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         struct lu_fid                   cfid    = {0};
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         if (child)
2041                 cfid = *lfsck_dto2fid(child);
2042         parent = lfsck_object_locate(dev, parent);
2043         if (IS_ERR(parent))
2044                 GOTO(log, rc = PTR_ERR(parent));
2045
2046         if (unlikely(!dt_try_as_dir(env, parent)))
2047                 GOTO(log, rc = -ENOTDIR);
2048
2049         if (!update || strcmp(name, name2) == 0)
2050                 rc = lfsck_lock(env, lfsck, parent, name, llh,
2051                                 MDS_INODELOCK_UPDATE, LCK_PW);
2052         else
2053                 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2054                                       MDS_INODELOCK_UPDATE, LCK_PW);
2055         if (rc != 0)
2056                 GOTO(log, rc);
2057
2058         th = dt_trans_create(env, dev);
2059         if (IS_ERR(th))
2060                 GOTO(unlock1, rc = PTR_ERR(th));
2061
2062         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2063         if (rc != 0)
2064                 GOTO(stop, rc);
2065
2066         if (update) {
2067                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2068                 LASSERT(!fid_is_zero(&cfid));
2069                 rec->rec_fid = &cfid;
2070                 rc = dt_declare_insert(env, parent,
2071                                        (const struct dt_rec *)rec,
2072                                        (const struct dt_key *)name2, th);
2073                 if (rc != 0)
2074                         GOTO(stop, rc);
2075         }
2076
2077         if (dec && S_ISDIR(type)) {
2078                 rc = dt_declare_ref_del(env, parent, th);
2079                 if (rc != 0)
2080                         GOTO(stop, rc);
2081         }
2082
2083         rc = dt_trans_start_local(env, dev, th);
2084         if (rc != 0)
2085                 GOTO(stop, rc);
2086
2087
2088         dt_write_lock(env, parent, 0);
2089         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2090                        (const struct dt_key *)name);
2091         /* Someone has removed the bad name entry by race. */
2092         if (rc == -ENOENT)
2093                 GOTO(unlock2, rc = 0);
2094
2095         if (rc != 0)
2096                 GOTO(unlock2, rc);
2097
2098         /* Someone has removed the bad name entry and reused it for other
2099          * object by race. */
2100         if (!lu_fid_eq(&tfid, &cfid))
2101                 GOTO(unlock2, rc = 0);
2102
2103         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2104                 GOTO(unlock2, rc = 1);
2105
2106         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2107         if (rc != 0)
2108                 GOTO(unlock2, rc);
2109
2110         if (update) {
2111                 rc = dt_insert(env, parent,
2112                                (const struct dt_rec *)rec,
2113                                (const struct dt_key *)name2, th, 1);
2114                 if (rc != 0)
2115                         GOTO(unlock2, rc);
2116         }
2117
2118         if (dec && S_ISDIR(type)) {
2119                 rc = dt_ref_del(env, parent, th);
2120                 if (rc != 0)
2121                         GOTO(unlock2, rc);
2122         }
2123
2124         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2125
2126 unlock2:
2127         dt_write_unlock(env, parent);
2128
2129 stop:
2130         dt_trans_stop(env, dev, th);
2131
2132         /* We are not sure whether the child will become orphan or not.
2133          * Record it in the LFSCK trace file for further checking in
2134          * the second-stage scanning. */
2135         if (!update && !dec && child && rc == 0)
2136                 lfsck_namespace_trace_update(env, com, &cfid,
2137                                              LNTF_CHECK_LINKEA, true);
2138
2139 unlock1:
2140         /* It is harmless even if unlock the unused lock_handle */
2141         lfsck_ibits_unlock(&lh, LCK_PW);
2142         lfsck_unlock(llh);
2143
2144 log:
2145         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2146                "entry for: parent "DFID", child "DFID", name %s, type "
2147                "in name entry %o, type claimed by child %o. repair it "
2148                "by %s with new name2 %s: rc = %d\n",
2149                lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2150                name, type, update ? lfsck_object_type(child) : 0,
2151                update ? "updating" : "removing", name2, rc);
2152
2153         if (rc != 0) {
2154                 struct lfsck_namespace *ns = com->lc_file_ram;
2155
2156                 ns->ln_flags |= LF_INCONSISTENT;
2157         }
2158
2159         return rc;
2160 }
2161
2162 /**
2163  * Update the ".." name entry for the given object.
2164  *
2165  * The object's ".." is corrupted, this function will update the ".." name
2166  * entry with the given pfid, and the linkEA with the given ldata.
2167  *
2168  * The caller should take the ldlm lock before the calling.
2169  *
2170  * \param[in] env       pointer to the thread context
2171  * \param[in] com       pointer to the lfsck component
2172  * \param[in] obj       pointer to the dt_object to be handled
2173  * \param[in] pfid      the new fid for the object's ".." name entry
2174  * \param[in] cname     the name for the @obj in the parent directory
2175  *
2176  * \retval              positive number for repaired cases
2177  * \retval              0 if nothing to be repaired
2178  * \retval              negative error number on failure
2179  */
2180 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2181                                                   struct lfsck_component *com,
2182                                                   struct dt_object *obj,
2183                                                   const struct lu_fid *pfid,
2184                                                   struct lu_name *cname)
2185 {
2186         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2187         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2188         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2189         struct dt_device                *dev    = lfsck_obj2dev(obj);
2190         struct thandle                  *th     = NULL;
2191         struct linkea_data               ldata  = { NULL };
2192         struct lu_buf                    linkea_buf;
2193         int                              rc     = 0;
2194         ENTRY;
2195
2196         LASSERT(!dt_object_remote(obj));
2197         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2198
2199         rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2200         if (rc != 0)
2201                 GOTO(log, rc);
2202
2203         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2204                        ldata.ld_leh->leh_len);
2205
2206         th = dt_trans_create(env, dev);
2207         if (IS_ERR(th))
2208                 GOTO(log, rc = PTR_ERR(th));
2209
2210         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2211         if (rc != 0)
2212                 GOTO(stop, rc);
2213
2214         rec->rec_type = S_IFDIR;
2215         rec->rec_fid = pfid;
2216         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2217                                (const struct dt_key *)dotdot, th);
2218         if (rc != 0)
2219                 GOTO(stop, rc);
2220
2221         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2222                                   XATTR_NAME_LINK, 0, th);
2223         if (rc != 0)
2224                 GOTO(stop, rc);
2225
2226         rc = dt_trans_start_local(env, dev, th);
2227         if (rc != 0)
2228                 GOTO(stop, rc);
2229
2230         dt_write_lock(env, obj, 0);
2231         if (unlikely(lfsck_is_dead_obj(obj)))
2232                 GOTO(unlock, rc = 0);
2233
2234         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2235                 GOTO(unlock, rc = 1);
2236
2237         /* The old ".." name entry maybe not exist. */
2238         dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2239
2240         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2241                        (const struct dt_key *)dotdot, th, 1);
2242         if (rc != 0)
2243                 GOTO(unlock, rc);
2244
2245         rc = lfsck_links_write(env, obj, &ldata, th);
2246
2247         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2248
2249 unlock:
2250         dt_write_unlock(env, obj);
2251
2252 stop:
2253         dt_trans_stop(env, dev, th);
2254
2255 log:
2256         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2257                "the object "DFID", new parent "DFID": rc = %d\n",
2258                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2259                PFID(pfid), rc);
2260
2261         if (rc != 0) {
2262                 struct lfsck_namespace *ns = com->lc_file_ram;
2263
2264                 ns->ln_flags |= LF_INCONSISTENT;
2265         }
2266
2267         return rc;
2268 }
2269
2270 /**
2271  * Handle orphan @obj during Double Scan Directory.
2272  *
2273  * Remove the @obj's current (invalid) linkEA entries, and insert
2274  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2275  * ${FID}-${PFID}-D-${conflict_version}
2276  *
2277  * The caller should take the ldlm lock before the calling.
2278  *
2279  * \param[in] env       pointer to the thread context
2280  * \param[in] com       pointer to the lfsck component
2281  * \param[in] obj       pointer to the orphan object to be handled
2282  * \param[in] pfid      the new fid for the object's ".." name entry
2283  * \param[in,out] lh    ldlm lock handler for the given @obj
2284  * \param[out] type     to tell the caller what the inconsistency is
2285  *
2286  * \retval              positive number for repaired cases
2287  * \retval              0 if nothing to be repaired
2288  * \retval              negative error number on failure
2289  */
2290 static int
2291 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2292                            struct lfsck_component *com,
2293                            struct dt_object *obj,
2294                            const struct lu_fid *pfid,
2295                            struct lustre_handle *lh,
2296                            enum lfsck_namespace_inconsistency_type *type)
2297 {
2298         struct lfsck_thread_info *info = lfsck_env_info(env);
2299         struct lfsck_namespace   *ns   = com->lc_file_ram;
2300         int                       rc;
2301         ENTRY;
2302
2303         /* Remove the unrecognized linkEA. */
2304         rc = lfsck_namespace_links_remove(env, com, obj);
2305         lfsck_ibits_unlock(lh, LCK_EX);
2306         if (rc < 0 && rc != -ENODATA)
2307                 RETURN(rc);
2308
2309         *type = LNIT_MUL_REF;
2310
2311         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2312          * ever tried to verify some remote MDT-object that resides on this
2313          * MDT, but this MDT failed to respond such request. So means there
2314          * may be some remote name entry on other MDT that references this
2315          * object with another name, so we cannot know whether this linkEA
2316          * is valid or not. So keep it there and maybe resolved when next
2317          * LFSCK run. */
2318         if (ns->ln_flags & LF_INCOMPLETE)
2319                 RETURN(0);
2320
2321         /* The unique linkEA is invalid, even if the ".." name entry may be
2322          * valid, we still cannot know via which name entry this directory
2323          * will be referenced. Then handle it as pure orphan. */
2324         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2325                  "-"DFID, PFID(pfid));
2326         rc = lfsck_namespace_insert_orphan(env, com, obj,
2327                                            info->lti_tmpbuf, "D", NULL);
2328
2329         RETURN(rc);
2330 }
2331
2332 /**
2333  * Double Scan Directory object for single linkEA entry case.
2334  *
2335  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2336  * then check whether the name is in the namespace or not, if not, add the
2337  * missing name entry back to namespace. If the linkEA entry is invalid,
2338  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2339  * as an orphan.
2340  *
2341  * \param[in] env       pointer to the thread context
2342  * \param[in] com       pointer to the lfsck component
2343  * \param[in] child     pointer to the directory to be double scanned
2344  * \param[in] pfid      the FID corresponding to the ".." entry
2345  * \param[in] ldata     pointer to the linkEA data for the given @child
2346  * \param[in,out] lh    ldlm lock handler for the given @child
2347  * \param[out] type     to tell the caller what the inconsistency is
2348  * \param[in] retry     if found inconsistency, but the caller does not hold
2349  *                      ldlm lock on the @child, then set @retry as true
2350  * \param[in] unknown   set if does not know how to repair the inconsistency
2351  *
2352  * \retval              positive number for repaired cases
2353  * \retval              0 if nothing to be repaired
2354  * \retval              negative error number on failure
2355  */
2356 static int
2357 lfsck_namespace_dsd_single(const struct lu_env *env,
2358                            struct lfsck_component *com,
2359                            struct dt_object *child,
2360                            const struct lu_fid *pfid,
2361                            struct linkea_data *ldata,
2362                            struct lustre_handle *lh,
2363                            enum lfsck_namespace_inconsistency_type *type,
2364                            bool *retry, bool *unknown)
2365 {
2366         struct lfsck_thread_info *info          = lfsck_env_info(env);
2367         struct lu_name           *cname         = &info->lti_name;
2368         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2369         struct lu_fid             tfid;
2370         struct lfsck_namespace   *ns            = com->lc_file_ram;
2371         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2372         struct dt_object         *parent        = NULL;
2373         struct lmv_mds_md_v1     *lmv;
2374         int                       rc            = 0;
2375         ENTRY;
2376
2377         rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2378                                                  info->lti_key,
2379                                                  sizeof(info->lti_key));
2380         /* The unique linkEA entry with bad parent will be handled as orphan. */
2381         if (rc != 0) {
2382                 if (!lustre_handle_is_used(lh) && retry != NULL)
2383                         *retry = true;
2384                 else
2385                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2386                                                         pfid, lh, type);
2387
2388                 GOTO(out, rc);
2389         }
2390
2391         parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2392         if (IS_ERR(parent))
2393                 GOTO(out, rc = PTR_ERR(parent));
2394
2395         /* We trust the unique linkEA entry in spite of whether it matches the
2396          * ".." name entry or not. Because even if the linkEA entry is wrong
2397          * and the ".." name entry is right, we still cannot know via which
2398          * name entry the child will be referenced, since all known entries
2399          * have been verified during the first-stage scanning. */
2400         if (!dt_object_exists(parent)) {
2401                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2402                  * has ever tried to verify some remote MDT-object that resides
2403                  * on this MDT, but this MDT failed to respond such request. So
2404                  * means there may be some remote name entry on other MDT that
2405                  * references this object with another name, so we cannot know
2406                  * whether this linkEA is valid or not. So keep it there and
2407                  * maybe resolved when next LFSCK run. */
2408                 if (ns->ln_flags & LF_INCOMPLETE)
2409                         GOTO(out, rc = 0);
2410
2411                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2412                         *retry = true;
2413
2414                         GOTO(out, rc = 0);
2415                 }
2416
2417                 lfsck_ibits_unlock(lh, LCK_EX);
2418
2419 lost_parent:
2420                 lmv = &info->lti_lmv;
2421                 rc = lfsck_read_stripe_lmv(env, child, lmv);
2422                 if (rc != 0 && rc != -ENODATA)
2423                         GOTO(out, rc);
2424
2425                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2426                         lmv = NULL;
2427                 } else if (lfsck_shard_name_to_index(env,
2428                                         cname->ln_name, cname->ln_namelen,
2429                                         S_IFDIR, cfid) < 0) {
2430                         /* It is an invalid name entry, we
2431                          * cannot trust the parent also. */
2432                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2433                                                 ldata, cname, &tfid, true);
2434                         if (rc < 0)
2435                                 GOTO(out, rc);
2436
2437                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2438                                  "-"DFID, PFID(pfid));
2439                         rc = lfsck_namespace_insert_orphan(env, com, child,
2440                                                 info->lti_tmpbuf, "S", NULL);
2441
2442                         GOTO(out, rc);
2443                 }
2444
2445                 /* Create the lost parent as an orphan. */
2446                 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2447                 if (rc >= 0) {
2448                         /* Add the missing name entry to the parent. */
2449                         rc = lfsck_namespace_insert_normal(env, com, parent,
2450                                                         child, cname->ln_name);
2451                         if (unlikely(rc == -EEXIST)) {
2452                                 /* Unfortunately, someone reused the name
2453                                  * under the parent by race. So we have
2454                                  * to remove the linkEA entry from
2455                                  * current child object. It means that the
2456                                  * LFSCK cannot recover the system
2457                                  * totally back to its original status,
2458                                  * but it is necessary to make the
2459                                  * current system to be consistent. */
2460                                 rc = lfsck_namespace_shrink_linkea(env,
2461                                                 com, child, ldata,
2462                                                 cname, &tfid, true);
2463                                 if (rc >= 0) {
2464                                         snprintf(info->lti_tmpbuf,
2465                                                  sizeof(info->lti_tmpbuf),
2466                                                  "-"DFID, PFID(pfid));
2467                                         rc = lfsck_namespace_insert_orphan(env,
2468                                                 com, child, info->lti_tmpbuf,
2469                                                 "D", NULL);
2470                                 }
2471                         }
2472                 }
2473
2474                 GOTO(out, rc);
2475         } /* !dt_object_exists(parent) */
2476
2477         /* The unique linkEA entry with bad parent will be handled as orphan. */
2478         if (unlikely(!dt_try_as_dir(env, parent))) {
2479                 if (!lustre_handle_is_used(lh) && retry != NULL)
2480                         *retry = true;
2481                 else
2482                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2483                                                         pfid, lh, type);
2484
2485                 GOTO(out, rc);
2486         }
2487
2488         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2489                        (const struct dt_key *)cname->ln_name);
2490         if (rc == -ENOENT) {
2491                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2492                  * has ever tried to verify some remote MDT-object that resides
2493                  * on this MDT, but this MDT failed to respond such request. So
2494                  * means there may be some remote name entry on other MDT that
2495                  * references this object with another name, so we cannot know
2496                  * whether this linkEA is valid or not. So keep it there and
2497                  * maybe resolved when next LFSCK run. */
2498                 if (ns->ln_flags & LF_INCOMPLETE)
2499                         GOTO(out, rc = 0);
2500
2501                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2502                         *retry = true;
2503
2504                         GOTO(out, rc = 0);
2505                 }
2506
2507                 lfsck_ibits_unlock(lh, LCK_EX);
2508                 rc = lfsck_namespace_check_name(env, parent, child, cname);
2509                 if (rc == -ENOENT)
2510                         goto lost_parent;
2511
2512                 if (rc < 0)
2513                         GOTO(out, rc);
2514
2515                 /* It is an invalid name entry, drop it. */
2516                 if (unlikely(rc > 0)) {
2517                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2518                                                 ldata, cname, &tfid, true);
2519                         if (rc >= 0) {
2520                                 snprintf(info->lti_tmpbuf,
2521                                          sizeof(info->lti_tmpbuf),
2522                                          "-"DFID, PFID(pfid));
2523                                 rc = lfsck_namespace_insert_orphan(env, com,
2524                                         child, info->lti_tmpbuf, "D", NULL);
2525                         }
2526
2527                         GOTO(out, rc);
2528                 }
2529
2530                 /* Add the missing name entry back to the namespace. */
2531                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2532                                                    cname->ln_name);
2533                 if (unlikely(rc == -ESTALE))
2534                         /* It may happen when the remote object has been
2535                          * removed, but the local MDT is not aware of that. */
2536                         goto lost_parent;
2537
2538                 if (unlikely(rc == -EEXIST)) {
2539                         /* Unfortunately, someone reused the name under the
2540                          * parent by race. So we have to remove the linkEA
2541                          * entry from current child object. It means that the
2542                          * LFSCK cannot recover the system totally back to
2543                          * its original status, but it is necessary to make
2544                          * the current system to be consistent.
2545                          *
2546                          * It also may be because of the LFSCK found some
2547                          * internal status of create operation. Under such
2548                          * case, nothing to be done. */
2549                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2550                                         parent, child, ldata, cname, &tfid);
2551                         if (rc >= 0) {
2552                                 snprintf(info->lti_tmpbuf,
2553                                          sizeof(info->lti_tmpbuf),
2554                                          "-"DFID, PFID(pfid));
2555                                 rc = lfsck_namespace_insert_orphan(env, com,
2556                                         child, info->lti_tmpbuf, "D", NULL);
2557                         }
2558                 }
2559
2560                 GOTO(out, rc);
2561         } /* rc == -ENOENT */
2562
2563         if (rc != 0)
2564                 GOTO(out, rc);
2565
2566         if (!lu_fid_eq(&tfid, cfid)) {
2567                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2568                         *retry = true;
2569
2570                         GOTO(out, rc = 0);
2571                 }
2572
2573                 lfsck_ibits_unlock(lh, LCK_EX);
2574                 /* The name entry references another MDT-object that
2575                  * may be created by the LFSCK for repairing dangling
2576                  * name entry. Try to replace it. */
2577                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2578                                                   &tfid, cname);
2579                 if (rc == 0)
2580                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2581                                                         pfid, lh, type);
2582
2583                 GOTO(out, rc);
2584         }
2585
2586         /* Zero FID may because the remote directroy object has invalid linkEA,
2587          * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2588          * how to repair the inconsistency, but the namespace LFSCK on the MDT
2589          * where its name entry resides may has more information (name, FID) to
2590          * repair such inconsistency. So here, keep the inconsistency to avoid
2591          * some imporper repairing. */
2592         if (fid_is_zero(pfid)) {
2593                 if (unknown)
2594                         *unknown = true;
2595
2596                 GOTO(out, rc = 0);
2597         }
2598
2599         /* The ".." name entry is wrong, update it. */
2600         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2601                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2602                         *retry = true;
2603
2604                         GOTO(out, rc = 0);
2605                 }
2606
2607                 *type = LNIT_UNMATCHED_PAIRS;
2608                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2609                                                 lfsck_dto2fid(parent), cname);
2610         }
2611
2612         GOTO(out, rc);
2613
2614 out:
2615         if (parent != NULL && !IS_ERR(parent))
2616                 lfsck_object_put(env, parent);
2617
2618         return rc;
2619 }
2620
2621 /**
2622  * Double Scan Directory object for multiple linkEA entries case.
2623  *
2624  * The given @child has multiple linkEA entries. There is at most one linkEA
2625  * entry will be valid, all the others will be removed. Firstly, the function
2626  * will try to find out the linkEA entry for which the name entry exists under
2627  * the given parent (@pfid). If there is no linkEA entry that matches the given
2628  * ".." name entry, then tries to find out the first linkEA entry that both the
2629  * parent and the name entry exist to rebuild a new ".." name entry.
