Whamcloud - gitweb
83d62dfe711981906917f6b50d808ec9dec15a33
[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) 2012, 2013, 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   0xA0629D03
45
46 enum lfsck_nameentry_check {
47         LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
48         LFSCK_NAMEENTRY_REMOVED         = 2, /* The entry has been removed. */
49         LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
50 };
51
52 static const char lfsck_namespace_name[] = "lfsck_namespace";
53
54 static struct lfsck_namespace_req *
55 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
56                                    struct lu_dirent *ent, __u16 type)
57 {
58         struct lfsck_namespace_req *lnr;
59         int                         size;
60
61         size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
62         OBD_ALLOC(lnr, size);
63         if (lnr == NULL)
64                 return ERR_PTR(-ENOMEM);
65
66         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
67         lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
68         lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
69         lnr->lnr_fid = ent->lde_fid;
70         lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
71         lnr->lnr_dir_cookie = ent->lde_hash;
72         lnr->lnr_attr = ent->lde_attrs;
73         lnr->lnr_size = size;
74         lnr->lnr_type = type;
75         lnr->lnr_namelen = ent->lde_namelen;
76         memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
77
78         return lnr;
79 }
80
81 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
82                                                struct lfsck_assistant_req *lar)
83 {
84         struct lfsck_namespace_req *lnr =
85                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
86
87         if (lnr->lnr_lmv != NULL)
88                 lfsck_lmv_put(env, lnr->lnr_lmv);
89
90         lu_object_put(env, &lnr->lnr_obj->do_lu);
91         OBD_FREE(lnr, lnr->lnr_size);
92 }
93
94 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
95                                       struct lfsck_namespace *src)
96 {
97         dst->ln_magic = le32_to_cpu(src->ln_magic);
98         dst->ln_status = le32_to_cpu(src->ln_status);
99         dst->ln_flags = le32_to_cpu(src->ln_flags);
100         dst->ln_success_count = le32_to_cpu(src->ln_success_count);
101         dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
102         dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
103         dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
104         dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
105         dst->ln_time_last_checkpoint =
106                                 le64_to_cpu(src->ln_time_last_checkpoint);
107         lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
108                                  &src->ln_pos_latest_start);
109         lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
110                                  &src->ln_pos_last_checkpoint);
111         lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
112                                  &src->ln_pos_first_inconsistent);
113         dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
114         dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
115         dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
116         dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
117         dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
118         dst->ln_objs_repaired_phase2 =
119                                 le64_to_cpu(src->ln_objs_repaired_phase2);
120         dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
121         dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
122         fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
123                       &src->ln_fid_latest_scanned_phase2);
124         dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
125         dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
126         dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
127         dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
128         dst->ln_unknown_inconsistency =
129                                 le64_to_cpu(src->ln_unknown_inconsistency);
130         dst->ln_unmatched_pairs_repaired =
131                                 le64_to_cpu(src->ln_unmatched_pairs_repaired);
132         dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
133         dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
134         dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
135         dst->ln_lost_dirent_repaired =
136                                 le64_to_cpu(src->ln_lost_dirent_repaired);
137         dst->ln_striped_dirs_scanned =
138                                 le64_to_cpu(src->ln_striped_dirs_scanned);
139         dst->ln_striped_dirs_repaired =
140                                 le64_to_cpu(src->ln_striped_dirs_repaired);
141         dst->ln_striped_dirs_failed =
142                                 le64_to_cpu(src->ln_striped_dirs_failed);
143         dst->ln_striped_dirs_disabled =
144                                 le64_to_cpu(src->ln_striped_dirs_disabled);
145         dst->ln_striped_dirs_skipped =
146                                 le64_to_cpu(src->ln_striped_dirs_skipped);
147         dst->ln_striped_shards_scanned =
148                                 le64_to_cpu(src->ln_striped_shards_scanned);
149         dst->ln_striped_shards_repaired =
150                                 le64_to_cpu(src->ln_striped_shards_repaired);
151         dst->ln_striped_shards_failed =
152                                 le64_to_cpu(src->ln_striped_shards_failed);
153         dst->ln_striped_shards_skipped =
154                                 le64_to_cpu(src->ln_striped_shards_skipped);
155         dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
156         dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
157         dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
158         dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
159         dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
160         dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
161 }
162
163 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
164                                       struct lfsck_namespace *src)
165 {
166         dst->ln_magic = cpu_to_le32(src->ln_magic);
167         dst->ln_status = cpu_to_le32(src->ln_status);
168         dst->ln_flags = cpu_to_le32(src->ln_flags);
169         dst->ln_success_count = cpu_to_le32(src->ln_success_count);
170         dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
171         dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
172         dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
173         dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
174         dst->ln_time_last_checkpoint =
175                                 cpu_to_le64(src->ln_time_last_checkpoint);
176         lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
177                                  &src->ln_pos_latest_start);
178         lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
179                                  &src->ln_pos_last_checkpoint);
180         lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
181                                  &src->ln_pos_first_inconsistent);
182         dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
183         dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
184         dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
185         dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
186         dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
187         dst->ln_objs_repaired_phase2 =
188                                 cpu_to_le64(src->ln_objs_repaired_phase2);
189         dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
190         dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
191         fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
192                       &src->ln_fid_latest_scanned_phase2);
193         dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
194         dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
195         dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
196         dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
197         dst->ln_unknown_inconsistency =
198                                 cpu_to_le64(src->ln_unknown_inconsistency);
199         dst->ln_unmatched_pairs_repaired =
200                                 cpu_to_le64(src->ln_unmatched_pairs_repaired);
201         dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
202         dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
203         dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
204         dst->ln_lost_dirent_repaired =
205                                 cpu_to_le64(src->ln_lost_dirent_repaired);
206         dst->ln_striped_dirs_scanned =
207                                 cpu_to_le64(src->ln_striped_dirs_scanned);
208         dst->ln_striped_dirs_repaired =
209                                 cpu_to_le64(src->ln_striped_dirs_repaired);
210         dst->ln_striped_dirs_failed =
211                                 cpu_to_le64(src->ln_striped_dirs_failed);
212         dst->ln_striped_dirs_disabled =
213                                 cpu_to_le64(src->ln_striped_dirs_disabled);
214         dst->ln_striped_dirs_skipped =
215                                 cpu_to_le64(src->ln_striped_dirs_skipped);
216         dst->ln_striped_shards_scanned =
217                                 cpu_to_le64(src->ln_striped_shards_scanned);
218         dst->ln_striped_shards_repaired =
219                                 cpu_to_le64(src->ln_striped_shards_repaired);
220         dst->ln_striped_shards_failed =
221                                 cpu_to_le64(src->ln_striped_shards_failed);
222         dst->ln_striped_shards_skipped =
223                                 cpu_to_le64(src->ln_striped_shards_skipped);
224         dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
225         dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
226         dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
227         dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
228         dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
229         dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
230 }
231
232 static void lfsck_namespace_record_failure(const struct lu_env *env,
233                                            struct lfsck_instance *lfsck,
234                                            struct lfsck_namespace *ns)
235 {
236         struct lfsck_position pos;
237
238         ns->ln_items_failed++;
239         lfsck_pos_fill(env, lfsck, &pos, false);
240         if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
241             lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
242                 ns->ln_pos_first_inconsistent = pos;
243
244                 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
245                        "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n",
246                        lfsck_lfsck2name(lfsck),
247                        ns->ln_pos_first_inconsistent.lp_oit_cookie,
248                        PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
249                        ns->ln_pos_first_inconsistent.lp_dir_cookie);
250         }
251 }
252
253 /**
254  * Load the MDT bitmap from the lfsck_namespace trace file.
255  *
256  * \param[in] env       pointer to the thread context
257  * \param[in] com       pointer to the lfsck component
258  *
259  * \retval              0 for success
260  * \retval              negative error number on failure or data corruption
261  */
262 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
263                                        struct lfsck_component *com)
264 {
265         struct dt_object                *obj    = com->lc_obj;
266         struct lfsck_assistant_data     *lad    = com->lc_data;
267         struct lfsck_namespace          *ns     = com->lc_file_ram;
268         cfs_bitmap_t                    *bitmap = lad->lad_bitmap;
269         ssize_t                          size;
270         __u32                            nbits;
271         int                              rc;
272         ENTRY;
273
274         if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
275             ns->ln_bitmap_size)
276                 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
277         else
278                 nbits = ns->ln_bitmap_size;
279
280         if (unlikely(nbits < BITS_PER_LONG))
281                 nbits = BITS_PER_LONG;
282
283         if (nbits > bitmap->size) {
284                 __u32 new_bits = bitmap->size;
285                 cfs_bitmap_t *new_bitmap;
286
287                 while (new_bits < nbits)
288                         new_bits <<= 1;
289
290                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
291                 if (new_bitmap == NULL)
292                         RETURN(-ENOMEM);
293
294                 lad->lad_bitmap = new_bitmap;
295                 CFS_FREE_BITMAP(bitmap);
296                 bitmap = new_bitmap;
297         }
298
299         if (ns->ln_bitmap_size == 0) {
300                 lad->lad_incomplete = 0;
301                 CFS_RESET_BITMAP(bitmap);
302
303                 RETURN(0);
304         }
305
306         size = (ns->ln_bitmap_size + 7) >> 3;
307         rc = dt_xattr_get(env, obj,
308                           lfsck_buf_get(env, bitmap->data, size),
309                           XATTR_NAME_LFSCK_BITMAP, BYPASS_CAPA);
310         if (rc != size)
311                 RETURN(rc >= 0 ? -EINVAL : rc);
312
313         if (cfs_bitmap_check_empty(bitmap))
314                 lad->lad_incomplete = 0;
315         else
316                 lad->lad_incomplete = 1;
317
318         RETURN(0);
319 }
320
321 /**
322  * \retval +ve: the lfsck_namespace is broken, the caller should reset it.
323  * \retval 0: succeed.
324  * \retval -ve: failed cases.
325  */
326 static int lfsck_namespace_load(const struct lu_env *env,
327                                 struct lfsck_component *com)
328 {
329         int len = com->lc_file_size;
330         int rc;
331
332         rc = dt_xattr_get(env, com->lc_obj,
333                           lfsck_buf_get(env, com->lc_file_disk, len),
334                           XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA);
335         if (rc == len) {
336                 struct lfsck_namespace *ns = com->lc_file_ram;
337
338                 lfsck_namespace_le_to_cpu(ns,
339                                 (struct lfsck_namespace *)com->lc_file_disk);
340                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
341                         CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
342                                "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
343                                ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
344                         rc = 1;
345                 } else {
346                         rc = 0;
347                 }
348         } else if (rc != -ENODATA) {
349                 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
350                        "expected = %d: rc = %d\n",
351                        lfsck_lfsck2name(com->lc_lfsck), len, rc);
352                 if (rc >= 0)
353                         rc = 1;
354         }
355         return rc;
356 }
357
358 static int lfsck_namespace_store(const struct lu_env *env,
359                                  struct lfsck_component *com)
360 {
361         struct dt_object                *obj    = com->lc_obj;
362         struct lfsck_instance           *lfsck  = com->lc_lfsck;
363         struct lfsck_namespace          *ns     = com->lc_file_ram;
364         struct lfsck_assistant_data     *lad    = com->lc_data;
365         cfs_bitmap_t                    *bitmap = NULL;
366         struct thandle                  *handle;
367         __u32                            nbits  = 0;
368         int                              len    = com->lc_file_size;
369         int                              rc;
370         ENTRY;
371
372         if (lad != NULL) {
373                 bitmap = lad->lad_bitmap;
374                 nbits = bitmap->size;
375
376                 LASSERT(nbits > 0);
377                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
378         }
379
380         ns->ln_bitmap_size = nbits;
381         lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
382                                   ns);
383         handle = dt_trans_create(env, lfsck->li_bottom);
384         if (IS_ERR(handle))
385                 GOTO(log, rc = PTR_ERR(handle));
386
387         rc = dt_declare_xattr_set(env, obj,
388                                   lfsck_buf_get(env, com->lc_file_disk, len),
389                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
390         if (rc != 0)
391                 GOTO(out, rc);
392
393         if (bitmap != NULL) {
394                 rc = dt_declare_xattr_set(env, obj,
395                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
396                                 XATTR_NAME_LFSCK_BITMAP, 0, handle);
397                 if (rc != 0)
398                         GOTO(out, rc);
399         }
400
401         rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
402         if (rc != 0)
403                 GOTO(out, rc);
404
405         rc = dt_xattr_set(env, obj,
406                           lfsck_buf_get(env, com->lc_file_disk, len),
407                           XATTR_NAME_LFSCK_NAMESPACE, 0, handle, BYPASS_CAPA);
408         if (rc == 0 && bitmap != NULL)
409                 rc = dt_xattr_set(env, obj,
410                                   lfsck_buf_get(env, bitmap->data, nbits >> 3),
411                                   XATTR_NAME_LFSCK_BITMAP, 0, handle,
412                                   BYPASS_CAPA);
413
414         GOTO(out, rc);
415
416 out:
417         dt_trans_stop(env, lfsck->li_bottom, handle);
418
419 log:
420         if (rc != 0)
421                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
422                        lfsck_lfsck2name(lfsck), rc);
423         return rc;
424 }
425
426 static int lfsck_namespace_init(const struct lu_env *env,
427                                 struct lfsck_component *com)
428 {
429         struct lfsck_namespace *ns = com->lc_file_ram;
430         int rc;
431
432         memset(ns, 0, sizeof(*ns));
433         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
434         ns->ln_status = LS_INIT;
435         down_write(&com->lc_sem);
436         rc = lfsck_namespace_store(env, com);
437         up_write(&com->lc_sem);
438         return rc;
439 }
440
441 /**
442  * Update the namespace LFSCK trace file for the given @fid
443  *
444  * \param[in] env       pointer to the thread context
445  * \param[in] com       pointer to the lfsck component
446  * \param[in] fid       the fid which flags to be updated in the lfsck
447  *                      trace file
448  * \param[in] add       true if add new flags, otherwise remove flags
449  *
450  * \retval              0 for succeed or nothing to be done
451  * \retval              negative error number on failure
452  */
453 int lfsck_namespace_trace_update(const struct lu_env *env,
454                                  struct lfsck_component *com,
455                                  const struct lu_fid *fid,
456                                  const __u8 flags, bool add)
457 {
458         struct lfsck_instance   *lfsck  = com->lc_lfsck;
459         struct dt_object        *obj    = com->lc_obj;
460         struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
461         struct dt_device        *dev    = lfsck->li_bottom;
462         struct thandle          *th     = NULL;
463         int                      rc     = 0;
464         __u8                     old    = 0;
465         __u8                     new    = 0;
466         ENTRY;
467
468         LASSERT(flags != 0);
469
470         down_write(&com->lc_sem);
471         fid_cpu_to_be(key, fid);
