lustre/lfsck/lfsck_namespace.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2017, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_namespace.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lu_object.h>
  34 #include <dt_object.h>
  35 #include <md_object.h>
  36 #include <lustre_fid.h>
  37 #include <lustre_lib.h>
  38 #include <lustre_net.h>
  39
  40 #include "lfsck_internal.h"
  41
  42 #define LFSCK_NAMESPACE_MAGIC_V1        0xA0629D03
  43 #define LFSCK_NAMESPACE_MAGIC_V2        0xA0621A0B
  44 #define LFSCK_NAMESPACE_MAGIC_V3        0xA06249FF
  45
  46 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
  47  * as the trace file magic. When downgrade to such old release, the old LFSCK
  48  * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
  49  * then it will reset the whole LFSCK, and will not cause start failure. The
  50  * similar case will happen when upgrade from such old release. */
  51 #define LFSCK_NAMESPACE_MAGIC           LFSCK_NAMESPACE_MAGIC_V3
  52
  53 enum lfsck_nameentry_check {
  54         LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
  55         LFSCK_NAMEENTRY_REMOVED         = 2, /* The entry has been removed. */
  56         LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
  57 };
  58
  59 static struct lfsck_namespace_req *
  60 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
  61                                    struct lfsck_assistant_object *lso,
  62                                    struct lu_dirent *ent, __u16 type)
  63 {
  64         struct lfsck_namespace_req *lnr;
  65         int                         size;
  66
  67         size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
  68         OBD_ALLOC(lnr, size);
  69         if (lnr == NULL)
  70                 return ERR_PTR(-ENOMEM);
  71
  72         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
  73         lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
  74         lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
  75         lnr->lnr_fid = ent->lde_fid;
  76         lnr->lnr_dir_cookie = ent->lde_hash;
  77         lnr->lnr_attr = ent->lde_attrs;
  78         lnr->lnr_size = size;
  79         lnr->lnr_type = type;
  80         lnr->lnr_namelen = ent->lde_namelen;
  81         memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
  82
  83         return lnr;
  84 }
  85
  86 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
  87                                                struct lfsck_assistant_req *lar)
  88 {
  89         struct lfsck_namespace_req *lnr =
  90                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
  91
  92         if (lnr->lnr_lmv != NULL)
  93                 lfsck_lmv_put(env, lnr->lnr_lmv);
  94
  95         lfsck_assistant_object_put(env, lar->lar_parent);
  96         OBD_FREE(lnr, lnr->lnr_size);
  97 }
  98
  99 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
 100                                       struct lfsck_namespace *src)
 101 {
 102         dst->ln_magic = le32_to_cpu(src->ln_magic);
 103         dst->ln_status = le32_to_cpu(src->ln_status);
 104         dst->ln_flags = le32_to_cpu(src->ln_flags);
 105         dst->ln_success_count = le32_to_cpu(src->ln_success_count);
 106         dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
 107         dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
 108         dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
 109         dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
 110         dst->ln_time_last_checkpoint =
 111                                 le64_to_cpu(src->ln_time_last_checkpoint);
 112         lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
 113                                  &src->ln_pos_latest_start);
 114         lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
 115                                  &src->ln_pos_last_checkpoint);
 116         lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
 117                                  &src->ln_pos_first_inconsistent);
 118         dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
 119         dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
 120         dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
 121         dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
 122         dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
 123         dst->ln_objs_repaired_phase2 =
 124                                 le64_to_cpu(src->ln_objs_repaired_phase2);
 125         dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
 126         dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
 127         fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
 128                       &src->ln_fid_latest_scanned_phase2);
 129         dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
 130         dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
 131         dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
 132         dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
 133         dst->ln_unknown_inconsistency =
 134                                 le64_to_cpu(src->ln_unknown_inconsistency);
 135         dst->ln_unmatched_pairs_repaired =
 136                                 le64_to_cpu(src->ln_unmatched_pairs_repaired);
 137         dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
 138         dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
 139         dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
 140         dst->ln_lost_dirent_repaired =
 141                                 le64_to_cpu(src->ln_lost_dirent_repaired);
 142         dst->ln_striped_dirs_scanned =
 143                                 le64_to_cpu(src->ln_striped_dirs_scanned);
 144         dst->ln_striped_dirs_repaired =
 145                                 le64_to_cpu(src->ln_striped_dirs_repaired);
 146         dst->ln_striped_dirs_failed =
 147                                 le64_to_cpu(src->ln_striped_dirs_failed);
 148         dst->ln_striped_dirs_disabled =
 149                                 le64_to_cpu(src->ln_striped_dirs_disabled);
 150         dst->ln_striped_dirs_skipped =
 151                                 le64_to_cpu(src->ln_striped_dirs_skipped);
 152         dst->ln_striped_shards_scanned =
 153                                 le64_to_cpu(src->ln_striped_shards_scanned);
 154         dst->ln_striped_shards_repaired =
 155                                 le64_to_cpu(src->ln_striped_shards_repaired);
 156         dst->ln_striped_shards_failed =
 157                                 le64_to_cpu(src->ln_striped_shards_failed);
 158         dst->ln_striped_shards_skipped =
 159                                 le64_to_cpu(src->ln_striped_shards_skipped);
 160         dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
 161         dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
 162         dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
 163         dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
 164         dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
 165         dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
 166         dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
 167         dst->ln_linkea_overflow_cleared =
 168                                 le64_to_cpu(src->ln_linkea_overflow_cleared);
 169         dst->ln_agent_entries_repaired =
 170                                 le64_to_cpu(src->ln_agent_entries_repaired);
 171 }
 172
 173 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
 174                                       struct lfsck_namespace *src)
 175 {
 176         dst->ln_magic = cpu_to_le32(src->ln_magic);
 177         dst->ln_status = cpu_to_le32(src->ln_status);
 178         dst->ln_flags = cpu_to_le32(src->ln_flags);
 179         dst->ln_success_count = cpu_to_le32(src->ln_success_count);
 180         dst->ln_run_time_phase1 = cpu_to_le64(src->ln_run_time_phase1);
 181         dst->ln_run_time_phase2 = cpu_to_le64(src->ln_run_time_phase2);
 182         dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
 183         dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
 184         dst->ln_time_last_checkpoint =
 185                                 cpu_to_le64(src->ln_time_last_checkpoint);
 186         lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
 187                                  &src->ln_pos_latest_start);
 188         lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
 189                                  &src->ln_pos_last_checkpoint);
 190         lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
 191                                  &src->ln_pos_first_inconsistent);
 192         dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
 193         dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
 194         dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
 195         dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
 196         dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
 197         dst->ln_objs_repaired_phase2 =
 198                                 cpu_to_le64(src->ln_objs_repaired_phase2);
 199         dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
 200         dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
 201         fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
 202                       &src->ln_fid_latest_scanned_phase2);
 203         dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
 204         dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
 205         dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
 206         dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
 207         dst->ln_unknown_inconsistency =
 208                                 cpu_to_le64(src->ln_unknown_inconsistency);
 209         dst->ln_unmatched_pairs_repaired =
 210                                 cpu_to_le64(src->ln_unmatched_pairs_repaired);
 211         dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
 212         dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
 213         dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
 214         dst->ln_lost_dirent_repaired =
 215                                 cpu_to_le64(src->ln_lost_dirent_repaired);
 216         dst->ln_striped_dirs_scanned =
 217                                 cpu_to_le64(src->ln_striped_dirs_scanned);
 218         dst->ln_striped_dirs_repaired =
 219                                 cpu_to_le64(src->ln_striped_dirs_repaired);
 220         dst->ln_striped_dirs_failed =
 221                                 cpu_to_le64(src->ln_striped_dirs_failed);
 222         dst->ln_striped_dirs_disabled =
 223                                 cpu_to_le64(src->ln_striped_dirs_disabled);
 224         dst->ln_striped_dirs_skipped =
 225                                 cpu_to_le64(src->ln_striped_dirs_skipped);
 226         dst->ln_striped_shards_scanned =
 227                                 cpu_to_le64(src->ln_striped_shards_scanned);
 228         dst->ln_striped_shards_repaired =
 229                                 cpu_to_le64(src->ln_striped_shards_repaired);
 230         dst->ln_striped_shards_failed =
 231                                 cpu_to_le64(src->ln_striped_shards_failed);
 232         dst->ln_striped_shards_skipped =
 233                                 cpu_to_le64(src->ln_striped_shards_skipped);
 234         dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
 235         dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
 236         dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
 237         dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
 238         dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
 239         dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
 240         dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
 241         dst->ln_linkea_overflow_cleared =
 242                                 cpu_to_le64(src->ln_linkea_overflow_cleared);
 243         dst->ln_agent_entries_repaired =
 244                                 cpu_to_le64(src->ln_agent_entries_repaired);
 245 }
 246
 247 static void lfsck_namespace_record_failure(const struct lu_env *env,
 248                                            struct lfsck_instance *lfsck,
 249                                            struct lfsck_namespace *ns)
 250 {
 251         struct lfsck_position pos;
 252
 253         ns->ln_items_failed++;
 254         lfsck_pos_fill(env, lfsck, &pos, false);
 255         if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
 256             lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
 257                 ns->ln_pos_first_inconsistent = pos;
 258
 259                 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
 260                        "inconsistency at the pos [%llu, "DFID", %#llx]\n",
 261                        lfsck_lfsck2name(lfsck),
 262                        ns->ln_pos_first_inconsistent.lp_oit_cookie,
 263                        PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
 264                        ns->ln_pos_first_inconsistent.lp_dir_cookie);
 265         }
 266 }
 267
 268 /**
 269  * Load the MDT bitmap from the lfsck_namespace trace file.
 270  *
 271  * \param[in] env       pointer to the thread context
 272  * \param[in] com       pointer to the lfsck component
 273  *
 274  * \retval              0 for success
 275  * \retval              negative error number on failure or data corruption
 276  */
 277 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
 278                                        struct lfsck_component *com)
 279 {
 280         struct dt_object                *obj    = com->lc_obj;
 281         struct lfsck_assistant_data     *lad    = com->lc_data;
 282         struct lfsck_namespace          *ns     = com->lc_file_ram;
 283         struct cfs_bitmap                       *bitmap = lad->lad_bitmap;
 284         ssize_t                          size;
 285         __u32                            nbits;
 286         int                              rc;
 287         ENTRY;
 288
 289         if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
 290             ns->ln_bitmap_size)
 291                 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
 292         else
 293                 nbits = ns->ln_bitmap_size;
 294
 295         if (unlikely(nbits < BITS_PER_LONG))
 296                 nbits = BITS_PER_LONG;
 297
 298         if (nbits > bitmap->size) {
 299                 __u32 new_bits = bitmap->size;
 300                 struct cfs_bitmap *new_bitmap;
 301
 302                 while (new_bits < nbits)
 303                         new_bits <<= 1;
 304
 305                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 306                 if (new_bitmap == NULL)
 307                         RETURN(-ENOMEM);
 308
 309                 lad->lad_bitmap = new_bitmap;
 310                 CFS_FREE_BITMAP(bitmap);
 311                 bitmap = new_bitmap;
 312         }
 313
 314         if (ns->ln_bitmap_size == 0) {
 315                 lad->lad_incomplete = 0;
 316                 CFS_RESET_BITMAP(bitmap);
 317
 318                 RETURN(0);
 319         }
 320
 321         size = (ns->ln_bitmap_size + 7) >> 3;
 322         rc = dt_xattr_get(env, obj,
 323                           lfsck_buf_get(env, bitmap->data, size),
 324                           XATTR_NAME_LFSCK_BITMAP);
 325         if (rc != size)
 326                 RETURN(rc >= 0 ? -EINVAL : rc);
 327
 328         if (cfs_bitmap_check_empty(bitmap))
 329                 lad->lad_incomplete = 0;
 330         else
 331                 lad->lad_incomplete = 1;
 332
 333         RETURN(0);
 334 }
 335
 336 /**
 337  * Load namespace LFSCK statistics information from the trace file.
 338  *
 339  * \param[in] env       pointer to the thread context
 340  * \param[in] com       pointer to the lfsck component
 341  *
 342  * \retval              0 for success
 343  * \retval              negative error number on failure
 344  */
 345 static int lfsck_namespace_load(const struct lu_env *env,
 346                                 struct lfsck_component *com)
 347 {
 348         int len = com->lc_file_size;
 349         int rc;
 350
 351         rc = dt_xattr_get(env, com->lc_obj,
 352                           lfsck_buf_get(env, com->lc_file_disk, len),
 353                           XATTR_NAME_LFSCK_NAMESPACE);
 354         if (rc == len) {
 355                 struct lfsck_namespace *ns = com->lc_file_ram;
 356
 357                 lfsck_namespace_le_to_cpu(ns,
 358                                 (struct lfsck_namespace *)com->lc_file_disk);
 359                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
 360                         CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
 361                                "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
 362                                ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
 363                         rc = -ESTALE;
 364                 } else {
 365                         rc = 0;
 366                 }
 367         } else if (rc != -ENODATA) {
 368                 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
 369                        "expected = %d: rc = %d\n",
 370                        lfsck_lfsck2name(com->lc_lfsck), len, rc);
 371                 if (rc >= 0)
 372                         rc = -ESTALE;
 373         }
 374
 375         return rc;
 376 }
 377
 378 static int lfsck_namespace_store(const struct lu_env *env,
 379                                  struct lfsck_component *com)
 380 {
 381         struct dt_object                *obj    = com->lc_obj;
 382         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 383         struct lfsck_namespace          *ns     = com->lc_file_ram;
 384         struct lfsck_assistant_data     *lad    = com->lc_data;
 385         struct dt_device                *dev    = lfsck_obj2dev(obj);
 386         struct cfs_bitmap               *bitmap = NULL;
 387         struct thandle                  *handle;
 388         __u32                            nbits  = 0;
 389         int                              len    = com->lc_file_size;
 390         int                              rc;
 391         ENTRY;
 392
 393         if (lad != NULL) {
 394                 bitmap = lad->lad_bitmap;
 395                 nbits = bitmap->size;
 396
 397                 LASSERT(nbits > 0);
 398                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 399         }
 400
 401         ns->ln_bitmap_size = nbits;
 402         lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
 403                                   ns);
 404         handle = dt_trans_create(env, dev);
 405         if (IS_ERR(handle))
 406                 GOTO(log, rc = PTR_ERR(handle));
 407
 408         rc = dt_declare_xattr_set(env, obj,
 409                                   lfsck_buf_get(env, com->lc_file_disk, len),
 410                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 411         if (rc != 0)
 412                 GOTO(out, rc);
 413
 414         if (bitmap != NULL) {
 415                 rc = dt_declare_xattr_set(env, obj,
 416                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 417                                 XATTR_NAME_LFSCK_BITMAP, 0, handle);
 418                 if (rc != 0)
 419                         GOTO(out, rc);
 420         }
 421
 422         rc = dt_trans_start_local(env, dev, handle);
 423         if (rc != 0)
 424                 GOTO(out, rc);
 425
 426         rc = dt_xattr_set(env, obj,
 427                           lfsck_buf_get(env, com->lc_file_disk, len),
 428                           XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 429         if (rc == 0 && bitmap != NULL)
 430                 rc = dt_xattr_set(env, obj,
 431                                   lfsck_buf_get(env, bitmap->data, nbits >> 3),
 432                                   XATTR_NAME_LFSCK_BITMAP, 0, handle);
 433
 434         GOTO(out, rc);
 435
 436 out:
 437         dt_trans_stop(env, dev, handle);
 438
 439 log:
 440         if (rc != 0)
 441                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
 442                        lfsck_lfsck2name(lfsck), rc);
 443         return rc;
 444 }
 445
 446 static int lfsck_namespace_init(const struct lu_env *env,
 447                                 struct lfsck_component *com)
 448 {
 449         struct lfsck_namespace *ns = com->lc_file_ram;
 450         int rc;
 451
 452         memset(ns, 0, sizeof(*ns));
 453         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
 454         ns->ln_status = LS_INIT;
 455         ns->ln_time_latest_reset = ktime_get_real_seconds();
 456         down_write(&com->lc_sem);
 457         rc = lfsck_namespace_store(env, com);
 458         if (rc == 0)
 459                 rc = lfsck_load_sub_trace_files(env, com,
 460                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
 461         up_write(&com->lc_sem);
 462
 463         return rc;
 464 }
 465
 466 /**
 467  * Update the namespace LFSCK trace file for the given @fid
 468  *
 469  * \param[in] env       pointer to the thread context
 470  * \param[in] com       pointer to the lfsck component
 471  * \param[in] fid       the fid which flags to be updated in the lfsck
 472  *                      trace file
 473  * \param[in] add       true if add new flags, otherwise remove flags
 474  *
 475  * \retval              0 for success or nothing to be done
 476  * \retval              negative error number on failure
 477  */
 478 int lfsck_namespace_trace_update(const struct lu_env *env,
 479                                  struct lfsck_component *com,
 480                                  const struct lu_fid *fid,
 481                                  const __u8 flags, bool add)
 482 {
 483         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 484         struct dt_object        *obj;
 485         struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
 486         struct dt_device        *dev;
 487         struct thandle          *th     = NULL;
 488         int                      idx;
 489         int                      rc     = 0;
 490         __u8                     old    = 0;
 491         __u8                     new    = 0;
 492         ENTRY;
 493
 494         LASSERT(flags != 0);
 495
 496         if (unlikely(!fid_is_sane(fid)))
 497                 RETURN(0);
 498
 499         idx = lfsck_sub_trace_file_fid2idx(fid);
 500         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 501         obj = com->lc_sub_trace_objs[idx].lsto_obj;
 502         if (unlikely(obj == NULL)) {
 503                 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 504                 RETURN(0);
 505         }
 506
 507         lfsck_object_get(obj);
 508         dev = lfsck_obj2dev(obj);
 509         fid_cpu_to_be(key, fid);
 510         rc = dt_lookup(env, obj, (struct dt_rec *)&old,
 511                        (const struct dt_key *)key);
 512         if (rc == -ENOENT) {
 513                 if (!add)
 514                         GOTO(unlock, rc = 0);
 515
 516                 old = 0;
 517                 new = flags;
 518         } else if (rc == 0) {
 519                 if (add) {
 520                         if ((old & flags) == flags)
 521                                 GOTO(unlock, rc = 0);
 522
 523                         new = old | flags;
 524                 } else {
 525                         if ((old & flags) == 0)
 526                                 GOTO(unlock, rc = 0);
 527
 528                         new = old & ~flags;
 529                 }
 530         } else {
 531                 GOTO(log, rc);
 532         }
 533
 534         th = dt_trans_create(env, dev);
 535         if (IS_ERR(th))
 536                 GOTO(log, rc = PTR_ERR(th));
 537
 538         if (old != 0) {
 539                 rc = dt_declare_delete(env, obj,
 540                                        (const struct dt_key *)key, th);
 541                 if (rc != 0)
 542                         GOTO(log, rc);
 543         }
 544
 545         if (new != 0) {
 546                 rc = dt_declare_insert(env, obj,
 547                                        (const struct dt_rec *)&new,
 548                                        (const struct dt_key *)key, th);
 549                 if (rc != 0)
 550                         GOTO(log, rc);
 551         }
 552
 553         rc = dt_trans_start_local(env, dev, th);
 554         if (rc != 0)
 555                 GOTO(log, rc);
 556
 557         if (old != 0) {
 558                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
 559                 if (rc != 0)
 560                         GOTO(log, rc);
 561         }
 562
 563         if (new != 0) {
 564                 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
 565                                (const struct dt_key *)key, th, 1);
 566                 if (rc != 0)
 567                         GOTO(log, rc);
 568         }
 569
 570         GOTO(log, rc);
 571
 572 log:
 573         if (th != NULL && !IS_ERR(th))
 574                 dt_trans_stop(env, dev, th);
 575
 576         CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
 577                "trace file, flags %x, old %x, new %x: rc = %d\n",
 578                lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
 579                (__u32)flags, (__u32)old, (__u32)new, rc);
 580
 581 unlock:
 582         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 583         lfsck_object_put(env, obj);
 584
 585         return rc;
 586 }
 587
 588 int lfsck_namespace_check_exist(const struct lu_env *env,
 589                                 struct dt_object *dir,
 590                                 struct dt_object *obj, const char *name)
 591 {
 592         struct lu_fid    *fid = &lfsck_env_info(env)->lti_fid;
 593         int               rc;
 594         ENTRY;
 595
 596         if (unlikely(lfsck_is_dead_obj(obj)))
 597                 RETURN(LFSCK_NAMEENTRY_DEAD);
 598
 599         rc = dt_lookup(env, dir, (struct dt_rec *)fid,
 600                        (const struct dt_key *)name);
 601         if (rc == -ENOENT)
 602                 RETURN(LFSCK_NAMEENTRY_REMOVED);
 603
 604         if (rc < 0)
 605                 RETURN(rc);
 606
 607         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
 608                 RETURN(LFSCK_NAMEENTRY_RECREATED);
 609
 610         RETURN(0);
 611 }
 612
 613 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
 614                                             struct dt_object *obj,
 615                                             struct thandle *handle)
 616 {
 617         int rc;
 618
 619         /* For remote updating LINKEA, there may be further LFSCK action
 620          * on remote MDT after the updating, so update the LINKEA ASAP. */
 621         if (dt_object_remote(obj))
 622                 handle->th_sync = 1;
 623
 624         /* For destroying all invalid linkEA entries. */
 625         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
 626         if (rc == 0)
 627                 /* For insert new linkEA entry. */
 628                 rc = dt_declare_xattr_set(env, obj,
 629                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
 630                         XATTR_NAME_LINK, 0, handle);
 631         return rc;
 632 }
 633
 634 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
 635                        struct linkea_data *ldata, bool with_rec)
 636 {
 637         int rc;
 638
 639         if (ldata->ld_buf->lb_buf == NULL)
 640                 return -ENOMEM;
 641
 642         if (!dt_object_exists(obj))
 643                 return -ENOENT;
 644
 645         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
 646         if (rc == -ERANGE) {
 647                 /* Buf was too small, figure out what we need. */
 648                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
 649                 if (unlikely(rc == 0))
 650                         return -ENODATA;
 651
 652                 if (rc < 0)
 653                         return rc;
 654
 655                 lu_buf_realloc(ldata->ld_buf, rc);
 656                 if (ldata->ld_buf->lb_buf == NULL)
 657                         return -ENOMEM;
 658
 659                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
 660         }
 661
 662         if (unlikely(rc == 0))
 663                 return -ENODATA;
 664
 665         if (rc > 0) {
 666                 if (with_rec)
 667                         rc = linkea_init_with_rec(ldata);
 668                 else
 669                         rc = linkea_init(ldata);
 670         }
 671
 672         return rc;
 673 }
 674
 675 /**
 676  * Remove linkEA for the given object.
 677  *
 678  * The caller should take the ldlm lock before the calling.
 679  *
 680  * \param[in] env       pointer to the thread context
 681  * \param[in] com       pointer to the lfsck component
 682  * \param[in] obj       pointer to the dt_object to be handled
 683  *
 684  * \retval              0 for repaired cases
 685  * \retval              negative error number on failure
 686  */
 687 static int lfsck_namespace_links_remove(const struct lu_env *env,
 688                                         struct lfsck_component *com,
 689                                         struct dt_object *obj)
 690 {
 691         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 692         struct dt_device                *dev    = lfsck_obj2dev(obj);
 693         struct thandle                  *th     = NULL;
 694         int                              rc     = 0;
 695         ENTRY;
 696
 697         LASSERT(dt_object_remote(obj) == 0);
 698
 699         th = dt_trans_create(env, dev);
 700         if (IS_ERR(th))
 701                 GOTO(log, rc = PTR_ERR(th));
 702
 703         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
 704         if (rc != 0)
 705                 GOTO(stop, rc);
 706
 707         rc = dt_trans_start_local(env, dev, th);
 708         if (rc != 0)
 709                 GOTO(stop, rc);
 710
 711         dt_write_lock(env, obj, 0);
 712         if (unlikely(lfsck_is_dead_obj(obj)))
 713                 GOTO(unlock, rc = -ENOENT);
 714
 715         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 716                 GOTO(unlock, rc = 0);
 717
 718         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
 719
 720         GOTO(unlock, rc);
 721
 722 unlock:
 723         dt_write_unlock(env, obj);
 724
 725 stop:
 726         dt_trans_stop(env, dev, th);
 727
 728 log:
 729         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
 730                "for the object "DFID": rc = %d\n",
 731                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
 732
 733         if (rc == 0) {
 734                 struct lfsck_namespace *ns = com->lc_file_ram;
 735
 736                 ns->ln_flags |= LF_INCONSISTENT;
 737         }
 738
 739         return rc;
 740 }
 741
 742 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
 743                              struct linkea_data *ldata, struct thandle *handle)
 744 {
 745         struct lu_buf buf;
 746         int rc;
 747
 748         lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
 749
 750 again:
 751         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
 752         if (unlikely(rc == -ENOSPC)) {
 753                 rc = linkea_overflow_shrink(ldata);
 754                 if (likely(rc > 0)) {
 755                         buf.lb_len = rc;
 756                         goto again;
 757                 }
 758         }
 759
 760         return rc;
 761 }
 762
 763 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
 764                                                struct lu_name *cname,
 765                                                struct lu_fid *pfid,
 766                                                char *buf, const int buflen)
 767 {
 768         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
 769         if (unlikely(ldata->ld_reclen <= 0 ||
 770                      ldata->ld_reclen + sizeof(struct link_ea_header) >
 771                         ldata->ld_leh->leh_len ||
 772                      cname->ln_namelen <= 0 ||
 773                      cname->ln_namelen > NAME_MAX ||
 774                      cname->ln_namelen >= buflen ||
 775                      !fid_is_sane(pfid)))
 776                 return -EINVAL;
 777
 778         /* To guarantee the 'name' is terminated with '0'. */
 779         memcpy(buf, cname->ln_name, cname->ln_namelen);
 780         buf[cname->ln_namelen] = 0;
 781         cname->ln_name = buf;
 782
 783         return 0;
 784 }
 785
 786 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
 787                                  const struct lu_name *lname)
 788 {
 789         LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
 790
 791         /* If current record is corrupted, all the subsequent
 792          * records will be dropped. */
 793         if (unlikely(ldata->ld_reclen <= 0 ||
 794                      ldata->ld_reclen + sizeof(struct link_ea_header) >
 795                         ldata->ld_leh->leh_len)) {
 796                 void *ptr = ldata->ld_lee;
 797
 798                 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
 799                 ldata->ld_leh->leh_reccount = 0;
 800                 linkea_first_entry(ldata);
 801                 while (ldata->ld_lee != NULL &&
 802                        (char *)ldata->ld_lee < (char *)ptr) {
 803                         int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
 804                                      ldata->ld_lee->lee_reclen[1];
 805
 806                         ldata->ld_leh->leh_len += reclen;
 807                         ldata->ld_leh->leh_reccount++;
 808                         ldata->ld_lee = (struct link_ea_entry *)
 809                                         ((char *)ldata->ld_lee + reclen);
 810                 }
 811
 812                 ldata->ld_lee = NULL;
 813         } else {
 814                 linkea_del_buf(ldata, lname);
 815         }
 816 }
 817
 818 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
 819                                                struct lu_name *cname,
 820                                                struct lu_fid *pfid,
 821                                                bool remove)
 822 {
 823         struct link_ea_entry    *oldlee;
 824         int                      oldlen;
 825         int                      repeated = 0;
 826
 827         oldlee = ldata->ld_lee;
 828         oldlen = ldata->ld_reclen;
 829         linkea_next_entry(ldata);
 830         while (ldata->ld_lee != NULL) {
 831                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
 832                                    ldata->ld_lee->lee_reclen[1];
 833                 if (unlikely(ldata->ld_reclen == oldlen &&
 834                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
 835                         repeated++;
 836                         if (!remove)
 837                                 break;
 838
 839                         lfsck_linkea_del_buf(ldata, cname);
 840                 } else {
 841                         linkea_next_entry(ldata);
 842                 }
 843         }
 844         ldata->ld_lee = oldlee;
 845         ldata->ld_reclen = oldlen;
 846
 847         return repeated;
 848 }
 849
 850 /**
 851  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
 852  *
 853  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
 854  * with the given type to generate the name, the detailed rules for name
 855  * have been described as following.
 856  *
 857  * The function also generates the linkEA corresponding to the name entry
 858  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
 859  *
 860  * \param[in] env       pointer to the thread context
 861  * \param[in] com       pointer to the lfsck component
 862  * \param[in] orphan    pointer to the orphan MDT-object
 863  * \param[in] infix     additional information for the orphan name, such as
 864  *                      the FID for original
 865  * \param[in] type      the type for describing why the orphan MDT-object is
 866  *                      created. The rules are as following:
 867  *
 868  *  type "D":           The MDT-object is a directory, it may knows its parent
 869  *                      but because there is no valid linkEA, the LFSCK cannot
 870  *                      know where to put it back to the namespace.
 871  *  type "O":           The MDT-object has no linkEA, and there is no name
 872  *                      entry that references the MDT-object.
 873  *
 874  *  type "S":           The orphan MDT-object is a shard of a striped directory
 875  *
 876  * \see lfsck_layout_recreate_parent() for more types.
 877  *
 878  * The orphan name will be like:
 879  * ${FID}-${infix}-${type}-${conflict_version}
 880  *
 881  * \param[out] count    if some others inserted some linkEA entries by race,
 882  *                      then return the linkEA entries count.