2630  *
2631  * \param[in] env       pointer to the thread context
2632  * \param[in] com       pointer to the lfsck component
2633  * \param[in] child     pointer to the directory to be double scanned
2634  * \param[in] pfid      the FID corresponding to the ".." entry
2635  * \param[in] ldata     pointer to the linkEA data for the given @child
2636  * \param[in,out] lh    ldlm lock handler for the given @child
2637  * \param[out] type     to tell the caller what the inconsistency is
2638  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2639  * \param[in] unknown   set if does not know how to repair the inconsistency
2640  *
2641  * \retval              positive number for repaired cases
2642  * \retval              0 if nothing to be repaired
2643  * \retval              negative error number on failure
2644  */
2645 static int
2646 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2647                              struct lfsck_component *com,
2648                              struct dt_object *child,
2649                              const struct lu_fid *pfid,
2650                              struct linkea_data *ldata,
2651                              struct lustre_handle *lh,
2652                              enum lfsck_namespace_inconsistency_type *type,
2653                              bool lpf, bool *unknown)
2654 {
2655         struct lfsck_thread_info *info          = lfsck_env_info(env);
2656         struct lu_name           *cname         = &info->lti_name;
2657         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2658         struct lu_fid            *pfid2         = &info->lti_fid3;
2659         struct lu_fid             tfid;
2660         struct lfsck_namespace   *ns            = com->lc_file_ram;
2661         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2662         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2663         struct dt_object         *parent        = NULL;
2664         struct linkea_data        ldata_new     = { NULL };
2665         int                       dirent_count  = 0;
2666         int                       rc            = 0;
2667         bool                      once          = true;
2668         ENTRY;
2669
2670 again:
2671         while (ldata->ld_lee != NULL) {
2672                 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2673                                                          info->lti_key,
2674                                                          sizeof(info->lti_key));
2675                 /* Drop invalid linkEA entry. */
2676                 if (rc != 0) {
2677                         lfsck_linkea_del_buf(ldata, cname);
2678                         continue;
2679                 }
2680
2681                 /* Drop repeated linkEA entries. */
2682                 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2683
2684                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2685                  * then it is possible that: the directry object has ever
2686                  * been lost, but its name entry was there. In the former
2687                  * LFSCK run, during the first-stage scanning, the LFSCK
2688                  * found the dangling name entry, but it did not recreate
2689                  * the lost object, and when moved to the second-stage
2690                  * scanning, some children objects of the lost directory
2691                  * object were found, then the LFSCK recreated such lost
2692                  * directory object as an orphan.
2693                  *
2694                  * When the LFSCK runs again, if the dangling name is still
2695                  * there, the LFSCK should move the orphan directory object
2696                  * back to the normal namespace. */
2697                 if (!lpf && !fid_is_zero(pfid) &&
2698                     !lu_fid_eq(pfid, &tfid) && once) {
2699                         linkea_next_entry(ldata);
2700                         continue;
2701                 }
2702
2703                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2704                 if (IS_ERR(parent))
2705                         RETURN(PTR_ERR(parent));
2706
2707                 if (!dt_object_exists(parent)) {
2708                         lfsck_object_put(env, parent);
2709                         if (ldata->ld_leh->leh_reccount > 1) {
2710                                 /* If it is NOT the last linkEA entry, then
2711                                  * there is still other chance to make the
2712                                  * child to be visible via other parent, then
2713                                  * remove this linkEA entry. */
2714                                 lfsck_linkea_del_buf(ldata, cname);
2715                                 continue;
2716                         }
2717
2718                         break;
2719                 }
2720
2721                 /* The linkEA entry with bad parent will be removed. */
2722                 if (unlikely(!dt_try_as_dir(env, parent))) {
2723                         lfsck_object_put(env, parent);
2724                         lfsck_linkea_del_buf(ldata, cname);
2725                         continue;
2726                 }
2727
2728                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2729                                (const struct dt_key *)cname->ln_name);
2730                 *pfid2 = *lfsck_dto2fid(parent);
2731                 if (rc == -ENOENT) {
2732                         lfsck_object_put(env, parent);
2733                         linkea_next_entry(ldata);
2734                         continue;
2735                 }
2736
2737                 if (rc != 0) {
2738                         lfsck_object_put(env, parent);
2739
2740                         RETURN(rc);
2741                 }
2742
2743                 if (lu_fid_eq(&tfid, cfid)) {
2744                         lfsck_object_put(env, parent);
2745                         /* If the parent (that is declared via linkEA entry)
2746                          * directory contains the specified child, but such
2747                          * parent does not match the dotdot name entry, then
2748                          * trust the linkEA. */
2749                         if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2750                                 *type = LNIT_UNMATCHED_PAIRS;
2751                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2752                                                 com, child, pfid2, cname);
2753
2754                                 RETURN(rc);
2755                         }
2756
2757 rebuild:
2758                         /* It is the most common case that we find the
2759                          * name entry corresponding to the linkEA entry
2760                          * that matches the ".." name entry. */
2761                         rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2762                                               cname, pfid2);
2763                         if (rc != 0)
2764                                 RETURN(rc);
2765
2766                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2767                                                             &ldata_new);
2768                         if (rc < 0)
2769                                 RETURN(rc);
2770
2771                         lfsck_linkea_del_buf(ldata, cname);
2772                         linkea_first_entry(ldata);
2773                         /* There may be some invalid dangling name entries under
2774                          * other parent directories, remove all of them. */
2775                         while (ldata->ld_lee != NULL) {
2776                                 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2777                                                 cname, &tfid, info->lti_key,
2778                                                 sizeof(info->lti_key));
2779                                 if (rc != 0)
2780                                         goto next;
2781
2782                                 parent = lfsck_object_find_bottom(env, lfsck,
2783                                                                   &tfid);
2784                                 if (IS_ERR(parent)) {
2785                                         rc = PTR_ERR(parent);
2786                                         if (rc != -ENOENT &&
2787                                             bk->lb_param & LPF_FAILOUT)
2788                                                 RETURN(rc);
2789
2790                                         goto next;
2791                                 }
2792
2793                                 if (!dt_object_exists(parent)) {
2794                                         lfsck_object_put(env, parent);
2795                                         goto next;
2796                                 }
2797
2798                                 rc = lfsck_namespace_repair_dirent(env, com,
2799                                         parent, child, cname->ln_name,
2800                                         cname->ln_name, S_IFDIR, false, true);
2801                                 lfsck_object_put(env, parent);
2802                                 if (rc < 0) {
2803                                         if (bk->lb_param & LPF_FAILOUT)
2804                                                 RETURN(rc);
2805
2806                                         goto next;
2807                                 }
2808
2809                                 dirent_count += rc;
2810
2811 next:
2812                                 lfsck_linkea_del_buf(ldata, cname);
2813                         }
2814
2815                         ns->ln_dirent_repaired += dirent_count;
2816
2817                         RETURN(rc);
2818                 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
2819
2820                 lfsck_ibits_unlock(lh, LCK_EX);
2821                 /* The name entry references another MDT-object that may be
2822                  * created by the LFSCK for repairing dangling name entry.
2823                  * Try to replace it. */
2824                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2825                                                   &tfid, cname);
2826                 lfsck_object_put(env, parent);
2827                 if (rc < 0)
2828                         RETURN(rc);
2829
2830                 if (rc > 0)
2831                         goto rebuild;
2832
2833                 lfsck_linkea_del_buf(ldata, cname);
2834         } /* while (ldata->ld_lee != NULL) */
2835
2836         /* If there is still linkEA overflow, return. */
2837         if (unlikely(ldata->ld_leh->leh_overflow_time))
2838                 RETURN(0);
2839
2840         linkea_first_entry(ldata);
2841         if (ldata->ld_leh->leh_reccount == 1) {
2842                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2843                                                 lh, type, NULL, unknown);
2844
2845                 RETURN(rc);
2846         }
2847
2848         /* All linkEA entries are invalid and removed, then handle the @child
2849          * as an orphan.*/
2850         if (ldata->ld_leh->leh_reccount == 0) {
2851                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2852                                                 type);
2853
2854                 RETURN(rc);
2855         }
2856
2857         /* If the dangling name entry for the orphan directory object has
2858          * been remvoed, then just check whether the directory object is
2859          * still under the .lustre/lost+found/MDTxxxx/ or not. */
2860         if (lpf) {
2861                 lpf = false;
2862                 goto again;
2863         }
2864
2865         /* There is no linkEA entry that matches the ".." name entry. Find
2866          * the first linkEA entry that both parent and name entry exist to
2867          * rebuild a new ".." name entry. */
2868         if (once) {
2869                 once = false;
2870                 goto again;
2871         }
2872
2873         RETURN(rc);
2874 }
2875
2876 /**
2877  * Repair the object's nlink attribute.
2878  *
2879  * If all the known name entries have been verified, then the object's hard
2880  * link attribute should match the object's linkEA entries count unless the
2881  * object's has too many hard link to be recorded in the linkEA. Such cases
2882  * should have been marked in the LFSCK trace file. Otherwise, trust the
2883  * linkEA to update the object's nlink attribute.
2884  *
2885  * \param[in] env       pointer to the thread context
2886  * \param[in] com       pointer to the lfsck component
2887  * \param[in] obj       pointer to the dt_object to be handled
2888  * \param[in,out] la    pointer to buffer to object's attribute before
2889  *                      and after the repairing
2890  *
2891  * \retval              positive number for repaired cases
2892  * \retval              0 if nothing to be repaired
2893  * \retval              negative error number on failure
2894  */
2895 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2896                                         struct lfsck_component *com,
2897                                         struct dt_object *obj,
2898                                         struct lu_attr *la)
2899 {
2900         struct lfsck_namespace          *ns     = com->lc_file_ram;
2901         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2902         struct dt_device                *dev    = lfsck_obj2dev(obj);
2903         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
2904         struct thandle                  *th     = NULL;
2905         struct linkea_data               ldata  = { NULL };
2906         struct lustre_handle             lh     = { 0 };
2907         __u32                            old    = la->la_nlink;
2908         int                              rc     = 0;
2909         ENTRY;
2910
2911         LASSERT(!dt_object_remote(obj));
2912
2913         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2914                               MDS_INODELOCK_UPDATE, LCK_PW);
2915         if (rc != 0)
2916                 GOTO(log, rc);
2917
2918         th = dt_trans_create(env, dev);
2919         if (IS_ERR(th))
2920                 GOTO(log, rc = PTR_ERR(th));
2921
2922         la->la_valid = LA_NLINK;
2923         rc = dt_declare_attr_set(env, obj, la, th);
2924         if (rc != 0)
2925                 GOTO(stop, rc);
2926
2927         rc = dt_trans_start_local(env, dev, th);
2928         if (rc != 0)
2929                 GOTO(stop, rc);
2930
2931         dt_write_lock(env, obj, 0);
2932         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2933          * ever tried to verify some remote MDT-object that resides on this
2934          * MDT, but this MDT failed to respond such request. So means there
2935          * may be some remote name entry on other MDT that references this
2936          * object with another name, so we cannot know whether this linkEA
2937          * is valid or not. So keep it there and maybe resolved when next
2938          * LFSCK run. */
2939         if (ns->ln_flags & LF_INCOMPLETE)
2940                 GOTO(unlock, rc = 0);
2941
2942         rc = dt_attr_get(env, obj, la);
2943         if (rc != 0)
2944                 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2945
2946         rc = lfsck_links_read2_with_rec(env, obj, &ldata);
2947         if (rc)
2948                 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2949
2950         /* XXX: Currently, we only update the nlink attribute if the known
2951          *      linkEA entries is larger than the nlink attribute. That is
2952          *      safe action. */
2953         if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
2954             unlikely(la->la_nlink == 0 ||
2955                      ldata.ld_leh->leh_overflow_time))
2956                 GOTO(unlock, rc = 0);
2957
2958         la->la_nlink = ldata.ld_leh->leh_reccount;
2959         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2960                 GOTO(unlock, rc = 1);
2961
2962         rc = dt_attr_set(env, obj, la, th);
2963
2964         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2965
2966 unlock:
2967         dt_write_unlock(env, obj);
2968
2969 stop:
2970         dt_trans_stop(env, dev, th);
2971
2972 log:
2973         lfsck_ibits_unlock(&lh, LCK_PW);
2974         CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2975                "nlink count from %u to %u: rc = %d\n",
2976                lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2977
2978         if (rc != 0)
2979                 ns->ln_flags |= LF_INCONSISTENT;
2980
2981         return rc;
2982 }
2983
2984 /**
2985  * Double scan the directory object for namespace LFSCK.
2986  *
2987  * This function will verify the <parent, child> pairs in the namespace tree:
2988  * the parent references the child via some name entry that should be in the
2989  * child's linkEA entry, the child should back references the parent via its
2990  * ".." name entry.
2991  *
2992  * The LFSCK will scan every linkEA entry in turn until find out the first
2993  * matched pairs. If found, then all other linkEA entries will be dropped.
2994  * If all the linkEA entries cannot match the ".." name entry, then there
2995  * are serveral possible cases:
2996  *
2997  * 1) If there is only one linkEA entry, then trust it as long as the PFID
2998  *    in the linkEA entry is valid.
2999  *
3000  * 2) If there are multiple linkEA entries, then try to find the linkEA
3001  *    that matches the ".." name entry. If found, then all other entries
3002  *    are invalid; otherwise, it is quite possible that the ".." name entry
3003  *    is corrupted. Under such case, the LFSCK will rebuild the ".." name
3004  *    entry according to the first valid linkEA entry (both the parent and
3005  *    the name entry should exist).
3006  *
3007  * 3) If the directory object has no (valid) linkEA entry, then the
3008  *    directory object will be handled as pure orphan and inserted
3009  *    in the .lustre/lost+found/MDTxxxx/ with the name:
3010  *    ${self_FID}-${PFID}-D-${conflict_version}
3011  *
3012  * \param[in] env       pointer to the thread context
3013  * \param[in] com       pointer to the lfsck component
3014  * \param[in] child     pointer to the directory object to be handled
3015  * \param[in] flags     to indicate the specical checking on the @child
3016  *
3017  * \retval              positive number for repaired cases
3018  * \retval              0 if nothing to be repaired
3019  * \retval              negative error number on failure
3020  */
3021 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3022                                            struct lfsck_component *com,
3023                                            struct dt_object *child, __u8 flags)
3024 {
3025         struct lfsck_thread_info *info          = lfsck_env_info(env);
3026         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
3027         struct lu_fid            *pfid          = &info->lti_fid2;
3028         struct lfsck_namespace   *ns            = com->lc_file_ram;
3029         struct lfsck_instance    *lfsck         = com->lc_lfsck;
3030         struct lustre_handle      lh            = { 0 };
3031         struct linkea_data        ldata         = { NULL };
3032         bool                      unknown       = false;
3033         bool                      lpf           = false;
3034         bool                      retry         = false;
3035         enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3036         int                       rc            = 0;
3037         ENTRY;
3038
3039         LASSERT(!dt_object_remote(child));
3040
3041         if (flags & LNTF_UNCERTAIN_LMV) {
3042                 if (flags & LNTF_RECHECK_NAME_HASH) {
3043                         rc = lfsck_namespace_scan_shard(env, com, child);
3044                         if (rc < 0)
3045                                 RETURN(rc);
3046
3047                         ns->ln_striped_shards_scanned++;
3048                 } else {
3049                         ns->ln_striped_shards_skipped++;
3050                 }
3051         }
3052
3053         flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3054         if (flags == 0)
3055                 RETURN(0);
3056
3057         if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3058             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3059                 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3060                        "the namespace LFSCK, then the LFSCK cannot guarantee"
3061                        "all the name entries have been verified in first-stage"
3062                        "scanning. So have to skip orphan related handling for"
3063                        "the directory object "DFID" with remote name entry\n",
3064                        lfsck_lfsck2name(lfsck), PFID(cfid));
3065
3066                 RETURN(0);
3067         }
3068
3069         if (unlikely(!dt_try_as_dir(env, child)))
3070                 GOTO(out, rc = -ENOTDIR);
3071
3072         /* We only take ldlm lock on the @child when required. When the
3073          * logic comes here for the first time, it is always false. */
3074         if (0) {
3075
3076 lock:
3077                 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3078                                       MDS_INODELOCK_UPDATE |
3079                                       MDS_INODELOCK_XATTR, LCK_EX);
3080                 if (rc != 0)
3081                         GOTO(out, rc);
3082         }
3083
3084         dt_read_lock(env, child, 0);
3085         if (unlikely(lfsck_is_dead_obj(child))) {
3086                 dt_read_unlock(env, child);
3087
3088                 GOTO(out, rc = 0);
3089         }
3090
3091         rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3092                        (const struct dt_key *)dotdot);
3093         if (rc != 0) {
3094                 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3095                         dt_read_unlock(env, child);
3096
3097                         GOTO(out, rc);
3098                 }
3099
3100                 if (!lustre_handle_is_used(&lh)) {
3101                         dt_read_unlock(env, child);
3102                         goto lock;
3103                 }
3104
3105                 fid_zero(pfid);
3106         } else if (lfsck->li_lpf_obj != NULL &&
3107                    lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3108                 lpf = true;
3109         } else if (unlikely(!fid_is_sane(pfid))) {
3110                 fid_zero(pfid);
3111         }
3112
3113         rc = lfsck_links_read(env, child, &ldata);
3114         dt_read_unlock(env, child);
3115         if (rc != 0) {
3116                 if (rc != -ENODATA && rc != -EINVAL)
3117                         GOTO(out, rc);
3118
3119                 if (!lustre_handle_is_used(&lh))
3120                         goto lock;
3121
3122                 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3123                         /* Remove the corrupted linkEA. */
3124                         rc = lfsck_namespace_links_remove(env, com, child);
3125                         if (rc == 0)
3126                                 /* Here, because of the crashed linkEA, we
3127                                  * cannot know whether there is some parent
3128                                  * that references the child directory via
3129                                  * some name entry or not. So keep it there,
3130                                  * when the LFSCK run next time, if there is
3131                                  * some parent that references this object,
3132                                  * then the LFSCK can rebuild the linkEA;
3133                                  * otherwise, this object will be handled
3134                                  * as orphan as above. */
3135                                 unknown = true;
3136                 } else {
3137                         /* 1. If we have neither ".." nor linkEA,
3138                          *    then it is an orphan.
3139                          *
3140                          * 2. If we only have the ".." name entry,
3141                          *    but no parent references this child
3142                          *    directory, then handle it as orphan. */
3143                         lfsck_ibits_unlock(&lh, LCK_EX);
3144                         type = LNIT_MUL_REF;
3145
3146                         /* If the LFSCK is marked as LF_INCOMPLETE,
3147                          * then means some MDT has ever tried to
3148                          * verify some remote MDT-object that resides
3149                          * on this MDT, but this MDT failed to respond
3150                          * such request. So means there may be some
3151                          * remote name entry on other MDT that
3152                          * references this object with another name,
3153                          * so we cannot know whether this linkEA is
3154                          * valid or not. So keep it there and maybe
3155                          * resolved when next LFSCK run. */
3156                         if (ns->ln_flags & LF_INCOMPLETE)
3157                                 GOTO(out, rc = 0);
3158
3159                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3160                                  "-"DFID, PFID(pfid));
3161                         rc = lfsck_namespace_insert_orphan(env, com, child,
3162                                                 info->lti_tmpbuf, "D", NULL);
3163                 }
3164
3165                 GOTO(out, rc);
3166         } /* rc != 0 */
3167
3168         linkea_first_entry(&ldata);
3169         /* This is the most common case: the object has unique linkEA entry. */
3170         if (ldata.ld_leh->leh_reccount == 1) {
3171                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3172                                                 &lh, &type, &retry, &unknown);
3173                 if (retry) {
3174                         LASSERT(!lustre_handle_is_used(&lh));
3175
3176                         retry = false;
3177                         goto lock;
3178                 }
3179
3180                 GOTO(out, rc);
3181         }
3182
3183         if (!lustre_handle_is_used(&lh))
3184                 goto lock;
3185
3186         if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3187                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3188                                                 &type);
3189
3190                 GOTO(out, rc);
3191         }
3192
3193         /* When we come here, the cases usually like that:
3194          * 1) The directory object has a corrupted linkEA entry. During the
3195          *    first-stage scanning, the LFSCK cannot know such corruption,
3196          *    then it appends the right linkEA entry according to the found
3197          *    name entry after the bad one.
3198          *
3199          * 2) The directory object has a right linkEA entry. During the
3200          *    first-stage scanning, the LFSCK finds some bad name entry,
3201          *    but the LFSCK cannot aware that at that time, then it adds
3202          *    the bad linkEA entry for further processing. */
3203         rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3204                                           &lh, &type, lpf, &unknown);
3205
3206         GOTO(out, rc);
3207
3208 out:
3209         lfsck_ibits_unlock(&lh, LCK_EX);
3210         if (rc > 0) {
3211                 switch (type) {
3212                 case LNIT_BAD_LINKEA:
3213                         ns->ln_linkea_repaired++;
3214                         break;
3215                 case LNIT_UNMATCHED_PAIRS:
3216                         ns->ln_unmatched_pairs_repaired++;
3217                         break;
3218                 case LNIT_MUL_REF:
3219                         ns->ln_mul_ref_repaired++;
3220                         break;
3221                 default:
3222                         break;
3223                 }
3224         }
3225
3226         if (unknown)
3227                 ns->ln_unknown_inconsistency++;
3228
3229         return rc;
3230 }
3231
3232 static inline bool
3233 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3234                                       struct lfsck_namespace *ns)
3235 {
3236         /* Both the leh_overflow_time and ln_time_latest_reset are
3237          * local time based, so need NOT to care about clock drift
3238          * among the servers. */
3239         return ldata->ld_leh->leh_overflow_time &&
3240                ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3241 }
3242
3243 /**
3244  * Clear the object's linkEA overflow timestamp.
