lustre/lfsck/lfsck_namespace.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2016, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_namespace.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lustre/lustre_idl.h>
  34 #include <lu_object.h>
  35 #include <dt_object.h>
  36 #include <md_object.h>
  37 #include <lustre_fid.h>
  38 #include <lustre_lib.h>
  39 #include <lustre_net.h>
  40 #include <lustre/lustre_user.h>
  41
  42 #include "lfsck_internal.h"
  43
  44 #define LFSCK_NAMESPACE_MAGIC_V1        0xA0629D03
  45 #define LFSCK_NAMESPACE_MAGIC_V2        0xA0621A0B
  46
  47 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
  48  * as the trace file magic. When downgrade to such old release, the old LFSCK
  49  * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
  50  * then it will reset the whole LFSCK, and will not cause start failure. The
  51  * similar case will happen when upgrade from such old release. */
  52 #define LFSCK_NAMESPACE_MAGIC           LFSCK_NAMESPACE_MAGIC_V2
  53
  54 enum lfsck_nameentry_check {
  55         LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
  56         LFSCK_NAMEENTRY_REMOVED         = 2, /* The entry has been removed. */
  57         LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
  58 };
  59
  60 static struct lfsck_namespace_req *
  61 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
  62                                    struct lfsck_assistant_object *lso,
  63                                    struct lu_dirent *ent, __u16 type)
  64 {
  65         struct lfsck_namespace_req *lnr;
  66         int                         size;
  67
  68         size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
  69         OBD_ALLOC(lnr, size);
  70         if (lnr == NULL)
  71                 return ERR_PTR(-ENOMEM);
  72
  73         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
  74         lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
  75         lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
  76         lnr->lnr_fid = ent->lde_fid;
  77         lnr->lnr_dir_cookie = ent->lde_hash;
  78         lnr->lnr_attr = ent->lde_attrs;
  79         lnr->lnr_size = size;
  80         lnr->lnr_type = type;
  81         lnr->lnr_namelen = ent->lde_namelen;
  82         memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
  83
  84         return lnr;
  85 }
  86
  87 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
  88                                                struct lfsck_assistant_req *lar)
  89 {
  90         struct lfsck_namespace_req *lnr =
  91                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
  92
  93         if (lnr->lnr_lmv != NULL)
  94                 lfsck_lmv_put(env, lnr->lnr_lmv);
  95
  96         lfsck_assistant_object_put(env, lar->lar_parent);
  97         OBD_FREE(lnr, lnr->lnr_size);
  98 }
  99
 100 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
 101                                       struct lfsck_namespace *src)
 102 {
 103         dst->ln_magic = le32_to_cpu(src->ln_magic);
 104         dst->ln_status = le32_to_cpu(src->ln_status);
 105         dst->ln_flags = le32_to_cpu(src->ln_flags);
 106         dst->ln_success_count = le32_to_cpu(src->ln_success_count);
 107         dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
 108         dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
 109         dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
 110         dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
 111         dst->ln_time_last_checkpoint =
 112                                 le64_to_cpu(src->ln_time_last_checkpoint);
 113         lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
 114                                  &src->ln_pos_latest_start);
 115         lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
 116                                  &src->ln_pos_last_checkpoint);
 117         lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
 118                                  &src->ln_pos_first_inconsistent);
 119         dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
 120         dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
 121         dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
 122         dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
 123         dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
 124         dst->ln_objs_repaired_phase2 =
 125                                 le64_to_cpu(src->ln_objs_repaired_phase2);
 126         dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
 127         dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
 128         fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
 129                       &src->ln_fid_latest_scanned_phase2);
 130         dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
 131         dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
 132         dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
 133         dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
 134         dst->ln_unknown_inconsistency =
 135                                 le64_to_cpu(src->ln_unknown_inconsistency);
 136         dst->ln_unmatched_pairs_repaired =
 137                                 le64_to_cpu(src->ln_unmatched_pairs_repaired);
 138         dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
 139         dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
 140         dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
 141         dst->ln_lost_dirent_repaired =
 142                                 le64_to_cpu(src->ln_lost_dirent_repaired);
 143         dst->ln_striped_dirs_scanned =
 144                                 le64_to_cpu(src->ln_striped_dirs_scanned);
 145         dst->ln_striped_dirs_repaired =
 146                                 le64_to_cpu(src->ln_striped_dirs_repaired);
 147         dst->ln_striped_dirs_failed =
 148                                 le64_to_cpu(src->ln_striped_dirs_failed);
 149         dst->ln_striped_dirs_disabled =
 150                                 le64_to_cpu(src->ln_striped_dirs_disabled);
 151         dst->ln_striped_dirs_skipped =
 152                                 le64_to_cpu(src->ln_striped_dirs_skipped);
 153         dst->ln_striped_shards_scanned =
 154                                 le64_to_cpu(src->ln_striped_shards_scanned);
 155         dst->ln_striped_shards_repaired =
 156                                 le64_to_cpu(src->ln_striped_shards_repaired);
 157         dst->ln_striped_shards_failed =
 158                                 le64_to_cpu(src->ln_striped_shards_failed);
 159         dst->ln_striped_shards_skipped =
 160                                 le64_to_cpu(src->ln_striped_shards_skipped);
 161         dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
 162         dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
 163         dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
 164         dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
 165         dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
 166         dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
 167         dst->ln_time_latest_reset = le32_to_cpu(src->ln_time_latest_reset);
 168         dst->ln_linkea_overflow_cleared =
 169                                 le64_to_cpu(src->ln_linkea_overflow_cleared);
 170 }
 171
 172 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
 173                                       struct lfsck_namespace *src)
 174 {
 175         dst->ln_magic = cpu_to_le32(src->ln_magic);
 176         dst->ln_status = cpu_to_le32(src->ln_status);
 177         dst->ln_flags = cpu_to_le32(src->ln_flags);
 178         dst->ln_success_count = cpu_to_le32(src->ln_success_count);
 179         dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
 180         dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
 181         dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
 182         dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
 183         dst->ln_time_last_checkpoint =
 184                                 cpu_to_le64(src->ln_time_last_checkpoint);
 185         lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
 186                                  &src->ln_pos_latest_start);
 187         lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
 188                                  &src->ln_pos_last_checkpoint);
 189         lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
 190                                  &src->ln_pos_first_inconsistent);
 191         dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
 192         dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
 193         dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
 194         dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
 195         dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
 196         dst->ln_objs_repaired_phase2 =
 197                                 cpu_to_le64(src->ln_objs_repaired_phase2);
 198         dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
 199         dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
 200         fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
 201                       &src->ln_fid_latest_scanned_phase2);
 202         dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
 203         dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
 204         dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
 205         dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
 206         dst->ln_unknown_inconsistency =
 207                                 cpu_to_le64(src->ln_unknown_inconsistency);
 208         dst->ln_unmatched_pairs_repaired =
 209                                 cpu_to_le64(src->ln_unmatched_pairs_repaired);
 210         dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
 211         dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
 212         dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
 213         dst->ln_lost_dirent_repaired =
 214                                 cpu_to_le64(src->ln_lost_dirent_repaired);
 215         dst->ln_striped_dirs_scanned =
 216                                 cpu_to_le64(src->ln_striped_dirs_scanned);
 217         dst->ln_striped_dirs_repaired =
 218                                 cpu_to_le64(src->ln_striped_dirs_repaired);
 219         dst->ln_striped_dirs_failed =
 220                                 cpu_to_le64(src->ln_striped_dirs_failed);
 221         dst->ln_striped_dirs_disabled =
 222                                 cpu_to_le64(src->ln_striped_dirs_disabled);
 223         dst->ln_striped_dirs_skipped =
 224                                 cpu_to_le64(src->ln_striped_dirs_skipped);
 225         dst->ln_striped_shards_scanned =
 226                                 cpu_to_le64(src->ln_striped_shards_scanned);
 227         dst->ln_striped_shards_repaired =
 228                                 cpu_to_le64(src->ln_striped_shards_repaired);
 229         dst->ln_striped_shards_failed =
 230                                 cpu_to_le64(src->ln_striped_shards_failed);
 231         dst->ln_striped_shards_skipped =
 232                                 cpu_to_le64(src->ln_striped_shards_skipped);
 233         dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
 234         dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
 235         dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
 236         dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
 237         dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
 238         dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
 239         dst->ln_time_latest_reset = cpu_to_le32(src->ln_time_latest_reset);
 240         dst->ln_linkea_overflow_cleared =
 241                                 cpu_to_le64(src->ln_linkea_overflow_cleared);
 242 }
 243
 244 static void lfsck_namespace_record_failure(const struct lu_env *env,
 245                                            struct lfsck_instance *lfsck,
 246                                            struct lfsck_namespace *ns)
 247 {
 248         struct lfsck_position pos;
 249
 250         ns->ln_items_failed++;
 251         lfsck_pos_fill(env, lfsck, &pos, false);
 252         if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
 253             lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
 254                 ns->ln_pos_first_inconsistent = pos;
 255
 256                 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
 257                        "inconsistency at the pos [%llu, "DFID", %#llx]\n",
 258                        lfsck_lfsck2name(lfsck),
 259                        ns->ln_pos_first_inconsistent.lp_oit_cookie,
 260                        PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
 261                        ns->ln_pos_first_inconsistent.lp_dir_cookie);
 262         }
 263 }
 264
 265 /**
 266  * Load the MDT bitmap from the lfsck_namespace trace file.
 267  *
 268  * \param[in] env       pointer to the thread context
 269  * \param[in] com       pointer to the lfsck component
 270  *
 271  * \retval              0 for success
 272  * \retval              negative error number on failure or data corruption
 273  */
 274 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
 275                                        struct lfsck_component *com)
 276 {
 277         struct dt_object                *obj    = com->lc_obj;
 278         struct lfsck_assistant_data     *lad    = com->lc_data;
 279         struct lfsck_namespace          *ns     = com->lc_file_ram;
 280         struct cfs_bitmap                       *bitmap = lad->lad_bitmap;
 281         ssize_t                          size;
 282         __u32                            nbits;
 283         int                              rc;
 284         ENTRY;
 285
 286         if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
 287             ns->ln_bitmap_size)
 288                 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
 289         else
 290                 nbits = ns->ln_bitmap_size;
 291
 292         if (unlikely(nbits < BITS_PER_LONG))
 293                 nbits = BITS_PER_LONG;
 294
 295         if (nbits > bitmap->size) {
 296                 __u32 new_bits = bitmap->size;
 297                 struct cfs_bitmap *new_bitmap;
 298
 299                 while (new_bits < nbits)
 300                         new_bits <<= 1;
 301
 302                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 303                 if (new_bitmap == NULL)
 304                         RETURN(-ENOMEM);
 305
 306                 lad->lad_bitmap = new_bitmap;
 307                 CFS_FREE_BITMAP(bitmap);
 308                 bitmap = new_bitmap;
 309         }
 310
 311         if (ns->ln_bitmap_size == 0) {
 312                 lad->lad_incomplete = 0;
 313                 CFS_RESET_BITMAP(bitmap);
 314
 315                 RETURN(0);
 316         }
 317
 318         size = (ns->ln_bitmap_size + 7) >> 3;
 319         rc = dt_xattr_get(env, obj,
 320                           lfsck_buf_get(env, bitmap->data, size),
 321                           XATTR_NAME_LFSCK_BITMAP);
 322         if (rc != size)
 323                 RETURN(rc >= 0 ? -EINVAL : rc);
 324
 325         if (cfs_bitmap_check_empty(bitmap))
 326                 lad->lad_incomplete = 0;
 327         else
 328                 lad->lad_incomplete = 1;
 329
 330         RETURN(0);
 331 }
 332
 333 /**
 334  * Load namespace LFSCK statistics information from the trace file.
 335  *
 336  * \param[in] env       pointer to the thread context
 337  * \param[in] com       pointer to the lfsck component
 338  *
 339  * \retval              0 for success
 340  * \retval              negative error number on failure
 341  */
 342 static int lfsck_namespace_load(const struct lu_env *env,
 343                                 struct lfsck_component *com)
 344 {
 345         int len = com->lc_file_size;
 346         int rc;
 347
 348         rc = dt_xattr_get(env, com->lc_obj,
 349                           lfsck_buf_get(env, com->lc_file_disk, len),
 350                           XATTR_NAME_LFSCK_NAMESPACE);
 351         if (rc == len) {
 352                 struct lfsck_namespace *ns = com->lc_file_ram;
 353
 354                 lfsck_namespace_le_to_cpu(ns,
 355                                 (struct lfsck_namespace *)com->lc_file_disk);
 356                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
 357                         CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
 358                                "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
 359                                ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
 360                         rc = -ESTALE;
 361                 } else {
 362                         rc = 0;
 363                 }
 364         } else if (rc != -ENODATA) {
 365                 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
 366                        "expected = %d: rc = %d\n",
 367                        lfsck_lfsck2name(com->lc_lfsck), len, rc);
 368                 if (rc >= 0)
 369                         rc = -ESTALE;
 370         }
 371
 372         return rc;
 373 }
 374
 375 static int lfsck_namespace_store(const struct lu_env *env,
 376                                  struct lfsck_component *com)
 377 {
 378         struct dt_object                *obj    = com->lc_obj;
 379         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 380         struct lfsck_namespace          *ns     = com->lc_file_ram;
 381         struct lfsck_assistant_data     *lad    = com->lc_data;
 382         struct dt_device                *dev    = lfsck_obj2dev(obj);
 383         struct cfs_bitmap               *bitmap = NULL;
 384         struct thandle                  *handle;
 385         __u32                            nbits  = 0;
 386         int                              len    = com->lc_file_size;
 387         int                              rc;
 388         ENTRY;
 389
 390         if (lad != NULL) {
 391                 bitmap = lad->lad_bitmap;
 392                 nbits = bitmap->size;
 393
 394                 LASSERT(nbits > 0);
 395                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 396         }
 397
 398         ns->ln_bitmap_size = nbits;
 399         lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
 400                                   ns);
 401         handle = dt_trans_create(env, dev);
 402         if (IS_ERR(handle))
 403                 GOTO(log, rc = PTR_ERR(handle));
 404
 405         rc = dt_declare_xattr_set(env, obj,
 406                                   lfsck_buf_get(env, com->lc_file_disk, len),
 407                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 408         if (rc != 0)
 409                 GOTO(out, rc);
 410
 411         if (bitmap != NULL) {
 412                 rc = dt_declare_xattr_set(env, obj,
 413                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 414                                 XATTR_NAME_LFSCK_BITMAP, 0, handle);
 415                 if (rc != 0)
 416                         GOTO(out, rc);
 417         }
 418
 419         rc = dt_trans_start_local(env, dev, handle);
 420         if (rc != 0)
 421                 GOTO(out, rc);
 422
 423         rc = dt_xattr_set(env, obj,
 424                           lfsck_buf_get(env, com->lc_file_disk, len),
 425                           XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 426         if (rc == 0 && bitmap != NULL)
 427                 rc = dt_xattr_set(env, obj,
 428                                   lfsck_buf_get(env, bitmap->data, nbits >> 3),
 429                                   XATTR_NAME_LFSCK_BITMAP, 0, handle);
 430
 431         GOTO(out, rc);
 432
 433 out:
 434         dt_trans_stop(env, dev, handle);
 435
 436 log:
 437         if (rc != 0)
 438                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
 439                        lfsck_lfsck2name(lfsck), rc);
 440         return rc;
 441 }
 442
 443 static int lfsck_namespace_init(const struct lu_env *env,
 444                                 struct lfsck_component *com)
 445 {
 446         struct lfsck_namespace *ns = com->lc_file_ram;
 447         int rc;
 448
 449         memset(ns, 0, sizeof(*ns));
 450         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
 451         ns->ln_status = LS_INIT;
 452         ns->ln_time_latest_reset = cfs_time_current_sec();
 453         down_write(&com->lc_sem);
 454         rc = lfsck_namespace_store(env, com);
 455         if (rc == 0)
 456                 rc = lfsck_load_sub_trace_files(env, com,
 457                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
 458         up_write(&com->lc_sem);
 459
 460         return rc;
 461 }
 462
 463 /**
 464  * Update the namespace LFSCK trace file for the given @fid
 465  *
 466  * \param[in] env       pointer to the thread context
 467  * \param[in] com       pointer to the lfsck component
 468  * \param[in] fid       the fid which flags to be updated in the lfsck
 469  *                      trace file
 470  * \param[in] add       true if add new flags, otherwise remove flags
 471  *
 472  * \retval              0 for success or nothing to be done
 473  * \retval              negative error number on failure
 474  */
 475 int lfsck_namespace_trace_update(const struct lu_env *env,
 476                                  struct lfsck_component *com,
 477                                  const struct lu_fid *fid,
 478                                  const __u8 flags, bool add)
 479 {
 480         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 481         struct dt_object        *obj;
 482         struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
 483         struct dt_device        *dev;
 484         struct thandle          *th     = NULL;
 485         int                      idx;
 486         int                      rc     = 0;
 487         __u8                     old    = 0;
 488         __u8                     new    = 0;
 489         ENTRY;
 490
 491         LASSERT(flags != 0);
 492
 493         if (unlikely(!fid_is_sane(fid)))
 494                 RETURN(0);
 495
 496         idx = lfsck_sub_trace_file_fid2idx(fid);
 497         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 498         obj = com->lc_sub_trace_objs[idx].lsto_obj;
 499         if (unlikely(obj == NULL)) {
 500                 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 501                 RETURN(0);
 502         }
 503
 504         lfsck_object_get(obj);
 505         dev = lfsck_obj2dev(obj);
 506         fid_cpu_to_be(key, fid);
 507         rc = dt_lookup(env, obj, (struct dt_rec *)&old,
 508                        (const struct dt_key *)key);
 509         if (rc == -ENOENT) {
 510                 if (!add)
 511                         GOTO(unlock, rc = 0);
 512
 513                 old = 0;
 514                 new = flags;
 515         } else if (rc == 0) {
 516                 if (add) {
 517                         if ((old & flags) == flags)
 518                                 GOTO(unlock, rc = 0);
 519
 520                         new = old | flags;
 521                 } else {
 522                         if ((old & flags) == 0)
 523                                 GOTO(unlock, rc = 0);
 524
 525                         new = old & ~flags;
 526                 }
 527         } else {
 528                 GOTO(log, rc);
 529         }
 530
 531         th = dt_trans_create(env, dev);
 532         if (IS_ERR(th))
 533                 GOTO(log, rc = PTR_ERR(th));
 534
 535         if (old != 0) {
 536                 rc = dt_declare_delete(env, obj,
 537                                        (const struct dt_key *)key, th);
 538                 if (rc != 0)
 539                         GOTO(log, rc);
 540         }
 541
 542         if (new != 0) {
 543                 rc = dt_declare_insert(env, obj,
 544                                        (const struct dt_rec *)&new,
 545                                        (const struct dt_key *)key, th);
 546                 if (rc != 0)
 547                         GOTO(log, rc);
 548         }
 549
 550         rc = dt_trans_start_local(env, dev, th);
 551         if (rc != 0)
 552                 GOTO(log, rc);
 553
 554         if (old != 0) {
 555                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
 556                 if (rc != 0)
 557                         GOTO(log, rc);
 558         }
 559
 560         if (new != 0) {
 561                 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
 562                                (const struct dt_key *)key, th, 1);
 563                 if (rc != 0)
 564                         GOTO(log, rc);
 565         }
 566
 567         GOTO(log, rc);
 568
 569 log:
 570         if (th != NULL && !IS_ERR(th))
 571                 dt_trans_stop(env, dev, th);
 572
 573         CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
 574                "trace file, flags %x, old %x, new %x: rc = %d\n",
 575                lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
 576                (__u32)flags, (__u32)old, (__u32)new, rc);
 577
 578 unlock:
 579         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 580         lfsck_object_put(env, obj);
 581
 582         return rc;
 583 }
 584
 585 int lfsck_namespace_check_exist(const struct lu_env *env,
 586                                 struct dt_object *dir,
 587                                 struct dt_object *obj, const char *name)
 588 {
 589         struct lu_fid    *fid = &lfsck_env_info(env)->lti_fid;
 590         int               rc;
 591         ENTRY;
 592
 593         if (unlikely(lfsck_is_dead_obj(obj)))
 594                 RETURN(LFSCK_NAMEENTRY_DEAD);
 595
 596         rc = dt_lookup(env, dir, (struct dt_rec *)fid,
 597                        (const struct dt_key *)name);
 598         if (rc == -ENOENT)
 599                 RETURN(LFSCK_NAMEENTRY_REMOVED);
 600
 601         if (rc < 0)
 602                 RETURN(rc);
 603
 604         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
 605                 RETURN(LFSCK_NAMEENTRY_RECREATED);
 606
 607         RETURN(0);
 608 }
 609
 610 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
 611                                             struct dt_object *obj,
 612                                             struct thandle *handle)
 613 {
 614         int rc;
 615
 616         /* For remote updating LINKEA, there may be further LFSCK action
 617          * on remote MDT after the updating, so update the LINKEA ASAP. */
 618         if (dt_object_remote(obj))
 619                 handle->th_sync = 1;
 620
 621         /* For destroying all invalid linkEA entries. */
 622         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
 623         if (rc == 0)
 624                 /* For insert new linkEA entry. */
 625                 rc = dt_declare_xattr_set(env, obj,
 626                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
 627                         XATTR_NAME_LINK, 0, handle);
 628         return rc;
 629 }
 630
 631 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
 632                        struct linkea_data *ldata, bool with_rec)
 633 {
 634         int rc;
 635
 636         if (ldata->ld_buf->lb_buf == NULL)
 637                 return -ENOMEM;
 638
 639         if (!dt_object_exists(obj))
 640                 return -ENOENT;
 641
 642         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
 643         if (rc == -ERANGE) {
 644                 /* Buf was too small, figure out what we need. */
 645                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
 646                 if (unlikely(rc == 0))
 647                         return -ENODATA;
 648
 649                 if (rc < 0)
 650                         return rc;
 651
 652                 lu_buf_realloc(ldata->ld_buf, rc);
 653                 if (ldata->ld_buf->lb_buf == NULL)
 654                         return -ENOMEM;
 655
 656                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
 657         }
 658
 659         if (unlikely(rc == 0))
 660                 return -ENODATA;
 661
 662         if (rc > 0) {
 663                 if (with_rec)
 664                         rc = linkea_init_with_rec(ldata);
 665                 else
 666                         rc = linkea_init(ldata);
 667         }
 668
 669         return rc;
 670 }
 671
 672 /**
 673  * Remove linkEA for the given object.
 674  *
 675  * The caller should take the ldlm lock before the calling.
 676  *
 677  * \param[in] env       pointer to the thread context
 678  * \param[in] com       pointer to the lfsck component
 679  * \param[in] obj       pointer to the dt_object to be handled
 680  *
 681  * \retval              0 for repaired cases
 682  * \retval              negative error number on failure
 683  */
 684 static int lfsck_namespace_links_remove(const struct lu_env *env,
 685                                         struct lfsck_component *com,
 686                                         struct dt_object *obj)
 687 {
 688         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 689         struct dt_device                *dev    = lfsck_obj2dev(obj);
 690         struct thandle                  *th     = NULL;
 691         int                              rc     = 0;
 692         ENTRY;
 693
 694         LASSERT(dt_object_remote(obj) == 0);
 695
 696         th = dt_trans_create(env, dev);
 697         if (IS_ERR(th))
 698                 GOTO(log, rc = PTR_ERR(th));
 699
 700         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
 701         if (rc != 0)
 702                 GOTO(stop, rc);
 703
 704         rc = dt_trans_start_local(env, dev, th);
 705         if (rc != 0)
 706                 GOTO(stop, rc);
 707
 708         dt_write_lock(env, obj, 0);
 709         if (unlikely(lfsck_is_dead_obj(obj)))
 710                 GOTO(unlock, rc = -ENOENT);
 711
 712         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 713                 GOTO(unlock, rc = 0);
 714
 715         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
 716
 717         GOTO(unlock, rc);
 718
 719 unlock:
 720         dt_write_unlock(env, obj);
 721
 722 stop:
 723         dt_trans_stop(env, dev, th);
 724
 725 log:
 726         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
 727                "for the object "DFID": rc = %d\n",
 728                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
 729
 730         if (rc == 0) {
 731                 struct lfsck_namespace *ns = com->lc_file_ram;
 732
 733                 ns->ln_flags |= LF_INCONSISTENT;
 734         }
 735
 736         return rc;
 737 }
 738
 739 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
 740                              struct linkea_data *ldata, struct thandle *handle)
 741 {
 742         struct lu_buf buf;
 743         int rc;
 744
 745         lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
 746
 747 again:
 748         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
 749         if (unlikely(rc == -ENOSPC)) {
 750                 rc = linkea_overflow_shrink(ldata);
 751                 if (likely(rc > 0)) {
 752                         buf.lb_len = rc;
 753                         goto again;
 754                 }
 755         }
 756
 757         return rc;
 758 }
 759
 760 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
 761                                                struct lu_name *cname,
 762                                                struct lu_fid *pfid,
 763                                                char *buf, const int buflen)
 764 {
 765         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
 766         if (unlikely(ldata->ld_reclen <= 0 ||
 767                      ldata->ld_reclen + sizeof(struct link_ea_header) >
 768                         ldata->ld_leh->leh_len ||
 769                      cname->ln_namelen <= 0 ||
 770                      cname->ln_namelen > NAME_MAX ||
 771                      cname->ln_namelen >= buflen ||
 772                      !fid_is_sane(pfid)))
 773                 return -EINVAL;
 774
 775         /* To guarantee the 'name' is terminated with '0'. */
 776         memcpy(buf, cname->ln_name, cname->ln_namelen);
 777         buf[cname->ln_namelen] = 0;
 778         cname->ln_name = buf;
 779
 780         return 0;
 781 }
 782
 783 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
 784                                  const struct lu_name *lname)
 785 {
 786         LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
 787
 788         /* If current record is corrupted, all the subsequent
 789          * records will be dropped. */
 790         if (unlikely(ldata->ld_reclen <= 0 ||
 791                      ldata->ld_reclen + sizeof(struct link_ea_header) >
 792                         ldata->ld_leh->leh_len)) {
 793                 void *ptr = ldata->ld_lee;
 794
 795                 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
 796                 ldata->ld_leh->leh_reccount = 0;
 797                 linkea_first_entry(ldata);
 798                 while (ldata->ld_lee != NULL &&
 799                        (char *)ldata->ld_lee < (char *)ptr) {
 800                         int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
 801                                      ldata->ld_lee->lee_reclen[1];
 802
 803                         ldata->ld_leh->leh_len += reclen;
 804                         ldata->ld_leh->leh_reccount++;
 805                         ldata->ld_lee = (struct link_ea_entry *)
 806                                         ((char *)ldata->ld_lee + reclen);
 807                 }
 808
 809                 ldata->ld_lee = NULL;
 810         } else {
 811                 linkea_del_buf(ldata, lname);
 812         }
 813 }
 814
 815 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
 816                                                struct lu_name *cname,
 817                                                struct lu_fid *pfid,
 818                                                bool remove)
 819 {
 820         struct link_ea_entry    *oldlee;
 821         int                      oldlen;
 822         int                      repeated = 0;
 823
 824         oldlee = ldata->ld_lee;
 825         oldlen = ldata->ld_reclen;
 826         linkea_next_entry(ldata);
 827         while (ldata->ld_lee != NULL) {
 828                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
 829                                    ldata->ld_lee->lee_reclen[1];
 830                 if (unlikely(ldata->ld_reclen == oldlen &&
 831                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
 832                         repeated++;
 833                         if (!remove)
 834                                 break;
 835
 836                         lfsck_linkea_del_buf(ldata, cname);
 837                 } else {
 838                         linkea_next_entry(ldata);
 839                 }
 840         }
 841         ldata->ld_lee = oldlee;
 842         ldata->ld_reclen = oldlen;
 843
 844         return repeated;
 845 }
 846
 847 /**
 848  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
 849  *
 850  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
 851  * with the given type to generate the name, the detailed rules for name
 852  * have been described as following.
 853  *
 854  * The function also generates the linkEA corresponding to the name entry
 855  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
 856  *
 857  * \param[in] env       pointer to the thread context
 858  * \param[in] com       pointer to the lfsck component
 859  * \param[in] orphan    pointer to the orphan MDT-object
 860  * \param[in] infix     additional information for the orphan name, such as
 861  *                      the FID for original
 862  * \param[in] type      the type for describing why the orphan MDT-object is
 863  *                      created. The rules are as following:
 864  *
 865  *  type "D":           The MDT-object is a directory, it may knows its parent
 866  *                      but because there is no valid linkEA, the LFSCK cannot
 867  *                      know where to put it back to the namespace.