 883  *
 884  * \retval              positive number for repaired cases
 885  * \retval              0 if needs to repair nothing
 886  * \retval              negative error number on failure
 887  */
 888 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
 889                                          struct lfsck_component *com,
 890                                          struct dt_object *orphan,
 891                                          const char *infix, const char *type,
 892                                          int *count)
 893 {
 894         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 895         struct lu_name                  *cname  = &info->lti_name;
 896         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
 897         struct lu_attr                  *la     = &info->lti_la2;
 898         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
 899         const struct lu_fid             *pfid;
 900         struct lu_fid                    tfid;
 901         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 902         struct dt_device                *dev    = lfsck_obj2dev(orphan);
 903         struct dt_object                *parent;
 904         struct thandle                  *th     = NULL;
 905         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
 906         struct lustre_handle             clh    = { 0 };
 907         struct linkea_data               ldata2 = { NULL };
 908         struct lu_buf                    linkea_buf;
 909         int                              namelen;
 910         int                              idx    = 0;
 911         int                              rc     = 0;
 912         bool                             exist  = false;
 913         ENTRY;
 914
 915         cname->ln_name = NULL;
 916         if (unlikely(lfsck->li_lpf_obj == NULL))
 917                 GOTO(log, rc = -ENXIO);
 918
 919         parent = lfsck->li_lpf_obj;
 920         pfid = lfsck_dto2fid(parent);
 921
 922 again:
 923         do {
 924                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
 925                                    PFID(cfid), infix, type, idx++);
 926                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
 927                                (const struct dt_key *)info->lti_key);
 928                 if (rc != 0 && rc != -ENOENT)
 929                         GOTO(log, rc);
 930
 931                 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
 932                         exist = true;
 933         } while (rc == 0 && !exist);
 934
 935         rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
 936                         MDS_INODELOCK_UPDATE, LCK_PW);
 937         if (rc != 0)
 938                 GOTO(log, rc);
 939
 940         /* Re-check whether the name conflict with othrs after taken
 941          * the ldlm lock. */
 942         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
 943                        (const struct dt_key *)info->lti_key);
 944         if (rc == 0) {
 945                 if (!lu_fid_eq(cfid, &tfid)) {
 946                         exist = false;
 947                         lfsck_unlock(pllh);
 948                         goto again;
 949                 }
 950
 951                 exist = true;
 952         } else if (rc != -ENOENT) {
 953                 GOTO(log, rc);
 954         } else {
 955                 exist = false;
 956         }
 957
 958         cname->ln_name = info->lti_key;
 959         cname->ln_namelen = namelen;
 960         rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
 961                               cname, pfid);
 962         if (rc != 0)
 963                 GOTO(log, rc);
 964
 965         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
 966                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
 967                               MDS_INODELOCK_XATTR, LCK_EX);
 968         if (rc != 0)
 969                 GOTO(log, rc);
 970
 971         lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
 972                        ldata2.ld_leh->leh_len);
 973         th = dt_trans_create(env, dev);
 974         if (IS_ERR(th))
 975                 GOTO(log, rc = PTR_ERR(th));
 976
 977         if (S_ISDIR(lfsck_object_type(orphan))) {
 978                 rc = dt_declare_delete(env, orphan,
 979                                        (const struct dt_key *)dotdot, th);
 980                 if (rc != 0)
 981                         GOTO(stop, rc);
 982
 983                 rec->rec_type = S_IFDIR;
 984                 rec->rec_fid = pfid;
 985                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
 986                                        (const struct dt_key *)dotdot, th);
 987                 if (rc != 0)
 988                         GOTO(stop, rc);
 989         }
 990
 991         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
 992                                   XATTR_NAME_LINK, 0, th);
 993         if (rc != 0)
 994                 GOTO(stop, rc);
 995
 996         if (!exist) {
 997                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
 998                 rec->rec_fid = cfid;
 999                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1000                                        (const struct dt_key *)cname->ln_name,
1001                                        th);
1002                 if (rc != 0)
1003                         GOTO(stop, rc);
1004
1005                 if (S_ISDIR(rec->rec_type)) {
1006                         rc = dt_declare_ref_add(env, parent, th);
1007                         if (rc != 0)
1008                                 GOTO(stop, rc);
1009                 }
1010         }
1011
1012         memset(la, 0, sizeof(*la));
1013         la->la_ctime = ktime_get_real_seconds();
1014         la->la_valid = LA_CTIME;
1015         rc = dt_declare_attr_set(env, orphan, la, th);
1016         if (rc != 0)
1017                 GOTO(stop, rc);
1018
1019         rc = dt_trans_start_local(env, dev, th);
1020         if (rc != 0)
1021                 GOTO(stop, rc);
1022
1023         dt_write_lock(env, orphan, 0);
1024         rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1025         if (likely(rc == -ENODATA || rc == -EINVAL)) {
1026                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1027                         GOTO(unlock, rc = 1);
1028
1029                 if (S_ISDIR(lfsck_object_type(orphan))) {
1030                         rc = dt_delete(env, orphan,
1031                                        (const struct dt_key *)dotdot, th);
1032                         if (rc != 0)
1033                                 GOTO(unlock, rc);
1034
1035                         rec->rec_type = S_IFDIR;
1036                         rec->rec_fid = pfid;
1037                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1038                                        (const struct dt_key *)dotdot, th, 1);
1039                         if (rc != 0)
1040                                 GOTO(unlock, rc);
1041                 }
1042
1043                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1044                                   th);
1045         } else {
1046                 if (rc == 0 && count != NULL)
1047                         *count = ldata2.ld_leh->leh_reccount;
1048
1049                 GOTO(unlock, rc);
1050         }
1051         dt_write_unlock(env, orphan);
1052
1053         if (rc == 0 && !exist) {
1054                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1055                 rec->rec_fid = cfid;
1056                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1057                                (const struct dt_key *)cname->ln_name, th, 1);
1058                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1059                         dt_write_lock(env, parent, 0);
1060                         rc = dt_ref_add(env, parent, th);
1061                         dt_write_unlock(env, parent);
1062                 }
1063         }
1064
1065         if (rc == 0)
1066                 rc = dt_attr_set(env, orphan, la, th);
1067
1068         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1069
1070 unlock:
1071         dt_write_unlock(env, orphan);
1072
1073 stop:
1074         dt_trans_stop(env, dev, th);
1075
1076 log:
1077         lfsck_ibits_unlock(&clh, LCK_EX);
1078         lfsck_unlock(pllh);
1079         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1080                "object "DFID", name = %s: rc = %d\n",
1081                lfsck_lfsck2name(lfsck), PFID(cfid),
1082                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1083
1084         if (rc != 0) {
1085                 struct lfsck_namespace *ns = com->lc_file_ram;
1086
1087                 ns->ln_flags |= LF_INCONSISTENT;
1088         }
1089
1090         return rc;
1091 }
1092
1093 /**
1094  * Add the specified name entry back to namespace.
1095  *
1096  * If there is a linkEA entry that back references a name entry under
1097  * some parent directory, but such parent directory does not have the
1098  * claimed name entry. On the other hand, the linkEA entries count is
1099  * not larger than the MDT-object's hard link count. Under such case,
1100  * it is quite possible that the name entry is lost. Then the LFSCK
1101  * should add the name entry back to the namespace.
1102  *
1103  * \param[in] env       pointer to the thread context
1104  * \param[in] com       pointer to the lfsck component
1105  * \param[in] parent    pointer to the directory under which the name entry
1106  *                      will be inserted into
1107  * \param[in] child     pointer to the object referenced by the name entry
1108  *                      that to be inserted into the parent
1109  * \param[in] name      the name for the child in the parent directory
1110  *
1111  * \retval              positive number for repaired cases
1112  * \retval              0 if nothing to be repaired
1113  * \retval              negative error number on failure
1114  */
1115 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1116                                          struct lfsck_component *com,
1117                                          struct dt_object *parent,
1118                                          struct dt_object *child,
1119                                          const char *name)
1120 {
1121         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1122         struct lu_attr                  *la     = &info->lti_la;
1123         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1124         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1125         /* The child and its name may be on different MDTs. */
1126         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1127         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
1128         struct dt_device                *dev    = lfsck->li_next;
1129         struct thandle                  *th     = NULL;
1130         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1131         int                              rc     = 0;
1132         ENTRY;
1133
1134         /* @parent/@child may be based on lfsck->li_bottom,
1135          * but here we need the object based on the lfsck->li_next. */
1136
1137         parent = lfsck_object_locate(dev, parent);
1138         if (IS_ERR(parent))
1139                 GOTO(log, rc = PTR_ERR(parent));
1140
1141         if (unlikely(!dt_try_as_dir(env, parent)))
1142                 GOTO(log, rc = -ENOTDIR);
1143
1144         child = lfsck_object_locate(dev, child);
1145         if (IS_ERR(child))
1146                 GOTO(log, rc = PTR_ERR(child));
1147
1148         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1149                 GOTO(log, rc = 1);
1150
1151         rc = lfsck_lock(env, lfsck, parent, name, llh,
1152                         MDS_INODELOCK_UPDATE, LCK_PW);
1153         if (rc != 0)
1154                 GOTO(log, rc);
1155
1156         th = dt_trans_create(env, dev);
1157         if (IS_ERR(th))
1158                 GOTO(unlock, rc = PTR_ERR(th));
1159
1160         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1161         rec->rec_fid = cfid;
1162         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1163                                (const struct dt_key *)name, th);
1164         if (rc != 0)
1165                 GOTO(stop, rc);
1166
1167         if (S_ISDIR(rec->rec_type)) {
1168                 rc = dt_declare_ref_add(env, parent, th);
1169                 if (rc != 0)
1170                         GOTO(stop, rc);
1171         }
1172
1173         memset(la, 0, sizeof(*la));
1174         la->la_ctime = ktime_get_real_seconds();
1175         la->la_valid = LA_CTIME;
1176         rc = dt_declare_attr_set(env, parent, la, th);
1177         if (rc != 0)
1178                 GOTO(stop, rc);
1179
1180         rc = dt_declare_attr_set(env, child, la, th);
1181         if (rc != 0)
1182                 GOTO(stop, rc);
1183
1184         rc = dt_trans_start_local(env, dev, th);
1185         if (rc != 0)
1186                 GOTO(stop, rc);
1187
1188         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1189                        (const struct dt_key *)name, th, 1);
1190         if (rc != 0)
1191                 GOTO(stop, rc);
1192
1193         if (S_ISDIR(rec->rec_type)) {
1194                 dt_write_lock(env, parent, 0);
1195                 rc = dt_ref_add(env, parent, th);
1196                 dt_write_unlock(env, parent);
1197                 if (rc != 0)
1198                         GOTO(stop, rc);
1199         }
1200
1201         la->la_ctime = ktime_get_real_seconds();
1202         rc = dt_attr_set(env, parent, la, th);
1203         if (rc != 0)
1204                 GOTO(stop, rc);
1205
1206         rc = dt_attr_set(env, child, la, th);
1207
1208         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1209
1210 stop:
1211         dt_trans_stop(env, dev, th);
1212
1213 unlock:
1214         lfsck_unlock(llh);
1215
1216 log:
1217         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1218                "the name %s and type %o to the parent "DFID": rc = %d\n",
1219                lfsck_lfsck2name(lfsck), PFID(cfid), name,
1220                lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1221
1222         if (rc != 0) {
1223                 struct lfsck_namespace *ns = com->lc_file_ram;
1224
1225                 ns->ln_flags |= LF_INCONSISTENT;
1226                 if (rc > 0)
1227                         ns->ln_lost_dirent_repaired++;
1228         }
1229
1230         return rc;
1231 }
1232
1233 /**
1234  * Create the specified orphan directory.
1235  *
1236  * For the case that the parent MDT-object stored in some MDT-object's
1237  * linkEA entry is lost, the LFSCK will re-create the parent object as
1238  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1239  * with the name ${FID}-P-${conflict_version}.
1240  *
1241  * \param[in] env       pointer to the thread context
1242  * \param[in] com       pointer to the lfsck component
1243  * \param[in] orphan    pointer to the orphan MDT-object to be created
1244  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1245  *
1246  * \retval              positive number for repaired cases
1247  * \retval              negative error number on failure
1248  */
1249 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1250                                              struct lfsck_component *com,
1251                                              struct dt_object *orphan,
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 lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1261         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1262         struct lu_fid                    tfid;
1263         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1264         struct lfsck_namespace          *ns     = com->lc_file_ram;
1265         struct dt_device                *dev    = lfsck_obj2dev(orphan);
1266         struct dt_object                *parent = NULL;
1267         struct thandle                  *th     = NULL;
1268         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1269         struct linkea_data               ldata  = { NULL };
1270         struct lu_buf                    linkea_buf;
1271         struct lu_buf                    lmv_buf;
1272         char                             name[32];
1273         int                              namelen;
1274         int                              idx    = 0;
1275         int                              rc     = 0;
1276         int                              rc1    = 0;
1277         ENTRY;
1278
1279         LASSERT(!dt_object_exists(orphan));
1280
1281         cname->ln_name = NULL;
1282         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1283                 GOTO(log, rc = 1);
1284
1285         if (dt_object_remote(orphan)) {
1286                 LASSERT(lfsck->li_lpf_root_obj != NULL);
1287
1288                 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1289                 if (idx < 0)
1290                         GOTO(log, rc = idx);
1291
1292                 snprintf(name, 8, "MDT%04x", idx);
1293                 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1294                                (struct dt_rec *)&tfid,
1295                                (const struct dt_key *)name);
1296                 if (rc != 0)
1297                         GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1298
1299                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1300                 if (IS_ERR(parent))
1301                         GOTO(log, rc = PTR_ERR(parent));
1302
1303                 if (unlikely(!dt_try_as_dir(env, parent)))
1304                         GOTO(log, rc = -ENOTDIR);
1305         } else {
1306                 if (unlikely(lfsck->li_lpf_obj == NULL))
1307                         GOTO(log, rc = -ENXIO);
1308
1309                 parent = lfsck->li_lpf_obj;
1310         }
1311
1312         dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1313         if (IS_ERR(dev))
1314                 GOTO(log, rc = PTR_ERR(dev));
1315
1316         idx = 0;
1317
1318 again:
1319         do {
1320                 namelen = snprintf(name, 31, DFID"-P-%d",
1321                                    PFID(cfid), idx++);
1322                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1323                                (const struct dt_key *)name);
1324                 if (rc != 0 && rc != -ENOENT)
1325                         GOTO(log, rc);
1326         } while (rc == 0);
1327
1328         rc = lfsck_lock(env, lfsck, parent, name, llh,
1329                         MDS_INODELOCK_UPDATE, LCK_PW);
1330         if (rc != 0)
1331                 GOTO(log, rc);
1332
1333         /* Re-check whether the name conflict with othrs after taken
1334          * the ldlm lock. */
1335         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1336                        (const struct dt_key *)name);
1337         if (unlikely(rc == 0)) {
1338                 lfsck_unlock(llh);
1339                 goto again;
1340         }
1341
1342         if (rc != -ENOENT)
1343                 GOTO(unlock1, rc);
1344
1345         cname->ln_name = name;
1346         cname->ln_namelen = namelen;
1347
1348         memset(la, 0, sizeof(*la));
1349         la->la_mode = S_IFDIR | 0700;
1350         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1351                        LA_ATIME | LA_MTIME | LA_CTIME;
1352
1353         orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1354                                    la->la_mode & S_IFMT);
1355
1356         memset(dof, 0, sizeof(*dof));
1357         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1358
1359         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1360                               cname, lfsck_dto2fid(parent));
1361         if (rc != 0)
1362                 GOTO(unlock1, rc);
1363
1364         th = dt_trans_create(env, dev);
1365         if (IS_ERR(th))
1366                 GOTO(unlock1, rc = PTR_ERR(th));
1367
1368         /* Sync the remote transaction to guarantee that the subsequent
1369          * lock against the @orphan can find the @orphan in time. */
1370         if (dt_object_remote(orphan))
1371                 th->th_sync = 1;
1372
1373         rc = dt_declare_create(env, orphan, la, hint, dof, th);
1374         if (rc != 0)
1375                 GOTO(stop, rc);
1376
1377         if (unlikely(!dt_try_as_dir(env, orphan)))
1378                 GOTO(stop, rc = -ENOTDIR);
1379
1380         rc = dt_declare_ref_add(env, orphan, th);
1381         if (rc != 0)
1382                 GOTO(stop, rc);
1383
1384         rec->rec_type = S_IFDIR;
1385         rec->rec_fid = cfid;
1386         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1387                                (const struct dt_key *)dot, th);
1388         if (rc != 0)
1389                 GOTO(stop, rc);
1390
1391         rec->rec_fid = lfsck_dto2fid(parent);
1392         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1393                                (const struct dt_key *)dotdot, th);
1394         if (rc != 0)
1395                 GOTO(stop, rc);
1396
1397         if (lmv != NULL) {
1398                 lmv->lmv_magic = LMV_MAGIC;
1399                 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1400                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1401                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1402                 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1403                                           XATTR_NAME_LMV, 0, th);
1404                 if (rc != 0)
1405                         GOTO(stop, rc);
1406         }
1407
1408         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1409                        ldata.ld_leh->leh_len);
1410         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1411                                   XATTR_NAME_LINK, 0, th);
1412         if (rc != 0)
1413                 GOTO(stop, rc);
1414
1415         rec->rec_fid = cfid;
1416         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1417                                (const struct dt_key *)name, th);
1418         if (rc == 0)
1419                 rc = dt_declare_ref_add(env, parent, th);
1420
1421         if (rc != 0)
1422                 GOTO(stop, rc);
1423
1424         rc = dt_trans_start_local(env, dev, th);
1425         if (rc != 0)
1426                 GOTO(stop, rc);
1427
1428         dt_write_lock(env, orphan, 0);
1429         rc = dt_create(env, orphan, la, hint, dof, th);
1430         if (rc != 0)
1431                 GOTO(unlock2, rc);
1432
1433         rc = dt_ref_add(env, orphan, th);
1434         if (rc != 0)
1435                 GOTO(unlock2, rc);
1436
1437         rec->rec_fid = cfid;
1438         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1439                        (const struct dt_key *)dot, th, 1);
1440         if (rc != 0)
1441                 GOTO(unlock2, rc);
1442
1443         rec->rec_fid = lfsck_dto2fid(parent);
1444         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1445                        (const struct dt_key *)dotdot, th, 1);
1446         if (rc != 0)
1447                 GOTO(unlock2, rc);
1448
1449         if (lmv != NULL) {
1450                 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1451                 if (rc != 0)
1452                         GOTO(unlock2, rc);
1453         }
1454
1455         rc = dt_xattr_set(env, orphan, &linkea_buf,
1456                           XATTR_NAME_LINK, 0, th);
1457         dt_write_unlock(env, orphan);
1458         if (rc != 0)
1459                 GOTO(stop, rc);
1460
1461         rec->rec_fid = cfid;
1462         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1463                        (const struct dt_key *)name, th, 1);
1464         if (rc == 0) {
1465                 dt_write_lock(env, parent, 0);
1466                 rc = dt_ref_add(env, parent, th);
1467                 dt_write_unlock(env, parent);
1468         }
1469
1470         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1471
1472 unlock2:
1473         dt_write_unlock(env, orphan);
1474
1475 stop:
1476         rc1 = dt_trans_stop(env, dev, th);
1477         if (rc1 != 0 && rc > 0)
1478                 rc = rc1;
1479
1480 unlock1:
1481         lfsck_unlock(llh);
1482
1483 log:
1484         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1485                "the object "DFID", name = %s: rc = %d\n",
1486                lfsck_lfsck2name(lfsck), PFID(cfid),
1487                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1488
1489         if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1490                 lfsck_object_put(env, parent);
1491
1492         if (rc != 0)
1493                 ns->ln_flags |= LF_INCONSISTENT;
1494
1495         return rc;
1496 }
1497
1498 /**
1499  * Remove the specified entry from the linkEA.
1500  *
1501  * Locate the linkEA entry with the given @cname and @pfid, then
1502  * remove this entry or the other entries those are repeated with
1503  * this entry.
1504  *
1505  * \param[in] env       pointer to the thread context
1506  * \param[in] com       pointer to the lfsck component
1507  * \param[in] obj       pointer to the dt_object to be handled
1508  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1509  * \param[in] cname     the name for the child in the parent directory
1510  * \param[in] pfid      the parent directory's FID for the linkEA
1511  * \param[in] next      if true, then remove the first found linkEA
1512  *                      entry, and move the ldata->ld_lee to next entry
1513  *
1514  * \retval              positive number for repaired cases
1515  * \retval              0 if nothing to be repaired
1516  * \retval              negative error number on failure
1517  */
1518 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1519                                          struct lfsck_component *com,
1520                                          struct dt_object *obj,
1521                                          struct linkea_data *ldata,
1522                                          struct lu_name *cname,
1523                                          struct lu_fid *pfid,
1524                                          bool next)
1525 {
1526         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1527         struct dt_device                *dev       = lfsck_obj2dev(obj);
1528         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1529         struct thandle                  *th        = NULL;
1530         struct lustre_handle             lh        = { 0 };
1531         struct linkea_data               ldata_new = { NULL };
1532         struct lu_buf                    linkea_buf;
1533         int                              buflen    = 0;
1534         int                              rc        = 0;
1535         ENTRY;
1536
1537         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1538                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1539                               LCK_EX);
1540         if (rc != 0)
1541                 GOTO(log, rc);
1542
1543         if (next)
1544                 lfsck_linkea_del_buf(ldata, cname);
1545         else
1546                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1547                                                     true);
1548         if (ldata->ld_leh->leh_reccount > 0 ||
1549             unlikely(ldata->ld_leh->leh_overflow_time)) {
1550                 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1551                                ldata->ld_leh->leh_len);
1552                 buflen = linkea_buf.lb_len;
1553         }
1554
1555 again:
1556         th = dt_trans_create(env, dev);
1557         if (IS_ERR(th))
1558                 GOTO(unlock1, rc = PTR_ERR(th));
1559
1560         if (buflen != 0)
1561                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1562                                           XATTR_NAME_LINK, 0, th);
1563         else
1564                 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1565         if (rc != 0)
1566                 GOTO(stop, rc);
1567
1568         rc = dt_trans_start_local(env, dev, th);
1569         if (rc != 0)
1570                 GOTO(stop, rc);
1571
1572         dt_write_lock(env, obj, 0);
1573         if (unlikely(lfsck_is_dead_obj(obj)))
1574                 GOTO(unlock2, rc = -ENOENT);
1575
1576         rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1577         if (rc)
1578                 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1579
1580         /* The specified linkEA entry has been removed by race. */
1581         rc = linkea_links_find(&ldata_new, cname, pfid);
1582         if (rc != 0)
1583                 GOTO(unlock2, rc = 0);
1584
1585         if (bk->lb_param & LPF_DRYRUN)
1586                 GOTO(unlock2, rc = 1);
1587
1588         if (next)
1589                 lfsck_linkea_del_buf(&ldata_new, cname);
1590         else
1591                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1592                                                     true);
1593
1594         if (buflen < ldata_new.ld_leh->leh_len) {
1595                 dt_write_unlock(env, obj);
1596                 dt_trans_stop(env, dev, th);
1597                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1598                                ldata_new.ld_leh->leh_len);
1599                 goto again;
1600         }
1601
1602         if (ldata_new.ld_leh->leh_reccount > 0 ||
1603             unlikely(ldata->ld_leh->leh_overflow_time))
1604                 rc = lfsck_links_write(env, obj, &ldata_new, th);
1605         else
1606                 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1607
1608         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1609
1610 unlock2:
1611         dt_write_unlock(env, obj);
1612
1613 stop:
1614         dt_trans_stop(env, dev, th);
1615
1616 unlock1:
1617         lfsck_ibits_unlock(&lh, LCK_EX);
1618
1619 log:
1620         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1621                "for the object: "DFID", parent "DFID", name %.*s\n",
1622                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1623                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1624                cname->ln_name);
1625
1626         if (rc != 0) {
1627                 struct lfsck_namespace *ns = com->lc_file_ram;
1628
1629                 ns->ln_flags |= LF_INCONSISTENT;
1630         }
1631
1632         return rc;
1633 }
1634
1635 /**
1636  * Conditionally remove the specified entry from the linkEA.
1637  *
1638  * Take the parent lock firstly, then check whether the specified
1639  * name entry exists or not: if yes, do nothing; otherwise, call
1640  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1641  *
1642  * \param[in] env       pointer to the thread context
1643  * \param[in] com       pointer to the lfsck component
1644  * \param[in] parent    pointer to the parent directory
1645  * \param[in] child     pointer to the child object that holds the linkEA
1646  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1647  * \param[in] cname     the name for the child in the parent directory
1648  * \param[in] pfid      the parent directory's FID for the linkEA
1649  *
1650  * \retval              positive number for repaired cases
1651  * \retval              0 if nothing to be repaired
1652  * \retval              negative error number on failure
1653  */
1654 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1655                                               struct lfsck_component *com,
1656                                               struct dt_object *parent,
1657                                               struct dt_object *child,
1658                                               struct linkea_data *ldata,
1659                                               struct lu_name *cname,
1660                                               struct lu_fid *pfid)
1661 {
1662         struct lfsck_thread_info *info  = lfsck_env_info(env);
1663         struct lu_fid            *cfid  = &info->lti_fid3;
1664         struct lfsck_lock_handle *llh   = &info->lti_llh;
1665         int                       rc;
1666         ENTRY;
1667
1668         rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1669                         MDS_INODELOCK_UPDATE, LCK_PR);
1670         if (rc != 0)
1671                 RETURN(rc);
1672
1673         dt_read_lock(env, parent, 0);
1674         if (unlikely(lfsck_is_dead_obj(parent))) {
1675                 dt_read_unlock(env, parent);
1676                 lfsck_unlock(llh);
1677                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1678                                                    cname, pfid, true);
1679
1680                 RETURN(rc);
1681         }
1682
1683         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1684                        (const struct dt_key *)cname->ln_name);
1685         dt_read_unlock(env, parent);
1686
1687         /* It is safe to release the ldlm lock, because when the logic come
1688          * here, we have got all the needed information above whether the
1689          * linkEA entry is valid or not. It is not important that others
1690          * may add new linkEA entry after the ldlm lock released. If other
1691          * has removed the specified linkEA entry by race, then it is OK,
1692          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1693          * such case. */
1694         lfsck_unlock(llh);
1695         if (rc == -ENOENT) {
1696                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1697                                                    cname, pfid, true);
1698
1699                 RETURN(rc);
1700         }
1701
1702         if (rc != 0)
1703                 RETURN(rc);
1704
1705         /* The LFSCK just found some internal status of cross-MDTs
1706          * create operation. That is normal. */
1707         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1708                 linkea_next_entry(ldata);
1709
1710                 RETURN(0);
1711         }
1712
1713         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1714                                            pfid, true);
1715
1716         RETURN(rc);
1717 }
1718
1719 /**
1720  * Conditionally replace name entry in the parent.
1721  *
1722  * As required, the LFSCK may re-create the lost MDT-object for dangling
1723  * name entry, but such repairing may be wrong because of bad FID in the
1724  * name entry. As the LFSCK processing, the real MDT-object may be found,
1725  * then the LFSCK should check whether the former re-created MDT-object
1726  * has been modified or not, if not, then destroy it and update the name
1727  * entry in the parent to reference the real MDT-object.