3245  *
3246  * If the MDT-object has too many hard links as to the linkEA cannot hold
3247  * all of them, then overflow timestamp will be set in the linkEA header.
3248  * If some hard links are removed after that, then it is possible to hold
3249  * other missed linkEA entries. If the namespace LFSCK have added all the
3250  * related linkEA entries, then it will remove the overflow timestamp.
3251  *
3252  * \param[in] env       pointer to the thread context
3253  * \param[in] com       pointer to the lfsck component
3254  * \param[in] ldata     pointer to the linkEA data for the given @obj
3255  * \param[in] obj       pointer to the dt_object to be handled
3256  *
3257  * \retval              positive number for repaired cases
3258  * \retval              0 if nothing to be repaired
3259  * \retval              negative error number on failure
3260  */
3261 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3262                                                  struct lfsck_component *com,
3263                                                  struct linkea_data *ldata,
3264                                                  struct dt_object *obj)
3265 {
3266         struct lfsck_namespace *ns = com->lc_file_ram;
3267         struct lfsck_instance *lfsck = com->lc_lfsck;
3268         struct dt_device *dev = lfsck_obj2dev(obj);
3269         struct thandle *th = NULL;
3270         struct lustre_handle lh = { 0 };
3271         struct lu_buf linkea_buf;
3272         int rc = 0;
3273         ENTRY;
3274
3275         LASSERT(!dt_object_remote(obj));
3276
3277         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3278                               MDS_INODELOCK_UPDATE, LCK_PW);
3279         if (rc != 0)
3280                 GOTO(log, rc);
3281
3282         th = dt_trans_create(env, dev);
3283         if (IS_ERR(th))
3284                 GOTO(log, rc = PTR_ERR(th));
3285
3286         rc = dt_declare_xattr_set(env, obj,
3287                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3288                         XATTR_NAME_LINK, 0, th);
3289         if (rc != 0)
3290                 GOTO(stop, rc);
3291
3292         rc = dt_trans_start_local(env, dev, th);
3293         if (rc != 0)
3294                 GOTO(stop, rc);
3295
3296         dt_write_lock(env, obj, 0);
3297         rc = lfsck_links_read(env, obj, ldata);
3298         if (rc != 0)
3299                 GOTO(unlock, rc);
3300
3301         if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3302                 GOTO(unlock, rc = 0);
3303
3304         ldata->ld_leh->leh_overflow_time = 0;
3305         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3306                 GOTO(unlock, rc = 1);
3307
3308         /* If all known entries are in the linkEA, then the 'leh_reccount'
3309          * should NOT be zero. */
3310         LASSERT(ldata->ld_leh->leh_reccount > 0);
3311
3312         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3313                        ldata->ld_leh->leh_len);
3314         rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3315         if (unlikely(rc == -ENOSPC))
3316                 rc = 0;
3317         else if (!rc)
3318                 rc = 1;
3319
3320         GOTO(unlock, rc);
3321
3322 unlock:
3323         dt_write_unlock(env, obj);
3324
3325 stop:
3326         dt_trans_stop(env, dev, th);
3327
3328 log:
3329         lfsck_ibits_unlock(&lh, LCK_PW);
3330         CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3331                DFID": rc = %d\n",
3332                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3333
3334         return rc;
3335 }
3336
3337 /**
3338  * Verify the object's agent entry.
3339  *
3340  * If the object claims to have agent entry but the linkEA does not contain
3341  * remote parent, then remove the agent entry. Otherwise, if the object has
3342  * no agent entry but its linkEA contains remote parent, then will generate
3343  * agent entry for it.
3344  *
3345  * \param[in] env       pointer to the thread context
3346  * \param[in] com       pointer to the lfsck component
3347  * \param[in] obj       pointer to the dt_object to be handled
3348  *
3349  * \retval              positive number for repaired cases
3350  * \retval              0 if nothing to be repaired
3351  * \retval              negative error number on failure
3352  */
3353 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3354                                              struct lfsck_component *com,
3355                                              struct dt_object *obj)
3356 {
3357         struct linkea_data ldata = { NULL };
3358         struct lfsck_thread_info *info = lfsck_env_info(env);
3359         struct lfsck_namespace *ns = com->lc_file_ram;
3360         struct lfsck_instance *lfsck = com->lc_lfsck;
3361         struct lu_fid *pfid = &info->lti_fid2;
3362         struct lu_name *cname = &info->lti_name;
3363         struct lu_seq_range *range = &info->lti_range;
3364         struct seq_server_site *ss = lfsck_dev_site(lfsck);
3365         __u32 idx = lfsck_dev_idx(lfsck);
3366         int rc;
3367         bool remote = false;
3368         ENTRY;
3369
3370         if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3371                 RETURN(0);
3372
3373         rc = lfsck_links_read_with_rec(env, obj, &ldata);
3374         if (rc == -ENOENT || rc == -ENODATA)
3375                 RETURN(0);
3376
3377         if (rc && rc != -EINVAL)
3378                 GOTO(out, rc);
3379
3380         /* We check the agent entry again after verifying the linkEA
3381          * successfully. So invalid linkEA should be dryrun mode. */
3382         if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3383                 RETURN(0);
3384
3385         linkea_first_entry(&ldata);
3386         while (ldata.ld_lee != NULL && !remote) {
3387                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3388                                     cname, pfid);
3389                 /* If parent FID is unknown, not verify agent entry. */
3390                 if (!fid_is_sane(pfid))
3391                         GOTO(out, rc = 0);
3392
3393                 fld_range_set_mdt(range);
3394                 rc = fld_server_lookup(env, ss->ss_server_fld,
3395                                        fid_seq(pfid), range);
3396                 if (rc)
3397                         GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3398
3399                 if (range->lsr_index != idx)
3400                         remote = true;
3401                 else
3402                         linkea_next_entry(&ldata);
3403         }
3404
3405         if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3406             (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3407                 struct dt_device *dev = lfsck_obj2dev(obj);
3408                 struct linkea_data ldata2 = { NULL };
3409                 struct lustre_handle lh = { 0 };
3410                 struct lu_buf linkea_buf;
3411                 struct thandle *handle;
3412
3413                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3414                         GOTO(out, rc = 1);
3415
3416                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3417                                       MDS_INODELOCK_UPDATE |
3418                                       MDS_INODELOCK_XATTR, LCK_EX);
3419                 if (rc)
3420                         GOTO(out, rc);
3421
3422                 handle = dt_trans_create(env, dev);
3423                 if (IS_ERR(handle))
3424                         GOTO(unlock, rc = PTR_ERR(handle));
3425
3426                 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3427                                ldata.ld_leh->leh_len);
3428                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3429                                 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3430                 if (rc)
3431                         GOTO(stop, rc);
3432
3433                 rc = dt_trans_start_local(env, dev, handle);
3434                 if (rc)
3435                         GOTO(stop, rc);
3436
3437                 dt_write_lock(env, obj, 0);
3438                 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3439                 if (rc) {
3440                         if (rc == -ENOENT || rc == -ENODATA)
3441                                 rc = 0;
3442                         GOTO(unlock2, rc);
3443                 }
3444
3445                 /* If someone changed linkEA by race, then the agent
3446                  * entry will be updated by lower layer automatically. */
3447                 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3448                     memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3449                            ldata.ld_leh->leh_len) != 0)
3450                         GOTO(unlock2, rc = 0);
3451
3452                 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3453                                   LU_XATTR_REPLACE, handle);
3454                 if (!rc)
3455                         rc = 1;
3456
3457                 GOTO(unlock2, rc);
3458
3459 unlock2:
3460                 dt_write_unlock(env, obj);
3461 stop:
3462                 dt_trans_stop(env, dev, handle);
3463 unlock:
3464                 lfsck_ibits_unlock(&lh, LCK_EX);
3465         }
3466
3467         GOTO(out, rc);
3468
3469 out:
3470         if (rc > 0)
3471                 ns->ln_agent_entries_repaired++;
3472         if (rc)
3473                 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3474                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3475         return rc;
3476 }
3477
3478 /**
3479  * Double scan the MDT-object for namespace LFSCK.
3480  *
3481  * If the MDT-object contains invalid or repeated linkEA entries, then drop
3482  * those entries from the linkEA; if the linkEA becomes empty or the object
3483  * has no linkEA, then it is an orphan and will be added into the directory
3484  * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3485  * the remote parent; if the name entry corresponding to some linkEA entry
3486  * is lost, then add the name entry back to the namespace.
3487  *
3488  * \param[in] env       pointer to the thread context
3489  * \param[in] com       pointer to the lfsck component
3490  * \param[in] child     pointer to the dt_object to be handled
3491  * \param[in] flags     some hints to indicate how the @child should be handled
3492  *
3493  * \retval              positive number for repaired cases
3494  * \retval              0 if nothing to be repaired
3495  * \retval              negative error number on failure
3496  */
3497 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3498                                            struct lfsck_component *com,
3499                                            struct dt_object *child, __u8 flags)
3500 {
3501         struct lfsck_thread_info *info     = lfsck_env_info(env);
3502         struct lu_attr           *la       = &info->lti_la;
3503         struct lu_name           *cname    = &info->lti_name;
3504         struct lu_fid            *pfid     = &info->lti_fid;
3505         struct lu_fid            *cfid     = &info->lti_fid2;
3506         struct lfsck_instance    *lfsck    = com->lc_lfsck;
3507         struct lfsck_namespace   *ns       = com->lc_file_ram;
3508         struct dt_object         *parent   = NULL;
3509         struct linkea_data        ldata    = { NULL };
3510         bool                      repaired = false;
3511         int                       count    = 0;
3512         int                       rc;
3513         ENTRY;
3514
3515         dt_read_lock(env, child, 0);
3516         if (unlikely(lfsck_is_dead_obj(child))) {
3517                 dt_read_unlock(env, child);
3518
3519                 RETURN(0);
3520         }
3521
3522         if (S_ISDIR(lfsck_object_type(child))) {
3523                 dt_read_unlock(env, child);
3524                 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3525                 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3526                         rc = lfsck_namespace_check_agent_entry(env, com, child);
3527
3528                 RETURN(rc);
3529         }
3530
3531         rc = lfsck_links_read(env, child, &ldata);
3532         dt_read_unlock(env, child);
3533
3534         if (rc == -EINVAL) {
3535                 struct lustre_handle lh = { 0 };
3536
3537                 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3538                                       MDS_INODELOCK_UPDATE |
3539                                       MDS_INODELOCK_XATTR, LCK_EX);
3540                 if (rc == 0) {
3541                         rc = lfsck_namespace_links_remove(env, com, child);
3542                         lfsck_ibits_unlock(&lh, LCK_EX);
3543                 }
3544
3545                 GOTO(out, rc);
3546         }
3547
3548         if (rc != 0)
3549                 GOTO(out, rc);
3550
3551         if (!(ns->ln_flags & LF_INCOMPLETE) &&
3552             unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3553                 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3554                                                            child);
3555                 if (rc < 0)
3556                         GOTO(out, rc);
3557
3558                 if (rc > 0)
3559                         ns->ln_linkea_overflow_cleared++;
3560         }
3561
3562         linkea_first_entry(&ldata);
3563         while (ldata.ld_lee != NULL) {
3564                 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3565                                                          info->lti_key,
3566                                                          sizeof(info->lti_key));
3567                 /* Invalid PFID in the linkEA entry. */
3568                 if (rc != 0) {
3569                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3570                                                 &ldata, cname, pfid, true);
3571                         if (rc < 0)
3572                                 GOTO(out, rc);
3573
3574                         if (rc > 0)
3575                                 repaired = true;
3576
3577                         continue;
3578                 }
3579
3580                 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3581                                                          false);
3582                 /* Found repeated linkEA entries */
3583                 if (rc > 0) {
3584                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3585                                                 &ldata, cname, pfid, false);
3586                         if (rc < 0)
3587                                 GOTO(out, rc);
3588
3589                         if (rc == 0)
3590                                 continue;
3591
3592                         repaired = true;
3593
3594                         /* fall through */
3595                 }
3596
3597                 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3598                 if (IS_ERR(parent))
3599                         GOTO(out, rc = PTR_ERR(parent));
3600
3601                 if (!dt_object_exists(parent)) {
3602
3603 lost_parent:
3604                         if (ldata.ld_leh->leh_reccount > 1) {
3605                                 /* If it is NOT the last linkEA entry, then
3606                                  * there is still other chance to make the
3607                                  * child to be visible via other parent, then
3608                                  * remove this linkEA entry. */
3609                                 rc = lfsck_namespace_shrink_linkea(env, com,
3610                                         child, &ldata, cname, pfid, true);
3611                         } else {
3612                                 /* If the LFSCK is marked as LF_INCOMPLETE,
3613                                  * then means some MDT has ever tried to
3614                                  * verify some remote MDT-object that resides
3615                                  * on this MDT, but this MDT failed to respond
3616                                  * such request. So means there may be some
3617                                  * remote name entry on other MDT that
3618                                  * references this object with another name,
3619                                  * so we cannot know whether this linkEA is
3620                                  * valid or not. So keep it there and maybe
3621                                  * resolved when next LFSCK run. */
3622                                 if (ns->ln_flags & LF_INCOMPLETE) {
3623                                         lfsck_object_put(env, parent);
3624
3625                                         GOTO(out, rc = 0);
3626                                 }
3627
3628                                 /* Create the lost parent as an orphan. */
3629                                 rc = lfsck_namespace_create_orphan_dir(env, com,
3630                                                                 parent, NULL);
3631                                 if (rc < 0) {
3632                                         lfsck_object_put(env, parent);
3633
3634                                         GOTO(out, rc);
3635                                 }
3636
3637                                 if (rc > 0)
3638                                         repaired = true;
3639
3640                                 /* Add the missing name entry to the parent. */
3641                                 rc = lfsck_namespace_insert_normal(env, com,
3642                                                 parent, child, cname->ln_name);
3643                                 if (unlikely(rc == -EEXIST))
3644                                         /* Unfortunately, someone reused the
3645                                          * name under the parent by race. So we
3646                                          * have to remove the linkEA entry from
3647                                          * current child object. It means that
3648                                          * the LFSCK cannot recover the system
3649                                          * totally back to its original status,
3650                                          * but it is necessary to make the
3651                                          * current system to be consistent. */
3652                                         rc = lfsck_namespace_shrink_linkea(env,
3653                                                         com, child, &ldata,
3654                                                         cname, pfid, true);
3655                                 else
3656                                         linkea_next_entry(&ldata);
3657                         }
3658
3659                         lfsck_object_put(env, parent);
3660                         if (rc < 0)
3661                                 GOTO(out, rc);
3662
3663                         if (rc > 0)
3664                                 repaired = true;
3665
3666                         continue;
3667                 } /* !dt_object_exists(parent) */
3668
3669                 /* The linkEA entry with bad parent will be removed. */
3670                 if (unlikely(!dt_try_as_dir(env, parent))) {
3671                         lfsck_object_put(env, parent);
3672                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3673                                                 &ldata, cname, pfid, true);
3674                         if (rc < 0)
3675                                 GOTO(out, rc);
3676
3677                         if (rc > 0)
3678                                 repaired = true;
3679
3680                         continue;
3681                 }
3682
3683                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3684                                (const struct dt_key *)cname->ln_name);
3685                 if (rc != 0 && rc != -ENOENT) {
3686                         lfsck_object_put(env, parent);
3687
3688                         GOTO(out, rc);
3689                 }
3690
3691                 if (rc == 0) {
3692                         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3693                                 /* It is the most common case that we
3694                                  * find the name entry corresponding
3695                                  * to the linkEA entry. */
3696                                 lfsck_object_put(env, parent);
3697                                 linkea_next_entry(&ldata);
3698                         } else {
3699                                 /* The name entry references another
3700                                  * MDT-object that may be created by
3701                                  * the LFSCK for repairing dangling
3702                                  * name entry. Try to replace it. */
3703                                 rc = lfsck_namespace_replace_cond(env, com,
3704                                                 parent, child, cfid, cname);
3705                                 lfsck_object_put(env, parent);
3706                                 if (rc < 0)
3707                                         GOTO(out, rc);
3708
3709                                 if (rc > 0) {
3710                                         repaired = true;
3711                                         linkea_next_entry(&ldata);
3712                                 } else {
3713                                         rc = lfsck_namespace_shrink_linkea(env,
3714                                                         com, child, &ldata,
3715                                                         cname, pfid, true);
3716                                         if (rc < 0)
3717                                                 GOTO(out, rc);
3718
3719                                         if (rc > 0)
3720                                                 repaired = true;
3721                                 }
3722                         }
3723
3724                         continue;
3725                 }
3726
3727                 /* The following handles -ENOENT case */
3728
3729                 rc = dt_attr_get(env, child, la);
3730                 if (rc != 0)
3731                         GOTO(out, rc);
3732
3733                 /* If there is no name entry in the parent dir and the object
3734                  * link count is fewer than the linkea entries count, then the
3735                  * linkea entry should be removed. */
3736                 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3737                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3738                                         parent, child, &ldata, cname, pfid);
3739                         lfsck_object_put(env, parent);
3740                         if (rc < 0)
3741                                 GOTO(out, rc);
3742
3743                         if (rc > 0)
3744                                 repaired = true;
3745
3746                         continue;
3747                 }
3748
3749                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3750                  * MDT has ever tried to verify some remote MDT-object that
3751                  * resides on this MDT, but this MDT failed to respond such
3752                  * request. So means there may be some remote name entry on
3753                  * other MDT that references this object with another name,
3754                  * so we cannot know whether this linkEA is valid or not.
3755                  * So keep it there and maybe resolved when next LFSCK run. */
3756                 if (ns->ln_flags & LF_INCOMPLETE) {
3757                         lfsck_object_put(env, parent);
3758
3759                         GOTO(out, rc = 0);
3760                 }
3761
3762                 rc = lfsck_namespace_check_name(env, parent, child, cname);
3763                 if (rc == -ENOENT)
3764                         goto lost_parent;
3765
3766                 if (rc < 0) {
3767                         lfsck_object_put(env, parent);
3768
3769                         GOTO(out, rc);
3770                 }
3771
3772                 /* It is an invalid name entry, drop it. */
3773                 if (unlikely(rc > 0)) {
3774                         lfsck_object_put(env, parent);
3775                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3776                                                 &ldata, cname, pfid, true);
3777                         if (rc < 0)
3778                                 GOTO(out, rc);
3779
3780                         if (rc > 0)
3781                                 repaired = true;
3782
3783                         continue;
3784                 }
3785
3786                 /* Add the missing name entry back to the namespace. */
3787                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3788                                                    cname->ln_name);
3789                 if (unlikely(rc == -ESTALE))
3790                         /* It may happen when the remote object has been
3791                          * removed, but the local MDT is not aware of that. */
3792                         goto lost_parent;
3793
3794                 if (unlikely(rc == -EEXIST))
3795                         /* Unfortunately, someone reused the name under the
3796                          * parent by race. So we have to remove the linkEA
3797                          * entry from current child object. It means that the
3798                          * LFSCK cannot recover the system totally back to
3799                          * its original status, but it is necessary to make
3800                          * the current system to be consistent.
3801                          *
3802                          * It also may be because of the LFSCK found some
3803                          * internal status of create operation. Under such
3804                          * case, nothing to be done. */
3805                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3806                                         parent, child, &ldata, cname, pfid);
3807                 else
3808                         linkea_next_entry(&ldata);
3809
3810                 lfsck_object_put(env, parent);
3811                 if (rc < 0)
3812                         GOTO(out, rc);
3813
3814                 if (rc > 0)
3815                         repaired = true;
3816         }
3817
3818         GOTO(out, rc = 0);
3819
3820 out:
3821         if (rc < 0 && rc != -ENODATA)
3822                 return rc;
3823
3824         if (rc == 0 && ldata.ld_leh != NULL)
3825                 count = ldata.ld_leh->leh_reccount;
3826
3827         if (count == 0) {
3828                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3829                  * MDT has ever tried to verify some remote MDT-object that
3830                  * resides on this MDT, but this MDT failed to respond such
3831                  * request. So means there may be some remote name entry on
3832                  * other MDT that references this object with another name,
3833                  * so we cannot know whether this linkEA is valid or not.