 868  *  type "O":           The MDT-object has no linkEA, and there is no name
 869  *                      entry that references the MDT-object.
 870  *
 871  *  type "S":           The orphan MDT-object is a shard of a striped directory
 872  *
 873  * \see lfsck_layout_recreate_parent() for more types.
 874  *
 875  * The orphan name will be like:
 876  * ${FID}-${infix}-${type}-${conflict_version}
 877  *
 878  * \param[out] count    if some others inserted some linkEA entries by race,
 879  *                      then return the linkEA entries count.
 880  *
 881  * \retval              positive number for repaired cases
 882  * \retval              0 if needs to repair nothing
 883  * \retval              negative error number on failure
 884  */
 885 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
 886                                          struct lfsck_component *com,
 887                                          struct dt_object *orphan,
 888                                          const char *infix, const char *type,
 889                                          int *count)
 890 {
 891         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 892         struct lu_name                  *cname  = &info->lti_name;
 893         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
 894         struct lu_attr                  *la     = &info->lti_la2;
 895         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
 896         const struct lu_fid             *pfid;
 897         struct lu_fid                    tfid;
 898         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 899         struct dt_device                *dev    = lfsck_obj2dev(orphan);
 900         struct dt_object                *parent;
 901         struct thandle                  *th     = NULL;
 902         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
 903         struct lustre_handle             clh    = { 0 };
 904         struct linkea_data               ldata2 = { NULL };
 905         struct lu_buf                    linkea_buf;
 906         int                              namelen;
 907         int                              idx    = 0;
 908         int                              rc     = 0;
 909         bool                             exist  = false;
 910         ENTRY;
 911
 912         cname->ln_name = NULL;
 913         if (unlikely(lfsck->li_lpf_obj == NULL))
 914                 GOTO(log, rc = -ENXIO);
 915
 916         parent = lfsck->li_lpf_obj;
 917         pfid = lfsck_dto2fid(parent);
 918
 919 again:
 920         do {
 921                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
 922                                    PFID(cfid), infix, type, idx++);
 923                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
 924                                (const struct dt_key *)info->lti_key);
 925                 if (rc != 0 && rc != -ENOENT)
 926                         GOTO(log, rc);
 927
 928                 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
 929                         exist = true;
 930         } while (rc == 0 && !exist);
 931
 932         rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
 933                         MDS_INODELOCK_UPDATE, LCK_PW);
 934         if (rc != 0)
 935                 GOTO(log, rc);
 936
 937         /* Re-check whether the name conflict with othrs after taken
 938          * the ldlm lock. */
 939         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
 940                        (const struct dt_key *)info->lti_key);
 941         if (rc == 0) {
 942                 if (!lu_fid_eq(cfid, &tfid)) {
 943                         exist = false;
 944                         lfsck_unlock(pllh);
 945                         goto again;
 946                 }
 947
 948                 exist = true;
 949         } else if (rc != -ENOENT) {
 950                 GOTO(log, rc);
 951         } else {
 952                 exist = false;
 953         }
 954
 955         cname->ln_name = info->lti_key;
 956         cname->ln_namelen = namelen;
 957         rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
 958                               cname, pfid);
 959         if (rc != 0)
 960                 GOTO(log, rc);
 961
 962         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
 963                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
 964                               MDS_INODELOCK_XATTR, LCK_EX);
 965         if (rc != 0)
 966                 GOTO(log, rc);
 967
 968         lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
 969                        ldata2.ld_leh->leh_len);
 970         th = dt_trans_create(env, dev);
 971         if (IS_ERR(th))
 972                 GOTO(log, rc = PTR_ERR(th));
 973
 974         if (S_ISDIR(lfsck_object_type(orphan))) {
 975                 rc = dt_declare_delete(env, orphan,
 976                                        (const struct dt_key *)dotdot, th);
 977                 if (rc != 0)
 978                         GOTO(stop, rc);
 979
 980                 rec->rec_type = S_IFDIR;
 981                 rec->rec_fid = pfid;
 982                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
 983                                        (const struct dt_key *)dotdot, th);
 984                 if (rc != 0)
 985                         GOTO(stop, rc);
 986         }
 987
 988         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
 989                                   XATTR_NAME_LINK, 0, th);
 990         if (rc != 0)
 991                 GOTO(stop, rc);
 992
 993         if (!exist) {
 994                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
 995                 rec->rec_fid = cfid;
 996                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 997                                        (const struct dt_key *)cname->ln_name,
 998                                        th);
 999                 if (rc != 0)
1000                         GOTO(stop, rc);
1001
1002                 if (S_ISDIR(rec->rec_type)) {
1003                         rc = dt_declare_ref_add(env, parent, th);
1004                         if (rc != 0)
1005                                 GOTO(stop, rc);
1006                 }
1007         }
1008
1009         memset(la, 0, sizeof(*la));
1010         la->la_ctime = cfs_time_current_sec();
1011         la->la_valid = LA_CTIME;
1012         rc = dt_declare_attr_set(env, orphan, la, th);
1013         if (rc != 0)
1014                 GOTO(stop, rc);
1015
1016         rc = dt_trans_start_local(env, dev, th);
1017         if (rc != 0)
1018                 GOTO(stop, rc);
1019
1020         dt_write_lock(env, orphan, 0);
1021         rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1022         if (likely(rc == -ENODATA || rc == -EINVAL)) {
1023                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1024                         GOTO(unlock, rc = 1);
1025
1026                 if (S_ISDIR(lfsck_object_type(orphan))) {
1027                         rc = dt_delete(env, orphan,
1028                                        (const struct dt_key *)dotdot, th);
1029                         if (rc != 0)
1030                                 GOTO(unlock, rc);
1031
1032                         rec->rec_type = S_IFDIR;
1033                         rec->rec_fid = pfid;
1034                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1035                                        (const struct dt_key *)dotdot, th, 1);
1036                         if (rc != 0)
1037                                 GOTO(unlock, rc);
1038                 }
1039
1040                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1041                                   th);
1042         } else {
1043                 if (rc == 0 && count != NULL)
1044                         *count = ldata2.ld_leh->leh_reccount;
1045
1046                 GOTO(unlock, rc);
1047         }
1048         dt_write_unlock(env, orphan);
1049
1050         if (rc == 0 && !exist) {
1051                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1052                 rec->rec_fid = cfid;
1053                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1054                                (const struct dt_key *)cname->ln_name, th, 1);
1055                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1056                         dt_write_lock(env, parent, 0);
1057                         rc = dt_ref_add(env, parent, th);
1058                         dt_write_unlock(env, parent);
1059                 }
1060         }
1061
1062         if (rc == 0)
1063                 rc = dt_attr_set(env, orphan, la, th);
1064
1065         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1066
1067 unlock:
1068         dt_write_unlock(env, orphan);
1069
1070 stop:
1071         dt_trans_stop(env, dev, th);
1072
1073 log:
1074         lfsck_ibits_unlock(&clh, LCK_EX);
1075         lfsck_unlock(pllh);
1076         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1077                "object "DFID", name = %s: rc = %d\n",
1078                lfsck_lfsck2name(lfsck), PFID(cfid),
1079                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1080
1081         if (rc != 0) {
1082                 struct lfsck_namespace *ns = com->lc_file_ram;
1083
1084                 ns->ln_flags |= LF_INCONSISTENT;
1085         }
1086
1087         return rc;
1088 }
1089
1090 /**
1091  * Add the specified name entry back to namespace.
1092  *
1093  * If there is a linkEA entry that back references a name entry under
1094  * some parent directory, but such parent directory does not have the
1095  * claimed name entry. On the other hand, the linkEA entries count is
1096  * not larger than the MDT-object's hard link count. Under such case,
1097  * it is quite possible that the name entry is lost. Then the LFSCK
1098  * should add the name entry back to the namespace.
1099  *
1100  * \param[in] env       pointer to the thread context
1101  * \param[in] com       pointer to the lfsck component
1102  * \param[in] parent    pointer to the directory under which the name entry
1103  *                      will be inserted into
1104  * \param[in] child     pointer to the object referenced by the name entry
1105  *                      that to be inserted into the parent
1106  * \param[in] name      the name for the child in the parent directory
1107  *
1108  * \retval              positive number for repaired cases
1109  * \retval              0 if nothing to be repaired
1110  * \retval              negative error number on failure
1111  */
1112 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1113                                          struct lfsck_component *com,
1114                                          struct dt_object *parent,
1115                                          struct dt_object *child,
1116                                          const char *name)
1117 {
1118         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1119         struct lu_attr                  *la     = &info->lti_la;
1120         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1121         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1122         /* The child and its name may be on different MDTs. */
1123         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1124         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
1125         struct dt_device                *dev    = lfsck->li_next;
1126         struct thandle                  *th     = NULL;
1127         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1128         int                              rc     = 0;
1129         ENTRY;
1130
1131         /* @parent/@child may be based on lfsck->li_bottom,
1132          * but here we need the object based on the lfsck->li_next. */
1133
1134         parent = lfsck_object_locate(dev, parent);
1135         if (IS_ERR(parent))
1136                 GOTO(log, rc = PTR_ERR(parent));
1137
1138         if (unlikely(!dt_try_as_dir(env, parent)))
1139                 GOTO(log, rc = -ENOTDIR);
1140
1141         child = lfsck_object_locate(dev, child);
1142         if (IS_ERR(child))
1143                 GOTO(log, rc = PTR_ERR(child));
1144
1145         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1146                 GOTO(log, rc = 1);
1147
1148         rc = lfsck_lock(env, lfsck, parent, name, llh,
1149                         MDS_INODELOCK_UPDATE, LCK_PW);
1150         if (rc != 0)
1151                 GOTO(log, rc);
1152
1153         th = dt_trans_create(env, dev);
1154         if (IS_ERR(th))
1155                 GOTO(unlock, rc = PTR_ERR(th));
1156
1157         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1158         rec->rec_fid = cfid;
1159         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1160                                (const struct dt_key *)name, th);
1161         if (rc != 0)
1162                 GOTO(stop, rc);
1163
1164         if (S_ISDIR(rec->rec_type)) {
1165                 rc = dt_declare_ref_add(env, parent, th);
1166                 if (rc != 0)
1167                         GOTO(stop, rc);
1168         }
1169
1170         memset(la, 0, sizeof(*la));
1171         la->la_ctime = cfs_time_current_sec();
1172         la->la_valid = LA_CTIME;
1173         rc = dt_declare_attr_set(env, parent, la, th);
1174         if (rc != 0)
1175                 GOTO(stop, rc);
1176
1177         rc = dt_declare_attr_set(env, child, la, th);
1178         if (rc != 0)
1179                 GOTO(stop, rc);
1180
1181         rc = dt_trans_start_local(env, dev, th);
1182         if (rc != 0)
1183                 GOTO(stop, rc);
1184
1185         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1186                        (const struct dt_key *)name, th, 1);
1187         if (rc != 0)
1188                 GOTO(stop, rc);
1189
1190         if (S_ISDIR(rec->rec_type)) {
1191                 dt_write_lock(env, parent, 0);
1192                 rc = dt_ref_add(env, parent, th);
1193                 dt_write_unlock(env, parent);
1194                 if (rc != 0)
1195                         GOTO(stop, rc);
1196         }
1197
1198         la->la_ctime = cfs_time_current_sec();
1199         rc = dt_attr_set(env, parent, la, th);
1200         if (rc != 0)
1201                 GOTO(stop, rc);
1202
1203         rc = dt_attr_set(env, child, la, th);
1204
1205         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1206
1207 stop:
1208         dt_trans_stop(env, dev, th);
1209
1210 unlock:
1211         lfsck_unlock(llh);
1212
1213 log:
1214         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1215                "the name %s and type %o to the parent "DFID": rc = %d\n",
1216                lfsck_lfsck2name(lfsck), PFID(cfid), name,
1217                lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1218
1219         if (rc != 0) {
1220                 struct lfsck_namespace *ns = com->lc_file_ram;
1221
1222                 ns->ln_flags |= LF_INCONSISTENT;
1223                 if (rc > 0)
1224                         ns->ln_lost_dirent_repaired++;
1225         }
1226
1227         return rc;
1228 }
1229
1230 /**
1231  * Create the specified orphan directory.
1232  *
1233  * For the case that the parent MDT-object stored in some MDT-object's
1234  * linkEA entry is lost, the LFSCK will re-create the parent object as
1235  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1236  * with the name ${FID}-P-${conflict_version}.
1237  *
1238  * \param[in] env       pointer to the thread context
1239  * \param[in] com       pointer to the lfsck component
1240  * \param[in] orphan    pointer to the orphan MDT-object to be created
1241  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1242  *
1243  * \retval              positive number for repaired cases
1244  * \retval              negative error number on failure
1245  */
1246 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1247                                              struct lfsck_component *com,
1248                                              struct dt_object *orphan,
1249                                              struct lmv_mds_md_v1 *lmv)
1250 {
1251         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1252         struct lu_attr                  *la     = &info->lti_la;
1253         struct dt_allocation_hint       *hint   = &info->lti_hint;
1254         struct dt_object_format         *dof    = &info->lti_dof;
1255         struct lu_name                  *cname  = &info->lti_name2;
1256         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1257         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1258         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1259         struct lu_fid                    tfid;
1260         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1261         struct lfsck_namespace          *ns     = com->lc_file_ram;
1262         struct dt_device                *dev    = lfsck_obj2dev(orphan);
1263         struct dt_object                *parent = NULL;
1264         struct thandle                  *th     = NULL;
1265         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1266         struct linkea_data               ldata  = { NULL };
1267         struct lu_buf                    linkea_buf;
1268         struct lu_buf                    lmv_buf;
1269         char                             name[32];
1270         int                              namelen;
1271         int                              idx    = 0;
1272         int                              rc     = 0;
1273         int                              rc1    = 0;
1274         ENTRY;
1275
1276         LASSERT(!dt_object_exists(orphan));
1277
1278         cname->ln_name = NULL;
1279         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1280                 GOTO(log, rc = 1);
1281
1282         if (dt_object_remote(orphan)) {
1283                 LASSERT(lfsck->li_lpf_root_obj != NULL);
1284
1285                 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1286                 if (idx < 0)
1287                         GOTO(log, rc = idx);
1288
1289                 snprintf(name, 8, "MDT%04x", idx);
1290                 rc = dt_lookup(env, lfsck->li_lpf_root_obj,
1291                                (struct dt_rec *)&tfid,
1292                                (const struct dt_key *)name);
1293                 if (rc != 0)
1294                         GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1295
1296                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1297                 if (IS_ERR(parent))
1298                         GOTO(log, rc = PTR_ERR(parent));
1299
1300                 if (unlikely(!dt_try_as_dir(env, parent)))
1301                         GOTO(log, rc = -ENOTDIR);
1302         } else {
1303                 if (unlikely(lfsck->li_lpf_obj == NULL))
1304                         GOTO(log, rc = -ENXIO);
1305
1306                 parent = lfsck->li_lpf_obj;
1307         }
1308
1309         dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1310         if (IS_ERR(dev))
1311                 GOTO(log, rc = PTR_ERR(dev));
1312
1313         idx = 0;
1314
1315 again:
1316         do {
1317                 namelen = snprintf(name, 31, DFID"-P-%d",
1318                                    PFID(cfid), idx++);
1319                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1320                                (const struct dt_key *)name);
1321                 if (rc != 0 && rc != -ENOENT)
1322                         GOTO(log, rc);
1323         } while (rc == 0);
1324
1325         rc = lfsck_lock(env, lfsck, parent, name, llh,
1326                         MDS_INODELOCK_UPDATE, LCK_PW);
1327         if (rc != 0)
1328                 GOTO(log, rc);
1329
1330         /* Re-check whether the name conflict with othrs after taken
1331          * the ldlm lock. */
1332         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1333                        (const struct dt_key *)name);
1334         if (unlikely(rc == 0)) {
1335                 lfsck_unlock(llh);
1336                 goto again;
1337         }
1338
1339         if (rc != -ENOENT)
1340                 GOTO(unlock1, rc);
1341
1342         cname->ln_name = name;
1343         cname->ln_namelen = namelen;
1344
1345         memset(la, 0, sizeof(*la));
1346         la->la_mode = S_IFDIR | 0700;
1347         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1348                        LA_ATIME | LA_MTIME | LA_CTIME;
1349
1350         orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1351                                    la->la_mode & S_IFMT);
1352
1353         memset(dof, 0, sizeof(*dof));
1354         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1355
1356         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1357                               cname, lfsck_dto2fid(parent));
1358         if (rc != 0)
1359                 GOTO(unlock1, rc);
1360
1361         th = dt_trans_create(env, dev);
1362         if (IS_ERR(th))
1363                 GOTO(unlock1, rc = PTR_ERR(th));
1364
1365         /* Sync the remote transaction to guarantee that the subsequent
1366          * lock against the @orphan can find the @orphan in time. */
1367         if (dt_object_remote(orphan))
1368                 th->th_sync = 1;
1369
1370         rc = dt_declare_create(env, orphan, la, hint, dof, th);
1371         if (rc != 0)
1372                 GOTO(stop, rc);
1373
1374         if (unlikely(!dt_try_as_dir(env, orphan)))
1375                 GOTO(stop, rc = -ENOTDIR);
1376
1377         rc = dt_declare_ref_add(env, orphan, th);
1378         if (rc != 0)
1379                 GOTO(stop, rc);
1380
1381         rec->rec_type = S_IFDIR;
1382         rec->rec_fid = cfid;
1383         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1384                                (const struct dt_key *)dot, th);
1385         if (rc != 0)
1386                 GOTO(stop, rc);
1387
1388         rec->rec_fid = lfsck_dto2fid(parent);
1389         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1390                                (const struct dt_key *)dotdot, th);
1391         if (rc != 0)
1392                 GOTO(stop, rc);
1393
1394         if (lmv != NULL) {
1395                 lmv->lmv_magic = LMV_MAGIC;
1396                 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1397                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1398                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1399                 rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
1400                                           XATTR_NAME_LMV, 0, th);
1401                 if (rc != 0)
1402                         GOTO(stop, rc);
1403         }
1404
1405         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1406                        ldata.ld_leh->leh_len);
1407         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1408                                   XATTR_NAME_LINK, 0, th);
1409         if (rc != 0)
1410                 GOTO(stop, rc);
1411
1412         rec->rec_fid = cfid;
1413         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1414                                (const struct dt_key *)name, th);
1415         if (rc == 0)
1416                 rc = dt_declare_ref_add(env, parent, th);
1417
1418         if (rc != 0)
1419                 GOTO(stop, rc);
1420
1421         rc = dt_trans_start_local(env, dev, th);
1422         if (rc != 0)
1423                 GOTO(stop, rc);
1424
1425         dt_write_lock(env, orphan, 0);
1426         rc = dt_create(env, orphan, la, hint, dof, th);
1427         if (rc != 0)
1428                 GOTO(unlock2, rc);
1429
1430         rc = dt_ref_add(env, orphan, th);
1431         if (rc != 0)
1432                 GOTO(unlock2, rc);
1433
1434         rec->rec_fid = cfid;
1435         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1436                        (const struct dt_key *)dot, th, 1);
1437         if (rc != 0)
1438                 GOTO(unlock2, rc);
1439
1440         rec->rec_fid = lfsck_dto2fid(parent);
1441         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1442                        (const struct dt_key *)dotdot, th, 1);
1443         if (rc != 0)
1444                 GOTO(unlock2, rc);
1445
1446         if (lmv != NULL) {
1447                 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1448                 if (rc != 0)
1449                         GOTO(unlock2, rc);
1450         }
1451
1452         rc = dt_xattr_set(env, orphan, &linkea_buf,
1453                           XATTR_NAME_LINK, 0, th);
1454         dt_write_unlock(env, orphan);
1455         if (rc != 0)
1456                 GOTO(stop, rc);
1457
1458         rec->rec_fid = cfid;
1459         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1460                        (const struct dt_key *)name, th, 1);
1461         if (rc == 0) {
1462                 dt_write_lock(env, parent, 0);
1463                 rc = dt_ref_add(env, parent, th);
1464                 dt_write_unlock(env, parent);
1465         }
1466
1467         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1468
1469 unlock2:
1470         dt_write_unlock(env, orphan);
1471
1472 stop:
1473         rc1 = dt_trans_stop(env, dev, th);
1474         if (rc1 != 0 && rc > 0)
1475                 rc = rc1;
1476
1477 unlock1:
1478         lfsck_unlock(llh);
1479
1480 log:
1481         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1482                "the object "DFID", name = %s: rc = %d\n",
1483                lfsck_lfsck2name(lfsck), PFID(cfid),
1484                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1485
1486         if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1487                 lfsck_object_put(env, parent);
1488
1489         if (rc != 0)
1490                 ns->ln_flags |= LF_INCONSISTENT;
1491
1492         return rc;
1493 }
1494
1495 /**
1496  * Remove the specified entry from the linkEA.
1497  *
1498  * Locate the linkEA entry with the given @cname and @pfid, then
1499  * remove this entry or the other entries those are repeated with
1500  * this entry.
1501  *
1502  * \param[in] env       pointer to the thread context
1503  * \param[in] com       pointer to the lfsck component
1504  * \param[in] obj       pointer to the dt_object to be handled
1505  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1506  * \param[in] cname     the name for the child in the parent directory
1507  * \param[in] pfid      the parent directory's FID for the linkEA
1508  * \param[in] next      if true, then remove the first found linkEA
1509  *                      entry, and move the ldata->ld_lee to next entry
1510  *
1511  * \retval              positive number for repaired cases
1512  * \retval              0 if nothing to be repaired
1513  * \retval              negative error number on failure
1514  */
1515 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1516                                          struct lfsck_component *com,
1517                                          struct dt_object *obj,
1518                                          struct linkea_data *ldata,
1519                                          struct lu_name *cname,
1520                                          struct lu_fid *pfid,
1521                                          bool next)
1522 {
1523         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1524         struct dt_device                *dev       = lfsck_obj2dev(obj);
1525         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1526         struct thandle                  *th        = NULL;
1527         struct lustre_handle             lh        = { 0 };
1528         struct linkea_data               ldata_new = { NULL };
1529         struct lu_buf                    linkea_buf;
1530         int                              buflen    = 0;
1531         int                              rc        = 0;
1532         ENTRY;
1533
1534         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1535                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1536                               LCK_EX);
1537         if (rc != 0)
1538                 GOTO(log, rc);
1539
1540         if (next)
1541                 lfsck_linkea_del_buf(ldata, cname);
1542         else
1543                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1544                                                     true);
1545         if (ldata->ld_leh->leh_reccount > 0 ||
1546             unlikely(ldata->ld_leh->leh_overflow_time)) {
1547                 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1548                                ldata->ld_leh->leh_len);
1549                 buflen = linkea_buf.lb_len;
1550         }
1551
1552 again:
1553         th = dt_trans_create(env, dev);
1554         if (IS_ERR(th))
1555                 GOTO(unlock1, rc = PTR_ERR(th));
1556
1557         if (buflen != 0)
1558                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1559                                           XATTR_NAME_LINK, 0, th);
1560         else
1561                 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1562         if (rc != 0)
1563                 GOTO(stop, rc);
1564
1565         rc = dt_trans_start_local(env, dev, th);
1566         if (rc != 0)
1567                 GOTO(stop, rc);
1568
1569         dt_write_lock(env, obj, 0);
1570         if (unlikely(lfsck_is_dead_obj(obj)))
1571                 GOTO(unlock2, rc = -ENOENT);
1572
1573         rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1574         if (rc)
1575                 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1576
1577         /* The specified linkEA entry has been removed by race. */
1578         rc = linkea_links_find(&ldata_new, cname, pfid);
1579         if (rc != 0)
1580                 GOTO(unlock2, rc = 0);
1581
1582         if (bk->lb_param & LPF_DRYRUN)
1583                 GOTO(unlock2, rc = 1);
1584
1585         if (next)
1586                 lfsck_linkea_del_buf(&ldata_new, cname);
1587         else
1588                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1589                                                     true);
1590
1591         if (buflen < ldata_new.ld_leh->leh_len) {
1592                 dt_write_unlock(env, obj);
1593                 dt_trans_stop(env, dev, th);
1594                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1595                                ldata_new.ld_leh->leh_len);
1596                 goto again;
1597         }
1598
1599         if (ldata_new.ld_leh->leh_reccount > 0 ||
1600             unlikely(ldata->ld_leh->leh_overflow_time))
1601                 rc = lfsck_links_write(env, obj, &ldata_new, th);
1602         else
1603                 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1604
1605         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1606
1607 unlock2:
1608         dt_write_unlock(env, obj);
1609
1610 stop:
1611         dt_trans_stop(env, dev, th);
1612
1613 unlock1:
1614         lfsck_ibits_unlock(&lh, LCK_EX);
1615
1616 log:
1617         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1618                "for the object: "DFID", parent "DFID", name %.*s\n",
1619                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1620                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1621                cname->ln_name);
1622
1623         if (rc != 0) {
1624                 struct lfsck_namespace *ns = com->lc_file_ram;
1625
1626                 ns->ln_flags |= LF_INCONSISTENT;
1627         }
1628
1629         return rc;
1630 }
1631
1632 /**
1633  * Conditionally remove the specified entry from the linkEA.
1634  *
1635  * Take the parent lock firstly, then check whether the specified
1636  * name entry exists or not: if yes, do nothing; otherwise, call
1637  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1638  *
1639  * \param[in] env       pointer to the thread context
1640  * \param[in] com       pointer to the lfsck component
1641  * \param[in] parent    pointer to the parent directory
1642  * \param[in] child     pointer to the child object that holds the linkEA
1643  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1644  * \param[in] cname     the name for the child in the parent directory
1645  * \param[in] pfid      the parent directory's FID for the linkEA
1646  *
1647  * \retval              positive number for repaired cases
1648  * \retval              0 if nothing to be repaired
1649  * \retval              negative error number on failure
1650  */
1651 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1652                                               struct lfsck_component *com,
1653                                               struct dt_object *parent,
1654                                               struct dt_object *child,
1655                                               struct linkea_data *ldata,
1656                                               struct lu_name *cname,
1657                                               struct lu_fid *pfid)
1658 {
1659         struct lfsck_thread_info *info  = lfsck_env_info(env);
1660         struct lu_fid            *cfid  = &info->lti_fid3;
1661         struct lfsck_lock_handle *llh   = &info->lti_llh;
1662         int                       rc;
1663         ENTRY;
1664
1665         rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1666                         MDS_INODELOCK_UPDATE, LCK_PR);
1667         if (rc != 0)
1668                 RETURN(rc);
1669
1670         dt_read_lock(env, parent, 0);
1671         if (unlikely(lfsck_is_dead_obj(parent))) {
1672                 dt_read_unlock(env, parent);
1673                 lfsck_unlock(llh);
1674                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1675                                                    cname, pfid, true);
1676
1677                 RETURN(rc);
1678         }
1679
1680         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1681                        (const struct dt_key *)cname->ln_name);
1682         dt_read_unlock(env, parent);
1683
1684         /* It is safe to release the ldlm lock, because when the logic come
1685          * here, we have got all the needed information above whether the
1686          * linkEA entry is valid or not. It is not important that others
1687          * may add new linkEA entry after the ldlm lock released. If other
1688          * has removed the specified linkEA entry by race, then it is OK,
1689          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1690          * such case. */
1691         lfsck_unlock(llh);
1692         if (rc == -ENOENT) {
1693                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1694                                                    cname, pfid, true);
1695
1696                 RETURN(rc);
1697         }
1698
1699         if (rc != 0)
1700                 RETURN(rc);
1701
1702         /* The LFSCK just found some internal status of cross-MDTs
1703          * create operation. That is normal. */
1704         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1705                 linkea_next_entry(ldata);
1706
1707                 RETURN(0);
1708         }
1709
1710         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1711                                            pfid, true);
1712
1713         RETURN(rc);
1714 }
1715
1716 /**
1717  * Conditionally replace name entry in the parent.
1718  *
1719  * As required, the LFSCK may re-create the lost MDT-object for dangling
1720  * name entry, but such repairing may be wrong because of bad FID in the
1721  * name entry. As the LFSCK processing, the real MDT-object may be found,
1722  * then the LFSCK should check whether the former re-created MDT-object
1723  * has been modified or not, if not, then destroy it and update the name
1724  * entry in the parent to reference the real MDT-object.