472         rc = dt_lookup(env, obj, (struct dt_rec *)&old,
473                        (const struct dt_key *)key, BYPASS_CAPA);
474         if (rc == -ENOENT) {
475                 if (!add)
476                         GOTO(unlock, rc = 0);
477
478                 old = 0;
479                 new = flags;
480         } else if (rc == 0) {
481                 if (add) {
482                         if ((old & flags) == flags)
483                                 GOTO(unlock, rc = 0);
484
485                         new = old | flags;
486                 } else {
487                         if ((old & flags) == 0)
488                                 GOTO(unlock, rc = 0);
489
490                         new = old & ~flags;
491                 }
492         } else {
493                 GOTO(log, rc);
494         }
495
496         th = dt_trans_create(env, dev);
497         if (IS_ERR(th))
498                 GOTO(log, rc = PTR_ERR(th));
499
500         if (old != 0) {
501                 rc = dt_declare_delete(env, obj,
502                                        (const struct dt_key *)key, th);
503                 if (rc != 0)
504                         GOTO(log, rc);
505         }
506
507         if (new != 0) {
508                 rc = dt_declare_insert(env, obj,
509                                        (const struct dt_rec *)&new,
510                                        (const struct dt_key *)key, th);
511                 if (rc != 0)
512                         GOTO(log, rc);
513         }
514
515         rc = dt_trans_start_local(env, dev, th);
516         if (rc != 0)
517                 GOTO(log, rc);
518
519         if (old != 0) {
520                 rc = dt_delete(env, obj, (const struct dt_key *)key,
521                                th, BYPASS_CAPA);
522                 if (rc != 0)
523                         GOTO(log, rc);
524         }
525
526         if (new != 0) {
527                 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
528                                (const struct dt_key *)key, th, BYPASS_CAPA, 1);
529                 if (rc != 0)
530                         GOTO(log, rc);
531         }
532
533         GOTO(log, rc);
534
535 log:
536         if (th != NULL && !IS_ERR(th))
537                 dt_trans_stop(env, dev, th);
538
539         CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
540                "trace file, flags %x, old %x, new %x: rc = %d\n",
541                lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
542                (__u32)flags, (__u32)old, (__u32)new, rc);
543
544 unlock:
545         up_write(&com->lc_sem);
546
547         return rc;
548 }
549
550 int lfsck_namespace_check_exist(const struct lu_env *env,
551                                 struct dt_object *dir,
552                                 struct dt_object *obj, const char *name)
553 {
554         struct lu_fid    *fid = &lfsck_env_info(env)->lti_fid;
555         int               rc;
556         ENTRY;
557
558         if (unlikely(lfsck_is_dead_obj(obj)))
559                 RETURN(LFSCK_NAMEENTRY_DEAD);
560
561         rc = dt_lookup(env, dir, (struct dt_rec *)fid,
562                        (const struct dt_key *)name, BYPASS_CAPA);
563         if (rc == -ENOENT)
564                 RETURN(LFSCK_NAMEENTRY_REMOVED);
565
566         if (rc < 0)
567                 RETURN(rc);
568
569         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
570                 RETURN(LFSCK_NAMEENTRY_RECREATED);
571
572         RETURN(0);
573 }
574
575 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
576                                             struct dt_object *obj,
577                                             struct thandle *handle)
578 {
579         int rc;
580
581         /* For destroying all invalid linkEA entries. */
582         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
583         if (rc != 0)
584                 return rc;
585
586         /* For insert new linkEA entry. */
587         rc = dt_declare_xattr_set(env, obj,
588                         lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
589                         XATTR_NAME_LINK, 0, handle);
590         return rc;
591 }
592
593 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
594                        struct linkea_data *ldata)
595 {
596         int rc;
597
598         if (ldata->ld_buf->lb_buf == NULL)
599                 return -ENOMEM;
600
601         if (!dt_object_exists(obj))
602                 return -ENOENT;
603
604         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, BYPASS_CAPA);
605         if (rc == -ERANGE) {
606                 /* Buf was too small, figure out what we need. */
607                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK,
608                                   BYPASS_CAPA);
609                 if (rc <= 0)
610                         return rc;
611
612                 lu_buf_realloc(ldata->ld_buf, rc);
613                 if (ldata->ld_buf->lb_buf == NULL)
614                         return -ENOMEM;
615
616                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK,
617                                   BYPASS_CAPA);
618         }
619
620         if (rc > 0)
621                 rc = linkea_init(ldata);
622
623         return rc;
624 }
625
626 /**
627  * Remove linkEA for the given object.
628  *
629  * The caller should take the ldlm lock before the calling.
630  *
631  * \param[in] env       pointer to the thread context
632  * \param[in] com       pointer to the lfsck component
633  * \param[in] obj       pointer to the dt_object to be handled
634  *
635  * \retval              0 for repaired cases
636  * \retval              negative error number on failure
637  */
638 static int lfsck_namespace_links_remove(const struct lu_env *env,
639                                         struct lfsck_component *com,
640                                         struct dt_object *obj)
641 {
642         struct lfsck_instance           *lfsck  = com->lc_lfsck;
643         struct dt_device                *dev    = lfsck->li_bottom;
644         struct thandle                  *th     = NULL;
645         int                              rc     = 0;
646         ENTRY;
647
648         LASSERT(dt_object_remote(obj) == 0);
649
650         th = dt_trans_create(env, dev);
651         if (IS_ERR(th))
652                 GOTO(log, rc = PTR_ERR(th));
653
654         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
655         if (rc != 0)
656                 GOTO(stop, rc);
657
658         rc = dt_trans_start_local(env, dev, th);
659         if (rc != 0)
660                 GOTO(stop, rc);
661
662         dt_write_lock(env, obj, 0);
663         if (unlikely(lfsck_is_dead_obj(obj)))
664                 GOTO(unlock, rc = -ENOENT);
665
666         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
667                 GOTO(unlock, rc = 0);
668
669         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th, BYPASS_CAPA);
670
671         GOTO(unlock, rc);
672
673 unlock:
674         dt_write_unlock(env, obj);
675
676 stop:
677         dt_trans_stop(env, dev, th);
678
679 log:
680         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
681                "for the object "DFID": rc = %d\n",
682                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
683
684         if (rc == 0) {
685                 struct lfsck_namespace *ns = com->lc_file_ram;
686
687                 ns->ln_flags |= LF_INCONSISTENT;
688         }
689
690         return rc;
691 }
692
693 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
694                              struct linkea_data *ldata, struct thandle *handle)
695 {
696         const struct lu_buf *buf = lfsck_buf_get_const(env,
697                                                        ldata->ld_buf->lb_buf,
698                                                        ldata->ld_leh->leh_len);
699
700         return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle,
701                             BYPASS_CAPA);
702 }
703
704 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
705                                                 struct lu_name *cname,
706                                                 struct lu_fid *pfid,
707                                                 char *buf)
708 {
709         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
710         /* To guarantee the 'name' is terminated with '0'. */
711         memcpy(buf, cname->ln_name, cname->ln_namelen);
712         buf[cname->ln_namelen] = 0;
713         cname->ln_name = buf;
714 }
715
716 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
717                                                struct lu_name *cname,
718                                                struct lu_fid *pfid,
719                                                bool remove)
720 {
721         struct link_ea_entry    *oldlee;
722         int                      oldlen;
723         int                      repeated = 0;
724
725         oldlee = ldata->ld_lee;
726         oldlen = ldata->ld_reclen;
727         linkea_next_entry(ldata);
728         while (ldata->ld_lee != NULL) {
729                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
730                                    ldata->ld_lee->lee_reclen[1];
731                 if (unlikely(ldata->ld_reclen == oldlen &&
732                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
733                         repeated++;
734                         if (!remove)
735                                 break;
736
737                         linkea_del_buf(ldata, cname);
738                 } else {
739                         linkea_next_entry(ldata);
740                 }
741         }
742         ldata->ld_lee = oldlee;
743         ldata->ld_reclen = oldlen;
744
745         return repeated;
746 }
747
748 /**
749  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
750  *
751  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
752  * with the given type to generate the name, the detailed rules for name
753  * have been described as following.
754  *
755  * The function also generates the linkEA corresponding to the name entry
756  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
757  *
758  * \param[in] env       pointer to the thread context
759  * \param[in] com       pointer to the lfsck component
760  * \param[in] orphan    pointer to the orphan MDT-object
761  * \param[in] infix     additional information for the orphan name, such as
762  *                      the FID for original
763  * \param[in] type      the type for describing why the orphan MDT-object is
764  *                      created. The rules are as following:
765  *
766  *  type "D":           The MDT-object is a directory, it may knows its parent
767  *                      but because there is no valid linkEA, the LFSCK cannot
768  *                      know where to put it back to the namespace.
769  *  type "O":           The MDT-object has no linkEA, and there is no name
770  *                      entry that references the MDT-object.
771  *
772  *  type "S":           The orphan MDT-object is a shard of a striped directory
773  *
774  * \see lfsck_layout_recreate_parent() for more types.
775  *
776  * The orphan name will be like:
777  * ${FID}-${infix}-${type}-${conflict_version}
778  *
779  * \param[out] count    if some others inserted some linkEA entries by race,
780  *                      then return the linkEA entries count.
781  *
782  * \retval              positive number for repaired cases
783  * \retval              0 if needs to repair nothing
784  * \retval              negative error number on failure
785  */
786 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
787                                          struct lfsck_component *com,
788                                          struct dt_object *orphan,
789                                          const char *infix, const char *type,
790                                          int *count)
791 {
792         struct lfsck_thread_info        *info   = lfsck_env_info(env);
793         struct lu_name                  *cname  = &info->lti_name;
794         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
795         struct lu_fid                   *tfid   = &info->lti_fid5;
796         struct lu_attr                  *la     = &info->lti_la3;
797         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
798         const struct lu_fid             *pfid;
799         struct lfsck_instance           *lfsck  = com->lc_lfsck;
800         struct dt_device                *dev    = lfsck->li_bottom;
801         struct dt_object                *parent;
802         struct thandle                  *th     = NULL;
803         struct lustre_handle             plh    = { 0 };
804         struct lustre_handle             clh    = { 0 };
805         struct linkea_data               ldata  = { 0 };
806         struct lu_buf                    linkea_buf;
807         int                              namelen;
808         int                              idx    = 0;
809         int                              rc     = 0;
810         bool                             exist  = false;
811         ENTRY;
812
813         cname->ln_name = NULL;
814         /* Create .lustre/lost+found/MDTxxxx when needed. */
815         if (unlikely(lfsck->li_lpf_obj == NULL)) {
816                 rc = lfsck_create_lpf(env, lfsck);
817                 if (rc != 0)
818                         GOTO(log, rc);
819         }
820
821         parent = lfsck->li_lpf_obj;
822         pfid = lfsck_dto2fid(parent);
823
824         /* Hold update lock on the parent to prevent others to access. */
825         rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
826                               MDS_INODELOCK_UPDATE, LCK_EX);
827         if (rc != 0)
828                 GOTO(log, rc);
829
830         do {
831                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
832                                    PFID(cfid), infix, type, idx++);
833                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
834                                (const struct dt_key *)info->lti_key,
835                                BYPASS_CAPA);
836                 if (rc != 0 && rc != -ENOENT)
837                         GOTO(log, rc);
838
839                 if (unlikely(rc == 0 && lu_fid_eq(cfid, tfid)))
840                         exist = true;
841         } while (rc == 0 && !exist);
842
843         cname->ln_name = info->lti_key;
844         cname->ln_namelen = namelen;
845         rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
846         if (rc != 0)
847                 GOTO(log, rc);
848
849         rc = linkea_add_buf(&ldata, cname, pfid);
850         if (rc != 0)
851                 GOTO(log, rc);
852
853         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
854                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP,
855                               LCK_EX);
856         if (rc != 0)
857                 GOTO(log, rc);
858
859         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
860                        ldata.ld_leh->leh_len);
861         th = dt_trans_create(env, dev);
862         if (IS_ERR(th))
863                 GOTO(log, rc = PTR_ERR(th));
864
865         if (S_ISDIR(lfsck_object_type(orphan))) {
866                 rc = dt_declare_delete(env, orphan,
867                                        (const struct dt_key *)dotdot, th);
868                 if (rc != 0)
869                         GOTO(stop, rc);
870
871                 rec->rec_type = S_IFDIR;
872                 rec->rec_fid = pfid;
873                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
874                                        (const struct dt_key *)dotdot, th);
875                 if (rc != 0)
876                         GOTO(stop, rc);
877         }
878
879         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
880                                   XATTR_NAME_LINK, 0, th);
881         if (rc != 0)
882                 GOTO(stop, rc);
883
884         if (!exist) {
885                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
886                 rec->rec_fid = cfid;
887                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
888                                        (const struct dt_key *)cname->ln_name,
889                                        th);
890                 if (rc != 0)
891                         GOTO(stop, rc);
892
893                 if (S_ISDIR(rec->rec_type)) {
894                         rc = dt_declare_ref_add(env, parent, th);
895                         if (rc != 0)
896                                 GOTO(stop, rc);
897                 }
898         }
899
900         memset(la, 0, sizeof(*la));
901         la->la_ctime = cfs_time_current_sec();
902         la->la_valid = LA_CTIME;
903         rc = dt_declare_attr_set(env, orphan, la, th);
904         if (rc != 0)
905                 GOTO(stop, rc);
906
907         rc = dt_trans_start_local(env, dev, th);
908         if (rc != 0)
909                 GOTO(stop, rc);
910
911         dt_write_lock(env, orphan, 0);
912         rc = lfsck_links_read(env, orphan, &ldata);
913         if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
914                    (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
915                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
916                         GOTO(unlock, rc = 1);
917
918                 if (S_ISDIR(lfsck_object_type(orphan))) {
919                         rc = dt_delete(env, orphan,
920                                        (const struct dt_key *)dotdot, th,
921                                        BYPASS_CAPA);
922                         if (rc != 0)
923                                 GOTO(unlock, rc);
924
925                         rec->rec_type = S_IFDIR;
926                         rec->rec_fid = pfid;
927                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
928                                        (const struct dt_key *)dotdot, th,
929                                        BYPASS_CAPA, 1);
930                         if (rc != 0)
931                                 GOTO(unlock, rc);
932                 }
933
934                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
935                                   th, BYPASS_CAPA);
936         } else {
937                 if (rc == 0 && count != NULL)
938                         *count = ldata.ld_leh->leh_reccount;
939
940                 GOTO(unlock, rc);
941         }
942         dt_write_unlock(env, orphan);
943
944         if (rc == 0 && !exist) {
945                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
946                 rec->rec_fid = cfid;
947                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
948                                (const struct dt_key *)cname->ln_name,
949                                th, BYPASS_CAPA, 1);
950                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
951                         dt_write_lock(env, parent, 0);
952                         rc = dt_ref_add(env, parent, th);
953                         dt_write_unlock(env, parent);
954                 }
955         }
956
957         if (rc != 0)
958                 GOTO(unlock, rc);
959
960         rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
961
962         GOTO(stop, rc = (rc == 0 ? 1 : rc));
963
964 unlock:
965         dt_write_unlock(env, orphan);
966
967 stop:
968         dt_trans_stop(env, dev, th);
969
970 log:
971         lfsck_ibits_unlock(&clh, LCK_EX);
972         lfsck_ibits_unlock(&plh, LCK_EX);
973         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
974                "object "DFID", name = %s: rc = %d\n",
975                lfsck_lfsck2name(lfsck), PFID(cfid),
976                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
977
978         if (rc != 0) {
979                 struct lfsck_namespace *ns = com->lc_file_ram;
980
981                 ns->ln_flags |= LF_INCONSISTENT;
982         }
983
984         return rc;
985 }
986
987 /**
988  * Add the specified name entry back to namespace.