3834                  * So keep it there and maybe resolved when next LFSCK run. */
3835                 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3836                     (ldata.ld_leh == NULL ||
3837                      !ldata.ld_leh->leh_overflow_time)) {
3838                         /* If the child becomes orphan, then insert it into
3839                          * the global .lustre/lost+found/MDTxxxx directory. */
3840                         rc = lfsck_namespace_insert_orphan(env, com, child,
3841                                                            "", "O", &count);
3842                         if (rc < 0)
3843                                 return rc;
3844
3845                         if (rc > 0) {
3846                                 ns->ln_mul_ref_repaired++;
3847                                 repaired = true;
3848                         }
3849                 }
3850         } else {
3851                 rc = dt_attr_get(env, child, la);
3852                 if (rc != 0)
3853                         return rc;
3854
3855                 if (la->la_nlink != 0 && la->la_nlink != count) {
3856                         if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
3857                                      !S_ISLNK(lfsck_object_type(child)))) {
3858                                 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
3859                                        "the object "DFID"'s nlink count %d "
3860                                        "does not match linkEA count %d, "
3861                                        "type %o, skip it.\n",
3862                                        lfsck_lfsck2name(lfsck),
3863                                        PFID(lfsck_dto2fid(child)),
3864                                        la->la_nlink, count,
3865                                        lfsck_object_type(child));
3866                         } else if (la->la_nlink < count &&
3867                                    likely(!ldata.ld_leh->leh_overflow_time)) {
3868                                 rc = lfsck_namespace_repair_nlink(env, com,
3869                                                                   child, la);
3870                                 if (rc > 0) {
3871                                         ns->ln_objs_nlink_repaired++;
3872                                         rc = 0;
3873                                 }
3874                         }
3875                 }
3876         }
3877
3878         if (repaired) {
3879                 if (la->la_nlink > 1)
3880                         ns->ln_mul_linked_repaired++;
3881
3882                 if (rc == 0)
3883                         rc = 1;
3884         }
3885
3886         if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3887                 rc = lfsck_namespace_check_agent_entry(env, com, child);
3888
3889         return rc;
3890 }
3891
3892 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3893                                             struct lfsck_namespace *ns,
3894                                             __u64 checked_phase1,
3895                                             __u64 checked_phase2,
3896                                             time64_t time_phase1,
3897                                             time64_t time_phase2, bool dryrun)
3898 {
3899         const char *postfix = dryrun ? "inconsistent" : "repaired";
3900
3901         seq_printf(m, "checked_phase1: %llu\n"
3902                    "checked_phase2: %llu\n"
3903                    "%s_phase1: %llu\n"
3904                    "%s_phase2: %llu\n"
3905                    "failed_phase1: %llu\n"
3906                    "failed_phase2: %llu\n"
3907                    "directories: %llu\n"
3908                    "dirent_%s: %llu\n"
3909                    "linkea_%s: %llu\n"
3910                    "nlinks_%s: %llu\n"
3911                    "multiple_linked_checked: %llu\n"
3912                    "multiple_linked_%s: %llu\n"
3913                    "unknown_inconsistency: %llu\n"
3914                    "unmatched_pairs_%s: %llu\n"
3915                    "dangling_%s: %llu\n"
3916                    "multiple_referenced_%s: %llu\n"
3917                    "bad_file_type_%s: %llu\n"
3918                    "lost_dirent_%s: %llu\n"
3919                    "local_lost_found_scanned: %llu\n"
3920                    "local_lost_found_moved: %llu\n"
3921                    "local_lost_found_skipped: %llu\n"
3922                    "local_lost_found_failed: %llu\n"
3923                    "striped_dirs_scanned: %llu\n"
3924                    "striped_dirs_%s: %llu\n"
3925                    "striped_dirs_failed: %llu\n"
3926                    "striped_dirs_disabled: %llu\n"
3927                    "striped_dirs_skipped: %llu\n"
3928                    "striped_shards_scanned: %llu\n"
3929                    "striped_shards_%s: %llu\n"
3930                    "striped_shards_failed: %llu\n"
3931                    "striped_shards_skipped: %llu\n"
3932                    "name_hash_%s: %llu\n"
3933                    "linkea_overflow_%s: %llu\n"
3934                    "agent_entries_%s: %llu\n"
3935                    "success_count: %u\n"
3936                    "run_time_phase1: %lld seconds\n"
3937                    "run_time_phase2: %lld seconds\n",
3938                    checked_phase1,
3939                    checked_phase2,
3940                    dryrun ? "inconsistent" : "updated",
3941                    ns->ln_items_repaired,
3942                    dryrun ? "inconsistent" : "updated",
3943                    ns->ln_objs_repaired_phase2,
3944                    ns->ln_items_failed,
3945                    ns->ln_objs_failed_phase2,
3946                    ns->ln_dirs_checked,
3947                    postfix, ns->ln_dirent_repaired,
3948                    postfix, ns->ln_linkea_repaired,
3949                    postfix, ns->ln_objs_nlink_repaired,
3950                    ns->ln_mul_linked_checked,
3951                    postfix, ns->ln_mul_linked_repaired,
3952                    ns->ln_unknown_inconsistency,
3953                    postfix, ns->ln_unmatched_pairs_repaired,
3954                    postfix, ns->ln_dangling_repaired,
3955                    postfix, ns->ln_mul_ref_repaired,
3956                    postfix, ns->ln_bad_type_repaired,
3957                    postfix, ns->ln_lost_dirent_repaired,
3958                    ns->ln_local_lpf_scanned,
3959                    ns->ln_local_lpf_moved,
3960                    ns->ln_local_lpf_skipped,
3961                    ns->ln_local_lpf_failed,
3962                    ns->ln_striped_dirs_scanned,
3963                    postfix, ns->ln_striped_dirs_repaired,
3964                    ns->ln_striped_dirs_failed,
3965                    ns->ln_striped_dirs_disabled,
3966                    ns->ln_striped_dirs_skipped,
3967                    ns->ln_striped_shards_scanned,
3968                    postfix, ns->ln_striped_shards_repaired,
3969                    ns->ln_striped_shards_failed,
3970                    ns->ln_striped_shards_skipped,
3971                    postfix, ns->ln_name_hash_repaired,
3972                    dryrun ? "inconsistent" : "cleared",
3973                    ns->ln_linkea_overflow_cleared,
3974                    postfix, ns->ln_agent_entries_repaired,
3975                    ns->ln_success_count,
3976                    time_phase1,
3977                    time_phase2);
3978 }
3979
3980 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3981                                         struct lfsck_component *com)
3982 {
3983         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3984         struct lfsck_namespace          *ns     = com->lc_file_ram;
3985
3986         while (!list_empty(&lfsck->li_list_lmv)) {
3987                 struct lfsck_lmv_unit   *llu;
3988                 struct lfsck_lmv        *llmv;
3989
3990                 llu = list_entry(lfsck->li_list_lmv.next,
3991                                  struct lfsck_lmv_unit, llu_link);
3992                 llmv = &llu->llu_lmv;
3993
3994                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3995                          "still in using: %u\n",
3996                          atomic_read(&llmv->ll_ref));
3997
3998                 ns->ln_striped_dirs_skipped++;
3999                 lfsck_lmv_put(env, llmv);
4000         }
4001 }
4002
4003 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4004                                                  struct lfsck_component *com,
4005                                                  struct dt_object *obj)
4006 {
4007         struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4008         int             rc;
4009
4010         rc = dt_attr_get(env, obj, la);
4011         if (rc != 0)
4012                 return rc;
4013
4014         /* zero-linkEA object may be orphan, but it also maybe because
4015          * of upgrading. Currently, we cannot record it for double scan.
4016          * Because it may cause the LFSCK trace file to be too large. */
4017
4018         /* "la_ctime" == 1 means that it has ever been removed from
4019          * backend /lost+found directory but not been added back to
4020          * the normal namespace yet. */
4021
4022         if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4023             unlikely(la->la_ctime == 1))
4024                 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4025                                                   LNTF_CHECK_LINKEA, true);
4026
4027         return rc;
4028 }
4029
4030 /* namespace APIs */
4031
4032 static int lfsck_namespace_reset(const struct lu_env *env,
4033                                  struct lfsck_component *com, bool init)
4034 {
4035         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4036         struct lfsck_namespace          *ns     = com->lc_file_ram;
4037         struct lfsck_assistant_data     *lad    = com->lc_data;
4038         struct dt_object                *root;
4039         int                              rc;
4040         ENTRY;
4041
4042         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4043         if (IS_ERR(root))
4044                 GOTO(log, rc = PTR_ERR(root));
4045
4046         if (unlikely(!dt_try_as_dir(env, root)))
4047                 GOTO(put, rc = -ENOTDIR);
4048
4049         down_write(&com->lc_sem);
4050         if (init) {
4051                 memset(ns, 0, sizeof(*ns));
4052         } else {
4053                 __u32 count = ns->ln_success_count;
4054                 time64_t last_time = ns->ln_time_last_complete;
4055
4056                 memset(ns, 0, sizeof(*ns));
4057                 ns->ln_success_count = count;
4058                 ns->ln_time_last_complete = last_time;
4059         }
4060         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4061         ns->ln_status = LS_INIT;
4062         ns->ln_time_latest_reset = ktime_get_real_seconds();
4063
4064         rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4065                                        &dt_lfsck_namespace_features,
4066                                        LFSCK_NAMESPACE, true);
4067         if (rc)
4068                 GOTO(out, rc);
4069
4070         rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4071                                         LFSCK_NAMESPACE, true);
4072         if (rc != 0)
4073                 GOTO(out, rc);
4074
4075         lad->lad_incomplete = 0;
4076         CFS_RESET_BITMAP(lad->lad_bitmap);
4077
4078         rc = lfsck_namespace_store(env, com);
4079
4080         GOTO(out, rc);
4081
4082 out:
4083         up_write(&com->lc_sem);
4084
4085 put:
4086         lfsck_object_put(env, root);
4087 log:
4088         CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4089                lfsck_lfsck2name(lfsck), rc);
4090         return rc;
4091 }
4092
4093 static void
4094 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4095                      bool new_checked)
4096 {
4097         struct lfsck_namespace *ns = com->lc_file_ram;
4098
4099         down_write(&com->lc_sem);
4100         if (new_checked)
4101                 com->lc_new_checked++;
4102         lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4103         up_write(&com->lc_sem);
4104 }
4105
4106 static void lfsck_namespace_close_dir(const struct lu_env *env,
4107                                       struct lfsck_component *com)
4108 {
4109         struct lfsck_namespace          *ns     = com->lc_file_ram;
4110         struct lfsck_assistant_data     *lad    = com->lc_data;
4111         struct lfsck_assistant_object   *lso    = NULL;
4112         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4113         struct lfsck_lmv                *llmv   = lfsck->li_lmv;
4114         struct lfsck_namespace_req      *lnr;
4115         __u32                            size   =
4116                                 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4117         bool                             wakeup = false;
4118         ENTRY;
4119
4120         if (llmv == NULL)
4121                 RETURN_EXIT;
4122
4123         OBD_ALLOC(lnr, size);
4124         if (lnr == NULL) {
4125                 ns->ln_striped_dirs_skipped++;
4126
4127                 RETURN_EXIT;
4128         }
4129
4130         lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4131                         NULL, lfsck->li_pos_current.lp_oit_cookie, true);
4132         if (IS_ERR(lso)) {
4133                 OBD_FREE(lnr, size);
4134                 ns->ln_striped_dirs_skipped++;
4135
4136                 RETURN_EXIT;
4137         }
4138
4139         /* Generate a dummy request to indicate that all shards' name entry
4140          * in this striped directory has been scanned for the first time. */
4141         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4142         lnr->lnr_lar.lar_parent = lso;
4143         lnr->lnr_lmv = lfsck_lmv_get(llmv);
4144         lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4145         lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4146         lnr->lnr_size = size;
4147
4148         spin_lock(&lad->lad_lock);
4149         if (lad->lad_assistant_status < 0 ||
4150             unlikely(!thread_is_running(&lfsck->li_thread) ||
4151                      !thread_is_running(&lad->lad_thread))) {
4152                 spin_unlock(&lad->lad_lock);
4153                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4154                 ns->ln_striped_dirs_skipped++;
4155
4156                 RETURN_EXIT;
4157         }
4158
4159         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4160         if (lad->lad_prefetched == 0)
4161                 wakeup = true;
4162
4163         lad->lad_prefetched++;
4164         spin_unlock(&lad->lad_lock);
4165         if (wakeup)
4166                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4167
4168         EXIT;
4169 }
4170
4171 static int lfsck_namespace_open_dir(const struct lu_env *env,
4172                                     struct lfsck_component *com)
4173 {
4174         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4175         struct lfsck_namespace  *ns     = com->lc_file_ram;
4176         struct lfsck_lmv        *llmv   = lfsck->li_lmv;
4177         int                      rc     = 0;
4178         ENTRY;
4179
4180         if (llmv == NULL)
4181                 RETURN(0);
4182
4183         if (llmv->ll_lmv_master) {
4184                 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4185
4186                 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4187                         lmv->lmv_master_mdt_index =
4188                                 lfsck_dev_idx(lfsck);
4189                         ns->ln_flags |= LF_INCONSISTENT;
4190                         llmv->ll_lmv_updated = 1;
4191                 }
4192         } else {
4193                 rc = lfsck_namespace_verify_stripe_slave(env, com,
4194                                         lfsck->li_obj_dir, llmv);
4195         }
4196
4197         RETURN(rc > 0 ? 0 : rc);
4198 }
4199
4200 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4201                                       struct lfsck_component *com, bool init)
4202 {
4203         struct lfsck_instance   *lfsck = com->lc_lfsck;
4204         struct lfsck_namespace  *ns    = com->lc_file_ram;
4205         int                      rc;
4206
4207         if (!init) {
4208                 rc = lfsck_checkpoint_generic(env, com);
4209                 if (rc != 0)
4210                         goto log;
4211         }
4212
4213         down_write(&com->lc_sem);
4214         if (init) {
4215                 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4216         } else {
4217                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4218                 ns->ln_run_time_phase1 += ktime_get_seconds() -
4219                                           lfsck->li_time_last_checkpoint;
4220                 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4221                 ns->ln_items_checked += com->lc_new_checked;
4222                 com->lc_new_checked = 0;
4223         }
4224
4225         rc = lfsck_namespace_store(env, com);
4226         up_write(&com->lc_sem);
4227
4228 log:
4229         CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4230                ", "DFID", %#llx], status = %d: rc = %d\n",
4231                lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4232                PFID(&lfsck->li_pos_current.lp_dir_parent),
4233                lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4234
4235         return rc > 0 ? 0 : rc;
4236 }
4237
4238 static int lfsck_namespace_prep(const struct lu_env *env,
4239                                 struct lfsck_component *com,
4240                                 struct lfsck_start_param *lsp)
4241 {
4242         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4243         struct lfsck_namespace  *ns     = com->lc_file_ram;
4244         struct lfsck_position   *pos    = &com->lc_pos_start;
4245         int                      rc;
4246
4247         rc = lfsck_namespace_load_bitmap(env, com);
4248         if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4249                 rc = lfsck_namespace_reset(env, com, false);
4250                 if (rc == 0)
4251                         rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4252
4253                 if (rc != 0) {
4254                         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4255                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4256
4257                         return rc;
4258                 }
4259         }
4260
4261         down_write(&com->lc_sem);
4262         ns->ln_time_latest_start = ktime_get_real_seconds();
4263         spin_lock(&lfsck->li_lock);
4264
4265         if (ns->ln_flags & LF_SCANNED_ONCE) {
4266                 if (!lfsck->li_drop_dryrun ||
4267                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4268                         ns->ln_status = LS_SCANNING_PHASE2;
4269                         list_move_tail(&com->lc_link,
4270                                        &lfsck->li_list_double_scan);
4271                         if (!list_empty(&com->lc_link_dir))
4272                                 list_del_init(&com->lc_link_dir);
4273                         lfsck_pos_set_zero(pos);
4274                 } else {
4275                         ns->ln_status = LS_SCANNING_PHASE1;
4276                         ns->ln_run_time_phase1 = 0;
4277                         ns->ln_run_time_phase2 = 0;
4278                         ns->ln_items_checked = 0;
4279                         ns->ln_items_repaired = 0;
4280                         ns->ln_items_failed = 0;
4281                         ns->ln_dirs_checked = 0;
4282                         ns->ln_objs_checked_phase2 = 0;
4283                         ns->ln_objs_repaired_phase2 = 0;
4284                         ns->ln_objs_failed_phase2 = 0;
4285                         ns->ln_objs_nlink_repaired = 0;
4286                         ns->ln_dirent_repaired = 0;
4287                         ns->ln_linkea_repaired = 0;
4288                         ns->ln_mul_linked_checked = 0;
4289                         ns->ln_mul_linked_repaired = 0;
4290                         ns->ln_unknown_inconsistency = 0;
4291                         ns->ln_unmatched_pairs_repaired = 0;
4292                         ns->ln_dangling_repaired = 0;
4293                         ns->ln_mul_ref_repaired = 0;
4294                         ns->ln_bad_type_repaired = 0;
4295                         ns->ln_lost_dirent_repaired = 0;
4296                         ns->ln_striped_dirs_scanned = 0;
4297                         ns->ln_striped_dirs_repaired = 0;
4298                         ns->ln_striped_dirs_failed = 0;
4299                         ns->ln_striped_dirs_disabled = 0;
4300                         ns->ln_striped_dirs_skipped = 0;
4301                         ns->ln_striped_shards_scanned = 0;
4302                         ns->ln_striped_shards_repaired = 0;
4303                         ns->ln_striped_shards_failed = 0;
4304                         ns->ln_striped_shards_skipped = 0;
4305                         ns->ln_name_hash_repaired = 0;
4306                         fid_zero(&ns->ln_fid_latest_scanned_phase2);
4307                         if (list_empty(&com->lc_link_dir))
4308                                 list_add_tail(&com->lc_link_dir,
4309                                               &lfsck->li_list_dir);
4310                         *pos = ns->ln_pos_first_inconsistent;
4311                 }
4312         } else {
4313                 ns->ln_status = LS_SCANNING_PHASE1;
4314                 if (list_empty(&com->lc_link_dir))
4315                         list_add_tail(&com->lc_link_dir,
4316                                       &lfsck->li_list_dir);
4317                 if (!lfsck->li_drop_dryrun ||
4318                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4319                         *pos = ns->ln_pos_last_checkpoint;
4320                         pos->lp_oit_cookie++;
4321                 } else {
4322                         *pos = ns->ln_pos_first_inconsistent;
4323                 }
4324         }
4325
4326         spin_unlock(&lfsck->li_lock);
4327         up_write(&com->lc_sem);
4328
4329         rc = lfsck_start_assistant(env, com, lsp);
4330
4331         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4332                DFID", %#llx]: rc = %d\n",
4333                lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4334                PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4335
4336         return rc;
4337 }
4338
4339 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4340                                     struct lfsck_component *com,
4341                                     struct dt_object *obj)
4342 {
4343         struct lfsck_thread_info *info  = lfsck_env_info(env);
4344         struct lfsck_namespace   *ns    = com->lc_file_ram;
4345         struct lfsck_instance    *lfsck = com->lc_lfsck;
4346         const struct lu_fid      *fid   = lfsck_dto2fid(obj);
4347         struct lu_fid            *pfid  = &info->lti_fid2;
4348         struct lu_name           *cname = &info->lti_name;
4349         struct lu_seq_range      *range = &info->lti_range;
4350         struct seq_server_site   *ss    = lfsck_dev_site(lfsck);
4351         struct linkea_data        ldata = { NULL };
4352         __u32                     idx   = lfsck_dev_idx(lfsck);
4353         int                       rc;
4354         bool remote = false;
4355         ENTRY;
4356
4357         rc = lfsck_links_read(env, obj, &ldata);
4358         if (rc == -ENOENT)
4359                 GOTO(out, rc = 0);
4360
4361         /* -EINVAL means crashed linkEA, should be verified. */
4362         if (rc == -EINVAL) {
4363                 rc = lfsck_namespace_trace_update(env, com, fid,
4364                                                   LNTF_CHECK_LINKEA, true);
4365                 if (rc == 0) {
4366                         struct lustre_handle lh = { 0 };
4367
4368                         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4369                                               MDS_INODELOCK_UPDATE |
4370                                               MDS_INODELOCK_XATTR, LCK_EX);
4371                         if (rc == 0) {
4372                                 rc = lfsck_namespace_links_remove(env, com,
4373                                                                   obj);
4374                                 lfsck_ibits_unlock(&lh, LCK_EX);
4375                         }
4376                 }
4377
4378                 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4379         }
4380
4381         if (rc && rc != -ENODATA)
4382                 GOTO(out, rc);
4383
4384         if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4385                 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4386
4387                 GOTO(out, rc);
4388         }
4389
4390         linkea_first_entry(&ldata);
4391         while (ldata.ld_lee != NULL) {
4392                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4393                                     cname, pfid);
4394                 if (!fid_is_sane(pfid)) {
4395                         rc = lfsck_namespace_trace_update(env, com, fid,
4396                                                   LNTF_CHECK_PARENT, true);
4397                 } else {
4398                         fld_range_set_mdt(range);
4399                         rc = fld_server_lookup(env, ss->ss_server_fld,
4400                                                fid_seq(pfid), range);
4401                         if ((rc == -ENOENT) ||
4402                             (!rc && range->lsr_index != idx)) {
4403                                 remote = true;
4404                                 break;
4405                         }
4406                 }
4407                 if (rc)
4408                         GOTO(out, rc);
4409
4410                 linkea_next_entry(&ldata);
4411         }
4412
4413         if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4414             (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4415                 rc = lfsck_namespace_trace_update(env, com, fid,
4416                                                   LNTF_CHECK_AGENT_ENTRY, true);
4417                 if (rc)
4418                         GOTO(out, rc);
4419         }
4420
4421         /* Record multiple-linked object. */
4422         if (ldata.ld_leh->leh_reccount > 1) {
4423                 rc = lfsck_namespace_trace_update(env, com, fid,
4424                                                   LNTF_CHECK_LINKEA, true);
4425
4426                 GOTO(out, rc);
4427         }
4428
4429         if (remote)
4430                 rc = lfsck_namespace_trace_update(env, com, fid,