1725  *
1726  * \param[in] env       pointer to the thread context
1727  * \param[in] com       pointer to the lfsck component
1728  * \param[in] parent    pointer to the parent directory
1729  * \param[in] child     pointer to the MDT-object that may be the real
1730  *                      MDT-object corresponding to the name entry in parent
1731  * \param[in] cfid      the current FID in the name entry
1732  * \param[in] cname     contains the name of the child in the parent directory
1733  *
1734  * \retval              positive number for repaired cases
1735  * \retval              0 if nothing to be repaired
1736  * \retval              negative error number on failure
1737  */
1738 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1739                                         struct lfsck_component *com,
1740                                         struct dt_object *parent,
1741                                         struct dt_object *child,
1742                                         const struct lu_fid *cfid,
1743                                         const struct lu_name *cname)
1744 {
1745         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1746         struct lu_attr                  *la     = &info->lti_la;
1747         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1748         struct lu_fid                    tfid;
1749         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1750         /* The child and its name may be on different MDTs. */
1751         struct dt_device                *dev    = lfsck->li_next;
1752         const char                      *name   = cname->ln_name;
1753         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1754         struct dt_object                *cobj   = NULL;
1755         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
1756         struct lustre_handle             clh    = { 0 };
1757         struct linkea_data               ldata  = { NULL };
1758         struct thandle                  *th     = NULL;
1759         bool                             exist  = true;
1760         int                              rc     = 0;
1761         ENTRY;
1762
1763         /* @parent/@child may be based on lfsck->li_bottom,
1764          * but here we need the object based on the lfsck->li_next. */
1765
1766         parent = lfsck_object_locate(dev, parent);
1767         if (IS_ERR(parent))
1768                 GOTO(log, rc = PTR_ERR(parent));
1769
1770         if (unlikely(!dt_try_as_dir(env, parent)))
1771                 GOTO(log, rc = -ENOTDIR);
1772
1773         rc = lfsck_lock(env, lfsck, parent, name, pllh,
1774                         MDS_INODELOCK_UPDATE, LCK_PW);
1775         if (rc != 0)
1776                 GOTO(log, rc);
1777
1778         if (!fid_is_sane(cfid)) {
1779                 exist = false;
1780                 goto replace;
1781         }
1782
1783         cobj = lfsck_object_find_by_dev(env, dev, cfid);
1784         if (IS_ERR(cobj)) {
1785                 rc = PTR_ERR(cobj);
1786                 if (rc == -ENOENT) {
1787                         exist = false;
1788                         goto replace;
1789                 }
1790
1791                 GOTO(log, rc);
1792         }
1793
1794         if (!dt_object_exists(cobj)) {
1795                 exist = false;
1796                 goto replace;
1797         }
1798
1799         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
1800                        (const struct dt_key *)name);
1801         if (rc == -ENOENT) {
1802                 exist = false;
1803                 goto replace;
1804         }
1805
1806         if (rc != 0)
1807                 GOTO(log, rc);
1808
1809         /* Someone changed the name entry, cannot replace it. */
1810         if (!lu_fid_eq(cfid, &tfid))
1811                 GOTO(log, rc = 0);
1812
1813         /* lock the object to be destroyed. */
1814         rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
1815                               MDS_INODELOCK_UPDATE |
1816                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1817                               LCK_EX);
1818         if (rc != 0)
1819                 GOTO(log, rc);
1820
1821         if (unlikely(lfsck_is_dead_obj(cobj))) {
1822                 exist = false;
1823                 goto replace;
1824         }
1825
1826         rc = dt_attr_get(env, cobj, la);
1827         if (rc != 0)
1828                 GOTO(log, rc);
1829
1830         /* The object has been modified by other(s), or it is not created by
1831          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1832         if (la->la_ctime != 0)
1833                 GOTO(log, rc);
1834
1835         if (S_ISREG(la->la_mode)) {
1836                 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
1837                 /* If someone has created related OST-object(s),
1838                  * then keep it. */
1839                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1840                         GOTO(log, rc = (rc > 0 ? 0 : rc));
1841         }
1842
1843 replace:
1844         dt_read_lock(env, child, 0);
1845         rc = lfsck_links_read2_with_rec(env, child, &ldata);
1846         dt_read_unlock(env, child);
1847
1848         /* Someone changed the child, no need to replace. */
1849         if (rc == -ENODATA)
1850                 GOTO(log, rc = 0);
1851
1852         if (rc != 0)
1853                 GOTO(log, rc);
1854
1855         rc = linkea_links_find(&ldata, cname, pfid);
1856         /* Someone moved the child, no need to replace. */
1857         if (rc != 0)
1858                 GOTO(log, rc = 0);
1859
1860         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1861                 GOTO(log, rc = 1);
1862
1863         th = dt_trans_create(env, dev);
1864         if (IS_ERR(th))
1865                 GOTO(log, rc = PTR_ERR(th));
1866
1867         if (exist) {
1868                 rc = dt_declare_destroy(env, cobj, th);
1869                 if (rc != 0)
1870                         GOTO(stop, rc);
1871         }
1872
1873         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1874         if (rc != 0)
1875                 GOTO(stop, rc);
1876
1877         rec->rec_type = S_IFDIR;
1878         rec->rec_fid = lfsck_dto2fid(child);
1879         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1880                                (const struct dt_key *)name, th);
1881         if (rc != 0)
1882                 GOTO(stop, rc);
1883
1884         rc = dt_trans_start_local(env, dev, th);
1885         if (rc != 0)
1886                 GOTO(stop, rc);
1887
1888         if (exist) {
1889                 rc = dt_destroy(env, cobj, th);
1890                 if (rc != 0)
1891                         GOTO(stop, rc);
1892         }
1893
1894         /* The old name entry maybe not exist. */
1895         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
1896         if (rc != 0 && rc != -ENOENT)
1897                 GOTO(stop, rc);
1898
1899         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1900                        (const struct dt_key *)name, th, 1);
1901
1902         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1903
1904 stop:
1905         dt_trans_stop(env, dev, th);
1906
1907 log:
1908         lfsck_ibits_unlock(&clh, LCK_EX);
1909         lfsck_unlock(pllh);
1910
1911         if (cobj != NULL && !IS_ERR(cobj))
1912                 lfsck_object_put(env, cobj);
1913
1914         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1915                "object "DFID" because of conflict with the object "DFID
1916                " under the parent "DFID" with name %s: rc = %d\n",
1917                lfsck_lfsck2name(lfsck), PFID(cfid),
1918                PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
1919
1920         return rc;
1921 }
1922
1923 /**
1924  * Overwrite the linkEA for the object with the given ldata.
1925  *
1926  * The caller should take the ldlm lock before the calling.
1927  *
1928  * \param[in] env       pointer to the thread context
1929  * \param[in] com       pointer to the lfsck component
1930  * \param[in] obj       pointer to the dt_object to be handled
1931  * \param[in] ldata     pointer to the new linkEA data
1932  *
1933  * \retval              positive number for repaired cases
1934  * \retval              0 if nothing to be repaired
1935  * \retval              negative error number on failure
1936  */
1937 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1938                                    struct lfsck_component *com,
1939                                    struct dt_object *obj,
1940                                    struct linkea_data *ldata)
1941 {
1942         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1943         struct dt_device                *dev    = lfsck_obj2dev(obj);
1944         struct thandle                  *th     = NULL;
1945         struct lu_buf                    linkea_buf;
1946         int                              rc     = 0;
1947         ENTRY;
1948
1949         th = dt_trans_create(env, dev);
1950         if (IS_ERR(th))
1951                 GOTO(log, rc = PTR_ERR(th));
1952
1953         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1954                        ldata->ld_leh->leh_len);
1955         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1956                                   XATTR_NAME_LINK, 0, th);
1957         if (rc != 0)
1958                 GOTO(stop, rc);
1959
1960         rc = dt_trans_start_local(env, dev, th);
1961         if (rc != 0)
1962                 GOTO(stop, rc);
1963
1964         dt_write_lock(env, obj, 0);
1965         if (unlikely(lfsck_is_dead_obj(obj)))
1966                 GOTO(unlock, rc = 0);
1967
1968         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1969                 GOTO(unlock, rc = 1);
1970
1971         rc = dt_xattr_set(env, obj, &linkea_buf,
1972                           XATTR_NAME_LINK, 0, th);
1973
1974         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1975
1976 unlock:
1977         dt_write_unlock(env, obj);
1978
1979 stop:
1980         dt_trans_stop(env, dev, th);
1981
1982 log:
1983         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1984                "object "DFID": rc = %d\n",
1985                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1986
1987         if (rc != 0) {
1988                 struct lfsck_namespace *ns = com->lc_file_ram;
1989
1990                 ns->ln_flags |= LF_INCONSISTENT;
1991         }
1992
1993         return rc;
1994 }
1995
1996 /**
1997  * Repair invalid name entry.
1998  *
1999  * If the name entry contains invalid information, such as bad file type
2000  * or (and) corrupted object FID, then either remove the name entry or
2001  * udpate the name entry with the given (right) information.
2002  *
2003  * \param[in] env       pointer to the thread context
2004  * \param[in] com       pointer to the lfsck component
2005  * \param[in] parent    pointer to the parent directory
2006  * \param[in] child     pointer to the object referenced by the name entry
2007  * \param[in] name      the old name of the child under the parent directory
2008  * \param[in] name2     the new name of the child under the parent directory
2009  * \param[in] type      the type claimed by the name entry
2010  * \param[in] update    update the name entry if true; otherwise, remove it
2011  * \param[in] dec       decrease the parent nlink count if true
2012  *
2013  * \retval              positive number for repaired successfully
2014  * \retval              0 if nothing to be repaired
2015  * \retval              negative error number on failure
2016  */
2017 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2018                                   struct lfsck_component *com,
2019                                   struct dt_object *parent,
2020                                   struct dt_object *child,
2021                                   const char *name, const char *name2,
2022                                   __u16 type, bool update, bool dec)
2023 {
2024         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2025         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2026         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2027         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
2028         struct lu_fid                    tfid;
2029         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2030         struct dt_device                *dev    = lfsck->li_next;
2031         struct thandle                  *th     = NULL;
2032         struct lfsck_lock_handle        *llh    = &info->lti_llh;
2033         struct lustre_handle             lh     = { 0 };
2034         int                              rc     = 0;
2035         ENTRY;
2036
2037         parent = lfsck_object_locate(dev, parent);
2038         if (IS_ERR(parent))
2039                 GOTO(log, rc = PTR_ERR(parent));
2040
2041         if (unlikely(!dt_try_as_dir(env, parent)))
2042                 GOTO(log, rc = -ENOTDIR);
2043
2044         if (!update || strcmp(name, name2) == 0)
2045                 rc = lfsck_lock(env, lfsck, parent, name, llh,
2046                                 MDS_INODELOCK_UPDATE, LCK_PW);
2047         else
2048                 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2049                                       MDS_INODELOCK_UPDATE, LCK_PW);
2050         if (rc != 0)
2051                 GOTO(log, rc);
2052
2053         th = dt_trans_create(env, dev);
2054         if (IS_ERR(th))
2055                 GOTO(unlock1, rc = PTR_ERR(th));
2056
2057         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2058         if (rc != 0)
2059                 GOTO(stop, rc);
2060
2061         if (update) {
2062                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2063                 rec->rec_fid = cfid;
2064                 rc = dt_declare_insert(env, parent,
2065                                        (const struct dt_rec *)rec,
2066                                        (const struct dt_key *)name2, th);
2067                 if (rc != 0)
2068                         GOTO(stop, rc);
2069         }
2070
2071         if (dec && S_ISDIR(type)) {
2072                 rc = dt_declare_ref_del(env, parent, th);
2073                 if (rc != 0)
2074                         GOTO(stop, rc);
2075         }
2076
2077         rc = dt_trans_start_local(env, dev, th);
2078         if (rc != 0)
2079                 GOTO(stop, rc);
2080
2081
2082         dt_write_lock(env, parent, 0);
2083         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2084                        (const struct dt_key *)name);
2085         /* Someone has removed the bad name entry by race. */
2086         if (rc == -ENOENT)
2087                 GOTO(unlock2, rc = 0);
2088
2089         if (rc != 0)
2090                 GOTO(unlock2, rc);
2091
2092         /* Someone has removed the bad name entry and reused it for other
2093          * object by race. */
2094         if (!lu_fid_eq(&tfid, cfid))
2095                 GOTO(unlock2, rc = 0);
2096
2097         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2098                 GOTO(unlock2, rc = 1);
2099
2100         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2101         if (rc != 0)
2102                 GOTO(unlock2, rc);
2103
2104         if (update) {
2105                 rc = dt_insert(env, parent,
2106                                (const struct dt_rec *)rec,
2107                                (const struct dt_key *)name2, th, 1);
2108                 if (rc != 0)
2109                         GOTO(unlock2, rc);
2110         }
2111
2112         if (dec && S_ISDIR(type)) {
2113                 rc = dt_ref_del(env, parent, th);
2114                 if (rc != 0)
2115                         GOTO(unlock2, rc);
2116         }
2117
2118         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2119
2120 unlock2:
2121         dt_write_unlock(env, parent);
2122
2123 stop:
2124         dt_trans_stop(env, dev, th);
2125
2126         /* We are not sure whether the child will become orphan or not.
2127          * Record it in the LFSCK trace file for further checking in
2128          * the second-stage scanning. */
2129         if (!update && !dec && rc == 0)
2130                 lfsck_namespace_trace_update(env, com, cfid,
2131                                              LNTF_CHECK_LINKEA, true);
2132
2133 unlock1:
2134         /* It is harmless even if unlock the unused lock_handle */
2135         lfsck_ibits_unlock(&lh, LCK_PW);
2136         lfsck_unlock(llh);
2137
2138 log:
2139         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2140                "entry for: parent "DFID", child "DFID", name %s, type "
2141                "in name entry %o, type claimed by child %o. repair it "
2142                "by %s with new name2 %s: rc = %d\n",
2143                lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2144                name, type, update ? lfsck_object_type(child) : 0,
2145                update ? "updating" : "removing", name2, rc);
2146
2147         if (rc != 0) {
2148                 struct lfsck_namespace *ns = com->lc_file_ram;
2149
2150                 ns->ln_flags |= LF_INCONSISTENT;
2151         }
2152
2153         return rc;
2154 }
2155
2156 /**
2157  * Update the ".." name entry for the given object.
2158  *
2159  * The object's ".." is corrupted, this function will update the ".." name
2160  * entry with the given pfid, and the linkEA with the given ldata.
2161  *
2162  * The caller should take the ldlm lock before the calling.
2163  *
2164  * \param[in] env       pointer to the thread context
2165  * \param[in] com       pointer to the lfsck component
2166  * \param[in] obj       pointer to the dt_object to be handled
2167  * \param[in] pfid      the new fid for the object's ".." name entry
2168  * \param[in] cname     the name for the @obj in the parent directory
2169  *
2170  * \retval              positive number for repaired cases
2171  * \retval              0 if nothing to be repaired
2172  * \retval              negative error number on failure
2173  */
2174 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2175                                                   struct lfsck_component *com,
2176                                                   struct dt_object *obj,
2177                                                   const struct lu_fid *pfid,
2178                                                   struct lu_name *cname)
2179 {
2180         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2181         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2182         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2183         struct dt_device                *dev    = lfsck_obj2dev(obj);
2184         struct thandle                  *th     = NULL;
2185         struct linkea_data               ldata  = { NULL };
2186         struct lu_buf                    linkea_buf;
2187         int                              rc     = 0;
2188         ENTRY;
2189
2190         LASSERT(!dt_object_remote(obj));
2191         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2192
2193         rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2194         if (rc != 0)
2195                 GOTO(log, rc);
2196
2197         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2198                        ldata.ld_leh->leh_len);
2199
2200         th = dt_trans_create(env, dev);
2201         if (IS_ERR(th))
2202                 GOTO(log, rc = PTR_ERR(th));
2203
2204         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2205         if (rc != 0)
2206                 GOTO(stop, rc);
2207
2208         rec->rec_type = S_IFDIR;
2209         rec->rec_fid = pfid;
2210         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2211                                (const struct dt_key *)dotdot, th);
2212         if (rc != 0)
2213                 GOTO(stop, rc);
2214
2215         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2216                                   XATTR_NAME_LINK, 0, th);
2217         if (rc != 0)
2218                 GOTO(stop, rc);
2219
2220         rc = dt_trans_start_local(env, dev, th);
2221         if (rc != 0)
2222                 GOTO(stop, rc);
2223
2224         dt_write_lock(env, obj, 0);
2225         if (unlikely(lfsck_is_dead_obj(obj)))
2226                 GOTO(unlock, rc = 0);
2227
2228         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2229                 GOTO(unlock, rc = 1);
2230
2231         /* The old ".." name entry maybe not exist. */
2232         dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2233
2234         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2235                        (const struct dt_key *)dotdot, th, 1);
2236         if (rc != 0)
2237                 GOTO(unlock, rc);
2238
2239         rc = lfsck_links_write(env, obj, &ldata, th);
2240
2241         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2242
2243 unlock:
2244         dt_write_unlock(env, obj);
2245
2246 stop:
2247         dt_trans_stop(env, dev, th);
2248
2249 log:
2250         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2251                "the object "DFID", new parent "DFID": rc = %d\n",
2252                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2253                PFID(pfid), rc);
2254
2255         if (rc != 0) {
2256                 struct lfsck_namespace *ns = com->lc_file_ram;
2257
2258                 ns->ln_flags |= LF_INCONSISTENT;
2259         }
2260
2261         return rc;
2262 }
2263
2264 /**
2265  * Handle orphan @obj during Double Scan Directory.
2266  *
2267  * Remove the @obj's current (invalid) linkEA entries, and insert
2268  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2269  * ${FID}-${PFID}-D-${conflict_version}
2270  *
2271  * The caller should take the ldlm lock before the calling.
2272  *
2273  * \param[in] env       pointer to the thread context
2274  * \param[in] com       pointer to the lfsck component
2275  * \param[in] obj       pointer to the orphan object to be handled
2276  * \param[in] pfid      the new fid for the object's ".." name entry
2277  * \param[in,out] lh    ldlm lock handler for the given @obj
2278  * \param[out] type     to tell the caller what the inconsistency is
2279  *
2280  * \retval              positive number for repaired cases
2281  * \retval              0 if nothing to be repaired
2282  * \retval              negative error number on failure
2283  */
2284 static int
2285 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2286                            struct lfsck_component *com,
2287                            struct dt_object *obj,
2288                            const struct lu_fid *pfid,
2289                            struct lustre_handle *lh,
2290                            enum lfsck_namespace_inconsistency_type *type)
2291 {
2292         struct lfsck_thread_info *info = lfsck_env_info(env);
2293         struct lfsck_namespace   *ns   = com->lc_file_ram;
2294         int                       rc;
2295         ENTRY;
2296
2297         /* Remove the unrecognized linkEA. */
2298         rc = lfsck_namespace_links_remove(env, com, obj);
2299         lfsck_ibits_unlock(lh, LCK_EX);
2300         if (rc < 0 && rc != -ENODATA)
2301                 RETURN(rc);
2302
2303         *type = LNIT_MUL_REF;
2304
2305         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2306          * ever tried to verify some remote MDT-object that resides on this
2307          * MDT, but this MDT failed to respond such request. So means there
2308          * may be some remote name entry on other MDT that references this
2309          * object with another name, so we cannot know whether this linkEA
2310          * is valid or not. So keep it there and maybe resolved when next
2311          * LFSCK run. */
2312         if (ns->ln_flags & LF_INCOMPLETE)
2313                 RETURN(0);
2314
2315         /* The unique linkEA is invalid, even if the ".." name entry may be
2316          * valid, we still cannot know via which name entry this directory
2317          * will be referenced. Then handle it as pure orphan. */
2318         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2319                  "-"DFID, PFID(pfid));
2320         rc = lfsck_namespace_insert_orphan(env, com, obj,
2321                                            info->lti_tmpbuf, "D", NULL);
2322
2323         RETURN(rc);
2324 }
2325
2326 /**
2327  * Double Scan Directory object for single linkEA entry case.
2328  *
2329  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2330  * then check whether the name is in the namespace or not, if not, add the
2331  * missing name entry back to namespace. If the linkEA entry is invalid,
2332  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2333  * as an orphan.
2334  *
2335  * \param[in] env       pointer to the thread context
2336  * \param[in] com       pointer to the lfsck component
2337  * \param[in] child     pointer to the directory to be double scanned
2338  * \param[in] pfid      the FID corresponding to the ".." entry
2339  * \param[in] ldata     pointer to the linkEA data for the given @child
2340  * \param[in,out] lh    ldlm lock handler for the given @child
2341  * \param[out] type     to tell the caller what the inconsistency is
2342  * \param[in] retry     if found inconsistency, but the caller does not hold
2343  *                      ldlm lock on the @child, then set @retry as true
2344  * \param[in] unknown   set if does not know how to repair the inconsistency
2345  *
2346  * \retval              positive number for repaired cases
2347  * \retval              0 if nothing to be repaired
2348  * \retval              negative error number on failure
2349  */
2350 static int
2351 lfsck_namespace_dsd_single(const struct lu_env *env,
2352                            struct lfsck_component *com,
2353                            struct dt_object *child,
2354                            const struct lu_fid *pfid,
2355                            struct linkea_data *ldata,
2356                            struct lustre_handle *lh,
2357                            enum lfsck_namespace_inconsistency_type *type,
2358                            bool *retry, bool *unknown)
2359 {
2360         struct lfsck_thread_info *info          = lfsck_env_info(env);
2361         struct lu_name           *cname         = &info->lti_name;
2362         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2363         struct lu_fid             tfid;
2364         struct lfsck_namespace   *ns            = com->lc_file_ram;
2365         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2366         struct dt_object         *parent        = NULL;
2367         struct lmv_mds_md_v1     *lmv;
2368         int                       rc            = 0;
2369         ENTRY;
2370
2371         rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2372                                                  info->lti_key,
2373                                                  sizeof(info->lti_key));
2374         /* The unique linkEA entry with bad parent will be handled as orphan. */
2375         if (rc != 0) {
2376                 if (!lustre_handle_is_used(lh) && retry != NULL)
2377                         *retry = true;
2378                 else
2379                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2380                                                         pfid, lh, type);
2381
2382                 GOTO(out, rc);
2383         }
2384
2385         parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2386         if (IS_ERR(parent))
2387                 GOTO(out, rc = PTR_ERR(parent));
2388
2389         /* We trust the unique linkEA entry in spite of whether it matches the
2390          * ".." name entry or not. Because even if the linkEA entry is wrong
2391          * and the ".." name entry is right, we still cannot know via which
2392          * name entry the child will be referenced, since all known entries
2393          * have been verified during the first-stage scanning. */
2394         if (!dt_object_exists(parent)) {
2395                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2396                  * has ever tried to verify some remote MDT-object that resides
2397                  * on this MDT, but this MDT failed to respond such request. So
2398                  * means there may be some remote name entry on other MDT that
2399                  * references this object with another name, so we cannot know
2400                  * whether this linkEA is valid or not. So keep it there and
2401                  * maybe resolved when next LFSCK run. */
2402                 if (ns->ln_flags & LF_INCOMPLETE)
2403                         GOTO(out, rc = 0);
2404
2405                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2406                         *retry = true;
2407
2408                         GOTO(out, rc = 0);
2409                 }
2410
2411                 lfsck_ibits_unlock(lh, LCK_EX);
2412
2413 lost_parent:
2414                 lmv = &info->lti_lmv;
2415                 rc = lfsck_read_stripe_lmv(env, child, lmv);
2416                 if (rc != 0 && rc != -ENODATA)
2417                         GOTO(out, rc);
2418
2419                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2420                         lmv = NULL;
2421                 } else if (lfsck_shard_name_to_index(env,
2422                                         cname->ln_name, cname->ln_namelen,
2423                                         S_IFDIR, cfid) < 0) {
2424                         /* It is an invalid name entry, we
2425                          * cannot trust the parent also. */
2426                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2427                                                 ldata, cname, &tfid, true);
2428                         if (rc < 0)
2429                                 GOTO(out, rc);
2430
2431                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2432                                  "-"DFID, PFID(pfid));
2433                         rc = lfsck_namespace_insert_orphan(env, com, child,
2434                                                 info->lti_tmpbuf, "S", NULL);
2435
2436                         GOTO(out, rc);
2437                 }
2438
2439                 /* Create the lost parent as an orphan. */
2440                 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2441                 if (rc >= 0) {
2442                         /* Add the missing name entry to the parent. */
2443                         rc = lfsck_namespace_insert_normal(env, com, parent,
2444                                                         child, cname->ln_name);
2445                         if (unlikely(rc == -EEXIST)) {
2446                                 /* Unfortunately, someone reused the name
2447                                  * under the parent by race. So we have
2448                                  * to remove the linkEA entry from
2449                                  * current child object. It means that the
2450                                  * LFSCK cannot recover the system
2451                                  * totally back to its original status,
2452                                  * but it is necessary to make the
2453                                  * current system to be consistent. */
2454                                 rc = lfsck_namespace_shrink_linkea(env,
2455                                                 com, child, ldata,
2456                                                 cname, &tfid, true);
2457                                 if (rc >= 0) {
2458                                         snprintf(info->lti_tmpbuf,
2459                                                  sizeof(info->lti_tmpbuf),
2460                                                  "-"DFID, PFID(pfid));
2461                                         rc = lfsck_namespace_insert_orphan(env,
2462                                                 com, child, info->lti_tmpbuf,
2463                                                 "D", NULL);
2464                                 }
2465                         }
2466                 }
2467
2468                 GOTO(out, rc);
2469         } /* !dt_object_exists(parent) */
2470
2471         /* The unique linkEA entry with bad parent will be handled as orphan. */
2472         if (unlikely(!dt_try_as_dir(env, parent))) {
2473                 if (!lustre_handle_is_used(lh) && retry != NULL)
2474                         *retry = true;
2475                 else
2476                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2477                                                         pfid, lh, type);
2478
2479                 GOTO(out, rc);
2480         }
2481
2482         rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2483                        (const struct dt_key *)cname->ln_name);
2484         if (rc == -ENOENT) {
2485                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2486                  * has ever tried to verify some remote MDT-object that resides
2487                  * on this MDT, but this MDT failed to respond such request. So
2488                  * means there may be some remote name entry on other MDT that
2489                  * references this object with another name, so we cannot know
2490                  * whether this linkEA is valid or not. So keep it there and
2491                  * maybe resolved when next LFSCK run. */
2492                 if (ns->ln_flags & LF_INCOMPLETE)
2493                         GOTO(out, rc = 0);
2494
2495                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2496                         *retry = true;
2497
2498                         GOTO(out, rc = 0);
2499                 }
2500
2501                 lfsck_ibits_unlock(lh, LCK_EX);
2502                 rc = lfsck_namespace_check_name(env, parent, child, cname);
2503                 if (rc == -ENOENT)
2504                         goto lost_parent;
2505
2506                 if (rc < 0)
2507                         GOTO(out, rc);
2508
2509                 /* It is an invalid name entry, drop it. */
2510                 if (unlikely(rc > 0)) {
2511                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2512                                                 ldata, cname, &tfid, true);
2513                         if (rc >= 0) {
2514                                 snprintf(info->lti_tmpbuf,
2515                                          sizeof(info->lti_tmpbuf),
2516                                          "-"DFID, PFID(pfid));
2517                                 rc = lfsck_namespace_insert_orphan(env, com,
2518                                         child, info->lti_tmpbuf, "D", NULL);
2519                         }
2520
2521                         GOTO(out, rc);
2522                 }
2523
2524                 /* Add the missing name entry back to the namespace. */
2525                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2526                                                    cname->ln_name);
2527                 if (unlikely(rc == -ESTALE))
2528                         /* It may happen when the remote object has been
2529                          * removed, but the local MDT is not aware of that. */
2530                         goto lost_parent;
2531
2532                 if (unlikely(rc == -EEXIST)) {
2533                         /* Unfortunately, someone reused the name under the
2534                          * parent by race. So we have to remove the linkEA
2535                          * entry from current child object. It means that the
2536                          * LFSCK cannot recover the system totally back to
2537                          * its original status, but it is necessary to make
2538                          * the current system to be consistent.