1728  *
1729  * \param[in] env       pointer to the thread context
1730  * \param[in] com       pointer to the lfsck component
1731  * \param[in] parent    pointer to the parent directory
1732  * \param[in] child     pointer to the MDT-object that may be the real
1733  *                      MDT-object corresponding to the name entry in parent
1734  * \param[in] cfid      the current FID in the name entry
1735  * \param[in] cname     contains the name of the child in the parent directory
1736  *
1737  * \retval              positive number for repaired cases
1738  * \retval              0 if nothing to be repaired
1739  * \retval              negative error number on failure
1740  */
1741 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1742                                         struct lfsck_component *com,
1743                                         struct dt_object *parent,
1744                                         struct dt_object *child,
1745                                         const struct lu_fid *cfid,
1746                                         const struct lu_name *cname)
1747 {
1748         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1749         struct lu_attr                  *la     = &info->lti_la;
1750         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1751         struct lu_fid                    tfid;
1752         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1753         /* The child and its name may be on different MDTs. */
1754         struct dt_device                *dev    = lfsck->li_next;
1755         const char                      *name   = cname->ln_name;
1756         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1757         struct dt_object                *cobj   = NULL;
1758         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
1759         struct lustre_handle             clh    = { 0 };
1760         struct linkea_data               ldata  = { NULL };
1761         struct thandle                  *th     = NULL;
1762         bool                             exist  = true;
1763         int                              rc     = 0;
1764         ENTRY;
1765
1766         /* @parent/@child may be based on lfsck->li_bottom,
1767          * but here we need the object based on the lfsck->li_next. */
1768
1769         parent = lfsck_object_locate(dev, parent);
1770         if (IS_ERR(parent))
1771                 GOTO(log, rc = PTR_ERR(parent));
1772
1773         if (unlikely(!dt_try_as_dir(env, parent)))
1774                 GOTO(log, rc = -ENOTDIR);
1775
1776         rc = lfsck_lock(env, lfsck, parent, name, pllh,
1777                         MDS_INODELOCK_UPDATE, LCK_PW);
1778         if (rc != 0)
1779                 GOTO(log, rc);
1780
1781         if (!fid_is_sane(cfid)) {
1782                 exist = false;
1783                 goto replace;
1784         }
1785
1786         cobj = lfsck_object_find_by_dev(env, dev, cfid);
1787         if (IS_ERR(cobj)) {
1788                 rc = PTR_ERR(cobj);
1789                 if (rc == -ENOENT) {
1790                         exist = false;
1791                         goto replace;
1792                 }
1793
1794                 GOTO(log, rc);
1795         }
1796
1797         if (!dt_object_exists(cobj)) {
1798                 exist = false;
1799                 goto replace;
1800         }
1801
1802         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1803                        (const struct dt_key *)name);
1804         if (rc == -ENOENT) {
1805                 exist = false;
1806                 goto replace;
1807         }
1808
1809         if (rc != 0)
1810                 GOTO(log, rc);
1811
1812         /* Someone changed the name entry, cannot replace it. */
1813         if (!lu_fid_eq(cfid, &tfid))
1814                 GOTO(log, rc = 0);
1815
1816         /* lock the object to be destroyed. */
1817         rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1818                               MDS_INODELOCK_UPDATE |
1819                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1820                               LCK_EX);
1821         if (rc != 0)
1822                 GOTO(log, rc);
1823
1824         if (unlikely(lfsck_is_dead_obj(cobj))) {
1825                 exist = false;
1826                 goto replace;
1827         }
1828
1829         rc = dt_attr_get(env, cobj, la);
1830         if (rc != 0)
1831                 GOTO(log, rc);
1832
1833         /* The object has been modified by other(s), or it is not created by
1834          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1835         if (la->la_ctime != 0)
1836                 GOTO(log, rc);
1837
1838         if (S_ISREG(la->la_mode)) {
1839                 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
1840                 /* If someone has created related OST-object(s),
1841                  * then keep it. */
1842                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1843                         GOTO(log, rc = (rc > 0 ? 0 : rc));
1844         }
1845
1846 replace:
1847         dt_read_lock(env, child, 0);
1848         rc = lfsck_links_read2_with_rec(env, child, &ldata);
1849         dt_read_unlock(env, child);
1850
1851         /* Someone changed the child, no need to replace. */
1852         if (rc == -ENODATA)
1853                 GOTO(log, rc = 0);
1854
1855         if (rc != 0)
1856                 GOTO(log, rc);
1857
1858         rc = linkea_links_find(&ldata, cname, pfid);
1859         /* Someone moved the child, no need to replace. */
1860         if (rc != 0)
1861                 GOTO(log, rc = 0);
1862
1863         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1864                 GOTO(log, rc = 1);
1865
1866         th = dt_trans_create(env, dev);
1867         if (IS_ERR(th))
1868                 GOTO(log, rc = PTR_ERR(th));
1869
1870         if (exist) {
1871                 rc = dt_declare_destroy(env, cobj, th);
1872                 if (rc != 0)
1873                         GOTO(stop, rc);
1874         }
1875
1876         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1877         if (rc != 0)
1878                 GOTO(stop, rc);
1879
1880         rec->rec_type = S_IFDIR;
1881         rec->rec_fid = lfsck_dto2fid(child);
1882         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1883                                (const struct dt_key *)name, th);
1884         if (rc != 0)
1885                 GOTO(stop, rc);
1886
1887         rc = dt_trans_start_local(env, dev, th);
1888         if (rc != 0)
1889                 GOTO(stop, rc);
1890
1891         if (exist) {
1892                 rc = dt_destroy(env, cobj, th);
1893                 if (rc != 0)
1894                         GOTO(stop, rc);
1895         }
1896
1897         /* The old name entry maybe not exist. */
1898         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
1899         if (rc != 0 && rc != -ENOENT)
1900                 GOTO(stop, rc);
1901
1902         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1903                        (const struct dt_key *)name, th, 1);
1904
1905         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1906
1907 stop:
1908         dt_trans_stop(env, dev, th);
1909
1910 log:
1911         lfsck_ibits_unlock(&clh, LCK_EX);
1912         lfsck_unlock(pllh);
1913
1914         if (cobj != NULL && !IS_ERR(cobj))
1915                 lfsck_object_put(env, cobj);
1916
1917         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1918                "object "DFID" because of conflict with the object "DFID
1919                " under the parent "DFID" with name %s: rc = %d\n",
1920                lfsck_lfsck2name(lfsck), PFID(cfid),
1921                PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
1922
1923         return rc;
1924 }
1925
1926 /**
1927  * Overwrite the linkEA for the object with the given ldata.
1928  *
1929  * The caller should take the ldlm lock before the calling.
1930  *
1931  * \param[in] env       pointer to the thread context
1932  * \param[in] com       pointer to the lfsck component
1933  * \param[in] obj       pointer to the dt_object to be handled
1934  * \param[in] ldata     pointer to the new linkEA data
1935  *
1936  * \retval              positive number for repaired cases
1937  * \retval              0 if nothing to be repaired
1938  * \retval              negative error number on failure
1939  */
1940 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1941                                    struct lfsck_component *com,
1942                                    struct dt_object *obj,
1943                                    struct linkea_data *ldata)
1944 {
1945         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1946         struct dt_device                *dev    = lfsck_obj2dev(obj);
1947         struct thandle                  *th     = NULL;
1948         struct lu_buf                    linkea_buf;
1949         int                              rc     = 0;
1950         ENTRY;
1951
1952         th = dt_trans_create(env, dev);
1953         if (IS_ERR(th))
1954                 GOTO(log, rc = PTR_ERR(th));
1955
1956         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1957                        ldata->ld_leh->leh_len);
1958         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1959                                   XATTR_NAME_LINK, 0, th);
1960         if (rc != 0)
1961                 GOTO(stop, rc);
1962
1963         rc = dt_trans_start_local(env, dev, th);
1964         if (rc != 0)
1965                 GOTO(stop, rc);
1966
1967         dt_write_lock(env, obj, 0);
1968         if (unlikely(lfsck_is_dead_obj(obj)))
1969                 GOTO(unlock, rc = 0);
1970
1971         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1972                 GOTO(unlock, rc = 1);
1973
1974         rc = dt_xattr_set(env, obj, &linkea_buf,
1975                           XATTR_NAME_LINK, 0, th);
1976
1977         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1978
1979 unlock:
1980         dt_write_unlock(env, obj);
1981
1982 stop:
1983         dt_trans_stop(env, dev, th);
1984
1985 log:
1986         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1987                "object "DFID": rc = %d\n",
1988                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1989
1990         if (rc != 0) {
1991                 struct lfsck_namespace *ns = com->lc_file_ram;
1992
1993                 ns->ln_flags |= LF_INCONSISTENT;
1994         }
1995
1996         return rc;
1997 }
1998
1999 /**
2000  * Repair invalid name entry.
2001  *
2002  * If the name entry contains invalid information, such as bad file type
2003  * or (and) corrupted object FID, then either remove the name entry or
2004  * udpate the name entry with the given (right) information.
2005  *
2006  * \param[in] env       pointer to the thread context
2007  * \param[in] com       pointer to the lfsck component
2008  * \param[in] parent    pointer to the parent directory
2009  * \param[in] child     pointer to the object referenced by the name entry
2010  * \param[in] name      the old name of the child under the parent directory
2011  * \param[in] name2     the new name of the child under the parent directory
2012  * \param[in] type      the type claimed by the name entry
2013  * \param[in] update    update the name entry if true; otherwise, remove it
2014  * \param[in] dec       decrease the parent nlink count if true
2015  *
2016  * \retval              positive number for repaired successfully
2017  * \retval              0 if nothing to be repaired
2018  * \retval              negative error number on failure
2019  */
2020 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2021                                   struct lfsck_component *com,
2022                                   struct dt_object *parent,
2023                                   struct dt_object *child,
2024                                   const char *name, const char *name2,
2025                                   __u16 type, bool update, bool dec)
2026 {
2027         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2028         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2029         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2030         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
2031         struct lu_fid                    tfid;
2032         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2033         struct dt_device                *dev    = lfsck->li_next;
2034         struct thandle                  *th     = NULL;
2035         struct lfsck_lock_handle        *llh    = &info->lti_llh;
2036         struct lustre_handle             lh     = { 0 };
2037         int                              rc     = 0;
2038         ENTRY;
2039
2040         parent = lfsck_object_locate(dev, parent);
2041         if (IS_ERR(parent))
2042                 GOTO(log, rc = PTR_ERR(parent));
2043
2044         if (unlikely(!dt_try_as_dir(env, parent)))
2045                 GOTO(log, rc = -ENOTDIR);
2046
2047         if (!update || strcmp(name, name2) == 0)
2048                 rc = lfsck_lock(env, lfsck, parent, name, llh,
2049                                 MDS_INODELOCK_UPDATE, LCK_PW);
2050         else
2051                 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2052                                       MDS_INODELOCK_UPDATE, LCK_PW);
2053         if (rc != 0)
2054                 GOTO(log, rc);
2055
2056         th = dt_trans_create(env, dev);
2057         if (IS_ERR(th))
2058                 GOTO(unlock1, rc = PTR_ERR(th));
2059
2060         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2061         if (rc != 0)
2062                 GOTO(stop, rc);
2063
2064         if (update) {
2065                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2066                 rec->rec_fid = cfid;
2067                 rc = dt_declare_insert(env, parent,
2068                                        (const struct dt_rec *)rec,
2069                                        (const struct dt_key *)name2, th);
2070                 if (rc != 0)
2071                         GOTO(stop, rc);
2072         }
2073
2074         if (dec && S_ISDIR(type)) {
2075                 rc = dt_declare_ref_del(env, parent, th);
2076                 if (rc != 0)
2077                         GOTO(stop, rc);
2078         }
2079
2080         rc = dt_trans_start_local(env, dev, th);
2081         if (rc != 0)
2082                 GOTO(stop, rc);
2083
2084
2085         dt_write_lock(env, parent, 0);
2086         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2087                        (const struct dt_key *)name);
2088         /* Someone has removed the bad name entry by race. */
2089         if (rc == -ENOENT)
2090                 GOTO(unlock2, rc = 0);
2091
2092         if (rc != 0)
2093                 GOTO(unlock2, rc);
2094
2095         /* Someone has removed the bad name entry and reused it for other
2096          * object by race. */
2097         if (!lu_fid_eq(&tfid, cfid))
2098                 GOTO(unlock2, rc = 0);
2099
2100         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2101                 GOTO(unlock2, rc = 1);
2102
2103         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2104         if (rc != 0)
2105                 GOTO(unlock2, rc);
2106
2107         if (update) {
2108                 rc = dt_insert(env, parent,
2109                                (const struct dt_rec *)rec,
2110                                (const struct dt_key *)name2, th, 1);
2111                 if (rc != 0)
2112                         GOTO(unlock2, rc);
2113         }
2114
2115         if (dec && S_ISDIR(type)) {
2116                 rc = dt_ref_del(env, parent, th);
2117                 if (rc != 0)
2118                         GOTO(unlock2, rc);
2119         }
2120
2121         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2122
2123 unlock2:
2124         dt_write_unlock(env, parent);
2125
2126 stop:
2127         dt_trans_stop(env, dev, th);
2128
2129         /* We are not sure whether the child will become orphan or not.
2130          * Record it in the LFSCK trace file for further checking in
2131          * the second-stage scanning. */
2132         if (!update && !dec && rc == 0)
2133                 lfsck_namespace_trace_update(env, com, cfid,
2134                                              LNTF_CHECK_LINKEA, true);
2135
2136 unlock1:
2137         /* It is harmless even if unlock the unused lock_handle */
2138         lfsck_ibits_unlock(&lh, LCK_PW);
2139         lfsck_unlock(llh);
2140
2141 log:
2142         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2143                "entry for: parent "DFID", child "DFID", name %s, type "
2144                "in name entry %o, type claimed by child %o. repair it "
2145                "by %s with new name2 %s: rc = %d\n",
2146                lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2147                name, type, update ? lfsck_object_type(child) : 0,
2148                update ? "updating" : "removing", name2, rc);
2149
2150         if (rc != 0) {
2151                 struct lfsck_namespace *ns = com->lc_file_ram;
2152
2153                 ns->ln_flags |= LF_INCONSISTENT;
2154         }
2155
2156         return rc;
2157 }
2158
2159 /**
2160  * Update the ".." name entry for the given object.
2161  *
2162  * The object's ".." is corrupted, this function will update the ".." name
2163  * entry with the given pfid, and the linkEA with the given ldata.
2164  *
2165  * The caller should take the ldlm lock before the calling.
2166  *
2167  * \param[in] env       pointer to the thread context
2168  * \param[in] com       pointer to the lfsck component
2169  * \param[in] obj       pointer to the dt_object to be handled
2170  * \param[in] pfid      the new fid for the object's ".." name entry
2171  * \param[in] cname     the name for the @obj in the parent directory
2172  *
2173  * \retval              positive number for repaired cases
2174  * \retval              0 if nothing to be repaired
2175  * \retval              negative error number on failure
2176  */
2177 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2178                                                   struct lfsck_component *com,
2179                                                   struct dt_object *obj,
2180                                                   const struct lu_fid *pfid,
2181                                                   struct lu_name *cname)
2182 {
2183         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2184         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2185         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2186         struct dt_device                *dev    = lfsck_obj2dev(obj);
2187         struct thandle                  *th     = NULL;
2188         struct linkea_data               ldata  = { NULL };
2189         struct lu_buf                    linkea_buf;
2190         int                              rc     = 0;
2191         ENTRY;
2192
2193         LASSERT(!dt_object_remote(obj));
2194         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2195
2196         rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2197         if (rc != 0)
2198                 GOTO(log, rc);
2199
2200         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2201                        ldata.ld_leh->leh_len);
2202
2203         th = dt_trans_create(env, dev);
2204         if (IS_ERR(th))
2205                 GOTO(log, rc = PTR_ERR(th));
2206
2207         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2208         if (rc != 0)
2209                 GOTO(stop, rc);
2210
2211         rec->rec_type = S_IFDIR;
2212         rec->rec_fid = pfid;
2213         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2214                                (const struct dt_key *)dotdot, th);
2215         if (rc != 0)
2216                 GOTO(stop, rc);
2217
2218         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2219                                   XATTR_NAME_LINK, 0, th);
2220         if (rc != 0)
2221                 GOTO(stop, rc);
2222
2223         rc = dt_trans_start_local(env, dev, th);
2224         if (rc != 0)
2225                 GOTO(stop, rc);
2226
2227         dt_write_lock(env, obj, 0);
2228         if (unlikely(lfsck_is_dead_obj(obj)))
2229                 GOTO(unlock, rc = 0);
2230
2231         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2232                 GOTO(unlock, rc = 1);
2233
2234         /* The old ".." name entry maybe not exist. */
2235         dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2236
2237         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2238                        (const struct dt_key *)dotdot, th, 1);
2239         if (rc != 0)
2240                 GOTO(unlock, rc);
2241
2242         rc = lfsck_links_write(env, obj, &ldata, th);
2243
2244         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2245
2246 unlock:
2247         dt_write_unlock(env, obj);
2248
2249 stop:
2250         dt_trans_stop(env, dev, th);
2251
2252 log:
2253         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2254                "the object "DFID", new parent "DFID": rc = %d\n",
2255                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2256                PFID(pfid), rc);
2257
2258         if (rc != 0) {
2259                 struct lfsck_namespace *ns = com->lc_file_ram;
2260
2261                 ns->ln_flags |= LF_INCONSISTENT;
2262         }
2263
2264         return rc;
2265 }
2266
2267 /**
2268  * Handle orphan @obj during Double Scan Directory.
2269  *
2270  * Remove the @obj's current (invalid) linkEA entries, and insert
2271  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2272  * ${FID}-${PFID}-D-${conflict_version}
2273  *
2274  * The caller should take the ldlm lock before the calling.
2275  *
2276  * \param[in] env       pointer to the thread context
2277  * \param[in] com       pointer to the lfsck component
2278  * \param[in] obj       pointer to the orphan object to be handled
2279  * \param[in] pfid      the new fid for the object's ".." name entry
2280  * \param[in,out] lh    ldlm lock handler for the given @obj
2281  * \param[out] type     to tell the caller what the inconsistency is
2282  *
2283  * \retval              positive number for repaired cases
2284  * \retval              0 if nothing to be repaired
2285  * \retval              negative error number on failure
2286  */
2287 static int
2288 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2289                            struct lfsck_component *com,
2290                            struct dt_object *obj,
2291                            const struct lu_fid *pfid,
2292                            struct lustre_handle *lh,
2293                            enum lfsck_namespace_inconsistency_type *type)
2294 {
2295         struct lfsck_thread_info *info = lfsck_env_info(env);
2296         struct lfsck_namespace   *ns   = com->lc_file_ram;
2297         int                       rc;
2298         ENTRY;
2299
2300         /* Remove the unrecognized linkEA. */
2301         rc = lfsck_namespace_links_remove(env, com, obj);
2302         lfsck_ibits_unlock(lh, LCK_EX);
2303         if (rc < 0 && rc != -ENODATA)
2304                 RETURN(rc);
2305
2306         *type = LNIT_MUL_REF;
2307
2308         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2309          * ever tried to verify some remote MDT-object that resides on this
2310          * MDT, but this MDT failed to respond such request. So means there
2311          * may be some remote name entry on other MDT that references this
2312          * object with another name, so we cannot know whether this linkEA
2313          * is valid or not. So keep it there and maybe resolved when next
2314          * LFSCK run. */
2315         if (ns->ln_flags & LF_INCOMPLETE)
2316                 RETURN(0);
2317
2318         /* The unique linkEA is invalid, even if the ".." name entry may be
2319          * valid, we still cannot know via which name entry this directory
2320          * will be referenced. Then handle it as pure orphan. */
2321         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2322                  "-"DFID, PFID(pfid));
2323         rc = lfsck_namespace_insert_orphan(env, com, obj,
2324                                            info->lti_tmpbuf, "D", NULL);
2325
2326         RETURN(rc);
2327 }
2328
2329 /**
2330  * Double Scan Directory object for single linkEA entry case.
2331  *
2332  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2333  * then check whether the name is in the namespace or not, if not, add the
2334  * missing name entry back to namespace. If the linkEA entry is invalid,
2335  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2336  * as an orphan.
2337  *
2338  * \param[in] env       pointer to the thread context
2339  * \param[in] com       pointer to the lfsck component
2340  * \param[in] child     pointer to the directory to be double scanned
2341  * \param[in] pfid      the FID corresponding to the ".." entry
2342  * \param[in] ldata     pointer to the linkEA data for the given @child
2343  * \param[in,out] lh    ldlm lock handler for the given @child
2344  * \param[out] type     to tell the caller what the inconsistency is
2345  * \param[in] retry     if found inconsistency, but the caller does not hold
2346  *                      ldlm lock on the @child, then set @retry as true
2347  * \param[in] unknown   set if does not know how to repair the inconsistency
2348  *
2349  * \retval              positive number for repaired cases
2350  * \retval              0 if nothing to be repaired
2351  * \retval              negative error number on failure
2352  */
2353 static int
2354 lfsck_namespace_dsd_single(const struct lu_env *env,
2355                            struct lfsck_component *com,
2356                            struct dt_object *child,
2357                            const struct lu_fid *pfid,
2358                            struct linkea_data *ldata,
2359                            struct lustre_handle *lh,
2360                            enum lfsck_namespace_inconsistency_type *type,
2361                            bool *retry, bool *unknown)
2362 {
2363         struct lfsck_thread_info *info          = lfsck_env_info(env);
2364         struct lu_name           *cname         = &info->lti_name;
2365         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2366         struct lu_fid             tfid;
2367         struct lfsck_namespace   *ns            = com->lc_file_ram;
2368         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2369         struct dt_object         *parent        = NULL;
2370         struct lmv_mds_md_v1     *lmv;
2371         int                       rc            = 0;
2372         ENTRY;
2373
2374         rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2375                                                  info->lti_key,
2376                                                  sizeof(info->lti_key));
2377         /* The unique linkEA entry with bad parent will be handled as orphan. */
2378         if (rc != 0) {
2379                 if (!lustre_handle_is_used(lh) && retry != NULL)
2380                         *retry = true;
2381                 else
2382                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2383                                                         pfid, lh, type);
2384
2385                 GOTO(out, rc);
2386         }
2387
2388         parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2389         if (IS_ERR(parent))
2390                 GOTO(out, rc = PTR_ERR(parent));
2391
2392         /* We trust the unique linkEA entry in spite of whether it matches the
2393          * ".." name entry or not. Because even if the linkEA entry is wrong
2394          * and the ".." name entry is right, we still cannot know via which
2395          * name entry the child will be referenced, since all known entries
2396          * have been verified during the first-stage scanning. */
2397         if (!dt_object_exists(parent)) {
2398                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2399                  * has ever tried to verify some remote MDT-object that resides
2400                  * on this MDT, but this MDT failed to respond such request. So
2401                  * means there may be some remote name entry on other MDT that
2402                  * references this object with another name, so we cannot know
2403                  * whether this linkEA is valid or not. So keep it there and
2404                  * maybe resolved when next LFSCK run. */
2405                 if (ns->ln_flags & LF_INCOMPLETE)
2406                         GOTO(out, rc = 0);
2407
2408                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2409                         *retry = true;
2410
2411                         GOTO(out, rc = 0);
2412                 }
2413
2414                 lfsck_ibits_unlock(lh, LCK_EX);
2415
2416 lost_parent:
2417                 lmv = &info->lti_lmv;
2418                 rc = lfsck_read_stripe_lmv(env, child, lmv);
2419                 if (rc != 0 && rc != -ENODATA)
2420                         GOTO(out, rc);
2421
2422                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2423                         lmv = NULL;
2424                 } else if (lfsck_shard_name_to_index(env,
2425                                         cname->ln_name, cname->ln_namelen,
2426                                         S_IFDIR, cfid) < 0) {
2427                         /* It is an invalid name entry, we
2428                          * cannot trust the parent also. */
2429                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2430                                                 ldata, cname, &tfid, true);
2431                         if (rc < 0)
2432                                 GOTO(out, rc);
2433
2434                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2435                                  "-"DFID, PFID(pfid));
2436                         rc = lfsck_namespace_insert_orphan(env, com, child,
2437                                                 info->lti_tmpbuf, "S", NULL);
2438
2439                         GOTO(out, rc);
2440                 }
2441
2442                 /* Create the lost parent as an orphan. */
2443                 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2444                 if (rc >= 0) {
2445                         /* Add the missing name entry to the parent. */
2446                         rc = lfsck_namespace_insert_normal(env, com, parent,
2447                                                         child, cname->ln_name);
2448                         if (unlikely(rc == -EEXIST)) {
2449                                 /* Unfortunately, someone reused the name
2450                                  * under the parent by race. So we have
2451                                  * to remove the linkEA entry from
2452                                  * current child object. It means that the
2453                                  * LFSCK cannot recover the system
2454                                  * totally back to its original status,
2455                                  * but it is necessary to make the
2456                                  * current system to be consistent. */
2457                                 rc = lfsck_namespace_shrink_linkea(env,
2458                                                 com, child, ldata,
2459                                                 cname, &tfid, true);
2460                                 if (rc >= 0) {
2461                                         snprintf(info->lti_tmpbuf,
2462                                                  sizeof(info->lti_tmpbuf),
2463                                                  "-"DFID, PFID(pfid));
2464                                         rc = lfsck_namespace_insert_orphan(env,
2465                                                 com, child, info->lti_tmpbuf,
2466                                                 "D", NULL);
2467                                 }
2468                         }
2469                 }
2470
2471                 GOTO(out, rc);
2472         } /* !dt_object_exists(parent) */
2473
2474         /* The unique linkEA entry with bad parent will be handled as orphan. */
2475         if (unlikely(!dt_try_as_dir(env, parent))) {
2476                 if (!lustre_handle_is_used(lh) && retry != NULL)
2477                         *retry = true;
2478                 else
2479                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2480                                                         pfid, lh, type);
2481
2482                 GOTO(out, rc);
2483         }
2484
2485         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2486                        (const struct dt_key *)cname->ln_name);
2487         if (rc == -ENOENT) {
2488                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2489                  * has ever tried to verify some remote MDT-object that resides
2490                  * on this MDT, but this MDT failed to respond such request. So
2491                  * means there may be some remote name entry on other MDT that
2492                  * references this object with another name, so we cannot know
2493                  * whether this linkEA is valid or not. So keep it there and
2494                  * maybe resolved when next LFSCK run. */
2495                 if (ns->ln_flags & LF_INCOMPLETE)
2496                         GOTO(out, rc = 0);
2497
2498                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2499                         *retry = true;
2500
2501                         GOTO(out, rc = 0);
2502                 }
2503
2504                 lfsck_ibits_unlock(lh, LCK_EX);
2505                 rc = lfsck_namespace_check_name(env, parent, child, cname);
2506                 if (rc == -ENOENT)
2507                         goto lost_parent;
2508
2509                 if (rc < 0)
2510                         GOTO(out, rc);
2511
2512                 /* It is an invalid name entry, drop it. */
2513                 if (unlikely(rc > 0)) {
2514                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2515                                                 ldata, cname, &tfid, true);
2516                         if (rc >= 0) {
2517                                 snprintf(info->lti_tmpbuf,
2518                                          sizeof(info->lti_tmpbuf),
2519                                          "-"DFID, PFID(pfid));
2520                                 rc = lfsck_namespace_insert_orphan(env, com,
2521                                         child, info->lti_tmpbuf, "D", NULL);
2522                         }
2523
2524                         GOTO(out, rc);
2525                 }
2526
2527                 /* Add the missing name entry back to the namespace. */
2528                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2529                                                    cname->ln_name);
2530                 if (unlikely(rc == -ESTALE))
2531                         /* It may happen when the remote object has been
2532                          * removed, but the local MDT is not aware of that. */
2533                         goto lost_parent;
2534
2535                 if (unlikely(rc == -EEXIST)) {
2536                         /* Unfortunately, someone reused the name under the
2537                          * parent by race. So we have to remove the linkEA
2538                          * entry from current child object. It means that the
2539                          * LFSCK cannot recover the system totally back to
2540                          * its original status, but it is necessary to make
2541                          * the current system to be consistent.
2542                          *
2543                          * It also may be because of the LFSCK found some
2544                          * internal status of create operation. Under such
2545                          * case, nothing to be done. */
2546                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2547                                         parent, child, ldata, cname, &tfid);
2548                         if (rc >= 0) {
2549                                 snprintf(info->lti_tmpbuf,
2550                                          sizeof(info->lti_tmpbuf),
2551                                          "-"DFID, PFID(pfid));
2552                                 rc = lfsck_namespace_insert_orphan(env, com,
2553                                         child, info->lti_tmpbuf, "D", NULL);
2554                         }
2555                 }
2556
2557                 GOTO(out, rc);
2558         } /* rc == -ENOENT */
2559
2560         if (rc != 0)
2561                 GOTO(out, rc);
2562
2563         if (!lu_fid_eq(&tfid, cfid)) {
2564                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2565                         *retry = true;
2566
2567                         GOTO(out, rc = 0);
2568                 }
2569
2570                 lfsck_ibits_unlock(lh, LCK_EX);
2571                 /* The name entry references another MDT-object that
2572                  * may be created by the LFSCK for repairing dangling
2573                  * name entry. Try to replace it. */
2574                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2575                                                   &tfid, cname);
2576                 if (rc == 0)
2577                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2578                                                         pfid, lh, type);
2579
2580                 GOTO(out, rc);
2581         }
2582
2583         /* Zero FID may because the remote directroy object has invalid linkEA,
2584          * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2585          * how to repair the inconsistency, but the namespace LFSCK on the MDT
2586          * where its name entry resides may has more information (name, FID) to
2587          * repair such inconsistency. So here, keep the inconsistency to avoid
2588          * some imporper repairing. */
2589         if (fid_is_zero(pfid)) {
2590                 if (unknown)
2591                         *unknown = true;
2592
2593                 GOTO(out, rc = 0);
2594         }
2595
2596         /* The ".." name entry is wrong, update it. */
2597         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2598                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2599                         *retry = true;
2600
2601                         GOTO(out, rc = 0);
2602                 }
2603
2604                 *type = LNIT_UNMATCHED_PAIRS;
2605                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2606                                                 lfsck_dto2fid(parent), cname);
2607         }
2608
2609         GOTO(out, rc);
2610
2611 out:
2612         if (parent != NULL && !IS_ERR(parent))
2613                 lfsck_object_put(env, parent);
2614
2615         return rc;
2616 }
2617
2618 /**
2619  * Double Scan Directory object for multiple linkEA entries case.
2620  *
2621  * The given @child has multiple linkEA entries. There is at most one linkEA
2622  * entry will be valid, all the others will be removed. Firstly, the function
2623  * will try to find out the linkEA entry for which the name entry exists under
2624  * the given parent (@pfid). If there is no linkEA entry that matches the given
2625  * ".." name entry, then tries to find out the first linkEA entry that both the
2626  * parent and the name entry exist to rebuild a new ".." name entry.