4431                                                   LNTF_CHECK_LINKEA, true);
4432         else
4433                 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4434
4435         GOTO(out, rc);
4436
4437 out:
4438         down_write(&com->lc_sem);
4439         if (S_ISDIR(lfsck_object_type(obj)))
4440                 ns->ln_dirs_checked++;
4441         if (rc != 0)
4442                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4443         up_write(&com->lc_sem);
4444
4445         return rc;
4446 }
4447
4448 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4449                                     struct lfsck_component *com,
4450                                     struct lfsck_assistant_object *lso,
4451                                     struct lu_dirent *ent, __u16 type)
4452 {
4453         struct lfsck_assistant_data     *lad     = com->lc_data;
4454         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4455         struct lfsck_namespace_req      *lnr;
4456         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4457         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4458         struct ptlrpc_thread            *athread = &lad->lad_thread;
4459         struct l_wait_info               lwi     = { 0 };
4460         bool                             wakeup  = false;
4461
4462         l_wait_event(mthread->t_ctl_waitq,
4463                      lad->lad_prefetched < bk->lb_async_windows ||
4464                      !thread_is_running(mthread) ||
4465                      !thread_is_running(athread),
4466                      &lwi);
4467
4468         if (unlikely(!thread_is_running(mthread) ||
4469                      !thread_is_running(athread)))
4470                 return 0;
4471
4472         if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4473                 return 0;
4474
4475         lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4476         if (IS_ERR(lnr)) {
4477                 struct lfsck_namespace *ns = com->lc_file_ram;
4478
4479                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4480                 return PTR_ERR(lnr);
4481         }
4482
4483         spin_lock(&lad->lad_lock);
4484         if (lad->lad_assistant_status < 0 ||
4485             unlikely(!thread_is_running(mthread) ||
4486                      !thread_is_running(athread))) {
4487                 spin_unlock(&lad->lad_lock);
4488                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4489                 return lad->lad_assistant_status;
4490         }
4491
4492         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4493         if (lad->lad_prefetched == 0)
4494                 wakeup = true;
4495
4496         lad->lad_prefetched++;
4497         spin_unlock(&lad->lad_lock);
4498         if (wakeup)
4499                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4500
4501         down_write(&com->lc_sem);
4502         com->lc_new_checked++;
4503         up_write(&com->lc_sem);
4504
4505         return 0;
4506 }
4507
4508 static int lfsck_namespace_post(const struct lu_env *env,
4509                                 struct lfsck_component *com,
4510                                 int result, bool init)
4511 {
4512         struct lfsck_instance   *lfsck = com->lc_lfsck;
4513         struct lfsck_namespace  *ns    = com->lc_file_ram;
4514         int                      rc;
4515         ENTRY;
4516
4517         lfsck_post_generic(env, com, &result);
4518
4519         down_write(&com->lc_sem);
4520         lfsck_namespace_release_lmv(env, com);
4521
4522         spin_lock(&lfsck->li_lock);
4523         if (!init)
4524                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4525         if (result > 0) {
4526                 ns->ln_status = LS_SCANNING_PHASE2;
4527                 ns->ln_flags |= LF_SCANNED_ONCE;
4528                 ns->ln_flags &= ~LF_UPGRADE;
4529                 list_del_init(&com->lc_link_dir);
4530                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4531         } else if (result == 0) {
4532                 if (lfsck->li_status != 0)
4533                         ns->ln_status = lfsck->li_status;
4534                 else
4535                         ns->ln_status = LS_STOPPED;
4536                 if (ns->ln_status != LS_PAUSED) {
4537                         list_del_init(&com->lc_link_dir);
4538                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4539                 }
4540         } else {
4541                 ns->ln_status = LS_FAILED;
4542                 list_del_init(&com->lc_link_dir);
4543                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4544         }
4545         spin_unlock(&lfsck->li_lock);
4546
4547         if (!init) {
4548                 ns->ln_run_time_phase1 += ktime_get_seconds() -
4549                                           lfsck->li_time_last_checkpoint;
4550                 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4551                 ns->ln_items_checked += com->lc_new_checked;
4552                 com->lc_new_checked = 0;
4553         }
4554
4555         rc = lfsck_namespace_store(env, com);
4556         up_write(&com->lc_sem);
4557
4558         CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4559                lfsck_lfsck2name(lfsck), rc);
4560
4561         RETURN(rc);
4562 }
4563
4564 static void
4565 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4566                      struct seq_file *m)
4567 {
4568         struct lfsck_instance   *lfsck = com->lc_lfsck;
4569         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4570         struct lfsck_namespace  *ns    = com->lc_file_ram;
4571
4572         down_read(&com->lc_sem);
4573         seq_printf(m, "name: lfsck_namespace\n"
4574                    "magic: %#x\n"
4575                    "version: %d\n"
4576                    "status: %s\n",
4577                    ns->ln_magic,
4578                    bk->lb_version,
4579                    lfsck_status2name(ns->ln_status));
4580
4581         lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4582
4583         lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4584
4585         lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4586
4587         lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4588
4589         lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4590
4591         lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4592
4593         lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4594                        "last_checkpoint_position");
4595
4596         lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4597                        "first_failure_position");
4598
4599         if (ns->ln_status == LS_SCANNING_PHASE1) {
4600                 struct lfsck_position pos;
4601                 time64_t duration = ktime_get_seconds() -
4602                                     lfsck->li_time_last_checkpoint;
4603                 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4604                 u64 speed = checked;
4605                 u64 new_checked = com->lc_new_checked;
4606                 time64_t rtime = ns->ln_run_time_phase1 + duration;
4607
4608                 if (duration != 0)
4609                         new_checked = div64_s64(new_checked, duration);
4610
4611                 if (rtime != 0)
4612                         speed = div64_s64(speed, rtime);
4613
4614                 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4615                                                 bk->lb_param & LPF_DRYRUN);
4616                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4617                            "average_speed_phase2: N/A\n"
4618                            "average_speed_total: %llu items/sec\n"
4619                            "real_time_speed_phase1: %llu items/sec\n"
4620                            "real_time_speed_phase2: N/A\n",
4621                            speed,
4622                            speed,
4623                            new_checked);
4624
4625                 if (likely(lfsck->li_di_oit)) {
4626                         const struct dt_it_ops *iops =
4627                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
4628
4629                         /* The low layer otable-based iteration position may NOT
4630                          * exactly match the namespace-based directory traversal
4631                          * cookie. Generally, it is not a serious issue. But the
4632                          * caller should NOT make assumption on that. */
4633                         pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4634                         if (!lfsck->li_current_oit_processed)
4635                                 pos.lp_oit_cookie--;
4636
4637                         spin_lock(&lfsck->li_lock);
4638                         if (lfsck->li_di_dir) {
4639                                 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4640                                 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4641                                         fid_zero(&pos.lp_dir_parent);
4642                                         pos.lp_dir_cookie = 0;
4643                                 } else {
4644                                         pos.lp_dir_parent =
4645                                         *lfsck_dto2fid(lfsck->li_obj_dir);
4646                                 }
4647                         } else {
4648                                 fid_zero(&pos.lp_dir_parent);
4649                                 pos.lp_dir_cookie = 0;
4650                         }
4651                         spin_unlock(&lfsck->li_lock);
4652                 } else {
4653                         pos = ns->ln_pos_last_checkpoint;
4654                 }
4655
4656                 lfsck_pos_dump(m, &pos, "current_position");
4657         } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4658                 time64_t duration = ktime_get_seconds() -
4659                                     com->lc_time_last_checkpoint;
4660                 __u64 checked = ns->ln_objs_checked_phase2 +
4661                                 com->lc_new_checked;
4662                 __u64 speed1 = ns->ln_items_checked;
4663                 __u64 speed2 = checked;
4664                 __u64 speed0 = speed1 + speed2;
4665                 __u64 new_checked = com->lc_new_checked;
4666                 time64_t rtime = ns->ln_run_time_phase2 + duration;
4667                 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4668
4669                 if (duration != 0)
4670                         new_checked = div64_s64(new_checked, duration);
4671
4672                 if (ns->ln_run_time_phase1 != 0)
4673                         speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4674                 else if (ns->ln_items_checked != 0)
4675                         time0++;
4676
4677                 if (rtime != 0)
4678                         speed2 = div64_s64(speed2, rtime);
4679                 else if (checked != 0)
4680                         time0++;
4681
4682                 if (time0 != 0)
4683                         speed0 = div64_s64(speed0, time0);
4684
4685                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4686                                                 checked,
4687                                                 ns->ln_run_time_phase1, rtime,
4688                                                 bk->lb_param & LPF_DRYRUN);
4689                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4690                            "average_speed_phase2: %llu objs/sec\n"
4691                            "average_speed_total: %llu items/sec\n"
4692                            "real_time_speed_phase1: N/A\n"
4693                            "real_time_speed_phase2: %llu objs/sec\n"
4694                            "current_position: "DFID"\n",
4695                            speed1,
4696                            speed2,
4697                            speed0,
4698                            new_checked,
4699                            PFID(&ns->ln_fid_latest_scanned_phase2));
4700         } else {
4701                 __u64 speed1 = ns->ln_items_checked;
4702                 __u64 speed2 = ns->ln_objs_checked_phase2;
4703                 __u64 speed0 = speed1 + speed2;
4704                 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4705
4706                 if (ns->ln_run_time_phase1 != 0)
4707                         speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4708                 else if (ns->ln_items_checked != 0)
4709                         time0++;
4710
4711                 if (ns->ln_run_time_phase2 != 0)
4712                         speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4713                 else if (ns->ln_objs_checked_phase2 != 0)
4714                         time0++;
4715
4716                 if (time0 != 0)
4717                         speed0 = div64_s64(speed0, time0);
4718
4719                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4720                                                 ns->ln_objs_checked_phase2,
4721                                                 ns->ln_run_time_phase1,
4722                                                 ns->ln_run_time_phase2,
4723                                                 bk->lb_param & LPF_DRYRUN);
4724                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4725                            "average_speed_phase2: %llu objs/sec\n"
4726                            "average_speed_total: %llu items/sec\n"
4727                            "real_time_speed_phase1: N/A\n"
4728                            "real_time_speed_phase2: N/A\n"
4729                            "current_position: N/A\n",
4730                            speed1,
4731                            speed2,
4732                            speed0);
4733         }
4734
4735         up_read(&com->lc_sem);
4736 }
4737
4738 static int lfsck_namespace_double_scan(const struct lu_env *env,
4739                                        struct lfsck_component *com)
4740 {
4741         struct lfsck_namespace          *ns     = com->lc_file_ram;
4742         struct lfsck_assistant_data     *lad    = com->lc_data;
4743         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4744         struct lfsck_tgt_desc           *ltd;
4745         struct lfsck_tgt_desc           *next;
4746         int                              rc;
4747
4748         rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4749         if (thread_is_stopped(&lad->lad_thread)) {
4750                 LASSERT(list_empty(&lad->lad_req_list));
4751                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4752
4753                 spin_lock(&ltds->ltd_lock);
4754                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4755                                          ltd_namespace_phase_list) {
4756                         list_del_init(&ltd->ltd_namespace_phase_list);
4757                 }
4758                 spin_unlock(&ltds->ltd_lock);
4759         }
4760
4761         return rc;
4762 }
4763
4764 static void lfsck_namespace_data_release(const struct lu_env *env,
4765                                          struct lfsck_component *com)
4766 {
4767         struct lfsck_assistant_data     *lad    = com->lc_data;
4768         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4769         struct lfsck_tgt_desc           *ltd;
4770         struct lfsck_tgt_desc           *next;
4771
4772         LASSERT(lad != NULL);
4773         LASSERT(thread_is_init(&lad->lad_thread) ||
4774                 thread_is_stopped(&lad->lad_thread));
4775         LASSERT(list_empty(&lad->lad_req_list));
4776
4777         com->lc_data = NULL;
4778         lfsck_namespace_release_lmv(env, com);
4779
4780         spin_lock(&ltds->ltd_lock);
4781         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4782                                  ltd_namespace_phase_list) {
4783                 list_del_init(&ltd->ltd_namespace_phase_list);
4784         }
4785         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4786                                  ltd_namespace_phase_list) {
4787                 list_del_init(&ltd->ltd_namespace_phase_list);
4788         }
4789         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4790                                  ltd_namespace_list) {
4791                 list_del_init(&ltd->ltd_namespace_list);
4792         }
4793         spin_unlock(&ltds->ltd_lock);
4794
4795         if (likely(lad->lad_bitmap != NULL))
4796                 CFS_FREE_BITMAP(lad->lad_bitmap);
4797
4798         OBD_FREE_PTR(lad);
4799 }
4800
4801 static void lfsck_namespace_quit(const struct lu_env *env,
4802                                  struct lfsck_component *com)
4803 {
4804         struct lfsck_assistant_data     *lad    = com->lc_data;
4805         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4806         struct lfsck_tgt_desc           *ltd;
4807         struct lfsck_tgt_desc           *next;
4808
4809         LASSERT(lad != NULL);
4810
4811         lfsck_quit_generic(env, com);
4812
4813         LASSERT(thread_is_init(&lad->lad_thread) ||
4814                 thread_is_stopped(&lad->lad_thread));
4815         LASSERT(list_empty(&lad->lad_req_list));
4816
4817         lfsck_namespace_release_lmv(env, com);
4818
4819         spin_lock(&ltds->ltd_lock);
4820         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4821                                  ltd_namespace_phase_list) {
4822                 list_del_init(&ltd->ltd_namespace_phase_list);
4823         }
4824         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4825                                  ltd_namespace_phase_list) {
4826                 list_del_init(&ltd->ltd_namespace_phase_list);
4827         }
4828         spin_unlock(&ltds->ltd_lock);
4829 }
4830
4831 static int lfsck_namespace_in_notify(const struct lu_env *env,
4832                                      struct lfsck_component *com,
4833                                      struct lfsck_request *lr)
4834 {
4835         struct lfsck_instance *lfsck = com->lc_lfsck;
4836         struct lfsck_namespace *ns = com->lc_file_ram;
4837         struct lfsck_assistant_data *lad = com->lc_data;
4838         struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4839         struct lfsck_tgt_desc *ltd;
4840         int rc = 0;
4841         bool fail = false;
4842         ENTRY;
4843
4844         switch (lr->lr_event) {
4845         case LE_SET_LMV_MASTER: {
4846                 struct dt_object        *obj;
4847
4848                 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4849                 if (IS_ERR(obj))
4850                         RETURN(PTR_ERR(obj));
4851
4852                 if (likely(dt_object_exists(obj)))
4853                         rc = lfsck_namespace_notify_lmv_master_local(env, com,
4854                                                                      obj);
4855
4856                 lfsck_object_put(env, obj);
4857
4858                 RETURN(rc > 0 ? 0 : rc);
4859         }
4860         case LE_SET_LMV_SLAVE: {
4861                 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4862                         ns->ln_striped_shards_repaired++;
4863
4864                 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4865                                                   LNTF_RECHECK_NAME_HASH, true);
4866
4867                 RETURN(rc > 0 ? 0 : rc);
4868         }
4869         case LE_PHASE1_DONE:
4870         case LE_PHASE2_DONE:
4871         case LE_PEER_EXIT:
4872                 break;
4873         default:
4874                 RETURN(-EINVAL);
4875         }
4876
4877         CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4878                "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4879                lr->lr_index, lr->lr_status, lr->lr_flags2);
4880
4881         spin_lock(&ltds->ltd_lock);
4882         ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
4883         if (ltd == NULL) {
4884                 spin_unlock(&ltds->ltd_lock);
4885
4886                 RETURN(-ENXIO);
4887         }
4888
4889         list_del_init(&ltd->ltd_namespace_phase_list);
4890         switch (lr->lr_event) {
4891         case LE_PHASE1_DONE:
4892                 if (lr->lr_status <= 0) {
4893                         ltd->ltd_namespace_done = 1;
4894                         list_del_init(&ltd->ltd_namespace_list);
4895                         CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4896                                "phase1 for namespace LFSCK: rc = %d.\n",
4897                                lfsck_lfsck2name(lfsck),
4898                                ltd->ltd_index, lr->lr_status);
4899                         ns->ln_flags |= LF_INCOMPLETE;
4900                         fail = true;
4901                         break;
4902                 }
4903
4904                 if (lr->lr_flags2 & LF_INCOMPLETE)
4905                         ns->ln_flags |= LF_INCOMPLETE;
4906
4907                 if (list_empty(&ltd->ltd_namespace_list))
4908                         list_add_tail(&ltd->ltd_namespace_list,
4909                                       &lad->lad_mdt_list);
4910                 list_add_tail(&ltd->ltd_namespace_phase_list,
4911                               &lad->lad_mdt_phase2_list);
4912                 break;
4913         case LE_PHASE2_DONE:
4914                 ltd->ltd_namespace_done = 1;
4915                 list_del_init(&ltd->ltd_namespace_list);
4916                 break;
4917         case LE_PEER_EXIT:
4918                 fail = true;
4919                 ltd->ltd_namespace_done = 1;
4920                 list_del_init(&ltd->ltd_namespace_list);
4921                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4922                         CDEBUG(D_LFSCK,
4923                                "%s: the peer MDT %x exit namespace LFSCK\n",
4924                                lfsck_lfsck2name(lfsck), ltd->ltd_index);
4925                         ns->ln_flags |= LF_INCOMPLETE;
4926                 }
4927                 break;
4928         default:
4929                 break;
4930         }
4931         spin_unlock(&ltds->ltd_lock);
4932
4933         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4934                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4935
4936                 memset(stop, 0, sizeof(*stop));
4937                 stop->ls_status = lr->lr_status;
4938                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4939                 lfsck_stop(env, lfsck->li_bottom, stop);
4940         } else if (lfsck_phase2_next_ready(lad)) {
4941                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4942         }
4943
4944         RETURN(0);
4945 }
4946
4947 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
4948 {
4949         *count += ns->ln_objs_nlink_repaired;
4950         *count += ns->ln_dirent_repaired;
4951         *count += ns->ln_linkea_repaired;
4952         *count += ns->ln_mul_linked_repaired;
4953         *count += ns->ln_unmatched_pairs_repaired;
4954         *count += ns->ln_dangling_repaired;
4955         *count += ns->ln_mul_ref_repaired;
4956         *count += ns->ln_bad_type_repaired;
4957         *count += ns->ln_lost_dirent_repaired;
4958         *count += ns->ln_striped_dirs_disabled;
4959         *count += ns->ln_striped_dirs_repaired;
4960         *count += ns->ln_striped_shards_repaired;
4961         *count += ns->ln_name_hash_repaired;
4962         *count += ns->ln_local_lpf_moved;
4963 }
4964
4965 static int lfsck_namespace_query_all(const struct lu_env *env,
4966                                      struct lfsck_component *com,
4967                                      __u32 *mdts_count, __u64 *repaired)
4968 {
4969         struct lfsck_namespace *ns = com->lc_file_ram;
4970         struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4971         struct lfsck_tgt_desc *ltd;
4972         int idx;
4973         int rc;
4974         ENTRY;
4975
4976         rc = lfsck_query_all(env, com);
4977         if (rc != 0)
4978                 RETURN(rc);
4979
4980         down_read(&ltds->ltd_rw_sem);
4981         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
4982                 ltd = lfsck_ltd2tgt(ltds, idx);
4983                 LASSERT(ltd != NULL);
4984
4985                 mdts_count[ltd->ltd_namespace_status]++;
4986                 *repaired += ltd->ltd_namespace_repaired;
4987         }
4988         up_read(&ltds->ltd_rw_sem);
4989
4990         down_read(&com->lc_sem);
4991         mdts_count[ns->ln_status]++;
4992         lfsck_namespace_repaired(ns, repaired);
4993         up_read(&com->lc_sem);
4994
4995         RETURN(0);
4996 }
4997
4998 static int lfsck_namespace_query(const struct lu_env *env,
4999                                  struct lfsck_component *com,
5000                                  struct lfsck_request *req,
5001                                  struct lfsck_reply *rep,
5002                                  struct lfsck_query *que, int idx)
5003 {
5004         struct lfsck_namespace *ns = com->lc_file_ram;
5005         int rc = 0;
5006
5007         if (que != NULL) {
5008                 LASSERT(com->lc_lfsck->li_master);
5009
5010                 rc = lfsck_namespace_query_all(env, com,
5011                                                que->lu_mdts_count[idx],
5012                                                &que->lu_repaired[idx]);
5013         } else {
5014                 down_read(&com->lc_sem);
5015                 rep->lr_status = ns->ln_status;
5016                 if (req->lr_flags & LEF_QUERY_ALL)
5017                         lfsck_namespace_repaired(ns, &rep->lr_repaired);
5018                 up_read(&com->lc_sem);
5019         }
5020
5021         return rc;
5022 }
5023
5024 static struct lfsck_operations lfsck_namespace_ops = {
5025         .lfsck_reset            = lfsck_namespace_reset,
5026         .lfsck_fail             = lfsck_namespace_fail,
5027         .lfsck_close_dir        = lfsck_namespace_close_dir,
5028         .lfsck_open_dir         = lfsck_namespace_open_dir,
5029         .lfsck_checkpoint       = lfsck_namespace_checkpoint,
5030         .lfsck_prep             = lfsck_namespace_prep,
5031         .lfsck_exec_oit         = lfsck_namespace_exec_oit,
5032         .lfsck_exec_dir         = lfsck_namespace_exec_dir,
5033         .lfsck_post             = lfsck_namespace_post,
5034         .lfsck_dump             = lfsck_namespace_dump,
5035         .lfsck_double_scan      = lfsck_namespace_double_scan,
5036         .lfsck_data_release     = lfsck_namespace_data_release,
5037         .lfsck_quit             = lfsck_namespace_quit,
5038         .lfsck_in_notify        = lfsck_namespace_in_notify,
5039         .lfsck_query            = lfsck_namespace_query,
5040 };
5041
5042 /**
5043  * Repair dangling name entry.