2539                          *
2540                          * It also may be because of the LFSCK found some
2541                          * internal status of create operation. Under such
2542                          * case, nothing to be done. */
2543                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2544                                         parent, child, ldata, cname, &tfid);
2545                         if (rc >= 0) {
2546                                 snprintf(info->lti_tmpbuf,
2547                                          sizeof(info->lti_tmpbuf),
2548                                          "-"DFID, PFID(pfid));
2549                                 rc = lfsck_namespace_insert_orphan(env, com,
2550                                         child, info->lti_tmpbuf, "D", NULL);
2551                         }
2552                 }
2553
2554                 GOTO(out, rc);
2555         } /* rc == -ENOENT */
2556
2557         if (rc != 0)
2558                 GOTO(out, rc);
2559
2560         if (!lu_fid_eq(&tfid, cfid)) {
2561                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2562                         *retry = true;
2563
2564                         GOTO(out, rc = 0);
2565                 }
2566
2567                 lfsck_ibits_unlock(lh, LCK_EX);
2568                 /* The name entry references another MDT-object that
2569                  * may be created by the LFSCK for repairing dangling
2570                  * name entry. Try to replace it. */
2571                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2572                                                   &tfid, cname);
2573                 if (rc == 0)
2574                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2575                                                         pfid, lh, type);
2576
2577                 GOTO(out, rc);
2578         }
2579
2580         /* Zero FID may because the remote directroy object has invalid linkEA,
2581          * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2582          * how to repair the inconsistency, but the namespace LFSCK on the MDT
2583          * where its name entry resides may has more information (name, FID) to
2584          * repair such inconsistency. So here, keep the inconsistency to avoid
2585          * some imporper repairing. */
2586         if (fid_is_zero(pfid)) {
2587                 if (unknown)
2588                         *unknown = true;
2589
2590                 GOTO(out, rc = 0);
2591         }
2592
2593         /* The ".." name entry is wrong, update it. */
2594         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2595                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2596                         *retry = true;
2597
2598                         GOTO(out, rc = 0);
2599                 }
2600
2601                 *type = LNIT_UNMATCHED_PAIRS;
2602                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2603                                                 lfsck_dto2fid(parent), cname);
2604         }
2605
2606         GOTO(out, rc);
2607
2608 out:
2609         if (parent != NULL && !IS_ERR(parent))
2610                 lfsck_object_put(env, parent);
2611
2612         return rc;
2613 }
2614
2615 /**
2616  * Double Scan Directory object for multiple linkEA entries case.
2617  *
2618  * The given @child has multiple linkEA entries. There is at most one linkEA
2619  * entry will be valid, all the others will be removed. Firstly, the function
2620  * will try to find out the linkEA entry for which the name entry exists under
2621  * the given parent (@pfid). If there is no linkEA entry that matches the given
2622  * ".." name entry, then tries to find out the first linkEA entry that both the
2623  * parent and the name entry exist to rebuild a new ".." name entry.
2624  *
2625  * \param[in] env       pointer to the thread context
2626  * \param[in] com       pointer to the lfsck component
2627  * \param[in] child     pointer to the directory to be double scanned
2628  * \param[in] pfid      the FID corresponding to the ".." entry
2629  * \param[in] ldata     pointer to the linkEA data for the given @child
2630  * \param[in,out] lh    ldlm lock handler for the given @child
2631  * \param[out] type     to tell the caller what the inconsistency is
2632  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2633  * \param[in] unknown   set if does not know how to repair the inconsistency
2634  *
2635  * \retval              positive number for repaired cases
2636  * \retval              0 if nothing to be repaired
2637  * \retval              negative error number on failure
2638  */
2639 static int
2640 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2641                              struct lfsck_component *com,
2642                              struct dt_object *child,
2643                              const struct lu_fid *pfid,
2644                              struct linkea_data *ldata,
2645                              struct lustre_handle *lh,
2646                              enum lfsck_namespace_inconsistency_type *type,
2647                              bool lpf, bool *unknown)
2648 {
2649         struct lfsck_thread_info *info          = lfsck_env_info(env);
2650         struct lu_name           *cname         = &info->lti_name;
2651         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2652         struct lu_fid            *pfid2         = &info->lti_fid3;
2653         struct lu_fid             tfid;
2654         struct lfsck_namespace   *ns            = com->lc_file_ram;
2655         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2656         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2657         struct dt_object         *parent        = NULL;
2658         struct linkea_data        ldata_new     = { NULL };
2659         int                       dirent_count  = 0;
2660         int                       rc            = 0;
2661         bool                      once          = true;
2662         ENTRY;
2663
2664 again:
2665         while (ldata->ld_lee != NULL) {
2666                 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2667                                                          info->lti_key,
2668                                                          sizeof(info->lti_key));
2669                 /* Drop invalid linkEA entry. */
2670                 if (rc != 0) {
2671                         lfsck_linkea_del_buf(ldata, cname);
2672                         continue;
2673                 }
2674
2675                 /* Drop repeated linkEA entries. */
2676                 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2677
2678                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2679                  * then it is possible that: the directry object has ever
2680                  * been lost, but its name entry was there. In the former
2681                  * LFSCK run, during the first-stage scanning, the LFSCK
2682                  * found the dangling name entry, but it did not recreate
2683                  * the lost object, and when moved to the second-stage
2684                  * scanning, some children objects of the lost directory
2685                  * object were found, then the LFSCK recreated such lost
2686                  * directory object as an orphan.
2687                  *
2688                  * When the LFSCK runs again, if the dangling name is still
2689                  * there, the LFSCK should move the orphan directory object
2690                  * back to the normal namespace. */
2691                 if (!lpf && !fid_is_zero(pfid) &&
2692                     !lu_fid_eq(pfid, &tfid) && once) {
2693                         linkea_next_entry(ldata);
2694                         continue;
2695                 }
2696
2697                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2698                 if (IS_ERR(parent))
2699                         RETURN(PTR_ERR(parent));
2700
2701                 if (!dt_object_exists(parent)) {
2702                         lfsck_object_put(env, parent);
2703                         if (ldata->ld_leh->leh_reccount > 1) {
2704                                 /* If it is NOT the last linkEA entry, then
2705                                  * there is still other chance to make the
2706                                  * child to be visible via other parent, then
2707                                  * remove this linkEA entry. */
2708                                 lfsck_linkea_del_buf(ldata, cname);
2709                                 continue;
2710                         }
2711
2712                         break;
2713                 }
2714
2715                 /* The linkEA entry with bad parent will be removed. */
2716                 if (unlikely(!dt_try_as_dir(env, parent))) {
2717                         lfsck_object_put(env, parent);
2718                         lfsck_linkea_del_buf(ldata, cname);
2719                         continue;
2720                 }
2721
2722                 rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
2723                                (const struct dt_key *)cname->ln_name);
2724                 *pfid2 = *lfsck_dto2fid(parent);
2725                 if (rc == -ENOENT) {
2726                         lfsck_object_put(env, parent);
2727                         linkea_next_entry(ldata);
2728                         continue;
2729                 }
2730
2731                 if (rc != 0) {
2732                         lfsck_object_put(env, parent);
2733
2734                         RETURN(rc);
2735                 }
2736
2737                 if (lu_fid_eq(&tfid, cfid)) {
2738                         lfsck_object_put(env, parent);
2739                         /* If the parent (that is declared via linkEA entry)
2740                          * directory contains the specified child, but such
2741                          * parent does not match the dotdot name entry, then
2742                          * trust the linkEA. */
2743                         if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2744                                 *type = LNIT_UNMATCHED_PAIRS;
2745                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2746                                                 com, child, pfid2, cname);
2747
2748                                 RETURN(rc);
2749                         }
2750
2751 rebuild:
2752                         /* It is the most common case that we find the
2753                          * name entry corresponding to the linkEA entry
2754                          * that matches the ".." name entry. */
2755                         rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2756                                               cname, pfid2);
2757                         if (rc != 0)
2758                                 RETURN(rc);
2759
2760                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2761                                                             &ldata_new);
2762                         if (rc < 0)
2763                                 RETURN(rc);
2764
2765                         lfsck_linkea_del_buf(ldata, cname);
2766                         linkea_first_entry(ldata);
2767                         /* There may be some invalid dangling name entries under
2768                          * other parent directories, remove all of them. */
2769                         while (ldata->ld_lee != NULL) {
2770                                 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2771                                                 cname, &tfid, info->lti_key,
2772                                                 sizeof(info->lti_key));
2773                                 if (rc != 0)
2774                                         goto next;
2775
2776                                 parent = lfsck_object_find_bottom(env, lfsck,
2777                                                                   &tfid);
2778                                 if (IS_ERR(parent)) {
2779                                         rc = PTR_ERR(parent);
2780                                         if (rc != -ENOENT &&
2781                                             bk->lb_param & LPF_FAILOUT)
2782                                                 RETURN(rc);
2783
2784                                         goto next;
2785                                 }
2786
2787                                 if (!dt_object_exists(parent)) {
2788                                         lfsck_object_put(env, parent);
2789                                         goto next;
2790                                 }
2791
2792                                 rc = lfsck_namespace_repair_dirent(env, com,
2793                                         parent, child, cname->ln_name,
2794                                         cname->ln_name, S_IFDIR, false, true);
2795                                 lfsck_object_put(env, parent);
2796                                 if (rc < 0) {
2797                                         if (bk->lb_param & LPF_FAILOUT)
2798                                                 RETURN(rc);
2799
2800                                         goto next;
2801                                 }
2802
2803                                 dirent_count += rc;
2804
2805 next:
2806                                 lfsck_linkea_del_buf(ldata, cname);
2807                         }
2808
2809                         ns->ln_dirent_repaired += dirent_count;
2810
2811                         RETURN(rc);
2812                 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
2813
2814                 lfsck_ibits_unlock(lh, LCK_EX);
2815                 /* The name entry references another MDT-object that may be
2816                  * created by the LFSCK for repairing dangling name entry.
2817                  * Try to replace it. */
2818                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2819                                                   &tfid, cname);
2820                 lfsck_object_put(env, parent);
2821                 if (rc < 0)
2822                         RETURN(rc);
2823
2824                 if (rc > 0)
2825                         goto rebuild;
2826
2827                 lfsck_linkea_del_buf(ldata, cname);
2828         } /* while (ldata->ld_lee != NULL) */
2829
2830         /* If there is still linkEA overflow, return. */
2831         if (unlikely(ldata->ld_leh->leh_overflow_time))
2832                 RETURN(0);
2833
2834         linkea_first_entry(ldata);
2835         if (ldata->ld_leh->leh_reccount == 1) {
2836                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2837                                                 lh, type, NULL, unknown);
2838
2839                 RETURN(rc);
2840         }
2841
2842         /* All linkEA entries are invalid and removed, then handle the @child
2843          * as an orphan.*/
2844         if (ldata->ld_leh->leh_reccount == 0) {
2845                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2846                                                 type);
2847
2848                 RETURN(rc);
2849         }
2850
2851         /* If the dangling name entry for the orphan directory object has
2852          * been remvoed, then just check whether the directory object is
2853          * still under the .lustre/lost+found/MDTxxxx/ or not. */
2854         if (lpf) {
2855                 lpf = false;
2856                 goto again;
2857         }
2858
2859         /* There is no linkEA entry that matches the ".." name entry. Find
2860          * the first linkEA entry that both parent and name entry exist to
2861          * rebuild a new ".." name entry. */
2862         if (once) {
2863                 once = false;
2864                 goto again;
2865         }
2866
2867         RETURN(rc);
2868 }
2869
2870 /**
2871  * Repair the object's nlink attribute.
2872  *
2873  * If all the known name entries have been verified, then the object's hard
2874  * link attribute should match the object's linkEA entries count unless the
2875  * object's has too many hard link to be recorded in the linkEA. Such cases
2876  * should have been marked in the LFSCK trace file. Otherwise, trust the
2877  * linkEA to update the object's nlink attribute.
2878  *
2879  * \param[in] env       pointer to the thread context
2880  * \param[in] com       pointer to the lfsck component
2881  * \param[in] obj       pointer to the dt_object to be handled
2882  * \param[in,out] la    pointer to buffer to object's attribute before
2883  *                      and after the repairing
2884  *
2885  * \retval              positive number for repaired cases
2886  * \retval              0 if nothing to be repaired
2887  * \retval              negative error number on failure
2888  */
2889 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2890                                         struct lfsck_component *com,
2891                                         struct dt_object *obj,
2892                                         struct lu_attr *la)
2893 {
2894         struct lfsck_namespace          *ns     = com->lc_file_ram;
2895         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2896         struct dt_device                *dev    = lfsck_obj2dev(obj);
2897         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
2898         struct thandle                  *th     = NULL;
2899         struct linkea_data               ldata  = { NULL };
2900         struct lustre_handle             lh     = { 0 };
2901         __u32                            old    = la->la_nlink;
2902         int                              rc     = 0;
2903         ENTRY;
2904
2905         LASSERT(!dt_object_remote(obj));
2906
2907         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
2908                               MDS_INODELOCK_UPDATE, LCK_PW);
2909         if (rc != 0)
2910                 GOTO(log, rc);
2911
2912         th = dt_trans_create(env, dev);
2913         if (IS_ERR(th))
2914                 GOTO(log, rc = PTR_ERR(th));
2915
2916         la->la_valid = LA_NLINK;
2917         rc = dt_declare_attr_set(env, obj, la, th);
2918         if (rc != 0)
2919                 GOTO(stop, rc);
2920
2921         rc = dt_trans_start_local(env, dev, th);
2922         if (rc != 0)
2923                 GOTO(stop, rc);
2924
2925         dt_write_lock(env, obj, 0);
2926         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2927          * ever tried to verify some remote MDT-object that resides on this
2928          * MDT, but this MDT failed to respond such request. So means there
2929          * may be some remote name entry on other MDT that references this
2930          * object with another name, so we cannot know whether this linkEA
2931          * is valid or not. So keep it there and maybe resolved when next
2932          * LFSCK run. */
2933         if (ns->ln_flags & LF_INCOMPLETE)
2934                 GOTO(unlock, rc = 0);
2935
2936         rc = dt_attr_get(env, obj, la);
2937         if (rc != 0)
2938                 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2939
2940         rc = lfsck_links_read2_with_rec(env, obj, &ldata);
2941         if (rc)
2942                 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2943
2944         /* XXX: Currently, we only update the nlink attribute if the known
2945          *      linkEA entries is larger than the nlink attribute. That is
2946          *      safe action. */
2947         if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
2948             unlikely(la->la_nlink == 0 ||
2949                      ldata.ld_leh->leh_overflow_time))
2950                 GOTO(unlock, rc = 0);
2951
2952         la->la_nlink = ldata.ld_leh->leh_reccount;
2953         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2954                 GOTO(unlock, rc = 1);
2955
2956         rc = dt_attr_set(env, obj, la, th);
2957
2958         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2959
2960 unlock:
2961         dt_write_unlock(env, obj);
2962
2963 stop:
2964         dt_trans_stop(env, dev, th);
2965
2966 log:
2967         lfsck_ibits_unlock(&lh, LCK_PW);
2968         CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2969                "nlink count from %u to %u: rc = %d\n",
2970                lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2971
2972         if (rc != 0)
2973                 ns->ln_flags |= LF_INCONSISTENT;
2974
2975         return rc;
2976 }
2977
2978 /**
2979  * Double scan the directory object for namespace LFSCK.
2980  *
2981  * This function will verify the <parent, child> pairs in the namespace tree:
2982  * the parent references the child via some name entry that should be in the
2983  * child's linkEA entry, the child should back references the parent via its
2984  * ".." name entry.
2985  *
2986  * The LFSCK will scan every linkEA entry in turn until find out the first
2987  * matched pairs. If found, then all other linkEA entries will be dropped.
2988  * If all the linkEA entries cannot match the ".." name entry, then there
2989  * are serveral possible cases:
2990  *
2991  * 1) If there is only one linkEA entry, then trust it as long as the PFID
2992  *    in the linkEA entry is valid.
2993  *
2994  * 2) If there are multiple linkEA entries, then try to find the linkEA
2995  *    that matches the ".." name entry. If found, then all other entries
2996  *    are invalid; otherwise, it is quite possible that the ".." name entry
2997  *    is corrupted. Under such case, the LFSCK will rebuild the ".." name
2998  *    entry according to the first valid linkEA entry (both the parent and
2999  *    the name entry should exist).
3000  *
3001  * 3) If the directory object has no (valid) linkEA entry, then the
3002  *    directory object will be handled as pure orphan and inserted
3003  *    in the .lustre/lost+found/MDTxxxx/ with the name:
3004  *    ${self_FID}-${PFID}-D-${conflict_version}
3005  *
3006  * \param[in] env       pointer to the thread context
3007  * \param[in] com       pointer to the lfsck component
3008  * \param[in] child     pointer to the directory object to be handled
3009  * \param[in] flags     to indicate the specical checking on the @child
3010  *
3011  * \retval              positive number for repaired cases
3012  * \retval              0 if nothing to be repaired
3013  * \retval              negative error number on failure
3014  */
3015 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3016                                            struct lfsck_component *com,
3017                                            struct dt_object *child, __u8 flags)
3018 {
3019         struct lfsck_thread_info *info          = lfsck_env_info(env);
3020         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
3021         struct lu_fid            *pfid          = &info->lti_fid2;
3022         struct lfsck_namespace   *ns            = com->lc_file_ram;
3023         struct lfsck_instance    *lfsck         = com->lc_lfsck;
3024         struct lustre_handle      lh            = { 0 };
3025         struct linkea_data        ldata         = { NULL };
3026         bool                      unknown       = false;
3027         bool                      lpf           = false;
3028         bool                      retry         = false;
3029         enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3030         int                       rc            = 0;
3031         ENTRY;
3032
3033         LASSERT(!dt_object_remote(child));
3034
3035         if (flags & LNTF_UNCERTAIN_LMV) {
3036                 if (flags & LNTF_RECHECK_NAME_HASH) {
3037                         rc = lfsck_namespace_scan_shard(env, com, child);
3038                         if (rc < 0)
3039                                 RETURN(rc);
3040
3041                         ns->ln_striped_shards_scanned++;
3042                 } else {
3043                         ns->ln_striped_shards_skipped++;
3044                 }
3045         }
3046
3047         flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3048         if (flags == 0)
3049                 RETURN(0);
3050
3051         if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3052             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3053                 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3054                        "the namespace LFSCK, then the LFSCK cannot guarantee"
3055                        "all the name entries have been verified in first-stage"
3056                        "scanning. So have to skip orphan related handling for"
3057                        "the directory object "DFID" with remote name entry\n",
3058                        lfsck_lfsck2name(lfsck), PFID(cfid));
3059
3060                 RETURN(0);
3061         }
3062
3063         if (unlikely(!dt_try_as_dir(env, child)))
3064                 GOTO(out, rc = -ENOTDIR);
3065
3066         /* We only take ldlm lock on the @child when required. When the
3067          * logic comes here for the first time, it is always false. */
3068         if (0) {
3069
3070 lock:
3071                 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3072                                       MDS_INODELOCK_UPDATE |
3073                                       MDS_INODELOCK_XATTR, LCK_EX);
3074                 if (rc != 0)
3075                         GOTO(out, rc);
3076         }
3077
3078         dt_read_lock(env, child, 0);
3079         if (unlikely(lfsck_is_dead_obj(child))) {
3080                 dt_read_unlock(env, child);
3081
3082                 GOTO(out, rc = 0);
3083         }
3084
3085         rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3086                        (const struct dt_key *)dotdot);
3087         if (rc != 0) {
3088                 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3089                         dt_read_unlock(env, child);
3090
3091                         GOTO(out, rc);
3092                 }
3093
3094                 if (!lustre_handle_is_used(&lh)) {
3095                         dt_read_unlock(env, child);
3096                         goto lock;
3097                 }
3098
3099                 fid_zero(pfid);
3100         } else if (lfsck->li_lpf_obj != NULL &&
3101                    lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3102                 lpf = true;
3103         } else if (unlikely(!fid_is_sane(pfid))) {
3104                 fid_zero(pfid);
3105         }
3106
3107         rc = lfsck_links_read(env, child, &ldata);
3108         dt_read_unlock(env, child);
3109         if (rc != 0) {
3110                 if (rc != -ENODATA && rc != -EINVAL)
3111                         GOTO(out, rc);
3112
3113                 if (!lustre_handle_is_used(&lh))
3114                         goto lock;
3115
3116                 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3117                         /* Remove the corrupted linkEA. */
3118                         rc = lfsck_namespace_links_remove(env, com, child);
3119                         if (rc == 0)
3120                                 /* Here, because of the crashed linkEA, we
3121                                  * cannot know whether there is some parent
3122                                  * that references the child directory via
3123                                  * some name entry or not. So keep it there,
3124                                  * when the LFSCK run next time, if there is
3125                                  * some parent that references this object,
3126                                  * then the LFSCK can rebuild the linkEA;
3127                                  * otherwise, this object will be handled
3128                                  * as orphan as above. */
3129                                 unknown = true;
3130                 } else {
3131                         /* 1. If we have neither ".." nor linkEA,
3132                          *    then it is an orphan.
3133                          *
3134                          * 2. If we only have the ".." name entry,
3135                          *    but no parent references this child
3136                          *    directory, then handle it as orphan. */
3137                         lfsck_ibits_unlock(&lh, LCK_EX);
3138                         type = LNIT_MUL_REF;
3139
3140                         /* If the LFSCK is marked as LF_INCOMPLETE,
3141                          * then means some MDT has ever tried to
3142                          * verify some remote MDT-object that resides
3143                          * on this MDT, but this MDT failed to respond
3144                          * such request. So means there may be some
3145                          * remote name entry on other MDT that
3146                          * references this object with another name,
3147                          * so we cannot know whether this linkEA is
3148                          * valid or not. So keep it there and maybe
3149                          * resolved when next LFSCK run. */
3150                         if (ns->ln_flags & LF_INCOMPLETE)
3151                                 GOTO(out, rc = 0);
3152
3153                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3154                                  "-"DFID, PFID(pfid));
3155                         rc = lfsck_namespace_insert_orphan(env, com, child,
3156                                                 info->lti_tmpbuf, "D", NULL);
3157                 }
3158
3159                 GOTO(out, rc);
3160         } /* rc != 0 */
3161
3162         linkea_first_entry(&ldata);
3163         /* This is the most common case: the object has unique linkEA entry. */
3164         if (ldata.ld_leh->leh_reccount == 1) {
3165                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3166                                                 &lh, &type, &retry, &unknown);
3167                 if (retry) {
3168                         LASSERT(!lustre_handle_is_used(&lh));
3169
3170                         retry = false;
3171                         goto lock;
3172                 }
3173
3174                 GOTO(out, rc);
3175         }
3176
3177         if (!lustre_handle_is_used(&lh))
3178                 goto lock;
3179
3180         if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3181                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3182                                                 &type);
3183
3184                 GOTO(out, rc);
3185         }
3186
3187         /* When we come here, the cases usually like that:
3188          * 1) The directory object has a corrupted linkEA entry. During the
3189          *    first-stage scanning, the LFSCK cannot know such corruption,
3190          *    then it appends the right linkEA entry according to the found
3191          *    name entry after the bad one.
3192          *
3193          * 2) The directory object has a right linkEA entry. During the
3194          *    first-stage scanning, the LFSCK finds some bad name entry,
3195          *    but the LFSCK cannot aware that at that time, then it adds
3196          *    the bad linkEA entry for further processing. */
3197         rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3198                                           &lh, &type, lpf, &unknown);
3199
3200         GOTO(out, rc);
3201
3202 out:
3203         lfsck_ibits_unlock(&lh, LCK_EX);
3204         if (rc > 0) {
3205                 switch (type) {
3206                 case LNIT_BAD_LINKEA:
3207                         ns->ln_linkea_repaired++;
3208                         break;
3209                 case LNIT_UNMATCHED_PAIRS:
3210                         ns->ln_unmatched_pairs_repaired++;
3211                         break;
3212                 case LNIT_MUL_REF:
3213                         ns->ln_mul_ref_repaired++;
3214                         break;
3215                 default:
3216                         break;
3217                 }
3218         }
3219
3220         if (unknown)
3221                 ns->ln_unknown_inconsistency++;
3222
3223         return rc;
3224 }
3225
3226 #define lfsck_time_before(a, b)         \
3227         (typecheck(__u32, a) &&         \
3228          typecheck(__u32, b) &&         \
3229          ((int)(a) - (int)(b) < 0))
3230
3231 static inline bool
3232 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3233                                       struct lfsck_namespace *ns)
3234 {
3235         /* Both the leh_overflow_time and ln_time_latest_reset are
3236          * local time based, so need NOT to care about clock drift
3237          * among the servers. */
3238         return ldata->ld_leh->leh_overflow_time &&
3239                lfsck_time_before(ldata->ld_leh->leh_overflow_time,
3240                                  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  * Double scan the MDT-object for namespace LFSCK.
3339  *
3340  * If the MDT-object contains invalid or repeated linkEA entries, then drop
3341  * those entries from the linkEA; if the linkEA becomes empty or the object
3342  * has no linkEA, then it is an orphan and will be added into the directory
3343  * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3344  * the remote parent; if the name entry corresponding to some linkEA entry
3345  * is lost, then add the name entry back to the namespace.