2627  *
2628  * \param[in] env       pointer to the thread context
2629  * \param[in] com       pointer to the lfsck component
2630  * \param[in] child     pointer to the directory to be double scanned
2631  * \param[in] pfid      the FID corresponding to the ".." entry
2632  * \param[in] ldata     pointer to the linkEA data for the given @child
2633  * \param[in,out] lh    ldlm lock handler for the given @child
2634  * \param[out] type     to tell the caller what the inconsistency is
2635  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2636  * \param[in] unknown   set if does not know how to repair the inconsistency
2637  *
2638  * \retval              positive number for repaired cases
2639  * \retval              0 if nothing to be repaired
2640  * \retval              negative error number on failure
2641  */
2642 static int
2643 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2644                              struct lfsck_component *com,
2645                              struct dt_object *child,
2646                              const struct lu_fid *pfid,
2647                              struct linkea_data *ldata,
2648                              struct lustre_handle *lh,
2649                              enum lfsck_namespace_inconsistency_type *type,
2650                              bool lpf, bool *unknown)
2651 {
2652         struct lfsck_thread_info *info          = lfsck_env_info(env);
2653         struct lu_name           *cname         = &info->lti_name;
2654         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2655         struct lu_fid            *pfid2         = &info->lti_fid3;
2656         struct lu_fid             tfid;
2657         struct lfsck_namespace   *ns            = com->lc_file_ram;
2658         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2659         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2660         struct dt_object         *parent        = NULL;
2661         struct linkea_data        ldata_new     = { NULL };
2662         int                       dirent_count  = 0;
2663         int                       rc            = 0;
2664         bool                      once          = true;
2665         ENTRY;
2666
2667 again:
2668         while (ldata->ld_lee != NULL) {
2669                 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2670                                                          info->lti_key,
2671                                                          sizeof(info->lti_key));
2672                 /* Drop invalid linkEA entry. */
2673                 if (rc != 0) {
2674                         lfsck_linkea_del_buf(ldata, cname);
2675                         continue;
2676                 }
2677
2678                 /* Drop repeated linkEA entries. */
2679                 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2680
2681                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2682                  * then it is possible that: the directry object has ever
2683                  * been lost, but its name entry was there. In the former
2684                  * LFSCK run, during the first-stage scanning, the LFSCK
2685                  * found the dangling name entry, but it did not recreate
2686                  * the lost object, and when moved to the second-stage
2687                  * scanning, some children objects of the lost directory
2688                  * object were found, then the LFSCK recreated such lost
2689                  * directory object as an orphan.
2690                  *
2691                  * When the LFSCK runs again, if the dangling name is still
2692                  * there, the LFSCK should move the orphan directory object
2693                  * back to the normal namespace. */
2694                 if (!lpf && !fid_is_zero(pfid) &&
2695                     !lu_fid_eq(pfid, &tfid) && once) {
2696                         linkea_next_entry(ldata);
2697                         continue;
2698                 }
2699
2700                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2701                 if (IS_ERR(parent))
2702                         RETURN(PTR_ERR(parent));
2703
2704                 if (!dt_object_exists(parent)) {
2705                         lfsck_object_put(env, parent);
2706                         if (ldata->ld_leh->leh_reccount > 1) {
2707                                 /* If it is NOT the last linkEA entry, then
2708                                  * there is still other chance to make the
2709                                  * child to be visible via other parent, then
2710                                  * remove this linkEA entry. */
2711                                 lfsck_linkea_del_buf(ldata, cname);
2712                                 continue;
2713                         }
2714
2715                         break;
2716                 }
2717
2718                 /* The linkEA entry with bad parent will be removed. */
2719                 if (unlikely(!dt_try_as_dir(env, parent))) {
2720                         lfsck_object_put(env, parent);
2721                         lfsck_linkea_del_buf(ldata, cname);
2722                         continue;
2723                 }
2724
2725                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2726                                (const struct dt_key *)cname->ln_name);
2727                 *pfid2 = *lfsck_dto2fid(parent);
2728                 if (rc == -ENOENT) {
2729                         lfsck_object_put(env, parent);
2730                         linkea_next_entry(ldata);
2731                         continue;
2732                 }
2733
2734                 if (rc != 0) {
2735                         lfsck_object_put(env, parent);
2736
2737                         RETURN(rc);
2738                 }
2739
2740                 if (lu_fid_eq(&tfid, cfid)) {
2741                         lfsck_object_put(env, parent);
2742                         /* If the parent (that is declared via linkEA entry)
2743                          * directory contains the specified child, but such
2744                          * parent does not match the dotdot name entry, then
2745                          * trust the linkEA. */
2746                         if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2747                                 *type = LNIT_UNMATCHED_PAIRS;
2748                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2749                                                 com, child, pfid2, cname);
2750
2751                                 RETURN(rc);
2752                         }
2753
2754 rebuild:
2755                         /* It is the most common case that we find the
2756                          * name entry corresponding to the linkEA entry
2757                          * that matches the ".." name entry. */
2758                         rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2759                                               cname, pfid2);
2760                         if (rc != 0)
2761                                 RETURN(rc);
2762
2763                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2764                                                             &ldata_new);
2765                         if (rc < 0)
2766                                 RETURN(rc);
2767
2768                         lfsck_linkea_del_buf(ldata, cname);
2769                         linkea_first_entry(ldata);
2770                         /* There may be some invalid dangling name entries under
2771                          * other parent directories, remove all of them. */
2772                         while (ldata->ld_lee != NULL) {
2773                                 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2774                                                 cname, &tfid, info->lti_key,
2775                                                 sizeof(info->lti_key));
2776                                 if (rc != 0)
2777                                         goto next;
2778
2779                                 parent = lfsck_object_find_bottom(env, lfsck,
2780                                                                   &tfid);
2781                                 if (IS_ERR(parent)) {
2782                                         rc = PTR_ERR(parent);
2783                                         if (rc != -ENOENT &&
2784                                             bk->lb_param & LPF_FAILOUT)
2785                                                 RETURN(rc);
2786
2787                                         goto next;
2788                                 }
2789
2790                                 if (!dt_object_exists(parent)) {
2791                                         lfsck_object_put(env, parent);
2792                                         goto next;
2793                                 }
2794
2795                                 rc = lfsck_namespace_repair_dirent(env, com,
2796                                         parent, child, cname->ln_name,
2797                                         cname->ln_name, S_IFDIR, false, true);
2798                                 lfsck_object_put(env, parent);
2799                                 if (rc < 0) {
2800                                         if (bk->lb_param & LPF_FAILOUT)
2801                                                 RETURN(rc);
2802
2803                                         goto next;
2804                                 }
2805
2806                                 dirent_count += rc;
2807
2808 next:
2809                                 lfsck_linkea_del_buf(ldata, cname);
2810                         }
2811
2812                         ns->ln_dirent_repaired += dirent_count;
2813
2814                         RETURN(rc);
2815                 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
2816
2817                 lfsck_ibits_unlock(lh, LCK_EX);
2818                 /* The name entry references another MDT-object that may be
2819                  * created by the LFSCK for repairing dangling name entry.
2820                  * Try to replace it. */
2821                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2822                                                   &tfid, cname);
2823                 lfsck_object_put(env, parent);
2824                 if (rc < 0)
2825                         RETURN(rc);
2826
2827                 if (rc > 0)
2828                         goto rebuild;
2829
2830                 lfsck_linkea_del_buf(ldata, cname);
2831         } /* while (ldata->ld_lee != NULL) */
2832
2833         /* If there is still linkEA overflow, return. */
2834         if (unlikely(ldata->ld_leh->leh_overflow_time))
2835                 RETURN(0);
2836
2837         linkea_first_entry(ldata);
2838         if (ldata->ld_leh->leh_reccount == 1) {
2839                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2840                                                 lh, type, NULL, unknown);
2841
2842                 RETURN(rc);
2843         }
2844
2845         /* All linkEA entries are invalid and removed, then handle the @child
2846          * as an orphan.*/
2847         if (ldata->ld_leh->leh_reccount == 0) {
2848                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2849                                                 type);
2850
2851                 RETURN(rc);
2852         }
2853
2854         /* If the dangling name entry for the orphan directory object has
2855          * been remvoed, then just check whether the directory object is
2856          * still under the .lustre/lost+found/MDTxxxx/ or not. */
2857         if (lpf) {
2858                 lpf = false;
2859                 goto again;
2860         }
2861
2862         /* There is no linkEA entry that matches the ".." name entry. Find
2863          * the first linkEA entry that both parent and name entry exist to
2864          * rebuild a new ".." name entry. */
2865         if (once) {
2866                 once = false;
2867                 goto again;
2868         }
2869
2870         RETURN(rc);
2871 }
2872
2873 /**
2874  * Repair the object's nlink attribute.
2875  *
2876  * If all the known name entries have been verified, then the object's hard
2877  * link attribute should match the object's linkEA entries count unless the
2878  * object's has too many hard link to be recorded in the linkEA. Such cases
2879  * should have been marked in the LFSCK trace file. Otherwise, trust the
2880  * linkEA to update the object's nlink attribute.
2881  *
2882  * \param[in] env       pointer to the thread context
2883  * \param[in] com       pointer to the lfsck component
2884  * \param[in] obj       pointer to the dt_object to be handled
2885  * \param[in,out] la    pointer to buffer to object's attribute before
2886  *                      and after the repairing
2887  *
2888  * \retval              positive number for repaired cases
2889  * \retval              0 if nothing to be repaired
2890  * \retval              negative error number on failure
2891  */
2892 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2893                                         struct lfsck_component *com,
2894                                         struct dt_object *obj,
2895                                         struct lu_attr *la)
2896 {
2897         struct lfsck_namespace          *ns     = com->lc_file_ram;
2898         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2899         struct dt_device                *dev    = lfsck_obj2dev(obj);
2900         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
2901         struct thandle                  *th     = NULL;
2902         struct linkea_data               ldata  = { NULL };
2903         struct lustre_handle             lh     = { 0 };
2904         __u32                            old    = la->la_nlink;
2905         int                              rc     = 0;
2906         ENTRY;
2907
2908         LASSERT(!dt_object_remote(obj));
2909
2910         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2911                               MDS_INODELOCK_UPDATE, LCK_PW);
2912         if (rc != 0)
2913                 GOTO(log, rc);
2914
2915         th = dt_trans_create(env, dev);
2916         if (IS_ERR(th))
2917                 GOTO(log, rc = PTR_ERR(th));
2918
2919         la->la_valid = LA_NLINK;
2920         rc = dt_declare_attr_set(env, obj, la, th);
2921         if (rc != 0)
2922                 GOTO(stop, rc);
2923
2924         rc = dt_trans_start_local(env, dev, th);
2925         if (rc != 0)
2926                 GOTO(stop, rc);
2927
2928         dt_write_lock(env, obj, 0);
2929         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2930          * ever tried to verify some remote MDT-object that resides on this
2931          * MDT, but this MDT failed to respond such request. So means there
2932          * may be some remote name entry on other MDT that references this
2933          * object with another name, so we cannot know whether this linkEA
2934          * is valid or not. So keep it there and maybe resolved when next
2935          * LFSCK run. */
2936         if (ns->ln_flags & LF_INCOMPLETE)
2937                 GOTO(unlock, rc = 0);
2938
2939         rc = dt_attr_get(env, obj, la);
2940         if (rc != 0)
2941                 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2942
2943         rc = lfsck_links_read2_with_rec(env, obj, &ldata);
2944         if (rc)
2945                 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2946
2947         /* XXX: Currently, we only update the nlink attribute if the known
2948          *      linkEA entries is larger than the nlink attribute. That is
2949          *      safe action. */
2950         if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
2951             unlikely(la->la_nlink == 0 ||
2952                      ldata.ld_leh->leh_overflow_time))
2953                 GOTO(unlock, rc = 0);
2954
2955         la->la_nlink = ldata.ld_leh->leh_reccount;
2956         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2957                 GOTO(unlock, rc = 1);
2958
2959         rc = dt_attr_set(env, obj, la, th);
2960
2961         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2962
2963 unlock:
2964         dt_write_unlock(env, obj);
2965
2966 stop:
2967         dt_trans_stop(env, dev, th);
2968
2969 log:
2970         lfsck_ibits_unlock(&lh, LCK_PW);
2971         CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2972                "nlink count from %u to %u: rc = %d\n",
2973                lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2974
2975         if (rc != 0)
2976                 ns->ln_flags |= LF_INCONSISTENT;
2977
2978         return rc;
2979 }
2980
2981 /**
2982  * Double scan the directory object for namespace LFSCK.
2983  *
2984  * This function will verify the <parent, child> pairs in the namespace tree:
2985  * the parent references the child via some name entry that should be in the
2986  * child's linkEA entry, the child should back references the parent via its
2987  * ".." name entry.
2988  *
2989  * The LFSCK will scan every linkEA entry in turn until find out the first
2990  * matched pairs. If found, then all other linkEA entries will be dropped.
2991  * If all the linkEA entries cannot match the ".." name entry, then there
2992  * are serveral possible cases:
2993  *
2994  * 1) If there is only one linkEA entry, then trust it as long as the PFID
2995  *    in the linkEA entry is valid.
2996  *
2997  * 2) If there are multiple linkEA entries, then try to find the linkEA
2998  *    that matches the ".." name entry. If found, then all other entries
2999  *    are invalid; otherwise, it is quite possible that the ".." name entry
3000  *    is corrupted. Under such case, the LFSCK will rebuild the ".." name
3001  *    entry according to the first valid linkEA entry (both the parent and
3002  *    the name entry should exist).
3003  *
3004  * 3) If the directory object has no (valid) linkEA entry, then the
3005  *    directory object will be handled as pure orphan and inserted
3006  *    in the .lustre/lost+found/MDTxxxx/ with the name:
3007  *    ${self_FID}-${PFID}-D-${conflict_version}
3008  *
3009  * \param[in] env       pointer to the thread context
3010  * \param[in] com       pointer to the lfsck component
3011  * \param[in] child     pointer to the directory object to be handled
3012  * \param[in] flags     to indicate the specical checking on the @child
3013  *
3014  * \retval              positive number for repaired cases
3015  * \retval              0 if nothing to be repaired
3016  * \retval              negative error number on failure
3017  */
3018 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3019                                            struct lfsck_component *com,
3020                                            struct dt_object *child, __u8 flags)
3021 {
3022         struct lfsck_thread_info *info          = lfsck_env_info(env);
3023         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
3024         struct lu_fid            *pfid          = &info->lti_fid2;
3025         struct lfsck_namespace   *ns            = com->lc_file_ram;
3026         struct lfsck_instance    *lfsck         = com->lc_lfsck;
3027         struct lustre_handle      lh            = { 0 };
3028         struct linkea_data        ldata         = { NULL };
3029         bool                      unknown       = false;
3030         bool                      lpf           = false;
3031         bool                      retry         = false;
3032         enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3033         int                       rc            = 0;
3034         ENTRY;
3035
3036         LASSERT(!dt_object_remote(child));
3037
3038         if (flags & LNTF_UNCERTAIN_LMV) {
3039                 if (flags & LNTF_RECHECK_NAME_HASH) {
3040                         rc = lfsck_namespace_scan_shard(env, com, child);
3041                         if (rc < 0)
3042                                 RETURN(rc);
3043
3044                         ns->ln_striped_shards_scanned++;
3045                 } else {
3046                         ns->ln_striped_shards_skipped++;
3047                 }
3048         }
3049
3050         flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3051         if (flags == 0)
3052                 RETURN(0);
3053
3054         if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3055             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3056                 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3057                        "the namespace LFSCK, then the LFSCK cannot guarantee"
3058                        "all the name entries have been verified in first-stage"
3059                        "scanning. So have to skip orphan related handling for"
3060                        "the directory object "DFID" with remote name entry\n",
3061                        lfsck_lfsck2name(lfsck), PFID(cfid));
3062
3063                 RETURN(0);
3064         }
3065
3066         if (unlikely(!dt_try_as_dir(env, child)))
3067                 GOTO(out, rc = -ENOTDIR);
3068
3069         /* We only take ldlm lock on the @child when required. When the
3070          * logic comes here for the first time, it is always false. */
3071         if (0) {
3072
3073 lock:
3074                 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3075                                       MDS_INODELOCK_UPDATE |
3076                                       MDS_INODELOCK_XATTR, LCK_EX);
3077                 if (rc != 0)
3078                         GOTO(out, rc);
3079         }
3080
3081         dt_read_lock(env, child, 0);
3082         if (unlikely(lfsck_is_dead_obj(child))) {
3083                 dt_read_unlock(env, child);
3084
3085                 GOTO(out, rc = 0);
3086         }
3087
3088         rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3089                        (const struct dt_key *)dotdot);
3090         if (rc != 0) {
3091                 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3092                         dt_read_unlock(env, child);
3093
3094                         GOTO(out, rc);
3095                 }
3096
3097                 if (!lustre_handle_is_used(&lh)) {
3098                         dt_read_unlock(env, child);
3099                         goto lock;
3100                 }
3101
3102                 fid_zero(pfid);
3103         } else if (lfsck->li_lpf_obj != NULL &&
3104                    lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3105                 lpf = true;
3106         } else if (unlikely(!fid_is_sane(pfid))) {
3107                 fid_zero(pfid);
3108         }
3109
3110         rc = lfsck_links_read(env, child, &ldata);
3111         dt_read_unlock(env, child);
3112         if (rc != 0) {
3113                 if (rc != -ENODATA && rc != -EINVAL)
3114                         GOTO(out, rc);
3115
3116                 if (!lustre_handle_is_used(&lh))
3117                         goto lock;
3118
3119                 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3120                         /* Remove the corrupted linkEA. */
3121                         rc = lfsck_namespace_links_remove(env, com, child);
3122                         if (rc == 0)
3123                                 /* Here, because of the crashed linkEA, we
3124                                  * cannot know whether there is some parent
3125                                  * that references the child directory via
3126                                  * some name entry or not. So keep it there,
3127                                  * when the LFSCK run next time, if there is
3128                                  * some parent that references this object,
3129                                  * then the LFSCK can rebuild the linkEA;
3130                                  * otherwise, this object will be handled
3131                                  * as orphan as above. */
3132                                 unknown = true;
3133                 } else {
3134                         /* 1. If we have neither ".." nor linkEA,
3135                          *    then it is an orphan.
3136                          *
3137                          * 2. If we only have the ".." name entry,
3138                          *    but no parent references this child
3139                          *    directory, then handle it as orphan. */
3140                         lfsck_ibits_unlock(&lh, LCK_EX);
3141                         type = LNIT_MUL_REF;
3142
3143                         /* If the LFSCK is marked as LF_INCOMPLETE,
3144                          * then means some MDT has ever tried to
3145                          * verify some remote MDT-object that resides
3146                          * on this MDT, but this MDT failed to respond
3147                          * such request. So means there may be some
3148                          * remote name entry on other MDT that
3149                          * references this object with another name,
3150                          * so we cannot know whether this linkEA is
3151                          * valid or not. So keep it there and maybe
3152                          * resolved when next LFSCK run. */
3153                         if (ns->ln_flags & LF_INCOMPLETE)
3154                                 GOTO(out, rc = 0);
3155
3156                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3157                                  "-"DFID, PFID(pfid));
3158                         rc = lfsck_namespace_insert_orphan(env, com, child,
3159                                                 info->lti_tmpbuf, "D", NULL);
3160                 }
3161
3162                 GOTO(out, rc);
3163         } /* rc != 0 */
3164
3165         linkea_first_entry(&ldata);
3166         /* This is the most common case: the object has unique linkEA entry. */
3167         if (ldata.ld_leh->leh_reccount == 1) {
3168                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3169                                                 &lh, &type, &retry, &unknown);
3170                 if (retry) {
3171                         LASSERT(!lustre_handle_is_used(&lh));
3172
3173                         retry = false;
3174                         goto lock;
3175                 }
3176
3177                 GOTO(out, rc);
3178         }
3179
3180         if (!lustre_handle_is_used(&lh))
3181                 goto lock;
3182
3183         if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3184                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3185                                                 &type);
3186
3187                 GOTO(out, rc);
3188         }
3189
3190         /* When we come here, the cases usually like that:
3191          * 1) The directory object has a corrupted linkEA entry. During the
3192          *    first-stage scanning, the LFSCK cannot know such corruption,
3193          *    then it appends the right linkEA entry according to the found
3194          *    name entry after the bad one.
3195          *
3196          * 2) The directory object has a right linkEA entry. During the
3197          *    first-stage scanning, the LFSCK finds some bad name entry,
3198          *    but the LFSCK cannot aware that at that time, then it adds
3199          *    the bad linkEA entry for further processing. */
3200         rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3201                                           &lh, &type, lpf, &unknown);
3202
3203         GOTO(out, rc);
3204
3205 out:
3206         lfsck_ibits_unlock(&lh, LCK_EX);
3207         if (rc > 0) {
3208                 switch (type) {
3209                 case LNIT_BAD_LINKEA:
3210                         ns->ln_linkea_repaired++;
3211                         break;
3212                 case LNIT_UNMATCHED_PAIRS:
3213                         ns->ln_unmatched_pairs_repaired++;
3214                         break;
3215                 case LNIT_MUL_REF:
3216                         ns->ln_mul_ref_repaired++;
3217                         break;
3218                 default:
3219                         break;
3220                 }
3221         }
3222
3223         if (unknown)
3224                 ns->ln_unknown_inconsistency++;
3225
3226         return rc;
3227 }
3228
3229 static inline bool
3230 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3231                                       struct lfsck_namespace *ns)
3232 {
3233         /* Both the leh_overflow_time and ln_time_latest_reset are
3234          * local time based, so need NOT to care about clock drift
3235          * among the servers. */
3236         return ldata->ld_leh->leh_overflow_time &&
3237                ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3238 }
3239
3240 /**
3241  * Clear the object's linkEA overflow timestamp.
3242  *
3243  * If the MDT-object has too many hard links as to the linkEA cannot hold
3244  * all of them, then overflow timestamp will be set in the linkEA header.
3245  * If some hard links are removed after that, then it is possible to hold
3246  * other missed linkEA entries. If the namespace LFSCK have added all the
3247  * related linkEA entries, then it will remove the overflow timestamp.
3248  *
3249  * \param[in] env       pointer to the thread context
3250  * \param[in] com       pointer to the lfsck component
3251  * \param[in] ldata     pointer to the linkEA data for the given @obj
3252  * \param[in] obj       pointer to the dt_object to be handled
3253  *
3254  * \retval              positive number for repaired cases
3255  * \retval              0 if nothing to be repaired
3256  * \retval              negative error number on failure
3257  */
3258 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3259                                                  struct lfsck_component *com,
3260                                                  struct linkea_data *ldata,
3261                                                  struct dt_object *obj)
3262 {
3263         struct lfsck_namespace *ns = com->lc_file_ram;
3264         struct lfsck_instance *lfsck = com->lc_lfsck;
3265         struct dt_device *dev = lfsck_obj2dev(obj);
3266         struct thandle *th = NULL;
3267         struct lustre_handle lh = { 0 };
3268         struct lu_buf linkea_buf;
3269         int rc = 0;
3270         ENTRY;
3271
3272         LASSERT(!dt_object_remote(obj));
3273
3274         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3275                               MDS_INODELOCK_UPDATE, LCK_PW);
3276         if (rc != 0)
3277                 GOTO(log, rc);
3278
3279         th = dt_trans_create(env, dev);
3280         if (IS_ERR(th))
3281                 GOTO(log, rc = PTR_ERR(th));
3282
3283         rc = dt_declare_xattr_set(env, obj,
3284                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3285                         XATTR_NAME_LINK, 0, th);
3286         if (rc != 0)
3287                 GOTO(stop, rc);
3288
3289         rc = dt_trans_start_local(env, dev, th);
3290         if (rc != 0)
3291                 GOTO(stop, rc);
3292
3293         dt_write_lock(env, obj, 0);
3294         rc = lfsck_links_read(env, obj, ldata);
3295         if (rc != 0)
3296                 GOTO(unlock, rc);
3297
3298         if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3299                 GOTO(unlock, rc = 0);
3300
3301         ldata->ld_leh->leh_overflow_time = 0;
3302         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3303                 GOTO(unlock, rc = 1);
3304
3305         /* If all known entries are in the linkEA, then the 'leh_reccount'
3306          * should NOT be zero. */
3307         LASSERT(ldata->ld_leh->leh_reccount > 0);
3308
3309         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3310                        ldata->ld_leh->leh_len);
3311         rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3312         if (unlikely(rc == -ENOSPC))
3313                 rc = 0;
3314         else if (!rc)
3315                 rc = 1;
3316
3317         GOTO(unlock, rc);
3318
3319 unlock:
3320         dt_write_unlock(env, obj);
3321
3322 stop:
3323         dt_trans_stop(env, dev, th);
3324
3325 log:
3326         lfsck_ibits_unlock(&lh, LCK_PW);
3327         CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3328                DFID": rc = %d\n",
3329                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3330
3331         return rc;
3332 }
3333
3334 /**
3335  * Verify the object's agent entry.
3336  *
3337  * If the object claims to have agent entry but the linkEA does not contain
3338  * remote parent, then remove the agent entry. Otherwise, if the object has
3339  * no agent entry but its linkEA contains remote parent, then will generate
3340  * agent entry for it.
3341  *
3342  * \param[in] env       pointer to the thread context
3343  * \param[in] com       pointer to the lfsck component
3344  * \param[in] obj       pointer to the dt_object to be handled
3345  *
3346  * \retval              positive number for repaired cases
3347  * \retval              0 if nothing to be repaired
3348  * \retval              negative error number on failure
3349  */
3350 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3351                                              struct lfsck_component *com,
3352                                              struct dt_object *obj)
3353 {
3354         struct linkea_data ldata = { NULL };
3355         struct lfsck_thread_info *info = lfsck_env_info(env);
3356         struct lfsck_namespace *ns = com->lc_file_ram;
3357         struct lfsck_instance *lfsck = com->lc_lfsck;
3358         struct lu_fid *pfid = &info->lti_fid2;
3359         struct lu_name *cname = &info->lti_name;
3360         struct lu_seq_range *range = &info->lti_range;
3361         struct seq_server_site *ss = lfsck_dev_site(lfsck);
3362         __u32 idx = lfsck_dev_idx(lfsck);
3363         int rc;
3364         bool remote = false;
3365         ENTRY;
3366
3367         if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3368                 RETURN(0);
3369
3370         rc = lfsck_links_read_with_rec(env, obj, &ldata);
3371         if (rc == -ENOENT || rc == -ENODATA)
3372                 RETURN(0);
3373
3374         if (rc && rc != -EINVAL)
3375                 GOTO(out, rc);
3376
3377         /* We check the agent entry again after verifying the linkEA
3378          * successfully. So invalid linkEA should be dryrun mode. */
3379         if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3380                 RETURN(0);
3381
3382         linkea_first_entry(&ldata);
3383         while (ldata.ld_lee != NULL && !remote) {
3384                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3385                                     cname, pfid);
3386                 /* If parent FID is unknown, not verify agent entry. */
3387                 if (!fid_is_sane(pfid))
3388                         GOTO(out, rc = 0);
3389
3390                 fld_range_set_mdt(range);
3391                 rc = fld_server_lookup(env, ss->ss_server_fld,
3392                                        fid_seq(pfid), range);
3393                 if (rc)
3394                         GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3395
3396                 if (range->lsr_index != idx)
3397                         remote = true;
3398                 else
3399                         linkea_next_entry(&ldata);
3400         }
3401
3402         if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3403             (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3404                 struct dt_device *dev = lfsck_obj2dev(obj);
3405                 struct linkea_data ldata2 = { NULL };
3406                 struct lustre_handle lh = { 0 };
3407                 struct lu_buf linkea_buf;
3408                 struct thandle *handle;
3409
3410                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3411                         GOTO(out, rc = 1);
3412
3413                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3414                                       MDS_INODELOCK_UPDATE |
3415                                       MDS_INODELOCK_XATTR, LCK_EX);
3416                 if (rc)
3417                         GOTO(out, rc);
3418
3419                 handle = dt_trans_create(env, dev);
3420                 if (IS_ERR(handle))
3421                         GOTO(unlock, rc = PTR_ERR(handle));
3422
3423                 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3424                                ldata.ld_leh->leh_len);
3425                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3426                                 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3427                 if (rc)
3428                         GOTO(stop, rc);
3429
3430                 rc = dt_trans_start_local(env, dev, handle);
3431                 if (rc)
3432                         GOTO(stop, rc);
3433
3434                 dt_write_lock(env, obj, 0);
3435                 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3436                 if (rc) {
3437                         if (rc == -ENOENT || rc == -ENODATA)
3438                                 rc = 0;
3439                         GOTO(unlock2, rc);
3440                 }
3441
3442                 /* If someone changed linkEA by race, then the agent
3443                  * entry will be updated by lower layer automatically. */
3444                 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3445                     memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3446                            ldata.ld_leh->leh_len) != 0)
3447                         GOTO(unlock2, rc = 0);
3448
3449                 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3450                                   LU_XATTR_REPLACE, handle);
3451                 if (!rc)
3452                         rc = 1;
3453
3454                 GOTO(unlock2, rc);
3455
3456 unlock2:
3457                 dt_write_unlock(env, obj);
3458 stop:
3459                 dt_trans_stop(env, dev, handle);
3460 unlock:
3461                 lfsck_ibits_unlock(&lh, LCK_EX);
3462         }
3463
3464         GOTO(out, rc);
3465
3466 out:
3467         if (rc > 0)
3468                 ns->ln_agent_entries_repaired++;
3469         if (rc)
3470                 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3471                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3472         return rc;
3473 }
3474
3475 /**
3476  * Double scan the MDT-object for namespace LFSCK.
3477  *
3478  * If the MDT-object contains invalid or repeated linkEA entries, then drop
3479  * those entries from the linkEA; if the linkEA becomes empty or the object
3480  * has no linkEA, then it is an orphan and will be added into the directory
3481  * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3482  * the remote parent; if the name entry corresponding to some linkEA entry
3483  * is lost, then add the name entry back to the namespace.