5044  *
5045  * For the name entry with dangling reference, we need to repare the
5046  * inconsistency according to the LFSCK sponsor's requirement:
5047  *
5048  * 1) Keep the inconsistency there and report the inconsistency case,
5049  *    then give the chance to the application to find related issues,
5050  *    and the users can make the decision about how to handle it with
5051  *    more human knownledge. (by default)
5052  *
5053  * 2) Re-create the missing MDT-object with the FID information.
5054  *
5055  * \param[in] env       pointer to the thread context
5056  * \param[in] com       pointer to the lfsck component
5057  * \param[in] parent    pointer to the dir object that contains the dangling
5058  *                      name entry
5059  * \param[in] child     pointer to the object corresponding to the dangling
5060  *                      name entry
5061  * \param[in] lnr       pointer to the namespace request that contains the
5062  *                      name's name, parent object, parent's LMV, and ect.
5063  *
5064  * \retval              positive number if no need to repair
5065  * \retval              zero for repaired successfully
5066  * \retval              negative error number on failure
5067  */
5068 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5069                                     struct lfsck_component *com,
5070                                     struct dt_object *parent,
5071                                     struct dt_object *child,
5072                                     struct lfsck_namespace_req *lnr)
5073 {
5074         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5075         struct lu_attr                  *la     = &info->lti_la;
5076         struct dt_allocation_hint       *hint   = &info->lti_hint;
5077         struct dt_object_format         *dof    = &info->lti_dof;
5078         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
5079         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
5080         const struct lu_name            *cname;
5081         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
5082         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
5083         struct linkea_data               ldata  = { NULL };
5084         struct lfsck_lock_handle        *llh    = &info->lti_llh;
5085         struct lu_buf                    linkea_buf;
5086         struct lu_buf                    lmv_buf;
5087         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5088         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5089         struct dt_device                *dev    = lfsck->li_next;
5090         struct thandle                  *th     = NULL;
5091         int                              rc     = 0;
5092         __u16                            type   = lnr->lnr_type;
5093         bool                             create;
5094         ENTRY;
5095
5096         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5097         if (bk->lb_param & LPF_CREATE_MDTOBJ)
5098                 create = true;
5099         else
5100                 create = false;
5101
5102         if (!create || bk->lb_param & LPF_DRYRUN)
5103                 GOTO(log, rc = 0);
5104
5105         /* We may need to create the sub-objects of the @child via LOD,
5106          * so make the modification based on lfsck->li_next. */
5107
5108         parent = lfsck_object_locate(dev, parent);
5109         if (IS_ERR(parent))
5110                 GOTO(log, rc = PTR_ERR(parent));
5111
5112         if (unlikely(!dt_try_as_dir(env, parent)))
5113                 GOTO(log, rc = -ENOTDIR);
5114
5115         child = lfsck_object_locate(dev, child);
5116         if (IS_ERR(child))
5117                 GOTO(log, rc = PTR_ERR(child));
5118
5119         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5120                               cname, pfid);
5121         if (rc != 0)
5122                 GOTO(log, rc);
5123
5124         rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5125                         MDS_INODELOCK_UPDATE, LCK_PR);
5126         if (rc != 0)
5127                 GOTO(log, rc);
5128
5129         rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5130         if (rc != 0)
5131                 GOTO(log, rc);
5132
5133         /* Set the ctime as zero, then others can know it is created for
5134          * repairing dangling name entry by LFSCK. And if the LFSCK made
5135          * wrong decision and the real MDT-object has been found later,
5136          * then the LFSCK has chance to fix the incosistency properly. */
5137         memset(la, 0, sizeof(*la));
5138         la->la_mode = (type & S_IFMT) | 0600;
5139         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
5140                         LA_ATIME | LA_MTIME | LA_CTIME;
5141
5142         child->do_ops->do_ah_init(env, hint, parent, child,
5143                                  la->la_mode & S_IFMT);
5144
5145         memset(dof, 0, sizeof(*dof));
5146         dof->dof_type = dt_mode_to_dft(type);
5147         /* If the target is a regular file, then the LFSCK will only create
5148          * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5149          * OST-objects will be created when write open. */
5150
5151         th = dt_trans_create(env, dev);
5152         if (IS_ERR(th))
5153                 GOTO(log, rc = PTR_ERR(th));
5154
5155         /* 1a. create child. */
5156         rc = dt_declare_create(env, child, la, hint, dof, th);
5157         if (rc != 0)
5158                 GOTO(stop, rc);
5159
5160         if (S_ISDIR(type)) {
5161                 if (unlikely(!dt_try_as_dir(env, child)))
5162                         GOTO(stop, rc = -ENOTDIR);
5163
5164                 /* 2a. increase child nlink */
5165                 rc = dt_declare_ref_add(env, child, th);
5166                 if (rc != 0)
5167                         GOTO(stop, rc);
5168
5169                 /* 3a. insert dot into child dir */
5170                 rec->rec_type = S_IFDIR;
5171                 rec->rec_fid = cfid;
5172                 rc = dt_declare_insert(env, child,
5173                                        (const struct dt_rec *)rec,
5174                                        (const struct dt_key *)dot, th);
5175                 if (rc != 0)
5176                         GOTO(stop, rc);
5177
5178                 /* 4a. insert dotdot into child dir */
5179                 rec->rec_fid = pfid;
5180                 rc = dt_declare_insert(env, child,
5181                                        (const struct dt_rec *)rec,
5182                                        (const struct dt_key *)dotdot, th);
5183                 if (rc != 0)
5184                         GOTO(stop, rc);
5185
5186                 /* 5a. generate slave LMV EA. */
5187                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5188                         int idx;
5189
5190                         idx = lfsck_shard_name_to_index(env,
5191                                         lnr->lnr_name, lnr->lnr_namelen,
5192                                         type, cfid);
5193                         if (unlikely(idx < 0))
5194                                 GOTO(stop, rc = idx);
5195
5196                         *lmv2 = lnr->lnr_lmv->ll_lmv;
5197                         lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5198                         lmv2->lmv_master_mdt_index = idx;
5199
5200                         lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5201                         lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5202                         rc = dt_declare_xattr_set(env, child, &lmv_buf,
5203                                                   XATTR_NAME_LMV, 0, th);
5204                         if (rc != 0)
5205                                 GOTO(stop, rc);
5206                 }
5207         }
5208
5209         /* 6a. insert linkEA for child */
5210         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5211                        ldata.ld_leh->leh_len);
5212         rc = dt_declare_xattr_set(env, child, &linkea_buf,
5213                                   XATTR_NAME_LINK, 0, th);
5214         if (rc != 0)
5215                 GOTO(stop, rc);
5216
5217         rc = dt_trans_start_local(env, dev, th);
5218         if (rc != 0)
5219                 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5220
5221         dt_write_lock(env, child, 0);
5222         /* 1b. create child */
5223         rc = dt_create(env, child, la, hint, dof, th);
5224         if (rc != 0)
5225                 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5226
5227         if (S_ISDIR(type)) {
5228                 /* 2b. increase child nlink */
5229                 rc = dt_ref_add(env, child, th);
5230                 if (rc != 0)
5231                         GOTO(unlock, rc);
5232
5233                 /* 3b. insert dot into child dir */
5234                 rec->rec_type = S_IFDIR;
5235                 rec->rec_fid = cfid;
5236                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5237                                (const struct dt_key *)dot, th, 1);
5238                 if (rc != 0)
5239                         GOTO(unlock, rc);
5240
5241                 /* 4b. insert dotdot into child dir */
5242                 rec->rec_fid = pfid;
5243                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5244                                (const struct dt_key *)dotdot, th, 1);
5245                 if (rc != 0)
5246                         GOTO(unlock, rc);
5247
5248                 /* 5b. generate slave LMV EA. */
5249                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5250                         rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5251                                           0, th);
5252                         if (rc != 0)
5253                                 GOTO(unlock, rc);
5254                 }
5255         }
5256
5257         /* 6b. insert linkEA for child. */
5258         rc = dt_xattr_set(env, child, &linkea_buf,
5259                           XATTR_NAME_LINK, 0, th);
5260
5261         GOTO(unlock, rc);
5262
5263 unlock:
5264         dt_write_unlock(env, child);
5265
5266 stop:
5267         dt_trans_stop(env, dev, th);
5268
5269 log:
5270         lfsck_unlock(llh);
5271         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5272                "reference for: parent "DFID", child "DFID", type %u, "
5273                "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5274                PFID(pfid), PFID(cfid), type, cname->ln_name,
5275                create ? "Create the lost MDT-object as required" :
5276                         "Keep the MDT-object there by default", rc);
5277
5278         if (rc <= 0) {
5279                 struct lfsck_namespace *ns = com->lc_file_ram;
5280
5281                 ns->ln_flags |= LF_INCONSISTENT;
5282         }
5283
5284         return rc;
5285 }
5286
5287 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5288                                                 struct lfsck_component *com,
5289                                                 struct lfsck_assistant_req *lar)
5290 {
5291         struct lfsck_thread_info   *info     = lfsck_env_info(env);
5292         struct lu_attr             *la       = &info->lti_la;
5293         struct lfsck_instance      *lfsck    = com->lc_lfsck;
5294         struct lfsck_bookmark      *bk       = &lfsck->li_bookmark_ram;
5295         struct lfsck_namespace     *ns       = com->lc_file_ram;
5296         struct lfsck_assistant_data *lad     = com->lc_data;
5297         struct linkea_data          ldata    = { NULL };
5298         const struct lu_name       *cname;
5299         struct thandle             *handle   = NULL;
5300         struct lfsck_namespace_req *lnr      =
5301                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5302         struct dt_object           *dir      = NULL;
5303         struct dt_object           *obj      = NULL;
5304         struct lfsck_assistant_object *lso   = lar->lar_parent;
5305         const struct lu_fid        *pfid     = &lso->lso_fid;
5306         struct dt_device           *dev      = NULL;
5307         struct lustre_handle        lh       = { 0 };
5308         bool                        repaired = false;
5309         bool                        dtlocked = false;
5310         bool                        remove;
5311         bool                        newdata;
5312         bool                        log      = false;
5313         bool                        bad_hash = false;
5314         bool                        bad_linkea = false;
5315         int                         idx      = 0;
5316         int                         count    = 0;
5317         int                         rc       = 0;
5318         enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5319         ENTRY;
5320
5321         if (lso->lso_dead)
5322                 RETURN(0);
5323
5324         la->la_nlink = 0;
5325         if (lnr->lnr_attr & LUDA_UPGRADE) {
5326                 ns->ln_flags |= LF_UPGRADE;
5327                 ns->ln_dirent_repaired++;
5328                 repaired = true;
5329         } else if (lnr->lnr_attr & LUDA_REPAIR) {
5330                 ns->ln_flags |= LF_INCONSISTENT;
5331                 ns->ln_dirent_repaired++;
5332                 repaired = true;
5333         }
5334
5335         if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5336                 if (strcmp(lnr->lnr_name, dotdot) != 0)
5337                         LBUG();
5338                 else
5339                         rc = lfsck_namespace_trace_update(env, com, pfid,
5340                                                 LNTF_CHECK_PARENT, true);
5341
5342                 GOTO(out, rc);
5343         }
5344
5345         if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5346                 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5347                        " for the name entry %.*s under "DFID"\n",
5348                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5349                        lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5350
5351                 if (strcmp(lnr->lnr_name, dotdot) != 0)
5352                         /* invalid FID means bad name entry, remove it. */
5353                         type = LNIT_BAD_DIRENT;
5354                 else
5355                         /* If the parent FID is invalid, we cannot remove
5356                          * the ".." entry directly. */
5357                         rc = lfsck_namespace_trace_update(env, com, pfid,
5358                                                 LNTF_CHECK_PARENT, true);
5359
5360                 GOTO(out, rc);
5361         }
5362
5363         if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5364                 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5365
5366                 RETURN(rc);
5367         }
5368
5369         if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5370                 GOTO(out, rc = 0);
5371
5372         if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5373                 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5374
5375                 RETURN(rc);
5376         }
5377
5378         idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5379         if (idx < 0)
5380                 GOTO(out, rc = idx);
5381
5382         if (idx == lfsck_dev_idx(lfsck)) {
5383                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5384                         GOTO(out, rc = 0);
5385
5386                 dev = lfsck->li_bottom;
5387         } else {
5388                 struct lfsck_tgt_desc *ltd;
5389
5390                 /* Usually, some local filesystem consistency verification
5391                  * tools can guarantee the local namespace tree consistenct.
5392                  * So the LFSCK will only verify the remote directory. */
5393                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5394                         rc = lfsck_namespace_trace_update(env, com, pfid,
5395                                                 LNTF_CHECK_PARENT, true);
5396
5397                         GOTO(out, rc);
5398                 }
5399
5400                 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5401                 if (unlikely(ltd == NULL)) {
5402                         CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5403                                "did not join the namespace LFSCK\n",
5404                                lfsck_lfsck2name(lfsck), idx);
5405                         lfsck_lad_set_bitmap(env, com, idx);
5406
5407                         GOTO(out, rc = -ENODEV);
5408                 }
5409
5410                 dev = ltd->ltd_tgt;
5411         }
5412
5413         obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5414         if (IS_ERR(obj))
5415                 GOTO(out, rc = PTR_ERR(obj));
5416
5417         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5418         if (dt_object_exists(obj) == 0) {
5419
5420 dangling:
5421                 if (dir == NULL) {
5422                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5423                         if (IS_ERR(dir)) {
5424                                 rc = PTR_ERR(dir);
5425
5426                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5427                         }
5428                 }
5429
5430                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5431                 if (rc == 0) {
5432                         if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5433                                         lnr->lnr_name, lnr->lnr_namelen)) {
5434                                 type = LNIT_BAD_DIRENT;
5435
5436                                 GOTO(out, rc);
5437                         }
5438
5439                         type = LNIT_DANGLING;
5440                         rc = lfsck_namespace_repair_dangling(env, com, dir,
5441                                                              obj, lnr);
5442                         if (rc == 0)
5443                                 repaired = true;
5444                 }
5445
5446                 GOTO(out, rc);
5447         }
5448
5449         if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5450
5451 again:
5452                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5453                                       MDS_INODELOCK_UPDATE |
5454                                       MDS_INODELOCK_XATTR, LCK_EX);
5455                 if (rc != 0)
5456                         GOTO(out, rc);
5457
5458                 handle = dt_trans_create(env, dev);
5459                 if (IS_ERR(handle))
5460                         GOTO(out, rc = PTR_ERR(handle));
5461
5462                 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5463                 if (rc != 0)
5464                         GOTO(stop, rc);
5465
5466                 rc = dt_trans_start_local(env, dev, handle);
5467                 if (rc != 0)
5468                         GOTO(stop, rc);
5469
5470                 dt_write_lock(env, obj, 0);
5471                 dtlocked = true;
5472         }
5473
5474         rc = lfsck_links_read(env, obj, &ldata);
5475         if (unlikely(rc == -ENOENT)) {
5476                 if (handle != NULL) {
5477                         dt_write_unlock(env, obj);
5478                         dtlocked = false;
5479
5480                         dt_trans_stop(env, dev, handle);
5481                         handle = NULL;
5482
5483                         lfsck_ibits_unlock(&lh, LCK_EX);
5484                 }
5485
5486                 /* It may happen when the remote object has been removed,
5487                  * but the local MDT is not aware of that. */
5488                 goto dangling;
5489         } else if (rc == 0) {
5490                 count = ldata.ld_leh->leh_reccount;
5491                 rc = linkea_links_find(&ldata, cname, pfid);
5492                 if ((rc == 0) &&
5493                     (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5494                         if ((lfsck_object_type(obj) & S_IFMT) !=
5495                             lnr->lnr_type) {
5496                                 ns->ln_flags |= LF_INCONSISTENT;
5497                                 type = LNIT_BAD_TYPE;
5498                         }
5499
5500                         goto stop;
5501                 }
5502
5503                 /* If the name entry hash does not match the slave striped
5504                  * directory, and the name entry does not match also, then
5505                  * it is quite possible that name entry is corrupted. */
5506                 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5507                                         lnr->lnr_name, lnr->lnr_namelen)) {
5508                         ns->ln_flags |= LF_INCONSISTENT;
5509                         type = LNIT_BAD_DIRENT;
5510
5511                         GOTO(stop, rc = 0);
5512                 }
5513
5514                 /* If the file type stored in the name entry does not match
5515                  * the file type claimed by the object, and the object does
5516                  * not recognize the name entry, then it is quite possible
5517                  * that the name entry is corrupted. */
5518                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5519                         ns->ln_flags |= LF_INCONSISTENT;
5520                         type = LNIT_BAD_DIRENT;
5521
5522                         GOTO(stop, rc = 0);
5523                 }
5524
5525                 /* For sub-dir object, we cannot make sure whether the sub-dir
5526                  * back references the parent via ".." name entry correctly or
5527                  * not in the LFSCK first-stage scanning. It may be that the
5528                  * (remote) sub-dir ".." name entry has no parent FID after
5529                  * file-level backup/restore and its linkEA may be wrong.