3346  *
3347  * \param[in] env       pointer to the thread context
3348  * \param[in] com       pointer to the lfsck component
3349  * \param[in] child     pointer to the dt_object to be handled
3350  * \param[in] flags     some hints to indicate how the @child should be handled
3351  *
3352  * \retval              positive number for repaired cases
3353  * \retval              0 if nothing to be repaired
3354  * \retval              negative error number on failure
3355  */
3356 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3357                                            struct lfsck_component *com,
3358                                            struct dt_object *child, __u8 flags)
3359 {
3360         struct lfsck_thread_info *info     = lfsck_env_info(env);
3361         struct lu_attr           *la       = &info->lti_la;
3362         struct lu_name           *cname    = &info->lti_name;
3363         struct lu_fid            *pfid     = &info->lti_fid;
3364         struct lu_fid            *cfid     = &info->lti_fid2;
3365         struct lfsck_instance    *lfsck    = com->lc_lfsck;
3366         struct lfsck_namespace   *ns       = com->lc_file_ram;
3367         struct dt_object         *parent   = NULL;
3368         struct linkea_data        ldata    = { NULL };
3369         bool                      repaired = false;
3370         int                       count    = 0;
3371         int                       rc;
3372         ENTRY;
3373
3374         dt_read_lock(env, child, 0);
3375         if (unlikely(lfsck_is_dead_obj(child))) {
3376                 dt_read_unlock(env, child);
3377
3378                 RETURN(0);
3379         }
3380
3381         if (S_ISDIR(lfsck_object_type(child))) {
3382                 dt_read_unlock(env, child);
3383                 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3384
3385                 RETURN(rc);
3386         }
3387
3388         rc = lfsck_links_read(env, child, &ldata);
3389         dt_read_unlock(env, child);
3390
3391         if (rc == -EINVAL) {
3392                 struct lustre_handle lh = { 0 };
3393
3394                 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3395                                       MDS_INODELOCK_UPDATE |
3396                                       MDS_INODELOCK_XATTR, LCK_EX);
3397                 if (rc == 0) {
3398                         rc = lfsck_namespace_links_remove(env, com, child);
3399                         lfsck_ibits_unlock(&lh, LCK_EX);
3400                 }
3401
3402                 GOTO(out, rc);
3403         }
3404
3405         if (rc != 0)
3406                 GOTO(out, rc);
3407
3408         if (!(ns->ln_flags & LF_INCOMPLETE) &&
3409             unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3410                 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3411                                                            child);
3412                 if (rc < 0)
3413                         GOTO(out, rc);
3414
3415                 if (rc > 0)
3416                         ns->ln_linkea_overflow_cleared++;
3417         }
3418
3419         linkea_first_entry(&ldata);
3420         while (ldata.ld_lee != NULL) {
3421                 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3422                                                          info->lti_key,
3423                                                          sizeof(info->lti_key));
3424                 /* Invalid PFID in the linkEA entry. */
3425                 if (rc != 0) {
3426                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3427                                                 &ldata, cname, pfid, true);
3428                         if (rc < 0)
3429                                 GOTO(out, rc);
3430
3431                         if (rc > 0)
3432                                 repaired = true;
3433
3434                         continue;
3435                 }
3436
3437                 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3438                                                          false);
3439                 /* Found repeated linkEA entries */
3440                 if (rc > 0) {
3441                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3442                                                 &ldata, cname, pfid, false);
3443                         if (rc < 0)
3444                                 GOTO(out, rc);
3445
3446                         if (rc == 0)
3447                                 continue;
3448
3449                         repaired = true;
3450
3451                         /* fall through */
3452                 }
3453
3454                 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3455                 if (IS_ERR(parent))
3456                         GOTO(out, rc = PTR_ERR(parent));
3457
3458                 if (!dt_object_exists(parent)) {
3459
3460 lost_parent:
3461                         if (ldata.ld_leh->leh_reccount > 1) {
3462                                 /* If it is NOT the last linkEA entry, then
3463                                  * there is still other chance to make the
3464                                  * child to be visible via other parent, then
3465                                  * remove this linkEA entry. */
3466                                 rc = lfsck_namespace_shrink_linkea(env, com,
3467                                         child, &ldata, cname, pfid, true);
3468                         } else {
3469                                 /* If the LFSCK is marked as LF_INCOMPLETE,
3470                                  * then means some MDT has ever tried to
3471                                  * verify some remote MDT-object that resides
3472                                  * on this MDT, but this MDT failed to respond
3473                                  * such request. So means there may be some
3474                                  * remote name entry on other MDT that
3475                                  * references this object with another name,
3476                                  * so we cannot know whether this linkEA is
3477                                  * valid or not. So keep it there and maybe
3478                                  * resolved when next LFSCK run. */
3479                                 if (ns->ln_flags & LF_INCOMPLETE) {
3480                                         lfsck_object_put(env, parent);
3481
3482                                         GOTO(out, rc = 0);
3483                                 }
3484
3485                                 /* Create the lost parent as an orphan. */
3486                                 rc = lfsck_namespace_create_orphan_dir(env, com,
3487                                                                 parent, NULL);
3488                                 if (rc < 0) {
3489                                         lfsck_object_put(env, parent);
3490
3491                                         GOTO(out, rc);
3492                                 }
3493
3494                                 if (rc > 0)
3495                                         repaired = true;
3496
3497                                 /* Add the missing name entry to the parent. */
3498                                 rc = lfsck_namespace_insert_normal(env, com,
3499                                                 parent, child, cname->ln_name);
3500                                 if (unlikely(rc == -EEXIST))
3501                                         /* Unfortunately, someone reused the
3502                                          * name under the parent by race. So we
3503                                          * have to remove the linkEA entry from
3504                                          * current child object. It means that
3505                                          * the LFSCK cannot recover the system
3506                                          * totally back to its original status,
3507                                          * but it is necessary to make the
3508                                          * current system to be consistent. */
3509                                         rc = lfsck_namespace_shrink_linkea(env,
3510                                                         com, child, &ldata,
3511                                                         cname, pfid, true);
3512                                 else
3513                                         linkea_next_entry(&ldata);
3514                         }
3515
3516                         lfsck_object_put(env, parent);
3517                         if (rc < 0)
3518                                 GOTO(out, rc);
3519
3520                         if (rc > 0)
3521                                 repaired = true;
3522
3523                         continue;
3524                 } /* !dt_object_exists(parent) */
3525
3526                 /* The linkEA entry with bad parent will be removed. */
3527                 if (unlikely(!dt_try_as_dir(env, parent))) {
3528                         lfsck_object_put(env, parent);
3529                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3530                                                 &ldata, cname, pfid, true);
3531                         if (rc < 0)
3532                                 GOTO(out, rc);
3533
3534                         if (rc > 0)
3535                                 repaired = true;
3536
3537                         continue;
3538                 }
3539
3540                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3541                                (const struct dt_key *)cname->ln_name);
3542                 if (rc != 0 && rc != -ENOENT) {
3543                         lfsck_object_put(env, parent);
3544
3545                         GOTO(out, rc);
3546                 }
3547
3548                 if (rc == 0) {
3549                         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3550                                 /* It is the most common case that we
3551                                  * find the name entry corresponding
3552                                  * to the linkEA entry. */
3553                                 lfsck_object_put(env, parent);
3554                                 linkea_next_entry(&ldata);
3555                         } else {
3556                                 /* The name entry references another
3557                                  * MDT-object that may be created by
3558                                  * the LFSCK for repairing dangling
3559                                  * name entry. Try to replace it. */
3560                                 rc = lfsck_namespace_replace_cond(env, com,
3561                                                 parent, child, cfid, cname);
3562                                 lfsck_object_put(env, parent);
3563                                 if (rc < 0)
3564                                         GOTO(out, rc);
3565
3566                                 if (rc > 0) {
3567                                         repaired = true;
3568                                         linkea_next_entry(&ldata);
3569                                 } else {
3570                                         rc = lfsck_namespace_shrink_linkea(env,
3571                                                         com, child, &ldata,
3572                                                         cname, pfid, true);
3573                                         if (rc < 0)
3574                                                 GOTO(out, rc);
3575
3576                                         if (rc > 0)
3577                                                 repaired = true;
3578                                 }
3579                         }
3580
3581                         continue;
3582                 }
3583
3584                 /* The following handles -ENOENT case */
3585
3586                 rc = dt_attr_get(env, child, la);
3587                 if (rc != 0)
3588                         GOTO(out, rc);
3589
3590                 /* If there is no name entry in the parent dir and the object
3591                  * link count is fewer than the linkea entries count, then the
3592                  * linkea entry should be removed. */
3593                 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3594                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3595                                         parent, child, &ldata, cname, pfid);
3596                         lfsck_object_put(env, parent);
3597                         if (rc < 0)
3598                                 GOTO(out, rc);
3599
3600                         if (rc > 0)
3601                                 repaired = true;
3602
3603                         continue;
3604                 }
3605
3606                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3607                  * MDT has ever tried to verify some remote MDT-object that
3608                  * resides on this MDT, but this MDT failed to respond such
3609                  * request. So means there may be some remote name entry on
3610                  * other MDT that references this object with another name,
3611                  * so we cannot know whether this linkEA is valid or not.
3612                  * So keep it there and maybe resolved when next LFSCK run. */
3613                 if (ns->ln_flags & LF_INCOMPLETE) {
3614                         lfsck_object_put(env, parent);
3615
3616                         GOTO(out, rc = 0);
3617                 }
3618
3619                 rc = lfsck_namespace_check_name(env, parent, child, cname);
3620                 if (rc == -ENOENT)
3621                         goto lost_parent;
3622
3623                 if (rc < 0) {
3624                         lfsck_object_put(env, parent);
3625
3626                         GOTO(out, rc);
3627                 }
3628
3629                 /* It is an invalid name entry, drop it. */
3630                 if (unlikely(rc > 0)) {
3631                         lfsck_object_put(env, parent);
3632                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3633                                                 &ldata, cname, pfid, true);
3634                         if (rc < 0)
3635                                 GOTO(out, rc);
3636
3637                         if (rc > 0)
3638                                 repaired = true;
3639
3640                         continue;
3641                 }
3642
3643                 /* Add the missing name entry back to the namespace. */
3644                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3645                                                    cname->ln_name);
3646                 if (unlikely(rc == -ESTALE))
3647                         /* It may happen when the remote object has been
3648                          * removed, but the local MDT is not aware of that. */
3649                         goto lost_parent;
3650
3651                 if (unlikely(rc == -EEXIST))
3652                         /* Unfortunately, someone reused the name under the
3653                          * parent by race. So we have to remove the linkEA
3654                          * entry from current child object. It means that the
3655                          * LFSCK cannot recover the system totally back to
3656                          * its original status, but it is necessary to make
3657                          * the current system to be consistent.
3658                          *
3659                          * It also may be because of the LFSCK found some
3660                          * internal status of create operation. Under such
3661                          * case, nothing to be done. */
3662                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3663                                         parent, child, &ldata, cname, pfid);
3664                 else
3665                         linkea_next_entry(&ldata);
3666
3667                 lfsck_object_put(env, parent);
3668                 if (rc < 0)
3669                         GOTO(out, rc);
3670
3671                 if (rc > 0)
3672                         repaired = true;
3673         }
3674
3675         GOTO(out, rc = 0);
3676
3677 out:
3678         if (rc < 0 && rc != -ENODATA)
3679                 return rc;
3680
3681         if (rc == 0 && ldata.ld_leh != NULL)
3682                 count = ldata.ld_leh->leh_reccount;
3683
3684         if (count == 0) {
3685                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3686                  * MDT has ever tried to verify some remote MDT-object that
3687                  * resides on this MDT, but this MDT failed to respond such
3688                  * request. So means there may be some remote name entry on
3689                  * other MDT that references this object with another name,
3690                  * so we cannot know whether this linkEA is valid or not.
3691                  * So keep it there and maybe resolved when next LFSCK run. */
3692                 if (!(ns->ln_flags & LF_INCOMPLETE) &&
3693                     (ldata.ld_leh == NULL ||
3694                      !ldata.ld_leh->leh_overflow_time)) {
3695                         /* If the child becomes orphan, then insert it into
3696                          * the global .lustre/lost+found/MDTxxxx directory. */
3697                         rc = lfsck_namespace_insert_orphan(env, com, child,
3698                                                            "", "O", &count);
3699                         if (rc < 0)
3700                                 return rc;
3701
3702                         if (rc > 0) {
3703                                 ns->ln_mul_ref_repaired++;
3704                                 repaired = true;
3705                         }
3706                 }
3707         } else {
3708                 rc = dt_attr_get(env, child, la);
3709                 if (rc != 0)
3710                         return rc;
3711
3712                 if (la->la_nlink != 0 && la->la_nlink != count) {
3713                         if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
3714                                      !S_ISLNK(lfsck_object_type(child)))) {
3715                                 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
3716                                        "the object "DFID"'s nlink count %d "
3717                                        "does not match linkEA count %d, "
3718                                        "type %o, skip it.\n",
3719                                        lfsck_lfsck2name(lfsck),
3720                                        PFID(lfsck_dto2fid(child)),
3721                                        la->la_nlink, count,
3722                                        lfsck_object_type(child));
3723                         } else if (la->la_nlink < count &&
3724                                    likely(!ldata.ld_leh->leh_overflow_time)) {
3725                                 rc = lfsck_namespace_repair_nlink(env, com,
3726                                                                   child, la);
3727                                 if (rc > 0) {
3728                                         ns->ln_objs_nlink_repaired++;
3729                                         rc = 0;
3730                                 }
3731                         }
3732                 }
3733         }
3734
3735         if (repaired) {
3736                 if (la->la_nlink > 1)
3737                         ns->ln_mul_linked_repaired++;
3738
3739                 if (rc == 0)
3740                         rc = 1;
3741         }
3742
3743         return rc;
3744 }
3745
3746 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3747                                             struct lfsck_namespace *ns,
3748                                             __u64 checked_phase1,
3749                                             __u64 checked_phase2,
3750                                             __u32 time_phase1,
3751                                             __u32 time_phase2)
3752 {
3753         seq_printf(m, "checked_phase1: %llu\n"
3754                    "checked_phase2: %llu\n"
3755                    "updated_phase1: %llu\n"
3756                    "updated_phase2: %llu\n"
3757                    "failed_phase1: %llu\n"
3758                    "failed_phase2: %llu\n"
3759                    "directories: %llu\n"
3760                    "dirent_repaired: %llu\n"
3761                    "linkea_repaired: %llu\n"
3762                    "nlinks_repaired: %llu\n"
3763                    "multiple_linked_checked: %llu\n"
3764                    "multiple_linked_repaired: %llu\n"
3765                    "unknown_inconsistency: %llu\n"
3766                    "unmatched_pairs_repaired: %llu\n"
3767                    "dangling_repaired: %llu\n"
3768                    "multiple_referenced_repaired: %llu\n"
3769                    "bad_file_type_repaired: %llu\n"
3770                    "lost_dirent_repaired: %llu\n"
3771                    "local_lost_found_scanned: %llu\n"
3772                    "local_lost_found_moved: %llu\n"
3773                    "local_lost_found_skipped: %llu\n"
3774                    "local_lost_found_failed: %llu\n"
3775                    "striped_dirs_scanned: %llu\n"
3776                    "striped_dirs_repaired: %llu\n"
3777                    "striped_dirs_failed: %llu\n"
3778                    "striped_dirs_disabled: %llu\n"
3779                    "striped_dirs_skipped: %llu\n"
3780                    "striped_shards_scanned: %llu\n"
3781                    "striped_shards_repaired: %llu\n"
3782                    "striped_shards_failed: %llu\n"
3783                    "striped_shards_skipped: %llu\n"
3784                    "name_hash_repaired: %llu\n"
3785                    "linkea_overflow_cleared: %llu\n"
3786                    "success_count: %u\n"
3787                    "run_time_phase1: %u seconds\n"
3788                    "run_time_phase2: %u seconds\n",
3789                    checked_phase1,
3790                    checked_phase2,
3791                    ns->ln_items_repaired,
3792                    ns->ln_objs_repaired_phase2,
3793                    ns->ln_items_failed,
3794                    ns->ln_objs_failed_phase2,
3795                    ns->ln_dirs_checked,
3796                    ns->ln_dirent_repaired,
3797                    ns->ln_linkea_repaired,
3798                    ns->ln_objs_nlink_repaired,
3799                    ns->ln_mul_linked_checked,
3800                    ns->ln_mul_linked_repaired,
3801                    ns->ln_unknown_inconsistency,
3802                    ns->ln_unmatched_pairs_repaired,
3803                    ns->ln_dangling_repaired,
3804                    ns->ln_mul_ref_repaired,
3805                    ns->ln_bad_type_repaired,
3806                    ns->ln_lost_dirent_repaired,
3807                    ns->ln_local_lpf_scanned,
3808                    ns->ln_local_lpf_moved,
3809                    ns->ln_local_lpf_skipped,
3810                    ns->ln_local_lpf_failed,
3811                    ns->ln_striped_dirs_scanned,
3812                    ns->ln_striped_dirs_repaired,
3813                    ns->ln_striped_dirs_failed,
3814                    ns->ln_striped_dirs_disabled,
3815                    ns->ln_striped_dirs_skipped,
3816                    ns->ln_striped_shards_scanned,
3817                    ns->ln_striped_shards_repaired,
3818                    ns->ln_striped_shards_failed,
3819                    ns->ln_striped_shards_skipped,
3820                    ns->ln_name_hash_repaired,
3821                    ns->ln_linkea_overflow_cleared,
3822                    ns->ln_success_count,
3823                    time_phase1,
3824                    time_phase2);
3825 }
3826
3827 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3828                                         struct lfsck_component *com)
3829 {
3830         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3831         struct lfsck_namespace          *ns     = com->lc_file_ram;
3832
3833         while (!list_empty(&lfsck->li_list_lmv)) {
3834                 struct lfsck_lmv_unit   *llu;
3835                 struct lfsck_lmv        *llmv;
3836
3837                 llu = list_entry(lfsck->li_list_lmv.next,
3838                                  struct lfsck_lmv_unit, llu_link);
3839                 llmv = &llu->llu_lmv;
3840
3841                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3842                          "still in using: %u\n",
3843                          atomic_read(&llmv->ll_ref));
3844
3845                 ns->ln_striped_dirs_skipped++;
3846                 lfsck_lmv_put(env, llmv);
3847         }
3848 }
3849
3850 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
3851                                                  struct lfsck_component *com,
3852                                                  struct dt_object *obj)
3853 {
3854         struct lu_attr *la = &lfsck_env_info(env)->lti_la;
3855         int             rc;
3856
3857         rc = dt_attr_get(env, obj, la);
3858         if (rc != 0)
3859                 return rc;
3860
3861         /* zero-linkEA object may be orphan, but it also maybe because
3862          * of upgrading. Currently, we cannot record it for double scan.
3863          * Because it may cause the LFSCK trace file to be too large. */
3864
3865         /* "la_ctime" == 1 means that it has ever been removed from
3866          * backend /lost+found directory but not been added back to
3867          * the normal namespace yet. */
3868
3869         if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
3870             unlikely(la->la_ctime == 1))
3871                 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
3872                                                   LNTF_CHECK_LINKEA, true);
3873
3874         return rc;
3875 }
3876
3877 /* namespace APIs */
3878
3879 static int lfsck_namespace_reset(const struct lu_env *env,
3880                                  struct lfsck_component *com, bool init)
3881 {
3882         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3883         struct lfsck_namespace          *ns     = com->lc_file_ram;
3884         struct lfsck_assistant_data     *lad    = com->lc_data;
3885         struct dt_object                *root;
3886         int                              rc;
3887         ENTRY;
3888
3889         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3890         if (IS_ERR(root))
3891                 GOTO(log, rc = PTR_ERR(root));
3892
3893         if (unlikely(!dt_try_as_dir(env, root)))
3894                 GOTO(put, rc = -ENOTDIR);
3895
3896         down_write(&com->lc_sem);
3897         if (init) {
3898                 memset(ns, 0, sizeof(*ns));
3899         } else {
3900                 __u32 count = ns->ln_success_count;
3901                 __u64 last_time = ns->ln_time_last_complete;
3902
3903                 memset(ns, 0, sizeof(*ns));
3904                 ns->ln_success_count = count;
3905                 ns->ln_time_last_complete = last_time;
3906         }
3907         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3908         ns->ln_status = LS_INIT;
3909         ns->ln_time_latest_reset = cfs_time_current_sec();
3910
3911         rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
3912                                        &dt_lfsck_namespace_features,
3913                                        LFSCK_NAMESPACE, true);
3914         if (rc)
3915                 GOTO(out, rc);
3916
3917         rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
3918                                         LFSCK_NAMESPACE, true);
3919         if (rc != 0)
3920                 GOTO(out, rc);
3921
3922         lad->lad_incomplete = 0;
3923         CFS_RESET_BITMAP(lad->lad_bitmap);
3924
3925         rc = lfsck_namespace_store(env, com);
3926
3927         GOTO(out, rc);
3928
3929 out:
3930         up_write(&com->lc_sem);
3931
3932 put:
3933         lfsck_object_put(env, root);
3934 log:
3935         CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3936                lfsck_lfsck2name(lfsck), rc);
3937         return rc;
3938 }
3939
3940 static void
3941 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3942                      bool new_checked)
3943 {
3944         struct lfsck_namespace *ns = com->lc_file_ram;
3945
3946         down_write(&com->lc_sem);
3947         if (new_checked)
3948                 com->lc_new_checked++;
3949         lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3950         up_write(&com->lc_sem);
3951 }
3952
3953 static void lfsck_namespace_close_dir(const struct lu_env *env,
3954                                       struct lfsck_component *com)
3955 {
3956         struct lfsck_namespace          *ns     = com->lc_file_ram;
3957         struct lfsck_assistant_data     *lad    = com->lc_data;
3958         struct lfsck_assistant_object   *lso    = NULL;
3959         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3960         struct lfsck_lmv                *llmv   = lfsck->li_lmv;
3961         struct lfsck_namespace_req      *lnr;
3962         __u32                            size   =
3963                                 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3964         bool                             wakeup = false;
3965         ENTRY;
3966
3967         if (llmv == NULL)
3968                 RETURN_EXIT;
3969
3970         OBD_ALLOC(lnr, size);
3971         if (lnr == NULL) {
3972                 ns->ln_striped_dirs_skipped++;
3973
3974                 RETURN_EXIT;
3975         }
3976
3977         lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
3978                         NULL, lfsck->li_pos_current.lp_oit_cookie, true);
3979         if (IS_ERR(lso)) {
3980                 OBD_FREE(lnr, size);
3981                 ns->ln_striped_dirs_skipped++;
3982
3983                 RETURN_EXIT;
3984         }
3985
3986         /* Generate a dummy request to indicate that all shards' name entry
3987          * in this striped directory has been scanned for the first time. */
3988         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3989         lnr->lnr_lar.lar_parent = lso;
3990         lnr->lnr_lmv = lfsck_lmv_get(llmv);
3991         lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3992         lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3993         lnr->lnr_size = size;
3994
3995         spin_lock(&lad->lad_lock);
3996         if (lad->lad_assistant_status < 0) {
3997                 spin_unlock(&lad->lad_lock);
3998                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3999                 ns->ln_striped_dirs_skipped++;
4000
4001                 RETURN_EXIT;
4002         }
4003
4004         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4005         if (lad->lad_prefetched == 0)
4006                 wakeup = true;
4007
4008         lad->lad_prefetched++;
4009         spin_unlock(&lad->lad_lock);
4010         if (wakeup)
4011                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4012
4013         EXIT;
4014 }
4015
4016 static int lfsck_namespace_open_dir(const struct lu_env *env,
4017                                     struct lfsck_component *com)
4018 {
4019         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4020         struct lfsck_namespace  *ns     = com->lc_file_ram;
4021         struct lfsck_lmv        *llmv   = lfsck->li_lmv;
4022         int                      rc     = 0;
4023         ENTRY;
4024
4025         if (llmv == NULL)
4026                 RETURN(0);
4027
4028         if (llmv->ll_lmv_master) {
4029                 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4030
4031                 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4032                         lmv->lmv_master_mdt_index =
4033                                 lfsck_dev_idx(lfsck);
4034                         ns->ln_flags |= LF_INCONSISTENT;
4035                         llmv->ll_lmv_updated = 1;
4036                 }
4037         } else {
4038                 rc = lfsck_namespace_verify_stripe_slave(env, com,
4039                                         lfsck->li_obj_dir, llmv);
4040         }
4041
4042         RETURN(rc > 0 ? 0 : rc);
4043 }
4044
4045 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4046                                       struct lfsck_component *com, bool init)
4047 {
4048         struct lfsck_instance   *lfsck = com->lc_lfsck;
4049         struct lfsck_namespace  *ns    = com->lc_file_ram;
4050         int                      rc;
4051
4052         if (!init) {
4053                 rc = lfsck_checkpoint_generic(env, com);
4054                 if (rc != 0)
4055                         goto log;
4056         }
4057
4058         down_write(&com->lc_sem);
4059         if (init) {
4060                 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4061         } else {
4062                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4063                 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4064                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4065                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4066                 ns->ln_items_checked += com->lc_new_checked;
4067                 com->lc_new_checked = 0;
4068         }
4069
4070         rc = lfsck_namespace_store(env, com);
4071         up_write(&com->lc_sem);
4072
4073 log:
4074         CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4075                ", "DFID", %#llx], status = %d: rc = %d\n",
4076                lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4077                PFID(&lfsck->li_pos_current.lp_dir_parent),
4078                lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4079
4080         return rc > 0 ? 0 : rc;
4081 }
4082
4083 static int lfsck_namespace_prep(const struct lu_env *env,
4084                                 struct lfsck_component *com,
4085                                 struct lfsck_start_param *lsp)
4086 {
4087         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4088         struct lfsck_namespace  *ns     = com->lc_file_ram;
4089         struct lfsck_position   *pos    = &com->lc_pos_start;
4090         int                      rc;
4091
4092         rc = lfsck_namespace_load_bitmap(env, com);
4093         if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4094                 rc = lfsck_namespace_reset(env, com, false);
4095                 if (rc == 0)
4096                         rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4097
4098                 if (rc != 0) {
4099                         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4100                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4101
4102                         return rc;
4103                 }
4104         }
4105
4106         down_write(&com->lc_sem);
4107         ns->ln_time_latest_start = cfs_time_current_sec();
4108         spin_lock(&lfsck->li_lock);