989  *
990  * If there is a linkEA entry that back references a name entry under
991  * some parent directory, but such parent directory does not have the
992  * claimed name entry. On the other hand, the linkEA entries count is
993  * not larger than the MDT-object's hard link count. Under such case,
994  * it is quite possible that the name entry is lost. Then the LFSCK
995  * should add the name entry back to the namespace.
996  *
997  * \param[in] env       pointer to the thread context
998  * \param[in] com       pointer to the lfsck component
999  * \param[in] parent    pointer to the directory under which the name entry
1000  *                      will be inserted into
1001  * \param[in] child     pointer to the object referenced by the name entry
1002  *                      that to be inserted into the parent
1003  * \param[in] name      the name for the child in the parent directory
1004  *
1005  * \retval              positive number for repaired cases
1006  * \retval              0 if nothing to be repaired
1007  * \retval              negative error number on failure
1008  */
1009 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1010                                          struct lfsck_component *com,
1011                                          struct dt_object *parent,
1012                                          struct dt_object *child,
1013                                          const char *name)
1014 {
1015         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1016         struct lu_attr                  *la     = &info->lti_la;
1017         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1018         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1019         struct dt_device                *dev    = lfsck->li_next;
1020         struct thandle                  *th     = NULL;
1021         struct lustre_handle             lh     = { 0 };
1022         int                              rc     = 0;
1023         ENTRY;
1024
1025         if (unlikely(!dt_try_as_dir(env, parent)))
1026                 GOTO(log, rc = -ENOTDIR);
1027
1028         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1029                 GOTO(log, rc = 1);
1030
1031         /* Hold update lock on the parent to prevent others to access. */
1032         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1033                               MDS_INODELOCK_UPDATE, LCK_EX);
1034         if (rc != 0)
1035                 GOTO(log, rc);
1036
1037         th = dt_trans_create(env, dev);
1038         if (IS_ERR(th))
1039                 GOTO(unlock, rc = PTR_ERR(th));
1040
1041         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1042         rec->rec_fid = lfsck_dto2fid(child);
1043         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1044                                (const struct dt_key *)name, th);
1045         if (rc != 0)
1046                 GOTO(stop, rc);
1047
1048         if (S_ISDIR(rec->rec_type)) {
1049                 rc = dt_declare_ref_add(env, parent, th);
1050                 if (rc != 0)
1051                         GOTO(stop, rc);
1052         }
1053
1054         memset(la, 0, sizeof(*la));
1055         la->la_ctime = cfs_time_current_sec();
1056         la->la_valid = LA_CTIME;
1057         rc = dt_declare_attr_set(env, parent, la, th);
1058         if (rc != 0)
1059                 GOTO(stop, rc);
1060
1061         rc = dt_declare_attr_set(env, child, la, th);
1062         if (rc != 0)
1063                 GOTO(stop, rc);
1064
1065         rc = dt_trans_start_local(env, dev, th);
1066         if (rc != 0)
1067                 GOTO(stop, rc);
1068
1069         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1070                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1071         if (rc != 0)
1072                 GOTO(stop, rc);
1073
1074         if (S_ISDIR(rec->rec_type)) {
1075                 dt_write_lock(env, parent, 0);
1076                 rc = dt_ref_add(env, parent, th);
1077                 dt_write_unlock(env, parent);
1078                 if (rc != 0)
1079                         GOTO(stop, rc);
1080         }
1081
1082         la->la_ctime = cfs_time_current_sec();
1083         rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA);
1084         if (rc != 0)
1085                 GOTO(stop, rc);
1086
1087         rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
1088
1089         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1090
1091 stop:
1092         dt_trans_stop(env, dev, th);
1093
1094 unlock:
1095         lfsck_ibits_unlock(&lh, LCK_EX);
1096
1097 log:
1098         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1099                "the name %s and type %o to the parent "DFID": rc = %d\n",
1100                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
1101                lfsck_object_type(child) & S_IFMT,
1102                PFID(lfsck_dto2fid(parent)), rc);
1103
1104         if (rc != 0) {
1105                 struct lfsck_namespace *ns = com->lc_file_ram;
1106
1107                 ns->ln_flags |= LF_INCONSISTENT;
1108                 if (rc > 0)
1109                         ns->ln_lost_dirent_repaired++;
1110         }
1111
1112         return rc;
1113 }
1114
1115 /**
1116  * Create the specified orphan MDT-object on remote MDT.
1117  *
1118  * The LFSCK instance on this MDT will send LFSCK RPC to remote MDT to
1119  * ask the remote LFSCK instance to create the specified orphan object
1120  * under .lustre/lost+found/MDTxxxx/ directory with the name:
1121  * ${FID}-P-${conflict_version}.
1122  *
1123  * \param[in] env       pointer to the thread context
1124  * \param[in] com       pointer to the lfsck component
1125  * \param[in] orphan    pointer to the orphan MDT-object
1126  * \param[in] type      the orphan's type to be created
1127  *
1128  *  type "P":           The orphan object to be created was a parent directory
1129  *                      of some MDT-object which linkEA shows that the @orphan
1130  *                      object is missing.
1131  *
1132  * \see lfsck_layout_recreate_parent() for more types.
1133  *
1134  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1135  *
1136  * \retval              positive number for repaired cases
1137  * \retval              0 if needs to repair nothing
1138  * \retval              negative error number on failure
1139  */
1140 static int lfsck_namespace_create_orphan_remote(const struct lu_env *env,
1141                                                 struct lfsck_component *com,
1142                                                 struct dt_object *orphan,
1143                                                 __u32 type,
1144                                                 struct lmv_mds_md_v1 *lmv)
1145 {
1146         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1147         struct lfsck_request            *lr     = &info->lti_lr;
1148         struct lu_seq_range             *range  = &info->lti_range;
1149         const struct lu_fid             *fid    = lfsck_dto2fid(orphan);
1150         struct lfsck_namespace          *ns     = com->lc_file_ram;
1151         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1152         struct seq_server_site          *ss     =
1153                         lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
1154         struct lfsck_tgt_desc           *ltd    = NULL;
1155         struct ptlrpc_request           *req    = NULL;
1156         int                              rc;
1157         ENTRY;
1158
1159         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1160                 GOTO(out, rc = 1);
1161
1162         fld_range_set_mdt(range);
1163         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1164         if (rc != 0)
1165                 GOTO(out, rc);
1166
1167         ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, range->lsr_index);
1168         if (ltd == NULL) {
1169                 ns->ln_flags |= LF_INCOMPLETE;
1170
1171                 GOTO(out, rc = -ENODEV);
1172         }
1173
1174         req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp),
1175                                    &RQF_LFSCK_NOTIFY);
1176         if (req == NULL)
1177                 GOTO(out, rc = -ENOMEM);
1178
1179         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1180         if (rc != 0) {
1181                 ptlrpc_request_free(req);
1182
1183                 GOTO(out, rc);
1184         }
1185
1186         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1187         memset(lr, 0, sizeof(*lr));
1188         lr->lr_event = LE_CREATE_ORPHAN;
1189         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1190         lr->lr_active = LFSCK_TYPE_NAMESPACE;
1191         lr->lr_fid = *fid;
1192         lr->lr_type = type;
1193         if (lmv != NULL) {
1194                 lr->lr_hash_type = lmv->lmv_hash_type;
1195                 lr->lr_stripe_count = lmv->lmv_stripe_count;
1196                 lr->lr_layout_version = lmv->lmv_layout_version;
1197                 memcpy(lr->lr_pool_name, lmv->lmv_pool_name,
1198                        sizeof(lr->lr_pool_name));
1199         }
1200
1201         ptlrpc_request_set_replen(req);
1202         rc = ptlrpc_queue_wait(req);
1203         ptlrpc_req_finished(req);
1204
1205         if (rc == 0)
1206                 rc = 1;
1207         else if (rc == -EEXIST)
1208                 rc = 0;
1209
1210         GOTO(out, rc);
1211
1212 out:
1213         CDEBUG(D_LFSCK, "%s: namespace LFSCK create object "
1214                DFID" on the MDT %x remotely: rc = %d\n",
1215                lfsck_lfsck2name(lfsck), PFID(fid),
1216                ltd != NULL ? ltd->ltd_index : -1, rc);
1217
1218         if (ltd != NULL)
1219                 lfsck_tgt_put(ltd);
1220
1221         return rc;
1222 }
1223
1224 /**
1225  * Create the specified orphan MDT-object locally.
1226  *
1227  * For the case that the parent MDT-object stored in some MDT-object's
1228  * linkEA entry is lost, the LFSCK will re-create the parent object as
1229  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1230  * with the name ${FID}-P-${conflict_version}.
1231  *
1232  * \param[in] env       pointer to the thread context
1233  * \param[in] com       pointer to the lfsck component
1234  * \param[in] orphan    pointer to the orphan MDT-object to be created
1235  * \param[in] type      the orphan's type to be created
1236  *
1237  *  type "P":           The orphan object to be created was a parent directory
1238  *                      of some MDT-object which linkEA shows that the @orphan
1239  *                      object is missing.
1240  *
1241  * \see lfsck_layout_recreate_parent() for more types.
1242  *
1243  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1244  *
1245  * \retval              positive number for repaired cases
1246  * \retval              negative error number on failure
1247  */
1248 static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
1249                                                struct lfsck_component *com,
1250                                                struct dt_object *orphan,
1251                                                __u32 type,
1252                                                struct lmv_mds_md_v1 *lmv)
1253 {
1254         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1255         struct lu_attr                  *la     = &info->lti_la;
1256         struct dt_allocation_hint       *hint   = &info->lti_hint;
1257         struct dt_object_format         *dof    = &info->lti_dof;
1258         struct lu_name                  *cname  = &info->lti_name2;
1259         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1260         struct lu_fid                   *tfid   = &info->lti_fid;
1261         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1262         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1263         const struct lu_fid             *pfid;
1264         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1265         struct dt_device                *dev    = lfsck->li_bottom;
1266         struct dt_object                *parent = NULL;
1267         struct dt_object                *child  = NULL;
1268         struct thandle                  *th     = NULL;
1269         struct lustre_handle             lh     = { 0 };
1270         struct linkea_data               ldata  = { 0 };
1271         struct lu_buf                    linkea_buf;
1272         struct lu_buf                    lmv_buf;
1273         char                             name[32];
1274         int                              namelen;
1275         int                              idx    = 0;
1276         int                              rc     = 0;
1277         ENTRY;
1278
1279         LASSERT(!dt_object_exists(orphan));
1280         LASSERT(!dt_object_remote(orphan));
1281
1282         /* @orphan maybe not attached to lfsck->li_bottom */
1283         child = lfsck_object_find_by_dev(env, dev, cfid);
1284         if (IS_ERR(child))
1285                 GOTO(log, rc = PTR_ERR(child));
1286
1287         cname->ln_name = NULL;
1288         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1289                 GOTO(log, rc = 1);
1290
1291         /* Create .lustre/lost+found/MDTxxxx when needed. */
1292         if (unlikely(lfsck->li_lpf_obj == NULL)) {
1293                 rc = lfsck_create_lpf(env, lfsck);
1294                 if (rc != 0)
1295                         GOTO(log, rc);
1296         }
1297
1298         parent = lfsck->li_lpf_obj;
1299         pfid = lfsck_dto2fid(parent);
1300
1301         /* Hold update lock on the parent to prevent others to access. */
1302         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1303                               MDS_INODELOCK_UPDATE, LCK_EX);
1304         if (rc != 0)
1305                 GOTO(log, rc);
1306
1307         do {
1308                 namelen = snprintf(name, 31, DFID"-P-%d",
1309                                    PFID(cfid), idx++);
1310                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1311                                (const struct dt_key *)name, BYPASS_CAPA);
1312                 if (rc != 0 && rc != -ENOENT)
1313                         GOTO(unlock1, rc);
1314         } while (rc == 0);
1315
1316         cname->ln_name = name;
1317         cname->ln_namelen = namelen;
1318
1319         memset(la, 0, sizeof(*la));
1320         la->la_mode = type | (S_ISDIR(type) ? 0700 : 0600);
1321         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1322                        LA_ATIME | LA_MTIME | LA_CTIME;
1323
1324         child->do_ops->do_ah_init(env, hint, parent, child,
1325                                   la->la_mode & S_IFMT);
1326
1327         memset(dof, 0, sizeof(*dof));
1328         dof->dof_type = dt_mode_to_dft(type);
1329
1330         rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1331         if (rc != 0)
1332                 GOTO(unlock1, rc);
1333
1334         rc = linkea_add_buf(&ldata, cname, pfid);
1335         if (rc != 0)
1336                 GOTO(unlock1, rc);
1337
1338         th = dt_trans_create(env, dev);
1339         if (IS_ERR(th))
1340                 GOTO(unlock1, rc = PTR_ERR(th));
1341
1342         rc = dt_declare_create(env, child, la, hint, dof, th);
1343         if (rc == 0 && S_ISDIR(type))
1344                 rc = dt_declare_ref_add(env, child, th);
1345
1346         if (rc != 0)
1347                 GOTO(stop, rc);
1348
1349         if (lmv != NULL) {
1350                 lmv->lmv_magic = LMV_MAGIC;
1351                 lmv->lmv_master_mdt_index = lfsck_dev_idx(dev);
1352                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1353                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1354                 rc = dt_declare_xattr_set(env, child, &lmv_buf,
1355                                           XATTR_NAME_LMV, 0, th);
1356                 if (rc != 0)
1357                         GOTO(stop, rc);
1358         }
1359
1360         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1361                        ldata.ld_leh->leh_len);
1362         rc = dt_declare_xattr_set(env, child, &linkea_buf,
1363                                   XATTR_NAME_LINK, 0, th);
1364         if (rc != 0)
1365                 GOTO(stop, rc);
1366
1367         rec->rec_type = type;
1368         rec->rec_fid = cfid;
1369         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1370                                (const struct dt_key *)name, th);
1371         if (rc == 0 && S_ISDIR(type))
1372                 rc = dt_declare_ref_add(env, parent, th);
1373
1374         if (rc != 0)
1375                 GOTO(stop, rc);
1376
1377         rc = dt_trans_start_local(env, dev, th);
1378         if (rc != 0)
1379                 GOTO(stop, rc);
1380
1381         dt_write_lock(env, child, 0);
1382         rc = dt_create(env, child, la, hint, dof, th);
1383         if (rc != 0)
1384                 GOTO(unlock2, rc);
1385
1386         if (S_ISDIR(type)) {
1387                 if (unlikely(!dt_try_as_dir(env, child)))
1388                         GOTO(unlock2, rc = -ENOTDIR);
1389
1390                 rec->rec_type = S_IFDIR;
1391                 rec->rec_fid = cfid;
1392                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1393                                (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
1394                 if (rc != 0)
1395                         GOTO(unlock2, rc);
1396
1397                 rec->rec_fid = pfid;
1398                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1399                                (const struct dt_key *)dotdot, th,
1400                                BYPASS_CAPA, 1);
1401                 if (rc != 0)
1402                         GOTO(unlock2, rc);
1403
1404                 rc = dt_ref_add(env, child, th);
1405                 if (rc != 0)
1406                         GOTO(unlock2, rc);
1407         }
1408
1409         if (lmv != NULL) {
1410                 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, 0,
1411                                   th, BYPASS_CAPA);
1412                 if (rc != 0)
1413                         GOTO(unlock2, rc);
1414         }
1415
1416         rc = dt_xattr_set(env, child, &linkea_buf,
1417                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1418         dt_write_unlock(env, child);
1419         if (rc != 0)
1420                 GOTO(stop, rc);
1421
1422         rec->rec_type = type;
1423         rec->rec_fid = cfid;
1424         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1425                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1426         if (rc == 0 && S_ISDIR(type)) {
1427                 dt_write_lock(env, parent, 0);
1428                 rc = dt_ref_add(env, parent, th);
1429                 dt_write_unlock(env, parent);
1430         }
1431
1432         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1433
1434 unlock2:
1435         dt_write_unlock(env, child);
1436
1437 stop:
1438         dt_trans_stop(env, dev, th);
1439
1440 unlock1:
1441         lfsck_ibits_unlock(&lh, LCK_EX);
1442
1443 log:
1444         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan locally for "
1445                "the object "DFID", name = %s, type %o: rc = %d\n",
1446                lfsck_lfsck2name(lfsck), PFID(cfid),
1447                cname->ln_name != NULL ? cname->ln_name : "<NULL>", type, rc);
1448
1449         if (child != NULL && !IS_ERR(child))
1450                 lfsck_object_put(env, child);
1451
1452         return rc;
1453 }
1454
1455 /**
1456  * Create the specified orphan MDT-object.