5530                  * So under such case, we should replace the linkEA according
5531                  * to current name entry. But this needs to be done during the
5532                  * LFSCK second-stage scanning. The LFSCK will record the name
5533                  * entry for further possible using. */
5534                 remove = false;
5535                 newdata = false;
5536                 goto nodata;
5537         } else if (unlikely(rc == -EINVAL)) {
5538                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5539                         type = LNIT_BAD_TYPE;
5540
5541                 count = 1;
5542                 /* The magic crashed, we are not sure whether there are more
5543                  * corrupt data in the linkea, so remove all linkea entries. */
5544                 remove = true;
5545                 newdata = true;
5546                 goto nodata;
5547         } else if (rc == -ENODATA) {
5548                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5549                         type = LNIT_BAD_TYPE;
5550
5551                 count = 1;
5552                 remove = false;
5553                 newdata = true;
5554
5555 nodata:
5556                 if (bk->lb_param & LPF_DRYRUN) {
5557                         if (rc == -ENODATA)
5558                                 ns->ln_flags |= LF_UPGRADE;
5559                         else
5560                                 ns->ln_flags |= LF_INCONSISTENT;
5561                         ns->ln_linkea_repaired++;
5562                         repaired = true;
5563                         log = true;
5564                         goto stop;
5565                 }
5566
5567                 if (!lustre_handle_is_used(&lh)) {
5568                         remove = false;
5569                         newdata = false;
5570                         type = LNIT_NONE;
5571
5572                         goto again;
5573                 }
5574
5575                 LASSERT(handle != NULL);
5576
5577                 if (dir == NULL) {
5578                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5579                         if (IS_ERR(dir)) {
5580                                 rc = PTR_ERR(dir);
5581
5582                                 GOTO(stop, rc == -ENOENT ? 0 : rc);
5583                         }
5584                 }
5585
5586                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5587                 if (rc != 0)
5588                         GOTO(stop, rc);
5589
5590                 bad_linkea = true;
5591                 if (!remove && newdata)
5592                         ns->ln_flags |= LF_UPGRADE;
5593                 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5594                         ns->ln_flags |= LF_INCONSISTENT;
5595
5596                 if (remove) {
5597                         LASSERT(newdata);
5598
5599                         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5600                         if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5601                                 GOTO(stop, rc);
5602                 }
5603
5604                 if (newdata) {
5605                         rc = linkea_data_new(&ldata,
5606                                         &lfsck_env_info(env)->lti_linkea_buf);
5607                         if (rc != 0)
5608                                 GOTO(stop, rc);
5609                 }
5610
5611                 rc = linkea_add_buf(&ldata, cname, pfid);
5612                 if (rc == 0)
5613                         rc = lfsck_links_write(env, obj, &ldata, handle);
5614                 if (rc != 0)
5615                         GOTO(stop, rc);
5616
5617                 count = ldata.ld_leh->leh_reccount;
5618                 if (!S_ISDIR(lfsck_object_type(obj)) ||
5619                     !dt_object_remote(obj)) {
5620                         ns->ln_linkea_repaired++;
5621                         repaired = true;
5622                         log = true;
5623                 }
5624         } else {
5625                 GOTO(stop, rc);
5626         }
5627
5628 stop:
5629         if (dtlocked)
5630                 dt_write_unlock(env, obj);
5631
5632         if (handle != NULL && !IS_ERR(handle))
5633                 dt_trans_stop(env, dev, handle);
5634
5635 out:
5636         lfsck_ibits_unlock(&lh, LCK_EX);
5637
5638         if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5639             !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5640                                              lnr->lnr_name, lnr->lnr_namelen) &&
5641             type != LNIT_BAD_DIRENT) {
5642                 ns->ln_flags |= LF_INCONSISTENT;
5643
5644                 log = false;
5645                 if (dir == NULL) {
5646                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5647                         if (IS_ERR(dir)) {
5648                                 rc = PTR_ERR(dir);
5649
5650                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5651                         }
5652                 }
5653
5654                 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5655                                                 lnr->lnr_lmv, lnr->lnr_name);
5656                 if (rc == 0)
5657                         bad_hash = true;
5658         }
5659
5660         if (rc >= 0) {
5661                 if (type != LNIT_NONE && dir == NULL) {
5662                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5663                         if (IS_ERR(dir)) {
5664                                 rc = PTR_ERR(dir);
5665
5666                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5667                         }
5668                 }
5669
5670                 switch (type) {
5671                 case LNIT_BAD_TYPE:
5672                         log = false;
5673                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5674                                         obj, lnr->lnr_name, lnr->lnr_name,
5675                                         lnr->lnr_type, true, false);
5676                         if (rc > 0)
5677                                 repaired = true;
5678                         break;
5679                 case LNIT_BAD_DIRENT:
5680                         log = false;
5681                         /* XXX: This is a bad dirent, we do not know whether
5682                          *      the original name entry reference a regular
5683                          *      file or a directory, then keep the parent's
5684                          *      nlink count unchanged here. */
5685                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5686                                         obj, lnr->lnr_name, lnr->lnr_name,
5687                                         lnr->lnr_type, false, false);
5688                         if (rc > 0)
5689                                 repaired = true;
5690                         break;
5691                 default:
5692                         break;
5693                 }
5694
5695                 if (obj != NULL && count == 1 &&
5696                     S_ISREG(lfsck_object_type(obj)))
5697                         dt_attr_get(env, obj, la);
5698         }
5699
5700 trace:
5701         down_write(&com->lc_sem);
5702         if (rc < 0) {
5703                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5704                        "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5705                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5706                        lnr->lnr_namelen, lnr->lnr_name, rc);
5707
5708                 lfsck_namespace_record_failure(env, lfsck, ns);
5709                 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5710                      rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5711                      rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5712                     dev != NULL && dev != lfsck->li_bottom)
5713                         lfsck_lad_set_bitmap(env, com, idx);
5714
5715                 if (!(bk->lb_param & LPF_FAILOUT))
5716                         rc = 0;
5717         } else {
5718                 if (repaired) {
5719                         ns->ln_items_repaired++;
5720                         if (log)
5721                                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5722                                        "repaired the entry: "DFID", parent "DFID
5723                                        ", name %.*s, type %d\n",
5724                                        lfsck_lfsck2name(lfsck),
5725                                        PFID(&lnr->lnr_fid), PFID(pfid),
5726                                        lnr->lnr_namelen, lnr->lnr_name, type);
5727
5728                         switch (type) {
5729                         case LNIT_DANGLING:
5730                                 ns->ln_dangling_repaired++;
5731                                 break;
5732                         case LNIT_BAD_TYPE:
5733                                 ns->ln_bad_type_repaired++;
5734                                 break;
5735                         case LNIT_BAD_DIRENT:
5736                                 ns->ln_dirent_repaired++;
5737                                 break;
5738                         default:
5739                                 break;
5740                         }
5741
5742                         if (bk->lb_param & LPF_DRYRUN &&
5743                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5744                                 lfsck_pos_fill(env, lfsck,
5745                                                &ns->ln_pos_first_inconsistent,
5746                                                false);
5747                 }
5748
5749                 if (bad_hash) {
5750                         ns->ln_name_hash_repaired++;
5751
5752                         /* Not count repeatedly. */
5753                         if (!repaired) {
5754                                 ns->ln_items_repaired++;
5755                                 if (log)
5756                                         CDEBUG(D_LFSCK, "%s: namespace LFSCK "
5757                                                "assistant repaired the entry: "
5758                                                DFID", parent "DFID
5759                                                ", name %.*s\n",
5760                                                lfsck_lfsck2name(lfsck),
5761                                                PFID(&lnr->lnr_fid), PFID(pfid),
5762                                                lnr->lnr_namelen, lnr->lnr_name);
5763                         }
5764
5765                         if (bk->lb_param & LPF_DRYRUN &&
5766                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5767                                 lfsck_pos_fill(env, lfsck,
5768                                                &ns->ln_pos_first_inconsistent,
5769                                                false);
5770                 }
5771
5772                 rc = 0;
5773         }
5774
5775         if (count > 1 || la->la_nlink > 1)
5776                 ns->ln_mul_linked_checked++;
5777
5778         up_write(&com->lc_sem);
5779
5780         if (obj != NULL && !IS_ERR(obj))
5781                 lfsck_object_put(env, obj);
5782
5783         if (dir != NULL && !IS_ERR(dir))
5784                 lfsck_object_put(env, dir);
5785
5786         lad->lad_advance_lock = bad_linkea;
5787
5788         return rc;
5789 }
5790
5791 /**
5792  * Handle one orphan under the backend /lost+found directory
5793  *
5794  * Insert the orphan FID into the namespace LFSCK trace file for further
5795  * processing (via the subsequent namespace LFSCK second-stage scanning).
5796  * At the same time, remove the orphan name entry from backend /lost+found
5797  * directory. There is an interval between the orphan name entry removed
5798  * from the backend /lost+found directory and the orphan FID in the LFSCK
5799  * trace file handled. In such interval, the LFSCK can be reset, then
5800  * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5801  * To guarantee that the orphans can be found when LFSCK run next time
5802  * without e2fsck again, when remove the orphan name entry, the LFSCK
5803  * will set the orphan's ctime attribute as 1. Since normal applications
5804  * cannot change the object's ctime attribute as 1. Then when LFSCK run
5805  * next time, it can record the object (that ctime is 1) in the namespace
5806  * LFSCK trace file during the first-stage scanning.
5807  *
5808  * \param[in] env       pointer to the thread context
5809  * \param[in] com       pointer to the lfsck component
5810  * \param[in] parent    pointer to the object for the backend /lost+found
5811  * \param[in] ent       pointer to the name entry for the target under the
5812  *                      backend /lost+found
5813  *
5814  * \retval              positive for repaired
5815  * \retval              0 if needs to repair nothing
5816  * \retval              negative error number on failure
5817  */
5818 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5819                                               struct lfsck_component *com,
5820                                               struct dt_object *parent,
5821                                               struct lu_dirent *ent)
5822 {
5823         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5824         struct lu_fid                   *key    = &info->lti_fid;
5825         struct lu_attr                  *la     = &info->lti_la;
5826         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5827         struct dt_object                *obj;
5828         struct dt_device                *dev    = lfsck->li_bottom;
5829         struct dt_object                *child  = NULL;
5830         struct thandle                  *th     = NULL;
5831         int                              idx;
5832         int                              rc     = 0;
5833         __u8                             flags  = 0;
5834         bool                             exist  = false;
5835         ENTRY;
5836
5837         child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5838         if (IS_ERR(child))
5839                 RETURN(PTR_ERR(child));
5840
5841         LASSERT(dt_object_exists(child));
5842         LASSERT(!dt_object_remote(child));
5843
5844         idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5845         obj = com->lc_sub_trace_objs[idx].lsto_obj;
5846         fid_cpu_to_be(key, &ent->lde_fid);
5847         rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5848                        (const struct dt_key *)key);
5849         if (rc == 0) {
5850                 exist = true;
5851                 flags |= LNTF_CHECK_ORPHAN;
5852         } else if (rc == -ENOENT) {
5853                 flags = LNTF_CHECK_ORPHAN;
5854         } else {
5855                 GOTO(out, rc);
5856         }
5857
5858         th = dt_trans_create(env, dev);
5859         if (IS_ERR(th))
5860                 GOTO(out, rc = PTR_ERR(th));
5861
5862         /* a1. remove name entry from backend /lost+found */
5863         rc = dt_declare_delete(env, parent,
5864                                (const struct dt_key *)ent->lde_name, th);
5865         if (rc != 0)
5866                 GOTO(stop, rc);
5867
5868         if (S_ISDIR(lfsck_object_type(child))) {
5869                 /* a2. decrease parent's nlink */
5870                 rc = dt_declare_ref_del(env, parent, th);
5871                 if (rc != 0)
5872                         GOTO(stop, rc);
5873         }
5874
5875         if (exist) {
5876                 /* a3. remove child's FID from the LFSCK trace file. */
5877                 rc = dt_declare_delete(env, obj,
5878                                        (const struct dt_key *)key, th);
5879                 if (rc != 0)
5880                         GOTO(stop, rc);
5881         } else {
5882                 /* a4. set child's ctime as 1 */
5883                 memset(la, 0, sizeof(*la));
5884                 la->la_ctime = 1;
5885                 la->la_valid = LA_CTIME;
5886                 rc = dt_declare_attr_set(env, child, la, th);
5887                 if (rc != 0)
5888                         GOTO(stop, rc);
5889         }
5890
5891         /* a5. insert child's FID into the LFSCK trace file. */
5892         rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5893                                (const struct dt_key *)key, th);
5894         if (rc != 0)
5895                 GOTO(stop, rc);
5896
5897         rc = dt_trans_start_local(env, dev, th);
5898         if (rc != 0)
5899                 GOTO(stop, rc);
5900
5901         /* b1. remove name entry from backend /lost+found */
5902         rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5903         if (rc != 0)
5904                 GOTO(stop, rc);
5905
5906         if (S_ISDIR(lfsck_object_type(child))) {
5907                 /* b2. decrease parent's nlink */
5908                 dt_write_lock(env, parent, 0);
5909                 rc = dt_ref_del(env, parent, th);
5910                 dt_write_unlock(env, parent);
5911                 if (rc != 0)
5912                         GOTO(stop, rc);
5913         }
5914
5915         if (exist) {
5916                 /* a3. remove child's FID from the LFSCK trace file. */
5917                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5918                 if (rc != 0)
5919                         GOTO(stop, rc);
5920         } else {
5921                 /* b4. set child's ctime as 1 */
5922                 rc = dt_attr_set(env, child, la, th);
5923                 if (rc != 0)
5924                         GOTO(stop, rc);
5925         }
5926
5927         /* b5. insert child's FID into the LFSCK trace file. */
5928         rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5929                        (const struct dt_key *)key, th, 1);
5930
5931         GOTO(stop, rc = (rc == 0 ? 1 : rc));
5932
5933 stop:
5934         dt_trans_stop(env, dev, th);
5935
5936 out:
5937         lfsck_object_put(env, child);
5938
5939         return rc;
5940 }
5941
5942 /**
5943  * Handle orphans under the backend /lost+found directory
5944  *
5945  * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5946  * and put them under the backend /lost+found directory that is invisible
5947  * to client. The LFSCK will scan such directory, for the original client
5948  * visible orphans, add their fids into the namespace LFSCK trace file,
5949  * then the subsenquent namespace LFSCK second-stage scanning can handle
5950  * them as other objects to be double scanned: either move back to normal
5951  * namespace, or to the global visible orphan directory:
5952  * /ROOT/.lustre/lost+found/MDTxxxx/
5953  *
5954  * \param[in] env       pointer to the thread context
5955  * \param[in] com       pointer to the lfsck component
5956  */
5957 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5958                                            struct lfsck_component *com)
5959 {
5960         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5961         struct lu_dirent                *ent    =
5962                                         (struct lu_dirent *)info->lti_key;
5963         struct lu_seq_range             *range  = &info->lti_range;
5964         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5965         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5966         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5967         struct lfsck_namespace          *ns     = com->lc_file_ram;
5968         struct dt_object                *parent;
5969         const struct dt_it_ops          *iops;
5970         struct dt_it                    *di;
5971         struct seq_server_site          *ss     = lfsck_dev_site(lfsck);
5972         __u64                            cookie;
5973         __u32                            idx    = lfsck_dev_idx(lfsck);
5974         int                              rc     = 0;
5975         __u16                            type;
5976         ENTRY;
5977
5978         parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5979                                           &LU_BACKEND_LPF_FID);
5980         if (IS_ERR(parent)) {
5981                 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5982                        lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5983                 RETURN_EXIT;
5984         }
5985
5986         /* It is normal that the /lost+found does not exist for ZFS backend. */
5987         if (!dt_object_exists(parent))
5988                 GOTO(out, rc = 0);
5989
5990         if (unlikely(!dt_try_as_dir(env, parent)))
5991                 GOTO(out, rc = -ENOTDIR);
5992
5993         CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5994                lfsck_lfsck2name(lfsck));
5995
5996         com->lc_new_scanned = 0;
5997         iops = &parent->do_index_ops->dio_it;
5998         di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
5999         if (IS_ERR(di))
6000                 GOTO(out, rc = PTR_ERR(di));
6001
6002         rc = iops->load(env, di, 0);
6003         if (rc == 0)
6004                 rc = iops->next(env, di);
6005         else if (rc > 0)
6006                 rc = 0;
6007
6008         while (rc == 0) {
6009                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6010                     unlikely(!thread_is_running(thread)))
6011                         break;
6012
6013                 rc = iops->rec(env, di, (struct dt_rec *)ent,
6014                                LUDA_64BITHASH | LUDA_TYPE);
6015                 if (rc == 0)
6016                         rc = lfsck_unpack_ent(ent, &cookie, &type);
6017
6018                 if (unlikely(rc != 0)) {
6019                         CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6020                                "/lost+found: rc = %d\n",
6021                                lfsck_lfsck2name(lfsck), rc);
6022
6023                         goto skip;
6024                 }
6025
6026                 /* skip dot and dotdot entries */
6027                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6028                         goto next;
6029
6030                 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6031                         goto skip;
6032
6033                 if (fid_is_norm(&ent->lde_fid)) {
6034                         fld_range_set_mdt(range);
6035                         rc = fld_local_lookup(env, ss->ss_server_fld,
6036                                               fid_seq(&ent->lde_fid), range);
6037                         if (rc != 0)
6038                                 goto skip;
6039                 } else if (idx != 0) {
6040                         /* If the returned FID is IGIF, then there are three
6041                          * possible cases:
6042                          *
6043                          * 1) The object is upgraded from old Lustre-1.8 with
6044                          *    IGIF assigned to such object.
6045                          * 2) The object is a backend local object and is
6046                          *    invisible to client.
6047                          * 3) The object lost its LMV EA, and since there is
6048                          *    no FID-in-dirent for the orphan in the backend
6049                          *    /lost+found directory, then the low layer will
6050                          *    return IGIF for such object.
6051                          *
6052                          * For MDTx (x != 0), it is either case 2) or case 3),
6053                          * but from the LFSCK view, they are indistinguishable.
6054                          * To be safe, the LFSCK will keep it there and report
6055                          * some message, then the adminstrator can handle that
6056                          * furtherly.
6057                          *
6058                          * For MDT0, it is more possible the case 1). The LFSCK
6059                          * will handle the orphan as an upgraded object. */
6060                         CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6061                                "in the backend /lost+found on the MDT %04x, "
6062                                "to be safe, skip it.\n",
6063                                lfsck_lfsck2name(lfsck), ent->lde_namelen,
6064                                ent->lde_name, PFID(&ent->lde_fid), idx);
6065                         goto skip;
6066                 }
6067
6068                 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6069
6070 skip:
6071                 down_write(&com->lc_sem);
6072                 com->lc_new_scanned++;
6073                 ns->ln_local_lpf_scanned++;
6074                 if (rc > 0)
6075                         ns->ln_local_lpf_moved++;
6076                 else if (rc == 0)
6077                         ns->ln_local_lpf_skipped++;
6078                 else
6079                         ns->ln_local_lpf_failed++;
6080                 up_write(&com->lc_sem);
6081
6082                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6083                         break;
6084
6085 next:
6086                 lfsck_control_speed_by_self(com);
6087                 if (unlikely(!thread_is_running(thread))) {
6088                         rc = 0;
6089                         break;
6090                 }
6091
6092                 rc = iops->next(env, di);
6093         }
6094
6095         iops->put(env, di);
6096         iops->fini(env, di);
6097
6098         EXIT;
6099
6100 out:
6101         CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6102                lfsck_lfsck2name(lfsck), rc);
6103
6104         lfsck_object_put(env, parent);
6105 }
6106
6107 /**
6108  * Rescan the striped directory after the master LMV EA reset.
6109  *
6110  * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6111  * the namespace LFSCK engine scan the striped directory for the first time,
6112  * it will be regarded as a normal directory. As the LFSCK processing, some
6113  * other LFSCK instance on other MDT will find the shard of this striped dir,
6114  * and find that the master MDT-object of the striped directory lost its LMV
6115  * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6116  * notify the LFSCK instance on this MDT to rescan the striped directory.
6117  *
6118  * \param[in] env       pointer to the thread context
6119  * \param[in] com       pointer to the lfsck component
6120  * \param[in] llu       the lfsck_lmv_unit that contains the striped directory
6121  *                      to be rescanned.