4109
4110         if (ns->ln_flags & LF_SCANNED_ONCE) {
4111                 if (!lfsck->li_drop_dryrun ||
4112                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4113                         ns->ln_status = LS_SCANNING_PHASE2;
4114                         list_move_tail(&com->lc_link,
4115                                        &lfsck->li_list_double_scan);
4116                         if (!list_empty(&com->lc_link_dir))
4117                                 list_del_init(&com->lc_link_dir);
4118                         lfsck_pos_set_zero(pos);
4119                 } else {
4120                         ns->ln_status = LS_SCANNING_PHASE1;
4121                         ns->ln_run_time_phase1 = 0;
4122                         ns->ln_run_time_phase2 = 0;
4123                         ns->ln_items_checked = 0;
4124                         ns->ln_items_repaired = 0;
4125                         ns->ln_items_failed = 0;
4126                         ns->ln_dirs_checked = 0;
4127                         ns->ln_objs_checked_phase2 = 0;
4128                         ns->ln_objs_repaired_phase2 = 0;
4129                         ns->ln_objs_failed_phase2 = 0;
4130                         ns->ln_objs_nlink_repaired = 0;
4131                         ns->ln_dirent_repaired = 0;
4132                         ns->ln_linkea_repaired = 0;
4133                         ns->ln_mul_linked_checked = 0;
4134                         ns->ln_mul_linked_repaired = 0;
4135                         ns->ln_unknown_inconsistency = 0;
4136                         ns->ln_unmatched_pairs_repaired = 0;
4137                         ns->ln_dangling_repaired = 0;
4138                         ns->ln_mul_ref_repaired = 0;
4139                         ns->ln_bad_type_repaired = 0;
4140                         ns->ln_lost_dirent_repaired = 0;
4141                         ns->ln_striped_dirs_scanned = 0;
4142                         ns->ln_striped_dirs_repaired = 0;
4143                         ns->ln_striped_dirs_failed = 0;
4144                         ns->ln_striped_dirs_disabled = 0;
4145                         ns->ln_striped_dirs_skipped = 0;
4146                         ns->ln_striped_shards_scanned = 0;
4147                         ns->ln_striped_shards_repaired = 0;
4148                         ns->ln_striped_shards_failed = 0;
4149                         ns->ln_striped_shards_skipped = 0;
4150                         ns->ln_name_hash_repaired = 0;
4151                         fid_zero(&ns->ln_fid_latest_scanned_phase2);
4152                         if (list_empty(&com->lc_link_dir))
4153                                 list_add_tail(&com->lc_link_dir,
4154                                               &lfsck->li_list_dir);
4155                         *pos = ns->ln_pos_first_inconsistent;
4156                 }
4157         } else {
4158                 ns->ln_status = LS_SCANNING_PHASE1;
4159                 if (list_empty(&com->lc_link_dir))
4160                         list_add_tail(&com->lc_link_dir,
4161                                       &lfsck->li_list_dir);
4162                 if (!lfsck->li_drop_dryrun ||
4163                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4164                         *pos = ns->ln_pos_last_checkpoint;
4165                         pos->lp_oit_cookie++;
4166                 } else {
4167                         *pos = ns->ln_pos_first_inconsistent;
4168                 }
4169         }
4170
4171         spin_unlock(&lfsck->li_lock);
4172         up_write(&com->lc_sem);
4173
4174         rc = lfsck_start_assistant(env, com, lsp);
4175
4176         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4177                DFID", %#llx]: rc = %d\n",
4178                lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4179                PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4180
4181         return rc;
4182 }
4183
4184 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4185                                     struct lfsck_component *com,
4186                                     struct dt_object *obj)
4187 {
4188         struct lfsck_thread_info *info  = lfsck_env_info(env);
4189         struct lfsck_namespace   *ns    = com->lc_file_ram;
4190         struct lfsck_instance    *lfsck = com->lc_lfsck;
4191         const struct lu_fid      *fid   = lfsck_dto2fid(obj);
4192         struct lu_fid            *pfid  = &info->lti_fid2;
4193         struct lu_name           *cname = &info->lti_name;
4194         struct lu_seq_range      *range = &info->lti_range;
4195         struct seq_server_site   *ss    = lfsck_dev_site(lfsck);
4196         struct linkea_data        ldata = { NULL };
4197         __u32                     idx   = lfsck_dev_idx(lfsck);
4198         int                       rc;
4199         ENTRY;
4200
4201         rc = lfsck_links_read(env, obj, &ldata);
4202         if (rc == -ENOENT)
4203                 GOTO(out, rc = 0);
4204
4205         /* -EINVAL means crashed linkEA, should be verified. */
4206         if (rc == -EINVAL) {
4207                 rc = lfsck_namespace_trace_update(env, com, fid,
4208                                                   LNTF_CHECK_LINKEA, true);
4209                 if (rc == 0) {
4210                         struct lustre_handle lh = { 0 };
4211
4212                         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4213                                               MDS_INODELOCK_UPDATE |
4214                                               MDS_INODELOCK_XATTR, LCK_EX);
4215                         if (rc == 0) {
4216                                 rc = lfsck_namespace_links_remove(env, com,
4217                                                                   obj);
4218                                 lfsck_ibits_unlock(&lh, LCK_EX);
4219                         }
4220                 }
4221
4222                 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4223         }
4224
4225         if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4226                 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4227
4228                 GOTO(out, rc);
4229         }
4230
4231         if (rc != 0)
4232                 GOTO(out, rc);
4233
4234         /* Record multiple-linked object. */
4235         if (ldata.ld_leh->leh_reccount > 1) {
4236                 rc = lfsck_namespace_trace_update(env, com, fid,
4237                                                   LNTF_CHECK_LINKEA, true);
4238
4239                 GOTO(out, rc);
4240         }
4241
4242         linkea_first_entry(&ldata);
4243         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4244         if (!fid_is_sane(pfid)) {
4245                 rc = lfsck_namespace_trace_update(env, com, fid,
4246                                                   LNTF_CHECK_PARENT, true);
4247         } else {
4248                 fld_range_set_mdt(range);
4249                 rc = fld_local_lookup(env, ss->ss_server_fld,
4250                                       fid_seq(pfid), range);
4251                 if ((rc == -ENOENT) ||
4252                     (rc == 0 && range->lsr_index != idx))
4253                         rc = lfsck_namespace_trace_update(env, com, fid,
4254                                                 LNTF_CHECK_LINKEA, true);
4255                 else
4256                         rc = lfsck_namespace_check_for_double_scan(env, com,
4257                                                                    obj);
4258         }
4259
4260         GOTO(out, rc);
4261
4262 out:
4263         down_write(&com->lc_sem);
4264         if (S_ISDIR(lfsck_object_type(obj)))
4265                 ns->ln_dirs_checked++;
4266         if (rc != 0)
4267                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4268         up_write(&com->lc_sem);
4269
4270         return rc;
4271 }
4272
4273 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4274                                     struct lfsck_component *com,
4275                                     struct lfsck_assistant_object *lso,
4276                                     struct lu_dirent *ent, __u16 type)
4277 {
4278         struct lfsck_assistant_data     *lad     = com->lc_data;
4279         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4280         struct lfsck_namespace_req      *lnr;
4281         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4282         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4283         struct ptlrpc_thread            *athread = &lad->lad_thread;
4284         struct l_wait_info               lwi     = { 0 };
4285         bool                             wakeup  = false;
4286
4287         l_wait_event(mthread->t_ctl_waitq,
4288                      lad->lad_prefetched < bk->lb_async_windows ||
4289                      !thread_is_running(mthread) ||
4290                      thread_is_stopped(athread),
4291                      &lwi);
4292
4293         if (unlikely(!thread_is_running(mthread)) ||
4294                      thread_is_stopped(athread))
4295                 return 0;
4296
4297         if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4298                 return 0;
4299
4300         lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4301         if (IS_ERR(lnr)) {
4302                 struct lfsck_namespace *ns = com->lc_file_ram;
4303
4304                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4305                 return PTR_ERR(lnr);
4306         }
4307
4308         spin_lock(&lad->lad_lock);
4309         if (lad->lad_assistant_status < 0) {
4310                 spin_unlock(&lad->lad_lock);
4311                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4312                 return lad->lad_assistant_status;
4313         }
4314
4315         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4316         if (lad->lad_prefetched == 0)
4317                 wakeup = true;
4318
4319         lad->lad_prefetched++;
4320         spin_unlock(&lad->lad_lock);
4321         if (wakeup)
4322                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4323
4324         down_write(&com->lc_sem);
4325         com->lc_new_checked++;
4326         up_write(&com->lc_sem);
4327
4328         return 0;
4329 }
4330
4331 static int lfsck_namespace_post(const struct lu_env *env,
4332                                 struct lfsck_component *com,
4333                                 int result, bool init)
4334 {
4335         struct lfsck_instance   *lfsck = com->lc_lfsck;
4336         struct lfsck_namespace  *ns    = com->lc_file_ram;
4337         int                      rc;
4338         ENTRY;
4339
4340         lfsck_post_generic(env, com, &result);
4341
4342         down_write(&com->lc_sem);
4343         lfsck_namespace_release_lmv(env, com);
4344
4345         spin_lock(&lfsck->li_lock);
4346         if (!init)
4347                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4348         if (result > 0) {
4349                 ns->ln_status = LS_SCANNING_PHASE2;
4350                 ns->ln_flags |= LF_SCANNED_ONCE;
4351                 ns->ln_flags &= ~LF_UPGRADE;
4352                 list_del_init(&com->lc_link_dir);
4353                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4354         } else if (result == 0) {
4355                 if (lfsck->li_status != 0)
4356                         ns->ln_status = lfsck->li_status;
4357                 else
4358                         ns->ln_status = LS_STOPPED;
4359                 if (ns->ln_status != LS_PAUSED) {
4360                         list_del_init(&com->lc_link_dir);
4361                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4362                 }
4363         } else {
4364                 ns->ln_status = LS_FAILED;
4365                 list_del_init(&com->lc_link_dir);
4366                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4367         }
4368         spin_unlock(&lfsck->li_lock);
4369
4370         if (!init) {
4371                 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4372                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4373                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4374                 ns->ln_items_checked += com->lc_new_checked;
4375                 com->lc_new_checked = 0;
4376         }
4377
4378         rc = lfsck_namespace_store(env, com);
4379         up_write(&com->lc_sem);
4380
4381         CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4382                lfsck_lfsck2name(lfsck), rc);
4383
4384         RETURN(rc);
4385 }
4386
4387 static void
4388 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4389                      struct seq_file *m)
4390 {
4391         struct lfsck_instance   *lfsck = com->lc_lfsck;
4392         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4393         struct lfsck_namespace  *ns    = com->lc_file_ram;
4394
4395         down_read(&com->lc_sem);
4396         seq_printf(m, "name: lfsck_namespace\n"
4397                    "magic: %#x\n"
4398                    "version: %d\n"
4399                    "status: %s\n",
4400                    ns->ln_magic,
4401                    bk->lb_version,
4402                    lfsck_status2name(ns->ln_status));
4403
4404         lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4405
4406         lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4407
4408         lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4409
4410         lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4411
4412         lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4413
4414         lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4415
4416         lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4417                        "last_checkpoint_position");
4418
4419         lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4420                        "first_failure_position");
4421
4422         if (ns->ln_status == LS_SCANNING_PHASE1) {
4423                 struct lfsck_position pos;
4424                 const struct dt_it_ops *iops;
4425                 cfs_duration_t duration = cfs_time_current() -
4426                                           lfsck->li_time_last_checkpoint;
4427                 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4428                 __u64 speed = checked;
4429                 __u64 new_checked = com->lc_new_checked *
4430                                     msecs_to_jiffies(MSEC_PER_SEC);
4431                 __u32 rtime = ns->ln_run_time_phase1 +
4432                               cfs_duration_sec(duration + HALF_SEC);
4433
4434                 if (duration != 0)
4435                         do_div(new_checked, duration);
4436
4437                 if (rtime != 0)
4438                         do_div(speed, rtime);
4439
4440                 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0);
4441                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4442                            "average_speed_phase2: N/A\n"
4443                            "average_speed_total: %llu items/sec\n"
4444                            "real_time_speed_phase1: %llu items/sec\n"
4445                            "real_time_speed_phase2: N/A\n",
4446                            speed,
4447                            speed,
4448                            new_checked);
4449
4450                 LASSERT(lfsck->li_di_oit != NULL);
4451
4452                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4453
4454                 /* The low layer otable-based iteration position may NOT
4455                  * exactly match the namespace-based directory traversal
4456                  * cookie. Generally, it is not a serious issue. But the
4457                  * caller should NOT make assumption on that. */
4458                 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4459                 if (!lfsck->li_current_oit_processed)
4460                         pos.lp_oit_cookie--;
4461
4462                 spin_lock(&lfsck->li_lock);
4463                 if (lfsck->li_di_dir != NULL) {
4464                         pos.lp_dir_cookie = lfsck->li_cookie_dir;
4465                         if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4466                                 fid_zero(&pos.lp_dir_parent);
4467                                 pos.lp_dir_cookie = 0;
4468                         } else {
4469                                 pos.lp_dir_parent =
4470                                         *lfsck_dto2fid(lfsck->li_obj_dir);
4471                         }
4472                 } else {
4473                         fid_zero(&pos.lp_dir_parent);
4474                         pos.lp_dir_cookie = 0;
4475                 }
4476                 spin_unlock(&lfsck->li_lock);
4477                 lfsck_pos_dump(m, &pos, "current_position");
4478         } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4479                 cfs_duration_t duration = cfs_time_current() -
4480                                           com->lc_time_last_checkpoint;
4481                 __u64 checked = ns->ln_objs_checked_phase2 +
4482                                 com->lc_new_checked;
4483                 __u64 speed1 = ns->ln_items_checked;
4484                 __u64 speed2 = checked;
4485                 __u64 speed0 = speed1 + speed2;
4486                 __u64 new_checked = com->lc_new_checked *
4487                                     msecs_to_jiffies(MSEC_PER_SEC);
4488                 __u32 rtime = ns->ln_run_time_phase2 +
4489                               cfs_duration_sec(duration + HALF_SEC);
4490                 __u32 time0 = ns->ln_run_time_phase1 + rtime;
4491
4492                 if (duration != 0)
4493                         do_div(new_checked, duration);
4494
4495                 if (ns->ln_run_time_phase1 != 0)
4496                         do_div(speed1, ns->ln_run_time_phase1);
4497                 else if (ns->ln_items_checked != 0)
4498                         time0++;
4499
4500                 if (rtime != 0)
4501                         do_div(speed2, rtime);
4502                 else if (checked != 0)
4503                         time0++;
4504
4505                 if (time0 != 0)
4506                         do_div(speed0, time0);
4507
4508                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4509                                                 checked,
4510                                                 ns->ln_run_time_phase1, rtime);
4511                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4512                            "average_speed_phase2: %llu objs/sec\n"
4513                            "average_speed_total: %llu items/sec\n"
4514                            "real_time_speed_phase1: N/A\n"
4515                            "real_time_speed_phase2: %llu objs/sec\n"
4516                            "current_position: "DFID"\n",
4517                            speed1,
4518                            speed2,
4519                            speed0,
4520                            new_checked,
4521                            PFID(&ns->ln_fid_latest_scanned_phase2));
4522         } else {
4523                 __u64 speed1 = ns->ln_items_checked;
4524                 __u64 speed2 = ns->ln_objs_checked_phase2;
4525                 __u64 speed0 = speed1 + speed2;
4526                 __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4527
4528                 if (ns->ln_run_time_phase1 != 0)
4529                         do_div(speed1, ns->ln_run_time_phase1);
4530                 else if (ns->ln_items_checked != 0)
4531                         time0++;
4532
4533                 if (ns->ln_run_time_phase2 != 0)
4534                         do_div(speed2, ns->ln_run_time_phase2);
4535                 else if (ns->ln_objs_checked_phase2 != 0)
4536                         time0++;
4537
4538                 if (time0 != 0)
4539                         do_div(speed0, time0);
4540
4541                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4542                                                 ns->ln_objs_checked_phase2,
4543                                                 ns->ln_run_time_phase1,
4544                                                 ns->ln_run_time_phase2);
4545                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4546                            "average_speed_phase2: %llu objs/sec\n"
4547                            "average_speed_total: %llu items/sec\n"
4548                            "real_time_speed_phase1: N/A\n"
4549                            "real_time_speed_phase2: N/A\n"
4550                            "current_position: N/A\n",
4551                            speed1,
4552                            speed2,
4553                            speed0);
4554         }
4555
4556         up_read(&com->lc_sem);
4557 }
4558
4559 static int lfsck_namespace_double_scan(const struct lu_env *env,
4560                                        struct lfsck_component *com)
4561 {
4562         struct lfsck_namespace          *ns     = com->lc_file_ram;
4563         struct lfsck_assistant_data     *lad    = com->lc_data;
4564         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4565         struct lfsck_tgt_desc           *ltd;
4566         struct lfsck_tgt_desc           *next;
4567         int                              rc;
4568
4569         rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4570         if (thread_is_stopped(&lad->lad_thread)) {
4571                 LASSERT(list_empty(&lad->lad_req_list));
4572                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4573
4574                 spin_lock(&ltds->ltd_lock);
4575                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4576                                          ltd_namespace_phase_list) {
4577                         list_del_init(&ltd->ltd_namespace_phase_list);
4578                 }
4579                 spin_unlock(&ltds->ltd_lock);
4580         }
4581
4582         return rc;
4583 }
4584
4585 static void lfsck_namespace_data_release(const struct lu_env *env,
4586                                          struct lfsck_component *com)
4587 {
4588         struct lfsck_assistant_data     *lad    = com->lc_data;
4589         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4590         struct lfsck_tgt_desc           *ltd;
4591         struct lfsck_tgt_desc           *next;
4592
4593         LASSERT(lad != NULL);
4594         LASSERT(thread_is_init(&lad->lad_thread) ||
4595                 thread_is_stopped(&lad->lad_thread));
4596         LASSERT(list_empty(&lad->lad_req_list));
4597
4598         com->lc_data = NULL;
4599         lfsck_namespace_release_lmv(env, com);
4600
4601         spin_lock(&ltds->ltd_lock);
4602         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4603                                  ltd_namespace_phase_list) {
4604                 list_del_init(&ltd->ltd_namespace_phase_list);
4605         }
4606         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4607                                  ltd_namespace_phase_list) {
4608                 list_del_init(&ltd->ltd_namespace_phase_list);
4609         }
4610         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4611                                  ltd_namespace_list) {
4612                 list_del_init(&ltd->ltd_namespace_list);
4613         }
4614         spin_unlock(&ltds->ltd_lock);
4615
4616         if (likely(lad->lad_bitmap != NULL))
4617                 CFS_FREE_BITMAP(lad->lad_bitmap);
4618
4619         OBD_FREE_PTR(lad);
4620 }
4621
4622 static void lfsck_namespace_quit(const struct lu_env *env,
4623                                  struct lfsck_component *com)
4624 {
4625         struct lfsck_assistant_data     *lad    = com->lc_data;
4626         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4627         struct lfsck_tgt_desc           *ltd;
4628         struct lfsck_tgt_desc           *next;
4629
4630         LASSERT(lad != NULL);
4631
4632         lfsck_quit_generic(env, com);
4633
4634         LASSERT(thread_is_init(&lad->lad_thread) ||
4635                 thread_is_stopped(&lad->lad_thread));
4636         LASSERT(list_empty(&lad->lad_req_list));
4637
4638         lfsck_namespace_release_lmv(env, com);
4639
4640         spin_lock(&ltds->ltd_lock);
4641         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4642                                  ltd_namespace_phase_list) {
4643                 list_del_init(&ltd->ltd_namespace_phase_list);
4644         }
4645         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4646                                  ltd_namespace_phase_list) {
4647                 list_del_init(&ltd->ltd_namespace_phase_list);
4648         }
4649         spin_unlock(&ltds->ltd_lock);
4650 }
4651
4652 static int lfsck_namespace_in_notify(const struct lu_env *env,
4653                                      struct lfsck_component *com,
4654                                      struct lfsck_request *lr,
4655                                      struct thandle *th)
4656 {
4657         struct lfsck_instance           *lfsck = com->lc_lfsck;
4658         struct lfsck_namespace          *ns    = com->lc_file_ram;
4659         struct lfsck_assistant_data     *lad   = com->lc_data;
4660         struct lfsck_tgt_descs          *ltds  = &lfsck->li_mdt_descs;
4661         struct lfsck_tgt_desc           *ltd;
4662         int                              rc    = 0;
4663         bool                             fail  = false;
4664         ENTRY;
4665
4666         switch (lr->lr_event) {
4667         case LE_SET_LMV_MASTER: {
4668                 struct dt_object        *obj;
4669
4670                 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
4671                 if (IS_ERR(obj))
4672                         RETURN(PTR_ERR(obj));
4673
4674                 if (likely(dt_object_exists(obj)))
4675                         rc = lfsck_namespace_notify_lmv_master_local(env, com,
4676                                                                      obj);
4677
4678                 lfsck_object_put(env, obj);
4679
4680                 RETURN(rc > 0 ? 0 : rc);
4681         }
4682         case LE_SET_LMV_SLAVE: {
4683                 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4684                         ns->ln_striped_shards_repaired++;
4685
4686                 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4687                                                   LNTF_RECHECK_NAME_HASH, true);
4688
4689                 RETURN(rc > 0 ? 0 : rc);
4690         }
4691         case LE_PHASE1_DONE:
4692         case LE_PHASE2_DONE:
4693         case LE_PEER_EXIT:
4694                 break;
4695         default:
4696                 RETURN(-EINVAL);
4697         }
4698
4699         CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4700                "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4701                lr->lr_index, lr->lr_status, lr->lr_flags2);
4702
4703         spin_lock(&ltds->ltd_lock);
4704         ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
4705         if (ltd == NULL) {
4706                 spin_unlock(&ltds->ltd_lock);
4707
4708                 RETURN(-ENXIO);
4709         }
4710
4711         list_del_init(&ltd->ltd_namespace_phase_list);
4712         switch (lr->lr_event) {
4713         case LE_PHASE1_DONE:
4714                 if (lr->lr_status <= 0) {
4715                         ltd->ltd_namespace_done = 1;
4716                         list_del_init(&ltd->ltd_namespace_list);
4717                         CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4718                                "phase1 for namespace LFSCK: rc = %d.\n",
4719                                lfsck_lfsck2name(lfsck),
4720                                ltd->ltd_index, lr->lr_status);
4721                         ns->ln_flags |= LF_INCOMPLETE;
4722                         fail = true;
4723                         break;
4724                 }
4725
4726                 if (lr->lr_flags2 & LF_INCOMPLETE)
4727                         ns->ln_flags |= LF_INCOMPLETE;
4728
4729                 if (list_empty(&ltd->ltd_namespace_list))
4730                         list_add_tail(&ltd->ltd_namespace_list,
4731                                       &lad->lad_mdt_list);
4732                 list_add_tail(&ltd->ltd_namespace_phase_list,
4733                               &lad->lad_mdt_phase2_list);
4734                 break;
4735         case LE_PHASE2_DONE:
4736                 ltd->ltd_namespace_done = 1;
4737                 list_del_init(&ltd->ltd_namespace_list);
4738                 break;
4739         case LE_PEER_EXIT:
4740                 fail = true;
4741                 ltd->ltd_namespace_done = 1;
4742                 list_del_init(&ltd->ltd_namespace_list);
4743                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4744                         CDEBUG(D_LFSCK,
4745                                "%s: the peer MDT %x exit namespace LFSCK\n",
4746                                lfsck_lfsck2name(lfsck), ltd->ltd_index);
4747                         ns->ln_flags |= LF_INCOMPLETE;
4748                 }
4749                 break;
4750         default:
4751                 break;
4752         }
4753         spin_unlock(&ltds->ltd_lock);
4754
4755         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4756                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4757
4758                 memset(stop, 0, sizeof(*stop));
4759                 stop->ls_status = lr->lr_status;
4760                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4761                 lfsck_stop(env, lfsck->li_bottom, stop);
4762         } else if (lfsck_phase2_next_ready(lad)) {
4763                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4764         }
4765
4766         RETURN(0);
4767 }
4768
4769 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
4770 {
4771         *count += ns->ln_objs_nlink_repaired;
4772         *count += ns->ln_dirent_repaired;
4773         *count += ns->ln_linkea_repaired;
4774         *count += ns->ln_mul_linked_repaired;
4775         *count += ns->ln_unmatched_pairs_repaired;
4776         *count += ns->ln_dangling_repaired;
4777         *count += ns->ln_mul_ref_repaired;
4778         *count += ns->ln_bad_type_repaired;
4779         *count += ns->ln_lost_dirent_repaired;
4780         *count += ns->ln_striped_dirs_disabled;
4781         *count += ns->ln_striped_dirs_repaired;
4782         *count += ns->ln_striped_shards_repaired;
4783         *count += ns->ln_name_hash_repaired;
4784         *count += ns->ln_local_lpf_moved;
4785 }
4786
4787 static int lfsck_namespace_query_all(const struct lu_env *env,
4788                                      struct lfsck_component *com,
4789                                      __u32 *mdts_count, __u64 *repaired)
4790 {
4791         struct lfsck_namespace *ns = com->lc_file_ram;
4792         struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4793         struct lfsck_tgt_desc *ltd;
4794         int idx;
4795         int rc;
4796         ENTRY;
4797
4798         rc = lfsck_query_all(env, com);
4799         if (rc != 0)
4800                 RETURN(rc);
4801
4802         down_read(&ltds->ltd_rw_sem);
4803         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
4804                 ltd = lfsck_ltd2tgt(ltds, idx);
4805                 LASSERT(ltd != NULL);
4806
4807                 mdts_count[ltd->ltd_namespace_status]++;
4808                 *repaired += ltd->ltd_namespace_repaired;
4809         }
4810         up_read(&ltds->ltd_rw_sem);
4811
4812         down_read(&com->lc_sem);
4813         mdts_count[ns->ln_status]++;
4814         lfsck_namespace_repaired(ns, repaired);
4815         up_read(&com->lc_sem);
4816
4817         RETURN(0);
4818 }
4819
4820 static int lfsck_namespace_query(const struct lu_env *env,
4821                                  struct lfsck_component *com,
4822                                  struct lfsck_request *req,
4823                                  struct lfsck_reply *rep,
4824                                  struct lfsck_query *que, int idx)
4825 {
4826         struct lfsck_namespace *ns = com->lc_file_ram;
4827         int rc = 0;
4828
4829         if (que != NULL) {
4830                 LASSERT(com->lc_lfsck->li_master);
4831
4832                 rc = lfsck_namespace_query_all(env, com,
4833                                                que->lu_mdts_count[idx],
4834                                                &que->lu_repaired[idx]);
4835         } else {
4836                 down_read(&com->lc_sem);
4837                 rep->lr_status = ns->ln_status;
4838                 if (req->lr_flags & LEF_QUERY_ALL)
4839                         lfsck_namespace_repaired(ns, &rep->lr_repaired);
4840                 up_read(&com->lc_sem);
4841         }
4842
4843         return rc;
4844 }
4845
4846 static struct lfsck_operations lfsck_namespace_ops = {
4847         .lfsck_reset            = lfsck_namespace_reset,
4848         .lfsck_fail             = lfsck_namespace_fail,
4849         .lfsck_close_dir        = lfsck_namespace_close_dir,
4850         .lfsck_open_dir         = lfsck_namespace_open_dir,
4851         .lfsck_checkpoint       = lfsck_namespace_checkpoint,
4852         .lfsck_prep             = lfsck_namespace_prep,
4853         .lfsck_exec_oit         = lfsck_namespace_exec_oit,
4854         .lfsck_exec_dir         = lfsck_namespace_exec_dir,
4855         .lfsck_post             = lfsck_namespace_post,
4856         .lfsck_dump             = lfsck_namespace_dump,
4857         .lfsck_double_scan      = lfsck_namespace_double_scan,
4858         .lfsck_data_release     = lfsck_namespace_data_release,
4859         .lfsck_quit             = lfsck_namespace_quit,
4860         .lfsck_in_notify        = lfsck_namespace_in_notify,
4861         .lfsck_query            = lfsck_namespace_query,
4862 };
4863
4864 /**
4865  * Repair dangling name entry.
4866  *
4867  * For the name entry with dangling reference, we need to repare the
4868  * inconsistency according to the LFSCK sponsor's requirement:
4869  *
4870  * 1) Keep the inconsistency there and report the inconsistency case,
4871  *    then give the chance to the application to find related issues,
4872  *    and the users can make the decision about how to handle it with
4873  *    more human knownledge. (by default)
4874  *
4875  * 2) Re-create the missing MDT-object with the FID information.
4876  *
4877  * \param[in] env       pointer to the thread context
4878  * \param[in] com       pointer to the lfsck component
4879  * \param[in] parent    pointer to the dir object that contains the dangling
4880  *                      name entry
4881  * \param[in] child     pointer to the object corresponding to the dangling
4882  *                      name entry
4883  * \param[in] lnr       pointer to the namespace request that contains the
4884  *                      name's name, parent object, parent's LMV, and ect.