1457  *
1458  * For the case that the parent MDT-object stored in some MDT-object's
1459  * linkEA entry is lost, the LFSCK will re-create the parent object as
1460  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1461  * with the name: ${FID}-P-${conflict_version}.
1462  *
1463  * \param[in] env       pointer to the thread context
1464  * \param[in] com       pointer to the lfsck component
1465  * \param[in] orphan    pointer to the orphan MDT-object
1466  *
1467  *  type "P":           The orphan object to be created was a parent directory
1468  *                      of some MDT-object which linkEA shows that the @orphan
1469  *                      object is missing.
1470  *
1471  * \see lfsck_layout_recreate_parent() for more types.
1472  *
1473  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1474  *
1475  * \retval              positive number for repaired cases
1476  * \retval              0 if needs to repair nothing
1477  * \retval              negative error number on failure
1478  */
1479 static int lfsck_namespace_create_orphan(const struct lu_env *env,
1480                                          struct lfsck_component *com,
1481                                          struct dt_object *orphan,
1482                                          struct lmv_mds_md_v1 *lmv)
1483 {
1484         struct lfsck_namespace *ns = com->lc_file_ram;
1485         int                     rc;
1486
1487         if (dt_object_remote(orphan))
1488                 rc = lfsck_namespace_create_orphan_remote(env, com, orphan,
1489                                                           S_IFDIR, lmv);
1490         else
1491                 rc = lfsck_namespace_create_orphan_local(env, com, orphan,
1492                                                          S_IFDIR, lmv);
1493
1494         if (rc != 0)
1495                 ns->ln_flags |= LF_INCONSISTENT;
1496
1497         return rc;
1498 }
1499
1500 /**
1501  * Remove the specified entry from the linkEA.
1502  *
1503  * Locate the linkEA entry with the given @cname and @pfid, then
1504  * remove this entry or the other entries those are repeated with
1505  * this entry.
1506  *
1507  * \param[in] env       pointer to the thread context
1508  * \param[in] com       pointer to the lfsck component
1509  * \param[in] obj       pointer to the dt_object to be handled
1510  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1511  * \param[in] cname     the name for the child in the parent directory
1512  * \param[in] pfid      the parent directory's FID for the linkEA
1513  * \param[in] next      if true, then remove the first found linkEA
1514  *                      entry, and move the ldata->ld_lee to next entry
1515  *
1516  * \retval              positive number for repaired cases
1517  * \retval              0 if nothing to be repaired
1518  * \retval              negative error number on failure
1519  */
1520 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1521                                          struct lfsck_component *com,
1522                                          struct dt_object *obj,
1523                                          struct linkea_data *ldata,
1524                                          struct lu_name *cname,
1525                                          struct lu_fid *pfid,
1526                                          bool next)
1527 {
1528         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1529         struct dt_device                *dev       = lfsck->li_bottom;
1530         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1531         struct thandle                  *th        = NULL;
1532         struct lustre_handle             lh        = { 0 };
1533         struct linkea_data               ldata_new = { 0 };
1534         struct lu_buf                    linkea_buf;
1535         int                              rc        = 0;
1536         ENTRY;
1537
1538         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1539                               MDS_INODELOCK_UPDATE |
1540                               MDS_INODELOCK_XATTR, LCK_EX);
1541         if (rc != 0)
1542                 GOTO(log, rc);
1543
1544         if (next)
1545                 linkea_del_buf(ldata, cname);
1546         else
1547                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1548                                                     true);
1549         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1550                        ldata->ld_leh->leh_len);
1551
1552 again:
1553         th = dt_trans_create(env, dev);
1554         if (IS_ERR(th))
1555                 GOTO(unlock1, rc = PTR_ERR(th));
1556
1557         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1558                                   XATTR_NAME_LINK, 0, th);
1559         if (rc != 0)
1560                 GOTO(stop, rc);
1561
1562         rc = dt_trans_start_local(env, dev, th);
1563         if (rc != 0)
1564                 GOTO(stop, rc);
1565
1566         dt_write_lock(env, obj, 0);
1567         if (unlikely(lfsck_is_dead_obj(obj)))
1568                 GOTO(unlock2, rc = -ENOENT);
1569
1570         rc = lfsck_links_read2(env, obj, &ldata_new);
1571         if (rc != 0)
1572                 GOTO(unlock2, rc);
1573
1574         /* The specified linkEA entry has been removed by race. */
1575         rc = linkea_links_find(&ldata_new, cname, pfid);
1576         if (rc != 0)
1577                 GOTO(unlock2, rc = 0);
1578
1579         if (bk->lb_param & LPF_DRYRUN)
1580                 GOTO(unlock2, rc = 1);
1581
1582         if (next)
1583                 linkea_del_buf(&ldata_new, cname);
1584         else
1585                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1586                                                     true);
1587
1588         if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1589                 dt_write_unlock(env, obj);
1590                 dt_trans_stop(env, dev, th);
1591                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1592                                ldata_new.ld_leh->leh_len);
1593                 goto again;
1594         }
1595
1596         lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1597                        ldata_new.ld_leh->leh_len);
1598         rc = dt_xattr_set(env, obj, &linkea_buf,
1599                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1600
1601         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1602
1603 unlock2:
1604         dt_write_unlock(env, obj);
1605
1606 stop:
1607         dt_trans_stop(env, dev, th);
1608
1609 unlock1:
1610         lfsck_ibits_unlock(&lh, LCK_EX);
1611
1612 log:
1613         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1614                "for the object: "DFID", parent "DFID", name %.*s\n",
1615                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1616                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1617                cname->ln_name);
1618
1619         if (rc != 0) {
1620                 struct lfsck_namespace *ns = com->lc_file_ram;
1621
1622                 ns->ln_flags |= LF_INCONSISTENT;
1623         }
1624
1625         return rc;
1626 }
1627
1628 /**
1629  * Conditionally remove the specified entry from the linkEA.
1630  *
1631  * Take the parent lock firstly, then check whether the specified
1632  * name entry exists or not: if yes, do nothing; otherwise, call
1633  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1634  *
1635  * \param[in] env       pointer to the thread context
1636  * \param[in] com       pointer to the lfsck component
1637  * \param[in] parent    pointer to the parent directory
1638  * \param[in] child     pointer to the child object that holds the linkEA
1639  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1640  * \param[in] cname     the name for the child in the parent directory
1641  * \param[in] pfid      the parent directory's FID for the linkEA
1642  *
1643  * \retval              positive number for repaired cases
1644  * \retval              0 if nothing to be repaired
1645  * \retval              negative error number on failure
1646  */
1647 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1648                                               struct lfsck_component *com,
1649                                               struct dt_object *parent,
1650                                               struct dt_object *child,
1651                                               struct linkea_data *ldata,
1652                                               struct lu_name *cname,
1653                                               struct lu_fid *pfid)
1654 {
1655         struct lu_fid           *cfid   = &lfsck_env_info(env)->lti_fid3;
1656         struct lustre_handle     lh     = { 0 };
1657         int                      rc;
1658         ENTRY;
1659
1660         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
1661                               MDS_INODELOCK_UPDATE, LCK_EX);
1662         if (rc != 0)
1663                 RETURN(rc);
1664
1665         dt_read_lock(env, parent, 0);
1666         if (unlikely(lfsck_is_dead_obj(parent))) {
1667                 dt_read_unlock(env, parent);
1668                 lfsck_ibits_unlock(&lh, LCK_EX);
1669                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1670                                                    cname, pfid, true);
1671
1672                 RETURN(rc);
1673         }
1674
1675         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1676                        (const struct dt_key *)cname->ln_name,
1677                        BYPASS_CAPA);
1678         dt_read_unlock(env, parent);
1679
1680         /* It is safe to release the ldlm lock, because when the logic come
1681          * here, we have got all the needed information above whether the
1682          * linkEA entry is valid or not. It is not important that others
1683          * may add new linkEA entry after the ldlm lock released. If other
1684          * has removed the specified linkEA entry by race, then it is OK,
1685          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1686          * such case. */
1687         lfsck_ibits_unlock(&lh, LCK_EX);
1688         if (rc == -ENOENT) {
1689                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1690                                                    cname, pfid, true);
1691
1692                 RETURN(rc);
1693         }
1694
1695         if (rc != 0)
1696                 RETURN(rc);
1697
1698         /* The LFSCK just found some internal status of cross-MDTs
1699          * create operation. That is normal. */
1700         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1701                 linkea_next_entry(ldata);
1702
1703                 RETURN(0);
1704         }
1705
1706         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1707                                            pfid, true);
1708
1709         RETURN(rc);
1710 }
1711
1712 /**
1713  * Conditionally replace name entry in the parent.
1714  *
1715  * As required, the LFSCK may re-create the lost MDT-object for dangling
1716  * name entry, but such repairing may be wrong because of bad FID in the
1717  * name entry. As the LFSCK processing, the real MDT-object may be found,
1718  * then the LFSCK should check whether the former re-created MDT-object
1719  * has been modified or not, if not, then destroy it and update the name
1720  * entry in the parent to reference the real MDT-object.