3484  *
3485  * \param[in] env       pointer to the thread context
3486  * \param[in] com       pointer to the lfsck component
3487  * \param[in] child     pointer to the dt_object to be handled
3488  * \param[in] flags     some hints to indicate how the @child should be handled
3489  *
3490  * \retval              positive number for repaired cases
3491  * \retval              0 if nothing to be repaired
3492  * \retval              negative error number on failure
3493  */
3494 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3495                                            struct lfsck_component *com,
3496                                            struct dt_object *child, __u8 flags)
3497 {
3498         struct lfsck_thread_info *info     = lfsck_env_info(env);
3499         struct lu_attr           *la       = &info->lti_la;
3500         struct lu_name           *cname    = &info->lti_name;
3501         struct lu_fid            *pfid     = &info->lti_fid;
3502         struct lu_fid            *cfid     = &info->lti_fid2;
3503         struct lfsck_instance    *lfsck    = com->lc_lfsck;
3504         struct lfsck_namespace   *ns       = com->lc_file_ram;
3505         struct dt_object         *parent   = NULL;
3506         struct linkea_data        ldata    = { NULL };
3507         bool                      repaired = false;
3508         int                       count    = 0;
3509         int                       rc;
3510         ENTRY;
3511
3512         dt_read_lock(env, child, 0);
3513         if (unlikely(lfsck_is_dead_obj(child))) {
3514                 dt_read_unlock(env, child);
3515
3516                 RETURN(0);
3517         }
3518
3519         if (S_ISDIR(lfsck_object_type(child))) {
3520                 dt_read_unlock(env, child);
3521                 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3522                 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3523                         rc = lfsck_namespace_check_agent_entry(env, com, child);
3524
3525                 RETURN(rc);
3526         }
3527
3528         rc = lfsck_links_read(env, child, &ldata);
3529         dt_read_unlock(env, child);
3530
3531         if (rc == -EINVAL) {
3532                 struct lustre_handle lh = { 0 };
3533
3534                 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3535                                       MDS_INODELOCK_UPDATE |
3536                                       MDS_INODELOCK_XATTR, LCK_EX);
3537                 if (rc == 0) {
3538                         rc = lfsck_namespace_links_remove(env, com, child);
3539                         lfsck_ibits_unlock(&lh, LCK_EX);
3540                 }
3541
3542                 GOTO(out, rc);
3543         }
3544
3545         if (rc != 0)
3546                 GOTO(out, rc);
3547
3548         if (!(ns->ln_flags & LF_INCOMPLETE) &&
3549             unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3550                 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3551                                                            child);
3552                 if (rc < 0)
3553                         GOTO(out, rc);
3554
3555                 if (rc > 0)
3556                         ns->ln_linkea_overflow_cleared++;
3557         }
3558
3559         linkea_first_entry(&ldata);
3560         while (ldata.ld_lee != NULL) {
3561                 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3562                                                          info->lti_key,
3563                                                          sizeof(info->lti_key));
3564                 /* Invalid PFID in the linkEA entry. */
3565                 if (rc != 0) {
3566                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3567                                                 &ldata, cname, pfid, true);
3568                         if (rc < 0)
3569                                 GOTO(out, rc);
3570
3571                         if (rc > 0)
3572                                 repaired = true;
3573
3574                         continue;
3575                 }
3576
3577                 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3578                                                          false);
3579                 /* Found repeated linkEA entries */
3580                 if (rc > 0) {
3581                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3582                                                 &ldata, cname, pfid, false);
3583                         if (rc < 0)
3584                                 GOTO(out, rc);
3585
3586                         if (rc == 0)
3587                                 continue;
3588
3589                         repaired = true;
3590
3591                         /* fall through */
3592                 }
3593
3594                 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3595                 if (IS_ERR(parent))
3596                         GOTO(out, rc = PTR_ERR(parent));
3597
3598                 if (!dt_object_exists(parent)) {
3599
3600 lost_parent:
3601                         if (ldata.ld_leh->leh_reccount > 1) {
3602                                 /* If it is NOT the last linkEA entry, then
3603                                  * there is still other chance to make the
3604                                  * child to be visible via other parent, then
3605                                  * remove this linkEA entry. */
3606                                 rc = lfsck_namespace_shrink_linkea(env, com,
3607                                         child, &ldata, cname, pfid, true);
3608                         } else {
3609                                 /* If the LFSCK is marked as LF_INCOMPLETE,
3610                                  * then means some MDT has ever tried to
3611                                  * verify some remote MDT-object that resides
3612                                  * on this MDT, but this MDT failed to respond
3613                                  * such request. So means there may be some
3614                                  * remote name entry on other MDT that
3615                                  * references this object with another name,
3616                                  * so we cannot know whether this linkEA is
3617                                  * valid or not. So keep it there and maybe
3618                                  * resolved when next LFSCK run. */
3619                                 if (ns->ln_flags & LF_INCOMPLETE) {
3620                                         lfsck_object_put(env, parent);
3621
3622                                         GOTO(out, rc = 0);
3623                                 }
3624
3625                                 /* Create the lost parent as an orphan. */
3626                                 rc = lfsck_namespace_create_orphan_dir(env, com,
3627                                                                 parent, NULL);
3628                                 if (rc < 0) {
3629                                         lfsck_object_put(env, parent);
3630
3631                                         GOTO(out, rc);
3632                                 }
3633
3634                                 if (rc > 0)
3635                                         repaired = true;
3636
3637                                 /* Add the missing name entry to the parent. */
3638                                 rc = lfsck_namespace_insert_normal(env, com,
3639                                                 parent, child, cname->ln_name);
3640                                 if (unlikely(rc == -EEXIST))
3641                                         /* Unfortunately, someone reused the
3642                                          * name under the parent by race. So we
3643                                          * have to remove the linkEA entry from
3644                                          * current child object. It means that
3645                                          * the LFSCK cannot recover the system
3646                                          * totally back to its original status,
3647                                          * but it is necessary to make the
3648                                          * current system to be consistent. */
3649                                         rc = lfsck_namespace_shrink_linkea(env,
3650                                                         com, child, &ldata,
3651                                                         cname, pfid, true);
3652                                 else
3653                                         linkea_next_entry(&ldata);
3654                         }
3655
3656                         lfsck_object_put(env, parent);
3657                         if (rc < 0)
3658                                 GOTO(out, rc);
3659
3660                         if (rc > 0)
3661                                 repaired = true;
3662
3663                         continue;
3664                 } /* !dt_object_exists(parent) */
3665
3666                 /* The linkEA entry with bad parent will be removed. */
3667                 if (unlikely(!dt_try_as_dir(env, parent))) {
3668                         lfsck_object_put(env, parent);
3669                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3670                                                 &ldata, cname, pfid, true);
3671                         if (rc < 0)
3672                                 GOTO(out, rc);
3673
3674                         if (rc > 0)
3675                                 repaired = true;
3676
3677                         continue;
3678                 }
3679
3680                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3681                                (const struct dt_key *)cname->ln_name);
3682                 if (rc != 0 && rc != -ENOENT) {
3683                         lfsck_object_put(env, parent);
3684
3685                         GOTO(out, rc);
3686                 }
3687
3688                 if (rc == 0) {
3689                         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3690                                 /* It is the most common case that we
3691                                  * find the name entry corresponding
3692                                  * to the linkEA entry. */
3693                                 lfsck_object_put(env, parent);
3694                                 linkea_next_entry(&ldata);
3695                         } else {
3696                                 /* The name entry references another
3697                                  * MDT-object that may be created by
3698                                  * the LFSCK for repairing dangling
3699                                  * name entry. Try to replace it. */
3700                                 rc = lfsck_namespace_replace_cond(env, com,
3701                                                 parent, child, cfid, cname);
3702                                 lfsck_object_put(env, parent);
3703                                 if (rc < 0)
3704                                         GOTO(out, rc);
3705
3706                                 if (rc > 0) {
3707                                         repaired = true;
3708                                         linkea_next_entry(&ldata);
3709                                 } else {
3710                                         rc = lfsck_namespace_shrink_linkea(env,
3711                                                         com, child, &ldata,
3712                                                         cname, pfid, true);
3713                                         if (rc < 0)
3714                                                 GOTO(out, rc);
3715
3716                                         if (rc > 0)
3717                                                 repaired = true;
3718                                 }
3719                         }
3720
3721                         continue;
3722                 }
3723
3724                 /* The following handles -ENOENT case */
3725
3726                 rc = dt_attr_get(env, child, la);
3727                 if (rc != 0)
3728                         GOTO(out, rc);
3729
3730                 /* If there is no name entry in the parent dir and the object
3731                  * link count is fewer than the linkea entries count, then the
3732                  * linkea entry should be removed. */
3733                 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3734                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3735                                         parent, child, &ldata, cname, pfid);
3736                         lfsck_object_put(env, parent);
3737                         if (rc < 0)
3738                                 GOTO(out, rc);
3739
3740                         if (rc > 0)
3741                                 repaired = true;
3742
3743                         continue;
3744                 }
3745
3746                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3747                  * MDT has ever tried to verify some remote MDT-object that
3748                  * resides on this MDT, but this MDT failed to respond such
3749                  * request. So means there may be some remote name entry on
3750                  * other MDT that references this object with another name,
3751                  * so we cannot know whether this linkEA is valid or not.
3752                  * So keep it there and maybe resolved when next LFSCK run. */
3753                 if (ns->ln_flags & LF_INCOMPLETE) {
3754                         lfsck_object_put(env, parent);
3755
3756                         GOTO(out, rc = 0);
3757                 }
3758
3759                 rc = lfsck_namespace_check_name(env, parent, child, cname);
3760                 if (rc == -ENOENT)
3761                         goto lost_parent;
3762
3763                 if (rc < 0) {
3764                         lfsck_object_put(env, parent);
3765
3766                         GOTO(out, rc);
3767                 }
3768
3769                 /* It is an invalid name entry, drop it. */
3770                 if (unlikely(rc > 0)) {
3771                         lfsck_object_put(env, parent);
3772                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3773                                                 &ldata, cname, pfid, true);
3774                         if (rc < 0)
3775                                 GOTO(out, rc);
3776
3777                         if (rc > 0)
3778                                 repaired = true;
3779
3780                         continue;
3781                 }
3782
3783                 /* Add the missing name entry back to the namespace. */
3784                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3785                                                    cname->ln_name);
3786                 if (unlikely(rc == -ESTALE))
3787                         /* It may happen when the remote object has been
3788                          * removed, but the local MDT is not aware of that. */
3789                         goto lost_parent;
3790
3791                 if (unlikely(rc == -EEXIST))
3792                         /* Unfortunately, someone reused the name under the
3793                          * parent by race. So we have to remove the linkEA
3794                          * entry from current child object. It means that the
3795                          * LFSCK cannot recover the system totally back to
3796                          * its original status, but it is necessary to make
3797                          * the current system to be consistent.
3798                          *
3799                          * It also may be because of the LFSCK found some
3800                          * internal status of create operation. Under such
3801                          * case, nothing to be done. */
3802                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3803                                         parent, child, &ldata, cname, pfid);
3804                 else
3805                         linkea_next_entry(&ldata);
3806
3807                 lfsck_object_put(env, parent);
3808                 if (rc < 0)
3809                         GOTO(out, rc);
3810
3811                 if (rc > 0)
3812                         repaired = true;
3813         }
3814
3815         GOTO(out, rc = 0);
3816
3817 out:
3818         if (rc < 0 && rc != -ENODATA)
3819                 return rc;
3820
3821         if (rc == 0 && ldata.ld_leh != NULL)
3822                 count = ldata.ld_leh->leh_reccount;
3823
3824         if (count == 0) {
3825                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3826                  * MDT has ever tried to verify some remote MDT-object that
3827                  * resides on this MDT, but this MDT failed to respond such
3828                  * request. So means there may be some remote name entry on
3829                  * other MDT that references this object with another name,
3830                  * so we cannot know whether this linkEA is valid or not.
3831                  * So keep it there and maybe resolved when next LFSCK run. */
3832                 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3833                     (ldata.ld_leh == NULL ||
3834                      !ldata.ld_leh->leh_overflow_time)) {
3835                         /* If the child becomes orphan, then insert it into
3836                          * the global .lustre/lost+found/MDTxxxx directory. */
3837                         rc = lfsck_namespace_insert_orphan(env, com, child,
3838                                                            "", "O", &count);
3839                         if (rc < 0)
3840                                 return rc;
3841
3842                         if (rc > 0) {
3843                                 ns->ln_mul_ref_repaired++;
3844                                 repaired = true;
3845                         }
3846                 }
3847         } else {
3848                 rc = dt_attr_get(env, child, la);
3849                 if (rc != 0)
3850                         return rc;
3851
3852                 if (la->la_nlink != 0 && la->la_nlink != count) {
3853                         if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
3854                                      !S_ISLNK(lfsck_object_type(child)))) {
3855                                 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
3856                                        "the object "DFID"'s nlink count %d "
3857                                        "does not match linkEA count %d, "
3858                                        "type %o, skip it.\n",
3859                                        lfsck_lfsck2name(lfsck),
3860                                        PFID(lfsck_dto2fid(child)),
3861                                        la->la_nlink, count,
3862                                        lfsck_object_type(child));
3863                         } else if (la->la_nlink < count &&
3864                                    likely(!ldata.ld_leh->leh_overflow_time)) {
3865                                 rc = lfsck_namespace_repair_nlink(env, com,
3866                                                                   child, la);
3867                                 if (rc > 0) {
3868                                         ns->ln_objs_nlink_repaired++;
3869                                         rc = 0;
3870                                 }
3871                         }
3872                 }
3873         }
3874
3875         if (repaired) {
3876                 if (la->la_nlink > 1)
3877                         ns->ln_mul_linked_repaired++;
3878
3879                 if (rc == 0)
3880                         rc = 1;
3881         }
3882
3883         if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3884                 rc = lfsck_namespace_check_agent_entry(env, com, child);
3885
3886         return rc;
3887 }
3888
3889 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3890                                             struct lfsck_namespace *ns,
3891                                             __u64 checked_phase1,
3892                                             __u64 checked_phase2,
3893                                             time64_t time_phase1,
3894                                             time64_t time_phase2, bool dryrun)
3895 {
3896         const char *postfix = dryrun ? "inconsistent" : "repaired";
3897
3898         seq_printf(m, "checked_phase1: %llu\n"
3899                    "checked_phase2: %llu\n"
3900                    "%s_phase1: %llu\n"
3901                    "%s_phase2: %llu\n"
3902                    "failed_phase1: %llu\n"
3903                    "failed_phase2: %llu\n"
3904                    "directories: %llu\n"
3905                    "dirent_%s: %llu\n"
3906                    "linkea_%s: %llu\n"
3907                    "nlinks_%s: %llu\n"
3908                    "multiple_linked_checked: %llu\n"
3909                    "multiple_linked_%s: %llu\n"
3910                    "unknown_inconsistency: %llu\n"
3911                    "unmatched_pairs_%s: %llu\n"
3912                    "dangling_%s: %llu\n"
3913                    "multiple_referenced_%s: %llu\n"
3914                    "bad_file_type_%s: %llu\n"
3915                    "lost_dirent_%s: %llu\n"
3916                    "local_lost_found_scanned: %llu\n"
3917                    "local_lost_found_moved: %llu\n"
3918                    "local_lost_found_skipped: %llu\n"
3919                    "local_lost_found_failed: %llu\n"
3920                    "striped_dirs_scanned: %llu\n"
3921                    "striped_dirs_%s: %llu\n"
3922                    "striped_dirs_failed: %llu\n"
3923                    "striped_dirs_disabled: %llu\n"
3924                    "striped_dirs_skipped: %llu\n"
3925                    "striped_shards_scanned: %llu\n"
3926                    "striped_shards_%s: %llu\n"
3927                    "striped_shards_failed: %llu\n"
3928                    "striped_shards_skipped: %llu\n"
3929                    "name_hash_%s: %llu\n"
3930                    "linkea_overflow_%s: %llu\n"
3931                    "agent_entries_%s: %llu\n"
3932                    "success_count: %u\n"
3933                    "run_time_phase1: %lld seconds\n"
3934                    "run_time_phase2: %lld seconds\n",
3935                    checked_phase1,
3936                    checked_phase2,
3937                    dryrun ? "inconsistent" : "updated",
3938                    ns->ln_items_repaired,
3939                    dryrun ? "inconsistent" : "updated",
3940                    ns->ln_objs_repaired_phase2,
3941                    ns->ln_items_failed,
3942                    ns->ln_objs_failed_phase2,
3943                    ns->ln_dirs_checked,
3944                    postfix, ns->ln_dirent_repaired,
3945                    postfix, ns->ln_linkea_repaired,
3946                    postfix, ns->ln_objs_nlink_repaired,
3947                    ns->ln_mul_linked_checked,
3948                    postfix, ns->ln_mul_linked_repaired,
3949                    ns->ln_unknown_inconsistency,
3950                    postfix, ns->ln_unmatched_pairs_repaired,
3951                    postfix, ns->ln_dangling_repaired,
3952                    postfix, ns->ln_mul_ref_repaired,
3953                    postfix, ns->ln_bad_type_repaired,
3954                    postfix, ns->ln_lost_dirent_repaired,
3955                    ns->ln_local_lpf_scanned,
3956                    ns->ln_local_lpf_moved,
3957                    ns->ln_local_lpf_skipped,
3958                    ns->ln_local_lpf_failed,
3959                    ns->ln_striped_dirs_scanned,
3960                    postfix, ns->ln_striped_dirs_repaired,
3961                    ns->ln_striped_dirs_failed,
3962                    ns->ln_striped_dirs_disabled,
3963                    ns->ln_striped_dirs_skipped,
3964                    ns->ln_striped_shards_scanned,
3965                    postfix, ns->ln_striped_shards_repaired,
3966                    ns->ln_striped_shards_failed,
3967                    ns->ln_striped_shards_skipped,
3968                    postfix, ns->ln_name_hash_repaired,
3969                    dryrun ? "inconsistent" : "cleared",
3970                    ns->ln_linkea_overflow_cleared,
3971                    postfix, ns->ln_agent_entries_repaired,
3972                    ns->ln_success_count,
3973                    time_phase1,
3974                    time_phase2);
3975 }
3976
3977 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3978                                         struct lfsck_component *com)
3979 {
3980         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3981         struct lfsck_namespace          *ns     = com->lc_file_ram;
3982
3983         while (!list_empty(&lfsck->li_list_lmv)) {
3984                 struct lfsck_lmv_unit   *llu;
3985                 struct lfsck_lmv        *llmv;
3986
3987                 llu = list_entry(lfsck->li_list_lmv.next,
3988                                  struct lfsck_lmv_unit, llu_link);
3989                 llmv = &llu->llu_lmv;
3990
3991                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3992                          "still in using: %u\n",
3993                          atomic_read(&llmv->ll_ref));
3994
3995                 ns->ln_striped_dirs_skipped++;
3996                 lfsck_lmv_put(env, llmv);
3997         }
3998 }
3999
4000 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4001                                                  struct lfsck_component *com,
4002                                                  struct dt_object *obj)
4003 {
4004         struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4005         int             rc;
4006
4007         rc = dt_attr_get(env, obj, la);
4008         if (rc != 0)
4009                 return rc;
4010
4011         /* zero-linkEA object may be orphan, but it also maybe because
4012          * of upgrading. Currently, we cannot record it for double scan.
4013          * Because it may cause the LFSCK trace file to be too large. */
4014
4015         /* "la_ctime" == 1 means that it has ever been removed from
4016          * backend /lost+found directory but not been added back to
4017          * the normal namespace yet. */
4018
4019         if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4020             unlikely(la->la_ctime == 1))
4021                 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4022                                                   LNTF_CHECK_LINKEA, true);
4023
4024         return rc;
4025 }
4026
4027 /* namespace APIs */
4028
4029 static int lfsck_namespace_reset(const struct lu_env *env,
4030                                  struct lfsck_component *com, bool init)
4031 {
4032         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4033         struct lfsck_namespace          *ns     = com->lc_file_ram;
4034         struct lfsck_assistant_data     *lad    = com->lc_data;
4035         struct dt_object                *root;
4036         int                              rc;
4037         ENTRY;
4038
4039         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4040         if (IS_ERR(root))
4041                 GOTO(log, rc = PTR_ERR(root));
4042
4043         if (unlikely(!dt_try_as_dir(env, root)))
4044                 GOTO(put, rc = -ENOTDIR);
4045
4046         down_write(&com->lc_sem);
4047         if (init) {
4048                 memset(ns, 0, sizeof(*ns));
4049         } else {
4050                 __u32 count = ns->ln_success_count;
4051                 time64_t last_time = ns->ln_time_last_complete;
4052
4053                 memset(ns, 0, sizeof(*ns));
4054                 ns->ln_success_count = count;
4055                 ns->ln_time_last_complete = last_time;
4056         }
4057         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4058         ns->ln_status = LS_INIT;
4059         ns->ln_time_latest_reset = ktime_get_real_seconds();
4060
4061         rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4062                                        &dt_lfsck_namespace_features,
4063                                        LFSCK_NAMESPACE, true);
4064         if (rc)
4065                 GOTO(out, rc);
4066
4067         rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4068                                         LFSCK_NAMESPACE, true);
4069         if (rc != 0)
4070                 GOTO(out, rc);
4071
4072         lad->lad_incomplete = 0;
4073         CFS_RESET_BITMAP(lad->lad_bitmap);
4074
4075         rc = lfsck_namespace_store(env, com);
4076
4077         GOTO(out, rc);
4078
4079 out:
4080         up_write(&com->lc_sem);
4081
4082 put:
4083         lfsck_object_put(env, root);
4084 log:
4085         CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4086                lfsck_lfsck2name(lfsck), rc);
4087         return rc;
4088 }
4089
4090 static void
4091 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4092                      bool new_checked)
4093 {
4094         struct lfsck_namespace *ns = com->lc_file_ram;
4095
4096         down_write(&com->lc_sem);
4097         if (new_checked)
4098                 com->lc_new_checked++;
4099         lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4100         up_write(&com->lc_sem);
4101 }
4102
4103 static void lfsck_namespace_close_dir(const struct lu_env *env,
4104                                       struct lfsck_component *com)
4105 {
4106         struct lfsck_namespace          *ns     = com->lc_file_ram;
4107         struct lfsck_assistant_data     *lad    = com->lc_data;
4108         struct lfsck_assistant_object   *lso    = NULL;
4109         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4110         struct lfsck_lmv                *llmv   = lfsck->li_lmv;
4111         struct lfsck_namespace_req      *lnr;
4112         __u32                            size   =
4113                                 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4114         bool                             wakeup = false;
4115         ENTRY;
4116
4117         if (llmv == NULL)
4118                 RETURN_EXIT;
4119
4120         OBD_ALLOC(lnr, size);
4121         if (lnr == NULL) {
4122                 ns->ln_striped_dirs_skipped++;
4123
4124                 RETURN_EXIT;
4125         }
4126
4127         lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4128                         NULL, lfsck->li_pos_current.lp_oit_cookie, true);
4129         if (IS_ERR(lso)) {
4130                 OBD_FREE(lnr, size);
4131                 ns->ln_striped_dirs_skipped++;
4132
4133                 RETURN_EXIT;
4134         }
4135
4136         /* Generate a dummy request to indicate that all shards' name entry
4137          * in this striped directory has been scanned for the first time. */
4138         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4139         lnr->lnr_lar.lar_parent = lso;
4140         lnr->lnr_lmv = lfsck_lmv_get(llmv);
4141         lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4142         lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4143         lnr->lnr_size = size;
4144
4145         spin_lock(&lad->lad_lock);
4146         if (lad->lad_assistant_status < 0 ||
4147             unlikely(!thread_is_running(&lfsck->li_thread) ||
4148                      !thread_is_running(&lad->lad_thread))) {
4149                 spin_unlock(&lad->lad_lock);
4150                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4151                 ns->ln_striped_dirs_skipped++;
4152
4153                 RETURN_EXIT;
4154         }
4155
4156         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4157         if (lad->lad_prefetched == 0)
4158                 wakeup = true;
4159
4160         lad->lad_prefetched++;
4161         spin_unlock(&lad->lad_lock);
4162         if (wakeup)
4163                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4164
4165         EXIT;
4166 }
4167
4168 static int lfsck_namespace_open_dir(const struct lu_env *env,
4169                                     struct lfsck_component *com)
4170 {
4171         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4172         struct lfsck_namespace  *ns     = com->lc_file_ram;
4173         struct lfsck_lmv        *llmv   = lfsck->li_lmv;
4174         int                      rc     = 0;
4175         ENTRY;
4176
4177         if (llmv == NULL)
4178                 RETURN(0);
4179
4180         if (llmv->ll_lmv_master) {
4181                 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4182
4183                 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4184                         lmv->lmv_master_mdt_index =
4185                                 lfsck_dev_idx(lfsck);
4186                         ns->ln_flags |= LF_INCONSISTENT;
4187                         llmv->ll_lmv_updated = 1;
4188                 }
4189         } else {
4190                 rc = lfsck_namespace_verify_stripe_slave(env, com,
4191                                         lfsck->li_obj_dir, llmv);
4192         }
4193
4194         RETURN(rc > 0 ? 0 : rc);
4195 }
4196
4197 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4198                                       struct lfsck_component *com, bool init)
4199 {
4200         struct lfsck_instance   *lfsck = com->lc_lfsck;
4201         struct lfsck_namespace  *ns    = com->lc_file_ram;
4202         int                      rc;
4203
4204         if (!init) {
4205                 rc = lfsck_checkpoint_generic(env, com);
4206                 if (rc != 0)
4207                         goto log;
4208         }
4209
4210         down_write(&com->lc_sem);
4211         if (init) {
4212                 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4213         } else {
4214                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4215                 ns->ln_run_time_phase1 += ktime_get_seconds() -
4216                                           lfsck->li_time_last_checkpoint;
4217                 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4218                 ns->ln_items_checked += com->lc_new_checked;
4219                 com->lc_new_checked = 0;
4220         }
4221
4222         rc = lfsck_namespace_store(env, com);
4223         up_write(&com->lc_sem);
4224
4225 log:
4226         CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4227                ", "DFID", %#llx], status = %d: rc = %d\n",
4228                lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4229                PFID(&lfsck->li_pos_current.lp_dir_parent),
4230                lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4231
4232         return rc > 0 ? 0 : rc;
4233 }
4234
4235 static int lfsck_namespace_prep(const struct lu_env *env,
4236                                 struct lfsck_component *com,
4237                                 struct lfsck_start_param *lsp)
4238 {
4239         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4240         struct lfsck_namespace  *ns     = com->lc_file_ram;
4241         struct lfsck_position   *pos    = &com->lc_pos_start;
4242         int                      rc;
4243
4244         rc = lfsck_namespace_load_bitmap(env, com);
4245         if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4246                 rc = lfsck_namespace_reset(env, com, false);
4247                 if (rc == 0)
4248                         rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4249
4250                 if (rc != 0) {
4251                         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4252                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4253
4254                         return rc;
4255                 }
4256         }
4257
4258         down_write(&com->lc_sem);
4259         ns->ln_time_latest_start = ktime_get_real_seconds();
4260         spin_lock(&lfsck->li_lock);
4261
4262         if (ns->ln_flags & LF_SCANNED_ONCE) {
4263                 if (!lfsck->li_drop_dryrun ||