4885  *
4886  * \retval              positive number if no need to repair
4887  * \retval              zero for repaired successfully
4888  * \retval              negative error number on failure
4889  */
4890 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4891                                     struct lfsck_component *com,
4892                                     struct dt_object *parent,
4893                                     struct dt_object *child,
4894                                     struct lfsck_namespace_req *lnr)
4895 {
4896         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4897         struct lu_attr                  *la     = &info->lti_la;
4898         struct dt_allocation_hint       *hint   = &info->lti_hint;
4899         struct dt_object_format         *dof    = &info->lti_dof;
4900         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
4901         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
4902         const struct lu_name            *cname;
4903         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
4904         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
4905         struct linkea_data               ldata  = { NULL };
4906         struct lfsck_lock_handle        *llh    = &info->lti_llh;
4907         struct lu_buf                    linkea_buf;
4908         struct lu_buf                    lmv_buf;
4909         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4910         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4911         struct dt_device                *dev    = lfsck->li_next;
4912         struct thandle                  *th     = NULL;
4913         int                              rc     = 0;
4914         __u16                            type   = lnr->lnr_type;
4915         bool                             create;
4916         ENTRY;
4917
4918         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4919         if (bk->lb_param & LPF_CREATE_MDTOBJ)
4920                 create = true;
4921         else
4922                 create = false;
4923
4924         if (!create || bk->lb_param & LPF_DRYRUN)
4925                 GOTO(log, rc = 0);
4926
4927         /* We may need to create the sub-objects of the @child via LOD,
4928          * so make the modification based on lfsck->li_next. */
4929
4930         parent = lfsck_object_locate(dev, parent);
4931         if (IS_ERR(parent))
4932                 GOTO(log, rc = PTR_ERR(parent));
4933
4934         if (unlikely(!dt_try_as_dir(env, parent)))
4935                 GOTO(log, rc = -ENOTDIR);
4936
4937         child = lfsck_object_locate(dev, child);
4938         if (IS_ERR(child))
4939                 GOTO(log, rc = PTR_ERR(child));
4940
4941         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
4942                               cname, pfid);
4943         if (rc != 0)
4944                 GOTO(log, rc);
4945
4946         rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
4947                         MDS_INODELOCK_UPDATE, LCK_PR);
4948         if (rc != 0)
4949                 GOTO(log, rc);
4950
4951         rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
4952         if (rc != 0)
4953                 GOTO(log, rc);
4954
4955         /* Set the ctime as zero, then others can know it is created for
4956          * repairing dangling name entry by LFSCK. And if the LFSCK made
4957          * wrong decision and the real MDT-object has been found later,
4958          * then the LFSCK has chance to fix the incosistency properly. */
4959         memset(la, 0, sizeof(*la));
4960         la->la_mode = (type & S_IFMT) | 0600;
4961         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4962                         LA_ATIME | LA_MTIME | LA_CTIME;
4963
4964         child->do_ops->do_ah_init(env, hint, parent, child,
4965                                  la->la_mode & S_IFMT);
4966
4967         memset(dof, 0, sizeof(*dof));
4968         dof->dof_type = dt_mode_to_dft(type);
4969         /* If the target is a regular file, then the LFSCK will only create
4970          * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4971          * OST-objects will be created when write open. */
4972
4973         th = dt_trans_create(env, dev);
4974         if (IS_ERR(th))
4975                 GOTO(log, rc = PTR_ERR(th));
4976
4977         /* 1a. create child. */
4978         rc = dt_declare_create(env, child, la, hint, dof, th);
4979         if (rc != 0)
4980                 GOTO(stop, rc);
4981
4982         if (S_ISDIR(type)) {
4983                 if (unlikely(!dt_try_as_dir(env, child)))
4984                         GOTO(stop, rc = -ENOTDIR);
4985
4986                 /* 2a. increase child nlink */
4987                 rc = dt_declare_ref_add(env, child, th);
4988                 if (rc != 0)
4989                         GOTO(stop, rc);
4990
4991                 /* 3a. insert dot into child dir */
4992                 rec->rec_type = S_IFDIR;
4993                 rec->rec_fid = cfid;
4994                 rc = dt_declare_insert(env, child,
4995                                        (const struct dt_rec *)rec,
4996                                        (const struct dt_key *)dot, th);
4997                 if (rc != 0)
4998                         GOTO(stop, rc);
4999
5000                 /* 4a. insert dotdot into child dir */
5001                 rec->rec_fid = pfid;
5002                 rc = dt_declare_insert(env, child,
5003                                        (const struct dt_rec *)rec,
5004                                        (const struct dt_key *)dotdot, th);
5005                 if (rc != 0)
5006                         GOTO(stop, rc);
5007
5008                 /* 5a. generate slave LMV EA. */
5009                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5010                         int idx;
5011
5012                         idx = lfsck_shard_name_to_index(env,
5013                                         lnr->lnr_name, lnr->lnr_namelen,
5014                                         type, cfid);
5015                         if (unlikely(idx < 0))
5016                                 GOTO(stop, rc = idx);
5017
5018                         *lmv2 = lnr->lnr_lmv->ll_lmv;
5019                         lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5020                         lmv2->lmv_master_mdt_index = idx;
5021
5022                         lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5023                         lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5024                         rc = dt_declare_xattr_set(env, child, &lmv_buf,
5025                                                   XATTR_NAME_LMV, 0, th);
5026                         if (rc != 0)
5027                                 GOTO(stop, rc);
5028                 }
5029         }
5030
5031         /* 6a. insert linkEA for child */
5032         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5033                        ldata.ld_leh->leh_len);
5034         rc = dt_declare_xattr_set(env, child, &linkea_buf,
5035                                   XATTR_NAME_LINK, 0, th);
5036         if (rc != 0)
5037                 GOTO(stop, rc);
5038
5039         rc = dt_trans_start_local(env, dev, th);
5040         if (rc != 0)
5041                 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5042
5043         dt_write_lock(env, child, 0);
5044         /* 1b. create child */
5045         rc = dt_create(env, child, la, hint, dof, th);
5046         if (rc != 0)
5047                 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5048
5049         if (S_ISDIR(type)) {
5050                 /* 2b. increase child nlink */
5051                 rc = dt_ref_add(env, child, th);
5052                 if (rc != 0)
5053                         GOTO(unlock, rc);
5054
5055                 /* 3b. insert dot into child dir */
5056                 rec->rec_type = S_IFDIR;
5057                 rec->rec_fid = cfid;
5058                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5059                                (const struct dt_key *)dot, th, 1);
5060                 if (rc != 0)
5061                         GOTO(unlock, rc);
5062
5063                 /* 4b. insert dotdot into child dir */
5064                 rec->rec_fid = pfid;
5065                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5066                                (const struct dt_key *)dotdot, th, 1);
5067                 if (rc != 0)
5068                         GOTO(unlock, rc);
5069
5070                 /* 5b. generate slave LMV EA. */
5071                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5072                         rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5073                                           0, th);
5074                         if (rc != 0)
5075                                 GOTO(unlock, rc);
5076                 }
5077         }
5078
5079         /* 6b. insert linkEA for child. */
5080         rc = dt_xattr_set(env, child, &linkea_buf,
5081                           XATTR_NAME_LINK, 0, th);
5082
5083         GOTO(unlock, rc);
5084
5085 unlock:
5086         dt_write_unlock(env, child);
5087
5088 stop:
5089         dt_trans_stop(env, dev, th);
5090
5091 log:
5092         lfsck_unlock(llh);
5093         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5094                "reference for: parent "DFID", child "DFID", type %u, "
5095                "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5096                PFID(pfid), PFID(cfid), type, cname->ln_name,
5097                create ? "Create the lost MDT-object as required" :
5098                         "Keep the MDT-object there by default", rc);
5099
5100         if (rc <= 0) {
5101                 struct lfsck_namespace *ns = com->lc_file_ram;
5102
5103                 ns->ln_flags |= LF_INCONSISTENT;
5104         }
5105
5106         return rc;
5107 }
5108
5109 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5110                                                 struct lfsck_component *com,
5111                                                 struct lfsck_assistant_req *lar)
5112 {
5113         struct lfsck_thread_info   *info     = lfsck_env_info(env);
5114         struct lu_attr             *la       = &info->lti_la;
5115         struct lfsck_instance      *lfsck    = com->lc_lfsck;
5116         struct lfsck_bookmark      *bk       = &lfsck->li_bookmark_ram;
5117         struct lfsck_namespace     *ns       = com->lc_file_ram;
5118         struct lfsck_assistant_data *lad     = com->lc_data;
5119         struct linkea_data          ldata    = { NULL };
5120         const struct lu_name       *cname;
5121         struct thandle             *handle   = NULL;
5122         struct lfsck_namespace_req *lnr      =
5123                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5124         struct dt_object           *dir      = NULL;
5125         struct dt_object           *obj      = NULL;
5126         struct lfsck_assistant_object *lso   = lar->lar_parent;
5127         const struct lu_fid        *pfid     = &lso->lso_fid;
5128         struct dt_device           *dev      = NULL;
5129         struct lustre_handle        lh       = { 0 };
5130         bool                        repaired = false;
5131         bool                        dtlocked = false;
5132         bool                        remove;
5133         bool                        newdata;
5134         bool                        log      = false;
5135         bool                        bad_hash = false;
5136         bool                        bad_linkea = false;
5137         int                         idx      = 0;
5138         int                         count    = 0;
5139         int                         rc       = 0;
5140         enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5141         ENTRY;
5142
5143         if (lso->lso_dead)
5144                 RETURN(0);
5145
5146         la->la_nlink = 0;
5147         if (lnr->lnr_attr & LUDA_UPGRADE) {
5148                 ns->ln_flags |= LF_UPGRADE;
5149                 ns->ln_dirent_repaired++;
5150                 repaired = true;
5151         } else if (lnr->lnr_attr & LUDA_REPAIR) {
5152                 ns->ln_flags |= LF_INCONSISTENT;
5153                 ns->ln_dirent_repaired++;
5154                 repaired = true;
5155         }
5156
5157         if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5158                 if (strcmp(lnr->lnr_name, dotdot) != 0)
5159                         LBUG();
5160                 else
5161                         rc = lfsck_namespace_trace_update(env, com, pfid,
5162                                                 LNTF_CHECK_PARENT, true);
5163
5164                 GOTO(out, rc);
5165         }
5166
5167         if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5168                 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5169                        " for the name entry %.*s under "DFID"\n",
5170                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5171                        lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5172
5173                 if (strcmp(lnr->lnr_name, dotdot) != 0)
5174                         /* invalid FID means bad name entry, remove it. */
5175                         type = LNIT_BAD_DIRENT;
5176                 else
5177                         /* If the parent FID is invalid, we cannot remove
5178                          * the ".." entry directly. */
5179                         rc = lfsck_namespace_trace_update(env, com, pfid,
5180                                                 LNTF_CHECK_PARENT, true);
5181
5182                 GOTO(out, rc);
5183         }
5184
5185         if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5186                 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5187
5188                 RETURN(rc);
5189         }
5190
5191         if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5192                 GOTO(out, rc = 0);
5193
5194         if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5195                 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5196
5197                 RETURN(rc);
5198         }
5199
5200         idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5201         if (idx < 0)
5202                 GOTO(out, rc = idx);
5203
5204         if (idx == lfsck_dev_idx(lfsck)) {
5205                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5206                         GOTO(out, rc = 0);
5207
5208                 dev = lfsck->li_bottom;
5209         } else {
5210                 struct lfsck_tgt_desc *ltd;
5211
5212                 /* Usually, some local filesystem consistency verification
5213                  * tools can guarantee the local namespace tree consistenct.
5214                  * So the LFSCK will only verify the remote directory. */
5215                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5216                         rc = lfsck_namespace_trace_update(env, com, pfid,
5217                                                 LNTF_CHECK_PARENT, true);
5218
5219                         GOTO(out, rc);
5220                 }
5221
5222                 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5223                 if (unlikely(ltd == NULL)) {
5224                         CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5225                                "did not join the namespace LFSCK\n",
5226                                lfsck_lfsck2name(lfsck), idx);
5227                         lfsck_lad_set_bitmap(env, com, idx);
5228
5229                         GOTO(out, rc = -ENODEV);
5230                 }
5231
5232                 dev = ltd->ltd_tgt;
5233         }
5234
5235         obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5236         if (IS_ERR(obj))
5237                 GOTO(out, rc = PTR_ERR(obj));
5238
5239         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5240         if (dt_object_exists(obj) == 0) {
5241
5242 dangling:
5243                 if (dir == NULL) {
5244                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5245                         if (IS_ERR(dir)) {
5246                                 rc = PTR_ERR(dir);
5247
5248                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5249                         }
5250                 }
5251
5252                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5253                 if (rc == 0) {
5254                         if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5255                                         lnr->lnr_name, lnr->lnr_namelen)) {
5256                                 type = LNIT_BAD_DIRENT;
5257
5258                                 GOTO(out, rc);
5259                         }
5260
5261                         type = LNIT_DANGLING;
5262                         rc = lfsck_namespace_repair_dangling(env, com, dir,
5263                                                              obj, lnr);
5264                         if (rc == 0)
5265                                 repaired = true;
5266                 }
5267
5268                 GOTO(out, rc);
5269         }
5270
5271         if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5272
5273 again:
5274                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5275                                       MDS_INODELOCK_UPDATE |
5276                                       MDS_INODELOCK_XATTR, LCK_EX);
5277                 if (rc != 0)
5278                         GOTO(out, rc);
5279
5280                 handle = dt_trans_create(env, dev);
5281                 if (IS_ERR(handle))
5282                         GOTO(out, rc = PTR_ERR(handle));
5283
5284                 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5285                 if (rc != 0)
5286                         GOTO(stop, rc);
5287
5288                 rc = dt_trans_start_local(env, dev, handle);
5289                 if (rc != 0)
5290                         GOTO(stop, rc);
5291
5292                 dt_write_lock(env, obj, 0);
5293                 dtlocked = true;
5294         }
5295
5296         rc = lfsck_links_read(env, obj, &ldata);
5297         if (unlikely(rc == -ENOENT)) {
5298                 if (handle != NULL) {
5299                         dt_write_unlock(env, obj);
5300                         dtlocked = false;
5301
5302                         dt_trans_stop(env, dev, handle);
5303                         handle = NULL;
5304
5305                         lfsck_ibits_unlock(&lh, LCK_EX);
5306                 }
5307
5308                 /* It may happen when the remote object has been removed,
5309                  * but the local MDT is not aware of that. */
5310                 goto dangling;
5311         } else if (rc == 0) {
5312                 count = ldata.ld_leh->leh_reccount;
5313                 rc = linkea_links_find(&ldata, cname, pfid);
5314                 if ((rc == 0) &&
5315                     (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5316                         if ((lfsck_object_type(obj) & S_IFMT) !=
5317                             lnr->lnr_type) {
5318                                 ns->ln_flags |= LF_INCONSISTENT;
5319                                 type = LNIT_BAD_TYPE;
5320                         }
5321
5322                         goto stop;
5323                 }
5324
5325                 /* If the name entry hash does not match the slave striped
5326                  * directory, and the name entry does not match also, then
5327                  * it is quite possible that name entry is corrupted. */
5328                 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5329                                         lnr->lnr_name, lnr->lnr_namelen)) {
5330                         ns->ln_flags |= LF_INCONSISTENT;
5331                         type = LNIT_BAD_DIRENT;
5332
5333                         GOTO(stop, rc = 0);
5334                 }
5335
5336                 /* If the file type stored in the name entry does not match
5337                  * the file type claimed by the object, and the object does
5338                  * not recognize the name entry, then it is quite possible
5339                  * that the name entry is corrupted. */
5340                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5341                         ns->ln_flags |= LF_INCONSISTENT;
5342                         type = LNIT_BAD_DIRENT;
5343
5344                         GOTO(stop, rc = 0);
5345                 }
5346
5347                 /* For sub-dir object, we cannot make sure whether the sub-dir
5348                  * back references the parent via ".." name entry correctly or
5349                  * not in the LFSCK first-stage scanning. It may be that the
5350                  * (remote) sub-dir ".." name entry has no parent FID after
5351                  * file-level backup/restore and its linkEA may be wrong.
5352                  * So under such case, we should replace the linkEA according
5353                  * to current name entry. But this needs to be done during the
5354                  * LFSCK second-stage scanning. The LFSCK will record the name
5355                  * entry for further possible using. */
5356                 remove = false;
5357                 newdata = false;
5358                 goto nodata;
5359         } else if (unlikely(rc == -EINVAL)) {
5360                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5361                         type = LNIT_BAD_TYPE;
5362
5363                 count = 1;
5364                 /* The magic crashed, we are not sure whether there are more
5365                  * corrupt data in the linkea, so remove all linkea entries. */
5366                 remove = true;
5367                 newdata = true;
5368                 goto nodata;
5369         } else if (rc == -ENODATA) {
5370                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5371                         type = LNIT_BAD_TYPE;
5372
5373                 count = 1;
5374                 remove = false;
5375                 newdata = true;
5376
5377 nodata:
5378                 if (bk->lb_param & LPF_DRYRUN) {
5379                         if (rc == -ENODATA)
5380                                 ns->ln_flags |= LF_UPGRADE;
5381                         else
5382                                 ns->ln_flags |= LF_INCONSISTENT;
5383                         ns->ln_linkea_repaired++;
5384                         repaired = true;
5385                         log = true;
5386                         goto stop;
5387                 }
5388
5389                 if (!lustre_handle_is_used(&lh)) {
5390                         remove = false;
5391                         newdata = false;
5392                         type = LNIT_NONE;
5393
5394                         goto again;
5395                 }
5396
5397                 LASSERT(handle != NULL);
5398
5399                 if (dir == NULL) {
5400                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5401                         if (IS_ERR(dir)) {
5402                                 rc = PTR_ERR(dir);
5403
5404                                 GOTO(stop, rc == -ENOENT ? 0 : rc);
5405                         }
5406                 }
5407
5408                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5409                 if (rc != 0)
5410                         GOTO(stop, rc);
5411
5412                 bad_linkea = true;
5413                 if (!remove && newdata)
5414                         ns->ln_flags |= LF_UPGRADE;
5415                 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5416                         ns->ln_flags |= LF_INCONSISTENT;
5417
5418                 if (remove) {
5419                         LASSERT(newdata);
5420
5421                         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5422                         if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5423                                 GOTO(stop, rc);
5424                 }
5425
5426                 if (newdata) {
5427                         rc = linkea_data_new(&ldata,
5428                                         &lfsck_env_info(env)->lti_linkea_buf);
5429                         if (rc != 0)
5430                                 GOTO(stop, rc);
5431                 }
5432
5433                 rc = linkea_add_buf(&ldata, cname, pfid);
5434                 if (rc == 0)
5435                         rc = lfsck_links_write(env, obj, &ldata, handle);
5436                 if (rc != 0)
5437                         GOTO(stop, rc);
5438
5439                 count = ldata.ld_leh->leh_reccount;
5440                 if (!S_ISDIR(lfsck_object_type(obj)) ||
5441                     !dt_object_remote(obj)) {
5442                         ns->ln_linkea_repaired++;
5443                         repaired = true;
5444                         log = true;
5445                 }
5446         } else {
5447                 GOTO(stop, rc);
5448         }
5449
5450 stop:
5451         if (dtlocked)
5452                 dt_write_unlock(env, obj);
5453
5454         if (handle != NULL && !IS_ERR(handle))
5455                 dt_trans_stop(env, dev, handle);
5456
5457 out:
5458         lfsck_ibits_unlock(&lh, LCK_EX);
5459
5460         if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5461             !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5462                                              lnr->lnr_name, lnr->lnr_namelen) &&
5463             type != LNIT_BAD_DIRENT) {
5464                 ns->ln_flags |= LF_INCONSISTENT;
5465
5466                 log = false;
5467                 if (dir == NULL) {
5468                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5469                         if (IS_ERR(dir)) {
5470                                 rc = PTR_ERR(dir);
5471
5472                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5473                         }
5474                 }
5475
5476                 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5477                                                 lnr->lnr_lmv, lnr->lnr_name);
5478                 if (rc == 0)
5479                         bad_hash = true;
5480         }
5481
5482         if (rc >= 0) {
5483                 if (type != LNIT_NONE && dir == NULL) {
5484                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5485                         if (IS_ERR(dir)) {
5486                                 rc = PTR_ERR(dir);
5487
5488                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5489                         }
5490                 }
5491
5492                 switch (type) {
5493                 case LNIT_BAD_TYPE:
5494                         log = false;
5495                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5496                                         obj, lnr->lnr_name, lnr->lnr_name,
5497                                         lnr->lnr_type, true, false);
5498                         if (rc > 0)
5499                                 repaired = true;
5500                         break;
5501                 case LNIT_BAD_DIRENT:
5502                         log = false;
5503                         /* XXX: This is a bad dirent, we do not know whether
5504                          *      the original name entry reference a regular
5505                          *      file or a directory, then keep the parent's
5506                          *      nlink count unchanged here. */
5507                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5508                                         obj, lnr->lnr_name, lnr->lnr_name,
5509                                         lnr->lnr_type, false, false);
5510                         if (rc > 0)
5511                                 repaired = true;
5512                         break;
5513                 default:
5514                         break;
5515                 }
5516
5517                 if (obj != NULL && count == 1 &&
5518                     S_ISREG(lfsck_object_type(obj)))
5519                         dt_attr_get(env, obj, la);
5520         }
5521
5522 trace:
5523         down_write(&com->lc_sem);
5524         if (rc < 0) {
5525                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5526                        "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5527                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5528                        lnr->lnr_namelen, lnr->lnr_name, rc);
5529
5530                 lfsck_namespace_record_failure(env, lfsck, ns);
5531                 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5532                      rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5533                      rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5534                     dev != NULL && dev != lfsck->li_bottom)
5535                         lfsck_lad_set_bitmap(env, com, idx);
5536
5537                 if (!(bk->lb_param & LPF_FAILOUT))
5538                         rc = 0;
5539         } else {
5540                 if (repaired) {
5541                         ns->ln_items_repaired++;
5542                         if (log)
5543                                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5544                                        "repaired the entry: "DFID", parent "DFID
5545                                        ", name %.*s, type %d\n",
5546                                        lfsck_lfsck2name(lfsck),
5547                                        PFID(&lnr->lnr_fid), PFID(pfid),
5548                                        lnr->lnr_namelen, lnr->lnr_name, type);
5549
5550                         switch (type) {
5551                         case LNIT_DANGLING:
5552                                 ns->ln_dangling_repaired++;
5553                                 break;
5554                         case LNIT_BAD_TYPE:
5555                                 ns->ln_bad_type_repaired++;
5556                                 break;
5557                         case LNIT_BAD_DIRENT:
5558                                 ns->ln_dirent_repaired++;
5559                                 break;
5560                         default:
5561                                 break;
5562                         }
5563
5564                         if (bk->lb_param & LPF_DRYRUN &&
5565                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5566                                 lfsck_pos_fill(env, lfsck,
5567                                                &ns->ln_pos_first_inconsistent,
5568                                                false);
5569                 }
5570
5571                 if (bad_hash) {
5572                         ns->ln_name_hash_repaired++;
5573
5574                         /* Not count repeatedly. */
5575                         if (!repaired) {
5576                                 ns->ln_items_repaired++;
5577                                 if (log)
5578                                         CDEBUG(D_LFSCK, "%s: namespace LFSCK "
5579                                                "assistant repaired the entry: "
5580                                                DFID", parent "DFID
5581                                                ", name %.*s\n",
5582                                                lfsck_lfsck2name(lfsck),
5583                                                PFID(&lnr->lnr_fid), PFID(pfid),
5584                                                lnr->lnr_namelen, lnr->lnr_name);
5585                         }
5586
5587                         if (bk->lb_param & LPF_DRYRUN &&
5588                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5589                                 lfsck_pos_fill(env, lfsck,
5590                                                &ns->ln_pos_first_inconsistent,
5591                                                false);
5592                 }
5593
5594                 rc = 0;
5595         }
5596
5597         if (count > 1 || la->la_nlink > 1)
5598                 ns->ln_mul_linked_checked++;
5599
5600         up_write(&com->lc_sem);
5601
5602         if (obj != NULL && !IS_ERR(obj))
5603                 lfsck_object_put(env, obj);
5604
5605         if (dir != NULL && !IS_ERR(dir))
5606                 lfsck_object_put(env, dir);
5607
5608         lad->lad_advance_lock = bad_linkea;
5609
5610         return rc;
5611 }
5612
5613 /**
5614  * Handle one orphan under the backend /lost+found directory
5615  *
5616  * Insert the orphan FID into the namespace LFSCK trace file for further
5617  * processing (via the subsequent namespace LFSCK second-stage scanning).
5618  * At the same time, remove the orphan name entry from backend /lost+found
5619  * directory. There is an interval between the orphan name entry removed
5620  * from the backend /lost+found directory and the orphan FID in the LFSCK
5621  * trace file handled. In such interval, the LFSCK can be reset, then
5622  * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5623  * To guarantee that the orphans can be found when LFSCK run next time
5624  * without e2fsck again, when remove the orphan name entry, the LFSCK
5625  * will set the orphan's ctime attribute as 1. Since normal applications
5626  * cannot change the object's ctime attribute as 1. Then when LFSCK run
5627  * next time, it can record the object (that ctime is 1) in the namespace
5628  * LFSCK trace file during the first-stage scanning.
5629  *
5630  * \param[in] env       pointer to the thread context
5631  * \param[in] com       pointer to the lfsck component
5632  * \param[in] parent    pointer to the object for the backend /lost+found
5633  * \param[in] ent       pointer to the name entry for the target under the
5634  *                      backend /lost+found
5635  *
5636  * \retval              positive for repaired
5637  * \retval              0 if needs to repair nothing
5638  * \retval              negative error number on failure
5639  */
5640 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5641                                               struct lfsck_component *com,
5642                                               struct dt_object *parent,
5643                                               struct lu_dirent *ent)
5644 {
5645         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5646         struct lu_fid                   *key    = &info->lti_fid;
5647         struct lu_attr                  *la     = &info->lti_la;
5648         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5649         struct dt_object                *obj;
5650         struct dt_device                *dev    = lfsck->li_bottom;
5651         struct dt_object                *child  = NULL;
5652         struct thandle                  *th     = NULL;
5653         int                              idx;
5654         int                              rc     = 0;
5655         __u8                             flags  = 0;
5656         bool                             exist  = false;
5657         ENTRY;
5658
5659         child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5660         if (IS_ERR(child))
5661                 RETURN(PTR_ERR(child));
5662
5663         LASSERT(dt_object_exists(child));
5664         LASSERT(!dt_object_remote(child));
5665
5666         idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
5667         obj = com->lc_sub_trace_objs[idx].lsto_obj;
5668         fid_cpu_to_be(key, &ent->lde_fid);
5669         rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5670                        (const struct dt_key *)key);
5671         if (rc == 0) {
5672                 exist = true;
5673                 flags |= LNTF_CHECK_ORPHAN;
5674         } else if (rc == -ENOENT) {
5675                 flags = LNTF_CHECK_ORPHAN;
5676         } else {
5677                 GOTO(out, rc);
5678         }
5679
5680         th = dt_trans_create(env, dev);
5681         if (IS_ERR(th))
5682                 GOTO(out, rc = PTR_ERR(th));
5683
5684         /* a1. remove name entry from backend /lost+found */
5685         rc = dt_declare_delete(env, parent,
5686                                (const struct dt_key *)ent->lde_name, th);
5687         if (rc != 0)
5688                 GOTO(stop, rc);
5689
5690         if (S_ISDIR(lfsck_object_type(child))) {
5691                 /* a2. decrease parent's nlink */
5692                 rc = dt_declare_ref_del(env, parent, th);
5693                 if (rc != 0)
5694                         GOTO(stop, rc);
5695         }
5696
5697         if (exist) {
5698                 /* a3. remove child's FID from the LFSCK trace file. */
5699                 rc = dt_declare_delete(env, obj,
5700                                        (const struct dt_key *)key, th);
5701                 if (rc != 0)
5702                         GOTO(stop, rc);
5703         } else {
5704                 /* a4. set child's ctime as 1 */
5705                 memset(la, 0, sizeof(*la));
5706                 la->la_ctime = 1;
5707                 la->la_valid = LA_CTIME;
5708                 rc = dt_declare_attr_set(env, child, la, th);
5709                 if (rc != 0)
5710                         GOTO(stop, rc);
5711         }
5712
5713         /* a5. insert child's FID into the LFSCK trace file. */
5714         rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5715                                (const struct dt_key *)key, th);
5716         if (rc != 0)
5717                 GOTO(stop, rc);
5718
5719         rc = dt_trans_start_local(env, dev, th);
5720         if (rc != 0)
5721                 GOTO(stop, rc);
5722
5723         /* b1. remove name entry from backend /lost+found */
5724         rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
5725         if (rc != 0)
5726                 GOTO(stop, rc);
5727
5728         if (S_ISDIR(lfsck_object_type(child))) {
5729                 /* b2. decrease parent's nlink */
5730                 dt_write_lock(env, parent, 0);
5731                 rc = dt_ref_del(env, parent, th);
5732                 dt_write_unlock(env, parent);
5733                 if (rc != 0)
5734                         GOTO(stop, rc);
5735         }
5736
5737         if (exist) {
5738                 /* a3. remove child's FID from the LFSCK trace file. */
5739                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
5740                 if (rc != 0)
5741                         GOTO(stop, rc);
5742         } else {
5743                 /* b4. set child's ctime as 1 */
5744                 rc = dt_attr_set(env, child, la, th);
5745                 if (rc != 0)
5746                         GOTO(stop, rc);
5747         }
5748
5749         /* b5. insert child's FID into the LFSCK trace file. */
5750         rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5751                        (const struct dt_key *)key, th, 1);
5752
5753         GOTO(stop, rc = (rc == 0 ? 1 : rc));
5754
5755 stop:
5756         dt_trans_stop(env, dev, th);
5757
5758 out:
5759         lfsck_object_put(env, child);
5760
5761         return rc;
5762 }
5763
5764 /**
5765  * Handle orphans under the backend /lost+found directory
5766  *
5767  * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5768  * and put them under the backend /lost+found directory that is invisible
5769  * to client. The LFSCK will scan such directory, for the original client
5770  * visible orphans, add their fids into the namespace LFSCK trace file,
5771  * then the subsenquent namespace LFSCK second-stage scanning can handle
5772  * them as other objects to be double scanned: either move back to normal
5773  * namespace, or to the global visible orphan directory:
5774  * /ROOT/.lustre/lost+found/MDTxxxx/
5775  *
5776  * \param[in] env       pointer to the thread context
5777  * \param[in] com       pointer to the lfsck component
5778  */
5779 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5780                                            struct lfsck_component *com)
5781 {
5782         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5783         struct lu_dirent                *ent    =
5784                                         (struct lu_dirent *)info->lti_key;
5785         struct lu_seq_range             *range  = &info->lti_range;
5786         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5787         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5788         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5789         struct lfsck_namespace          *ns     = com->lc_file_ram;
5790         struct dt_object                *parent;
5791         const struct dt_it_ops          *iops;
5792         struct dt_it                    *di;
5793         struct seq_server_site          *ss     = lfsck_dev_site(lfsck);
5794         __u64                            cookie;
5795         __u32                            idx    = lfsck_dev_idx(lfsck);
5796         int                              rc     = 0;
5797         __u16                            type;
5798         ENTRY;
5799
5800         parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
5801                                           &LU_BACKEND_LPF_FID);
5802         if (IS_ERR(parent)) {
5803                 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5804                        lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5805                 RETURN_EXIT;
5806         }
5807
5808         /* It is normal that the /lost+found does not exist for ZFS backend. */
5809         if (!dt_object_exists(parent))
5810                 GOTO(out, rc = 0);
5811
5812         if (unlikely(!dt_try_as_dir(env, parent)))
5813                 GOTO(out, rc = -ENOTDIR);
5814
5815         CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5816                lfsck_lfsck2name(lfsck));
5817
5818         com->lc_new_scanned = 0;
5819         iops = &parent->do_index_ops->dio_it;
5820         di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
5821         if (IS_ERR(di))
5822                 GOTO(out, rc = PTR_ERR(di));
5823
5824         rc = iops->load(env, di, 0);
5825         if (rc == 0)
5826                 rc = iops->next(env, di);
5827         else if (rc > 0)
5828                 rc = 0;
5829
5830         while (rc == 0) {
5831                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
5832                     unlikely(!thread_is_running(thread)))
5833                         break;
5834
5835                 rc = iops->rec(env, di, (struct dt_rec *)ent,
5836                                LUDA_64BITHASH | LUDA_TYPE);
5837                 if (rc == 0)
5838                         rc = lfsck_unpack_ent(ent, &cookie, &type);
5839
5840                 if (unlikely(rc != 0)) {
5841                         CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5842                                "/lost+found: rc = %d\n",
5843                                lfsck_lfsck2name(lfsck), rc);
5844
5845                         goto skip;
5846                 }
5847
5848                 /* skip dot and dotdot entries */
5849                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5850                         goto next;
5851
5852                 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5853                         goto skip;
5854
5855                 if (fid_is_norm(&ent->lde_fid)) {
5856                         fld_range_set_mdt(range);
5857                         rc = fld_local_lookup(env, ss->ss_server_fld,
5858                                               fid_seq(&ent->lde_fid), range);
5859                         if (rc != 0)
5860                                 goto skip;
5861                 } else if (idx != 0) {
5862                         /* If the returned FID is IGIF, then there are three
5863                          * possible cases:
5864                          *
5865                          * 1) The object is upgraded from old Lustre-1.8 with
5866                          *    IGIF assigned to such object.