1721  *
1722  * \param[in] env       pointer to the thread context
1723  * \param[in] com       pointer to the lfsck component
1724  * \param[in] parent    pointer to the parent directory
1725  * \param[in] child     pointer to the MDT-object that may be the real
1726  *                      MDT-object corresponding to the name entry in parent
1727  * \param[in] cfid      the current FID in the name entry
1728  * \param[in] cname     contains the name of the child in the parent directory
1729  *
1730  * \retval              positive number for repaired cases
1731  * \retval              0 if nothing to be repaired
1732  * \retval              negative error number on failure
1733  */
1734 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1735                                         struct lfsck_component *com,
1736                                         struct dt_object *parent,
1737                                         struct dt_object *child,
1738                                         const struct lu_fid *cfid,
1739                                         const struct lu_name *cname)
1740 {
1741         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1742         struct lu_fid                   *tfid   = &info->lti_fid5;
1743         struct lu_attr                  *la     = &info->lti_la;
1744         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1745         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1746         struct dt_device                *dev    = lfsck->li_next;
1747         const char                      *name   = cname->ln_name;
1748         struct dt_object                *obj    = NULL;
1749         struct lustre_handle             plh    = { 0 };
1750         struct lustre_handle             clh    = { 0 };
1751         struct linkea_data               ldata  = { 0 };
1752         struct thandle                  *th     = NULL;
1753         bool                             exist  = true;
1754         int                              rc     = 0;
1755         ENTRY;
1756
1757         rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
1758                               MDS_INODELOCK_UPDATE, LCK_EX);
1759         if (rc != 0)
1760                 GOTO(log, rc);
1761
1762         if (!fid_is_sane(cfid)) {
1763                 exist = false;
1764                 goto replace;
1765         }
1766
1767         obj = lfsck_object_find(env, lfsck, cfid);
1768         if (IS_ERR(obj)) {
1769                 rc = PTR_ERR(obj);
1770                 if (rc == -ENOENT) {
1771                         exist = false;
1772                         goto replace;
1773                 }
1774
1775                 GOTO(log, rc);
1776         }
1777
1778         if (!dt_object_exists(obj)) {
1779                 exist = false;
1780                 goto replace;
1781         }
1782
1783         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1784                        (const struct dt_key *)name, BYPASS_CAPA);
1785         if (rc == -ENOENT) {
1786                 exist = false;
1787                 goto replace;
1788         }
1789
1790         if (rc != 0)
1791                 GOTO(log, rc);
1792
1793         /* Someone changed the name entry, cannot replace it. */
1794         if (!lu_fid_eq(cfid, tfid))
1795                 GOTO(log, rc = 0);
1796
1797         /* lock the object to be destroyed. */
1798         rc = lfsck_ibits_lock(env, lfsck, obj, &clh,
1799                               MDS_INODELOCK_UPDATE |
1800                               MDS_INODELOCK_XATTR, LCK_EX);
1801         if (rc != 0)
1802                 GOTO(log, rc);
1803
1804         if (unlikely(lfsck_is_dead_obj(obj))) {
1805                 exist = false;
1806                 goto replace;
1807         }
1808
1809         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
1810         if (rc != 0)
1811                 GOTO(log, rc);
1812
1813         /* The object has been modified by other(s), or it is not created by
1814          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1815         if (la->la_ctime != 0)
1816                 GOTO(log, rc);
1817
1818         if (S_ISREG(la->la_mode)) {
1819                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
1820                                   BYPASS_CAPA);
1821                 /* If someone has created related OST-object(s),
1822                  * then keep it. */
1823                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1824                         GOTO(log, rc = (rc > 0 ? 0 : rc));
1825         }
1826
1827 replace:
1828         dt_read_lock(env, child, 0);
1829         rc = lfsck_links_read2(env, child, &ldata);
1830         dt_read_unlock(env, child);
1831
1832         /* Someone changed the child, no need to replace. */
1833         if (rc == -ENODATA)
1834                 GOTO(log, rc = 0);
1835
1836         if (rc != 0)
1837                 GOTO(log, rc);
1838
1839         rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(parent));
1840         /* Someone moved the child, no need to replace. */
1841         if (rc != 0)
1842                 GOTO(log, rc = 0);
1843
1844         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1845                 GOTO(log, rc = 1);
1846
1847         th = dt_trans_create(env, dev);
1848         if (IS_ERR(th))
1849                 GOTO(log, rc = PTR_ERR(th));
1850
1851         if (exist) {
1852                 rc = dt_declare_destroy(env, obj, th);
1853                 if (rc != 0)
1854                         GOTO(stop, rc);
1855         }
1856
1857         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1858         if (rc != 0)
1859                 GOTO(stop, rc);
1860
1861         rec->rec_type = S_IFDIR;
1862         rec->rec_fid = lfsck_dto2fid(child);
1863         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1864                                (const struct dt_key *)name, th);
1865         if (rc != 0)
1866                 GOTO(stop, rc);
1867
1868         rc = dt_trans_start(env, dev, th);
1869         if (rc != 0)
1870                 GOTO(stop, rc);
1871
1872         if (exist) {
1873                 rc = dt_destroy(env, obj, th);
1874                 if (rc != 0)
1875                         GOTO(stop, rc);
1876         }
1877
1878         /* The old name entry maybe not exist. */
1879         dt_delete(env, parent, (const struct dt_key *)name, th,
1880                   BYPASS_CAPA);
1881
1882         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1883                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1884
1885         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1886
1887 stop:
1888         dt_trans_stop(env, dev, th);
1889
1890 log:
1891         lfsck_ibits_unlock(&clh, LCK_EX);
1892         lfsck_ibits_unlock(&plh, LCK_EX);
1893         if (obj != NULL && !IS_ERR(obj))
1894                 lfsck_object_put(env, obj);
1895
1896         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1897                "object "DFID" because of conflict with the object "DFID
1898                " under the parent "DFID" with name %s: rc = %d\n",
1899                lfsck_lfsck2name(lfsck), PFID(cfid),
1900                PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)),
1901                name, rc);
1902
1903         return rc;
1904 }
1905
1906 /**
1907  * Overwrite the linkEA for the object with the given ldata.
1908  *
1909  * The caller should take the ldlm lock before the calling.
1910  *
1911  * \param[in] env       pointer to the thread context
1912  * \param[in] com       pointer to the lfsck component
1913  * \param[in] obj       pointer to the dt_object to be handled
1914  * \param[in] ldata     pointer to the new linkEA data
1915  *
1916  * \retval              positive number for repaired cases
1917  * \retval              0 if nothing to be repaired
1918  * \retval              negative error number on failure
1919  */
1920 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1921                                    struct lfsck_component *com,
1922                                    struct dt_object *obj,
1923                                    struct linkea_data *ldata)
1924 {
1925         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1926         struct dt_device                *dev    = lfsck->li_bottom;
1927         struct thandle                  *th     = NULL;
1928         struct lu_buf                    linkea_buf;
1929         int                              rc     = 0;
1930         ENTRY;
1931
1932         LASSERT(!dt_object_remote(obj));
1933
1934         th = dt_trans_create(env, dev);
1935         if (IS_ERR(th))
1936                 GOTO(log, rc = PTR_ERR(th));
1937
1938         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1939                        ldata->ld_leh->leh_len);
1940         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1941                                   XATTR_NAME_LINK, 0, th);
1942         if (rc != 0)
1943                 GOTO(stop, rc);
1944
1945         rc = dt_trans_start_local(env, dev, th);
1946         if (rc != 0)
1947                 GOTO(stop, rc);
1948
1949         dt_write_lock(env, obj, 0);
1950         if (unlikely(lfsck_is_dead_obj(obj)))
1951                 GOTO(unlock, rc = 0);
1952
1953         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1954                 GOTO(unlock, rc = 1);
1955
1956         rc = dt_xattr_set(env, obj, &linkea_buf,
1957                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1958
1959         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1960
1961 unlock:
1962         dt_write_unlock(env, obj);
1963
1964 stop:
1965         dt_trans_stop(env, dev, th);
1966
1967 log:
1968         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1969                "object "DFID": rc = %d\n",
1970                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1971
1972         if (rc != 0) {
1973                 struct lfsck_namespace *ns = com->lc_file_ram;
1974
1975                 ns->ln_flags |= LF_INCONSISTENT;
1976         }
1977
1978         return rc;
1979 }
1980
1981 /**
1982  * Repair invalid name entry.
1983  *
1984  * If the name entry contains invalid information, such as bad file type
1985  * or (and) corrupted object FID, then either remove the name entry or
1986  * udpate the name entry with the given (right) information.
1987  *
1988  * \param[in] env       pointer to the thread context
1989  * \param[in] com       pointer to the lfsck component
1990  * \param[in] parent    pointer to the parent directory
1991  * \param[in] child     pointer to the object referenced by the name entry
1992  * \param[in] name      the old name of the child under the parent directory
1993  * \param[in] name2     the new name of the child under the parent directory
1994  * \param[in] type      the type claimed by the name entry
1995  * \param[in] update    update the name entry if true; otherwise, remove it
1996  * \param[in] dec       decrease the parent nlink count if true
1997  *
1998  * \retval              positive number for repaired successfully
1999  * \retval              0 if nothing to be repaired
2000  * \retval              negative error number on failure
2001  */
2002 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2003                                   struct lfsck_component *com,
2004                                   struct dt_object *parent,
2005                                   struct dt_object *child,
2006                                   const char *name, const char *name2,
2007                                   __u16 type, bool update, bool dec)
2008 {
2009         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2010         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2011         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
2012         struct lu_fid                   *tfid   = &info->lti_fid5;
2013         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2014         struct dt_device                *dev    = lfsck->li_next;
2015         struct thandle                  *th     = NULL;
2016         struct lustre_handle             lh     = { 0 };
2017         int                              rc     = 0;
2018         ENTRY;
2019
2020         if (unlikely(!dt_try_as_dir(env, parent)))
2021                 GOTO(log, rc = -ENOTDIR);
2022
2023         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2024                               MDS_INODELOCK_UPDATE, LCK_EX);
2025         if (rc != 0)
2026                 GOTO(log, rc);
2027
2028         th = dt_trans_create(env, dev);
2029         if (IS_ERR(th))
2030                 GOTO(unlock1, rc = PTR_ERR(th));
2031
2032         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2033         if (rc != 0)
2034                 GOTO(stop, rc);
2035
2036         if (update) {
2037                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2038                 rec->rec_fid = cfid;
2039                 rc = dt_declare_insert(env, parent,
2040                                        (const struct dt_rec *)rec,
2041                                        (const struct dt_key *)name2, th);
2042                 if (rc != 0)
2043                         GOTO(stop, rc);
2044         }
2045
2046         if (dec) {
2047                 rc = dt_declare_ref_del(env, parent, th);
2048                 if (rc != 0)
2049                         GOTO(stop, rc);
2050         }
2051
2052         rc = dt_trans_start(env, dev, th);
2053         if (rc != 0)
2054                 GOTO(stop, rc);
2055
2056         dt_write_lock(env, parent, 0);
2057         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2058                        (const struct dt_key *)name, BYPASS_CAPA);
2059         /* Someone has removed the bad name entry by race. */
2060         if (rc == -ENOENT)
2061                 GOTO(unlock2, rc = 0);
2062
2063         if (rc != 0)
2064                 GOTO(unlock2, rc);
2065
2066         /* Someone has removed the bad name entry and reused it for other
2067          * object by race. */
2068         if (!lu_fid_eq(tfid, cfid))
2069                 GOTO(unlock2, rc = 0);
2070
2071         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2072                 GOTO(unlock2, rc = 1);
2073
2074         rc = dt_delete(env, parent, (const struct dt_key *)name, th,
2075                        BYPASS_CAPA);
2076         if (rc != 0)
2077                 GOTO(unlock2, rc);
2078
2079         if (update) {
2080                 rc = dt_insert(env, parent,
2081                                (const struct dt_rec *)rec,
2082                                (const struct dt_key *)name2, th,
2083                                BYPASS_CAPA, 1);
2084                 if (rc != 0)
2085                         GOTO(unlock2, rc);
2086         }
2087
2088         if (dec) {
2089                 rc = dt_ref_del(env, parent, th);
2090                 if (rc != 0)
2091                         GOTO(unlock2, rc);
2092         }
2093
2094         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2095
2096 unlock2:
2097         dt_write_unlock(env, parent);
2098
2099 stop:
2100         dt_trans_stop(env, dev, th);
2101
2102         /* We are not sure whether the child will become orphan or not.
2103          * Record it in the LFSCK trace file for further checking in
2104          * the second-stage scanning. */
2105         if (!update && !dec && rc == 0)
2106                 lfsck_namespace_trace_update(env, com, cfid,
2107                                              LNTF_CHECK_LINKEA, true);
2108
2109 unlock1:
2110         lfsck_ibits_unlock(&lh, LCK_EX);
2111
2112 log:
2113         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2114                "entry for: parent "DFID", child "DFID", name %s, type "
2115                "in name entry %o, type claimed by child %o. repair it "
2116                "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
2117                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2118                name, type, update ? lfsck_object_type(child) : 0,
2119                update ? "updating" : "removing", name2, rc);
2120
2121         if (rc != 0) {
2122                 struct lfsck_namespace *ns = com->lc_file_ram;
2123
2124                 ns->ln_flags |= LF_INCONSISTENT;
2125         }
2126
2127         return rc;
2128 }
2129
2130 /**
2131  * Update the ".." name entry for the given object.
2132  *
2133  * The object's ".." is corrupted, this function will update the ".." name
2134  * entry with the given pfid, and the linkEA with the given ldata.
2135  *
2136  * The caller should take the ldlm lock before the calling.
2137  *
2138  * \param[in] env       pointer to the thread context
2139  * \param[in] com       pointer to the lfsck component
2140  * \param[in] obj       pointer to the dt_object to be handled
2141  * \param[in] pfid      the new fid for the object's ".." name entry
2142  * \param[in] cname     the name for the @obj in the parent directory
2143  *
2144  * \retval              positive number for repaired cases
2145  * \retval              0 if nothing to be repaired
2146  * \retval              negative error number on failure
2147  */
2148 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2149                                                   struct lfsck_component *com,
2150                                                   struct dt_object *obj,
2151                                                   const struct lu_fid *pfid,
2152                                                   struct lu_name *cname)
2153 {
2154         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2155         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2156         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2157         struct dt_device                *dev    = lfsck->li_bottom;
2158         struct thandle                  *th     = NULL;
2159         struct linkea_data               ldata  = { 0 };
2160         struct lu_buf                    linkea_buf;
2161         int                              rc     = 0;
2162         ENTRY;
2163
2164         LASSERT(!dt_object_remote(obj));
2165         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2166
2167         rc = linkea_data_new(&ldata, &info->lti_big_buf);
2168         if (rc != 0)
2169                 GOTO(log, rc);
2170
2171         rc = linkea_add_buf(&ldata, cname, pfid);
2172         if (rc != 0)
2173                 GOTO(log, rc);
2174
2175         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2176                        ldata.ld_leh->leh_len);
2177
2178         th = dt_trans_create(env, dev);
2179         if (IS_ERR(th))
2180                 GOTO(log, rc = PTR_ERR(th));
2181
2182         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2183         if (rc != 0)
2184                 GOTO(stop, rc);
2185
2186         rec->rec_type = S_IFDIR;
2187         rec->rec_fid = pfid;
2188         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2189                                (const struct dt_key *)dotdot, th);
2190         if (rc != 0)
2191                 GOTO(stop, rc);
2192
2193         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2194                                   XATTR_NAME_LINK, 0, th);
2195         if (rc != 0)
2196                 GOTO(stop, rc);
2197
2198         rc = dt_trans_start_local(env, dev, th);
2199         if (rc != 0)
2200                 GOTO(stop, rc);
2201
2202         dt_write_lock(env, obj, 0);
2203         if (unlikely(lfsck_is_dead_obj(obj)))
2204                 GOTO(unlock, rc = 0);
2205
2206         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2207                 GOTO(unlock, rc = 1);
2208
2209         /* The old ".." name entry maybe not exist. */
2210         dt_delete(env, obj, (const struct dt_key *)dotdot, th,
2211                   BYPASS_CAPA);
2212
2213         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2214                        (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
2215         if (rc != 0)
2216                 GOTO(unlock, rc);
2217
2218         rc = dt_xattr_set(env, obj, &linkea_buf,
2219                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2220
2221         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2222
2223 unlock:
2224         dt_write_unlock(env, obj);
2225
2226 stop:
2227         dt_trans_stop(env, dev, th);
2228
2229 log:
2230         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2231                "the object "DFID", new parent "DFID": rc = %d\n",
2232                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2233                PFID(pfid), rc);
2234
2235         if (rc != 0) {
2236                 struct lfsck_namespace *ns = com->lc_file_ram;
2237
2238                 ns->ln_flags |= LF_INCONSISTENT;
2239         }
2240
2241         return rc;
2242 }
2243
2244 /**
2245  * Handle orphan @obj during Double Scan Directory.