4264                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4265                         ns->ln_status = LS_SCANNING_PHASE2;
4266                         list_move_tail(&com->lc_link,
4267                                        &lfsck->li_list_double_scan);
4268                         if (!list_empty(&com->lc_link_dir))
4269                                 list_del_init(&com->lc_link_dir);
4270                         lfsck_pos_set_zero(pos);
4271                 } else {
4272                         ns->ln_status = LS_SCANNING_PHASE1;
4273                         ns->ln_run_time_phase1 = 0;
4274                         ns->ln_run_time_phase2 = 0;
4275                         ns->ln_items_checked = 0;
4276                         ns->ln_items_repaired = 0;
4277                         ns->ln_items_failed = 0;
4278                         ns->ln_dirs_checked = 0;
4279                         ns->ln_objs_checked_phase2 = 0;
4280                         ns->ln_objs_repaired_phase2 = 0;
4281                         ns->ln_objs_failed_phase2 = 0;
4282                         ns->ln_objs_nlink_repaired = 0;
4283                         ns->ln_dirent_repaired = 0;
4284                         ns->ln_linkea_repaired = 0;
4285                         ns->ln_mul_linked_checked = 0;
4286                         ns->ln_mul_linked_repaired = 0;
4287                         ns->ln_unknown_inconsistency = 0;
4288                         ns->ln_unmatched_pairs_repaired = 0;
4289                         ns->ln_dangling_repaired = 0;
4290                         ns->ln_mul_ref_repaired = 0;
4291                         ns->ln_bad_type_repaired = 0;
4292                         ns->ln_lost_dirent_repaired = 0;
4293                         ns->ln_striped_dirs_scanned = 0;
4294                         ns->ln_striped_dirs_repaired = 0;
4295                         ns->ln_striped_dirs_failed = 0;
4296                         ns->ln_striped_dirs_disabled = 0;
4297                         ns->ln_striped_dirs_skipped = 0;
4298                         ns->ln_striped_shards_scanned = 0;
4299                         ns->ln_striped_shards_repaired = 0;
4300                         ns->ln_striped_shards_failed = 0;
4301                         ns->ln_striped_shards_skipped = 0;
4302                         ns->ln_name_hash_repaired = 0;
4303                         fid_zero(&ns->ln_fid_latest_scanned_phase2);
4304                         if (list_empty(&com->lc_link_dir))
4305                                 list_add_tail(&com->lc_link_dir,
4306                                               &lfsck->li_list_dir);
4307                         *pos = ns->ln_pos_first_inconsistent;
4308                 }
4309         } else {
4310                 ns->ln_status = LS_SCANNING_PHASE1;
4311                 if (list_empty(&com->lc_link_dir))
4312                         list_add_tail(&com->lc_link_dir,
4313                                       &lfsck->li_list_dir);
4314                 if (!lfsck->li_drop_dryrun ||
4315                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4316                         *pos = ns->ln_pos_last_checkpoint;
4317                         pos->lp_oit_cookie++;
4318                 } else {
4319                         *pos = ns->ln_pos_first_inconsistent;
4320                 }
4321         }
4322
4323         spin_unlock(&lfsck->li_lock);
4324         up_write(&com->lc_sem);
4325
4326         rc = lfsck_start_assistant(env, com, lsp);
4327
4328         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4329                DFID", %#llx]: rc = %d\n",
4330                lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4331                PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4332
4333         return rc;
4334 }
4335
4336 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4337                                     struct lfsck_component *com,
4338                                     struct dt_object *obj)
4339 {
4340         struct lfsck_thread_info *info  = lfsck_env_info(env);
4341         struct lfsck_namespace   *ns    = com->lc_file_ram;
4342         struct lfsck_instance    *lfsck = com->lc_lfsck;
4343         const struct lu_fid      *fid   = lfsck_dto2fid(obj);
4344         struct lu_fid            *pfid  = &info->lti_fid2;
4345         struct lu_name           *cname = &info->lti_name;
4346         struct lu_seq_range      *range = &info->lti_range;
4347         struct seq_server_site   *ss    = lfsck_dev_site(lfsck);
4348         struct linkea_data        ldata = { NULL };
4349         __u32                     idx   = lfsck_dev_idx(lfsck);
4350         int                       rc;
4351         bool remote = false;
4352         ENTRY;
4353
4354         rc = lfsck_links_read(env, obj, &ldata);
4355         if (rc == -ENOENT)
4356                 GOTO(out, rc = 0);
4357
4358         /* -EINVAL means crashed linkEA, should be verified. */
4359         if (rc == -EINVAL) {
4360                 rc = lfsck_namespace_trace_update(env, com, fid,
4361                                                   LNTF_CHECK_LINKEA, true);
4362                 if (rc == 0) {
4363                         struct lustre_handle lh = { 0 };
4364
4365                         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4366                                               MDS_INODELOCK_UPDATE |
4367                                               MDS_INODELOCK_XATTR, LCK_EX);
4368                         if (rc == 0) {
4369                                 rc = lfsck_namespace_links_remove(env, com,
4370                                                                   obj);
4371                                 lfsck_ibits_unlock(&lh, LCK_EX);
4372                         }
4373                 }
4374
4375                 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4376         }
4377
4378         if (rc && rc != -ENODATA)
4379                 GOTO(out, rc);
4380
4381         if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4382                 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4383
4384                 GOTO(out, rc);
4385         }
4386
4387         linkea_first_entry(&ldata);
4388         while (ldata.ld_lee != NULL) {
4389                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4390                                     cname, pfid);
4391                 if (!fid_is_sane(pfid)) {
4392                         rc = lfsck_namespace_trace_update(env, com, fid,
4393                                                   LNTF_CHECK_PARENT, true);
4394                 } else {
4395                         fld_range_set_mdt(range);
4396                         rc = fld_server_lookup(env, ss->ss_server_fld,
4397                                                fid_seq(pfid), range);
4398                         if ((rc == -ENOENT) ||
4399                             (!rc && range->lsr_index != idx)) {
4400                                 remote = true;
4401                                 break;
4402                         }
4403                 }
4404                 if (rc)
4405                         GOTO(out, rc);
4406
4407                 linkea_next_entry(&ldata);
4408         }
4409
4410         if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4411             (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4412                 rc = lfsck_namespace_trace_update(env, com, fid,
4413                                                   LNTF_CHECK_AGENT_ENTRY, true);
4414                 if (rc)
4415                         GOTO(out, rc);
4416         }
4417
4418         /* Record multiple-linked object. */
4419         if (ldata.ld_leh->leh_reccount > 1) {
4420                 rc = lfsck_namespace_trace_update(env, com, fid,
4421                                                   LNTF_CHECK_LINKEA, true);
4422
4423                 GOTO(out, rc);
4424         }
4425
4426         if (remote)
4427                 rc = lfsck_namespace_trace_update(env, com, fid,
4428                                                   LNTF_CHECK_LINKEA, true);
4429         else
4430                 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4431
4432         GOTO(out, rc);
4433
4434 out:
4435         down_write(&com->lc_sem);
4436         if (S_ISDIR(lfsck_object_type(obj)))
4437                 ns->ln_dirs_checked++;
4438         if (rc != 0)
4439                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4440         up_write(&com->lc_sem);
4441
4442         return rc;
4443 }
4444
4445 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4446                                     struct lfsck_component *com,
4447                                     struct lfsck_assistant_object *lso,
4448                                     struct lu_dirent *ent, __u16 type)
4449 {
4450         struct lfsck_assistant_data     *lad     = com->lc_data;
4451         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4452         struct lfsck_namespace_req      *lnr;
4453         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4454         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4455         struct ptlrpc_thread            *athread = &lad->lad_thread;
4456         struct l_wait_info               lwi     = { 0 };
4457         bool                             wakeup  = false;
4458
4459         l_wait_event(mthread->t_ctl_waitq,
4460                      lad->lad_prefetched < bk->lb_async_windows ||
4461                      !thread_is_running(mthread) ||
4462                      !thread_is_running(athread),
4463                      &lwi);
4464
4465         if (unlikely(!thread_is_running(mthread) ||
4466                      !thread_is_running(athread)))
4467                 return 0;
4468
4469         if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4470                 return 0;
4471
4472         lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4473         if (IS_ERR(lnr)) {
4474                 struct lfsck_namespace *ns = com->lc_file_ram;
4475
4476                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4477                 return PTR_ERR(lnr);
4478         }
4479
4480         spin_lock(&lad->lad_lock);
4481         if (lad->lad_assistant_status < 0 ||
4482             unlikely(!thread_is_running(mthread) ||
4483                      !thread_is_running(athread))) {
4484                 spin_unlock(&lad->lad_lock);
4485                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4486                 return lad->lad_assistant_status;
4487         }
4488
4489         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4490         if (lad->lad_prefetched == 0)
4491                 wakeup = true;
4492
4493         lad->lad_prefetched++;
4494         spin_unlock(&lad->lad_lock);
4495         if (wakeup)
4496                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4497
4498         down_write(&com->lc_sem);
4499         com->lc_new_checked++;
4500         up_write(&com->lc_sem);
4501
4502         return 0;
4503 }
4504
4505 static int lfsck_namespace_post(const struct lu_env *env,
4506                                 struct lfsck_component *com,
4507                                 int result, bool init)
4508 {
4509         struct lfsck_instance   *lfsck = com->lc_lfsck;
4510         struct lfsck_namespace  *ns    = com->lc_file_ram;
4511         int                      rc;
4512         ENTRY;
4513
4514         lfsck_post_generic(env, com, &result);
4515
4516         down_write(&com->lc_sem);
4517         lfsck_namespace_release_lmv(env, com);
4518
4519         spin_lock(&lfsck->li_lock);
4520         if (!init)
4521                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4522         if (result > 0) {
4523                 ns->ln_status = LS_SCANNING_PHASE2;
4524                 ns->ln_flags |= LF_SCANNED_ONCE;
4525                 ns->ln_flags &= ~LF_UPGRADE;
4526                 list_del_init(&com->lc_link_dir);
4527                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4528         } else if (result == 0) {
4529                 if (lfsck->li_status != 0)
4530                         ns->ln_status = lfsck->li_status;
4531                 else
4532                         ns->ln_status = LS_STOPPED;
4533                 if (ns->ln_status != LS_PAUSED) {
4534                         list_del_init(&com->lc_link_dir);
4535                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4536                 }
4537         } else {
4538                 ns->ln_status = LS_FAILED;
4539                 list_del_init(&com->lc_link_dir);
4540                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4541         }
4542         spin_unlock(&lfsck->li_lock);
4543
4544         if (!init) {
4545                 ns->ln_run_time_phase1 += ktime_get_seconds() -
4546                                           lfsck->li_time_last_checkpoint;
4547                 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4548                 ns->ln_items_checked += com->lc_new_checked;
4549                 com->lc_new_checked = 0;
4550         }
4551
4552         rc = lfsck_namespace_store(env, com);
4553         up_write(&com->lc_sem);
4554
4555         CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4556                lfsck_lfsck2name(lfsck), rc);
4557
4558         RETURN(rc);
4559 }
4560
4561 static void
4562 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4563                      struct seq_file *m)
4564 {
4565         struct lfsck_instance   *lfsck = com->lc_lfsck;
4566         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4567         struct lfsck_namespace  *ns    = com->lc_file_ram;
4568
4569         down_read(&com->lc_sem);
4570         seq_printf(m, "name: lfsck_namespace\n"
4571                    "magic: %#x\n"
4572                    "version: %d\n"
4573                    "status: %s\n",
4574                    ns->ln_magic,
4575                    bk->lb_version,
4576                    lfsck_status2name(ns->ln_status));
4577
4578         lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4579
4580         lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4581
4582         lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4583
4584         lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4585
4586         lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4587
4588         lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4589
4590         lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4591                        "last_checkpoint_position");
4592
4593         lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4594                        "first_failure_position");
4595
4596         if (ns->ln_status == LS_SCANNING_PHASE1) {
4597                 struct lfsck_position pos;
4598                 time64_t duration = ktime_get_seconds() -
4599                                     lfsck->li_time_last_checkpoint;
4600                 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4601                 u64 speed = checked;
4602                 u64 new_checked = com->lc_new_checked;
4603                 time64_t rtime = ns->ln_run_time_phase1 + duration;
4604
4605                 if (duration != 0)
4606                         new_checked = div64_s64(new_checked, duration);
4607
4608                 if (rtime != 0)
4609                         speed = div64_s64(speed, rtime);
4610
4611                 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4612                                                 bk->lb_param & LPF_DRYRUN);
4613                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4614                            "average_speed_phase2: N/A\n"
4615                            "average_speed_total: %llu items/sec\n"
4616                            "real_time_speed_phase1: %llu items/sec\n"
4617                            "real_time_speed_phase2: N/A\n",
4618                            speed,
4619                            speed,
4620                            new_checked);
4621
4622                 if (likely(lfsck->li_di_oit)) {
4623                         const struct dt_it_ops *iops =
4624                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
4625
4626                         /* The low layer otable-based iteration position may NOT
4627                          * exactly match the namespace-based directory traversal
4628                          * cookie. Generally, it is not a serious issue. But the
4629                          * caller should NOT make assumption on that. */
4630                         pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4631                         if (!lfsck->li_current_oit_processed)
4632                                 pos.lp_oit_cookie--;
4633
4634                         spin_lock(&lfsck->li_lock);
4635                         if (lfsck->li_di_dir) {
4636                                 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4637                                 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4638                                         fid_zero(&pos.lp_dir_parent);
4639                                         pos.lp_dir_cookie = 0;
4640                                 } else {
4641                                         pos.lp_dir_parent =
4642                                         *lfsck_dto2fid(lfsck->li_obj_dir);
4643                                 }
4644                         } else {
4645                                 fid_zero(&pos.lp_dir_parent);
4646                                 pos.lp_dir_cookie = 0;
4647                         }
4648                         spin_unlock(&lfsck->li_lock);
4649                 } else {
4650                         pos = ns->ln_pos_last_checkpoint;
4651                 }
4652
4653                 lfsck_pos_dump(m, &pos, "current_position");
4654         } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4655                 time64_t duration = ktime_get_seconds() -
4656                                     com->lc_time_last_checkpoint;
4657                 __u64 checked = ns->ln_objs_checked_phase2 +
4658                                 com->lc_new_checked;
4659                 __u64 speed1 = ns->ln_items_checked;
4660                 __u64 speed2 = checked;
4661                 __u64 speed0 = speed1 + speed2;
4662                 __u64 new_checked = com->lc_new_checked;
4663                 time64_t rtime = ns->ln_run_time_phase2 + duration;
4664                 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4665
4666                 if (duration != 0)
4667                         new_checked = div64_s64(new_checked, duration);
4668
4669                 if (ns->ln_run_time_phase1 != 0)
4670                         speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4671                 else if (ns->ln_items_checked != 0)
4672                         time0++;
4673
4674                 if (rtime != 0)
4675                         speed2 = div64_s64(speed2, rtime);
4676                 else if (checked != 0)
4677                         time0++;
4678
4679                 if (time0 != 0)
4680                         speed0 = div64_s64(speed0, time0);
4681
4682                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4683                                                 checked,
4684                                                 ns->ln_run_time_phase1, rtime,
4685                                                 bk->lb_param & LPF_DRYRUN);
4686                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4687                            "average_speed_phase2: %llu objs/sec\n"
4688                            "average_speed_total: %llu items/sec\n"
4689                            "real_time_speed_phase1: N/A\n"
4690                            "real_time_speed_phase2: %llu objs/sec\n"
4691                            "current_position: "DFID"\n",
4692                            speed1,
4693                            speed2,
4694                            speed0,
4695                            new_checked,
4696                            PFID(&ns->ln_fid_latest_scanned_phase2));
4697         } else {
4698                 __u64 speed1 = ns->ln_items_checked;
4699                 __u64 speed2 = ns->ln_objs_checked_phase2;
4700                 __u64 speed0 = speed1 + speed2;
4701                 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4702
4703                 if (ns->ln_run_time_phase1 != 0)
4704                         speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4705                 else if (ns->ln_items_checked != 0)
4706                         time0++;
4707
4708                 if (ns->ln_run_time_phase2 != 0)
4709                         speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4710                 else if (ns->ln_objs_checked_phase2 != 0)
4711                         time0++;
4712
4713                 if (time0 != 0)
4714                         speed0 = div64_s64(speed0, time0);
4715
4716                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4717                                                 ns->ln_objs_checked_phase2,
4718                                                 ns->ln_run_time_phase1,
4719                                                 ns->ln_run_time_phase2,
4720                                                 bk->lb_param & LPF_DRYRUN);
4721                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4722                            "average_speed_phase2: %llu objs/sec\n"
4723                            "average_speed_total: %llu items/sec\n"
4724                            "real_time_speed_phase1: N/A\n"
4725                            "real_time_speed_phase2: N/A\n"
4726                            "current_position: N/A\n",
4727                            speed1,
4728                            speed2,
4729                            speed0);
4730         }
4731
4732         up_read(&com->lc_sem);
4733 }
4734
4735 static int lfsck_namespace_double_scan(const struct lu_env *env,
4736                                        struct lfsck_component *com)
4737 {
4738         struct lfsck_namespace          *ns     = com->lc_file_ram;
4739         struct lfsck_assistant_data     *lad    = com->lc_data;
4740         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4741         struct lfsck_tgt_desc           *ltd;
4742         struct lfsck_tgt_desc           *next;
4743         int                              rc;
4744
4745         rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4746         if (thread_is_stopped(&lad->lad_thread)) {
4747                 LASSERT(list_empty(&lad->lad_req_list));
4748                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4749
4750                 spin_lock(&ltds->ltd_lock);
4751                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4752                                          ltd_namespace_phase_list) {
4753                         list_del_init(&ltd->ltd_namespace_phase_list);
4754                 }
4755                 spin_unlock(&ltds->ltd_lock);
4756         }
4757
4758         return rc;
4759 }
4760
4761 static void lfsck_namespace_data_release(const struct lu_env *env,
4762                                          struct lfsck_component *com)
4763 {
4764         struct lfsck_assistant_data     *lad    = com->lc_data;
4765         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4766         struct lfsck_tgt_desc           *ltd;
4767         struct lfsck_tgt_desc           *next;
4768
4769         LASSERT(lad != NULL);
4770         LASSERT(thread_is_init(&lad->lad_thread) ||
4771                 thread_is_stopped(&lad->lad_thread));
4772         LASSERT(list_empty(&lad->lad_req_list));
4773
4774         com->lc_data = NULL;
4775         lfsck_namespace_release_lmv(env, com);
4776
4777         spin_lock(&ltds->ltd_lock);
4778         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4779                                  ltd_namespace_phase_list) {
4780                 list_del_init(&ltd->ltd_namespace_phase_list);
4781         }
4782         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4783                                  ltd_namespace_phase_list) {
4784                 list_del_init(&ltd->ltd_namespace_phase_list);
4785         }
4786         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4787                                  ltd_namespace_list) {
4788                 list_del_init(&ltd->ltd_namespace_list);
4789         }
4790         spin_unlock(&ltds->ltd_lock);
4791
4792         if (likely(lad->lad_bitmap != NULL))
4793                 CFS_FREE_BITMAP(lad->lad_bitmap);
4794
4795         OBD_FREE_PTR(lad);
4796 }
4797
4798 static void lfsck_namespace_quit(const struct lu_env *env,
4799                                  struct lfsck_component *com)
4800 {
4801         struct lfsck_assistant_data     *lad    = com->lc_data;
4802         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4803         struct lfsck_tgt_desc           *ltd;
4804         struct lfsck_tgt_desc           *next;
4805
4806         LASSERT(lad != NULL);
4807
4808         lfsck_quit_generic(env, com);
4809
4810         LASSERT(thread_is_init(&lad->lad_thread) ||
4811                 thread_is_stopped(&lad->lad_thread));
4812         LASSERT(list_empty(&lad->lad_req_list));
4813
4814         lfsck_namespace_release_lmv(env, com);
4815
4816         spin_lock(&ltds->ltd_lock);
4817         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4818                                  ltd_namespace_phase_list) {
4819                 list_del_init(&ltd->ltd_namespace_phase_list);
4820         }
4821         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4822                                  ltd_namespace_phase_list) {
4823                 list_del_init(&ltd->ltd_namespace_phase_list);
4824         }
4825         spin_unlock(&ltds->ltd_lock);
4826 }
4827
4828 static int lfsck_namespace_in_notify(const struct lu_env *env,
4829                                      struct lfsck_component *com,
4830                                      struct lfsck_request *lr)
4831 {
4832         struct lfsck_instance *lfsck = com->lc_lfsck;
4833         struct lfsck_namespace *ns = com->lc_file_ram;
4834         struct lfsck_assistant_data *lad = com->lc_data;
4835         struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4836         struct lfsck_tgt_desc *ltd;
4837         int rc = 0;
4838         bool fail = false;
4839         ENTRY;
4840
4841         switch (lr->lr_event) {
4842         case LE_SET_LMV_MASTER: {
4843                 struct dt_object        *obj;
4844
4845                 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4846                 if (IS_ERR(obj))
4847                         RETURN(PTR_ERR(obj));
4848
4849                 if (likely(dt_object_exists(obj)))
4850                         rc = lfsck_namespace_notify_lmv_master_local(env, com,
4851                                                                      obj);
4852
4853                 lfsck_object_put(env, obj);
4854
4855                 RETURN(rc > 0 ? 0 : rc);
4856         }
4857         case LE_SET_LMV_SLAVE: {
4858                 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4859                         ns->ln_striped_shards_repaired++;
4860
4861                 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4862                                                   LNTF_RECHECK_NAME_HASH, true);
4863
4864                 RETURN(rc > 0 ? 0 : rc);
4865         }
4866         case LE_PHASE1_DONE:
4867         case LE_PHASE2_DONE:
4868         case LE_PEER_EXIT:
4869                 break;
4870         default:
4871                 RETURN(-EINVAL);
4872         }
4873
4874         CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4875                "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4876                lr->lr_index, lr->lr_status, lr->lr_flags2);
4877
4878         spin_lock(&ltds->ltd_lock);
4879         ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
4880         if (ltd == NULL) {
4881                 spin_unlock(&ltds->ltd_lock);
4882
4883                 RETURN(-ENXIO);
4884         }
4885
4886         list_del_init(&ltd->ltd_namespace_phase_list);
4887         switch (lr->lr_event) {
4888         case LE_PHASE1_DONE:
4889                 if (lr->lr_status <= 0) {
4890                         ltd->ltd_namespace_done = 1;
4891                         list_del_init(&ltd->ltd_namespace_list);
4892                         CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4893                                "phase1 for namespace LFSCK: rc = %d.\n",
4894                                lfsck_lfsck2name(lfsck),
4895                                ltd->ltd_index, lr->lr_status);
4896                         ns->ln_flags |= LF_INCOMPLETE;
4897                         fail = true;
4898                         break;
4899                 }
4900
4901                 if (lr->lr_flags2 & LF_INCOMPLETE)
4902                         ns->ln_flags |= LF_INCOMPLETE;
4903
4904                 if (list_empty(&ltd->ltd_namespace_list))
4905                         list_add_tail(&ltd->ltd_namespace_list,
4906                                       &lad->lad_mdt_list);
4907                 list_add_tail(&ltd->ltd_namespace_phase_list,
4908                               &lad->lad_mdt_phase2_list);
4909                 break;
4910         case LE_PHASE2_DONE:
4911                 ltd->ltd_namespace_done = 1;
4912                 list_del_init(&ltd->ltd_namespace_list);
4913                 break;
4914         case LE_PEER_EXIT:
4915                 fail = true;
4916                 ltd->ltd_namespace_done = 1;
4917                 list_del_init(&ltd->ltd_namespace_list);
4918                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4919                         CDEBUG(D_LFSCK,
4920                                "%s: the peer MDT %x exit namespace LFSCK\n",
4921                                lfsck_lfsck2name(lfsck), ltd->ltd_index);
4922                         ns->ln_flags |= LF_INCOMPLETE;
4923                 }
4924                 break;
4925         default:
4926                 break;
4927         }
4928         spin_unlock(&ltds->ltd_lock);
4929
4930         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4931                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4932
4933                 memset(stop, 0, sizeof(*stop));
4934                 stop->ls_status = lr->lr_status;
4935                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4936                 lfsck_stop(env, lfsck->li_bottom, stop);
4937         } else if (lfsck_phase2_next_ready(lad)) {
4938                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4939         }
4940
4941         RETURN(0);
4942 }
4943
4944 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
4945 {
4946         *count += ns->ln_objs_nlink_repaired;
4947         *count += ns->ln_dirent_repaired;
4948         *count += ns->ln_linkea_repaired;
4949         *count += ns->ln_mul_linked_repaired;
4950         *count += ns->ln_unmatched_pairs_repaired;
4951         *count += ns->ln_dangling_repaired;
4952         *count += ns->ln_mul_ref_repaired;
4953         *count += ns->ln_bad_type_repaired;
4954         *count += ns->ln_lost_dirent_repaired;
4955         *count += ns->ln_striped_dirs_disabled;
4956         *count += ns->ln_striped_dirs_repaired;
4957         *count += ns->ln_striped_shards_repaired;
4958         *count += ns->ln_name_hash_repaired;
4959         *count += ns->ln_local_lpf_moved;
4960 }
4961
4962 static int lfsck_namespace_query_all(const struct lu_env *env,
4963                                      struct lfsck_component *com,
4964                                      __u32 *mdts_count, __u64 *repaired)
4965 {
4966         struct lfsck_namespace *ns = com->lc_file_ram;
4967         struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4968         struct lfsck_tgt_desc *ltd;
4969         int idx;
4970         int rc;
4971         ENTRY;
4972
4973         rc = lfsck_query_all(env, com);
4974         if (rc != 0)
4975                 RETURN(rc);
4976
4977         down_read(&ltds->ltd_rw_sem);
4978         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
4979                 ltd = lfsck_ltd2tgt(ltds, idx);
4980                 LASSERT(ltd != NULL);
4981
4982                 mdts_count[ltd->ltd_namespace_status]++;
4983                 *repaired += ltd->ltd_namespace_repaired;
4984         }
4985         up_read(&ltds->ltd_rw_sem);
4986
4987         down_read(&com->lc_sem);
4988         mdts_count[ns->ln_status]++;
4989         lfsck_namespace_repaired(ns, repaired);
4990         up_read(&com->lc_sem);
4991
4992         RETURN(0);
4993 }
4994
4995 static int lfsck_namespace_query(const struct lu_env *env,
4996                                  struct lfsck_component *com,
4997                                  struct lfsck_request *req,
4998                                  struct lfsck_reply *rep,
4999                                  struct lfsck_query *que, int idx)
5000 {
5001         struct lfsck_namespace *ns = com->lc_file_ram;
5002         int rc = 0;
5003
5004         if (que != NULL) {
5005                 LASSERT(com->lc_lfsck->li_master);
5006
5007                 rc = lfsck_namespace_query_all(env, com,
5008                                                que->lu_mdts_count[idx],
5009                                                &que->lu_repaired[idx]);
5010         } else {
5011                 down_read(&com->lc_sem);
5012                 rep->lr_status = ns->ln_status;
5013                 if (req->lr_flags & LEF_QUERY_ALL)
5014                         lfsck_namespace_repaired(ns, &rep->lr_repaired);
5015                 up_read(&com->lc_sem);
5016         }
5017
5018         return rc;
5019 }
5020
5021 static struct lfsck_operations lfsck_namespace_ops = {
5022         .lfsck_reset            = lfsck_namespace_reset,
5023         .lfsck_fail             = lfsck_namespace_fail,
5024         .lfsck_close_dir        = lfsck_namespace_close_dir,
5025         .lfsck_open_dir         = lfsck_namespace_open_dir,
5026         .lfsck_checkpoint       = lfsck_namespace_checkpoint,
5027         .lfsck_prep             = lfsck_namespace_prep,
5028         .lfsck_exec_oit         = lfsck_namespace_exec_oit,
5029         .lfsck_exec_dir         = lfsck_namespace_exec_dir,
5030         .lfsck_post             = lfsck_namespace_post,
5031         .lfsck_dump             = lfsck_namespace_dump,
5032         .lfsck_double_scan      = lfsck_namespace_double_scan,
5033         .lfsck_data_release     = lfsck_namespace_data_release,
5034         .lfsck_quit             = lfsck_namespace_quit,
5035         .lfsck_in_notify        = lfsck_namespace_in_notify,
5036         .lfsck_query            = lfsck_namespace_query,
5037 };
5038
5039 /**
5040  * Repair dangling name entry.
5041  *
5042  * For the name entry with dangling reference, we need to repare the
5043  * inconsistency according to the LFSCK sponsor's requirement:
5044  *
5045  * 1) Keep the inconsistency there and report the inconsistency case,
5046  *    then give the chance to the application to find related issues,
5047  *    and the users can make the decision about how to handle it with
5048  *    more human knownledge. (by default)
5049  *
5050  * 2) Re-create the missing MDT-object with the FID information.
5051  *
5052  * \param[in] env       pointer to the thread context
5053  * \param[in] com       pointer to the lfsck component
5054  * \param[in] parent    pointer to the dir object that contains the dangling
5055  *                      name entry
5056  * \param[in] child     pointer to the object corresponding to the dangling
5057  *                      name entry
5058  * \param[in] lnr       pointer to the namespace request that contains the
5059  *                      name's name, parent object, parent's LMV, and ect.