6122  *
6123  * \retval              positive number for success
6124  * \retval              0 for LFSCK stopped/paused
6125  * \retval              negative error number on failure
6126  */
6127 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6128                                               struct lfsck_component *com,
6129                                               struct lfsck_lmv_unit *llu)
6130 {
6131         struct lfsck_thread_info        *info   = lfsck_env_info(env);
6132         struct lfsck_instance           *lfsck  = com->lc_lfsck;
6133         struct lfsck_assistant_data     *lad    = com->lc_data;
6134         struct dt_object                *dir;
6135         const struct dt_it_ops          *iops;
6136         struct dt_it                    *di;
6137         struct lu_dirent                *ent    =
6138                         (struct lu_dirent *)info->lti_key;
6139         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
6140         struct ptlrpc_thread            *thread = &lfsck->li_thread;
6141         struct lfsck_assistant_object   *lso    = NULL;
6142         struct lfsck_namespace_req      *lnr;
6143         struct lfsck_assistant_req      *lar;
6144         int                              rc;
6145         __u16                            type;
6146         ENTRY;
6147
6148         LASSERT(list_empty(&lad->lad_req_list));
6149
6150         lfsck->li_lmv = &llu->llu_lmv;
6151         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6152         rc = lfsck_open_dir(env, lfsck, 0);
6153         if (rc != 0)
6154                 RETURN(rc);
6155
6156         dir = lfsck->li_obj_dir;
6157         di = lfsck->li_di_dir;
6158         iops = &dir->do_index_ops->dio_it;
6159         do {
6160                 rc = iops->rec(env, di, (struct dt_rec *)ent,
6161                                lfsck->li_args_dir);
6162                 if (rc == 0)
6163                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6164                                               &type);
6165
6166                 if (rc != 0) {
6167                         if (bk->lb_param & LPF_FAILOUT)
6168                                 GOTO(out, rc);
6169
6170                         goto next;
6171                 }
6172
6173                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6174                         goto next;
6175
6176                 if (lso == NULL) {
6177                         lso = lfsck_assistant_object_init(env,
6178                                 lfsck_dto2fid(dir), NULL,
6179                                 lfsck->li_pos_current.lp_oit_cookie, true);
6180                         if (IS_ERR(lso)) {
6181                                 if (bk->lb_param & LPF_FAILOUT)
6182                                         GOTO(out, rc = PTR_ERR(lso));
6183
6184                                 lso = NULL;
6185                                 goto next;
6186                         }
6187                 }
6188
6189                 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6190                 if (IS_ERR(lnr)) {
6191                         if (bk->lb_param & LPF_FAILOUT)
6192                                 GOTO(out, rc = PTR_ERR(lnr));
6193
6194                         goto next;
6195                 }
6196
6197                 lar = &lnr->lnr_lar;
6198                 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6199                 lfsck_namespace_assistant_req_fini(env, lar);
6200                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6201                         GOTO(out, rc);
6202
6203                 if (unlikely(!thread_is_running(thread)))
6204                         GOTO(out, rc = 0);
6205
6206 next:
6207                 rc = iops->next(env, di);
6208         } while (rc == 0);
6209
6210 out:
6211         if (lso != NULL && !IS_ERR(lso))
6212                 lfsck_assistant_object_put(env, lso);
6213
6214         lfsck_close_dir(env, lfsck, rc);
6215         if (rc <= 0)
6216                 RETURN(rc);
6217
6218         /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6219         if (list_empty(&lad->lad_req_list))
6220                 RETURN(1);
6221
6222         spin_lock(&lad->lad_lock);
6223         lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6224                           lar_list);
6225         list_del_init(&lar->lar_list);
6226         spin_unlock(&lad->lad_lock);
6227
6228         rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6229         lfsck_namespace_assistant_req_fini(env, lar);
6230
6231         RETURN(rc == 0 ? 1 : rc);
6232 }
6233
6234 static int
6235 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6236                                            struct lfsck_component *com,
6237                                            struct dt_object *obj, bool first)
6238 {
6239         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6240         struct ptlrpc_thread    *thread = &lfsck->li_thread;
6241         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
6242         struct lfsck_namespace  *ns     = com->lc_file_ram;
6243         const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
6244         struct dt_object        *target;
6245         struct dt_it            *di;
6246         struct dt_key           *key;
6247         struct lu_fid            fid;
6248         int                      rc;
6249         __u8                     flags  = 0;
6250         ENTRY;
6251
6252         di = iops->init(env, obj, 0);
6253         if (IS_ERR(di))
6254                 RETURN(PTR_ERR(di));
6255
6256         if (first)
6257                 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6258         else
6259                 fid_zero(&fid);
6260         rc = iops->get(env, di, (const struct dt_key *)&fid);
6261         if (rc < 0)
6262                 GOTO(fini, rc);
6263
6264         if (first) {
6265                 /* The start one either has been processed or does not exist,
6266                  * skip it. */
6267                 rc = iops->next(env, di);
6268                 if (rc != 0)
6269                         GOTO(put, rc);
6270         }
6271
6272         do {
6273                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6274                     unlikely(!thread_is_running(thread)))
6275                         GOTO(put, rc = 0);
6276
6277                 key = iops->key(env, di);
6278                 if (IS_ERR(key)) {
6279                         rc = PTR_ERR(key);
6280                         if (rc == -ENOENT)
6281                                 GOTO(put, rc = 1);
6282
6283                         goto checkpoint;
6284                 }
6285
6286                 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6287                 if (!fid_is_sane(&fid)) {
6288                         rc = 0;
6289                         goto checkpoint;
6290                 }
6291
6292                 target = lfsck_object_find_bottom(env, lfsck, &fid);
6293                 if (IS_ERR(target)) {
6294                         rc = PTR_ERR(target);
6295                         goto checkpoint;
6296                 }
6297
6298                 if (dt_object_exists(target)) {
6299                         rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6300                         if (rc == 0) {
6301                                 rc = lfsck_namespace_double_scan_one(env, com,
6302                                                                 target, flags);
6303                                 if (rc == -ENOENT)
6304                                         rc = 0;
6305                         }
6306                 }
6307
6308                 lfsck_object_put(env, target);
6309
6310 checkpoint:
6311                 down_write(&com->lc_sem);
6312                 com->lc_new_checked++;
6313                 com->lc_new_scanned++;
6314                 if (rc >= 0)
6315                         ns->ln_fid_latest_scanned_phase2 = fid;
6316
6317                 if (rc > 0)
6318                         ns->ln_objs_repaired_phase2++;
6319                 else if (rc < 0)
6320                         ns->ln_objs_failed_phase2++;
6321                 up_write(&com->lc_sem);
6322
6323                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6324                         GOTO(put, rc);
6325
6326                 if (unlikely(com->lc_time_next_checkpoint <=
6327                              ktime_get_seconds()) &&
6328                     com->lc_new_checked != 0) {
6329                         down_write(&com->lc_sem);
6330                         ns->ln_run_time_phase2 += ktime_get_seconds() -
6331                                                   com->lc_time_last_checkpoint;
6332                         ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6333                         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6334                         com->lc_new_checked = 0;
6335                         lfsck_namespace_store(env, com);
6336                         up_write(&com->lc_sem);
6337
6338                         com->lc_time_last_checkpoint = ktime_get_seconds();
6339                         com->lc_time_next_checkpoint =
6340                                 com->lc_time_last_checkpoint +
6341                                 LFSCK_CHECKPOINT_INTERVAL;
6342                 }
6343
6344                 lfsck_control_speed_by_self(com);
6345                 if (unlikely(!thread_is_running(thread)))
6346                         GOTO(put, rc = 0);
6347
6348                 rc = iops->next(env, di);
6349         } while (rc == 0);
6350
6351         GOTO(put, rc);
6352
6353 put:
6354         iops->put(env, di);
6355
6356 fini:
6357         iops->fini(env, di);
6358
6359         return rc;
6360 }
6361
6362 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6363                                                 struct lfsck_component *com)
6364 {
6365         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6366         struct lfsck_namespace  *ns     = com->lc_file_ram;
6367         int                      rc;
6368         int                      i;
6369         ENTRY;
6370
6371         while (!list_empty(&lfsck->li_list_lmv)) {
6372                 struct lfsck_lmv_unit *llu;
6373
6374                 spin_lock(&lfsck->li_lock);
6375                 llu = list_entry(lfsck->li_list_lmv.next,
6376                                  struct lfsck_lmv_unit, llu_link);
6377                 list_del_init(&llu->llu_link);
6378                 spin_unlock(&lfsck->li_lock);
6379
6380                 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6381                 if (rc <= 0)
6382                         RETURN(rc);
6383         }
6384
6385         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6386                lfsck_lfsck2name(lfsck));
6387
6388         lfsck_namespace_scan_local_lpf(env, com);
6389
6390         com->lc_new_checked = 0;
6391         com->lc_new_scanned = 0;
6392         com->lc_time_last_checkpoint = ktime_get_seconds();
6393         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6394                                        LFSCK_CHECKPOINT_INTERVAL;
6395
6396         i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6397         rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6398                                 com->lc_sub_trace_objs[i].lsto_obj, true);
6399         while (rc > 0 && ++i < LFSCK_STF_COUNT)
6400                 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6401                                 com->lc_sub_trace_objs[i].lsto_obj, false);
6402
6403         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6404                "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6405
6406         RETURN(rc);
6407 }
6408
6409 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6410                                                struct lfsck_component *com,
6411                                                struct lfsck_position *pos)
6412 {
6413         struct lfsck_assistant_data     *lad = com->lc_data;
6414         struct lfsck_namespace_req      *lnr;
6415
6416         if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6417             LS_SCANNING_PHASE1)
6418                 return;
6419
6420         if (list_empty(&lad->lad_req_list))
6421                 return;
6422
6423         lnr = list_entry(lad->lad_req_list.next,
6424                          struct lfsck_namespace_req,
6425                          lnr_lar.lar_list);
6426         pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6427         pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6428         pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6429 }
6430
6431 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6432                                               struct lfsck_component *com,
6433                                               int rc)
6434 {
6435         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6436         struct lfsck_namespace  *ns     = com->lc_file_ram;
6437
6438         down_write(&com->lc_sem);
6439         ns->ln_run_time_phase2 += ktime_get_seconds() -
6440                                   com->lc_time_last_checkpoint;
6441         ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6442         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6443         com->lc_new_checked = 0;
6444
6445         if (rc > 0) {
6446                 if (ns->ln_flags & LF_INCOMPLETE)
6447                         ns->ln_status = LS_PARTIAL;
6448                 else
6449                         ns->ln_status = LS_COMPLETED;
6450                 ns->ln_flags &= ~LF_SCANNED_ONCE;
6451                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6452                         ns->ln_flags &= ~LF_INCONSISTENT;
6453                 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6454                 ns->ln_success_count++;
6455         } else if (rc == 0) {
6456                 if (lfsck->li_status != 0)
6457                         ns->ln_status = lfsck->li_status;
6458                 else
6459                         ns->ln_status = LS_STOPPED;
6460         } else {
6461                 ns->ln_status = LS_FAILED;
6462         }
6463
6464         rc = lfsck_namespace_store(env, com);
6465         up_write(&com->lc_sem);
6466
6467         return rc;
6468 }
6469
6470 static int
6471 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6472                                                   struct ptlrpc_request *req,
6473                                                   void *args, int rc)
6474 {
6475         if (rc == 0) {
6476                 struct lfsck_async_interpret_args *laia = args;
6477                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
6478
6479                 ltd->ltd_synced_failures = 1;
6480         }
6481
6482         return 0;
6483 }
6484
6485 /**
6486  * Notify remote LFSCK instances about former failures.
6487  *
6488  * The local LFSCK instance has recorded which MDTs have ever failed to respond
6489  * some LFSCK verification requests (maybe because of network issues or the MDT
6490  * itself trouble). During the respond gap the MDT may missed some name entries
6491  * verification, then the MDT cannot know whether related MDT-objects have been
6492  * referenced by related name entries or not, then in the second-stage scanning,
6493  * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6494  * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6495  * name entry for repairing the orphan.
6496  *
6497  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6498  * it will scan the bitmap for the ever failed MDTs, and notify them that they
6499  * have ever missed some name entries verification and should skip the handling
6500  * for orphan MDT-objects.
6501  *
6502  * \param[in] env       pointer to the thread context
6503  * \param[in] com       pointer to the lfsck component
6504  * \param[in] lr        pointer to the lfsck request
6505  */
6506 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6507                                                     struct lfsck_component *com,
6508                                                     struct lfsck_request *lr)
6509 {
6510         struct lfsck_async_interpret_args *laia  =
6511                                 &lfsck_env_info(env)->lti_laia2;
6512         struct lfsck_assistant_data       *lad   = com->lc_data;
6513         struct lfsck_namespace            *ns    = com->lc_file_ram;
6514         struct lfsck_instance             *lfsck = com->lc_lfsck;
6515         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
6516         struct lfsck_tgt_desc             *ltd;
6517         struct ptlrpc_request_set         *set;
6518         __u32                              idx;
6519         int                                rc    = 0;
6520         ENTRY;
6521
6522         if (!lad->lad_incomplete)
6523                 RETURN_EXIT;
6524
6525         set = ptlrpc_prep_set();
6526         if (set == NULL)
6527                 GOTO(out, rc = -ENOMEM);
6528
6529         lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6530         memset(laia, 0, sizeof(*laia));
6531         lad->lad_touch_gen++;
6532
6533         down_read(&ltds->ltd_rw_sem);
6534         cfs_foreach_bit(lad->lad_bitmap, idx) {
6535                 ltd = lfsck_ltd2tgt(ltds, idx);
6536                 if (unlikely(!ltd))
6537                         continue;
6538
6539                 laia->laia_ltd = ltd;
6540                 rc = lfsck_async_request(env, ltd, lr, set,
6541                         lfsck_namespace_assistant_sync_failures_interpret,
6542                         laia, LFSCK_NOTIFY);
6543                 if (rc != 0)
6544                         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6545                                "to sync failure with MDT %x: rc = %d\n",
6546                                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6547         }
6548         up_read(&ltds->ltd_rw_sem);
6549
6550         rc = ptlrpc_set_wait(set);
6551         ptlrpc_set_destroy(set);
6552
6553         GOTO(out, rc);
6554
6555 out:
6556         if (rc != 0)
6557                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6558                        "to sync failure with MDTs, and related MDTs "
6559                        "may handle orphan improperly: rc = %d\n",
6560                        lfsck_lfsck2name(lfsck), rc);
6561
6562         EXIT;
6563 }
6564
6565 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6566         .la_handler_p1          = lfsck_namespace_assistant_handler_p1,
6567         .la_handler_p2          = lfsck_namespace_assistant_handler_p2,
6568         .la_fill_pos            = lfsck_namespace_assistant_fill_pos,
6569         .la_double_scan_result  = lfsck_namespace_double_scan_result,
6570         .la_req_fini            = lfsck_namespace_assistant_req_fini,
6571         .la_sync_failures       = lfsck_namespace_assistant_sync_failures,
6572 };
6573
6574 /**
6575  * Verify the specified linkEA entry for the given directory object.
6576  * If the object has no such linkEA entry or it has more other linkEA
6577  * entries, then re-generate the linkEA with the given information.
6578  *
6579  * \param[in] env       pointer to the thread context
6580  * \param[in] obj       pointer to the dt_object to be handled
6581  * \param[in] cname     the name for the child in the parent directory
6582  * \param[in] pfid      the parent directory's FID for the linkEA
6583  *
6584  * \retval              0 for success
6585  * \retval              negative error number on failure
6586  */
6587 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6588                         const struct lu_name *cname, const struct lu_fid *pfid)
6589 {
6590         struct dt_device        *dev    = lfsck_obj2dev(obj);
6591         struct linkea_data       ldata  = { NULL };
6592         struct lu_buf            linkea_buf;
6593         struct thandle          *th;
6594         int                      rc;
6595         int                      fl     = LU_XATTR_CREATE;
6596         bool                     dirty  = false;
6597         ENTRY;
6598
6599         LASSERT(S_ISDIR(lfsck_object_type(obj)));
6600
6601         rc = lfsck_links_read_with_rec(env, obj, &ldata);
6602         if (rc == -ENODATA) {
6603                 dirty = true;
6604         } else if (rc == 0) {
6605                 fl = LU_XATTR_REPLACE;
6606                 if (ldata.ld_leh->leh_reccount != 1) {
6607                         dirty = true;
6608                 } else {
6609                         rc = linkea_links_find(&ldata, cname, pfid);
6610                         if (rc != 0)
6611                                 dirty = true;
6612                 }
6613         }
6614
6615         if (!dirty)
6616                 RETURN(rc);
6617
6618         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6619                               cname, pfid);
6620         if (rc != 0)
6621                 RETURN(rc);
6622
6623         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6624                        ldata.ld_leh->leh_len);
6625         th = dt_trans_create(env, dev);
6626         if (IS_ERR(th))
6627                 RETURN(PTR_ERR(th));
6628
6629         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6630                                   XATTR_NAME_LINK, fl, th);
6631         if (rc != 0)
6632                 GOTO(stop, rc);
6633
6634         rc = dt_trans_start_local(env, dev, th);
6635         if (rc != 0)
6636                 GOTO(stop, rc);
6637
6638         dt_write_lock(env, obj, 0);
6639         rc = dt_xattr_set(env, obj, &linkea_buf,
6640                           XATTR_NAME_LINK, fl, th);
6641         dt_write_unlock(env, obj);
6642
6643         GOTO(stop, rc);
6644
6645 stop:
6646         dt_trans_stop(env, dev, th);
6647         return rc;
6648 }
6649
6650 /**
6651  * Get the name and parent directory's FID from the first linkEA entry.
6652  *
6653  * \param[in] env       pointer to the thread context
6654  * \param[in] obj       pointer to the object which get linkEA from
6655  * \param[out] name     pointer to the buffer to hold the name
6656  *                      in the first linkEA entry
6657  * \param[out] pfid     pointer to the buffer to hold the parent
6658  *                      directory's FID in the first linkEA entry
6659  *
6660  * \retval              0 for success
6661  * \retval              negative error number on failure
6662  */
6663 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6664                           char *name, struct lu_fid *pfid)
6665 {
6666         struct lu_name           *cname = &lfsck_env_info(env)->lti_name;
6667         struct linkea_data        ldata = { NULL };
6668         int                       rc;
6669
6670         rc = lfsck_links_read_with_rec(env, obj, &ldata);
6671         if (rc)
6672                 return rc;
6673
6674         linkea_first_entry(&ldata);
6675         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6676         /* To guarantee the 'name' is terminated with '0'. */
6677         memcpy(name, cname->ln_name, cname->ln_namelen);
6678         name[cname->ln_namelen] = 0;
6679
6680         return 0;
6681 }
6682
6683 /**
6684  * Update the object's name entry with the given FID.
6685  *
6686  * \param[in] env       pointer to the thread context
6687  * \param[in] lfsck     pointer to the lfsck instance
6688  * \param[in] dir       pointer to the directory that holds
6689  *                      the name entry
6690  * \param[in] name      the name for the entry to be updated
6691  * \param[in] fid       the new FID for the name entry referenced
6692  * \param[in] type      the type for the name entry to be updated
6693  *
6694  * \retval              0 for success
6695  * \retval              negative error number on failure
6696  */
6697 int lfsck_update_name_entry(const struct lu_env *env,
6698                             struct lfsck_instance *lfsck,
6699                             struct dt_object *dir, const char *name,
6700                             const struct lu_fid *fid, __u32 type)
6701 {
6702         struct lfsck_thread_info *info   = lfsck_env_info(env);
6703         struct dt_insert_rec     *rec    = &info->lti_dt_rec;
6704         struct lfsck_lock_handle *llh    = &info->lti_llh;
6705         struct dt_device         *dev    = lfsck_obj2dev(dir);
6706         struct thandle           *th;
6707         int                       rc;
6708         bool                      exists = true;
6709         ENTRY;
6710
6711         rc = lfsck_lock(env, lfsck, dir, name, llh,
6712                         MDS_INODELOCK_UPDATE, LCK_PW);
6713         if (rc != 0)
6714                 RETURN(rc);
6715
6716         th = dt_trans_create(env, dev);
6717         if (IS_ERR(th))
6718                 GOTO(unlock, rc = PTR_ERR(th));
6719
6720         rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6721         if (rc != 0)
6722                 GOTO(stop, rc);
6723
6724         rec->rec_type = type;
6725         rec->rec_fid = fid;
6726         rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6727                                (const struct dt_key *)name, th);
6728         if (rc != 0)
6729                 GOTO(stop, rc);
6730
6731         rc = dt_declare_ref_add(env, dir, th);
6732         if (rc != 0)
6733                 GOTO(stop, rc);
6734
6735         rc = dt_trans_start_local(env, dev, th);
6736         if (rc != 0)
6737                 GOTO(stop, rc);
6738
6739         rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6740         if (rc == -ENOENT) {
6741                 exists = false;
6742                 rc = 0;
6743         }
6744
6745         if (rc != 0)
6746                 GOTO(stop, rc);
6747
6748         rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6749                        (const struct dt_key *)name, th, 1);
6750         if (rc == 0 && S_ISDIR(type) && !exists) {
6751                 dt_write_lock(env, dir, 0);
6752                 rc = dt_ref_add(env, dir, th);
6753                 dt_write_unlock(env, dir);
6754         }
6755
6756         GOTO(stop, rc);
6757
6758 stop:
6759         dt_trans_stop(env, dev, th);
6760
6761 unlock:
6762         lfsck_unlock(llh);
6763         CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6764                " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6765                PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6766
6767         return rc;
6768 }
6769
6770 int lfsck_namespace_setup(const struct lu_env *env,
6771                           struct lfsck_instance *lfsck)
6772 {
6773         struct lfsck_component  *com;
6774         struct lfsck_namespace  *ns;
6775         struct dt_object        *root = NULL;
6776         struct dt_object        *obj;
6777         int                      i;
6778         int                      rc;
6779         ENTRY;
6780
6781         LASSERT(lfsck->li_master);
6782
6783         OBD_ALLOC_PTR(com);
6784         if (com == NULL)
6785                 RETURN(-ENOMEM);
6786
6787         INIT_LIST_HEAD(&com->lc_link);
6788         INIT_LIST_HEAD(&com->lc_link_dir);
6789         init_rwsem(&com->lc_sem);
6790         atomic_set(&com->lc_ref, 1);
6791         com->lc_lfsck = lfsck;
6792         com->lc_type = LFSCK_TYPE_NAMESPACE;
6793         com->lc_ops = &lfsck_namespace_ops;
6794         com->lc_data = lfsck_assistant_data_init(
6795                         &lfsck_namespace_assistant_ops,
6796                         LFSCK_NAMESPACE);
6797         if (com->lc_data == NULL)
6798                 GOTO(out, rc = -ENOMEM);
6799
6800         com->lc_file_size = sizeof(struct lfsck_namespace);
6801         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6802         if (com->lc_file_ram == NULL)
6803                 GOTO(out, rc = -ENOMEM);
6804
6805         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6806         if (com->lc_file_disk == NULL)
6807                 GOTO(out, rc = -ENOMEM);
6808
6809         for (i = 0; i < LFSCK_STF_COUNT; i++)
6810                 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6811
6812         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6813         if (IS_ERR(root))
6814                 GOTO(out, rc = PTR_ERR(root));
6815
6816         if (unlikely(!dt_try_as_dir(env, root)))
6817                 GOTO(out, rc = -ENOTDIR);
6818
6819         obj = local_index_find_or_create(env, lfsck->li_los, root,
6820                                          LFSCK_NAMESPACE,
6821                                          S_IFREG | S_IRUGO | S_IWUSR,
6822                                          &dt_lfsck_namespace_features);
6823         if (IS_ERR(obj))
6824                 GOTO(out, rc = PTR_ERR(obj));
6825
6826         com->lc_obj = obj;
6827         rc = lfsck_namespace_load(env, com);
6828         if (rc == -ENODATA)
6829                 rc = lfsck_namespace_init(env, com);
6830         else if (rc < 0)
6831                 rc = lfsck_namespace_reset(env, com, true);
6832         else
6833                 rc = lfsck_load_sub_trace_files(env, com,
6834                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
6835         if (rc != 0)
6836                 GOTO(out, rc);
6837
6838         ns = com->lc_file_ram;
6839         switch (ns->ln_status) {
6840         case LS_INIT:
6841         case LS_COMPLETED:
6842         case LS_FAILED:
6843         case LS_STOPPED:
6844                 spin_lock(&lfsck->li_lock);
6845                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6846                 spin_unlock(&lfsck->li_lock);
6847                 break;
6848         default:
6849                 CERROR("%s: unknown lfsck_namespace status %d\n",
6850                        lfsck_lfsck2name(lfsck), ns->ln_status);
6851                 /* fall through */
6852         case LS_SCANNING_PHASE1:
6853         case LS_SCANNING_PHASE2:
6854                 /* No need to store the status to disk right now.
6855                  * If the system crashed before the status stored,
6856                  * it will be loaded back when next time. */
6857                 ns->ln_status = LS_CRASHED;
6858                 /* fall through */
6859         case LS_PAUSED:
6860         case LS_CRASHED:
6861                 spin_lock(&lfsck->li_lock);
6862                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6863                 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6864                 spin_unlock(&lfsck->li_lock);
6865                 break;
6866         }
6867
6868         GOTO(out, rc = 0);
6869
6870 out:
6871         if (root != NULL && !IS_ERR(root))
6872                 lfsck_object_put(env, root);
6873         if (rc != 0) {
6874                 lfsck_component_cleanup(env, com);
6875                 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6876                        lfsck_lfsck2name(lfsck), rc);
6877         }
6878         return rc;
6879 }