2246  *
2247  * Remove the @obj's current (invalid) linkEA entries, and insert
2248  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2249  * ${FID}-${PFID}-D-${conflict_version}
2250  *
2251  * The caller should take the ldlm lock before the calling.
2252  *
2253  * \param[in] env       pointer to the thread context
2254  * \param[in] com       pointer to the lfsck component
2255  * \param[in] obj       pointer to the orphan object to be handled
2256  * \param[in] pfid      the new fid for the object's ".." name entry
2257  * \param[in,out] lh    ldlm lock handler for the given @obj
2258  * \param[out] type     to tell the caller what the inconsistency is
2259  *
2260  * \retval              positive number for repaired cases
2261  * \retval              0 if nothing to be repaired
2262  * \retval              negative error number on failure
2263  */
2264 static int
2265 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2266                            struct lfsck_component *com,
2267                            struct dt_object *obj,
2268                            const struct lu_fid *pfid,
2269                            struct lustre_handle *lh,
2270                            enum lfsck_namespace_inconsistency_type *type)
2271 {
2272         struct lfsck_thread_info *info = lfsck_env_info(env);
2273         struct lfsck_namespace   *ns   = com->lc_file_ram;
2274         int                       rc;
2275         ENTRY;
2276
2277         /* Remove the unrecognized linkEA. */
2278         rc = lfsck_namespace_links_remove(env, com, obj);
2279         lfsck_ibits_unlock(lh, LCK_EX);
2280         if (rc < 0 && rc != -ENODATA)
2281                 RETURN(rc);
2282
2283         *type = LNIT_MUL_REF;
2284
2285         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2286          * ever tried to verify some remote MDT-object that resides on this
2287          * MDT, but this MDT failed to respond such request. So means there
2288          * may be some remote name entry on other MDT that references this
2289          * object with another name, so we cannot know whether this linkEA
2290          * is valid or not. So keep it there and maybe resolved when next
2291          * LFSCK run. */
2292         if (ns->ln_flags & LF_INCOMPLETE)
2293                 RETURN(0);
2294
2295         /* The unique linkEA is invalid, even if the ".." name entry may be
2296          * valid, we still cannot know via which name entry this directory
2297          * will be referenced. Then handle it as pure orphan. */
2298         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2299                  "-"DFID, PFID(pfid));
2300         rc = lfsck_namespace_insert_orphan(env, com, obj,
2301                                            info->lti_tmpbuf, "D", NULL);
2302
2303         RETURN(rc);
2304 }
2305
2306 /**
2307  * Double Scan Directory object for single linkEA entry case.
2308  *
2309  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2310  * then check whether the name is in the namespace or not, if not, add the
2311  * missing name entry back to namespace. If the linkEA entry is invalid,
2312  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2313  * as an orphan.
2314  *
2315  * \param[in] env       pointer to the thread context
2316  * \param[in] com       pointer to the lfsck component
2317  * \param[in] child     pointer to the directory to be double scanned
2318  * \param[in] pfid      the FID corresponding to the ".." entry
2319  * \param[in] ldata     pointer to the linkEA data for the given @child
2320  * \param[in,out] lh    ldlm lock handler for the given @child
2321  * \param[out] type     to tell the caller what the inconsistency is
2322  * \param[in] retry     if found inconsistency, but the caller does not hold
2323  *                      ldlm lock on the @child, then set @retry as true
2324  *
2325  * \retval              positive number for repaired cases
2326  * \retval              0 if nothing to be repaired
2327  * \retval              negative error number on failure
2328  */
2329 static int
2330 lfsck_namespace_dsd_single(const struct lu_env *env,
2331                            struct lfsck_component *com,
2332                            struct dt_object *child,
2333                            const struct lu_fid *pfid,
2334                            struct linkea_data *ldata,
2335                            struct lustre_handle *lh,
2336                            enum lfsck_namespace_inconsistency_type *type,
2337                            bool *retry)
2338 {
2339         struct lfsck_thread_info *info          = lfsck_env_info(env);
2340         struct lu_name           *cname         = &info->lti_name;
2341         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2342         struct lu_fid            *tfid          = &info->lti_fid3;
2343         struct lfsck_namespace   *ns            = com->lc_file_ram;
2344         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2345         struct dt_object         *parent        = NULL;
2346         struct lmv_mds_md_v1     *lmv;
2347         int                       rc            = 0;
2348         ENTRY;
2349
2350         lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid, info->lti_key);
2351         /* The unique linkEA entry with bad parent will be handled as orphan. */
2352         if (!fid_is_sane(tfid)) {
2353                 if (!lustre_handle_is_used(lh) && retry != NULL)
2354                         *retry = true;
2355                 else
2356                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2357                                                         pfid, lh, type);
2358
2359                 GOTO(out, rc);
2360         }
2361
2362         parent = lfsck_object_find_bottom(env, lfsck, tfid);
2363         if (IS_ERR(parent))
2364                 GOTO(out, rc = PTR_ERR(parent));
2365
2366         /* We trust the unique linkEA entry in spite of whether it matches the
2367          * ".." name entry or not. Because even if the linkEA entry is wrong
2368          * and the ".." name entry is right, we still cannot know via which
2369          * name entry the child will be referenced, since all known entries
2370          * have been verified during the first-stage scanning. */
2371         if (!dt_object_exists(parent)) {
2372                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2373                  * has ever tried to verify some remote MDT-object that resides
2374                  * on this MDT, but this MDT failed to respond such request. So
2375                  * means there may be some remote name entry on other MDT that
2376                  * references this object with another name, so we cannot know
2377                  * whether this linkEA is valid or not. So keep it there and
2378                  * maybe resolved when next LFSCK run. */
2379                 if (ns->ln_flags & LF_INCOMPLETE)
2380                         GOTO(out, rc = 0);
2381
2382                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2383                         *retry = true;
2384
2385                         GOTO(out, rc = 0);
2386                 }
2387
2388                 lfsck_ibits_unlock(lh, LCK_EX);
2389
2390 lost_parent:
2391                 lmv = &info->lti_lmv;
2392                 rc = lfsck_read_stripe_lmv(env, child, lmv);
2393                 if (rc != 0 && rc != -ENODATA)
2394                         GOTO(out, rc);
2395
2396                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2397                         lmv = NULL;
2398                 } else if (lfsck_shard_name_to_index(env,
2399                                         cname->ln_name, cname->ln_namelen,
2400                                         S_IFDIR, cfid) < 0) {
2401                         /* It is an invalid name entry, we
2402                          * cannot trust the parent also. */
2403                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2404                                                 ldata, cname, tfid, true);
2405                         if (rc < 0)
2406                                 GOTO(out, rc);
2407
2408                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2409                                  "-"DFID, PFID(pfid));
2410                         rc = lfsck_namespace_insert_orphan(env, com, child,
2411                                                 info->lti_tmpbuf, "S", NULL);
2412
2413                         GOTO(out, rc);
2414                 }
2415
2416                 /* Create the lost parent as an orphan. */
2417                 rc = lfsck_namespace_create_orphan(env, com, parent, lmv);
2418                 if (rc >= 0) {
2419                         /* Add the missing name entry to the parent. */
2420                         rc = lfsck_namespace_insert_normal(env, com, parent,
2421                                                         child, cname->ln_name);
2422                         if (unlikely(rc == -EEXIST)) {
2423                                 /* Unfortunately, someone reused the name
2424                                  * under the parent by race. So we have
2425                                  * to remove the linkEA entry from
2426                                  * current child object. It means that the
2427                                  * LFSCK cannot recover the system
2428                                  * totally back to its original status,
2429                                  * but it is necessary to make the
2430                                  * current system to be consistent. */
2431                                 rc = lfsck_namespace_shrink_linkea(env,
2432                                                 com, child, ldata,
2433                                                 cname, tfid, true);
2434                                 if (rc >= 0) {
2435                                         snprintf(info->lti_tmpbuf,
2436                                                  sizeof(info->lti_tmpbuf),
2437                                                  "-"DFID, PFID(pfid));
2438                                         rc = lfsck_namespace_insert_orphan(env,
2439                                                 com, child, info->lti_tmpbuf,
2440                                                 "D", NULL);
2441                                 }
2442                         }
2443                 }
2444
2445                 GOTO(out, rc);
2446         }
2447
2448         /* The unique linkEA entry with bad parent will be handled as orphan. */
2449         if (unlikely(!dt_try_as_dir(env, parent))) {
2450                 if (!lustre_handle_is_used(lh) && retry != NULL)
2451                         *retry = true;
2452                 else
2453                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2454                                                         pfid, lh, type);
2455
2456                 GOTO(out, rc);
2457         }
2458
2459         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2460                        (const struct dt_key *)cname->ln_name, BYPASS_CAPA);
2461         if (rc == -ENOENT) {
2462                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2463                  * has ever tried to verify some remote MDT-object that resides
2464                  * on this MDT, but this MDT failed to respond such request. So
2465                  * means there may be some remote name entry on other MDT that
2466                  * references this object with another name, so we cannot know
2467                  * whether this linkEA is valid or not. So keep it there and
2468                  * maybe resolved when next LFSCK run. */
2469                 if (ns->ln_flags & LF_INCOMPLETE)
2470                         GOTO(out, rc = 0);
2471
2472                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2473                         *retry = true;
2474
2475                         GOTO(out, rc = 0);
2476                 }
2477
2478                 lfsck_ibits_unlock(lh, LCK_EX);
2479                 rc = lfsck_namespace_check_name(env, parent, child, cname);
2480                 if (rc == -ENOENT)
2481                         goto lost_parent;
2482
2483                 if (rc < 0)
2484                         GOTO(out, rc);
2485
2486                 /* It is an invalid name entry, drop it. */
2487                 if (unlikely(rc > 0)) {
2488                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2489                                                 ldata, cname, tfid, true);
2490                         if (rc >= 0) {
2491                                 snprintf(info->lti_tmpbuf,
2492                                          sizeof(info->lti_tmpbuf),
2493                                          "-"DFID, PFID(pfid));
2494                                 rc = lfsck_namespace_insert_orphan(env, com,
2495                                         child, info->lti_tmpbuf, "D", NULL);
2496                         }
2497
2498                         GOTO(out, rc);
2499                 }
2500
2501                 /* Add the missing name entry back to the namespace. */
2502                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2503                                                    cname->ln_name);
2504                 if (unlikely(rc == -ESTALE))
2505                         /* It may happen when the remote object has been
2506                          * removed, but the local MDT is not aware of that. */
2507                         goto lost_parent;
2508
2509                 if (unlikely(rc == -EEXIST)) {
2510                         /* Unfortunately, someone reused the name under the
2511                          * parent by race. So we have to remove the linkEA
2512                          * entry from current child object. It means that the
2513                          * LFSCK cannot recover the system totally back to
2514                          * its original status, but it is necessary to make
2515                          * the current system to be consistent.
2516                          *
2517                          * It also may be because of the LFSCK found some
2518                          * internal status of create operation. Under such
2519                          * case, nothing to be done. */
2520                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2521                                         parent, child, ldata, cname, tfid);
2522                         if (rc >= 0) {
2523                                 snprintf(info->lti_tmpbuf,
2524                                          sizeof(info->lti_tmpbuf),
2525                                          "-"DFID, PFID(pfid));
2526                                 rc = lfsck_namespace_insert_orphan(env, com,
2527                                         child, info->lti_tmpbuf, "D", NULL);
2528                         }
2529                 }
2530
2531                 GOTO(out, rc);
2532         }
2533
2534         if (rc != 0)
2535                 GOTO(out, rc);
2536
2537         if (!lu_fid_eq(tfid, cfid)) {
2538                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2539                         *retry = true;
2540
2541                         GOTO(out, rc = 0);
2542                 }
2543
2544                 lfsck_ibits_unlock(lh, LCK_EX);
2545                 /* The name entry references another MDT-object that
2546                  * may be created by the LFSCK for repairing dangling
2547                  * name entry. Try to replace it. */
2548                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2549                                                   tfid, cname);
2550                 if (rc == 0)
2551                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2552                                                         pfid, lh, type);
2553
2554                 GOTO(out, rc);
2555         }
2556
2557         if (fid_is_zero(pfid))
2558                 GOTO(out, rc = 0);
2559
2560         /* The ".." name entry is wrong, update it. */
2561         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2562                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2563                         *retry = true;
2564
2565                         GOTO(out, rc = 0);
2566                 }
2567
2568                 *type = LNIT_UNMATCHED_PAIRS;
2569                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2570                                                 lfsck_dto2fid(parent), cname);
2571         }
2572
2573         GOTO(out, rc);
2574
2575 out:
2576         if (parent != NULL && !IS_ERR(parent))
2577                 lfsck_object_put(env, parent);
2578
2579         return rc;
2580 }
2581
2582 /**
2583  * Double Scan Directory object for multiple linkEA entries case.
2584  *
2585  * The given @child has multiple linkEA entries. There is at most one linkEA
2586  * entry will be valid, all the others will be removed. Firstly, the function
2587  * will try to find out the linkEA entry for which the name entry exists under
2588  * the given parent (@pfid). If there is no linkEA entry that matches the given
2589  * ".." name entry, then tries to find out the first linkEA entry that both the
2590  * parent and the name entry exist to rebuild a new ".." name entry.
