEA
2994  *    that matches the ".." name entry. If found, then all other entries
2995  *    are invalid; otherwise, it is quite possible that the ".." name entry
2996  *    is corrupted. Under such case, the LFSCK will rebuild the ".." name
2997  *    entry according to the first valid linkEA entry (both the parent and
2998  *    the name entry should exist).
2999  *
3000  * 3) If the directory object has no (valid) linkEA entry, then the
3001  *    directory object will be handled as pure orphan and inserted
3002  *    in the .lustre/lost+found/MDTxxxx/ with the name:
3003  *    ${self_FID}-${PFID}-D-${conflict_version}
3004  *
3005  * \param[in] env       pointer to the thread context
3006  * \param[in] com       pointer to the lfsck component
3007  * \param[in] child     pointer to the directory object to be handled
3008  * \param[in] flags     to indicate the specical checking on the @child
3009  *
3010  * \retval              positive number for repaired cases
3011  * \retval              0 if nothing to be repaired
3012  * \retval              negative error number on failure
3013  */
3014 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3015                                            struct lfsck_component *com,
3016                                            struct dt_object *child, __u8 flags)
3017 {
3018         struct lfsck_thread_info *info          = lfsck_env_info(env);
3019         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
3020         struct lu_fid            *pfid          = &info->lti_fid2;
3021         struct lfsck_namespace   *ns            = com->lc_file_ram;
3022         struct lfsck_instance    *lfsck         = com->lc_lfsck;
3023         struct lustre_handle      lh            = { 0 };
3024         struct linkea_data        ldata         = { NULL };
3025         bool                      unknown       = false;
3026         bool                      lpf           = false;
3027         bool                      retry         = false;
3028         enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3029         int                       rc            = 0;
3030         ENTRY;
3031
3032         LASSERT(!dt_object_remote(child));
3033
3034         if (flags & LNTF_UNCERTAIN_LMV) {