5867                          * 2) The object is a backend local object and is
5868                          *    invisible to client.
5869                          * 3) The object lost its LMV EA, and since there is
5870                          *    no FID-in-dirent for the orphan in the backend
5871                          *    /lost+found directory, then the low layer will
5872                          *    return IGIF for such object.
5873                          *
5874                          * For MDTx (x != 0), it is either case 2) or case 3),
5875                          * but from the LFSCK view, they are indistinguishable.
5876                          * To be safe, the LFSCK will keep it there and report
5877                          * some message, then the adminstrator can handle that
5878                          * furtherly.
5879                          *
5880                          * For MDT0, it is more possible the case 1). The LFSCK
5881                          * will handle the orphan as an upgraded object. */
5882                         CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5883                                "in the backend /lost+found on the MDT %04x, "
5884                                "to be safe, skip it.\n",
5885                                lfsck_lfsck2name(lfsck), ent->lde_namelen,
5886                                ent->lde_name, PFID(&ent->lde_fid), idx);
5887                         goto skip;
5888                 }
5889
5890                 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5891
5892 skip:
5893                 down_write(&com->lc_sem);
5894                 com->lc_new_scanned++;
5895                 ns->ln_local_lpf_scanned++;
5896                 if (rc > 0)
5897                         ns->ln_local_lpf_moved++;
5898                 else if (rc == 0)
5899                         ns->ln_local_lpf_skipped++;
5900                 else
5901                         ns->ln_local_lpf_failed++;
5902                 up_write(&com->lc_sem);
5903
5904                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5905                         break;
5906
5907 next:
5908                 lfsck_control_speed_by_self(com);
5909                 if (unlikely(!thread_is_running(thread))) {
5910                         rc = 0;
5911                         break;
5912                 }
5913
5914                 rc = iops->next(env, di);
5915         }
5916
5917         iops->put(env, di);
5918         iops->fini(env, di);
5919
5920         EXIT;
5921
5922 out:
5923         CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5924                lfsck_lfsck2name(lfsck), rc);
5925
5926         lfsck_object_put(env, parent);
5927 }
5928
5929 /**
5930  * Rescan the striped directory after the master LMV EA reset.
5931  *
5932  * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5933  * the namespace LFSCK engine scan the striped directory for the first time,
5934  * it will be regarded as a normal directory. As the LFSCK processing, some
5935  * other LFSCK instance on other MDT will find the shard of this striped dir,
5936  * and find that the master MDT-object of the striped directory lost its LMV
5937  * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5938  * notify the LFSCK instance on this MDT to rescan the striped directory.
5939  *
5940  * \param[in] env       pointer to the thread context
5941  * \param[in] com       pointer to the lfsck component
5942  * \param[in] llu       the lfsck_lmv_unit that contains the striped directory
5943  *                      to be rescanned.
5944  *
5945  * \retval              positive number for success
5946  * \retval              0 for LFSCK stopped/paused
5947  * \retval              negative error number on failure
5948  */
5949 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5950                                               struct lfsck_component *com,
5951                                               struct lfsck_lmv_unit *llu)
5952 {
5953         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5954         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5955         struct lfsck_assistant_data     *lad    = com->lc_data;
5956         struct dt_object                *dir;
5957         const struct dt_it_ops          *iops;
5958         struct dt_it                    *di;
5959         struct lu_dirent                *ent    =
5960                         (struct lu_dirent *)info->lti_key;
5961         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5962         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5963         struct lfsck_assistant_object   *lso    = NULL;
5964         struct lfsck_namespace_req      *lnr;
5965         struct lfsck_assistant_req      *lar;
5966         int                              rc;
5967         __u16                            type;
5968         ENTRY;
5969
5970         LASSERT(list_empty(&lad->lad_req_list));
5971
5972         lfsck->li_lmv = &llu->llu_lmv;
5973         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5974         rc = lfsck_open_dir(env, lfsck, 0);
5975         if (rc != 0)
5976                 RETURN(rc);
5977
5978         dir = lfsck->li_obj_dir;
5979         di = lfsck->li_di_dir;
5980         iops = &dir->do_index_ops->dio_it;
5981         do {
5982                 rc = iops->rec(env, di, (struct dt_rec *)ent,
5983                                lfsck->li_args_dir);
5984                 if (rc == 0)
5985                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5986                                               &type);
5987
5988                 if (rc != 0) {
5989                         if (bk->lb_param & LPF_FAILOUT)
5990                                 GOTO(out, rc);
5991
5992                         goto next;
5993                 }
5994
5995                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5996                         goto next;
5997
5998                 if (lso == NULL) {
5999                         lso = lfsck_assistant_object_init(env,
6000                                 lfsck_dto2fid(dir), NULL,
6001                                 lfsck->li_pos_current.lp_oit_cookie, true);
6002                         if (IS_ERR(lso)) {
6003                                 if (bk->lb_param & LPF_FAILOUT)
6004                                         GOTO(out, rc = PTR_ERR(lso));
6005
6006                                 lso = NULL;
6007                                 goto next;
6008                         }
6009                 }
6010
6011                 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6012                 if (IS_ERR(lnr)) {
6013                         if (bk->lb_param & LPF_FAILOUT)
6014                                 GOTO(out, rc = PTR_ERR(lnr));
6015
6016                         goto next;
6017                 }
6018
6019                 lar = &lnr->lnr_lar;
6020                 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6021                 lfsck_namespace_assistant_req_fini(env, lar);
6022                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6023                         GOTO(out, rc);
6024
6025                 if (unlikely(!thread_is_running(thread)))
6026                         GOTO(out, rc = 0);
6027
6028 next:
6029                 rc = iops->next(env, di);
6030         } while (rc == 0);
6031
6032 out:
6033         if (lso != NULL && !IS_ERR(lso))
6034                 lfsck_assistant_object_put(env, lso);
6035
6036         lfsck_close_dir(env, lfsck, rc);
6037         if (rc <= 0)
6038                 RETURN(rc);
6039
6040         /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6041         if (list_empty(&lad->lad_req_list))
6042                 RETURN(1);
6043
6044         spin_lock(&lad->lad_lock);
6045         lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6046                           lar_list);
6047         list_del_init(&lar->lar_list);
6048         spin_unlock(&lad->lad_lock);
6049
6050         rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6051         lfsck_namespace_assistant_req_fini(env, lar);
6052
6053         RETURN(rc == 0 ? 1 : rc);
6054 }
6055
6056 static int
6057 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6058                                            struct lfsck_component *com,
6059                                            struct dt_object *obj, bool first)
6060 {
6061         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6062         struct ptlrpc_thread    *thread = &lfsck->li_thread;
6063         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
6064         struct lfsck_namespace  *ns     = com->lc_file_ram;
6065         const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
6066         struct dt_object        *target;
6067         struct dt_it            *di;
6068         struct dt_key           *key;
6069         struct lu_fid            fid;
6070         int                      rc;
6071         __u8                     flags  = 0;
6072         ENTRY;
6073
6074         di = iops->init(env, obj, 0);
6075         if (IS_ERR(di))
6076                 RETURN(PTR_ERR(di));
6077
6078         if (first)
6079                 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6080         else
6081                 fid_zero(&fid);
6082         rc = iops->get(env, di, (const struct dt_key *)&fid);
6083         if (rc < 0)
6084                 GOTO(fini, rc);
6085
6086         if (first) {
6087                 /* The start one either has been processed or does not exist,
6088                  * skip it. */
6089                 rc = iops->next(env, di);
6090                 if (rc != 0)
6091                         GOTO(put, rc);
6092         }
6093
6094         do {
6095                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6096                     unlikely(!thread_is_running(thread)))
6097                         GOTO(put, rc = 0);
6098
6099                 key = iops->key(env, di);
6100                 if (IS_ERR(key)) {
6101                         rc = PTR_ERR(key);
6102                         if (rc == -ENOENT)
6103                                 GOTO(put, rc = 1);
6104
6105                         goto checkpoint;
6106                 }
6107
6108                 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6109                 if (!fid_is_sane(&fid)) {
6110                         rc = 0;
6111                         goto checkpoint;
6112                 }
6113
6114                 target = lfsck_object_find_bottom(env, lfsck, &fid);
6115                 if (IS_ERR(target)) {
6116                         rc = PTR_ERR(target);
6117                         goto checkpoint;
6118                 }
6119
6120                 if (dt_object_exists(target)) {
6121                         rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6122                         if (rc == 0) {
6123                                 rc = lfsck_namespace_double_scan_one(env, com,
6124                                                                 target, flags);
6125                                 if (rc == -ENOENT)
6126                                         rc = 0;
6127                         }
6128                 }
6129
6130                 lfsck_object_put(env, target);
6131
6132 checkpoint:
6133                 down_write(&com->lc_sem);
6134                 com->lc_new_checked++;
6135                 com->lc_new_scanned++;
6136                 if (rc >= 0)
6137                         ns->ln_fid_latest_scanned_phase2 = fid;
6138
6139                 if (rc > 0)
6140                         ns->ln_objs_repaired_phase2++;
6141                 else if (rc < 0)
6142                         ns->ln_objs_failed_phase2++;
6143                 up_write(&com->lc_sem);
6144
6145                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6146                         GOTO(put, rc);
6147
6148                 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
6149                                               cfs_time_current())) &&
6150                     com->lc_new_checked != 0) {
6151                         down_write(&com->lc_sem);
6152                         ns->ln_run_time_phase2 +=
6153                                 cfs_duration_sec(cfs_time_current() +
6154                                 HALF_SEC - com->lc_time_last_checkpoint);
6155                         ns->ln_time_last_checkpoint = cfs_time_current_sec();
6156                         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6157                         com->lc_new_checked = 0;
6158                         lfsck_namespace_store(env, com);
6159                         up_write(&com->lc_sem);
6160
6161                         com->lc_time_last_checkpoint = cfs_time_current();
6162                         com->lc_time_next_checkpoint =
6163                                 com->lc_time_last_checkpoint +
6164                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6165                 }
6166
6167                 lfsck_control_speed_by_self(com);
6168                 if (unlikely(!thread_is_running(thread)))
6169                         GOTO(put, rc = 0);
6170
6171                 rc = iops->next(env, di);
6172         } while (rc == 0);
6173
6174         GOTO(put, rc);
6175
6176 put:
6177         iops->put(env, di);
6178
6179 fini:
6180         iops->fini(env, di);
6181
6182         return rc;
6183 }
6184
6185 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6186                                                 struct lfsck_component *com)
6187 {
6188         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6189         struct lfsck_namespace  *ns     = com->lc_file_ram;
6190         int                      rc;
6191         int                      i;
6192         ENTRY;
6193
6194         while (!list_empty(&lfsck->li_list_lmv)) {
6195                 struct lfsck_lmv_unit *llu;
6196
6197                 spin_lock(&lfsck->li_lock);
6198                 llu = list_entry(lfsck->li_list_lmv.next,
6199                                  struct lfsck_lmv_unit, llu_link);
6200                 list_del_init(&llu->llu_link);
6201                 spin_unlock(&lfsck->li_lock);
6202
6203                 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6204                 if (rc <= 0)
6205                         RETURN(rc);
6206         }
6207
6208         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6209                lfsck_lfsck2name(lfsck));
6210
6211         lfsck_namespace_scan_local_lpf(env, com);
6212
6213         com->lc_new_checked = 0;
6214         com->lc_new_scanned = 0;
6215         com->lc_time_last_checkpoint = cfs_time_current();
6216         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6217                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6218
6219         i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6220         rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6221                                 com->lc_sub_trace_objs[i].lsto_obj, true);
6222         while (rc > 0 && ++i < LFSCK_STF_COUNT)
6223                 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6224                                 com->lc_sub_trace_objs[i].lsto_obj, false);
6225
6226         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6227                "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6228
6229         RETURN(rc);
6230 }
6231
6232 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6233                                                struct lfsck_component *com,
6234                                                struct lfsck_position *pos)
6235 {
6236         struct lfsck_assistant_data     *lad = com->lc_data;
6237         struct lfsck_namespace_req      *lnr;
6238
6239         if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6240             LS_SCANNING_PHASE1)
6241                 return;
6242
6243         if (list_empty(&lad->lad_req_list))
6244                 return;
6245
6246         lnr = list_entry(lad->lad_req_list.next,
6247                          struct lfsck_namespace_req,
6248                          lnr_lar.lar_list);
6249         pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6250         pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6251         pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6252 }
6253
6254 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6255                                               struct lfsck_component *com,
6256                                               int rc)
6257 {
6258         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6259         struct lfsck_namespace  *ns     = com->lc_file_ram;
6260
6261         down_write(&com->lc_sem);
6262         ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
6263                                   HALF_SEC - com->lc_time_last_checkpoint);
6264         ns->ln_time_last_checkpoint = cfs_time_current_sec();
6265         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6266         com->lc_new_checked = 0;
6267
6268         if (rc > 0) {
6269                 if (ns->ln_flags & LF_INCOMPLETE)
6270                         ns->ln_status = LS_PARTIAL;
6271                 else
6272                         ns->ln_status = LS_COMPLETED;
6273                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6274                         ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
6275                 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6276                 ns->ln_success_count++;
6277         } else if (rc == 0) {
6278                 if (lfsck->li_status != 0)
6279                         ns->ln_status = lfsck->li_status;
6280                 else
6281                         ns->ln_status = LS_STOPPED;
6282         } else {
6283                 ns->ln_status = LS_FAILED;
6284         }
6285
6286         rc = lfsck_namespace_store(env, com);
6287         up_write(&com->lc_sem);
6288
6289         return rc;
6290 }
6291
6292 static int
6293 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6294                                                   struct ptlrpc_request *req,
6295                                                   void *args, int rc)
6296 {
6297         if (rc == 0) {
6298                 struct lfsck_async_interpret_args *laia = args;
6299                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
6300
6301                 ltd->ltd_synced_failures = 1;
6302         }
6303
6304         return 0;
6305 }
6306
6307 /**
6308  * Notify remote LFSCK instances about former failures.
6309  *
6310  * The local LFSCK instance has recorded which MDTs have ever failed to respond
6311  * some LFSCK verification requests (maybe because of network issues or the MDT
6312  * itself trouble). During the respond gap the MDT may missed some name entries
6313  * verification, then the MDT cannot know whether related MDT-objects have been
6314  * referenced by related name entries or not, then in the second-stage scanning,
6315  * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6316  * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6317  * name entry for repairing the orphan.
6318  *
6319  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6320  * it will scan the bitmap for the ever failed MDTs, and notify them that they
6321  * have ever missed some name entries verification and should skip the handling
6322  * for orphan MDT-objects.
6323  *
6324  * \param[in] env       pointer to the thread context
6325  * \param[in] com       pointer to the lfsck component
6326  * \param[in] lr        pointer to the lfsck request
6327  */
6328 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6329                                                     struct lfsck_component *com,
6330                                                     struct lfsck_request *lr)
6331 {
6332         struct lfsck_async_interpret_args *laia  =
6333                                 &lfsck_env_info(env)->lti_laia2;
6334         struct lfsck_assistant_data       *lad   = com->lc_data;
6335         struct lfsck_namespace            *ns    = com->lc_file_ram;
6336         struct lfsck_instance             *lfsck = com->lc_lfsck;
6337         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
6338         struct lfsck_tgt_desc             *ltd;
6339         struct ptlrpc_request_set         *set;
6340         __u32                              idx;
6341         int                                rc    = 0;
6342         ENTRY;
6343
6344         if (!lad->lad_incomplete)
6345                 RETURN_EXIT;
6346
6347         set = ptlrpc_prep_set();
6348         if (set == NULL)
6349                 GOTO(out, rc = -ENOMEM);
6350
6351         lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6352         memset(laia, 0, sizeof(*laia));
6353         lad->lad_touch_gen++;
6354
6355         down_read(&ltds->ltd_rw_sem);
6356         cfs_foreach_bit(lad->lad_bitmap, idx) {
6357                 ltd = lfsck_ltd2tgt(ltds, idx);
6358                 LASSERT(ltd != NULL);
6359
6360                 laia->laia_ltd = ltd;
6361                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6362                         lfsck_namespace_assistant_sync_failures_interpret,
6363                         laia, LFSCK_NOTIFY);
6364                 if (rc != 0)
6365                         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6366                                "to sync failure with MDT %x: rc = %d\n",
6367                                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6368         }
6369         up_read(&ltds->ltd_rw_sem);
6370
6371         rc = ptlrpc_set_wait(set);
6372         ptlrpc_set_destroy(set);
6373
6374         GOTO(out, rc);
6375
6376 out:
6377         if (rc != 0)
6378                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6379                        "to sync failure with MDTs, and related MDTs "
6380                        "may handle orphan improperly: rc = %d\n",
6381                        lfsck_lfsck2name(lfsck), rc);
6382
6383         EXIT;
6384 }
6385
6386 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6387         .la_handler_p1          = lfsck_namespace_assistant_handler_p1,
6388         .la_handler_p2          = lfsck_namespace_assistant_handler_p2,
6389         .la_fill_pos            = lfsck_namespace_assistant_fill_pos,
6390         .la_double_scan_result  = lfsck_namespace_double_scan_result,
6391         .la_req_fini            = lfsck_namespace_assistant_req_fini,
6392         .la_sync_failures       = lfsck_namespace_assistant_sync_failures,
6393 };
6394
6395 /**
6396  * Verify the specified linkEA entry for the given directory object.
6397  * If the object has no such linkEA entry or it has more other linkEA
6398  * entries, then re-generate the linkEA with the given information.
6399  *
6400  * \param[in] env       pointer to the thread context
6401  * \param[in] obj       pointer to the dt_object to be handled
6402  * \param[in] cname     the name for the child in the parent directory
6403  * \param[in] pfid      the parent directory's FID for the linkEA
6404  *
6405  * \retval              0 for success
6406  * \retval              negative error number on failure
6407  */
6408 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6409                         const struct lu_name *cname, const struct lu_fid *pfid)
6410 {
6411         struct dt_device        *dev    = lfsck_obj2dev(obj);
6412         struct linkea_data       ldata  = { NULL };
6413         struct lu_buf            linkea_buf;
6414         struct thandle          *th;
6415         int                      rc;
6416         int                      fl     = LU_XATTR_CREATE;
6417         bool                     dirty  = false;
6418         ENTRY;
6419
6420         LASSERT(S_ISDIR(lfsck_object_type(obj)));
6421
6422         rc = lfsck_links_read_with_rec(env, obj, &ldata);
6423         if (rc == -ENODATA) {
6424                 dirty = true;
6425         } else if (rc == 0) {
6426                 fl = LU_XATTR_REPLACE;
6427                 if (ldata.ld_leh->leh_reccount != 1) {
6428                         dirty = true;
6429                 } else {
6430                         rc = linkea_links_find(&ldata, cname, pfid);
6431                         if (rc != 0)
6432                                 dirty = true;
6433                 }
6434         }
6435
6436         if (!dirty)
6437                 RETURN(rc);
6438
6439         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6440                               cname, pfid);
6441         if (rc != 0)
6442                 RETURN(rc);
6443
6444         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6445                        ldata.ld_leh->leh_len);
6446         th = dt_trans_create(env, dev);
6447         if (IS_ERR(th))
6448                 RETURN(PTR_ERR(th));
6449
6450         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6451                                   XATTR_NAME_LINK, fl, th);
6452         if (rc != 0)
6453                 GOTO(stop, rc);
6454
6455         rc = dt_trans_start_local(env, dev, th);
6456         if (rc != 0)
6457                 GOTO(stop, rc);
6458
6459         dt_write_lock(env, obj, 0);
6460         rc = dt_xattr_set(env, obj, &linkea_buf,
6461                           XATTR_NAME_LINK, fl, th);
6462         dt_write_unlock(env, obj);
6463
6464         GOTO(stop, rc);
6465
6466 stop:
6467         dt_trans_stop(env, dev, th);
6468         return rc;
6469 }
6470
6471 /**
6472  * Get the name and parent directory's FID from the first linkEA entry.
6473  *
6474  * \param[in] env       pointer to the thread context
6475  * \param[in] obj       pointer to the object which get linkEA from
6476  * \param[out] name     pointer to the buffer to hold the name
6477  *                      in the first linkEA entry
6478  * \param[out] pfid     pointer to the buffer to hold the parent
6479  *                      directory's FID in the first linkEA entry
6480  *
6481  * \retval              0 for success
6482  * \retval              negative error number on failure
6483  */
6484 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6485                           char *name, struct lu_fid *pfid)
6486 {
6487         struct lu_name           *cname = &lfsck_env_info(env)->lti_name;
6488         struct linkea_data        ldata = { NULL };
6489         int                       rc;
6490
6491         rc = lfsck_links_read_with_rec(env, obj, &ldata);
6492         if (rc)
6493                 return rc;
6494
6495         linkea_first_entry(&ldata);
6496         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6497         /* To guarantee the 'name' is terminated with '0'. */
6498         memcpy(name, cname->ln_name, cname->ln_namelen);
6499         name[cname->ln_namelen] = 0;
6500
6501         return 0;
6502 }
6503
6504 /**
6505  * Update the object's name entry with the given FID.
6506  *
6507  * \param[in] env       pointer to the thread context
6508  * \param[in] lfsck     pointer to the lfsck instance
6509  * \param[in] dir       pointer to the directory that holds
6510  *                      the name entry
6511  * \param[in] name      the name for the entry to be updated
6512  * \param[in] fid       the new FID for the name entry referenced
6513  * \param[in] type      the type for the name entry to be updated
6514  *
6515  * \retval              0 for success
6516  * \retval              negative error number on failure
6517  */
6518 int lfsck_update_name_entry(const struct lu_env *env,
6519                             struct lfsck_instance *lfsck,
6520                             struct dt_object *dir, const char *name,
6521                             const struct lu_fid *fid, __u32 type)
6522 {
6523         struct lfsck_thread_info *info   = lfsck_env_info(env);
6524         struct dt_insert_rec     *rec    = &info->lti_dt_rec;
6525         struct lfsck_lock_handle *llh    = &info->lti_llh;
6526         struct dt_device         *dev    = lfsck_obj2dev(dir);
6527         struct thandle           *th;
6528         int                       rc;
6529         bool                      exists = true;
6530         ENTRY;
6531
6532         rc = lfsck_lock(env, lfsck, dir, name, llh,
6533                         MDS_INODELOCK_UPDATE, LCK_PW);
6534         if (rc != 0)
6535                 RETURN(rc);
6536
6537         th = dt_trans_create(env, dev);
6538         if (IS_ERR(th))
6539                 GOTO(unlock, rc = PTR_ERR(th));
6540
6541         rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6542         if (rc != 0)
6543                 GOTO(stop, rc);
6544
6545         rec->rec_type = type;
6546         rec->rec_fid = fid;
6547         rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6548                                (const struct dt_key *)name, th);
6549         if (rc != 0)
6550                 GOTO(stop, rc);
6551
6552         rc = dt_declare_ref_add(env, dir, th);
6553         if (rc != 0)
6554                 GOTO(stop, rc);
6555
6556         rc = dt_trans_start_local(env, dev, th);
6557         if (rc != 0)
6558                 GOTO(stop, rc);
6559
6560         rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6561         if (rc == -ENOENT) {
6562                 exists = false;
6563                 rc = 0;
6564         }
6565
6566         if (rc != 0)
6567                 GOTO(stop, rc);
6568
6569         rc = dt_insert(env, dir, (const struct dt_rec *)rec,
6570                        (const struct dt_key *)name, th, 1);
6571         if (rc == 0 && S_ISDIR(type) && !exists) {
6572                 dt_write_lock(env, dir, 0);
6573                 rc = dt_ref_add(env, dir, th);
6574                 dt_write_unlock(env, dir);
6575         }
6576
6577         GOTO(stop, rc);
6578
6579 stop:
6580         dt_trans_stop(env, dev, th);
6581
6582 unlock:
6583         lfsck_unlock(llh);
6584         CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6585                " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6586                PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
6587
6588         return rc;
6589 }
6590
6591 int lfsck_namespace_setup(const struct lu_env *env,
6592                           struct lfsck_instance *lfsck)
6593 {
6594         struct lfsck_component  *com;
6595         struct lfsck_namespace  *ns;
6596         struct dt_object        *root = NULL;
6597         struct dt_object        *obj;
6598         int                      i;
6599         int                      rc;
6600         ENTRY;
6601
6602         LASSERT(lfsck->li_master);
6603
6604         OBD_ALLOC_PTR(com);
6605         if (com == NULL)
6606                 RETURN(-ENOMEM);
6607
6608         INIT_LIST_HEAD(&com->lc_link);
6609         INIT_LIST_HEAD(&com->lc_link_dir);
6610         init_rwsem(&com->lc_sem);
6611         atomic_set(&com->lc_ref, 1);
6612         com->lc_lfsck = lfsck;
6613         com->lc_type = LFSCK_TYPE_NAMESPACE;
6614         com->lc_ops = &lfsck_namespace_ops;
6615         com->lc_data = lfsck_assistant_data_init(
6616                         &lfsck_namespace_assistant_ops,
6617                         LFSCK_NAMESPACE);
6618         if (com->lc_data == NULL)
6619                 GOTO(out, rc = -ENOMEM);
6620
6621         com->lc_file_size = sizeof(struct lfsck_namespace);
6622         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6623         if (com->lc_file_ram == NULL)
6624                 GOTO(out, rc = -ENOMEM);
6625
6626         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6627         if (com->lc_file_disk == NULL)
6628                 GOTO(out, rc = -ENOMEM);
6629
6630         for (i = 0; i < LFSCK_STF_COUNT; i++)
6631                 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6632
6633         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6634         if (IS_ERR(root))
6635                 GOTO(out, rc = PTR_ERR(root));
6636
6637         if (unlikely(!dt_try_as_dir(env, root)))
6638                 GOTO(out, rc = -ENOTDIR);
6639
6640         obj = local_index_find_or_create(env, lfsck->li_los, root,
6641                                          LFSCK_NAMESPACE,
6642                                          S_IFREG | S_IRUGO | S_IWUSR,
6643                                          &dt_lfsck_namespace_features);
6644         if (IS_ERR(obj))
6645                 GOTO(out, rc = PTR_ERR(obj));
6646
6647         com->lc_obj = obj;
6648         rc = lfsck_namespace_load(env, com);
6649         if (rc == -ENODATA)
6650                 rc = lfsck_namespace_init(env, com);
6651         else if (rc < 0)
6652                 rc = lfsck_namespace_reset(env, com, true);
6653         else
6654                 rc = lfsck_load_sub_trace_files(env, com,
6655                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
6656         if (rc != 0)
6657                 GOTO(out, rc);
6658
6659         ns = com->lc_file_ram;
6660         switch (ns->ln_status) {
6661         case LS_INIT:
6662         case LS_COMPLETED:
6663         case LS_FAILED:
6664         case LS_STOPPED:
6665                 spin_lock(&lfsck->li_lock);
6666                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6667                 spin_unlock(&lfsck->li_lock);
6668                 break;
6669         default:
6670                 CERROR("%s: unknown lfsck_namespace status %d\n",
6671                        lfsck_lfsck2name(lfsck), ns->ln_status);
6672                 /* fall through */
6673         case LS_SCANNING_PHASE1:
6674         case LS_SCANNING_PHASE2:
6675                 /* No need to store the status to disk right now.
6676                  * If the system crashed before the status stored,
6677                  * it will be loaded back when next time. */
6678                 ns->ln_status = LS_CRASHED;
6679                 /* fall through */
6680         case LS_PAUSED:
6681         case LS_CRASHED:
6682                 spin_lock(&lfsck->li_lock);
6683                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6684                 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6685                 spin_unlock(&lfsck->li_lock);
6686                 break;
6687         }
6688
6689         GOTO(out, rc = 0);
6690
6691 out:
6692         if (root != NULL && !IS_ERR(root))
6693                 lfsck_object_put(env, root);
6694         if (rc != 0) {
6695                 lfsck_component_cleanup(env, com);
6696                 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6697                        lfsck_lfsck2name(lfsck), rc);
6698         }
6699         return rc;
6700 }