2591  *
2592  * \param[in] env       pointer to the thread context
2593  * \param[in] com       pointer to the lfsck component
2594  * \param[in] child     pointer to the directory to be double scanned
2595  * \param[in] pfid      the FID corresponding to the ".." entry
2596  * \param[in] ldata     pointer to the linkEA data for the given @child
2597  * \param[in,out] lh    ldlm lock handler for the given @child
2598  * \param[out] type     to tell the caller what the inconsistency is
2599  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2600  *
2601  * \retval              positive number for repaired cases
2602  * \retval              0 if nothing to be repaired
2603  * \retval              negative error number on failure
2604  */
2605 static int
2606 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2607                              struct lfsck_component *com,
2608                              struct dt_object *child,
2609                              const struct lu_fid *pfid,
2610                              struct linkea_data *ldata,
2611                              struct lustre_handle *lh,
2612                              enum lfsck_namespace_inconsistency_type *type,
2613                              bool lpf)
2614 {
2615         struct lfsck_thread_info *info          = lfsck_env_info(env);
2616         struct lu_name           *cname         = &info->lti_name;
2617         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2618         struct lu_fid            *tfid          = &info->lti_fid3;
2619         struct lu_fid            *pfid2         = &info->lti_fid4;
2620         struct lfsck_namespace   *ns            = com->lc_file_ram;
2621         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2622         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2623         struct dt_object         *parent        = NULL;
2624         struct linkea_data        ldata_new     = { 0 };
2625         int                       dirent_count  = 0;
2626         int                       linkea_count  = 0;
2627         int                       rc            = 0;
2628         bool                      once          = true;
2629         ENTRY;
2630
2631 again:
2632         while (ldata->ld_lee != NULL) {
2633                 lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid,
2634                                                     info->lti_key);
2635                 /* Drop repeated linkEA entries. */
2636                 lfsck_namespace_filter_linkea_entry(ldata, cname, tfid, true);
2637                 /* Drop invalid linkEA entry. */
2638                 if (!fid_is_sane(tfid)) {
2639                         linkea_del_buf(ldata, cname);
2640                         linkea_count++;
2641                         continue;
2642                 }
2643
2644                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2645                  * then it is possible that: the directry object has ever
2646                  * been lost, but its name entry was there. In the former
2647                  * LFSCK run, during the first-stage scanning, the LFSCK
2648                  * found the dangling name entry, but it did not recreate
2649                  * the lost object, and when moved to the second-stage
2650                  * scanning, some children objects of the lost directory
2651                  * object were found, then the LFSCK recreated such lost
2652                  * directory object as an orphan.
2653                  *
2654                  * When the LFSCK runs again, if the dangling name is still
2655                  * there, the LFSCK should move the orphan directory object
2656                  * back to the normal namespace. */
2657                 if (!lpf && !lu_fid_eq(pfid, tfid) && once) {
2658                         linkea_next_entry(ldata);
2659                         continue;
2660                 }
2661
2662                 parent = lfsck_object_find_bottom(env, lfsck, tfid);
2663                 if (IS_ERR(parent))
2664                         RETURN(PTR_ERR(parent));
2665
2666                 if (!dt_object_exists(parent)) {
2667                         lfsck_object_put(env, parent);
2668                         if (ldata->ld_leh->leh_reccount > 1) {
2669                                 /* If it is NOT the last linkEA entry, then
2670                                  * there is still other chance to make the
2671                                  * child to be visible via other parent, then
2672                                  * remove this linkEA entry. */
2673                                 linkea_del_buf(ldata, cname);
2674                                 linkea_count++;
2675                                 continue;
2676                         }
2677
2678                         break;
2679                 }
2680
2681                 /* The linkEA entry with bad parent will be removed. */
2682                 if (unlikely(!dt_try_as_dir(env, parent))) {
2683                         lfsck_object_put(env, parent);
2684                         linkea_del_buf(ldata, cname);
2685                         linkea_count++;
2686                         continue;
2687                 }
2688
2689                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2690                                (const struct dt_key *)cname->ln_name,
2691                                BYPASS_CAPA);
2692                 *pfid2 = *lfsck_dto2fid(parent);
2693                 if (rc == -ENOENT) {
2694                         lfsck_object_put(env, parent);
2695                         linkea_next_entry(ldata);
2696                         continue;
2697                 }
2698
2699                 if (rc != 0) {
2700                         lfsck_object_put(env, parent);
2701
2702                         RETURN(rc);
2703                 }
2704
2705                 if (lu_fid_eq(tfid, cfid)) {
2706                         lfsck_object_put(env, parent);
2707                         if (!lu_fid_eq(pfid, pfid2)) {
2708                                 *type = LNIT_UNMATCHED_PAIRS;
2709                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2710                                                 com, child, pfid2, cname);
2711
2712                                 RETURN(rc);
2713                         }
2714
2715 rebuild:
2716                         /* It is the most common case that we find the
2717                          * name entry corresponding to the linkEA entry
2718                          * that matches the ".." name entry. */
2719                         rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2720                         if (rc != 0)
2721                                 RETURN(rc);
2722
2723                         rc = linkea_add_buf(&ldata_new, cname, pfid2);
2724                         if (rc != 0)
2725                                 RETURN(rc);
2726
2727                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2728                                                             &ldata_new);
2729                         if (rc < 0)
2730                                 RETURN(rc);
2731
2732                         linkea_del_buf(ldata, cname);
2733                         linkea_count++;
2734                         linkea_first_entry(ldata);
2735                         /* There may be some invalid dangling name entries under
2736                          * other parent directories, remove all of them. */
2737                         while (ldata->ld_lee != NULL) {
2738                                 lfsck_namespace_unpack_linkea_entry(ldata,
2739                                                 cname, tfid, info->lti_key);
2740                                 if (!fid_is_sane(tfid))
2741                                         goto next;
2742
2743                                 parent = lfsck_object_find_bottom(env, lfsck,
2744                                                                   tfid);
2745                                 if (IS_ERR(parent)) {
2746                                         rc = PTR_ERR(parent);
2747                                         if (rc != -ENOENT &&
2748                                             bk->lb_param & LPF_FAILOUT)
2749                                                 RETURN(rc);
2750
2751                                         goto next;
2752                                 }
2753
2754                                 if (!dt_object_exists(parent)) {
2755                                         lfsck_object_put(env, parent);
2756                                         goto next;
2757                                 }
2758
2759                                 rc = lfsck_namespace_repair_dirent(env, com,
2760                                         parent, child, cname->ln_name,
2761                                         cname->ln_name, S_IFDIR, false, true);
2762                                 lfsck_object_put(env, parent);
2763                                 if (rc < 0) {
2764                                         if (bk->lb_param & LPF_FAILOUT)
2765                                                 RETURN(rc);
2766
2767                                         goto next;
2768                                 }
2769
2770                                 dirent_count += rc;
2771
2772 next:
2773                                 linkea_del_buf(ldata, cname);
2774                         }
2775
2776                         ns->ln_dirent_repaired += dirent_count;
2777
2778                         RETURN(rc);
2779                 }
2780
2781                 lfsck_ibits_unlock(lh, LCK_EX);
2782                 /* The name entry references another MDT-object that may be
2783                  * created by the LFSCK for repairing dangling name entry.
2784                  * Try to replace it. */
2785                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2786                                                   tfid, cname);
2787                 lfsck_object_put(env, parent);
2788                 if (rc < 0)
2789                         RETURN(rc);
2790
2791                 if (rc > 0)
2792                         goto rebuild;
2793
2794                 linkea_del_buf(ldata, cname);
2795         }
2796
2797         linkea_first_entry(ldata);
2798         if (ldata->ld_leh->leh_reccount == 1) {
2799                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2800                                                 lh, type, NULL);
2801
2802                 if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
2803                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2804                                                             ldata);
2805
2806                 RETURN(rc);
2807         }
2808
2809         /* All linkEA entries are invalid and removed, then handle the @child
2810          * as an orphan.*/
2811         if (ldata->ld_leh->leh_reccount == 0) {
2812                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2813                                                 type);
2814
2815                 RETURN(rc);
2816         }
2817
2818         /* If the dangling name entry for the orphan directory object has
2819          * been remvoed, then just check whether the directory object is
2820          * still under the .lustre/lost+found/MDTxxxx/ or not. */
2821         if (lpf) {
2822                 lpf = false;
2823                 goto again;
2824         }
2825
2826         /* There is no linkEA entry that matches the ".." name entry. Find
2827          * the first linkEA entry that both parent and name entry exist to
2828          * rebuild a new ".." name entry. */
2829         if (once) {
2830                 once = false;
2831                 goto again;
2832         }
2833
2834         RETURN(rc);
2835 }
2836
2837 /**
2838  * Repair the object's nlink attribute.
2839  *
2840  * If all the known name entries have been verified, then the object's hard
2841  * link attribute should match the object's linkEA entries count unless the
2842  * object's has too much hard link to be recorded in the linkEA. Such cases
2843  * should have been marked in the LFSCK trace file. Otherwise, trust the
2844  * linkEA to update the object's nlink attribute.
2845  *
2846  * \param[in] env       pointer to the thread context
2847  * \param[in] com       pointer to the lfsck component
2848  * \param[in] obj       pointer to the dt_object to be handled
2849  * \param[in,out] nlink pointer to buffer to object's hard lock count before
2850  *                      and after the repairing
2851  *
2852  * \retval              positive number for repaired cases
2853  * \retval              0 if nothing to be repaired
2854  * \retval              negative error number on failure
2855  */
2856 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2857                                         struct lfsck_component *com,
2858                                         struct dt_object *obj, __u32 *nlink)
2859 {
2860         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2861         struct lu_attr                  *la     = &info->lti_la3;
2862         struct lu_fid                   *tfid   = &info->lti_fid3;
2863         struct lfsck_namespace          *ns     = com->lc_file_ram;
2864         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2865         struct dt_device                *dev    = lfsck->li_bottom;
2866         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
2867         struct dt_object                *child  = NULL;
2868         struct thandle                  *th     = NULL;
2869         struct linkea_data               ldata  = { 0 };
2870         struct lustre_handle             lh     = { 0 };
2871         __u32                            old    = *nlink;
2872         int                              rc     = 0;
2873         __u8                             flags;
2874         ENTRY;
2875
2876         LASSERT(!dt_object_remote(obj));
2877         LASSERT(S_ISREG(lfsck_object_type(obj)));
2878
2879         child = lfsck_object_find_by_dev(env, dev, cfid);
2880         if (IS_ERR(child))
2881                 GOTO(log, rc = PTR_ERR(child));
2882
2883         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
2884                               MDS_INODELOCK_UPDATE |
2885                               MDS_INODELOCK_XATTR, LCK_EX);
2886         if (rc != 0)
2887                 GOTO(log, rc);
2888
2889         th = dt_trans_create(env, dev);
2890         if (IS_ERR(th))
2891                 GOTO(log, rc = PTR_ERR(th));
2892
2893         la->la_valid = LA_NLINK;
2894         rc = dt_declare_attr_set(env, child, la, th);
2895         if (rc != 0)
2896                 GOTO(stop, rc);
2897
2898         rc = dt_trans_start_local(env, dev, th);
2899         if (rc != 0)
2900                 GOTO(stop, rc);
2901
2902         dt_write_lock(env, child, 0);
2903         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2904          * ever tried to verify some remote MDT-object that resides on this
2905          * MDT, but this MDT failed to respond such request. So means there
2906          * may be some remote name entry on other MDT that references this
2907          * object with another name, so we cannot know whether this linkEA
2908          * is valid or not. So keep it there and maybe resolved when next
2909          * LFSCK run. */
2910         if (ns->ln_flags & LF_INCOMPLETE)
2911                 GOTO(unlock, rc = 0);
2912
2913         fid_cpu_to_be(tfid, cfid);
2914         rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)&flags,
2915                        (const struct dt_key *)tfid, BYPASS_CAPA);
2916         if (rc != 0)
2917                 GOTO(unlock, rc);
2918
2919         if (flags & LNTF_SKIP_NLINK)
2920                 GOTO(unlock, rc = 0);
2921
2922         rc = lfsck_links_read2(env, child, &ldata);
2923         if (rc == -ENODATA)
2924                 GOTO(unlock, rc = 0);
2925
2926         if (rc != 0)
2927                 GOTO(unlock, rc);
2928
2929         if (*nlink == ldata.ld_leh->leh_reccount)
2930                 GOTO(unlock, rc = 0);
2931
2932         la->la_nlink = *nlink = ldata.ld_leh->leh_reccount;
2933         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2934                 GOTO(unlock, rc = 1);
2935
2936         rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
2937
2938         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2939
2940 unlock:
2941         dt_write_unlock(env, child);
2942
2943 stop:
2944         dt_trans_stop(env, dev, th);
2945
2946 log:
2947         lfsck_ibits_unlock(&lh, LCK_EX);
2948         if (child != NULL && !IS_ERR(child))
2949                 lfsck_object_put(env, child);
2950
2951         CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2952                "nlink count from %u to %u: rc = %d\n",
2953                lfsck_lfsck2name(lfsck), PFID(cfid), old, *nlink, rc);
2954
2955         if (rc != 0)
2956                 ns->ln_flags |= LF_INCONSISTENT;
2957
2958         return rc;
2959 }
2960
2961 /**
2962  * Double scan the directory object for namespace LFSCK.
2963  *
2964  * This function will verify the <parent, child> pairs in the namespace tree:
2965  * the parent references the child via some name entry that should be in the
2966  * child's linkEA entry, the child should back references the parent via its
2967  * ".." name entry.
2968  *
2969  * The LFSCK will scan every linkEA entry in turn until find out the first
2970  * matched pairs. If found, then all other linkEA entries will be dropped.
2971  * If all the linkEA entries cannot match the ".." name entry, then there
2972  * are serveral possible cases:
2973  *
2974  * 1) If there is only one linkEA entry, then trust it as long as the PFID
2975  *    in the linkEA entry is valid.
2976  *
2977  * 2) If there are multiple linkEA entries, then try to find the linkEA
2978  *    that matches the ".." name entry. If found, then all other entries
2979  *    are invalid; otherwise, it is quite possible that the ".." name entry
2980  *    is corrupted. Under such case, the LFSCK will rebuild the ".." name
2981  *    entry according to the first valid linkEA entry (both the parent and
2982  *    the name entry should exist).