5060  *
5061  * \retval              positive number if no need to repair
5062  * \retval              zero for repaired successfully
5063  * \retval              negative error number on failure
5064  */
5065 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5066                                     struct lfsck_component *com,
5067                                     struct dt_object *parent,
5068                                     struct dt_object *child,
5069                                     struct lfsck_namespace_req *lnr)
5070 {
5071         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5072         struct lu_attr                  *la     = &info->lti_la;
5073         struct dt_allocation_hint       *hint   = &info->lti_hint;
5074         struct dt_object_format         *dof    = &info->lti_dof;
5075         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
5076         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
5077         const struct lu_name            *cname;
5078         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
5079         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
5080         struct linkea_data               ldata  = { NULL };
5081         struct lfsck_lock_handle        *llh    = &info->lti_llh;
5082         struct lu_buf                    linkea_buf;
5083         struct lu_buf                    lmv_buf;
5084         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5085         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5086         struct dt_device                *dev    = lfsck->li_next;
5087         struct thandle                  *th     = NULL;
5088         int                              rc     = 0;
5089         __u16                            type   = lnr->lnr_type;
5090         bool                             create;
5091         ENTRY;
5092
5093         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5094         if (bk->lb_param & LPF_CREATE_MDTOBJ)
5095                 create = true;
5096         else
5097                 create = false;
5098
5099         if (!create || bk->lb_param & LPF_DRYRUN)
5100                 GOTO(log, rc = 0);
5101
5102         /* We may need to create the sub-objects of the @child via LOD,
5103          * so make the modification based on lfsck->li_next. */
5104
5105         parent = lfsck_object_locate(dev, parent);
5106         if (IS_ERR(parent))
5107                 GOTO(log, rc = PTR_ERR(parent));
5108
5109         if (unlikely(!dt_try_as_dir(env, parent)))
5110                 GOTO(log, rc = -ENOTDIR);
5111
5112         child = lfsck_object_locate(dev, child);
5113         if (IS_ERR(child))
5114                 GOTO(log, rc = PTR_ERR(child));
5115
5116         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5117                               cname, pfid);
5118         if (rc != 0)
5119                 GOTO(log, rc);
5120
5121         rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5122                         MDS_INODELOCK_UPDATE, LCK_PR);
5123         if (rc != 0)
5124                 GOTO(log, rc);
5125
5126         rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5127         if (rc != 0)
5128                 GOTO(log, rc);
5129
5130         /* Set the ctime as zero, then others can know it is created for
5131          * repairing dangling name entry by LFSCK. And if the LFSCK made
5132          * wrong decision and the real MDT-object has been found later,
5133          * then the LFSCK has chance to fix the incosistency properly. */
5134         memset(la, 0, sizeof(*la));
5135         la->la_mode = (type & S_IFMT) | 0600;
5136         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
5137                         LA_ATIME | LA_MTIME | LA_CTIME;
5138
5139         child->do_ops->do_ah_init(env, hint, parent, child,
5140                                  la->la_mode & S_IFMT);
5141
5142         memset(dof, 0, sizeof(*dof));
5143         dof->dof_type = dt_mode_to_dft(type);
5144         /* If the target is a regular file, then the LFSCK will only create
5145          * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5146          * OST-objects will be created when write open. */
5147
5148         th = dt_trans_create(env, dev);
5149         if (IS_ERR(th))
5150                 GOTO(log, rc = PTR_ERR(th));
5151
5152         /* 1a. create child. */
5153         rc = dt_declare_create(env, child, la, hint, dof, th);
5154         if (rc != 0)
5155                 GOTO(stop, rc);
5156
5157         if (S_ISDIR(type)) {
5158                 if (unlikely(!dt_try_as_dir(env, child)))
5159                         GOTO(stop, rc = -ENOTDIR);
5160
5161                 /* 2a. increase child nlink */
5162                 rc = dt_declare_ref_add(env, child, th);
5163                 if (rc != 0)
5164                         GOTO(stop, rc);
5165
5166                 /* 3a. insert dot into child dir */
5167                 rec->rec_type = S_IFDIR;
5168                 rec->rec_fid = cfid;
5169                 rc = dt_declare_insert(env, child,
5170                                        (const struct dt_rec *)rec,
5171                                        (const struct dt_key *)dot, th);
5172                 if (rc != 0)
5173                         GOTO(stop, rc);
5174
5175                 /* 4a. insert dotdot into child dir */
5176                 rec->rec_fid = pfid;
5177                 rc = dt_declare_insert(env, child,
5178                                        (const struct dt_rec *)rec,
5179                                        (const struct dt_key *)dotdot, th);
5180                 if (rc != 0)
5181                         GOTO(stop, rc);
5182
5183                 /* 5a. generate slave LMV EA. */
5184                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5185                         int idx;
5186
5187                         idx = lfsck_shard_name_to_index(env,
5188                                         lnr->lnr_name, lnr->lnr_namelen,
5189                                         type, cfid);
5190                         if (unlikely(idx < 0))
5191                                 GOTO(stop, rc = idx);
5192
5193                         *lmv2 = lnr->lnr_lmv->ll_lmv;
5194                         lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5195                         lmv2->lmv_master_mdt_index = idx;
5196
5197                         lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5198                         lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5199                         rc = dt_declare_xattr_set(env, child, &lmv_buf,
5200                                                   XATTR_NAME_LMV, 0, th);
5201                         if (rc != 0)
5202                                 GOTO(stop, rc);
5203                 }
5204         }
5205
5206         /* 6a. insert linkEA for child */
5207         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5208                        ldata.ld_leh->leh_len);
5209         rc = dt_declare_xattr_set(env, child, &linkea_buf,
5210                                   XATTR_NAME_LINK, 0, th);
5211         if (rc != 0)
5212                 GOTO(stop, rc);
5213
5214         rc = dt_trans_start_local(env, dev, th);
5215         if (rc != 0)
5216                 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5217
5218         dt_write_lock(env, child, 0);
5219         /* 1b. create child */
5220         rc = dt_create(env, child, la, hint, dof, th);
5221         if (rc != 0)
5222                 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5223
5224         if (S_ISDIR(type)) {
5225                 /* 2b. increase child nlink */
5226                 rc = dt_ref_add(env, child, th);
5227                 if (rc != 0)
5228                         GOTO(unlock, rc);
5229
5230                 /* 3b. insert dot into child dir */
5231                 rec->rec_type = S_IFDIR;
5232                 rec->rec_fid = cfid;
5233                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5234                                (const struct dt_key *)dot, th, 1);
5235                 if (rc != 0)
5236                         GOTO(unlock, rc);
5237
5238                 /* 4b. insert dotdot into child dir */
5239                 rec->rec_fid = pfid;
5240                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5241                                (const struct dt_key *)dotdot, th, 1);
5242                 if (rc != 0)
5243                         GOTO(unlock, rc);
5244
5245                 /* 5b. generate slave LMV EA. */
5246                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5247                         rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5248                                           0, th);
5249                         if (rc != 0)
5250                                 GOTO(unlock, rc);
5251                 }
5252         }
5253
5254         /* 6b. insert linkEA for child. */
5255         rc = dt_xattr_set(env, child, &linkea_buf,
5256                           XATTR_NAME_LINK, 0, th);
5257
5258         GOTO(unlock, rc);
5259
5260 unlock:
5261         dt_write_unlock(env, child);
5262
5263 stop:
5264         dt_trans_stop(env, dev, th);
5265
5266 log:
5267         lfsck_unlock(llh);
5268         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5269                "reference for: parent "DFID", child "DFID", type %u, "
5270                "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5271                PFID(pfid), PFID(cfid), type, cname->ln_name,
5272                create ? "Create the lost MDT-object as required" :
5273                         "Keep the MDT-object there by default", rc);
5274
5275         if (rc <= 0) {
5276                 struct lfsck_namespace *ns = com->lc_file_ram;
5277
5278                 ns->ln_flags |= LF_INCONSISTENT;
5279         }
5280
5281         return rc;
5282 }
5283
5284 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5285                                                 struct lfsck_component *com,
5286                                                 struct lfsck_assistant_req *lar)
5287 {
5288         struct lfsck_thread_info   *info     = lfsck_env_info(env);
5289         struct lu_attr             *la       = &info->lti_la;
5290         struct lfsck_instance      *lfsck    = com->lc_lfsck;
5291         struct lfsck_bookmark      *bk       = &lfsck->li_bookmark_ram;
5292         struct lfsck_namespace     *ns       = com->lc_file_ram;
5293         struct lfsck_assistant_data *lad     = com->lc_data;
5294         struct linkea_data          ldata    = { NULL };
5295         const struct lu_name       *cname;
5296         struct thandle             *handle   = NULL;
5297         struct lfsck_namespace_req *lnr      =
5298                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5299         struct dt_object           *dir      = NULL;
5300         struct dt_object           *obj      = NULL;
5301         struct lfsck_assistant_object *lso   = lar->lar_parent;
5302         const struct lu_fid        *pfid     = &lso->lso_fid;
5303         struct dt_device           *dev      = NULL;
5304         struct lustre_handle        lh       = { 0 };
5305         bool                        repaired = false;
5306         bool                        dtlocked = false;
5307         bool                        remove;
5308         bool                        newdata;
5309         bool                        log      = false;
5310         bool                        bad_hash = false;
5311         bool                        bad_linkea = false;
5312         int                         idx      = 0;
5313         int                         count    = 0;
5314         int                         rc       = 0;
5315         enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5316         ENTRY;
5317
5318         if (lso->lso_dead)
5319                 RETURN(0);
5320
5321         la->la_nlink = 0;
5322         if (lnr->lnr_attr & LUDA_UPGRADE) {
5323                 ns->ln_flags |= LF_UPGRADE;
5324                 ns->ln_dirent_repaired++;
5325                 repaired = true;
5326         } else if (lnr->lnr_attr & LUDA_REPAIR) {
5327                 ns->ln_flags |= LF_INCONSISTENT;
5328                 ns->ln_dirent_repaired++;
5329                 repaired = true;
5330         }
5331
5332         if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5333                 if (strcmp(lnr->lnr_name, dotdot) != 0)
5334                         LBUG();
5335                 else
5336                         rc = lfsck_namespace_trace_update(env, com, pfid,
5337                                                 LNTF_CHECK_PARENT, true);
5338
5339                 GOTO(out, rc);
5340         }
5341
5342         if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5343                 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5344                        " for the name entry %.*s under "DFID"\n",
5345                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5346                        lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5347
5348                 if (strcmp(lnr->lnr_name, dotdot) != 0)
5349                         /* invalid FID means bad name entry, remove it. */
5350                         type = LNIT_BAD_DIRENT;
5351                 else
5352                         /* If the parent FID is invalid, we cannot remove
5353                          * the ".." entry directly. */
5354                         rc = lfsck_namespace_trace_update(env, com, pfid,
5355                                                 LNTF_CHECK_PARENT, true);
5356
5357                 GOTO(out, rc);
5358         }
5359
5360         if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5361                 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5362
5363                 RETURN(rc);
5364         }
5365
5366         if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5367                 GOTO(out, rc = 0);
5368
5369         if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5370                 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5371
5372                 RETURN(rc);
5373         }
5374
5375         idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5376         if (idx < 0)
5377                 GOTO(out, rc = idx);
5378
5379         if (idx == lfsck_dev_idx(lfsck)) {
5380                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5381                         GOTO(out, rc = 0);
5382
5383                 dev = lfsck->li_bottom;
5384         } else {
5385                 struct lfsck_tgt_desc *ltd;
5386
5387                 /* Usually, some local filesystem consistency verification
5388                  * tools can guarantee the local namespace tree consistenct.
5389                  * So the LFSCK will only verify the remote directory. */
5390                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5391                         rc = lfsck_namespace_trace_update(env, com, pfid,
5392                                                 LNTF_CHECK_PARENT, true);
5393
5394                         GOTO(out, rc);
5395                 }
5396
5397                 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5398                 if (unlikely(ltd == NULL)) {
5399                         CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5400                                "did not join the namespace LFSCK\n",
5401                                lfsck_lfsck2name(lfsck), idx);
5402                         lfsck_lad_set_bitmap(env, com, idx);
5403
5404                         GOTO(out, rc = -ENODEV);
5405                 }
5406
5407                 dev = ltd->ltd_tgt;
5408         }
5409
5410         obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5411         if (IS_ERR(obj))
5412                 GOTO(out, rc = PTR_ERR(obj));
5413
5414         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5415         if (dt_object_exists(obj) == 0) {
5416
5417 dangling:
5418                 if (dir == NULL) {
5419                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5420                         if (IS_ERR(dir)) {
5421                                 rc = PTR_ERR(dir);
5422
5423                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5424                         }
5425                 }
5426
5427                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5428                 if (rc == 0) {
5429                         if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5430                                         lnr->lnr_name, lnr->lnr_namelen)) {
5431                                 type = LNIT_BAD_DIRENT;
5432
5433                                 GOTO(out, rc);
5434                         }
5435
5436                         type = LNIT_DANGLING;
5437                         rc = lfsck_namespace_repair_dangling(env, com, dir,
5438                                                              obj, lnr);
5439                         if (rc == 0)
5440                                 repaired = true;
5441                 }
5442
5443                 GOTO(out, rc);
5444         }
5445
5446         if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5447
5448 again:
5449                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5450                                       MDS_INODELOCK_UPDATE |
5451                                       MDS_INODELOCK_XATTR, LCK_EX);
5452                 if (rc != 0)
5453                         GOTO(out, rc);
5454
5455                 handle = dt_trans_create(env, dev);
5456                 if (IS_ERR(handle))
5457                         GOTO(out, rc = PTR_ERR(handle));
5458
5459                 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5460                 if (rc != 0)
5461                         GOTO(stop, rc);
5462
5463                 rc = dt_trans_start_local(env, dev, handle);
5464                 if (rc != 0)
5465                         GOTO(stop, rc);
5466
5467                 dt_write_lock(env, obj, 0);
5468                 dtlocked = true;
5469         }
5470
5471         rc = lfsck_links_read(env, obj, &ldata);
5472         if (unlikely(rc == -ENOENT)) {
5473                 if (handle != NULL) {
5474                         dt_write_unlock(env, obj);
5475                         dtlocked = false;
5476
5477                         dt_trans_stop(env, dev, handle);
5478                         handle = NULL;
5479
5480                         lfsck_ibits_unlock(&lh, LCK_EX);
5481                 }
5482
5483                 /* It may happen when the remote object has been removed,
5484                  * but the local MDT is not aware of that. */
5485                 goto dangling;
5486         } else if (rc == 0) {
5487                 count = ldata.ld_leh->leh_reccount;
5488                 rc = linkea_links_find(&ldata, cname, pfid);
5489                 if ((rc == 0) &&
5490                     (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5491                         if ((lfsck_object_type(obj) & S_IFMT) !=
5492                             lnr->lnr_type) {
5493                                 ns->ln_flags |= LF_INCONSISTENT;
5494                                 type = LNIT_BAD_TYPE;
5495                         }
5496
5497                         goto stop;
5498                 }
5499
5500                 /* If the name entry hash does not match the slave striped
5501                  * directory, and the name entry does not match also, then
5502                  * it is quite possible that name entry is corrupted. */
5503                 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5504                                         lnr->lnr_name, lnr->lnr_namelen)) {
5505                         ns->ln_flags |= LF_INCONSISTENT;
5506                         type = LNIT_BAD_DIRENT;
5507
5508                         GOTO(stop, rc = 0);
5509                 }
5510
5511                 /* If the file type stored in the name entry does not match
5512                  * the file type claimed by the object, and the object does
5513                  * not recognize the name entry, then it is quite possible
5514                  * that the name entry is corrupted. */
5515                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5516                         ns->ln_flags |= LF_INCONSISTENT;
5517                         type = LNIT_BAD_DIRENT;
5518
5519                         GOTO(stop, rc = 0);
5520                 }
5521
5522                 /* For sub-dir object, we cannot make sure whether the sub-dir
5523                  * back references the parent via ".." name entry correctly or
5524                  * not in the LFSCK first-stage scanning. It may be that the
5525                  * (remote) sub-dir ".." name entry has no parent FID after
5526                  * file-level backup/restore and its linkEA may be wrong.
5527                  * So under such case, we should replace the linkEA according
5528                  * to current name entry. But this needs to be done during the
5529                  * LFSCK second-stage scanning. The LFSCK will record the name
5530                  * entry for further possible using. */
5531                 remove = false;
5532                 newdata = false;
5533                 goto nodata;
5534         } else if (unlikely(rc == -EINVAL)) {
5535                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5536                         type = LNIT_BAD_TYPE;
5537
5538                 count = 1;
5539                 /* The magic crashed, we are not sure whether there are more
5540                  * corrupt data in the linkea, so remove all linkea entries. */
5541                 remove = true;
5542                 newdata = true;
5543                 goto nodata;
5544         } else if (rc == -ENODATA) {
5545                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5546                         type = LNIT_BAD_TYPE;
5547
5548                 count = 1;
5549                 remove = false;
5550                 newdata = true;
5551
5552 nodata:
5553                 if (bk->lb_param & LPF_DRYRUN) {
5554                         if (rc == -ENODATA)
5555                                 ns->ln_flags |= LF_UPGRADE;
5556                         else
5557                                 ns->ln_flags |= LF_INCONSISTENT;
5558                         ns->ln_linkea_repaired++;
5559                         repaired = true;
5560                         log = true;
5561                         goto stop;
5562                 }
5563
5564                 if (!lustre_handle_is_used(&lh)) {
5565                         remove = false;
5566                         newdata = false;
5567                         type = LNIT_NONE;
5568
5569                         goto again;
5570                 }
5571
5572                 LASSERT(handle != NULL);
5573
5574                 if (dir == NULL) {
5575                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5576                         if (IS_ERR(dir)) {
5577                                 rc = PTR_ERR(dir);
5578
5579                                 GOTO(stop, rc == -ENOENT ? 0 : rc);
5580                         }
5581                 }
5582
5583                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5584                 if (rc != 0)
5585                         GOTO(stop, rc);
5586
5587                 bad_linkea = true;
5588                 if (!remove && newdata)
5589                         ns->ln_flags |= LF_UPGRADE;
5590                 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5591                         ns->ln_flags |= LF_INCONSISTENT;
5592
5593                 if (remove) {
5594                         LASSERT(newdata);
5595
5596                         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5597                         if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5598                                 GOTO(stop, rc);
5599                 }
5600
5601                 if (newdata) {
5602                         rc = linkea_data_new(&ldata,
5603                                         &lfsck_env_info(env)->lti_linkea_buf);
5604                         if (rc != 0)
5605                                 GOTO(stop, rc);
5606                 }
5607
5608                 rc = linkea_add_buf(&ldata, cname, pfid);
5609                 if (rc == 0)
5610                         rc = lfsck_links_write(env, obj, &ldata, handle);
5611                 if (rc != 0)
5612                         GOTO(stop, rc);
5613
5614                 count = ldata.ld_leh->leh_reccount;
5615                 if (!S_ISDIR(lfsck_object_type(obj)) ||
5616                     !dt_object_remote(obj)) {
5617                         ns->ln_linkea_repaired++;
5618                         repaired = true;
5619                         log = true;
5620                 }
5621         } else {
5622                 GOTO(stop, rc);
5623         }
5624
5625 stop:
5626         if (dtlocked)
5627                 dt_write_unlock(env, obj);
5628
5629         if (handle != NULL && !IS_ERR(handle))
5630                 dt_trans_stop(env, dev, handle);
5631
5632 out:
5633         lfsck_ibits_unlock(&lh, LCK_EX);
5634
5635         if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5636             !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5637                                              lnr->lnr_name, lnr->lnr_namelen) &&
5638             type != LNIT_BAD_DIRENT) {
5639                 ns->ln_flags |= LF_INCONSISTENT;
5640
5641                 log = false;
5642                 if (dir == NULL) {
5643                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5644                         if (IS_ERR(dir)) {
5645                                 rc = PTR_ERR(dir);
5646
5647                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5648                         }
5649                 }
5650
5651                 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5652                                                 lnr->lnr_lmv, lnr->lnr_name);
5653                 if (rc == 0)
5654                         bad_hash = true;
5655         }
5656
5657         if (rc >= 0) {
5658                 if (type != LNIT_NONE && dir == NULL) {
5659                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5660                         if (IS_ERR(dir)) {
5661                                 rc = PTR_ERR(dir);
5662
5663                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5664                         }
5665                 }
5666
5667                 switch (type) {
5668                 case LNIT_BAD_TYPE:
5669                         log = false;
5670                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5671                                         obj, lnr->lnr_name, lnr->lnr_name,
5672                                         lnr->lnr_type, true, false);
5673                         if (rc > 0)
5674                                 repaired = true;
5675                         break;
5676                 case LNIT_BAD_DIRENT:
5677                         log = false;
5678                         /* XXX: This is a bad dirent, we do not know whether
5679                          *      the original name entry reference a regular
5680                          *      file or a directory, then keep the parent's
5681                          *      nlink count unchanged here. */
5682                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5683                                         obj, lnr->lnr_name, lnr->lnr_name,
5684                                         lnr->lnr_type, false, false);
5685                         if (rc > 0)
5686                                 repaired = true;
5687                         break;
5688                 default:
5689                         break;
5690                 }
5691
5692                 if (obj != NULL && count == 1 &&
5693                     S_ISREG(lfsck_object_type(obj)))
5694                         dt_attr_get(env, obj, la);
5695         }
5696
5697 trace:
5698         down_write(&com->lc_sem);
5699         if (rc < 0) {
5700                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5701                        "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5702                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5703                        lnr->lnr_namelen, lnr->lnr_name, rc);
5704
5705                 lfsck_namespace_record_failure(env, lfsck, ns);
5706                 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5707                      rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5708                      rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5709                     dev != NULL && dev != lfsck->li_bottom)
5710                         lfsck_lad_set_bitmap(env, com, idx);
5711
5712                 if (!(bk->lb_param & LPF_FAILOUT))
5713                         rc = 0;
5714         } else {
5715                 if (repaired) {
5716                         ns->ln_items_repaired++;
5717                         if (log)
5718                                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5719                                        "repaired the entry: "DFID", parent "DFID
5720                                        ", name %.*s, type %d\n",
5721                                        lfsck_lfsck2name(lfsck),
5722                                        PFID(&lnr->lnr_fid), PFID(pfid),
5723                                        lnr->lnr_namelen, lnr->lnr_name, type);
5724
5725                         switch (type) {
5726                         case LNIT_DANGLING:
5727                                 ns->ln_dangling_repaired++;
5728                                 break;
5729                         case LNIT_BAD_TYPE:
5730                                 ns->ln_bad_type_repaired++;
5731                                 break;
5732                         case LNIT_BAD_DIRENT:
5733                                 ns->ln_dirent_repaired++;
5734                                 break;
5735                         default:
5736                                 break;
5737                         }
5738
5739                         if (bk->lb_param & LPF_DRYRUN &&
5740                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5741                                 lfsck_pos_fill(env, lfsck,
5742                                                &ns->ln_pos_first_inconsistent,
5743                                                false);
5744                 }
5745
5746                 if (bad_hash) {
5747                         ns->ln_name_hash_repaired++;
5748
5749                         /* Not count repeatedly. */
5750                         if (!repaired) {
5751                                 ns->ln_items_repaired++;
5752                                 if (log)
5753                                         CDEBUG(D_LFSCK, "%s: namespace LFSCK "
5754                                                "assistant repaired the entry: "
5755                                                DFID", parent "DFID
5756                                                ", name %.*s\n",
5757                                                lfsck_lfsck2name(lfsck),
5758                                                PFID(&lnr->lnr_fid), PFID(pfid),
5759                                                lnr->lnr_namelen, lnr->lnr_name);
5760                         }
5761
5762                         if (bk->lb_param & LPF_DRYRUN &&
5763                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5764                                 lfsck_pos_fill(env, lfsck,
5765                                                &ns->ln_pos_first_inconsistent,
5766                                                false);
5767                 }
5768
5769                 rc = 0;
5770         }
5771
5772         if (count > 1 || la->la_nlink > 1)
5773                 ns->ln_mul_linked_checked++;
5774
5775         up_write(&com->lc_sem);
5776
5777         if (obj != NULL && !IS_ERR(obj))
5778                 lfsck_object_put(env, obj);
5779
5780         if (dir != NULL && !IS_ERR(dir))
5781                 lfsck_object_put(env, dir);
5782
5783         lad->lad_advance_lock = bad_linkea;
5784
5785         return rc;
5786 }
5787
5788 /**
5789  * Handle one orphan under the backend /lost+found directory
5790  *
5791  * Insert the orphan FID into the namespace LFSCK trace file for further
5792  * processing (via the subsequent namespace LFSCK second-stage scanning).
5793  * At the same time, remove the orphan name entry from backend /lost+found
5794  * directory. There is an interval between the orphan name entry removed
5795  * from the backend /lost+found directory and the orphan FID in the LFSCK
5796  * trace file handled. In such interval, the LFSCK can be reset, then
5797  * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5798  * To guarantee that the orphans can be found when LFSCK run next time
5799  * without e2fsck again, when remove the orphan name entry, the LFSCK
5800  * will set the orphan's ctime attribute as 1. Since normal applications
5801  * cannot change the object's ctime attribute as 1. Then when LFSCK run
5802  * next time, it can record the object (that ctime is 1) in the namespace
5803  * LFSCK trace file during the first-stage scanning.
5804  *
5805  * \param[in] env       pointer to the thread context
5806  * \param[in] com       pointer to the lfsck component
5807  * \param[in] parent    pointer to the object for the backend /lost+found
5808  * \param[in] ent       pointer to the name entry for the target under the
5809  *                      backend /lost+found
5810  *
5811  * \retval              positive for repaired
5812  * \retval              0 if needs to repair nothing
5813  * \retval              negative error number on failure
5814  */
5815 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5816                                               struct lfsck_component *com,
5817                                               struct dt_object *parent,
5818                                               struct lu_dirent *ent)
5819 {
5820         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5821         struct lu_fid                   *key    = &info->lti_fid;
5822         struct lu_attr                  *la     = &info->lti_la;
5823         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5824         struct dt_object                *obj;
5825         struct dt_device                *dev    = lfsck->li_bottom;
5826         struct dt_object                *child  = NULL;
5827         struct thandle                  *th     = NULL;
5828         int                              idx;
5829         int                              rc     = 0;
5830         __u8                             flags  = 0;
5831         bool                             exist  = false;
5832         ENTRY;
5833
5834         child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5835         if (IS_ERR(child))
5836                 RETURN(PTR_ERR(child));
5837
5838         LASSERT(dt_object_exists(child));
5839         LASSERT(!dt_object_remote(child));
5840
5841         idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5842         obj = com->lc_sub_trace_objs[idx].lsto_obj;
5843         fid_cpu_to_be(key, &ent->lde_fid);
5844         rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5845                        (const struct dt_key *)key);
5846         if (rc == 0) {
5847                 exist = true;
5848                 flags |= LNTF_CHECK_ORPHAN;
5849         } else if (rc == -ENOENT) {
5850                 flags = LNTF_CHECK_ORPHAN;
5851         } else {
5852                 GOTO(out, rc);
5853         }
5854
5855         th = dt_trans_create(env, dev);
5856         if (IS_ERR(th))
5857                 GOTO(out, rc = PTR_ERR(th));
5858
5859         /* a1. remove name entry from backend /lost+found */
5860         rc = dt_declare_delete(env, parent,
5861                                (const struct dt_key *)ent->lde_name, th);
5862         if (rc != 0)
5863                 GOTO(stop, rc);
5864
5865         if (S_ISDIR(lfsck_object_type(child))) {
5866                 /* a2. decrease parent's nlink */
5867                 rc = dt_declare_ref_del(env, parent, th);
5868                 if (rc != 0)
5869                         GOTO(stop, rc);
5870         }
5871
5872         if (exist) {
5873                 /* a3. remove child's FID from the LFSCK trace file. */
5874                 rc = dt_declare_delete(env, obj,
5875                                        (const struct dt_key *)key, th);
5876                 if (rc != 0)
5877                         GOTO(stop, rc);
5878         } else {
5879                 /* a4. set child's ctime as 1 */
5880                 memset(la, 0, sizeof(*la));
5881                 la->la_ctime = 1;
5882                 la->la_valid = LA_CTIME;
5883                 rc = dt_declare_attr_set(env, child, la, th);
5884                 if (rc != 0)
5885                         GOTO(stop, rc);
5886         }
5887
5888         /* a5. insert child's FID into the LFSCK trace file. */
5889         rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5890                                (const struct dt_key *)key, th);
5891         if (rc != 0)
5892                 GOTO(stop, rc);
5893
5894         rc = dt_trans_start_local(env, dev, th);
5895         if (rc != 0)
5896                 GOTO(stop, rc);
5897
5898         /* b1. remove name entry from backend /lost+found */
5899         rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5900         if (rc != 0)
5901                 GOTO(stop, rc);
5902
5903         if (S_ISDIR(lfsck_object_type(child))) {
5904                 /* b2. decrease parent's nlink */
5905                 dt_write_lock(env, parent, 0);
5906                 rc = dt_ref_del(env, parent, th);
5907                 dt_write_unlock(env, parent);
5908                 if (rc != 0)
5909                         GOTO(stop, rc);
5910         }
5911
5912         if (exist) {
5913                 /* a3. remove child's FID from the LFSCK trace file. */
5914                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5915                 if (rc != 0)
5916                         GOTO(stop, rc);
5917         } else {
5918                 /* b4. set child's ctime as 1 */
5919                 rc = dt_attr_set(env, child, la, th);
5920                 if (rc != 0)
5921                         GOTO(stop, rc);
5922         }
5923
5924         /* b5. insert child's FID into the LFSCK trace file. */
5925         rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5926                        (const struct dt_key *)key, th, 1);
5927
5928         GOTO(stop, rc = (rc == 0 ? 1 : rc));
5929
5930 stop:
5931         dt_trans_stop(env, dev, th);
5932
5933 out:
5934         lfsck_object_put(env, child);
5935
5936         return rc;
5937 }
5938
5939 /**
5940  * Handle orphans under the backend /lost+found directory
5941  *
5942  * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5943  * and put them under the backend /lost+found directory that is invisible
5944  * to client. The LFSCK will scan such directory, for the original client
5945  * visible orphans, add their fids into the namespace LFSCK trace file,
5946  * then the subsenquent namespace LFSCK second-stage scanning can handle
5947  * them as other objects to be double scanned: either move back to normal
5948  * namespace, or to the global visible orphan directory:
5949  * /ROOT/.lustre/lost+found/MDTxxxx/
5950  *
5951  * \param[in] env       pointer to the thread context
5952  * \param[in] com       pointer to the lfsck component
5953  */
5954 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5955                                            struct lfsck_component *com)
5956 {
5957         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5958         struct lu_dirent                *ent    =
5959                                         (struct lu_dirent *)info->lti_key;
5960         struct lu_seq_range             *range  = &info->lti_range;
5961         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5962         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5963         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5964         struct lfsck_namespace          *ns     = com->lc_file_ram;
5965         struct dt_object                *parent;
5966         const struct dt_it_ops          *iops;
5967         struct dt_it                    *di;
5968         struct seq_server_site          *ss     = lfsck_dev_site(lfsck);
5969         __u64                            cookie;
5970         __u32                            idx    = lfsck_dev_idx(lfsck);
5971         int                              rc     = 0;
5972         __u16                            type;
5973         ENTRY;
5974
5975         parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5976                                           &LU_BACKEND_LPF_FID);
5977         if (IS_ERR(parent)) {
5978                 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5979                        lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5980                 RETURN_EXIT;
5981         }
5982
5983         /* It is normal that the /lost+found does not exist for ZFS backend. */
5984         if (!dt_object_exists(parent))
5985                 GOTO(out, rc = 0);
5986
5987         if (unlikely(!dt_try_as_dir(env, parent)))
5988                 GOTO(out, rc = -ENOTDIR);
5989
5990         CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5991                lfsck_lfsck2name(lfsck));
5992
5993         com->lc_new_scanned = 0;
5994         iops = &parent->do_index_ops->dio_it;
5995         di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
5996         if (IS_ERR(di))
5997                 GOTO(out, rc = PTR_ERR(di));
5998
5999         rc = iops->load(env, di, 0);
6000         if (rc == 0)
6001                 rc = iops->next(env, di);
6002         else if (rc > 0)
6003                 rc = 0;
6004
6005         while (rc == 0) {
6006                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6007                     unlikely(!thread_is_running(thread)))
6008                         break;
6009
6010                 rc = iops->rec(env, di, (struct dt_rec *)ent,
6011                                LUDA_64BITHASH | LUDA_TYPE);
6012                 if (rc == 0)
6013                         rc = lfsck_unpack_ent(ent, &cookie, &type);
6014
6015                 if (unlikely(rc != 0)) {
6016                         CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6017                                "/lost+found: rc = %d\n",
6018                                lfsck_lfsck2name(lfsck), rc);
6019
6020                         goto skip;
6021                 }
6022
6023                 /* skip dot and dotdot entries */
6024                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6025                         goto next;
6026
6027                 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6028                         goto skip;
6029
6030                 if (fid_is_norm(&ent->lde_fid)) {
6031                         fld_range_set_mdt(range);
6032                         rc = fld_local_lookup(env, ss->ss_server_fld,
6033                                               fid_seq(&ent->lde_fid), range);
6034                         if (rc != 0)
6035                                 goto skip;
6036                 } else if (idx != 0) {
6037                         /* If the returned FID is IGIF, then there are three
6038                          * possible cases:
6039                          *
6040                          * 1) The object is upgraded from old Lustre-1.8 with
6041                          *    IGIF assigned to such object.