2983  *
2984  * 3) If the directory object has no (valid) linkEA entry, then the
2985  *    directory object will be handled as pure orphan and inserted
2986  *    in the .lustre/lost+found/MDTxxxx/ with the name:
2987  *    ${self_FID}-${PFID}-D-${conflict_version}
2988  *
2989  * \param[in] env       pointer to the thread context
2990  * \param[in] com       pointer to the lfsck component
2991  * \param[in] child     pointer to the directory object to be handled
2992  * \param[in] flags     to indicate the specical checking on the @child
2993  *
2994  * \retval              positive number for repaired cases
2995  * \retval              0 if nothing to be repaired
2996  * \retval              negative error number on failure
2997  */
2998 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
2999                                            struct lfsck_component *com,
3000                                            struct dt_object *child, __u8 flags)
3001 {
3002         struct lfsck_thread_info *info          = lfsck_env_info(env);
3003         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
3004         struct lu_fid            *pfid          = &info->lti_fid2;
3005         struct lfsck_namespace   *ns            = com->lc_file_ram;
3006         struct lfsck_instance    *lfsck         = com->lc_lfsck;
3007         struct lustre_handle      lh            = { 0 };
3008         struct linkea_data        ldata         = { 0 };
3009         bool                      unknown       = false;
3010         bool                      lpf           = false;
3011         bool                      retry         = false;
3012         enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3013         int                       rc            = 0;
3014         ENTRY;
3015
3016         LASSERT(!dt_object_remote(child));
3017
3018         if (flags & LNTF_UNCERTAIN_LMV) {
3019                 if (flags & LNTF_RECHECK_NAME_HASH) {
3020                         rc = lfsck_namespace_scan_shard(env, com, child);
3021                         if (rc < 0)
3022                                 RETURN(rc);
3023
3024                         ns->ln_striped_shards_scanned++;
3025                 } else {
3026                         ns->ln_striped_shards_skipped++;
3027                 }
3028         }
3029
3030         flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3031         if (flags == 0)
3032                 RETURN(0);
3033
3034         if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3035             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3036                 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3037                        "the namespace LFSCK, then the LFSCK cannot guarantee"
3038                        "all the name entries have been verified in first-stage"
3039                        "scanning. So have to skip orphan related handling for"
3040                        "the directory object "DFID" with remote name entry\n",
3041                        lfsck_lfsck2name(lfsck), PFID(cfid));
3042
3043                 RETURN(0);
3044         }
3045
3046         if (unlikely(!dt_try_as_dir(env, child)))
3047                 GOTO(out, rc = -ENOTDIR);
3048
3049         /* We only take ldlm lock on the @child when required. When the
3050          * logic comes here for the first time, it is always false. */
3051         if (0) {
3052
3053 lock:
3054                 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3055                                       MDS_INODELOCK_UPDATE |
3056                                       MDS_INODELOCK_XATTR, LCK_EX);
3057                 if (rc != 0)
3058                         GOTO(out, rc);
3059         }
3060
3061         dt_read_lock(env, child, 0);
3062         if (unlikely(lfsck_is_dead_obj(child))) {
3063                 dt_read_unlock(env, child);
3064
3065                 GOTO(out, rc = 0);
3066         }
3067
3068         rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3069                        (const struct dt_key *)dotdot, BYPASS_CAPA);
3070         if (rc != 0) {
3071                 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3072                         dt_read_unlock(env, child);
3073
3074                         GOTO(out, rc);
3075                 }
3076
3077                 if (!lustre_handle_is_used(&lh)) {
3078                         dt_read_unlock(env, child);
3079                         goto lock;
3080                 }
3081
3082                 fid_zero(pfid);
3083         } else if (lfsck->li_lpf_obj != NULL &&
3084                    lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3085                 lpf = true;
3086         } else if (unlikely(!fid_is_sane(pfid))) {
3087                 fid_zero(pfid);
3088         }
3089
3090         rc = lfsck_links_read(env, child, &ldata);
3091         dt_read_unlock(env, child);
3092         if (rc != 0) {
3093                 if (rc != -ENODATA && rc != -EINVAL)
3094                         GOTO(out, rc);
3095
3096                 if (!lustre_handle_is_used(&lh))
3097                         goto lock;
3098
3099                 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3100                         /* Remove the corrupted linkEA. */
3101                         rc = lfsck_namespace_links_remove(env, com, child);
3102                         if (rc == 0)
3103                                 /* Here, because of the crashed linkEA, we
3104                                  * cannot know whether there is some parent
3105                                  * that references the child directory via
3106                                  * some name entry or not. So keep it there,
3107                                  * when the LFSCK run next time, if there is
3108                                  * some parent that references this object,
3109                                  * then the LFSCK can rebuild the linkEA;
3110                                  * otherwise, this object will be handled
3111                                  * as orphan as above. */
3112                                 unknown = true;
3113                 } else {
3114                         /* 1. If we have neither ".." nor linkEA,
3115                          *    then it is an orphan.
3116                          *
3117                          * 2. If we only have the ".." name entry,
3118                          *    but no parent references this child
3119                          *    directory, then handle it as orphan. */
3120                         lfsck_ibits_unlock(&lh, LCK_EX);
3121                         type = LNIT_MUL_REF;
3122
3123                         /* If the LFSCK is marked as LF_INCOMPLETE,
3124                          * then means some MDT has ever tried to
3125                          * verify some remote MDT-object that resides
3126                          * on this MDT, but this MDT failed to respond
3127                          * such request. So means there may be some
3128                          * remote name entry on other MDT that
3129                          * references this object with another name,
3130                          * so we cannot know whether this linkEA is
3131                          * valid or not. So keep it there and maybe
3132                          * resolved when next LFSCK run. */
3133                         if (ns->ln_flags & LF_INCOMPLETE)
3134                                 GOTO(out, rc = 0);
3135
3136                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3137                                  "-"DFID, PFID(pfid));
3138                         rc = lfsck_namespace_insert_orphan(env, com, child,
3139                                                 info->lti_tmpbuf, "D", NULL);
3140                 }
3141
3142                 GOTO(out, rc);
3143         }
3144
3145         linkea_first_entry(&ldata);
3146         /* This is the most common case: the object has unique linkEA entry. */
3147         if (ldata.ld_leh->leh_reccount == 1) {
3148                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3149                                                 &lh, &type, &retry);
3150                 if (retry) {
3151                         LASSERT(!lustre_handle_is_used(&lh));
3152
3153                         retry = false;
3154                         goto lock;
3155                 }
3156
3157                 GOTO(out, rc);
3158         }
3159
3160         if (!lustre_handle_is_used(&lh))
3161                 goto lock;
3162
3163         if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3164                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3165                                                 &type);
3166
3167                 GOTO(out, rc);
3168         }
3169
3170         /* When we come here, the cases usually like that:
3171          * 1) The directory object has a corrupted linkEA entry. During the
3172          *    first-stage scanning, the LFSCK cannot know such corruption,
3173          *    then it appends the right linkEA entry according to the found
3174          *    name entry after the bad one.
3175          *
3176          * 2) The directory object has a right linkEA entry. During the
3177          *    first-stage scanning, the LFSCK finds some bad name entry,
3178          *    but the LFSCK cannot aware that at that time, then it adds
3179          *    the bad linkEA entry for further processing. */
3180         rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3181                                           &lh, &type, lpf);
3182
3183         GOTO(out, rc);
3184
3185 out:
3186         lfsck_ibits_unlock(&lh, LCK_EX);
3187         if (rc > 0) {
3188                 switch (type) {
3189                 case LNIT_BAD_LINKEA:
3190                         ns->ln_linkea_repaired++;
3191                         break;
3192                 case LNIT_UNMATCHED_PAIRS:
3193                         ns->ln_unmatched_pairs_repaired++;
3194                         break;
3195                 case LNIT_MUL_REF:
3196                         ns->ln_mul_ref_repaired++;
3197                         break;
3198                 default:
3199                         break;
3200                 }
3201         }
3202
3203         if (unknown)
3204                 ns->ln_unknown_inconsistency++;
3205
3206         return rc;
3207 }
3208
3209 /**
3210  * Double scan the MDT-object for namespace LFSCK.
3211  *
3212  * If the MDT-object contains invalid or repeated linkEA entries, then drop
3213  * those entries from the linkEA; if the linkEA becomes empty or the object
3214  * has no linkEA, then it is an orphan and will be added into the directory
3215  * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3216  * the remote parent; if the name entry corresponding to some linkEA entry
3217  * is lost, then add the name entry back to the namespace.
3218  *
3219  * \param[in] env       pointer to the thread context
3220  * \param[in] com       pointer to the lfsck component
3221  * \param[in] child     pointer to the dt_object to be handled
3222  * \param[in] flags     some hints to indicate how the @child should be handled
3223  *
3224  * \retval              positive number for repaired cases
3225  * \retval              0 if nothing to be repaired
3226  * \retval              negative error number on failure
3227  */
3228 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3229                                            struct lfsck_component *com,
3230                                            struct dt_object *child, __u8 flags)
3231 {
3232         struct lfsck_thread_info *info     = lfsck_env_info(env);
3233         struct lu_attr           *la       = &info->lti_la;
3234         struct lu_name           *cname    = &info->lti_name;
3235         struct lu_fid            *pfid     = &info->lti_fid;
3236         struct lu_fid            *cfid     = &info->lti_fid2;
3237         struct lfsck_instance    *lfsck    = com->lc_lfsck;
3238         struct lfsck_namespace   *ns       = com->lc_file_ram;
3239         struct dt_object         *parent   = NULL;
3240         struct linkea_data        ldata    = { 0 };
3241         bool                      repaired = false;
3242         int                       count    = 0;
3243         int                       rc;
3244         ENTRY;
3245
3246         dt_read_lock(env, child, 0);
3247         if (unlikely(lfsck_is_dead_obj(child))) {
3248                 dt_read_unlock(env, child);
3249
3250                 RETURN(0);
3251         }
3252
3253         if (S_ISDIR(lfsck_object_type(child))) {
3254                 dt_read_unlock(env, child);
3255                 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3256
3257                 RETURN(rc);
3258         }
3259
3260         rc = lfsck_links_read(env, child, &ldata);
3261         dt_read_unlock(env, child);
3262         if (rc != 0)
3263                 GOTO(out, rc);
3264
3265         linkea_first_entry(&ldata);
3266         while (ldata.ld_lee != NULL) {
3267                 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3268                                                     info->lti_key);
3269                 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3270                                                          false);
3271                 /* Found repeated linkEA entries */
3272                 if (rc > 0) {
3273                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3274                                                 &ldata, cname, pfid, false);
3275                         if (rc < 0)
3276                                 GOTO(out, rc);
3277
3278                         if (rc == 0)
3279                                 continue;
3280
3281                         repaired = true;
3282
3283                         /* fall through */
3284                 }
3285
3286                 /* Invalid PFID in the linkEA entry. */
3287                 if (!fid_is_sane(pfid)) {
3288                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3289                                                 &ldata, cname, pfid, true);
3290                         if (rc < 0)
3291                                 GOTO(out, rc);
3292
3293                         if (rc > 0)
3294                                 repaired = true;
3295
3296                         continue;
3297                 }
3298
3299                 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3300                 if (IS_ERR(parent))
3301                         GOTO(out, rc = PTR_ERR(parent));
3302
3303                 if (!dt_object_exists(parent)) {
3304
3305 lost_parent:
3306                         if (ldata.ld_leh->leh_reccount > 1) {
3307                                 /* If it is NOT the last linkEA entry, then
3308                                  * there is still other chance to make the
3309                                  * child to be visible via other parent, then
3310                                  * remove this linkEA entry. */
3311                                 rc = lfsck_namespace_shrink_linkea(env, com,
3312                                         child, &ldata, cname, pfid, true);
3313                         } else {
3314                                 /* If the LFSCK is marked as LF_INCOMPLETE,
3315                                  * then means some MDT has ever tried to
3316                                  * verify some remote MDT-object that resides
3317                                  * on this MDT, but this MDT failed to respond
3318                                  * such request. So means there may be some
3319                                  * remote name entry on other MDT that
3320                                  * references this object with another name,
3321                                  * so we cannot know whether this linkEA is
3322                                  * valid or not. So keep it there and maybe
3323                                  * resolved when next LFSCK run. */
3324                                 if (ns->ln_flags & LF_INCOMPLETE) {
3325                                         lfsck_object_put(env, parent);
3326
3327                                         GOTO(out, rc = 0);
3328                                 }
3329
3330                                 /* Create the lost parent as an orphan. */
3331                                 rc = lfsck_namespace_create_orphan(env, com,
3332                                                                 parent, NULL);
3333                                 if (rc < 0) {
3334                                         lfsck_object_put(env, parent);
3335
3336                                         GOTO(out, rc);
3337                                 }
3338
3339                                 if (rc > 0)
3340                                         repaired = true;
3341
3342                                 /* Add the missing name entry to the parent. */
3343                                 rc = lfsck_namespace_insert_normal(env, com,
3344                                                 parent, child, cname->ln_name);
3345                                 if (unlikely(rc == -EEXIST))
3346                                         /* Unfortunately, someone reused the
3347                                          * name under the parent by race. So we
3348                                          * have to remove the linkEA entry from
3349                                          * current child object. It means that
3350                                          * the LFSCK cannot recover the system
3351                                          * totally back to its original status,
3352                                          * but it is necessary to make the
3353                                          * current system to be consistent. */
3354                                         rc = lfsck_namespace_shrink_linkea(env,
3355                                                         com, child, &ldata,
3356                                                         cname, pfid, true);
3357                                 else
3358                                         linkea_next_entry(&ldata);
3359                         }
3360
3361                         lfsck_object_put(env, parent);
3362                         if (rc < 0)
3363                                 GOTO(out, rc);
3364
3365                         if (rc > 0)
3366                                 repaired = true;
3367
3368                         continue;
3369                 }
3370
3371                 /* The linkEA entry with bad parent will be removed. */
3372                 if (unlikely(!dt_try_as_dir(env, parent))) {
3373                         lfsck_object_put(env, parent);
3374                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3375                                                 &ldata, cname, pfid, true);
3376                         if (rc < 0)
3377                                 GOTO(out, rc);
3378
3379                         if (rc > 0)
3380                                 repaired = true;
3381
3382                         continue;
3383                 }
3384
3385                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3386                                (const struct dt_key *)cname->ln_name,
3387                                BYPASS_CAPA);
3388                 if (rc != 0 && rc != -ENOENT) {
3389                         lfsck_object_put(env, parent);
3390
3391                         GOTO(out, rc);
3392                 }
3393
3394                 if (rc == 0) {
3395                         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3396                                 /* It is the most common case that we
3397                                  * find the name entry corresponding
3398                                  * to the linkEA entry. */
3399                                 lfsck_object_put(env, parent);
3400                                 linkea_next_entry(&ldata);
3401                         } else {
3402                                 /* The name entry references another
3403                                  * MDT-object that may be created by
3404                                  * the LFSCK for repairing dangling
3405                                  * name entry. Try to replace it. */
3406                                 rc = lfsck_namespace_replace_cond(env, com,
3407                                                 parent, child, cfid, cname);
3408                                 lfsck_object_put(env, parent);
3409                                 if (rc < 0)
3410                                         GOTO(out, rc);
3411
3412                                 if (rc > 0) {
3413                                         repaired = true;
3414                                         linkea_next_entry(&ldata);
3415                                 } else {
3416                                         rc = lfsck_namespace_shrink_linkea(env,
3417                                                         com, child, &ldata,
3418                                                         cname, pfid, true);
3419                                         if (rc < 0)
3420                                                 GOTO(out, rc);
3421
3422                                         if (rc > 0)
3423                                                 repaired = true;
3424                                 }
3425                         }
3426
3427                         continue;
3428                 }
3429
3430                 rc = dt_at