6042                          * 2) The object is a backend local object and is
6043                          *    invisible to client.
6044                          * 3) The object lost its LMV EA, and since there is
6045                          *    no FID-in-dirent for the orphan in the backend
6046                          *    /lost+found directory, then the low layer will
6047                          *    return IGIF for such object.
6048                          *
6049                          * For MDTx (x != 0), it is either case 2) or case 3),
6050                          * but from the LFSCK view, they are indistinguishable.
6051                          * To be safe, the LFSCK will keep it there and report
6052                          * some message, then the adminstrator can handle that
6053                          * furtherly.
6054                          *
6055                          * For MDT0, it is more possible the case 1). The LFSCK
6056                          * will handle the orphan as an upgraded object. */
6057                         CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6058                                "in the backend /lost+found on the MDT %04x, "
6059                                "to be safe, skip it.\n",
6060                                lfsck_lfsck2name(lfsck), ent->lde_namelen,
6061                                ent->lde_name, PFID(&ent->lde_fid), idx);
6062                         goto skip;
6063                 }
6064
6065                 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6066
6067 skip:
6068                 down_write(&com->lc_sem);
6069                 com->lc_new_scanned++;
6070                 ns->ln_local_lpf_scanned++;
6071                 if (rc > 0)
6072                         ns->ln_local_lpf_moved++;
6073                 else if (rc == 0)
6074                         ns->ln_local_lpf_skipped++;
6075                 else
6076                         ns->ln_local_lpf_failed++;
6077                 up_write(&com->lc_sem);
6078
6079                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6080                         break;
6081
6082 next:
6083                 lfsck_control_speed_by_self(com);
6084                 if (unlikely(!thread_is_running(thread))) {
6085                         rc = 0;
6086                         break;
6087                 }
6088
6089                 rc = iops->next(env, di);
6090         }
6091
6092         iops->put(env, di);
6093         iops->fini(env, di);
6094
6095         EXIT;
6096
6097 out:
6098         CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6099                lfsck_lfsck2name(lfsck), rc);
6100
6101         lfsck_object_put(env, parent);
6102 }
6103
6104 /**
6105  * Rescan the striped directory after the master LMV EA reset.
6106  *
6107  * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6108  * the namespace LFSCK engine scan the striped directory for the first time,
6109  * it will be regarded as a normal directory. As the LFSCK processing, some
6110  * other LFSCK instance on other MDT will find the shard of this striped dir,
6111  * and find that the master MDT-object of the striped directory lost its LMV
6112  * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6113  * notify the LFSCK instance on this MDT to rescan the striped directory.
6114  *
6115  * \param[in] env       pointer to the thread context
6116  * \param[in] com       pointer to the lfsck component
6117  * \param[in] llu       the lfsck_lmv_unit that contains the striped directory
6118  *                      to be rescanned.
6119  *
6120  * \retval              positive number for success
6121  * \retval              0 for LFSCK stopped/paused
6122  * \retval              negative error number on failure
6123  */
6124 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6125                                               struct lfsck_component *com,
6126                                               struct lfsck_lmv_unit *llu)
6127 {
6128         struct lfsck_thread_info        *info   = lfsck_env_info(env);
6129         struct lfsck_instance           *lfsck  = com->lc_lfsck;
6130         struct lfsck_assistant_data     *lad    = com->lc_data;
6131         struct dt_object                *dir;
6132         const struct dt_it_ops          *iops;
6133         struct dt_it                    *di;
6134         struct lu_dirent                *ent    =
6135                         (struct lu_dirent *)info->lti_key;
6136         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
6137         struct ptlrpc_thread            *thread = &lfsck->li_thread;
6138         struct lfsck_assistant_object   *lso    = NULL;
6139         struct lfsck_namespace_req      *lnr;
6140         struct lfsck_assistant_req      *lar;
6141         int                              rc;
6142         __u16                            type;
6143         ENTRY;
6144
6145         LASSERT(list_empty(&lad->lad_req_list));
6146
6147         lfsck->li_lmv = &llu->llu_lmv;
6148         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6149         rc = lfsck_open_dir(env, lfsck, 0);
6150         if (rc != 0)
6151                 RETURN(rc);
6152
6153         dir = lfsck->li_obj_dir;
6154         di = lfsck->li_di_dir;
6155         iops = &dir->do_index_ops->dio_it;
6156         do {
6157                 rc = iops->rec(env, di, (struct dt_rec *)ent,
6158                                lfsck->li_args_dir);
6159                 if (rc == 0)
6160                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6161                                               &type);
6162
6163                 if (rc != 0) {
6164                         if (bk->lb_param & LPF_FAILOUT)
6165                                 GOTO(out, rc);
6166
6167                         goto next;
6168                 }
6169
6170                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6171                         goto next;
6172
6173                 if (lso == NULL) {
6174                         lso = lfsck_assistant_object_init(env,
6175                                 lfsck_dto2fid(dir), NULL,
6176                                 lfsck->li_pos_current.lp_oit_cookie, true);
6177                         if (IS_ERR(lso)) {
6178                                 if (bk->lb_param & LPF_FAILOUT)
6179                                         GOTO(out, rc = PTR_ERR(lso));
6180
6181                                 lso = NULL;
6182                                 goto next;
6183                         }
6184                 }
6185
6186                 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6187                 if (IS_ERR(lnr)) {
6188                         if (bk->lb_param & LPF_FAILOUT)
6189                                 GOTO(out, rc = PTR_ERR(lnr));
6190
6191                         goto next;
6192                 }
6193
6194                 lar = &lnr->lnr_lar;
6195                 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6196                 lfsck_namespace_assistant_req_fini(env, lar);
6197                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6198                         GOTO(out, rc);
6199
6200                 if (unlikely(!thread_is_running(thread)))
6201                         GOTO(out, rc = 0);
6202
6203 next:
6204                 rc = iops->next(env, di);
6205         } while (rc == 0);
6206
6207 out:
6208         if (lso != NULL && !IS_ERR(lso))
6209                 lfsck_assistant_object_put(env, lso);
6210
6211         lfsck_close_dir(env, lfsck, rc);
6212         if (rc <= 0)
6213                 RETURN(rc);
6214
6215         /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6216         if (list_empty(&lad->lad_req_list))
6217                 RETURN(1);
6218
6219         spin_lock(&lad->lad_lock);
6220         lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6221                           lar_list);
6222         list_del_init(&lar->lar_list);
6223         spin_unlock(&lad->lad_lock);
6224
6225         rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6226         lfsck_namespace_assistant_req_fini(env, lar);
6227
6228         RETURN(rc == 0 ? 1 : rc);
6229 }
6230
6231 static int
6232 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6233                                            struct lfsck_component *com,
6234                                            struct dt_object *obj, bool first)
6235 {
6236         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6237         struct ptlrpc_thread    *thread = &lfsck->li_thread;
6238         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
6239         struct lfsck_namespace  *ns     = com->lc_file_ram;
6240         const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
6241         struct dt_object        *target;
6242         struct dt_it            *di;
6243         struct dt_key           *key;
6244         struct lu_fid            fid;
6245         int                      rc;
6246         __u8                     flags  = 0;
6247         ENTRY;
6248
6249         di = iops->init(env, obj, 0);
6250         if (IS_ERR(di))
6251                 RETURN(PTR_ERR(di));
6252
6253         if (first)
6254                 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6255         else
6256                 fid_zero(&fid);
6257         rc = iops->get(env, di, (const struct dt_key *)&fid);
6258         if (rc < 0)
6259                 GOTO(fini, rc);
6260
6261         if (first) {
6262                 /* The start one either has been processed or does not exist,
6263                  * skip it. */
6264                 rc = iops->next(env, di);
6265                 if (rc != 0)
6266                         GOTO(put, rc);
6267         }
6268
6269         do {
6270                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6271                     unlikely(!thread_is_running(thread)))
6272                         GOTO(put, rc = 0);
6273
6274                 key = iops->key(env, di);
6275                 if (IS_ERR(key)) {
6276                         rc = PTR_ERR(key);
6277                         if (rc == -ENOENT)
6278                                 GOTO(put, rc = 1);
6279
6280                         goto checkpoint;
6281                 }
6282
6283                 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6284                 if (!fid_is_sane(&fid)) {
6285                         rc = 0;
6286                         goto checkpoint;
6287                 }
6288
6289                 target = lfsck_object_find_bottom(env, lfsck, &fid);
6290                 if (IS_ERR(target)) {
6291                         rc = PTR_ERR(target);
6292                         goto checkpoint;
6293                 }
6294
6295                 if (dt_object_exists(target)) {
6296                         rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6297                         if (rc == 0) {
6298                                 rc = lfsck_namespace_double_scan_one(env, com,
6299                                                                 target, flags);
6300                                 if (rc == -ENOENT)
6301                                         rc = 0;
6302                         }
6303                 }
6304
6305                 lfsck_object_put(env, target);
6306
6307 checkpoint:
6308                 down_write(&com->lc_sem);
6309                 com->lc_new_checked++;
6310                 com->lc_new_scanned++;
6311                 if (rc >= 0)
6312                         ns->ln_fid_latest_scanned_phase2 = fid;
6313
6314                 if (rc > 0)
6315                         ns->ln_objs_repaired_phase2++;
6316                 else if (rc < 0)
6317                         ns->ln_objs_failed_phase2++;
6318                 up_write(&com->lc_sem);
6319
6320                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6321                         GOTO(put, rc);
6322
6323                 if (unlikely(com->lc_time_next_checkpoint <=
6324                              ktime_get_seconds()) &&
6325                     com->lc_new_checked != 0) {
6326                         down_write(&com->lc_sem);
6327                         ns->ln_run_time_phase2 += ktime_get_seconds() -
6328                                                   com->lc_time_last_checkpoint;
6329                         ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6330                         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6331                         com->lc_new_checked = 0;
6332                         lfsck_namespace_store(env, com);
6333                         up_write(&com->lc_sem);
6334
6335                         com->lc_time_last_checkpoint = ktime_get_seconds();
6336                         com->lc_time_next_checkpoint =
6337                                 com->lc_time_last_checkpoint +
6338                                 LFSCK_CHECKPOINT_INTERVAL;
6339                 }
6340
6341                 lfsck_control_speed_by_self(com);
6342                 if (unlikely(!thread_is_running(thread)))
6343                         GOTO(put, rc = 0);
6344
6345                 rc = iops->next(env, di);
6346         } while (rc == 0);
6347
6348         GOTO(put, rc);
6349
6350 put:
6351         iops->put(env, di);
6352
6353 fini:
6354         iops->fini(env, di);
6355
6356         return rc;
6357 }
6358
6359 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6360                                                 struct lfsck_component *com)
6361 {
6362         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6363         struct lfsck_namespace  *ns     = com->lc_file_ram;
6364         int                      rc;
6365         int                      i;
6366         ENTRY;
6367
6368         while (!list_empty(&lfsck->li_list_lmv)) {
6369                 struct lfsck_lmv_unit *llu;
6370
6371                 spin_lock(&lfsck->li_lock);
6372                 llu = list_entry(lfsck->li_list_lmv.next,
6373                                  struct lfsck_lmv_unit, llu_link);
6374                 list_del_init(&llu->llu_link);
6375                 spin_unlock(&lfsck->li_lock);
6376
6377                 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6378                 if (rc <= 0)
6379                         RETURN(rc);
6380         }
6381
6382         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6383                lfsck_lfsck2name(lfsck));
6384
6385         lfsck_namespace_scan_local_lpf(env, com);
6386
6387         com->lc_new_checked = 0;
6388         com->lc_new_scanned = 0;
6389         com->lc_time_last_checkpoint = ktime_get_seconds();
6390         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6391                                        LFSCK_CHECKPOINT_INTERVAL;
6392
6393         i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6394         rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6395                                 com->lc_sub_trace_objs[i].lsto_obj, true);
6396         while (rc > 0 && ++i < LFSCK_STF_COUNT)
6397                 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6398                                 com->lc_sub_trace_objs[i].lsto_obj, false);
6399
6400         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6401                "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6402
6403         RETURN(rc);
6404 }
6405
6406 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6407                                                struct lfsck_component *com,
6408                                                struct lfsck_position *pos)
6409 {
6410         struct lfsck_assistant_data     *lad = com->lc_data;
6411         struct lfsck_namespace_req      *lnr;
6412
6413         if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6414             LS_SCANNING_PHASE1)
6415                 return;
6416
6417         if (list_empty(&lad->lad_req_list))
6418                 return;
6419
6420         lnr = list_entry(lad->lad_req_list.next,
6421                          struct lfsck_namespace_req,
6422                          lnr_lar.lar_list);
6423         pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6424         pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6425         pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6426 }
6427
6428 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6429                                               struct lfsck_component *com,
6430                                               int rc)
6431 {
6432         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6433         struct lfsck_namespace  *ns     = com->lc_file_ram;
6434
6435         down_write(&com->lc_sem);
6436         ns->ln_run_time_phase2 += ktime_get_seconds() -
6437                                   com->lc_time_last_checkpoint;
6438         ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6439         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6440         com->lc_new_checked = 0;
6441
6442         if (rc > 0) {
6443                 if (ns->ln_flags & LF_INCOMPLETE)
6444                         ns->ln_status = LS_PARTIAL;
6445                 else
6446                         ns->ln_status = LS_COMPLETED;
6447                 ns->ln_flags &= ~LF_SCANNED_ONCE;
6448                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6449                         ns->ln_flags &= ~LF_INCONSISTENT;
6450                 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6451                 ns->ln_success_count++;
6452         } else if (rc == 0) {
6453                 if (lfsck->li_status != 0)
6454                         ns->ln_status = lfsck->li_status;
6455                 else
6456                         ns->ln_status = LS_STOPPED;
6457         } else {
6458                 ns->ln_status = LS_FAILED;
6459         }
6460
6461         rc = lfsck_namespace_store(env, com);
6462         up_write(&com->lc_sem);
6463
6464         return rc;
6465 }
6466
6467 static int
6468 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6469                                                   struct ptlrpc_request *req,
6470                                                   void *args, int rc)
6471 {
6472         if (rc == 0) {
6473                 struct lfsck_async_interpret_args *laia = args;
6474                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
6475
6476                 ltd->ltd_synced_failures = 1;
6477         }
6478
6479         return 0;
6480 }
6481
6482 /**
6483  * Notify remote LFSCK instances about former failures.
6484  *
6485  * The local LFSCK instance has recorded which MDTs have ever failed to respond
6486  * some LFSCK verification requests (maybe because of network issues or the MDT
6487  * itself trouble). During the respond gap the MDT may missed some name entries
6488  * verification, then the MDT cannot know whether related MDT-objects have been
6489  * referenced by related name entries or not, then in the second-stage scanning,
6490  * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6491  * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6492  * name entry for repairing the orphan.
6493  *
6494  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6495  * it will scan the bitmap for the ever failed MDTs, and notify them that they
6496  * have ever missed some name entries verification and should skip the handling
6497  * for orphan MDT-objects.
6498  *
6499  * \param[in] env       pointer to the thread context
6500  * \param[in] com       pointer to the lfsck component
6501  * \param[in] lr        pointer to the lfsck request
6502  */
6503 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6504                                                     struct lfsck_component *com,
6505                                                     struct lfsck_request *lr)
6506 {
6507         struct lfsck_async_interpret_args *laia  =
6508                                 &lfsck_env_info(env)->lti_laia2;
6509         struct lfsck_assistant_data       *lad   = com->lc_data;
6510         struct lfsck_namespace            *ns    = com->lc_file_ram;
6511         struct lfsck_instance             *lfsck = com->lc_lfsck;
6512         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
6513         struct lfsck_tgt_desc             *ltd;
6514         struct ptlrpc_request_set         *set;
6515         __u32                              idx;
6516         int                                rc    = 0;
6517         ENTRY;
6518
6519         if (!lad->lad_incomplete)
6520                 RETURN_EXIT;
6521
6522         set = ptlrpc_prep_set();
6523         if (set == NULL)
6524                 GOTO(out, rc = -ENOMEM);
6525
6526         lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6527         memset(laia, 0, sizeof(*laia));
6528         lad->lad_touch_gen++;
6529
6530         down_read(&ltds->ltd_rw_sem);
6531         cfs_foreach_bit(lad->lad_bitmap, idx) {
6532                 ltd = lfsck_ltd2tgt(ltds, idx);
6533                 if (unlikely(!ltd))
6534                         continue;
6535
6536                 laia->laia_ltd = ltd;
6537                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6538                         lfsck_namespace_assistant_sync_failures_interpret,
6539                         laia, LFSCK_NOTIFY);
6540                 if (rc != 0)
6541                         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6542                                "to sync failure with MDT %x: rc = %d\n",
6543                                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6544         }
6545         up_read(&ltds->ltd_rw_sem);
6546
6547         rc = ptlrpc_set_wait(set);
6548         ptlrpc_set_destroy(set);
6549
6550         GOTO(out, rc);
6551
6552 out:
6553         if (rc != 0)
6554                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6555                        "to sync failure with MDTs, and related MDTs "
6556                        "may handle orphan improperly: rc = %d\n",
6557                        lfsck_lfsck2name(lfsck), rc);
6558
6559         EXIT;
6560 }
6561
6562 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6563         .la_handler_p1          = lfsck_namespace_assistant_handler_p1,
6564         .la_handler_p2          = lfsck_namespace_assistant_handler_p2,
6565         .la_fill_pos            = lfsck_namespace_assistant_fill_pos,
6566         .la_double_scan_result  = lfsck_namespace_double_scan_result,
6567         .la_req_fini            = lfsck_namespace_assistant_req_fini,
6568         .la_sync_failures       = lfsck_namespace_assistant_sync_failures,
6569 };
6570
6571 /**
6572  * Verify the specified linkEA entry for the given directory object.
6573  * If the object has no such linkEA entry or it has more other linkEA
6574  * entries, then re-generate the linkEA with the given information.
6575  *
6576  * \param[in] env       pointer to the thread context
6577  * \param[in] obj       pointer to the dt_object to be handled
6578  * \param[in] cname     the name for the child in the parent directory
6579  * \param[in] pfid      the parent directory's FID for the linkEA
6580  *
6581  * \retval              0 for success
6582  * \retval              negative error number on failure
6583  */
6584 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6585                         const struct lu_name *cname, const struct lu_fid *pfid)
6586 {
6587         struct dt_device        *dev    = lfsck_obj2dev(obj);
6588         struct linkea_data       ldata  = { NULL };
6589         struct lu_buf            linkea_buf;
6590         struct thandle          *th;
6591         int                      rc;
6592         int                      fl     = LU_XATTR_CREATE;
6593         bool                     dirty  = false;
6594         ENTRY;
6595
6596         LASSERT(S_ISDIR(lfsck_object_type(obj)));
6597
6598         rc = lfsck_links_read_with_rec(env, obj, &ldata);
6599         if (rc == -ENODATA) {
6600                 dirty = true;
6601         } else if (rc == 0) {
6602                 fl = LU_XATTR_REPLACE;
6603                 if (ldata.ld_leh->leh_reccount != 1) {
6604                         dirty = true;
6605                 } else {
6606                         rc = linkea_links_find(&ldata, cname, pfid);
6607                         if (rc != 0)
6608                                 dirty = true;
6609                 }
6610         }
6611
6612         if (!dirty)
6613                 RETURN(rc);
6614
6615         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6616                               cname, pfid);
6617         if (rc != 0)
6618                 RETURN(rc);
6619
6620         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6621                        ldata.ld_leh->leh_len);
6622         th = dt_trans_create(env, dev);
6623         if (IS_ERR(th))
6624                 RETURN(PTR_ERR(th));
6625
6626         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6627                                   XATTR_NAME_LINK, fl, th);
6628         if (rc != 0)
6629                 GOTO(stop, rc);
6630
6631         rc = dt_trans_start_local(env, dev, th);
6632         if (rc != 0)
6633                 GOTO(stop, rc);
6634
6635         dt_write_lock(env, obj, 0);
6636         rc = dt_xattr_set(env, obj, &linkea_buf,
6637                           XATTR_NAME_LINK, fl, th);
6638         dt_write_unlock(env, obj);
6639
6640         GOTO(stop, rc);
6641
6642 stop:
6643         dt_trans_stop(env, dev, th);
6644         return rc;
6645 }
6646
6647 /**
6648  * Get the name and parent directory's FID from the first linkEA entry.
6649  *
6650  * \param[in] env       pointer to the thread context
6651  * \param[in] obj       pointer to the object which get linkEA from
6652  * \param[out] name     pointer to the buffer to hold the name
6653  *                      in the first linkEA entry
6654  * \param[out] pfid     pointer to the buffer to hold the parent
6655  *                      directory's FID in the first linkEA entry
6656  *
6657  * \retval              0 for success
6658  * \retval              negative error number on failure
6659  */
6660 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6661                           char *name, struct lu_fid *pfid)
6662 {
6663         struct lu_name           *cname = &lfsck_env_info(env)->lti_name;
6664         struct linkea_data        ldata = { NULL };
6665         int                       rc;
6666
6667         rc = lfsck_links_read_with_rec(env, obj, &ldata);
6668         if (rc)
6669                 return rc;
6670
6671         linkea_first_entry(&ldata);
6672         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6673         /* To guarantee the 'name' is terminated with '0'. */
6674         memcpy(name, cname->ln_name, cname->ln_namelen);
6675         name[cname->ln_namelen] = 0;
6676
6677         return 0;
6678 }
6679
6680 /**
6681  * Update the object's name entry with the given FID.
6682  *
6683  * \param[in] env       pointer to the thread context
6684  * \param[in] lfsck     pointer to the lfsck instance
6685  * \param[in] dir       pointer to the directory that holds
6686  *                      the name entry
6687  * \param[in] name      the name for the entry to be updated
6688  * \param[in] fid       the new FID for the name entry referenced
6689  * \param[in] type      the type for the name entry to be updated
6690  *
6691  * \retval              0 for success
6692  * \retval              negative error number on failure
6693  */
6694 int lfsck_update_name_entry(const struct lu_env *env,
6695                             struct lfsck_instance *lfsck,
6696                             struct dt_object *dir, const char *name,
6697                             const struct lu_fid *fid, __u32 type)
6698 {
6699         struct lfsck_thread_info *info   = lfsck_env_info(env);
6700         struct dt_insert_rec     *rec    = &info->lti_dt_rec;
6701         struct lfsck_lock_handle *llh    = &info->lti_llh;
6702         struct dt_device         *dev    = lfsck_obj2dev(dir);
6703         struct thandle           *th;
6704         int                       rc;
6705         bool                      exists = true;
6706         ENTRY;
6707
6708         rc = lfsck_lock(env, lfsck, dir, name, llh,
6709                         MDS_INODELOCK_UPDATE, LCK_PW);
6710         if (rc != 0)
6711                 RETURN(rc);
6712
6713         th = dt_trans_create(env, dev);
6714         if (IS_ERR(th))
6715                 GOTO(unlock, rc = PTR_ERR(th));
6716
6717         rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6718         if (rc != 0)
6719                 GOTO(stop, rc);
6720
6721         rec->rec_type = type;
6722         rec->rec_fid = fid;
6723         rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6724                                (const struct dt_key *)name, th);
6725         if (rc != 0)
6726                 GOTO(stop, rc);
6727
6728         rc = dt_declare_ref_add(env, dir, th);
6729         if (rc != 0)
6730                 GOTO(stop, rc);
6731
6732         rc = dt_trans_start_local(env, dev, th);
6733         if (rc != 0)
6734                 GOTO(stop, rc);
6735
6736         rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6737         if (rc == -ENOENT) {
6738                 exists = false;
6739                 rc = 0;
6740         }
6741
6742         if (rc != 0)
6743                 GOTO(stop, rc);
6744
6745         rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6746                        (const struct dt_key *)name, th, 1);
6747         if (rc == 0 && S_ISDIR(type) && !exists) {
6748                 dt_write_lock(env, dir, 0);
6749                 rc = dt_ref_add(env, dir, th);
6750                 dt_write_unlock(env, dir);
6751         }
6752
6753         GOTO(stop, rc);
6754
6755 stop:
6756         dt_trans_stop(env, dev, th);
6757
6758 unlock:
6759         lfsck_unlock(llh);
6760         CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6761                " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6762                PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6763
6764         return rc;
6765 }
6766
6767 int lfsck_namespace_setup(const struct lu_env *env,
6768                           struct lfsck_instance *lfsck)
6769 {
6770         struct lfsck_component  *com;
6771         struct lfsck_namespace  *ns;
6772         struct dt_object        *root = NULL;
6773         struct dt_object        *obj;
6774         int                      i;
6775         int                      rc;
6776         ENTRY;
6777
6778         LASSERT(lfsck->li_master);
6779
6780         OBD_ALLOC_PTR(com);
6781         if (com == NULL)
6782                 RETURN(-ENOMEM);
6783
6784         INIT_LIST_HEAD(&com->lc_link);
6785         INIT_LIST_HEAD(&com->lc_link_dir);
6786         init_rwsem(&com->lc_sem);
6787         atomic_set(&com->lc_ref, 1);
6788         com->lc_lfsck = lfsck;
6789         com->lc_type = LFSCK_TYPE_NAMESPACE;
6790         com->lc_ops = &lfsck_namespace_ops;
6791         com->lc_data = lfsck_assistant_data_init(
6792                         &lfsck_namespace_assistant_ops,
6793                         LFSCK_NAMESPACE);
6794         if (com->lc_data == NULL)
6795                 GOTO(out, rc = -ENOMEM);
6796
6797         com->lc_file_size = sizeof(struct lfsck_namespace);
6798         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6799         if (com->lc_file_ram == NULL)
6800                 GOTO(out, rc = -ENOMEM);
6801
6802         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6803         if (com->lc_file_disk == NULL)
6804                 GOTO(out, rc = -ENOMEM);
6805
6806         for (i = 0; i < LFSCK_STF_COUNT; i++)
6807                 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6808
6809         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6810         if (IS_ERR(root))
6811                 GOTO(out, rc = PTR_ERR(root));
6812
6813         if (unlikely(!dt_try_as_dir(env, root)))
6814                 GOTO(out, rc = -ENOTDIR);
6815
6816         obj = local_index_find_or_create(env, lfsck->li_los, root,
6817                                          LFSCK_NAMESPACE,
6818                                          S_IFREG | S_IRUGO | S_IWUSR,
6819                                          &dt_lfsck_namespace_features);
6820         if (IS_ERR(obj))
6821                 GOTO(out, rc = PTR_ERR(obj));
6822
6823         com->lc_obj = obj;
6824         rc = lfsck_namespace_load(env, com);
6825         if (rc == -ENODATA)
6826                 rc = lfsck_namespace_init(env, com);
6827         else if (rc < 0)
6828                 rc = lfsck_namespace_reset(env, com, true);
6829         else
6830                 rc = lfsck_load_sub_trace_files(env, com,
6831                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
6832         if (rc != 0)
6833                 GOTO(out, rc);
6834
6835         ns = com->lc_file_ram;
6836         switch (ns->ln_status) {
6837         case LS_INIT:
6838         case LS_COMPLETED:
6839         case LS_FAILED:
6840         case LS_STOPPED:
6841                 spin_lock(&lfsck->li_lock);
6842                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6843                 spin_unlock(&lfsck->li_lock);
6844                 break;
6845         default:
6846                 CERROR("%s: unknown lfsck_namespace status %d\n",
6847                        lfsck_lfsck2name(lfsck), ns->ln_status);
6848                 /* fall through */
6849         case LS_SCANNING_PHASE1:
6850         case LS_SCANNING_PHASE2:
6851                 /* No need to store the status to disk right now.
6852                  * If the system crashed before the status stored,
6853                  * it will be loaded back when next time. */
6854                 ns->ln_status = LS_CRASHED;
6855                 /* fall through */
6856         case LS_PAUSED:
6857         case LS_CRASHED:
6858                 spin_lock(&lfsck->li_lock);
6859                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6860                 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6861                 spin_unlock(&lfsck->li_lock);
6862                 break;
6863         }
6864
6865         GOTO(out, rc = 0);
6866
6867 out:
6868         if (root != NULL && !IS_ERR(root))
6869                 lfsck_object_put(env, root);
6870         if (rc != 0) {
6871                 lfsck_component_cleanup(env, com);
6872                 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6873                        lfsck_lfsck2name(lfsck), rc);
6874         }
6875         return rc;
6876 }